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/mount.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/vfs.h>
35 #include <linux/writeback.h> /* for the emergency remount stuff */
36 #include <linux/idr.h>
37 #include <linux/kobject.h>
38 #include <linux/mutex.h>
39 #include <linux/file.h>
40 #include <linux/backing-dev.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 const struct super_operations default_op;
61 if (security_sb_alloc(s)) {
66 INIT_LIST_HEAD(&s->s_files);
67 INIT_LIST_HEAD(&s->s_instances);
68 INIT_HLIST_HEAD(&s->s_anon);
69 INIT_LIST_HEAD(&s->s_inodes);
70 INIT_LIST_HEAD(&s->s_dentry_lru);
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);
81 * sget() can have s_umount recursion.
83 * When it cannot find a suitable sb, it allocates a new
84 * one (this one), and tries again to find a suitable old
87 * In case that succeeds, it will acquire the s_umount
88 * lock of the old one. Since these are clearly distrinct
89 * locks, and this object isn't exposed yet, there's no
92 * Annotate this by putting this lock in a different
95 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
97 atomic_set(&s->s_active, 1);
98 mutex_init(&s->s_vfs_rename_mutex);
99 mutex_init(&s->s_dquot.dqio_mutex);
100 mutex_init(&s->s_dquot.dqonoff_mutex);
101 init_rwsem(&s->s_dquot.dqptr_sem);
102 init_waitqueue_head(&s->s_wait_unfrozen);
103 s->s_maxbytes = MAX_NON_LFS;
104 s->dq_op = sb_dquot_ops;
105 s->s_qcop = sb_quotactl_ops;
106 s->s_op = &default_op;
107 s->s_time_gran = 1000000000;
114 * destroy_super - frees a superblock
115 * @s: superblock to free
117 * Frees a superblock.
119 static inline void destroy_super(struct super_block *s)
127 /* Superblock refcounting */
130 * Drop a superblock's refcount. Returns non-zero if the superblock was
131 * destroyed. The caller must hold sb_lock.
133 static int __put_super(struct super_block *sb)
137 if (!--sb->s_count) {
145 * Drop a superblock's refcount.
146 * Returns non-zero if the superblock is about to be destroyed and
147 * at least is already removed from super_blocks list, so if we are
148 * making a loop through super blocks then we need to restart.
149 * The caller must hold sb_lock.
151 int __put_super_and_need_restart(struct super_block *sb)
153 /* check for race with generic_shutdown_super() */
154 if (list_empty(&sb->s_list)) {
155 /* super block is removed, need to restart... */
159 /* can't be the last, since s_list is still in use */
161 BUG_ON(sb->s_count == 0);
166 * put_super - drop a temporary reference to superblock
167 * @sb: superblock in question
169 * Drops a temporary reference, frees superblock if there's no
172 void put_super(struct super_block *sb)
176 spin_unlock(&sb_lock);
181 * deactivate_super - drop an active reference to superblock
182 * @s: superblock to deactivate
184 * Drops an active reference to superblock, acquiring a temprory one if
185 * there is no active references left. In that case we lock superblock,
186 * tell fs driver to shut it down and drop the temporary reference we
189 void deactivate_super(struct super_block *s)
191 struct file_system_type *fs = s->s_type;
192 if (atomic_dec_and_test(&s->s_active)) {
194 down_write(&s->s_umount);
201 EXPORT_SYMBOL(deactivate_super);
204 * deactivate_locked_super - drop an active reference to superblock
205 * @s: superblock to deactivate
207 * Equivalent of up_write(&s->s_umount); deactivate_super(s);, except that
208 * it does not unlock it until it's all over. As the result, it's safe to
209 * use to dispose of new superblock on ->get_sb() failure exits - nobody
210 * will see the sucker until it's all over. Equivalent using up_write +
211 * deactivate_super is safe for that purpose only if superblock is either
212 * safe to use or has NULL ->s_root when we unlock.
214 void deactivate_locked_super(struct super_block *s)
216 struct file_system_type *fs = s->s_type;
217 if (atomic_dec_and_test(&s->s_active)) {
223 up_write(&s->s_umount);
227 EXPORT_SYMBOL(deactivate_locked_super);
230 * grab_super - acquire an active reference
231 * @s: reference we are trying to make active
233 * Tries to acquire an active reference. grab_super() is used when we
234 * had just found a superblock in super_blocks or fs_type->fs_supers
235 * and want to turn it into a full-blown active reference. grab_super()
236 * is called with sb_lock held and drops it. Returns 1 in case of
237 * success, 0 if we had failed (superblock contents was already dead or
238 * dying when grab_super() had been called).
240 static int grab_super(struct super_block *s) __releases(sb_lock)
242 if (atomic_inc_not_zero(&s->s_active)) {
243 spin_unlock(&sb_lock);
244 down_write(&s->s_umount);
247 /* it's going away */
249 spin_unlock(&sb_lock);
250 /* usually that'll be enough for it to die... */
251 down_write(&s->s_umount);
252 up_write(&s->s_umount);
254 /* ... but in case it wasn't, let's at least yield() */
260 * Superblock locking. We really ought to get rid of these two.
262 void lock_super(struct super_block * sb)
265 mutex_lock(&sb->s_lock);
268 void unlock_super(struct super_block * sb)
271 mutex_unlock(&sb->s_lock);
274 EXPORT_SYMBOL(lock_super);
275 EXPORT_SYMBOL(unlock_super);
278 * generic_shutdown_super - common helper for ->kill_sb()
279 * @sb: superblock to kill
281 * generic_shutdown_super() does all fs-independent work on superblock
282 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
283 * that need destruction out of superblock, call generic_shutdown_super()
284 * and release aforementioned objects. Note: dentries and inodes _are_
285 * taken care of and do not need specific handling.
287 * Upon calling this function, the filesystem may no longer alter or
288 * rearrange the set of dentries belonging to this super_block, nor may it
289 * change the attachments of dentries to inodes.
291 void generic_shutdown_super(struct super_block *sb)
293 const struct super_operations *sop = sb->s_op;
297 shrink_dcache_for_umount(sb);
300 sb->s_flags &= ~MS_ACTIVE;
302 /* bad name - it should be evict_inodes() */
303 invalidate_inodes(sb);
308 /* Forget any remaining inodes */
309 if (invalidate_inodes(sb)) {
310 printk("VFS: Busy inodes after unmount of %s. "
311 "Self-destruct in 5 seconds. Have a nice day...\n",
317 /* should be initialized for __put_super_and_need_restart() */
318 list_del_init(&sb->s_list);
319 list_del(&sb->s_instances);
320 spin_unlock(&sb_lock);
321 up_write(&sb->s_umount);
324 EXPORT_SYMBOL(generic_shutdown_super);
327 * sget - find or create a superblock
328 * @type: filesystem type superblock should belong to
329 * @test: comparison callback
330 * @set: setup callback
331 * @data: argument to each of them
333 struct super_block *sget(struct file_system_type *type,
334 int (*test)(struct super_block *,void *),
335 int (*set)(struct super_block *,void *),
338 struct super_block *s = NULL;
339 struct super_block *old;
345 list_for_each_entry(old, &type->fs_supers, s_instances) {
346 if (!test(old, data))
348 if (!grab_super(old))
351 up_write(&s->s_umount);
358 spin_unlock(&sb_lock);
359 s = alloc_super(type);
361 return ERR_PTR(-ENOMEM);
367 spin_unlock(&sb_lock);
368 up_write(&s->s_umount);
373 strlcpy(s->s_id, type->name, sizeof(s->s_id));
374 list_add_tail(&s->s_list, &super_blocks);
375 list_add(&s->s_instances, &type->fs_supers);
376 spin_unlock(&sb_lock);
377 get_filesystem(type);
383 void drop_super(struct super_block *sb)
385 up_read(&sb->s_umount);
389 EXPORT_SYMBOL(drop_super);
392 * sync_supers - helper for periodic superblock writeback
394 * Call the write_super method if present on all dirty superblocks in
395 * the system. This is for the periodic writeback used by most older
396 * filesystems. For data integrity superblock writeback use
397 * sync_filesystems() instead.
399 * Note: check the dirty flag before waiting, so we don't
400 * hold up the sync while mounting a device. (The newly
401 * mounted device won't need syncing.)
403 void sync_supers(void)
405 struct super_block *sb;
409 list_for_each_entry(sb, &super_blocks, s_list) {
410 if (sb->s_op->write_super && sb->s_dirt) {
412 spin_unlock(&sb_lock);
414 down_read(&sb->s_umount);
415 if (sb->s_root && sb->s_dirt)
416 sb->s_op->write_super(sb);
417 up_read(&sb->s_umount);
420 if (__put_super_and_need_restart(sb))
424 spin_unlock(&sb_lock);
428 * get_super - get the superblock of a device
429 * @bdev: device to get the superblock for
431 * Scans the superblock list and finds the superblock of the file system
432 * mounted on the device given. %NULL is returned if no match is found.
435 struct super_block * get_super(struct block_device *bdev)
437 struct super_block *sb;
444 list_for_each_entry(sb, &super_blocks, s_list) {
445 if (sb->s_bdev == bdev) {
447 spin_unlock(&sb_lock);
448 down_read(&sb->s_umount);
451 up_read(&sb->s_umount);
452 /* restart only when sb is no longer on the list */
454 if (__put_super_and_need_restart(sb))
458 spin_unlock(&sb_lock);
462 EXPORT_SYMBOL(get_super);
465 * get_active_super - get an active reference to the superblock of a device
466 * @bdev: device to get the superblock for
468 * Scans the superblock list and finds the superblock of the file system
469 * mounted on the device given. Returns the superblock with an active
470 * reference and s_umount held exclusively or %NULL if none was found.
472 struct super_block *get_active_super(struct block_device *bdev)
474 struct super_block *sb;
480 list_for_each_entry(sb, &super_blocks, s_list) {
481 if (sb->s_bdev != bdev)
484 if (grab_super(sb)) /* drops sb_lock */
489 spin_unlock(&sb_lock);
493 struct super_block * user_get_super(dev_t dev)
495 struct super_block *sb;
499 list_for_each_entry(sb, &super_blocks, s_list) {
500 if (sb->s_dev == dev) {
502 spin_unlock(&sb_lock);
503 down_read(&sb->s_umount);
506 up_read(&sb->s_umount);
507 /* restart only when sb is no longer on the list */
509 if (__put_super_and_need_restart(sb))
513 spin_unlock(&sb_lock);
517 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
519 struct super_block *s;
524 s = user_get_super(new_decode_dev(dev));
527 err = vfs_statfs(s->s_root, &sbuf);
532 memset(&tmp,0,sizeof(struct ustat));
533 tmp.f_tfree = sbuf.f_bfree;
534 tmp.f_tinode = sbuf.f_ffree;
536 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
542 * do_remount_sb - asks filesystem to change mount options.
543 * @sb: superblock in question
544 * @flags: numeric part of options
545 * @data: the rest of options
546 * @force: whether or not to force the change
548 * Alters the mount options of a mounted file system.
550 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
553 int remount_rw, remount_ro;
555 if (sb->s_frozen != SB_UNFROZEN)
559 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
563 if (flags & MS_RDONLY)
565 shrink_dcache_sb(sb);
568 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
569 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
571 /* If we are remounting RDONLY and current sb is read/write,
572 make sure there are no rw files opened */
576 else if (!fs_may_remount_ro(sb))
578 retval = vfs_dq_off(sb, 1);
579 if (retval < 0 && retval != -ENOSYS)
583 if (sb->s_op->remount_fs) {
584 retval = sb->s_op->remount_fs(sb, &flags, data);
588 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
590 vfs_dq_quota_on_remount(sb);
592 * Some filesystems modify their metadata via some other path than the
593 * bdev buffer cache (eg. use a private mapping, or directories in
594 * pagecache, etc). Also file data modifications go via their own
595 * mappings. So If we try to mount readonly then copy the filesystem
596 * from bdev, we could get stale data, so invalidate it to give a best
597 * effort at coherency.
599 if (remount_ro && sb->s_bdev)
600 invalidate_bdev(sb->s_bdev);
604 static void do_emergency_remount(struct work_struct *work)
606 struct super_block *sb;
609 list_for_each_entry(sb, &super_blocks, s_list) {
611 spin_unlock(&sb_lock);
612 down_write(&sb->s_umount);
613 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
615 * What lock protects sb->s_flags??
617 do_remount_sb(sb, MS_RDONLY, NULL, 1);
619 up_write(&sb->s_umount);
623 spin_unlock(&sb_lock);
625 printk("Emergency Remount complete\n");
628 void emergency_remount(void)
630 struct work_struct *work;
632 work = kmalloc(sizeof(*work), GFP_ATOMIC);
634 INIT_WORK(work, do_emergency_remount);
640 * Unnamed block devices are dummy devices used by virtual
641 * filesystems which don't use real block-devices. -- jrs
644 static DEFINE_IDA(unnamed_dev_ida);
645 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
646 static int unnamed_dev_start = 0; /* don't bother trying below it */
648 int set_anon_super(struct super_block *s, void *data)
654 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
656 spin_lock(&unnamed_dev_lock);
657 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
659 unnamed_dev_start = dev + 1;
660 spin_unlock(&unnamed_dev_lock);
661 if (error == -EAGAIN)
662 /* We raced and lost with another CPU. */
667 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
668 spin_lock(&unnamed_dev_lock);
669 ida_remove(&unnamed_dev_ida, dev);
670 if (unnamed_dev_start > dev)
671 unnamed_dev_start = dev;
672 spin_unlock(&unnamed_dev_lock);
675 s->s_dev = MKDEV(0, dev & MINORMASK);
676 s->s_bdi = &noop_backing_dev_info;
680 EXPORT_SYMBOL(set_anon_super);
682 void kill_anon_super(struct super_block *sb)
684 int slot = MINOR(sb->s_dev);
686 generic_shutdown_super(sb);
687 spin_lock(&unnamed_dev_lock);
688 ida_remove(&unnamed_dev_ida, slot);
689 if (slot < unnamed_dev_start)
690 unnamed_dev_start = slot;
691 spin_unlock(&unnamed_dev_lock);
694 EXPORT_SYMBOL(kill_anon_super);
696 void kill_litter_super(struct super_block *sb)
699 d_genocide(sb->s_root);
703 EXPORT_SYMBOL(kill_litter_super);
705 static int ns_test_super(struct super_block *sb, void *data)
707 return sb->s_fs_info == data;
710 static int ns_set_super(struct super_block *sb, void *data)
712 sb->s_fs_info = data;
713 return set_anon_super(sb, NULL);
716 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
717 int (*fill_super)(struct super_block *, void *, int),
718 struct vfsmount *mnt)
720 struct super_block *sb;
722 sb = sget(fs_type, ns_test_super, ns_set_super, data);
729 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
731 deactivate_locked_super(sb);
735 sb->s_flags |= MS_ACTIVE;
738 simple_set_mnt(mnt, sb);
742 EXPORT_SYMBOL(get_sb_ns);
745 static int set_bdev_super(struct super_block *s, void *data)
748 s->s_dev = s->s_bdev->bd_dev;
751 * We set the bdi here to the queue backing, file systems can
752 * overwrite this in ->fill_super()
754 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
758 static int test_bdev_super(struct super_block *s, void *data)
760 return (void *)s->s_bdev == data;
763 int get_sb_bdev(struct file_system_type *fs_type,
764 int flags, const char *dev_name, void *data,
765 int (*fill_super)(struct super_block *, void *, int),
766 struct vfsmount *mnt)
768 struct block_device *bdev;
769 struct super_block *s;
770 fmode_t mode = FMODE_READ;
773 if (!(flags & MS_RDONLY))
776 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
778 return PTR_ERR(bdev);
781 * once the super is inserted into the list by sget, s_umount
782 * will protect the lockfs code from trying to start a snapshot
783 * while we are mounting
785 mutex_lock(&bdev->bd_fsfreeze_mutex);
786 if (bdev->bd_fsfreeze_count > 0) {
787 mutex_unlock(&bdev->bd_fsfreeze_mutex);
791 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
792 mutex_unlock(&bdev->bd_fsfreeze_mutex);
797 if ((flags ^ s->s_flags) & MS_RDONLY) {
798 deactivate_locked_super(s);
803 close_bdev_exclusive(bdev, mode);
805 char b[BDEVNAME_SIZE];
809 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
810 sb_set_blocksize(s, block_size(bdev));
811 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
813 deactivate_locked_super(s);
817 s->s_flags |= MS_ACTIVE;
821 simple_set_mnt(mnt, s);
827 close_bdev_exclusive(bdev, mode);
832 EXPORT_SYMBOL(get_sb_bdev);
834 void kill_block_super(struct super_block *sb)
836 struct block_device *bdev = sb->s_bdev;
837 fmode_t mode = sb->s_mode;
839 bdev->bd_super = NULL;
840 generic_shutdown_super(sb);
842 close_bdev_exclusive(bdev, mode);
845 EXPORT_SYMBOL(kill_block_super);
848 int get_sb_nodev(struct file_system_type *fs_type,
849 int flags, void *data,
850 int (*fill_super)(struct super_block *, void *, int),
851 struct vfsmount *mnt)
854 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
861 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
863 deactivate_locked_super(s);
866 s->s_flags |= MS_ACTIVE;
867 simple_set_mnt(mnt, s);
871 EXPORT_SYMBOL(get_sb_nodev);
873 static int compare_single(struct super_block *s, void *p)
878 int get_sb_single(struct file_system_type *fs_type,
879 int flags, void *data,
880 int (*fill_super)(struct super_block *, void *, int),
881 struct vfsmount *mnt)
883 struct super_block *s;
886 s = sget(fs_type, compare_single, set_anon_super, NULL);
891 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
893 deactivate_locked_super(s);
896 s->s_flags |= MS_ACTIVE;
898 do_remount_sb(s, flags, data, 0);
900 simple_set_mnt(mnt, s);
904 EXPORT_SYMBOL(get_sb_single);
907 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
909 struct vfsmount *mnt;
910 char *secdata = NULL;
914 return ERR_PTR(-ENODEV);
917 mnt = alloc_vfsmnt(name);
921 if (flags & MS_KERNMOUNT)
922 mnt->mnt_flags = MNT_INTERNAL;
924 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
925 secdata = alloc_secdata();
929 error = security_sb_copy_data(data, secdata);
931 goto out_free_secdata;
934 error = type->get_sb(type, flags, name, data, mnt);
936 goto out_free_secdata;
937 BUG_ON(!mnt->mnt_sb);
938 WARN_ON(!mnt->mnt_sb->s_bdi);
940 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
945 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
946 * but s_maxbytes was an unsigned long long for many releases. Throw
947 * this warning for a little while to try and catch filesystems that
948 * violate this rule. This warning should be either removed or
949 * converted to a BUG() in 2.6.34.
951 WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
952 "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
954 mnt->mnt_mountpoint = mnt->mnt_root;
955 mnt->mnt_parent = mnt;
956 up_write(&mnt->mnt_sb->s_umount);
957 free_secdata(secdata);
961 deactivate_locked_super(mnt->mnt_sb);
963 free_secdata(secdata);
967 return ERR_PTR(error);
970 EXPORT_SYMBOL_GPL(vfs_kern_mount);
972 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
975 const char *subtype = strchr(fstype, '.');
984 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
986 if (!mnt->mnt_sb->s_subtype)
996 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
998 struct file_system_type *type = get_fs_type(fstype);
999 struct vfsmount *mnt;
1001 return ERR_PTR(-ENODEV);
1002 mnt = vfs_kern_mount(type, flags, name, data);
1003 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1004 !mnt->mnt_sb->s_subtype)
1005 mnt = fs_set_subtype(mnt, fstype);
1006 put_filesystem(type);
1009 EXPORT_SYMBOL_GPL(do_kern_mount);
1011 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1013 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1016 EXPORT_SYMBOL_GPL(kern_mount_data);