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 * We don't call trylock_super() here as it is a scalability bottleneck,
124 * so we're exposed to partial setup state. The shrinker rwsem does not
125 * protect filesystem operations backing list_lru_shrink_count() or
126 * s_op->nr_cached_objects(). Counts can change between
127 * super_cache_count and super_cache_scan, so we really don't need locks
130 * However, if we are currently mounting the superblock, the underlying
131 * filesystem might be in a state of partial construction and hence it
132 * is dangerous to access it. trylock_super() uses a SB_BORN check to
133 * avoid this situation, so do the same here. The memory barrier is
134 * matched with the one in mount_fs() as we don't hold locks here.
136 if (!(sb
->s_flags
& SB_BORN
))
140 if (sb
->s_op
&& sb
->s_op
->nr_cached_objects
)
141 total_objects
= sb
->s_op
->nr_cached_objects(sb
, sc
);
143 total_objects
+= list_lru_shrink_count(&sb
->s_dentry_lru
, sc
);
144 total_objects
+= list_lru_shrink_count(&sb
->s_inode_lru
, sc
);
146 total_objects
= vfs_pressure_ratio(total_objects
);
147 return total_objects
;
150 static void destroy_super_work(struct work_struct
*work
)
152 struct super_block
*s
= container_of(work
, struct super_block
,
156 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++)
157 percpu_free_rwsem(&s
->s_writers
.rw_sem
[i
]);
161 static void destroy_super_rcu(struct rcu_head
*head
)
163 struct super_block
*s
= container_of(head
, struct super_block
, rcu
);
164 INIT_WORK(&s
->destroy_work
, destroy_super_work
);
165 schedule_work(&s
->destroy_work
);
168 /* Free a superblock that has never been seen by anyone */
169 static void destroy_unused_super(struct super_block
*s
)
173 up_write(&s
->s_umount
);
174 list_lru_destroy(&s
->s_dentry_lru
);
175 list_lru_destroy(&s
->s_inode_lru
);
177 put_user_ns(s
->s_user_ns
);
179 free_prealloced_shrinker(&s
->s_shrink
);
180 /* no delays needed */
181 destroy_super_work(&s
->destroy_work
);
185 * alloc_super - create new superblock
186 * @type: filesystem type superblock should belong to
187 * @flags: the mount flags
188 * @user_ns: User namespace for the super_block
190 * Allocates and initializes a new &struct super_block. alloc_super()
191 * returns a pointer new superblock or %NULL if allocation had failed.
193 static struct super_block
*alloc_super(struct file_system_type
*type
, int flags
,
194 struct user_namespace
*user_ns
)
196 struct super_block
*s
= kzalloc(sizeof(struct super_block
), GFP_USER
);
197 static const struct super_operations default_op
;
203 INIT_LIST_HEAD(&s
->s_mounts
);
204 s
->s_user_ns
= get_user_ns(user_ns
);
205 init_rwsem(&s
->s_umount
);
206 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
208 * sget() can have s_umount recursion.
210 * When it cannot find a suitable sb, it allocates a new
211 * one (this one), and tries again to find a suitable old
214 * In case that succeeds, it will acquire the s_umount
215 * lock of the old one. Since these are clearly distrinct
216 * locks, and this object isn't exposed yet, there's no
219 * Annotate this by putting this lock in a different
222 down_write_nested(&s
->s_umount
, SINGLE_DEPTH_NESTING
);
224 if (security_sb_alloc(s
))
227 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++) {
228 if (__percpu_init_rwsem(&s
->s_writers
.rw_sem
[i
],
230 &type
->s_writers_key
[i
]))
233 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
234 s
->s_bdi
= &noop_backing_dev_info
;
236 if (s
->s_user_ns
!= &init_user_ns
)
237 s
->s_iflags
|= SB_I_NODEV
;
238 INIT_HLIST_NODE(&s
->s_instances
);
239 INIT_HLIST_BL_HEAD(&s
->s_anon
);
240 mutex_init(&s
->s_sync_lock
);
241 INIT_LIST_HEAD(&s
->s_inodes
);
242 spin_lock_init(&s
->s_inode_list_lock
);
243 INIT_LIST_HEAD(&s
->s_inodes_wb
);
244 spin_lock_init(&s
->s_inode_wblist_lock
);
246 if (list_lru_init_memcg(&s
->s_dentry_lru
))
248 if (list_lru_init_memcg(&s
->s_inode_lru
))
251 atomic_set(&s
->s_active
, 1);
252 mutex_init(&s
->s_vfs_rename_mutex
);
253 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
254 init_rwsem(&s
->s_dquot
.dqio_sem
);
255 s
->s_maxbytes
= MAX_NON_LFS
;
256 s
->s_op
= &default_op
;
257 s
->s_time_gran
= 1000000000;
258 s
->cleancache_poolid
= CLEANCACHE_NO_POOL
;
260 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
261 s
->s_shrink
.scan_objects
= super_cache_scan
;
262 s
->s_shrink
.count_objects
= super_cache_count
;
263 s
->s_shrink
.batch
= 1024;
264 s
->s_shrink
.flags
= SHRINKER_NUMA_AWARE
| SHRINKER_MEMCG_AWARE
;
265 if (prealloc_shrinker(&s
->s_shrink
))
270 destroy_unused_super(s
);
274 /* Superblock refcounting */
277 * Drop a superblock's refcount. The caller must hold sb_lock.
279 static void __put_super(struct super_block
*s
)
282 list_del_init(&s
->s_list
);
283 WARN_ON(s
->s_dentry_lru
.node
);
284 WARN_ON(s
->s_inode_lru
.node
);
285 WARN_ON(!list_empty(&s
->s_mounts
));
287 put_user_ns(s
->s_user_ns
);
289 call_rcu(&s
->rcu
, destroy_super_rcu
);
294 * put_super - drop a temporary reference to superblock
295 * @sb: superblock in question
297 * Drops a temporary reference, frees superblock if there's no
300 static void put_super(struct super_block
*sb
)
304 spin_unlock(&sb_lock
);
309 * deactivate_locked_super - drop an active reference to superblock
310 * @s: superblock to deactivate
312 * Drops an active reference to superblock, converting it into a temporary
313 * one if there is no other active references left. In that case we
314 * tell fs driver to shut it down and drop the temporary reference we
317 * Caller holds exclusive lock on superblock; that lock is released.
319 void deactivate_locked_super(struct super_block
*s
)
321 struct file_system_type
*fs
= s
->s_type
;
322 if (atomic_dec_and_test(&s
->s_active
)) {
323 cleancache_invalidate_fs(s
);
324 unregister_shrinker(&s
->s_shrink
);
328 * Since list_lru_destroy() may sleep, we cannot call it from
329 * put_super(), where we hold the sb_lock. Therefore we destroy
330 * the lru lists right now.
332 list_lru_destroy(&s
->s_dentry_lru
);
333 list_lru_destroy(&s
->s_inode_lru
);
338 up_write(&s
->s_umount
);
342 EXPORT_SYMBOL(deactivate_locked_super
);
345 * deactivate_super - drop an active reference to superblock
346 * @s: superblock to deactivate
348 * Variant of deactivate_locked_super(), except that superblock is *not*
349 * locked by caller. If we are going to drop the final active reference,
350 * lock will be acquired prior to that.
352 void deactivate_super(struct super_block
*s
)
354 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
355 down_write(&s
->s_umount
);
356 deactivate_locked_super(s
);
360 EXPORT_SYMBOL(deactivate_super
);
363 * grab_super - acquire an active reference
364 * @s: reference we are trying to make active
366 * Tries to acquire an active reference. grab_super() is used when we
367 * had just found a superblock in super_blocks or fs_type->fs_supers
368 * and want to turn it into a full-blown active reference. grab_super()
369 * is called with sb_lock held and drops it. Returns 1 in case of
370 * success, 0 if we had failed (superblock contents was already dead or
371 * dying when grab_super() had been called). Note that this is only
372 * called for superblocks not in rundown mode (== ones still on ->fs_supers
373 * of their type), so increment of ->s_count is OK here.
375 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
378 spin_unlock(&sb_lock
);
379 down_write(&s
->s_umount
);
380 if ((s
->s_flags
& SB_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
384 up_write(&s
->s_umount
);
390 * trylock_super - try to grab ->s_umount shared
391 * @sb: reference we are trying to grab
393 * Try to prevent fs shutdown. This is used in places where we
394 * cannot take an active reference but we need to ensure that the
395 * filesystem is not shut down while we are working on it. It returns
396 * false if we cannot acquire s_umount or if we lose the race and
397 * filesystem already got into shutdown, and returns true with the s_umount
398 * lock held in read mode in case of success. On successful return,
399 * the caller must drop the s_umount lock when done.
401 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
402 * The reason why it's safe is that we are OK with doing trylock instead
403 * of down_read(). There's a couple of places that are OK with that, but
404 * it's very much not a general-purpose interface.
406 bool trylock_super(struct super_block
*sb
)
408 if (down_read_trylock(&sb
->s_umount
)) {
409 if (!hlist_unhashed(&sb
->s_instances
) &&
410 sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
412 up_read(&sb
->s_umount
);
419 * generic_shutdown_super - common helper for ->kill_sb()
420 * @sb: superblock to kill
422 * generic_shutdown_super() does all fs-independent work on superblock
423 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
424 * that need destruction out of superblock, call generic_shutdown_super()
425 * and release aforementioned objects. Note: dentries and inodes _are_
426 * taken care of and do not need specific handling.
428 * Upon calling this function, the filesystem may no longer alter or
429 * rearrange the set of dentries belonging to this super_block, nor may it
430 * change the attachments of dentries to inodes.
432 void generic_shutdown_super(struct super_block
*sb
)
434 const struct super_operations
*sop
= sb
->s_op
;
437 shrink_dcache_for_umount(sb
);
439 sb
->s_flags
&= ~SB_ACTIVE
;
441 fsnotify_unmount_inodes(sb
);
442 cgroup_writeback_umount();
446 if (sb
->s_dio_done_wq
) {
447 destroy_workqueue(sb
->s_dio_done_wq
);
448 sb
->s_dio_done_wq
= NULL
;
454 if (!list_empty(&sb
->s_inodes
)) {
455 printk("VFS: Busy inodes after unmount of %s. "
456 "Self-destruct in 5 seconds. Have a nice day...\n",
461 /* should be initialized for __put_super_and_need_restart() */
462 hlist_del_init(&sb
->s_instances
);
463 spin_unlock(&sb_lock
);
464 up_write(&sb
->s_umount
);
465 if (sb
->s_bdi
!= &noop_backing_dev_info
) {
467 sb
->s_bdi
= &noop_backing_dev_info
;
471 EXPORT_SYMBOL(generic_shutdown_super
);
474 * sget_userns - find or create a superblock
475 * @type: filesystem type superblock should belong to
476 * @test: comparison callback
477 * @set: setup callback
478 * @flags: mount flags
479 * @user_ns: User namespace for the super_block
480 * @data: argument to each of them
482 struct super_block
*sget_userns(struct file_system_type
*type
,
483 int (*test
)(struct super_block
*,void *),
484 int (*set
)(struct super_block
*,void *),
485 int flags
, struct user_namespace
*user_ns
,
488 struct super_block
*s
= NULL
;
489 struct super_block
*old
;
492 if (!(flags
& (SB_KERNMOUNT
|SB_SUBMOUNT
)) &&
493 !(type
->fs_flags
& FS_USERNS_MOUNT
) &&
494 !capable(CAP_SYS_ADMIN
))
495 return ERR_PTR(-EPERM
);
499 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
500 if (!test(old
, data
))
502 if (user_ns
!= old
->s_user_ns
) {
503 spin_unlock(&sb_lock
);
504 destroy_unused_super(s
);
505 return ERR_PTR(-EBUSY
);
507 if (!grab_super(old
))
509 destroy_unused_super(s
);
514 spin_unlock(&sb_lock
);
515 s
= alloc_super(type
, (flags
& ~SB_SUBMOUNT
), user_ns
);
517 return ERR_PTR(-ENOMEM
);
523 spin_unlock(&sb_lock
);
524 destroy_unused_super(s
);
528 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
529 list_add_tail(&s
->s_list
, &super_blocks
);
530 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
531 spin_unlock(&sb_lock
);
532 get_filesystem(type
);
533 register_shrinker_prepared(&s
->s_shrink
);
537 EXPORT_SYMBOL(sget_userns
);
540 * sget - find or create a superblock
541 * @type: filesystem type superblock should belong to
542 * @test: comparison callback
543 * @set: setup callback
544 * @flags: mount flags
545 * @data: argument to each of them
547 struct super_block
*sget(struct file_system_type
*type
,
548 int (*test
)(struct super_block
*,void *),
549 int (*set
)(struct super_block
*,void *),
553 struct user_namespace
*user_ns
= current_user_ns();
555 /* We don't yet pass the user namespace of the parent
556 * mount through to here so always use &init_user_ns
557 * until that changes.
559 if (flags
& SB_SUBMOUNT
)
560 user_ns
= &init_user_ns
;
562 /* Ensure the requestor has permissions over the target filesystem */
563 if (!(flags
& (SB_KERNMOUNT
|SB_SUBMOUNT
)) && !ns_capable(user_ns
, CAP_SYS_ADMIN
))
564 return ERR_PTR(-EPERM
);
566 return sget_userns(type
, test
, set
, flags
, user_ns
, data
);
571 void drop_super(struct super_block
*sb
)
573 up_read(&sb
->s_umount
);
577 EXPORT_SYMBOL(drop_super
);
579 void drop_super_exclusive(struct super_block
*sb
)
581 up_write(&sb
->s_umount
);
584 EXPORT_SYMBOL(drop_super_exclusive
);
587 * iterate_supers - call function for all active superblocks
588 * @f: function to call
589 * @arg: argument to pass to it
591 * Scans the superblock list and calls given function, passing it
592 * locked superblock and given argument.
594 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
596 struct super_block
*sb
, *p
= NULL
;
599 list_for_each_entry(sb
, &super_blocks
, s_list
) {
600 if (hlist_unhashed(&sb
->s_instances
))
603 spin_unlock(&sb_lock
);
605 down_read(&sb
->s_umount
);
606 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
608 up_read(&sb
->s_umount
);
617 spin_unlock(&sb_lock
);
621 * iterate_supers_type - call function for superblocks of given type
623 * @f: function to call
624 * @arg: argument to pass to it
626 * Scans the superblock list and calls given function, passing it
627 * locked superblock and given argument.
629 void iterate_supers_type(struct file_system_type
*type
,
630 void (*f
)(struct super_block
*, void *), void *arg
)
632 struct super_block
*sb
, *p
= NULL
;
635 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
637 spin_unlock(&sb_lock
);
639 down_read(&sb
->s_umount
);
640 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
642 up_read(&sb
->s_umount
);
651 spin_unlock(&sb_lock
);
654 EXPORT_SYMBOL(iterate_supers_type
);
656 static struct super_block
*__get_super(struct block_device
*bdev
, bool excl
)
658 struct super_block
*sb
;
665 list_for_each_entry(sb
, &super_blocks
, s_list
) {
666 if (hlist_unhashed(&sb
->s_instances
))
668 if (sb
->s_bdev
== bdev
) {
670 spin_unlock(&sb_lock
);
672 down_read(&sb
->s_umount
);
674 down_write(&sb
->s_umount
);
676 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
679 up_read(&sb
->s_umount
);
681 up_write(&sb
->s_umount
);
682 /* nope, got unmounted */
688 spin_unlock(&sb_lock
);
693 * get_super - get the superblock of a device
694 * @bdev: device to get the superblock for
696 * Scans the superblock list and finds the superblock of the file system
697 * mounted on the device given. %NULL is returned if no match is found.
699 struct super_block
*get_super(struct block_device
*bdev
)
701 return __get_super(bdev
, false);
703 EXPORT_SYMBOL(get_super
);
705 static struct super_block
*__get_super_thawed(struct block_device
*bdev
,
709 struct super_block
*s
= __get_super(bdev
, excl
);
710 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
713 up_read(&s
->s_umount
);
715 up_write(&s
->s_umount
);
716 wait_event(s
->s_writers
.wait_unfrozen
,
717 s
->s_writers
.frozen
== SB_UNFROZEN
);
723 * get_super_thawed - get thawed superblock of a device
724 * @bdev: device to get the superblock for
726 * Scans the superblock list and finds the superblock of the file system
727 * mounted on the device. The superblock is returned once it is thawed
728 * (or immediately if it was not frozen). %NULL is returned if no match
731 struct super_block
*get_super_thawed(struct block_device
*bdev
)
733 return __get_super_thawed(bdev
, false);
735 EXPORT_SYMBOL(get_super_thawed
);
738 * get_super_exclusive_thawed - get thawed superblock of a device
739 * @bdev: device to get the superblock for
741 * Scans the superblock list and finds the superblock of the file system
742 * mounted on the device. The superblock is returned once it is thawed
743 * (or immediately if it was not frozen) and s_umount semaphore is held
744 * in exclusive mode. %NULL is returned if no match is found.
746 struct super_block
*get_super_exclusive_thawed(struct block_device
*bdev
)
748 return __get_super_thawed(bdev
, true);
750 EXPORT_SYMBOL(get_super_exclusive_thawed
);
753 * get_active_super - get an active reference to the superblock of a device
754 * @bdev: device to get the superblock for
756 * Scans the superblock list and finds the superblock of the file system
757 * mounted on the device given. Returns the superblock with an active
758 * reference or %NULL if none was found.
760 struct super_block
*get_active_super(struct block_device
*bdev
)
762 struct super_block
*sb
;
769 list_for_each_entry(sb
, &super_blocks
, s_list
) {
770 if (hlist_unhashed(&sb
->s_instances
))
772 if (sb
->s_bdev
== bdev
) {
775 up_write(&sb
->s_umount
);
779 spin_unlock(&sb_lock
);
783 struct super_block
*user_get_super(dev_t dev
)
785 struct super_block
*sb
;
789 list_for_each_entry(sb
, &super_blocks
, s_list
) {
790 if (hlist_unhashed(&sb
->s_instances
))
792 if (sb
->s_dev
== dev
) {
794 spin_unlock(&sb_lock
);
795 down_read(&sb
->s_umount
);
797 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
799 up_read(&sb
->s_umount
);
800 /* nope, got unmounted */
806 spin_unlock(&sb_lock
);
811 * do_remount_sb - asks filesystem to change mount options.
812 * @sb: superblock in question
813 * @sb_flags: revised superblock flags
814 * @data: the rest of options
815 * @force: whether or not to force the change
817 * Alters the mount options of a mounted file system.
819 int do_remount_sb(struct super_block
*sb
, int sb_flags
, void *data
, int force
)
824 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
828 if (!(sb_flags
& SB_RDONLY
) && bdev_read_only(sb
->s_bdev
))
832 remount_ro
= (sb_flags
& SB_RDONLY
) && !sb_rdonly(sb
);
835 if (!hlist_empty(&sb
->s_pins
)) {
836 up_write(&sb
->s_umount
);
837 group_pin_kill(&sb
->s_pins
);
838 down_write(&sb
->s_umount
);
841 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
843 remount_ro
= (sb_flags
& SB_RDONLY
) && !sb_rdonly(sb
);
846 shrink_dcache_sb(sb
);
848 /* If we are remounting RDONLY and current sb is read/write,
849 make sure there are no rw files opened */
852 sb
->s_readonly_remount
= 1;
855 retval
= sb_prepare_remount_readonly(sb
);
861 if (sb
->s_op
->remount_fs
) {
862 retval
= sb
->s_op
->remount_fs(sb
, &sb_flags
, data
);
865 goto cancel_readonly
;
866 /* If forced remount, go ahead despite any errors */
867 WARN(1, "forced remount of a %s fs returned %i\n",
868 sb
->s_type
->name
, retval
);
871 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (sb_flags
& MS_RMT_MASK
);
872 /* Needs to be ordered wrt mnt_is_readonly() */
874 sb
->s_readonly_remount
= 0;
877 * Some filesystems modify their metadata via some other path than the
878 * bdev buffer cache (eg. use a private mapping, or directories in
879 * pagecache, etc). Also file data modifications go via their own
880 * mappings. So If we try to mount readonly then copy the filesystem
881 * from bdev, we could get stale data, so invalidate it to give a best
882 * effort at coherency.
884 if (remount_ro
&& sb
->s_bdev
)
885 invalidate_bdev(sb
->s_bdev
);
889 sb
->s_readonly_remount
= 0;
893 static void do_emergency_remount(struct work_struct
*work
)
895 struct super_block
*sb
, *p
= NULL
;
898 list_for_each_entry(sb
, &super_blocks
, s_list
) {
899 if (hlist_unhashed(&sb
->s_instances
))
902 spin_unlock(&sb_lock
);
903 down_write(&sb
->s_umount
);
904 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& SB_BORN
) &&
907 * What lock protects sb->s_flags??
909 do_remount_sb(sb
, SB_RDONLY
, NULL
, 1);
911 up_write(&sb
->s_umount
);
919 spin_unlock(&sb_lock
);
921 printk("Emergency Remount complete\n");
924 void emergency_remount(void)
926 struct work_struct
*work
;
928 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
930 INIT_WORK(work
, do_emergency_remount
);
936 * Unnamed block devices are dummy devices used by virtual
937 * filesystems which don't use real block-devices. -- jrs
940 static DEFINE_IDA(unnamed_dev_ida
);
941 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
942 /* Many userspace utilities consider an FSID of 0 invalid.
943 * Always return at least 1 from get_anon_bdev.
945 static int unnamed_dev_start
= 1;
947 int get_anon_bdev(dev_t
*p
)
953 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
955 spin_lock(&unnamed_dev_lock
);
956 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
958 unnamed_dev_start
= dev
+ 1;
959 spin_unlock(&unnamed_dev_lock
);
960 if (error
== -EAGAIN
)
961 /* We raced and lost with another CPU. */
966 if (dev
>= (1 << MINORBITS
)) {
967 spin_lock(&unnamed_dev_lock
);
968 ida_remove(&unnamed_dev_ida
, dev
);
969 if (unnamed_dev_start
> dev
)
970 unnamed_dev_start
= dev
;
971 spin_unlock(&unnamed_dev_lock
);
974 *p
= MKDEV(0, dev
& MINORMASK
);
977 EXPORT_SYMBOL(get_anon_bdev
);
979 void free_anon_bdev(dev_t dev
)
981 int slot
= MINOR(dev
);
982 spin_lock(&unnamed_dev_lock
);
983 ida_remove(&unnamed_dev_ida
, slot
);
984 if (slot
< unnamed_dev_start
)
985 unnamed_dev_start
= slot
;
986 spin_unlock(&unnamed_dev_lock
);
988 EXPORT_SYMBOL(free_anon_bdev
);
990 int set_anon_super(struct super_block
*s
, void *data
)
992 return get_anon_bdev(&s
->s_dev
);
995 EXPORT_SYMBOL(set_anon_super
);
997 void kill_anon_super(struct super_block
*sb
)
999 dev_t dev
= sb
->s_dev
;
1000 generic_shutdown_super(sb
);
1001 free_anon_bdev(dev
);
1004 EXPORT_SYMBOL(kill_anon_super
);
1006 void kill_litter_super(struct super_block
*sb
)
1009 d_genocide(sb
->s_root
);
1010 kill_anon_super(sb
);
1013 EXPORT_SYMBOL(kill_litter_super
);
1015 static int ns_test_super(struct super_block
*sb
, void *data
)
1017 return sb
->s_fs_info
== data
;
1020 static int ns_set_super(struct super_block
*sb
, void *data
)
1022 sb
->s_fs_info
= data
;
1023 return set_anon_super(sb
, NULL
);
1026 struct dentry
*mount_ns(struct file_system_type
*fs_type
,
1027 int flags
, void *data
, void *ns
, struct user_namespace
*user_ns
,
1028 int (*fill_super
)(struct super_block
*, void *, int))
1030 struct super_block
*sb
;
1032 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1033 * over the namespace.
1035 if (!(flags
& SB_KERNMOUNT
) && !ns_capable(user_ns
, CAP_SYS_ADMIN
))
1036 return ERR_PTR(-EPERM
);
1038 sb
= sget_userns(fs_type
, ns_test_super
, ns_set_super
, flags
,
1041 return ERR_CAST(sb
);
1045 err
= fill_super(sb
, data
, flags
& SB_SILENT
? 1 : 0);
1047 deactivate_locked_super(sb
);
1048 return ERR_PTR(err
);
1051 sb
->s_flags
|= SB_ACTIVE
;
1054 return dget(sb
->s_root
);
1057 EXPORT_SYMBOL(mount_ns
);
1060 static int set_bdev_super(struct super_block
*s
, void *data
)
1063 s
->s_dev
= s
->s_bdev
->bd_dev
;
1064 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_bdi
);
1069 static int test_bdev_super(struct super_block
*s
, void *data
)
1071 return (void *)s
->s_bdev
== data
;
1074 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
1075 int flags
, const char *dev_name
, void *data
,
1076 int (*fill_super
)(struct super_block
*, void *, int))
1078 struct block_device
*bdev
;
1079 struct super_block
*s
;
1080 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1083 if (!(flags
& SB_RDONLY
))
1084 mode
|= FMODE_WRITE
;
1086 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1088 return ERR_CAST(bdev
);
1090 if (current_user_ns() != &init_user_ns
) {
1092 * For userns mounts, disallow mounting if bdev is open for
1095 if (!atomic_dec_unless_positive(&bdev
->bd_inode
->i_writecount
)) {
1099 if (bdev
->bd_contains
!= bdev
&&
1100 !atomic_dec_unless_positive(&bdev
->bd_contains
->bd_inode
->i_writecount
)) {
1101 atomic_inc(&bdev
->bd_inode
->i_writecount
);
1108 * once the super is inserted into the list by sget, s_umount
1109 * will protect the lockfs code from trying to start a snapshot
1110 * while we are mounting
1112 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1113 if (bdev
->bd_fsfreeze_count
> 0) {
1114 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1118 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| SB_NOSEC
,
1120 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1125 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1126 deactivate_locked_super(s
);
1132 * s_umount nests inside bd_mutex during
1133 * __invalidate_device(). blkdev_put() acquires
1134 * bd_mutex and can't be called under s_umount. Drop
1135 * s_umount temporarily. This is safe as we're
1136 * holding an active reference.
1138 up_write(&s
->s_umount
);
1139 blkdev_put(bdev
, mode
);
1140 down_write(&s
->s_umount
);
1143 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1144 sb_set_blocksize(s
, block_size(bdev
));
1145 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1147 deactivate_locked_super(s
);
1151 s
->s_flags
|= SB_ACTIVE
;
1155 return dget(s
->s_root
);
1160 if (current_user_ns() != &init_user_ns
) {
1161 atomic_inc(&bdev
->bd_inode
->i_writecount
);
1162 if (bdev
->bd_contains
!= bdev
)
1163 atomic_inc(&bdev
->bd_contains
->bd_inode
->i_writecount
);
1166 blkdev_put(bdev
, mode
);
1168 return ERR_PTR(error
);
1170 EXPORT_SYMBOL(mount_bdev
);
1172 void kill_block_super(struct super_block
*sb
)
1174 struct block_device
*bdev
= sb
->s_bdev
;
1175 fmode_t mode
= sb
->s_mode
;
1177 bdev
->bd_super
= NULL
;
1178 generic_shutdown_super(sb
);
1179 sync_blockdev(bdev
);
1180 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1181 if (sb
->s_user_ns
!= &init_user_ns
) {
1182 atomic_inc(&bdev
->bd_inode
->i_writecount
);
1183 if (bdev
->bd_contains
!= bdev
)
1184 atomic_inc(&bdev
->bd_contains
->bd_inode
->i_writecount
);
1186 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1189 EXPORT_SYMBOL(kill_block_super
);
1192 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1193 int flags
, void *data
,
1194 int (*fill_super
)(struct super_block
*, void *, int))
1197 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1202 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1204 deactivate_locked_super(s
);
1205 return ERR_PTR(error
);
1207 s
->s_flags
|= SB_ACTIVE
;
1208 return dget(s
->s_root
);
1210 EXPORT_SYMBOL(mount_nodev
);
1212 static int compare_single(struct super_block
*s
, void *p
)
1217 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1218 int flags
, void *data
,
1219 int (*fill_super
)(struct super_block
*, void *, int))
1221 struct super_block
*s
;
1224 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1228 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1230 deactivate_locked_super(s
);
1231 return ERR_PTR(error
);
1233 s
->s_flags
|= SB_ACTIVE
;
1235 do_remount_sb(s
, flags
, data
, 0);
1237 return dget(s
->s_root
);
1239 EXPORT_SYMBOL(mount_single
);
1242 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1244 struct dentry
*root
;
1245 struct super_block
*sb
;
1246 char *secdata
= NULL
;
1247 int error
= -ENOMEM
;
1249 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1250 secdata
= alloc_secdata();
1254 error
= security_sb_copy_data(data
, secdata
);
1256 goto out_free_secdata
;
1259 root
= type
->mount(type
, flags
, name
, data
);
1261 error
= PTR_ERR(root
);
1262 goto out_free_secdata
;
1266 WARN_ON(!sb
->s_bdi
);
1269 * Write barrier is for super_cache_count(). We place it before setting
1270 * SB_BORN as the data dependency between the two functions is the
1271 * superblock structure contents that we just set up, not the SB_BORN
1275 sb
->s_flags
|= SB_BORN
;
1277 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1282 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1283 * but s_maxbytes was an unsigned long long for many releases. Throw
1284 * this warning for a little while to try and catch filesystems that
1285 * violate this rule.
1287 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1288 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1290 up_write(&sb
->s_umount
);
1291 free_secdata(secdata
);
1295 deactivate_locked_super(sb
);
1297 free_secdata(secdata
);
1299 return ERR_PTR(error
);
1303 * Setup private BDI for given superblock. It gets automatically cleaned up
1304 * in generic_shutdown_super().
1306 int super_setup_bdi_name(struct super_block
*sb
, char *fmt
, ...)
1308 struct backing_dev_info
*bdi
;
1312 bdi
= bdi_alloc(GFP_KERNEL
);
1316 bdi
->name
= sb
->s_type
->name
;
1318 va_start(args
, fmt
);
1319 err
= bdi_register_va(bdi
, fmt
, args
);
1325 WARN_ON(sb
->s_bdi
!= &noop_backing_dev_info
);
1330 EXPORT_SYMBOL(super_setup_bdi_name
);
1333 * Setup private BDI for given superblock. I gets automatically cleaned up
1334 * in generic_shutdown_super().
1336 int super_setup_bdi(struct super_block
*sb
)
1338 static atomic_long_t bdi_seq
= ATOMIC_LONG_INIT(0);
1340 return super_setup_bdi_name(sb
, "%.28s-%ld", sb
->s_type
->name
,
1341 atomic_long_inc_return(&bdi_seq
));
1343 EXPORT_SYMBOL(super_setup_bdi
);
1346 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1349 void __sb_end_write(struct super_block
*sb
, int level
)
1351 percpu_up_read(sb
->s_writers
.rw_sem
+ level
-1);
1353 EXPORT_SYMBOL(__sb_end_write
);
1356 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1359 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1361 bool force_trylock
= false;
1364 #ifdef CONFIG_LOCKDEP
1366 * We want lockdep to tell us about possible deadlocks with freezing
1367 * but it's it bit tricky to properly instrument it. Getting a freeze
1368 * protection works as getting a read lock but there are subtle
1369 * problems. XFS for example gets freeze protection on internal level
1370 * twice in some cases, which is OK only because we already hold a
1371 * freeze protection also on higher level. Due to these cases we have
1372 * to use wait == F (trylock mode) which must not fail.
1377 for (i
= 0; i
< level
- 1; i
++)
1378 if (percpu_rwsem_is_held(sb
->s_writers
.rw_sem
+ i
)) {
1379 force_trylock
= true;
1384 if (wait
&& !force_trylock
)
1385 percpu_down_read(sb
->s_writers
.rw_sem
+ level
-1);
1387 ret
= percpu_down_read_trylock(sb
->s_writers
.rw_sem
+ level
-1);
1389 WARN_ON(force_trylock
&& !ret
);
1392 EXPORT_SYMBOL(__sb_start_write
);
1395 * sb_wait_write - wait until all writers to given file system finish
1396 * @sb: the super for which we wait
1397 * @level: type of writers we wait for (normal vs page fault)
1399 * This function waits until there are no writers of given type to given file
1402 static void sb_wait_write(struct super_block
*sb
, int level
)
1404 percpu_down_write(sb
->s_writers
.rw_sem
+ level
-1);
1408 * We are going to return to userspace and forget about these locks, the
1409 * ownership goes to the caller of thaw_super() which does unlock().
1411 static void lockdep_sb_freeze_release(struct super_block
*sb
)
1415 for (level
= SB_FREEZE_LEVELS
- 1; level
>= 0; level
--)
1416 percpu_rwsem_release(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1420 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1422 static void lockdep_sb_freeze_acquire(struct super_block
*sb
)
1426 for (level
= 0; level
< SB_FREEZE_LEVELS
; ++level
)
1427 percpu_rwsem_acquire(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1430 static void sb_freeze_unlock(struct super_block
*sb
)
1434 for (level
= SB_FREEZE_LEVELS
- 1; level
>= 0; level
--)
1435 percpu_up_write(sb
->s_writers
.rw_sem
+ level
);
1439 * freeze_super - lock the filesystem and force it into a consistent state
1440 * @sb: the super to lock
1442 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1443 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1446 * During this function, sb->s_writers.frozen goes through these values:
1448 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1450 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1451 * writes should be blocked, though page faults are still allowed. We wait for
1452 * all writes to complete and then proceed to the next stage.
1454 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1455 * but internal fs threads can still modify the filesystem (although they
1456 * should not dirty new pages or inodes), writeback can run etc. After waiting
1457 * for all running page faults we sync the filesystem which will clean all
1458 * dirty pages and inodes (no new dirty pages or inodes can be created when
1461 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1462 * modification are blocked (e.g. XFS preallocation truncation on inode
1463 * reclaim). This is usually implemented by blocking new transactions for
1464 * filesystems that have them and need this additional guard. After all
1465 * internal writers are finished we call ->freeze_fs() to finish filesystem
1466 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1467 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1469 * sb->s_writers.frozen is protected by sb->s_umount.
1471 int freeze_super(struct super_block
*sb
)
1475 atomic_inc(&sb
->s_active
);
1476 down_write(&sb
->s_umount
);
1477 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1478 deactivate_locked_super(sb
);
1482 if (!(sb
->s_flags
& SB_BORN
)) {
1483 up_write(&sb
->s_umount
);
1484 return 0; /* sic - it's "nothing to do" */
1487 if (sb_rdonly(sb
)) {
1488 /* Nothing to do really... */
1489 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1490 up_write(&sb
->s_umount
);
1494 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1495 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1496 up_write(&sb
->s_umount
);
1497 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1498 down_write(&sb
->s_umount
);
1500 /* Now we go and block page faults... */
1501 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1502 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1504 /* All writers are done so after syncing there won't be dirty data */
1505 sync_filesystem(sb
);
1507 /* Now wait for internal filesystem counter */
1508 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1509 sb_wait_write(sb
, SB_FREEZE_FS
);
1511 if (sb
->s_op
->freeze_fs
) {
1512 ret
= sb
->s_op
->freeze_fs(sb
);
1515 "VFS:Filesystem freeze failed\n");
1516 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1517 sb_freeze_unlock(sb
);
1518 wake_up(&sb
->s_writers
.wait_unfrozen
);
1519 deactivate_locked_super(sb
);
1524 * For debugging purposes so that fs can warn if it sees write activity
1525 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1527 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1528 lockdep_sb_freeze_release(sb
);
1529 up_write(&sb
->s_umount
);
1532 EXPORT_SYMBOL(freeze_super
);
1535 * thaw_super -- unlock filesystem
1536 * @sb: the super to thaw
1538 * Unlocks the filesystem and marks it writeable again after freeze_super().
1540 int thaw_super(struct super_block
*sb
)
1544 down_write(&sb
->s_umount
);
1545 if (sb
->s_writers
.frozen
!= SB_FREEZE_COMPLETE
) {
1546 up_write(&sb
->s_umount
);
1550 if (sb_rdonly(sb
)) {
1551 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1555 lockdep_sb_freeze_acquire(sb
);
1557 if (sb
->s_op
->unfreeze_fs
) {
1558 error
= sb
->s_op
->unfreeze_fs(sb
);
1561 "VFS:Filesystem thaw failed\n");
1562 lockdep_sb_freeze_release(sb
);
1563 up_write(&sb
->s_umount
);
1568 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1569 sb_freeze_unlock(sb
);
1571 wake_up(&sb
->s_writers
.wait_unfrozen
);
1572 deactivate_locked_super(sb
);
1575 EXPORT_SYMBOL(thaw_super
);