1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * super.c contains code to handle: - mount structures
9 * - filesystem drivers list
11 * - umount system call
14 * GK 2/5/95 - Changed to support mounting the root fs via NFS
16 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18 * Added options to /proc/mounts:
19 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
24 #include <linux/export.h>
25 #include <linux/slab.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h> /* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include <linux/user_namespace.h>
41 static LIST_HEAD(super_blocks
);
42 static DEFINE_SPINLOCK(sb_lock
);
44 static char *sb_writers_name
[SB_FREEZE_LEVELS
] = {
51 * One thing we have to be careful of with a per-sb shrinker is that we don't
52 * drop the last active reference to the superblock from within the shrinker.
53 * If that happens we could trigger unregistering the shrinker from within the
54 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
55 * take a passive reference to the superblock to avoid this from occurring.
57 static unsigned long super_cache_scan(struct shrinker
*shrink
,
58 struct shrink_control
*sc
)
60 struct super_block
*sb
;
67 sb
= container_of(shrink
, struct super_block
, s_shrink
);
70 * Deadlock avoidance. We may hold various FS locks, and we don't want
71 * to recurse into the FS that called us in clear_inode() and friends..
73 if (!(sc
->gfp_mask
& __GFP_FS
))
76 if (!trylock_super(sb
))
79 if (sb
->s_op
->nr_cached_objects
)
80 fs_objects
= sb
->s_op
->nr_cached_objects(sb
, sc
);
82 inodes
= list_lru_shrink_count(&sb
->s_inode_lru
, sc
);
83 dentries
= list_lru_shrink_count(&sb
->s_dentry_lru
, sc
);
84 total_objects
= dentries
+ inodes
+ fs_objects
+ 1;
88 /* proportion the scan between the caches */
89 dentries
= mult_frac(sc
->nr_to_scan
, dentries
, total_objects
);
90 inodes
= mult_frac(sc
->nr_to_scan
, inodes
, total_objects
);
91 fs_objects
= mult_frac(sc
->nr_to_scan
, fs_objects
, total_objects
);
94 * prune the dcache first as the icache is pinned by it, then
95 * prune the icache, followed by the filesystem specific caches
97 * Ensure that we always scan at least one object - memcg kmem
98 * accounting uses this to fully empty the caches.
100 sc
->nr_to_scan
= dentries
+ 1;
101 freed
= prune_dcache_sb(sb
, sc
);
102 sc
->nr_to_scan
= inodes
+ 1;
103 freed
+= prune_icache_sb(sb
, sc
);
106 sc
->nr_to_scan
= fs_objects
+ 1;
107 freed
+= sb
->s_op
->free_cached_objects(sb
, sc
);
110 up_read(&sb
->s_umount
);
114 static unsigned long super_cache_count(struct shrinker
*shrink
,
115 struct shrink_control
*sc
)
117 struct super_block
*sb
;
118 long total_objects
= 0;
120 sb
= container_of(shrink
, struct super_block
, s_shrink
);
123 * Don't call trylock_super as it is a potential
124 * scalability bottleneck. The counts could get updated
125 * between super_cache_count and super_cache_scan anyway.
126 * Call to super_cache_count with shrinker_rwsem held
127 * ensures the safety of call to list_lru_shrink_count() and
128 * s_op->nr_cached_objects().
130 if (sb
->s_op
&& sb
->s_op
->nr_cached_objects
)
131 total_objects
= sb
->s_op
->nr_cached_objects(sb
, sc
);
133 total_objects
+= list_lru_shrink_count(&sb
->s_dentry_lru
, sc
);
134 total_objects
+= list_lru_shrink_count(&sb
->s_inode_lru
, sc
);
136 total_objects
= vfs_pressure_ratio(total_objects
);
137 return total_objects
;
140 static void destroy_super_work(struct work_struct
*work
)
142 struct super_block
*s
= container_of(work
, struct super_block
,
146 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++)
147 percpu_free_rwsem(&s
->s_writers
.rw_sem
[i
]);
151 static void destroy_super_rcu(struct rcu_head
*head
)
153 struct super_block
*s
= container_of(head
, struct super_block
, rcu
);
154 INIT_WORK(&s
->destroy_work
, destroy_super_work
);
155 schedule_work(&s
->destroy_work
);
158 /* Free a superblock that has never been seen by anyone */
159 static void destroy_unused_super(struct super_block
*s
)
163 up_write(&s
->s_umount
);
164 list_lru_destroy(&s
->s_dentry_lru
);
165 list_lru_destroy(&s
->s_inode_lru
);
167 put_user_ns(s
->s_user_ns
);
169 free_prealloced_shrinker(&s
->s_shrink
);
170 /* no delays needed */
171 destroy_super_work(&s
->destroy_work
);
175 * alloc_super - create new superblock
176 * @type: filesystem type superblock should belong to
177 * @flags: the mount flags
178 * @user_ns: User namespace for the super_block
180 * Allocates and initializes a new &struct super_block. alloc_super()
181 * returns a pointer new superblock or %NULL if allocation had failed.
183 static struct super_block
*alloc_super(struct file_system_type
*type
, int flags
,
184 struct user_namespace
*user_ns
)
186 struct super_block
*s
= kzalloc(sizeof(struct super_block
), GFP_USER
);
187 static const struct super_operations default_op
;
193 INIT_LIST_HEAD(&s
->s_mounts
);
194 s
->s_user_ns
= get_user_ns(user_ns
);
195 init_rwsem(&s
->s_umount
);
196 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
198 * sget() can have s_umount recursion.
200 * When it cannot find a suitable sb, it allocates a new
201 * one (this one), and tries again to find a suitable old
204 * In case that succeeds, it will acquire the s_umount
205 * lock of the old one. Since these are clearly distrinct
206 * locks, and this object isn't exposed yet, there's no
209 * Annotate this by putting this lock in a different
212 down_write_nested(&s
->s_umount
, SINGLE_DEPTH_NESTING
);
214 if (security_sb_alloc(s
))
217 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++) {
218 if (__percpu_init_rwsem(&s
->s_writers
.rw_sem
[i
],
220 &type
->s_writers_key
[i
]))
223 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
224 s
->s_bdi
= &noop_backing_dev_info
;
226 if (s
->s_user_ns
!= &init_user_ns
)
227 s
->s_iflags
|= SB_I_NODEV
;
228 INIT_HLIST_NODE(&s
->s_instances
);
229 INIT_HLIST_BL_HEAD(&s
->s_anon
);
230 mutex_init(&s
->s_sync_lock
);
231 INIT_LIST_HEAD(&s
->s_inodes
);
232 spin_lock_init(&s
->s_inode_list_lock
);
233 INIT_LIST_HEAD(&s
->s_inodes_wb
);
234 spin_lock_init(&s
->s_inode_wblist_lock
);
236 if (list_lru_init_memcg(&s
->s_dentry_lru
))
238 if (list_lru_init_memcg(&s
->s_inode_lru
))
241 atomic_set(&s
->s_active
, 1);
242 mutex_init(&s
->s_vfs_rename_mutex
);
243 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
244 init_rwsem(&s
->s_dquot
.dqio_sem
);
245 s
->s_maxbytes
= MAX_NON_LFS
;
246 s
->s_op
= &default_op
;
247 s
->s_time_gran
= 1000000000;
248 s
->cleancache_poolid
= CLEANCACHE_NO_POOL
;
250 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
251 s
->s_shrink
.scan_objects
= super_cache_scan
;
252 s
->s_shrink
.count_objects
= super_cache_count
;
253 s
->s_shrink
.batch
= 1024;
254 s
->s_shrink
.flags
= SHRINKER_NUMA_AWARE
| SHRINKER_MEMCG_AWARE
;
255 if (prealloc_shrinker(&s
->s_shrink
))
260 destroy_unused_super(s
);
264 /* Superblock refcounting */
267 * Drop a superblock's refcount. The caller must hold sb_lock.
269 static void __put_super(struct super_block
*s
)
272 list_del_init(&s
->s_list
);
273 WARN_ON(s
->s_dentry_lru
.node
);
274 WARN_ON(s
->s_inode_lru
.node
);
275 WARN_ON(!list_empty(&s
->s_mounts
));
277 put_user_ns(s
->s_user_ns
);
279 call_rcu(&s
->rcu
, destroy_super_rcu
);
284 * put_super - drop a temporary reference to superblock
285 * @sb: superblock in question
287 * Drops a temporary reference, frees superblock if there's no
290 static void put_super(struct super_block
*sb
)
294 spin_unlock(&sb_lock
);
299 * deactivate_locked_super - drop an active reference to superblock
300 * @s: superblock to deactivate
302 * Drops an active reference to superblock, converting it into a temporary
303 * one if there is no other active references left. In that case we
304 * tell fs driver to shut it down and drop the temporary reference we
307 * Caller holds exclusive lock on superblock; that lock is released.
309 void deactivate_locked_super(struct super_block
*s
)
311 struct file_system_type
*fs
= s
->s_type
;
312 if (atomic_dec_and_test(&s
->s_active
)) {
313 cleancache_invalidate_fs(s
);
314 unregister_shrinker(&s
->s_shrink
);
318 * Since list_lru_destroy() may sleep, we cannot call it from
319 * put_super(), where we hold the sb_lock. Therefore we destroy
320 * the lru lists right now.
322 list_lru_destroy(&s
->s_dentry_lru
);
323 list_lru_destroy(&s
->s_inode_lru
);
328 up_write(&s
->s_umount
);
332 EXPORT_SYMBOL(deactivate_locked_super
);
335 * deactivate_super - drop an active reference to superblock
336 * @s: superblock to deactivate
338 * Variant of deactivate_locked_super(), except that superblock is *not*
339 * locked by caller. If we are going to drop the final active reference,
340 * lock will be acquired prior to that.
342 void deactivate_super(struct super_block
*s
)
344 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
345 down_write(&s
->s_umount
);
346 deactivate_locked_super(s
);
350 EXPORT_SYMBOL(deactivate_super
);
353 * grab_super - acquire an active reference
354 * @s: reference we are trying to make active
356 * Tries to acquire an active reference. grab_super() is used when we
357 * had just found a superblock in super_blocks or fs_type->fs_supers
358 * and want to turn it into a full-blown active reference. grab_super()
359 * is called with sb_lock held and drops it. Returns 1 in case of
360 * success, 0 if we had failed (superblock contents was already dead or
361 * dying when grab_super() had been called). Note that this is only
362 * called for superblocks not in rundown mode (== ones still on ->fs_supers
363 * of their type), so increment of ->s_count is OK here.
365 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
368 spin_unlock(&sb_lock
);
369 down_write(&s
->s_umount
);
370 if ((s
->s_flags
& SB_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
374 up_write(&s
->s_umount
);
380 * trylock_super - try to grab ->s_umount shared
381 * @sb: reference we are trying to grab
383 * Try to prevent fs shutdown. This is used in places where we
384 * cannot take an active reference but we need to ensure that the
385 * filesystem is not shut down while we are working on it. It returns
386 * false if we cannot acquire s_umount or if we lose the race and
387 * filesystem already got into shutdown, and returns true with the s_umount
388 * lock held in read mode in case of success. On successful return,
389 * the caller must drop the s_umount lock when done.
391 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
392 * The reason why it's safe is that we are OK with doing trylock instead
393 * of down_read(). There's a couple of places that are OK with that, but
394 * it's very much not a general-purpose interface.
396 bool trylock_super(struct super_block
*sb
)
398 if (down_read_trylock(&sb
->s_umount
)) {
399 if (!hlist_unhashed(&sb
->s_instances
) &&
400 sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
402 up_read(&sb
->s_umount
);
409 * generic_shutdown_super - common helper for ->kill_sb()
410 * @sb: superblock to kill
412 * generic_shutdown_super() does all fs-independent work on superblock
413 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
414 * that need destruction out of superblock, call generic_shutdown_super()
415 * and release aforementioned objects. Note: dentries and inodes _are_
416 * taken care of and do not need specific handling.
418 * Upon calling this function, the filesystem may no longer alter or
419 * rearrange the set of dentries belonging to this super_block, nor may it
420 * change the attachments of dentries to inodes.
422 void generic_shutdown_super(struct super_block
*sb
)
424 const struct super_operations
*sop
= sb
->s_op
;
427 shrink_dcache_for_umount(sb
);
429 sb
->s_flags
&= ~SB_ACTIVE
;
431 fsnotify_unmount_inodes(sb
);
432 cgroup_writeback_umount();
436 if (sb
->s_dio_done_wq
) {
437 destroy_workqueue(sb
->s_dio_done_wq
);
438 sb
->s_dio_done_wq
= NULL
;
444 if (!list_empty(&sb
->s_inodes
)) {
445 printk("VFS: Busy inodes after unmount of %s. "
446 "Self-destruct in 5 seconds. Have a nice day...\n",
451 /* should be initialized for __put_super_and_need_restart() */
452 hlist_del_init(&sb
->s_instances
);
453 spin_unlock(&sb_lock
);
454 up_write(&sb
->s_umount
);
455 if (sb
->s_bdi
!= &noop_backing_dev_info
) {
457 sb
->s_bdi
= &noop_backing_dev_info
;
461 EXPORT_SYMBOL(generic_shutdown_super
);
464 * sget_userns - find or create a superblock
465 * @type: filesystem type superblock should belong to
466 * @test: comparison callback
467 * @set: setup callback
468 * @flags: mount flags
469 * @user_ns: User namespace for the super_block
470 * @data: argument to each of them
472 struct super_block
*sget_userns(struct file_system_type
*type
,
473 int (*test
)(struct super_block
*,void *),
474 int (*set
)(struct super_block
*,void *),
475 int flags
, struct user_namespace
*user_ns
,
478 struct super_block
*s
= NULL
;
479 struct super_block
*old
;
482 if (!(flags
& (SB_KERNMOUNT
|SB_SUBMOUNT
)) &&
483 !(type
->fs_flags
& FS_USERNS_MOUNT
) &&
484 !capable(CAP_SYS_ADMIN
))
485 return ERR_PTR(-EPERM
);
489 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
490 if (!test(old
, data
))
492 if (user_ns
!= old
->s_user_ns
) {
493 spin_unlock(&sb_lock
);
494 destroy_unused_super(s
);
495 return ERR_PTR(-EBUSY
);
497 if (!grab_super(old
))
499 destroy_unused_super(s
);
504 spin_unlock(&sb_lock
);
505 s
= alloc_super(type
, (flags
& ~SB_SUBMOUNT
), user_ns
);
507 return ERR_PTR(-ENOMEM
);
513 spin_unlock(&sb_lock
);
514 destroy_unused_super(s
);
518 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
519 list_add_tail(&s
->s_list
, &super_blocks
);
520 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
521 spin_unlock(&sb_lock
);
522 get_filesystem(type
);
523 register_shrinker_prepared(&s
->s_shrink
);
527 EXPORT_SYMBOL(sget_userns
);
530 * sget - find or create a superblock
531 * @type: filesystem type superblock should belong to
532 * @test: comparison callback
533 * @set: setup callback
534 * @flags: mount flags
535 * @data: argument to each of them
537 struct super_block
*sget(struct file_system_type
*type
,
538 int (*test
)(struct super_block
*,void *),
539 int (*set
)(struct super_block
*,void *),
543 struct user_namespace
*user_ns
= current_user_ns();
545 /* We don't yet pass the user namespace of the parent
546 * mount through to here so always use &init_user_ns
547 * until that changes.
549 if (flags
& SB_SUBMOUNT
)
550 user_ns
= &init_user_ns
;
552 /* Ensure the requestor has permissions over the target filesystem */
553 if (!(flags
& (SB_KERNMOUNT
|SB_SUBMOUNT
)) && !ns_capable(user_ns
, CAP_SYS_ADMIN
))
554 return ERR_PTR(-EPERM
);
556 return sget_userns(type
, test
, set
, flags
, user_ns
, data
);
561 void drop_super(struct super_block
*sb
)
563 up_read(&sb
->s_umount
);
567 EXPORT_SYMBOL(drop_super
);
569 void drop_super_exclusive(struct super_block
*sb
)
571 up_write(&sb
->s_umount
);
574 EXPORT_SYMBOL(drop_super_exclusive
);
577 * iterate_supers - call function for all active superblocks
578 * @f: function to call
579 * @arg: argument to pass to it
581 * Scans the superblock list and calls given function, passing it
582 * locked superblock and given argument.
584 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
586 struct super_block
*sb
, *p
= NULL
;
589 list_for_each_entry(sb
, &super_blocks
, s_list
) {
590 if (hlist_unhashed(&sb
->s_instances
))
593 spin_unlock(&sb_lock
);
595 down_read(&sb
->s_umount
);
596 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
598 up_read(&sb
->s_umount
);
607 spin_unlock(&sb_lock
);
611 * iterate_supers_type - call function for superblocks of given type
613 * @f: function to call
614 * @arg: argument to pass to it
616 * Scans the superblock list and calls given function, passing it
617 * locked superblock and given argument.
619 void iterate_supers_type(struct file_system_type
*type
,
620 void (*f
)(struct super_block
*, void *), void *arg
)
622 struct super_block
*sb
, *p
= NULL
;
625 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
627 spin_unlock(&sb_lock
);
629 down_read(&sb
->s_umount
);
630 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
632 up_read(&sb
->s_umount
);
641 spin_unlock(&sb_lock
);
644 EXPORT_SYMBOL(iterate_supers_type
);
646 static struct super_block
*__get_super(struct block_device
*bdev
, bool excl
)
648 struct super_block
*sb
;
655 list_for_each_entry(sb
, &super_blocks
, s_list
) {
656 if (hlist_unhashed(&sb
->s_instances
))
658 if (sb
->s_bdev
== bdev
) {
660 spin_unlock(&sb_lock
);
662 down_read(&sb
->s_umount
);
664 down_write(&sb
->s_umount
);
666 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
669 up_read(&sb
->s_umount
);
671 up_write(&sb
->s_umount
);
672 /* nope, got unmounted */
678 spin_unlock(&sb_lock
);
683 * get_super - get the superblock of a device
684 * @bdev: device to get the superblock for
686 * Scans the superblock list and finds the superblock of the file system
687 * mounted on the device given. %NULL is returned if no match is found.
689 struct super_block
*get_super(struct block_device
*bdev
)
691 return __get_super(bdev
, false);
693 EXPORT_SYMBOL(get_super
);
695 static struct super_block
*__get_super_thawed(struct block_device
*bdev
,
699 struct super_block
*s
= __get_super(bdev
, excl
);
700 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
703 up_read(&s
->s_umount
);
705 up_write(&s
->s_umount
);
706 wait_event(s
->s_writers
.wait_unfrozen
,
707 s
->s_writers
.frozen
== SB_UNFROZEN
);
713 * get_super_thawed - get thawed superblock of a device
714 * @bdev: device to get the superblock for
716 * Scans the superblock list and finds the superblock of the file system
717 * mounted on the device. The superblock is returned once it is thawed
718 * (or immediately if it was not frozen). %NULL is returned if no match
721 struct super_block
*get_super_thawed(struct block_device
*bdev
)
723 return __get_super_thawed(bdev
, false);
725 EXPORT_SYMBOL(get_super_thawed
);
728 * get_super_exclusive_thawed - get thawed superblock of a device
729 * @bdev: device to get the superblock for
731 * Scans the superblock list and finds the superblock of the file system
732 * mounted on the device. The superblock is returned once it is thawed
733 * (or immediately if it was not frozen) and s_umount semaphore is held
734 * in exclusive mode. %NULL is returned if no match is found.
736 struct super_block
*get_super_exclusive_thawed(struct block_device
*bdev
)
738 return __get_super_thawed(bdev
, true);
740 EXPORT_SYMBOL(get_super_exclusive_thawed
);
743 * get_active_super - get an active reference to the superblock of a device
744 * @bdev: device to get the superblock for
746 * Scans the superblock list and finds the superblock of the file system
747 * mounted on the device given. Returns the superblock with an active
748 * reference or %NULL if none was found.
750 struct super_block
*get_active_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
) {
765 up_write(&sb
->s_umount
);
769 spin_unlock(&sb_lock
);
773 struct super_block
*user_get_super(dev_t dev
)
775 struct super_block
*sb
;
779 list_for_each_entry(sb
, &super_blocks
, s_list
) {
780 if (hlist_unhashed(&sb
->s_instances
))
782 if (sb
->s_dev
== dev
) {
784 spin_unlock(&sb_lock
);
785 down_read(&sb
->s_umount
);
787 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
789 up_read(&sb
->s_umount
);
790 /* nope, got unmounted */
796 spin_unlock(&sb_lock
);
801 * do_remount_sb - asks filesystem to change mount options.
802 * @sb: superblock in question
803 * @sb_flags: revised superblock flags
804 * @data: the rest of options
805 * @force: whether or not to force the change
807 * Alters the mount options of a mounted file system.
809 int do_remount_sb(struct super_block
*sb
, int sb_flags
, void *data
, int force
)
814 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
818 if (!(sb_flags
& SB_RDONLY
) && bdev_read_only(sb
->s_bdev
))
822 remount_ro
= (sb_flags
& SB_RDONLY
) && !sb_rdonly(sb
);
825 if (!hlist_empty(&sb
->s_pins
)) {
826 up_write(&sb
->s_umount
);
827 group_pin_kill(&sb
->s_pins
);
828 down_write(&sb
->s_umount
);
831 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
833 remount_ro
= (sb_flags
& SB_RDONLY
) && !sb_rdonly(sb
);
836 shrink_dcache_sb(sb
);
838 /* If we are remounting RDONLY and current sb is read/write,
839 make sure there are no rw files opened */
842 sb
->s_readonly_remount
= 1;
845 retval
= sb_prepare_remount_readonly(sb
);
851 if (sb
->s_op
->remount_fs
) {
852 retval
= sb
->s_op
->remount_fs(sb
, &sb_flags
, data
);
855 goto cancel_readonly
;
856 /* If forced remount, go ahead despite any errors */
857 WARN(1, "forced remount of a %s fs returned %i\n",
858 sb
->s_type
->name
, retval
);
861 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (sb_flags
& MS_RMT_MASK
);
862 /* Needs to be ordered wrt mnt_is_readonly() */
864 sb
->s_readonly_remount
= 0;
867 * Some filesystems modify their metadata via some other path than the
868 * bdev buffer cache (eg. use a private mapping, or directories in
869 * pagecache, etc). Also file data modifications go via their own
870 * mappings. So If we try to mount readonly then copy the filesystem
871 * from bdev, we could get stale data, so invalidate it to give a best
872 * effort at coherency.
874 if (remount_ro
&& sb
->s_bdev
)
875 invalidate_bdev(sb
->s_bdev
);
879 sb
->s_readonly_remount
= 0;
883 static void do_emergency_remount(struct work_struct
*work
)
885 struct super_block
*sb
, *p
= NULL
;
888 list_for_each_entry(sb
, &super_blocks
, s_list
) {
889 if (hlist_unhashed(&sb
->s_instances
))
892 spin_unlock(&sb_lock
);
893 down_write(&sb
->s_umount
);
894 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& SB_BORN
) &&
897 * What lock protects sb->s_flags??
899 do_remount_sb(sb
, SB_RDONLY
, NULL
, 1);
901 up_write(&sb
->s_umount
);
909 spin_unlock(&sb_lock
);
911 printk("Emergency Remount complete\n");
914 void emergency_remount(void)
916 struct work_struct
*work
;
918 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
920 INIT_WORK(work
, do_emergency_remount
);
926 * Unnamed block devices are dummy devices used by virtual
927 * filesystems which don't use real block-devices. -- jrs
930 static DEFINE_IDA(unnamed_dev_ida
);
931 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
932 /* Many userspace utilities consider an FSID of 0 invalid.
933 * Always return at least 1 from get_anon_bdev.
935 static int unnamed_dev_start
= 1;
937 int get_anon_bdev(dev_t
*p
)
943 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
945 spin_lock(&unnamed_dev_lock
);
946 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
948 unnamed_dev_start
= dev
+ 1;
949 spin_unlock(&unnamed_dev_lock
);
950 if (error
== -EAGAIN
)
951 /* We raced and lost with another CPU. */
956 if (dev
>= (1 << MINORBITS
)) {
957 spin_lock(&unnamed_dev_lock
);
958 ida_remove(&unnamed_dev_ida
, dev
);
959 if (unnamed_dev_start
> dev
)
960 unnamed_dev_start
= dev
;
961 spin_unlock(&unnamed_dev_lock
);
964 *p
= MKDEV(0, dev
& MINORMASK
);
967 EXPORT_SYMBOL(get_anon_bdev
);
969 void free_anon_bdev(dev_t dev
)
971 int slot
= MINOR(dev
);
972 spin_lock(&unnamed_dev_lock
);
973 ida_remove(&unnamed_dev_ida
, slot
);
974 if (slot
< unnamed_dev_start
)
975 unnamed_dev_start
= slot
;
976 spin_unlock(&unnamed_dev_lock
);
978 EXPORT_SYMBOL(free_anon_bdev
);
980 int set_anon_super(struct super_block
*s
, void *data
)
982 return get_anon_bdev(&s
->s_dev
);
985 EXPORT_SYMBOL(set_anon_super
);
987 void kill_anon_super(struct super_block
*sb
)
989 dev_t dev
= sb
->s_dev
;
990 generic_shutdown_super(sb
);
994 EXPORT_SYMBOL(kill_anon_super
);
996 void kill_litter_super(struct super_block
*sb
)
999 d_genocide(sb
->s_root
);
1000 kill_anon_super(sb
);
1003 EXPORT_SYMBOL(kill_litter_super
);
1005 static int ns_test_super(struct super_block
*sb
, void *data
)
1007 return sb
->s_fs_info
== data
;
1010 static int ns_set_super(struct super_block
*sb
, void *data
)
1012 sb
->s_fs_info
= data
;
1013 return set_anon_super(sb
, NULL
);
1016 struct dentry
*mount_ns(struct file_system_type
*fs_type
,
1017 int flags
, void *data
, void *ns
, struct user_namespace
*user_ns
,
1018 int (*fill_super
)(struct super_block
*, void *, int))
1020 struct super_block
*sb
;
1022 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1023 * over the namespace.
1025 if (!(flags
& SB_KERNMOUNT
) && !ns_capable(user_ns
, CAP_SYS_ADMIN
))
1026 return ERR_PTR(-EPERM
);
1028 sb
= sget_userns(fs_type
, ns_test_super
, ns_set_super
, flags
,
1031 return ERR_CAST(sb
);
1035 err
= fill_super(sb
, data
, flags
& SB_SILENT
? 1 : 0);
1037 deactivate_locked_super(sb
);
1038 return ERR_PTR(err
);
1041 sb
->s_flags
|= SB_ACTIVE
;
1044 return dget(sb
->s_root
);
1047 EXPORT_SYMBOL(mount_ns
);
1050 static int set_bdev_super(struct super_block
*s
, void *data
)
1053 s
->s_dev
= s
->s_bdev
->bd_dev
;
1054 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_bdi
);
1059 static int test_bdev_super(struct super_block
*s
, void *data
)
1061 return (void *)s
->s_bdev
== data
;
1064 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
1065 int flags
, const char *dev_name
, void *data
,
1066 int (*fill_super
)(struct super_block
*, void *, int))
1068 struct block_device
*bdev
;
1069 struct super_block
*s
;
1070 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1073 if (!(flags
& SB_RDONLY
))
1074 mode
|= FMODE_WRITE
;
1076 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1078 return ERR_CAST(bdev
);
1080 if (current_user_ns() != &init_user_ns
) {
1082 * For userns mounts, disallow mounting if bdev is open for
1085 if (!atomic_dec_unless_positive(&bdev
->bd_inode
->i_writecount
)) {
1089 if (bdev
->bd_contains
!= bdev
&&
1090 !atomic_dec_unless_positive(&bdev
->bd_contains
->bd_inode
->i_writecount
)) {
1091 atomic_inc(&bdev
->bd_inode
->i_writecount
);
1098 * once the super is inserted into the list by sget, s_umount
1099 * will protect the lockfs code from trying to start a snapshot
1100 * while we are mounting
1102 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1103 if (bdev
->bd_fsfreeze_count
> 0) {
1104 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1108 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| SB_NOSEC
,
1110 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1115 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1116 deactivate_locked_super(s
);
1122 * s_umount nests inside bd_mutex during
1123 * __invalidate_device(). blkdev_put() acquires
1124 * bd_mutex and can't be called under s_umount. Drop
1125 * s_umount temporarily. This is safe as we're
1126 * holding an active reference.
1128 up_write(&s
->s_umount
);
1129 blkdev_put(bdev
, mode
);
1130 down_write(&s
->s_umount
);
1133 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1134 sb_set_blocksize(s
, block_size(bdev
));
1135 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1137 deactivate_locked_super(s
);
1141 s
->s_flags
|= SB_ACTIVE
;
1145 return dget(s
->s_root
);
1150 if (current_user_ns() != &init_user_ns
) {
1151 atomic_inc(&bdev
->bd_inode
->i_writecount
);
1152 if (bdev
->bd_contains
!= bdev
)
1153 atomic_inc(&bdev
->bd_contains
->bd_inode
->i_writecount
);
1156 blkdev_put(bdev
, mode
);
1158 return ERR_PTR(error
);
1160 EXPORT_SYMBOL(mount_bdev
);
1162 void kill_block_super(struct super_block
*sb
)
1164 struct block_device
*bdev
= sb
->s_bdev
;
1165 fmode_t mode
= sb
->s_mode
;
1167 bdev
->bd_super
= NULL
;
1168 generic_shutdown_super(sb
);
1169 sync_blockdev(bdev
);
1170 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1171 if (sb
->s_user_ns
!= &init_user_ns
) {
1172 atomic_inc(&bdev
->bd_inode
->i_writecount
);
1173 if (bdev
->bd_contains
!= bdev
)
1174 atomic_inc(&bdev
->bd_contains
->bd_inode
->i_writecount
);
1176 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1179 EXPORT_SYMBOL(kill_block_super
);
1182 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1183 int flags
, void *data
,
1184 int (*fill_super
)(struct super_block
*, void *, int))
1187 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1192 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1194 deactivate_locked_super(s
);
1195 return ERR_PTR(error
);
1197 s
->s_flags
|= SB_ACTIVE
;
1198 return dget(s
->s_root
);
1200 EXPORT_SYMBOL(mount_nodev
);
1202 static int compare_single(struct super_block
*s
, void *p
)
1207 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1208 int flags
, void *data
,
1209 int (*fill_super
)(struct super_block
*, void *, int))
1211 struct super_block
*s
;
1214 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1218 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1220 deactivate_locked_super(s
);
1221 return ERR_PTR(error
);
1223 s
->s_flags
|= SB_ACTIVE
;
1225 do_remount_sb(s
, flags
, data
, 0);
1227 return dget(s
->s_root
);
1229 EXPORT_SYMBOL(mount_single
);
1232 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1234 struct dentry
*root
;
1235 struct super_block
*sb
;
1236 char *secdata
= NULL
;
1237 int error
= -ENOMEM
;
1239 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1240 secdata
= alloc_secdata();
1244 error
= security_sb_copy_data(data
, secdata
);
1246 goto out_free_secdata
;
1249 root
= type
->mount(type
, flags
, name
, data
);
1251 error
= PTR_ERR(root
);
1252 goto out_free_secdata
;
1256 WARN_ON(!sb
->s_bdi
);
1257 sb
->s_flags
|= SB_BORN
;
1259 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1264 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1265 * but s_maxbytes was an unsigned long long for many releases. Throw
1266 * this warning for a little while to try and catch filesystems that
1267 * violate this rule.
1269 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1270 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1272 up_write(&sb
->s_umount
);
1273 free_secdata(secdata
);
1277 deactivate_locked_super(sb
);
1279 free_secdata(secdata
);
1281 return ERR_PTR(error
);
1285 * Setup private BDI for given superblock. It gets automatically cleaned up
1286 * in generic_shutdown_super().
1288 int super_setup_bdi_name(struct super_block
*sb
, char *fmt
, ...)
1290 struct backing_dev_info
*bdi
;
1294 bdi
= bdi_alloc(GFP_KERNEL
);
1298 bdi
->name
= sb
->s_type
->name
;
1300 va_start(args
, fmt
);
1301 err
= bdi_register_va(bdi
, fmt
, args
);
1307 WARN_ON(sb
->s_bdi
!= &noop_backing_dev_info
);
1312 EXPORT_SYMBOL(super_setup_bdi_name
);
1315 * Setup private BDI for given superblock. I gets automatically cleaned up
1316 * in generic_shutdown_super().
1318 int super_setup_bdi(struct super_block
*sb
)
1320 static atomic_long_t bdi_seq
= ATOMIC_LONG_INIT(0);
1322 return super_setup_bdi_name(sb
, "%.28s-%ld", sb
->s_type
->name
,
1323 atomic_long_inc_return(&bdi_seq
));
1325 EXPORT_SYMBOL(super_setup_bdi
);
1328 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1331 void __sb_end_write(struct super_block
*sb
, int level
)
1333 percpu_up_read(sb
->s_writers
.rw_sem
+ level
-1);
1335 EXPORT_SYMBOL(__sb_end_write
);
1338 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1341 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1343 bool force_trylock
= false;
1346 #ifdef CONFIG_LOCKDEP
1348 * We want lockdep to tell us about possible deadlocks with freezing
1349 * but it's it bit tricky to properly instrument it. Getting a freeze
1350 * protection works as getting a read lock but there are subtle
1351 * problems. XFS for example gets freeze protection on internal level
1352 * twice in some cases, which is OK only because we already hold a
1353 * freeze protection also on higher level. Due to these cases we have
1354 * to use wait == F (trylock mode) which must not fail.
1359 for (i
= 0; i
< level
- 1; i
++)
1360 if (percpu_rwsem_is_held(sb
->s_writers
.rw_sem
+ i
)) {
1361 force_trylock
= true;
1366 if (wait
&& !force_trylock
)
1367 percpu_down_read(sb
->s_writers
.rw_sem
+ level
-1);
1369 ret
= percpu_down_read_trylock(sb
->s_writers
.rw_sem
+ level
-1);
1371 WARN_ON(force_trylock
&& !ret
);
1374 EXPORT_SYMBOL(__sb_start_write
);
1377 * sb_wait_write - wait until all writers to given file system finish
1378 * @sb: the super for which we wait
1379 * @level: type of writers we wait for (normal vs page fault)
1381 * This function waits until there are no writers of given type to given file
1384 static void sb_wait_write(struct super_block
*sb
, int level
)
1386 percpu_down_write(sb
->s_writers
.rw_sem
+ level
-1);
1390 * We are going to return to userspace and forget about these locks, the
1391 * ownership goes to the caller of thaw_super() which does unlock().
1393 static void lockdep_sb_freeze_release(struct super_block
*sb
)
1397 for (level
= SB_FREEZE_LEVELS
- 1; level
>= 0; level
--)
1398 percpu_rwsem_release(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1402 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1404 static void lockdep_sb_freeze_acquire(struct super_block
*sb
)
1408 for (level
= 0; level
< SB_FREEZE_LEVELS
; ++level
)
1409 percpu_rwsem_acquire(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1412 static void sb_freeze_unlock(struct super_block
*sb
)
1416 for (level
= SB_FREEZE_LEVELS
- 1; level
>= 0; level
--)
1417 percpu_up_write(sb
->s_writers
.rw_sem
+ level
);
1421 * freeze_super - lock the filesystem and force it into a consistent state
1422 * @sb: the super to lock
1424 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1425 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1428 * During this function, sb->s_writers.frozen goes through these values:
1430 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1432 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1433 * writes should be blocked, though page faults are still allowed. We wait for
1434 * all writes to complete and then proceed to the next stage.
1436 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1437 * but internal fs threads can still modify the filesystem (although they
1438 * should not dirty new pages or inodes), writeback can run etc. After waiting
1439 * for all running page faults we sync the filesystem which will clean all
1440 * dirty pages and inodes (no new dirty pages or inodes can be created when
1443 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1444 * modification are blocked (e.g. XFS preallocation truncation on inode
1445 * reclaim). This is usually implemented by blocking new transactions for
1446 * filesystems that have them and need this additional guard. After all
1447 * internal writers are finished we call ->freeze_fs() to finish filesystem
1448 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1449 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1451 * sb->s_writers.frozen is protected by sb->s_umount.
1453 int freeze_super(struct super_block
*sb
)
1457 atomic_inc(&sb
->s_active
);
1458 down_write(&sb
->s_umount
);
1459 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1460 deactivate_locked_super(sb
);
1464 if (!(sb
->s_flags
& SB_BORN
)) {
1465 up_write(&sb
->s_umount
);
1466 return 0; /* sic - it's "nothing to do" */
1469 if (sb_rdonly(sb
)) {
1470 /* Nothing to do really... */
1471 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1472 up_write(&sb
->s_umount
);
1476 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1477 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1478 up_write(&sb
->s_umount
);
1479 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1480 down_write(&sb
->s_umount
);
1482 /* Now we go and block page faults... */
1483 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1484 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1486 /* All writers are done so after syncing there won't be dirty data */
1487 sync_filesystem(sb
);
1489 /* Now wait for internal filesystem counter */
1490 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1491 sb_wait_write(sb
, SB_FREEZE_FS
);
1493 if (sb
->s_op
->freeze_fs
) {
1494 ret
= sb
->s_op
->freeze_fs(sb
);
1497 "VFS:Filesystem freeze failed\n");
1498 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1499 sb_freeze_unlock(sb
);
1500 wake_up(&sb
->s_writers
.wait_unfrozen
);
1501 deactivate_locked_super(sb
);
1506 * For debugging purposes so that fs can warn if it sees write activity
1507 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1509 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1510 lockdep_sb_freeze_release(sb
);
1511 up_write(&sb
->s_umount
);
1514 EXPORT_SYMBOL(freeze_super
);
1517 * thaw_super -- unlock filesystem
1518 * @sb: the super to thaw
1520 * Unlocks the filesystem and marks it writeable again after freeze_super().
1522 int thaw_super(struct super_block
*sb
)
1526 down_write(&sb
->s_umount
);
1527 if (sb
->s_writers
.frozen
!= SB_FREEZE_COMPLETE
) {
1528 up_write(&sb
->s_umount
);
1532 if (sb_rdonly(sb
)) {
1533 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1537 lockdep_sb_freeze_acquire(sb
);
1539 if (sb
->s_op
->unfreeze_fs
) {
1540 error
= sb
->s_op
->unfreeze_fs(sb
);
1543 "VFS:Filesystem thaw failed\n");
1544 lockdep_sb_freeze_release(sb
);
1545 up_write(&sb
->s_umount
);
1550 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1551 sb_freeze_unlock(sb
);
1553 wake_up(&sb
->s_writers
.wait_unfrozen
);
1554 deactivate_locked_super(sb
);
1557 EXPORT_SYMBOL(thaw_super
);