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 <linux/file.h>
41 #include <asm/uaccess.h>
45 LIST_HEAD(super_blocks);
46 DEFINE_SPINLOCK(sb_lock);
49 * alloc_super - create new superblock
50 * @type: filesystem type superblock should belong to
52 * Allocates and initializes a new &struct super_block. alloc_super()
53 * returns a pointer new superblock or %NULL if allocation had failed.
55 static struct super_block *alloc_super(struct file_system_type *type)
57 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
58 static struct super_operations default_op;
61 if (security_sb_alloc(s)) {
66 INIT_LIST_HEAD(&s->s_dirty);
67 INIT_LIST_HEAD(&s->s_io);
68 INIT_LIST_HEAD(&s->s_more_io);
69 INIT_LIST_HEAD(&s->s_files);
70 INIT_LIST_HEAD(&s->s_instances);
71 INIT_HLIST_HEAD(&s->s_anon);
72 INIT_LIST_HEAD(&s->s_inodes);
73 INIT_LIST_HEAD(&s->s_dentry_lru);
74 init_rwsem(&s->s_umount);
75 mutex_init(&s->s_lock);
76 lockdep_set_class(&s->s_umount, &type->s_umount_key);
78 * The locking rules for s_lock are up to the
79 * filesystem. For example ext3fs has different
80 * lock ordering than usbfs:
82 lockdep_set_class(&s->s_lock, &type->s_lock_key);
84 * sget() can have s_umount recursion.
86 * When it cannot find a suitable sb, it allocates a new
87 * one (this one), and tries again to find a suitable old
90 * In case that succeeds, it will acquire the s_umount
91 * lock of the old one. Since these are clearly distrinct
92 * locks, and this object isn't exposed yet, there's no
95 * Annotate this by putting this lock in a different
98 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
100 atomic_set(&s->s_active, 1);
101 mutex_init(&s->s_vfs_rename_mutex);
102 mutex_init(&s->s_dquot.dqio_mutex);
103 mutex_init(&s->s_dquot.dqonoff_mutex);
104 init_rwsem(&s->s_dquot.dqptr_sem);
105 init_waitqueue_head(&s->s_wait_unfrozen);
106 s->s_maxbytes = MAX_NON_LFS;
107 s->dq_op = sb_dquot_ops;
108 s->s_qcop = sb_quotactl_ops;
109 s->s_op = &default_op;
110 s->s_time_gran = 1000000000;
117 * destroy_super - frees a superblock
118 * @s: superblock to free
120 * Frees a superblock.
122 static inline void destroy_super(struct super_block *s)
130 /* Superblock refcounting */
133 * Drop a superblock's refcount. Returns non-zero if the superblock was
134 * destroyed. The caller must hold sb_lock.
136 static int __put_super(struct super_block *sb)
140 if (!--sb->s_count) {
148 * Drop a superblock's refcount.
149 * Returns non-zero if the superblock is about to be destroyed and
150 * at least is already removed from super_blocks list, so if we are
151 * making a loop through super blocks then we need to restart.
152 * The caller must hold sb_lock.
154 int __put_super_and_need_restart(struct super_block *sb)
156 /* check for race with generic_shutdown_super() */
157 if (list_empty(&sb->s_list)) {
158 /* super block is removed, need to restart... */
162 /* can't be the last, since s_list is still in use */
164 BUG_ON(sb->s_count == 0);
169 * put_super - drop a temporary reference to superblock
170 * @sb: superblock in question
172 * Drops a temporary reference, frees superblock if there's no
175 static void put_super(struct super_block *sb)
179 spin_unlock(&sb_lock);
184 * deactivate_super - drop an active reference to superblock
185 * @s: superblock to deactivate
187 * Drops an active reference to superblock, acquiring a temprory one if
188 * there is no active references left. In that case we lock superblock,
189 * tell fs driver to shut it down and drop the temporary reference we
192 void deactivate_super(struct super_block *s)
194 struct file_system_type *fs = s->s_type;
195 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
196 s->s_count -= S_BIAS-1;
197 spin_unlock(&sb_lock);
199 down_write(&s->s_umount);
206 EXPORT_SYMBOL(deactivate_super);
209 * deactivate_locked_super - drop an active reference to superblock
210 * @s: superblock to deactivate
212 * Equivalent of up_write(&s->s_umount); deactivate_super(s);, except that
213 * it does not unlock it until it's all over. As the result, it's safe to
214 * use to dispose of new superblock on ->get_sb() failure exits - nobody
215 * will see the sucker until it's all over. Equivalent using up_write +
216 * deactivate_super is safe for that purpose only if superblock is either
217 * safe to use or has NULL ->s_root when we unlock.
219 void deactivate_locked_super(struct super_block *s)
221 struct file_system_type *fs = s->s_type;
222 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
223 s->s_count -= S_BIAS-1;
224 spin_unlock(&sb_lock);
230 up_write(&s->s_umount);
234 EXPORT_SYMBOL(deactivate_locked_super);
237 * grab_super - acquire an active reference
238 * @s: reference we are trying to make active
240 * Tries to acquire an active reference. grab_super() is used when we
241 * had just found a superblock in super_blocks or fs_type->fs_supers
242 * and want to turn it into a full-blown active reference. grab_super()
243 * is called with sb_lock held and drops it. Returns 1 in case of
244 * success, 0 if we had failed (superblock contents was already dead or
245 * dying when grab_super() had been called).
247 static int grab_super(struct super_block *s) __releases(sb_lock)
250 spin_unlock(&sb_lock);
251 down_write(&s->s_umount);
254 if (s->s_count > S_BIAS) {
255 atomic_inc(&s->s_active);
257 spin_unlock(&sb_lock);
260 spin_unlock(&sb_lock);
262 up_write(&s->s_umount);
269 * Superblock locking. We really ought to get rid of these two.
271 void lock_super(struct super_block * sb)
274 mutex_lock(&sb->s_lock);
277 void unlock_super(struct super_block * sb)
280 mutex_unlock(&sb->s_lock);
283 EXPORT_SYMBOL(lock_super);
284 EXPORT_SYMBOL(unlock_super);
287 * Do the filesystem syncing work. For simple filesystems sync_inodes_sb(sb, 0)
288 * just dirties buffers with inodes so we have to submit IO for these buffers
289 * via __sync_blockdev(). This also speeds up the wait == 1 case since in that
290 * case write_inode() functions do sync_dirty_buffer() and thus effectively
291 * write one block at a time.
293 static int __fsync_super(struct super_block *sb, int wait)
296 sync_inodes_sb(sb, wait);
298 if (sb->s_dirt && sb->s_op->write_super)
299 sb->s_op->write_super(sb);
301 if (sb->s_op->sync_fs)
302 sb->s_op->sync_fs(sb, wait);
303 return __sync_blockdev(sb->s_bdev, wait);
307 * Write out and wait upon all dirty data associated with this
308 * superblock. Filesystem data as well as the underlying block
309 * device. Takes the superblock lock.
311 int fsync_super(struct super_block *sb)
315 ret = __fsync_super(sb, 0);
318 return __fsync_super(sb, 1);
320 EXPORT_SYMBOL_GPL(fsync_super);
323 * generic_shutdown_super - common helper for ->kill_sb()
324 * @sb: superblock to kill
326 * generic_shutdown_super() does all fs-independent work on superblock
327 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
328 * that need destruction out of superblock, call generic_shutdown_super()
329 * and release aforementioned objects. Note: dentries and inodes _are_
330 * taken care of and do not need specific handling.
332 * Upon calling this function, the filesystem may no longer alter or
333 * rearrange the set of dentries belonging to this super_block, nor may it
334 * change the attachments of dentries to inodes.
336 void generic_shutdown_super(struct super_block *sb)
338 const struct super_operations *sop = sb->s_op;
342 shrink_dcache_for_umount(sb);
345 sb->s_flags &= ~MS_ACTIVE;
347 /* bad name - it should be evict_inodes() */
348 invalidate_inodes(sb);
351 if (sop->write_super && sb->s_dirt)
352 sop->write_super(sb);
356 /* Forget any remaining inodes */
357 if (invalidate_inodes(sb)) {
358 printk("VFS: Busy inodes after unmount of %s. "
359 "Self-destruct in 5 seconds. Have a nice day...\n",
367 /* should be initialized for __put_super_and_need_restart() */
368 list_del_init(&sb->s_list);
369 list_del(&sb->s_instances);
370 spin_unlock(&sb_lock);
371 up_write(&sb->s_umount);
374 EXPORT_SYMBOL(generic_shutdown_super);
377 * sget - find or create a superblock
378 * @type: filesystem type superblock should belong to
379 * @test: comparison callback
380 * @set: setup callback
381 * @data: argument to each of them
383 struct super_block *sget(struct file_system_type *type,
384 int (*test)(struct super_block *,void *),
385 int (*set)(struct super_block *,void *),
388 struct super_block *s = NULL;
389 struct super_block *old;
395 list_for_each_entry(old, &type->fs_supers, s_instances) {
396 if (!test(old, data))
398 if (!grab_super(old))
401 up_write(&s->s_umount);
408 spin_unlock(&sb_lock);
409 s = alloc_super(type);
411 return ERR_PTR(-ENOMEM);
417 spin_unlock(&sb_lock);
418 up_write(&s->s_umount);
423 strlcpy(s->s_id, type->name, sizeof(s->s_id));
424 list_add_tail(&s->s_list, &super_blocks);
425 list_add(&s->s_instances, &type->fs_supers);
426 spin_unlock(&sb_lock);
427 get_filesystem(type);
433 void drop_super(struct super_block *sb)
435 up_read(&sb->s_umount);
439 EXPORT_SYMBOL(drop_super);
441 static inline void write_super(struct super_block *sb)
444 if (sb->s_root && sb->s_dirt)
445 if (sb->s_op->write_super)
446 sb->s_op->write_super(sb);
451 * Note: check the dirty flag before waiting, so we don't
452 * hold up the sync while mounting a device. (The newly
453 * mounted device won't need syncing.)
455 void sync_supers(void)
457 struct super_block *sb;
461 list_for_each_entry(sb, &super_blocks, s_list) {
464 spin_unlock(&sb_lock);
465 down_read(&sb->s_umount);
467 up_read(&sb->s_umount);
469 if (__put_super_and_need_restart(sb))
473 spin_unlock(&sb_lock);
477 * Sync all the data for all the filesystems (called by sys_sync() and
480 * This operation is careful to avoid the livelock which could easily happen
481 * if two or more filesystems are being continuously dirtied. s_need_sync
482 * is used only here. We set it against all filesystems and then clear it as
483 * we sync them. So redirtied filesystems are skipped.
485 * But if process A is currently running sync_filesystems and then process B
486 * calls sync_filesystems as well, process B will set all the s_need_sync
487 * flags again, which will cause process A to resync everything. Fix that with
490 void sync_filesystems(int wait)
492 struct super_block *sb;
493 static DEFINE_MUTEX(mutex);
495 mutex_lock(&mutex); /* Could be down_interruptible */
497 list_for_each_entry(sb, &super_blocks, s_list) {
498 if (sb->s_flags & MS_RDONLY)
504 list_for_each_entry(sb, &super_blocks, s_list) {
505 if (!sb->s_need_sync)
508 if (sb->s_flags & MS_RDONLY)
509 continue; /* hm. Was remounted r/o meanwhile */
511 spin_unlock(&sb_lock);
512 down_read(&sb->s_umount);
514 __fsync_super(sb, wait);
515 up_read(&sb->s_umount);
516 /* restart only when sb is no longer on the list */
518 if (__put_super_and_need_restart(sb))
521 spin_unlock(&sb_lock);
522 mutex_unlock(&mutex);
526 * get_super - get the superblock of a device
527 * @bdev: device to get the superblock for
529 * Scans the superblock list and finds the superblock of the file system
530 * mounted on the device given. %NULL is returned if no match is found.
533 struct super_block * get_super(struct block_device *bdev)
535 struct super_block *sb;
542 list_for_each_entry(sb, &super_blocks, s_list) {
543 if (sb->s_bdev == bdev) {
545 spin_unlock(&sb_lock);
546 down_read(&sb->s_umount);
549 up_read(&sb->s_umount);
550 /* restart only when sb is no longer on the list */
552 if (__put_super_and_need_restart(sb))
556 spin_unlock(&sb_lock);
560 EXPORT_SYMBOL(get_super);
562 struct super_block * user_get_super(dev_t dev)
564 struct super_block *sb;
568 list_for_each_entry(sb, &super_blocks, s_list) {
569 if (sb->s_dev == dev) {
571 spin_unlock(&sb_lock);
572 down_read(&sb->s_umount);
575 up_read(&sb->s_umount);
576 /* restart only when sb is no longer on the list */
578 if (__put_super_and_need_restart(sb))
582 spin_unlock(&sb_lock);
586 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
588 struct super_block *s;
593 s = user_get_super(new_decode_dev(dev));
596 err = vfs_statfs(s->s_root, &sbuf);
601 memset(&tmp,0,sizeof(struct ustat));
602 tmp.f_tfree = sbuf.f_bfree;
603 tmp.f_tinode = sbuf.f_ffree;
605 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
611 * do_remount_sb - asks filesystem to change mount options.
612 * @sb: superblock in question
613 * @flags: numeric part of options
614 * @data: the rest of options
615 * @force: whether or not to force the change
617 * Alters the mount options of a mounted file system.
619 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
625 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
628 if (flags & MS_RDONLY)
630 shrink_dcache_sb(sb);
633 /* If we are remounting RDONLY and current sb is read/write,
634 make sure there are no rw files opened */
635 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
638 else if (!fs_may_remount_ro(sb))
640 retval = vfs_dq_off(sb, 1);
641 if (retval < 0 && retval != -ENOSYS)
644 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
646 if (sb->s_op->remount_fs) {
648 retval = sb->s_op->remount_fs(sb, &flags, data);
653 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
655 vfs_dq_quota_on_remount(sb);
659 static void do_emergency_remount(struct work_struct *work)
661 struct super_block *sb;
664 list_for_each_entry(sb, &super_blocks, s_list) {
666 spin_unlock(&sb_lock);
667 down_read(&sb->s_umount);
668 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
670 * ->remount_fs needs lock_kernel().
672 * What lock protects sb->s_flags??
675 do_remount_sb(sb, MS_RDONLY, NULL, 1);
681 spin_unlock(&sb_lock);
683 printk("Emergency Remount complete\n");
686 void emergency_remount(void)
688 struct work_struct *work;
690 work = kmalloc(sizeof(*work), GFP_ATOMIC);
692 INIT_WORK(work, do_emergency_remount);
698 * Unnamed block devices are dummy devices used by virtual
699 * filesystems which don't use real block-devices. -- jrs
702 static DEFINE_IDA(unnamed_dev_ida);
703 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
705 int set_anon_super(struct super_block *s, void *data)
711 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
713 spin_lock(&unnamed_dev_lock);
714 error = ida_get_new(&unnamed_dev_ida, &dev);
715 spin_unlock(&unnamed_dev_lock);
716 if (error == -EAGAIN)
717 /* We raced and lost with another CPU. */
722 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
723 spin_lock(&unnamed_dev_lock);
724 ida_remove(&unnamed_dev_ida, dev);
725 spin_unlock(&unnamed_dev_lock);
728 s->s_dev = MKDEV(0, dev & MINORMASK);
732 EXPORT_SYMBOL(set_anon_super);
734 void kill_anon_super(struct super_block *sb)
736 int slot = MINOR(sb->s_dev);
738 generic_shutdown_super(sb);
739 spin_lock(&unnamed_dev_lock);
740 ida_remove(&unnamed_dev_ida, slot);
741 spin_unlock(&unnamed_dev_lock);
744 EXPORT_SYMBOL(kill_anon_super);
746 void kill_litter_super(struct super_block *sb)
749 d_genocide(sb->s_root);
753 EXPORT_SYMBOL(kill_litter_super);
755 static int ns_test_super(struct super_block *sb, void *data)
757 return sb->s_fs_info == data;
760 static int ns_set_super(struct super_block *sb, void *data)
762 sb->s_fs_info = data;
763 return set_anon_super(sb, NULL);
766 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
767 int (*fill_super)(struct super_block *, void *, int),
768 struct vfsmount *mnt)
770 struct super_block *sb;
772 sb = sget(fs_type, ns_test_super, ns_set_super, data);
779 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
781 deactivate_locked_super(sb);
785 sb->s_flags |= MS_ACTIVE;
788 simple_set_mnt(mnt, sb);
792 EXPORT_SYMBOL(get_sb_ns);
795 static int set_bdev_super(struct super_block *s, void *data)
798 s->s_dev = s->s_bdev->bd_dev;
802 static int test_bdev_super(struct super_block *s, void *data)
804 return (void *)s->s_bdev == data;
807 int get_sb_bdev(struct file_system_type *fs_type,
808 int flags, const char *dev_name, void *data,
809 int (*fill_super)(struct super_block *, void *, int),
810 struct vfsmount *mnt)
812 struct block_device *bdev;
813 struct super_block *s;
814 fmode_t mode = FMODE_READ;
817 if (!(flags & MS_RDONLY))
820 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
822 return PTR_ERR(bdev);
825 * once the super is inserted into the list by sget, s_umount
826 * will protect the lockfs code from trying to start a snapshot
827 * while we are mounting
829 down(&bdev->bd_mount_sem);
830 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
831 up(&bdev->bd_mount_sem);
836 if ((flags ^ s->s_flags) & MS_RDONLY) {
837 deactivate_locked_super(s);
842 close_bdev_exclusive(bdev, mode);
844 char b[BDEVNAME_SIZE];
848 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
849 sb_set_blocksize(s, block_size(bdev));
850 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
852 deactivate_locked_super(s);
856 s->s_flags |= MS_ACTIVE;
860 simple_set_mnt(mnt, s);
866 close_bdev_exclusive(bdev, mode);
871 EXPORT_SYMBOL(get_sb_bdev);
873 void kill_block_super(struct super_block *sb)
875 struct block_device *bdev = sb->s_bdev;
876 fmode_t mode = sb->s_mode;
878 bdev->bd_super = NULL;
879 generic_shutdown_super(sb);
881 close_bdev_exclusive(bdev, mode);
884 EXPORT_SYMBOL(kill_block_super);
887 int get_sb_nodev(struct file_system_type *fs_type,
888 int flags, void *data,
889 int (*fill_super)(struct super_block *, void *, int),
890 struct vfsmount *mnt)
893 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
900 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
902 deactivate_locked_super(s);
905 s->s_flags |= MS_ACTIVE;
906 simple_set_mnt(mnt, s);
910 EXPORT_SYMBOL(get_sb_nodev);
912 static int compare_single(struct super_block *s, void *p)
917 int get_sb_single(struct file_system_type *fs_type,
918 int flags, void *data,
919 int (*fill_super)(struct super_block *, void *, int),
920 struct vfsmount *mnt)
922 struct super_block *s;
925 s = sget(fs_type, compare_single, set_anon_super, NULL);
930 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
932 deactivate_locked_super(s);
935 s->s_flags |= MS_ACTIVE;
937 do_remount_sb(s, flags, data, 0);
938 simple_set_mnt(mnt, s);
942 EXPORT_SYMBOL(get_sb_single);
945 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
947 struct vfsmount *mnt;
948 char *secdata = NULL;
952 return ERR_PTR(-ENODEV);
955 mnt = alloc_vfsmnt(name);
959 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
960 secdata = alloc_secdata();
964 error = security_sb_copy_data(data, secdata);
966 goto out_free_secdata;
969 error = type->get_sb(type, flags, name, data, mnt);
971 goto out_free_secdata;
972 BUG_ON(!mnt->mnt_sb);
974 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
978 mnt->mnt_mountpoint = mnt->mnt_root;
979 mnt->mnt_parent = mnt;
980 up_write(&mnt->mnt_sb->s_umount);
981 free_secdata(secdata);
985 deactivate_locked_super(mnt->mnt_sb);
987 free_secdata(secdata);
991 return ERR_PTR(error);
994 EXPORT_SYMBOL_GPL(vfs_kern_mount);
996 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
999 const char *subtype = strchr(fstype, '.');
1008 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1010 if (!mnt->mnt_sb->s_subtype)
1016 return ERR_PTR(err);
1020 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1022 struct file_system_type *type = get_fs_type(fstype);
1023 struct vfsmount *mnt;
1025 return ERR_PTR(-ENODEV);
1026 mnt = vfs_kern_mount(type, flags, name, data);
1027 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1028 !mnt->mnt_sb->s_subtype)
1029 mnt = fs_set_subtype(mnt, fstype);
1030 put_filesystem(type);
1033 EXPORT_SYMBOL_GPL(do_kern_mount);
1035 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1037 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1040 EXPORT_SYMBOL_GPL(kern_mount_data);