4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
8 * - filesystem drivers list
10 * - umount system call
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
23 #include <linux/module.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/smp_lock.h>
27 #include <linux/acct.h>
28 #include <linux/blkdev.h>
29 #include <linux/quotaops.h>
30 #include <linux/namei.h>
31 #include <linux/buffer_head.h> /* for fsync_super() */
32 #include <linux/mount.h>
33 #include <linux/security.h>
34 #include <linux/syscalls.h>
35 #include <linux/vfs.h>
36 #include <linux/writeback.h> /* for the emergency remount stuff */
37 #include <linux/idr.h>
38 #include <linux/kobject.h>
39 #include <linux/mutex.h>
40 #include <asm/uaccess.h>
43 LIST_HEAD(super_blocks
);
44 DEFINE_SPINLOCK(sb_lock
);
47 * alloc_super - create new superblock
48 * @type: filesystem type superblock should belong to
50 * Allocates and initializes a new &struct super_block. alloc_super()
51 * returns a pointer new superblock or %NULL if allocation had failed.
53 static struct super_block
*alloc_super(struct file_system_type
*type
)
55 struct super_block
*s
= kzalloc(sizeof(struct super_block
), GFP_USER
);
56 static struct super_operations default_op
;
59 if (security_sb_alloc(s
)) {
64 INIT_LIST_HEAD(&s
->s_dirty
);
65 INIT_LIST_HEAD(&s
->s_io
);
66 INIT_LIST_HEAD(&s
->s_more_io
);
67 INIT_LIST_HEAD(&s
->s_files
);
68 INIT_LIST_HEAD(&s
->s_instances
);
69 INIT_HLIST_HEAD(&s
->s_anon
);
70 INIT_LIST_HEAD(&s
->s_inodes
);
71 init_rwsem(&s
->s_umount
);
72 mutex_init(&s
->s_lock
);
73 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
75 * The locking rules for s_lock are up to the
76 * filesystem. For example ext3fs has different
77 * lock ordering than usbfs:
79 lockdep_set_class(&s
->s_lock
, &type
->s_lock_key
);
80 down_write(&s
->s_umount
);
82 atomic_set(&s
->s_active
, 1);
83 mutex_init(&s
->s_vfs_rename_mutex
);
84 mutex_init(&s
->s_dquot
.dqio_mutex
);
85 mutex_init(&s
->s_dquot
.dqonoff_mutex
);
86 init_rwsem(&s
->s_dquot
.dqptr_sem
);
87 init_waitqueue_head(&s
->s_wait_unfrozen
);
88 s
->s_maxbytes
= MAX_NON_LFS
;
89 s
->dq_op
= sb_dquot_ops
;
90 s
->s_qcop
= sb_quotactl_ops
;
91 s
->s_op
= &default_op
;
92 s
->s_time_gran
= 1000000000;
99 * destroy_super - frees a superblock
100 * @s: superblock to free
102 * Frees a superblock.
104 static inline void destroy_super(struct super_block
*s
)
112 /* Superblock refcounting */
115 * Drop a superblock's refcount. Returns non-zero if the superblock was
116 * destroyed. The caller must hold sb_lock.
118 int __put_super(struct super_block
*sb
)
122 if (!--sb
->s_count
) {
130 * Drop a superblock's refcount.
131 * Returns non-zero if the superblock is about to be destroyed and
132 * at least is already removed from super_blocks list, so if we are
133 * making a loop through super blocks then we need to restart.
134 * The caller must hold sb_lock.
136 int __put_super_and_need_restart(struct super_block
*sb
)
138 /* check for race with generic_shutdown_super() */
139 if (list_empty(&sb
->s_list
)) {
140 /* super block is removed, need to restart... */
144 /* can't be the last, since s_list is still in use */
146 BUG_ON(sb
->s_count
== 0);
151 * put_super - drop a temporary reference to superblock
152 * @sb: superblock in question
154 * Drops a temporary reference, frees superblock if there's no
157 static void put_super(struct super_block
*sb
)
161 spin_unlock(&sb_lock
);
166 * deactivate_super - drop an active reference to superblock
167 * @s: superblock to deactivate
169 * Drops an active reference to superblock, acquiring a temprory one if
170 * there is no active references left. In that case we lock superblock,
171 * tell fs driver to shut it down and drop the temporary reference we
174 void deactivate_super(struct super_block
*s
)
176 struct file_system_type
*fs
= s
->s_type
;
177 if (atomic_dec_and_lock(&s
->s_active
, &sb_lock
)) {
178 s
->s_count
-= S_BIAS
-1;
179 spin_unlock(&sb_lock
);
181 down_write(&s
->s_umount
);
188 EXPORT_SYMBOL(deactivate_super
);
191 * grab_super - acquire an active reference
192 * @s: reference we are trying to make active
194 * Tries to acquire an active reference. grab_super() is used when we
195 * had just found a superblock in super_blocks or fs_type->fs_supers
196 * and want to turn it into a full-blown active reference. grab_super()
197 * is called with sb_lock held and drops it. Returns 1 in case of
198 * success, 0 if we had failed (superblock contents was already dead or
199 * dying when grab_super() had been called).
201 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
204 spin_unlock(&sb_lock
);
205 down_write(&s
->s_umount
);
208 if (s
->s_count
> S_BIAS
) {
209 atomic_inc(&s
->s_active
);
211 spin_unlock(&sb_lock
);
214 spin_unlock(&sb_lock
);
216 up_write(&s
->s_umount
);
223 * Superblock locking. We really ought to get rid of these two.
225 void lock_super(struct super_block
* sb
)
228 mutex_lock(&sb
->s_lock
);
231 void unlock_super(struct super_block
* sb
)
234 mutex_unlock(&sb
->s_lock
);
237 EXPORT_SYMBOL(lock_super
);
238 EXPORT_SYMBOL(unlock_super
);
241 * Write out and wait upon all dirty data associated with this
242 * superblock. Filesystem data as well as the underlying block
243 * device. Takes the superblock lock. Requires a second blkdev
244 * flush by the caller to complete the operation.
246 void __fsync_super(struct super_block
*sb
)
248 sync_inodes_sb(sb
, 0);
251 if (sb
->s_dirt
&& sb
->s_op
->write_super
)
252 sb
->s_op
->write_super(sb
);
254 if (sb
->s_op
->sync_fs
)
255 sb
->s_op
->sync_fs(sb
, 1);
256 sync_blockdev(sb
->s_bdev
);
257 sync_inodes_sb(sb
, 1);
261 * Write out and wait upon all dirty data associated with this
262 * superblock. Filesystem data as well as the underlying block
263 * device. Takes the superblock lock.
265 int fsync_super(struct super_block
*sb
)
268 return sync_blockdev(sb
->s_bdev
);
272 * generic_shutdown_super - common helper for ->kill_sb()
273 * @sb: superblock to kill
275 * generic_shutdown_super() does all fs-independent work on superblock
276 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
277 * that need destruction out of superblock, call generic_shutdown_super()
278 * and release aforementioned objects. Note: dentries and inodes _are_
279 * taken care of and do not need specific handling.
281 * Upon calling this function, the filesystem may no longer alter or
282 * rearrange the set of dentries belonging to this super_block, nor may it
283 * change the attachments of dentries to inodes.
285 void generic_shutdown_super(struct super_block
*sb
)
287 const struct super_operations
*sop
= sb
->s_op
;
290 shrink_dcache_for_umount(sb
);
293 sb
->s_flags
&= ~MS_ACTIVE
;
294 /* bad name - it should be evict_inodes() */
295 invalidate_inodes(sb
);
298 if (sop
->write_super
&& sb
->s_dirt
)
299 sop
->write_super(sb
);
303 /* Forget any remaining inodes */
304 if (invalidate_inodes(sb
)) {
305 printk("VFS: Busy inodes after unmount of %s. "
306 "Self-destruct in 5 seconds. Have a nice day...\n",
314 /* should be initialized for __put_super_and_need_restart() */
315 list_del_init(&sb
->s_list
);
316 list_del(&sb
->s_instances
);
317 spin_unlock(&sb_lock
);
318 up_write(&sb
->s_umount
);
321 EXPORT_SYMBOL(generic_shutdown_super
);
324 * sget - find or create a superblock
325 * @type: filesystem type superblock should belong to
326 * @test: comparison callback
327 * @set: setup callback
328 * @data: argument to each of them
330 struct super_block
*sget(struct file_system_type
*type
,
331 int (*test
)(struct super_block
*,void *),
332 int (*set
)(struct super_block
*,void *),
335 struct super_block
*s
= NULL
;
336 struct super_block
*old
;
342 list_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
343 if (!test(old
, data
))
345 if (!grab_super(old
))
353 spin_unlock(&sb_lock
);
354 s
= alloc_super(type
);
356 return ERR_PTR(-ENOMEM
);
362 spin_unlock(&sb_lock
);
367 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
368 list_add_tail(&s
->s_list
, &super_blocks
);
369 list_add(&s
->s_instances
, &type
->fs_supers
);
370 spin_unlock(&sb_lock
);
371 get_filesystem(type
);
377 void drop_super(struct super_block
*sb
)
379 up_read(&sb
->s_umount
);
383 EXPORT_SYMBOL(drop_super
);
385 static inline void write_super(struct super_block
*sb
)
388 if (sb
->s_root
&& sb
->s_dirt
)
389 if (sb
->s_op
->write_super
)
390 sb
->s_op
->write_super(sb
);
395 * Note: check the dirty flag before waiting, so we don't
396 * hold up the sync while mounting a device. (The newly
397 * mounted device won't need syncing.)
399 void sync_supers(void)
401 struct super_block
*sb
;
405 list_for_each_entry(sb
, &super_blocks
, s_list
) {
408 spin_unlock(&sb_lock
);
409 down_read(&sb
->s_umount
);
411 up_read(&sb
->s_umount
);
413 if (__put_super_and_need_restart(sb
))
417 spin_unlock(&sb_lock
);
421 * Call the ->sync_fs super_op against all filesystems which are r/w and
422 * which implement it.
424 * This operation is careful to avoid the livelock which could easily happen
425 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
426 * is used only here. We set it against all filesystems and then clear it as
427 * we sync them. So redirtied filesystems are skipped.
429 * But if process A is currently running sync_filesystems and then process B
430 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
431 * flags again, which will cause process A to resync everything. Fix that with
434 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
436 void sync_filesystems(int wait
)
438 struct super_block
*sb
;
439 static DEFINE_MUTEX(mutex
);
441 mutex_lock(&mutex
); /* Could be down_interruptible */
443 list_for_each_entry(sb
, &super_blocks
, s_list
) {
444 if (!sb
->s_op
->sync_fs
)
446 if (sb
->s_flags
& MS_RDONLY
)
448 sb
->s_need_sync_fs
= 1;
452 list_for_each_entry(sb
, &super_blocks
, s_list
) {
453 if (!sb
->s_need_sync_fs
)
455 sb
->s_need_sync_fs
= 0;
456 if (sb
->s_flags
& MS_RDONLY
)
457 continue; /* hm. Was remounted r/o meanwhile */
459 spin_unlock(&sb_lock
);
460 down_read(&sb
->s_umount
);
461 if (sb
->s_root
&& (wait
|| sb
->s_dirt
))
462 sb
->s_op
->sync_fs(sb
, wait
);
463 up_read(&sb
->s_umount
);
464 /* restart only when sb is no longer on the list */
466 if (__put_super_and_need_restart(sb
))
469 spin_unlock(&sb_lock
);
470 mutex_unlock(&mutex
);
474 * get_super - get the superblock of a device
475 * @bdev: device to get the superblock for
477 * Scans the superblock list and finds the superblock of the file system
478 * mounted on the device given. %NULL is returned if no match is found.
481 struct super_block
* get_super(struct block_device
*bdev
)
483 struct super_block
*sb
;
490 list_for_each_entry(sb
, &super_blocks
, s_list
) {
491 if (sb
->s_bdev
== bdev
) {
493 spin_unlock(&sb_lock
);
494 down_read(&sb
->s_umount
);
497 up_read(&sb
->s_umount
);
498 /* restart only when sb is no longer on the list */
500 if (__put_super_and_need_restart(sb
))
504 spin_unlock(&sb_lock
);
508 EXPORT_SYMBOL(get_super
);
510 struct super_block
* user_get_super(dev_t dev
)
512 struct super_block
*sb
;
516 list_for_each_entry(sb
, &super_blocks
, s_list
) {
517 if (sb
->s_dev
== dev
) {
519 spin_unlock(&sb_lock
);
520 down_read(&sb
->s_umount
);
523 up_read(&sb
->s_umount
);
524 /* restart only when sb is no longer on the list */
526 if (__put_super_and_need_restart(sb
))
530 spin_unlock(&sb_lock
);
534 asmlinkage
long sys_ustat(unsigned dev
, struct ustat __user
* ubuf
)
536 struct super_block
*s
;
541 s
= user_get_super(new_decode_dev(dev
));
544 err
= vfs_statfs(s
->s_root
, &sbuf
);
549 memset(&tmp
,0,sizeof(struct ustat
));
550 tmp
.f_tfree
= sbuf
.f_bfree
;
551 tmp
.f_tinode
= sbuf
.f_ffree
;
553 err
= copy_to_user(ubuf
,&tmp
,sizeof(struct ustat
)) ? -EFAULT
: 0;
559 * mark_files_ro - mark all files read-only
560 * @sb: superblock in question
562 * All files are marked read-only. We don't care about pending
563 * delete files so this should be used in 'force' mode only.
566 static void mark_files_ro(struct super_block
*sb
)
571 list_for_each_entry(f
, &sb
->s_files
, f_u
.fu_list
) {
572 if (S_ISREG(f
->f_path
.dentry
->d_inode
->i_mode
) && file_count(f
))
573 f
->f_mode
&= ~FMODE_WRITE
;
579 * do_remount_sb - asks filesystem to change mount options.
580 * @sb: superblock in question
581 * @flags: numeric part of options
582 * @data: the rest of options
583 * @force: whether or not to force the change
585 * Alters the mount options of a mounted file system.
587 int do_remount_sb(struct super_block
*sb
, int flags
, void *data
, int force
)
592 if (!(flags
& MS_RDONLY
) && bdev_read_only(sb
->s_bdev
))
595 if (flags
& MS_RDONLY
)
597 shrink_dcache_sb(sb
);
600 /* If we are remounting RDONLY and current sb is read/write,
601 make sure there are no rw files opened */
602 if ((flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
)) {
605 else if (!fs_may_remount_ro(sb
))
610 if (sb
->s_op
->remount_fs
) {
612 retval
= sb
->s_op
->remount_fs(sb
, &flags
, data
);
617 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (flags
& MS_RMT_MASK
);
621 static void do_emergency_remount(unsigned long foo
)
623 struct super_block
*sb
;
626 list_for_each_entry(sb
, &super_blocks
, s_list
) {
628 spin_unlock(&sb_lock
);
629 down_read(&sb
->s_umount
);
630 if (sb
->s_root
&& sb
->s_bdev
&& !(sb
->s_flags
& MS_RDONLY
)) {
632 * ->remount_fs needs lock_kernel().
634 * What lock protects sb->s_flags??
637 do_remount_sb(sb
, MS_RDONLY
, NULL
, 1);
643 spin_unlock(&sb_lock
);
644 printk("Emergency Remount complete\n");
647 void emergency_remount(void)
649 pdflush_operation(do_emergency_remount
, 0);
653 * Unnamed block devices are dummy devices used by virtual
654 * filesystems which don't use real block-devices. -- jrs
657 static struct idr unnamed_dev_idr
;
658 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
660 int set_anon_super(struct super_block
*s
, void *data
)
666 if (idr_pre_get(&unnamed_dev_idr
, GFP_ATOMIC
) == 0)
668 spin_lock(&unnamed_dev_lock
);
669 error
= idr_get_new(&unnamed_dev_idr
, NULL
, &dev
);
670 spin_unlock(&unnamed_dev_lock
);
671 if (error
== -EAGAIN
)
672 /* We raced and lost with another CPU. */
677 if ((dev
& MAX_ID_MASK
) == (1 << MINORBITS
)) {
678 spin_lock(&unnamed_dev_lock
);
679 idr_remove(&unnamed_dev_idr
, dev
);
680 spin_unlock(&unnamed_dev_lock
);
683 s
->s_dev
= MKDEV(0, dev
& MINORMASK
);
687 EXPORT_SYMBOL(set_anon_super
);
689 void kill_anon_super(struct super_block
*sb
)
691 int slot
= MINOR(sb
->s_dev
);
693 generic_shutdown_super(sb
);
694 spin_lock(&unnamed_dev_lock
);
695 idr_remove(&unnamed_dev_idr
, slot
);
696 spin_unlock(&unnamed_dev_lock
);
699 EXPORT_SYMBOL(kill_anon_super
);
701 void __init
unnamed_dev_init(void)
703 idr_init(&unnamed_dev_idr
);
706 void kill_litter_super(struct super_block
*sb
)
709 d_genocide(sb
->s_root
);
713 EXPORT_SYMBOL(kill_litter_super
);
716 static int set_bdev_super(struct super_block
*s
, void *data
)
719 s
->s_dev
= s
->s_bdev
->bd_dev
;
723 static int test_bdev_super(struct super_block
*s
, void *data
)
725 return (void *)s
->s_bdev
== data
;
728 int get_sb_bdev(struct file_system_type
*fs_type
,
729 int flags
, const char *dev_name
, void *data
,
730 int (*fill_super
)(struct super_block
*, void *, int),
731 struct vfsmount
*mnt
)
733 struct block_device
*bdev
;
734 struct super_block
*s
;
737 bdev
= open_bdev_excl(dev_name
, flags
, fs_type
);
739 return PTR_ERR(bdev
);
742 * once the super is inserted into the list by sget, s_umount
743 * will protect the lockfs code from trying to start a snapshot
744 * while we are mounting
746 down(&bdev
->bd_mount_sem
);
747 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, bdev
);
748 up(&bdev
->bd_mount_sem
);
753 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
754 up_write(&s
->s_umount
);
760 close_bdev_excl(bdev
);
762 char b
[BDEVNAME_SIZE
];
765 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
766 sb_set_blocksize(s
, block_size(bdev
));
767 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
769 up_write(&s
->s_umount
);
774 s
->s_flags
|= MS_ACTIVE
;
777 return simple_set_mnt(mnt
, s
);
782 close_bdev_excl(bdev
);
787 EXPORT_SYMBOL(get_sb_bdev
);
789 void kill_block_super(struct super_block
*sb
)
791 struct block_device
*bdev
= sb
->s_bdev
;
793 generic_shutdown_super(sb
);
795 close_bdev_excl(bdev
);
798 EXPORT_SYMBOL(kill_block_super
);
801 int get_sb_nodev(struct file_system_type
*fs_type
,
802 int flags
, void *data
,
803 int (*fill_super
)(struct super_block
*, void *, int),
804 struct vfsmount
*mnt
)
807 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, NULL
);
814 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
816 up_write(&s
->s_umount
);
820 s
->s_flags
|= MS_ACTIVE
;
821 return simple_set_mnt(mnt
, s
);
824 EXPORT_SYMBOL(get_sb_nodev
);
826 static int compare_single(struct super_block
*s
, void *p
)
831 int get_sb_single(struct file_system_type
*fs_type
,
832 int flags
, void *data
,
833 int (*fill_super
)(struct super_block
*, void *, int),
834 struct vfsmount
*mnt
)
836 struct super_block
*s
;
839 s
= sget(fs_type
, compare_single
, set_anon_super
, NULL
);
844 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
846 up_write(&s
->s_umount
);
850 s
->s_flags
|= MS_ACTIVE
;
852 do_remount_sb(s
, flags
, data
, 0);
853 return simple_set_mnt(mnt
, s
);
856 EXPORT_SYMBOL(get_sb_single
);
859 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
861 struct vfsmount
*mnt
;
862 char *secdata
= NULL
;
866 return ERR_PTR(-ENODEV
);
869 mnt
= alloc_vfsmnt(name
);
873 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
874 secdata
= alloc_secdata();
878 error
= security_sb_copy_data(data
, secdata
);
880 goto out_free_secdata
;
883 error
= type
->get_sb(type
, flags
, name
, data
, mnt
);
885 goto out_free_secdata
;
886 BUG_ON(!mnt
->mnt_sb
);
888 error
= security_sb_kern_mount(mnt
->mnt_sb
, secdata
);
892 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
893 mnt
->mnt_parent
= mnt
;
894 up_write(&mnt
->mnt_sb
->s_umount
);
895 free_secdata(secdata
);
899 up_write(&mnt
->mnt_sb
->s_umount
);
900 deactivate_super(mnt
->mnt_sb
);
902 free_secdata(secdata
);
906 return ERR_PTR(error
);
909 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
911 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
914 const char *subtype
= strchr(fstype
, '.');
923 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
925 if (!mnt
->mnt_sb
->s_subtype
)
935 do_kern_mount(const char *fstype
, int flags
, const char *name
, void *data
)
937 struct file_system_type
*type
= get_fs_type(fstype
);
938 struct vfsmount
*mnt
;
940 return ERR_PTR(-ENODEV
);
941 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
942 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
943 !mnt
->mnt_sb
->s_subtype
)
944 mnt
= fs_set_subtype(mnt
, fstype
);
945 put_filesystem(type
);
948 EXPORT_SYMBOL_GPL(do_kern_mount
);
950 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
952 return vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
955 EXPORT_SYMBOL_GPL(kern_mount_data
);