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 int __put_super(struct super_block *sb)
137 if (!--sb->s_count) {
138 list_del_init(&sb->s_list);
146 * Drop a superblock's refcount.
147 * Returns non-zero if the superblock is about to be destroyed and
148 * at least is already removed from super_blocks list, so if we are
149 * making a loop through super blocks then we need to restart.
150 * The caller must hold sb_lock.
152 int __put_super_and_need_restart(struct super_block *sb)
154 /* check for race with generic_shutdown_super() */
155 if (list_empty(&sb->s_instances)) {
156 /* super block is removed, need to restart... */
160 /* can't be the last, since s_list is still in use */
162 BUG_ON(sb->s_count == 0);
167 * put_super - drop a temporary reference to superblock
168 * @sb: superblock in question
170 * Drops a temporary reference, frees superblock if there's no
173 void put_super(struct super_block *sb)
177 spin_unlock(&sb_lock);
182 * deactivate_locked_super - drop an active reference to superblock
183 * @s: superblock to deactivate
185 * Drops an active reference to superblock, converting it into a temprory
186 * one if there is no other active references left. In that case we
187 * tell fs driver to shut it down and drop the temporary reference we
190 * Caller holds exclusive lock on superblock; that lock is released.
192 void deactivate_locked_super(struct super_block *s)
194 struct file_system_type *fs = s->s_type;
195 if (atomic_dec_and_test(&s->s_active)) {
201 up_write(&s->s_umount);
205 EXPORT_SYMBOL(deactivate_locked_super);
208 * deactivate_super - drop an active reference to superblock
209 * @s: superblock to deactivate
211 * Variant of deactivate_locked_super(), except that superblock is *not*
212 * locked by caller. If we are going to drop the final active reference,
213 * lock will be acquired prior to that.
215 void deactivate_super(struct super_block *s)
217 if (!atomic_add_unless(&s->s_active, -1, 1)) {
218 down_write(&s->s_umount);
219 deactivate_locked_super(s);
223 EXPORT_SYMBOL(deactivate_super);
226 * grab_super - acquire an active reference
227 * @s: reference we are trying to make active
229 * Tries to acquire an active reference. grab_super() is used when we
230 * had just found a superblock in super_blocks or fs_type->fs_supers
231 * and want to turn it into a full-blown active reference. grab_super()
232 * is called with sb_lock held and drops it. Returns 1 in case of
233 * success, 0 if we had failed (superblock contents was already dead or
234 * dying when grab_super() had been called).
236 static int grab_super(struct super_block *s) __releases(sb_lock)
238 if (atomic_inc_not_zero(&s->s_active)) {
239 spin_unlock(&sb_lock);
240 down_write(&s->s_umount);
243 /* it's going away */
245 spin_unlock(&sb_lock);
246 /* wait for it to die */
247 down_write(&s->s_umount);
248 up_write(&s->s_umount);
254 * Superblock locking. We really ought to get rid of these two.
256 void lock_super(struct super_block * sb)
259 mutex_lock(&sb->s_lock);
262 void unlock_super(struct super_block * sb)
265 mutex_unlock(&sb->s_lock);
268 EXPORT_SYMBOL(lock_super);
269 EXPORT_SYMBOL(unlock_super);
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;
291 shrink_dcache_for_umount(sb);
294 sb->s_flags &= ~MS_ACTIVE;
296 /* bad name - it should be evict_inodes() */
297 invalidate_inodes(sb);
302 /* Forget any remaining inodes */
303 if (invalidate_inodes(sb)) {
304 printk("VFS: Busy inodes after unmount of %s. "
305 "Self-destruct in 5 seconds. Have a nice day...\n",
311 /* should be initialized for __put_super_and_need_restart() */
312 list_del_init(&sb->s_instances);
313 spin_unlock(&sb_lock);
314 up_write(&sb->s_umount);
317 EXPORT_SYMBOL(generic_shutdown_super);
320 * sget - find or create a superblock
321 * @type: filesystem type superblock should belong to
322 * @test: comparison callback
323 * @set: setup callback
324 * @data: argument to each of them
326 struct super_block *sget(struct file_system_type *type,
327 int (*test)(struct super_block *,void *),
328 int (*set)(struct super_block *,void *),
331 struct super_block *s = NULL;
332 struct super_block *old;
338 list_for_each_entry(old, &type->fs_supers, s_instances) {
339 if (!test(old, data))
341 if (!grab_super(old))
344 up_write(&s->s_umount);
351 spin_unlock(&sb_lock);
352 s = alloc_super(type);
354 return ERR_PTR(-ENOMEM);
360 spin_unlock(&sb_lock);
361 up_write(&s->s_umount);
366 strlcpy(s->s_id, type->name, sizeof(s->s_id));
367 list_add_tail(&s->s_list, &super_blocks);
368 list_add(&s->s_instances, &type->fs_supers);
369 spin_unlock(&sb_lock);
370 get_filesystem(type);
376 void drop_super(struct super_block *sb)
378 up_read(&sb->s_umount);
382 EXPORT_SYMBOL(drop_super);
385 * sync_supers - helper for periodic superblock writeback
387 * Call the write_super method if present on all dirty superblocks in
388 * the system. This is for the periodic writeback used by most older
389 * filesystems. For data integrity superblock writeback use
390 * sync_filesystems() instead.
392 * Note: check the dirty flag before waiting, so we don't
393 * hold up the sync while mounting a device. (The newly
394 * mounted device won't need syncing.)
396 void sync_supers(void)
398 struct super_block *sb;
402 list_for_each_entry(sb, &super_blocks, s_list) {
403 if (list_empty(&sb->s_instances))
405 if (sb->s_op->write_super && sb->s_dirt) {
407 spin_unlock(&sb_lock);
409 down_read(&sb->s_umount);
410 if (sb->s_root && sb->s_dirt)
411 sb->s_op->write_super(sb);
412 up_read(&sb->s_umount);
415 if (__put_super_and_need_restart(sb))
419 spin_unlock(&sb_lock);
423 * get_super - get the superblock of a device
424 * @bdev: device to get the superblock for
426 * Scans the superblock list and finds the superblock of the file system
427 * mounted on the device given. %NULL is returned if no match is found.
430 struct super_block * get_super(struct block_device *bdev)
432 struct super_block *sb;
439 list_for_each_entry(sb, &super_blocks, s_list) {
440 if (list_empty(&sb->s_instances))
442 if (sb->s_bdev == bdev) {
444 spin_unlock(&sb_lock);
445 down_read(&sb->s_umount);
448 up_read(&sb->s_umount);
449 /* restart only when sb is no longer on the list */
451 if (__put_super_and_need_restart(sb))
455 spin_unlock(&sb_lock);
459 EXPORT_SYMBOL(get_super);
462 * get_active_super - get an active reference to the superblock of a device
463 * @bdev: device to get the superblock for
465 * Scans the superblock list and finds the superblock of the file system
466 * mounted on the device given. Returns the superblock with an active
467 * reference and s_umount held exclusively or %NULL if none was found.
469 struct super_block *get_active_super(struct block_device *bdev)
471 struct super_block *sb;
477 list_for_each_entry(sb, &super_blocks, s_list) {
478 if (list_empty(&sb->s_instances))
480 if (sb->s_bdev != bdev)
483 if (grab_super(sb)) /* drops sb_lock */
488 spin_unlock(&sb_lock);
492 struct super_block * user_get_super(dev_t dev)
494 struct super_block *sb;
498 list_for_each_entry(sb, &super_blocks, s_list) {
499 if (list_empty(&sb->s_instances))
501 if (sb->s_dev == dev) {
503 spin_unlock(&sb_lock);
504 down_read(&sb->s_umount);
507 up_read(&sb->s_umount);
508 /* restart only when sb is no longer on the list */
510 if (__put_super_and_need_restart(sb))
514 spin_unlock(&sb_lock);
518 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
520 struct super_block *s;
525 s = user_get_super(new_decode_dev(dev));
528 err = vfs_statfs(s->s_root, &sbuf);
533 memset(&tmp,0,sizeof(struct ustat));
534 tmp.f_tfree = sbuf.f_bfree;
535 tmp.f_tinode = sbuf.f_ffree;
537 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
543 * do_remount_sb - asks filesystem to change mount options.
544 * @sb: superblock in question
545 * @flags: numeric part of options
546 * @data: the rest of options
547 * @force: whether or not to force the change
549 * Alters the mount options of a mounted file system.
551 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
554 int remount_rw, remount_ro;
556 if (sb->s_frozen != SB_UNFROZEN)
560 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
564 if (flags & MS_RDONLY)
566 shrink_dcache_sb(sb);
569 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
570 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
572 /* If we are remounting RDONLY and current sb is read/write,
573 make sure there are no rw files opened */
577 else if (!fs_may_remount_ro(sb))
579 retval = vfs_dq_off(sb, 1);
580 if (retval < 0 && retval != -ENOSYS)
584 if (sb->s_op->remount_fs) {
585 retval = sb->s_op->remount_fs(sb, &flags, data);
589 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
591 vfs_dq_quota_on_remount(sb);
593 * Some filesystems modify their metadata via some other path than the
594 * bdev buffer cache (eg. use a private mapping, or directories in
595 * pagecache, etc). Also file data modifications go via their own
596 * mappings. So If we try to mount readonly then copy the filesystem
597 * from bdev, we could get stale data, so invalidate it to give a best
598 * effort at coherency.
600 if (remount_ro && sb->s_bdev)
601 invalidate_bdev(sb->s_bdev);
605 static void do_emergency_remount(struct work_struct *work)
607 struct super_block *sb;
610 list_for_each_entry(sb, &super_blocks, s_list) {
611 if (list_empty(&sb->s_instances))
614 spin_unlock(&sb_lock);
615 down_write(&sb->s_umount);
616 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
618 * What lock protects sb->s_flags??
620 do_remount_sb(sb, MS_RDONLY, NULL, 1);
622 up_write(&sb->s_umount);
626 spin_unlock(&sb_lock);
628 printk("Emergency Remount complete\n");
631 void emergency_remount(void)
633 struct work_struct *work;
635 work = kmalloc(sizeof(*work), GFP_ATOMIC);
637 INIT_WORK(work, do_emergency_remount);
643 * Unnamed block devices are dummy devices used by virtual
644 * filesystems which don't use real block-devices. -- jrs
647 static DEFINE_IDA(unnamed_dev_ida);
648 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
649 static int unnamed_dev_start = 0; /* don't bother trying below it */
651 int set_anon_super(struct super_block *s, void *data)
657 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
659 spin_lock(&unnamed_dev_lock);
660 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
662 unnamed_dev_start = dev + 1;
663 spin_unlock(&unnamed_dev_lock);
664 if (error == -EAGAIN)
665 /* We raced and lost with another CPU. */
670 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
671 spin_lock(&unnamed_dev_lock);
672 ida_remove(&unnamed_dev_ida, dev);
673 if (unnamed_dev_start > dev)
674 unnamed_dev_start = dev;
675 spin_unlock(&unnamed_dev_lock);
678 s->s_dev = MKDEV(0, dev & MINORMASK);
679 s->s_bdi = &noop_backing_dev_info;
683 EXPORT_SYMBOL(set_anon_super);
685 void kill_anon_super(struct super_block *sb)
687 int slot = MINOR(sb->s_dev);
689 generic_shutdown_super(sb);
690 spin_lock(&unnamed_dev_lock);
691 ida_remove(&unnamed_dev_ida, slot);
692 if (slot < unnamed_dev_start)
693 unnamed_dev_start = slot;
694 spin_unlock(&unnamed_dev_lock);
697 EXPORT_SYMBOL(kill_anon_super);
699 void kill_litter_super(struct super_block *sb)
702 d_genocide(sb->s_root);
706 EXPORT_SYMBOL(kill_litter_super);
708 static int ns_test_super(struct super_block *sb, void *data)
710 return sb->s_fs_info == data;
713 static int ns_set_super(struct super_block *sb, void *data)
715 sb->s_fs_info = data;
716 return set_anon_super(sb, NULL);
719 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
720 int (*fill_super)(struct super_block *, void *, int),
721 struct vfsmount *mnt)
723 struct super_block *sb;
725 sb = sget(fs_type, ns_test_super, ns_set_super, data);
732 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
734 deactivate_locked_super(sb);
738 sb->s_flags |= MS_ACTIVE;
741 simple_set_mnt(mnt, sb);
745 EXPORT_SYMBOL(get_sb_ns);
748 static int set_bdev_super(struct super_block *s, void *data)
751 s->s_dev = s->s_bdev->bd_dev;
754 * We set the bdi here to the queue backing, file systems can
755 * overwrite this in ->fill_super()
757 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
761 static int test_bdev_super(struct super_block *s, void *data)
763 return (void *)s->s_bdev == data;
766 int get_sb_bdev(struct file_system_type *fs_type,
767 int flags, const char *dev_name, void *data,
768 int (*fill_super)(struct super_block *, void *, int),
769 struct vfsmount *mnt)
771 struct block_device *bdev;
772 struct super_block *s;
773 fmode_t mode = FMODE_READ;
776 if (!(flags & MS_RDONLY))
779 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
781 return PTR_ERR(bdev);
784 * once the super is inserted into the list by sget, s_umount
785 * will protect the lockfs code from trying to start a snapshot
786 * while we are mounting
788 mutex_lock(&bdev->bd_fsfreeze_mutex);
789 if (bdev->bd_fsfreeze_count > 0) {
790 mutex_unlock(&bdev->bd_fsfreeze_mutex);
794 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
795 mutex_unlock(&bdev->bd_fsfreeze_mutex);
800 if ((flags ^ s->s_flags) & MS_RDONLY) {
801 deactivate_locked_super(s);
806 close_bdev_exclusive(bdev, mode);
808 char b[BDEVNAME_SIZE];
812 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
813 sb_set_blocksize(s, block_size(bdev));
814 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
816 deactivate_locked_super(s);
820 s->s_flags |= MS_ACTIVE;
824 simple_set_mnt(mnt, s);
830 close_bdev_exclusive(bdev, mode);
835 EXPORT_SYMBOL(get_sb_bdev);
837 void kill_block_super(struct super_block *sb)
839 struct block_device *bdev = sb->s_bdev;
840 fmode_t mode = sb->s_mode;
842 bdev->bd_super = NULL;
843 generic_shutdown_super(sb);
845 close_bdev_exclusive(bdev, mode);
848 EXPORT_SYMBOL(kill_block_super);
851 int get_sb_nodev(struct file_system_type *fs_type,
852 int flags, void *data,
853 int (*fill_super)(struct super_block *, void *, int),
854 struct vfsmount *mnt)
857 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
864 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
866 deactivate_locked_super(s);
869 s->s_flags |= MS_ACTIVE;
870 simple_set_mnt(mnt, s);
874 EXPORT_SYMBOL(get_sb_nodev);
876 static int compare_single(struct super_block *s, void *p)
881 int get_sb_single(struct file_system_type *fs_type,
882 int flags, void *data,
883 int (*fill_super)(struct super_block *, void *, int),
884 struct vfsmount *mnt)
886 struct super_block *s;
889 s = sget(fs_type, compare_single, set_anon_super, NULL);
894 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
896 deactivate_locked_super(s);
899 s->s_flags |= MS_ACTIVE;
901 do_remount_sb(s, flags, data, 0);
903 simple_set_mnt(mnt, s);
907 EXPORT_SYMBOL(get_sb_single);
910 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
912 struct vfsmount *mnt;
913 char *secdata = NULL;
917 return ERR_PTR(-ENODEV);
920 mnt = alloc_vfsmnt(name);
924 if (flags & MS_KERNMOUNT)
925 mnt->mnt_flags = MNT_INTERNAL;
927 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
928 secdata = alloc_secdata();
932 error = security_sb_copy_data(data, secdata);
934 goto out_free_secdata;
937 error = type->get_sb(type, flags, name, data, mnt);
939 goto out_free_secdata;
940 BUG_ON(!mnt->mnt_sb);
941 WARN_ON(!mnt->mnt_sb->s_bdi);
943 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
948 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
949 * but s_maxbytes was an unsigned long long for many releases. Throw
950 * this warning for a little while to try and catch filesystems that
951 * violate this rule. This warning should be either removed or
952 * converted to a BUG() in 2.6.34.
954 WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
955 "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
957 mnt->mnt_mountpoint = mnt->mnt_root;
958 mnt->mnt_parent = mnt;
959 up_write(&mnt->mnt_sb->s_umount);
960 free_secdata(secdata);
964 deactivate_locked_super(mnt->mnt_sb);
966 free_secdata(secdata);
970 return ERR_PTR(error);
973 EXPORT_SYMBOL_GPL(vfs_kern_mount);
975 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
978 const char *subtype = strchr(fstype, '.');
987 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
989 if (!mnt->mnt_sb->s_subtype)
999 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1001 struct file_system_type *type = get_fs_type(fstype);
1002 struct vfsmount *mnt;
1004 return ERR_PTR(-ENODEV);
1005 mnt = vfs_kern_mount(type, flags, name, data);
1006 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1007 !mnt->mnt_sb->s_subtype)
1008 mnt = fs_set_subtype(mnt, fstype);
1009 put_filesystem(type);
1012 EXPORT_SYMBOL_GPL(do_kern_mount);
1014 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1016 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1019 EXPORT_SYMBOL_GPL(kern_mount_data);