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_lock(&s->s_active, &sb_lock)) {
193 s->s_count -= S_BIAS-1;
194 spin_unlock(&sb_lock);
196 down_write(&s->s_umount);
203 EXPORT_SYMBOL(deactivate_super);
206 * deactivate_locked_super - drop an active reference to superblock
207 * @s: superblock to deactivate
209 * Equivalent of up_write(&s->s_umount); deactivate_super(s);, except that
210 * it does not unlock it until it's all over. As the result, it's safe to
211 * use to dispose of new superblock on ->get_sb() failure exits - nobody
212 * will see the sucker until it's all over. Equivalent using up_write +
213 * deactivate_super is safe for that purpose only if superblock is either
214 * safe to use or has NULL ->s_root when we unlock.
216 void deactivate_locked_super(struct super_block *s)
218 struct file_system_type *fs = s->s_type;
219 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
220 s->s_count -= S_BIAS-1;
221 spin_unlock(&sb_lock);
227 up_write(&s->s_umount);
231 EXPORT_SYMBOL(deactivate_locked_super);
234 * grab_super - acquire an active reference
235 * @s: reference we are trying to make active
237 * Tries to acquire an active reference. grab_super() is used when we
238 * had just found a superblock in super_blocks or fs_type->fs_supers
239 * and want to turn it into a full-blown active reference. grab_super()
240 * is called with sb_lock held and drops it. Returns 1 in case of
241 * success, 0 if we had failed (superblock contents was already dead or
242 * dying when grab_super() had been called).
244 static int grab_super(struct super_block *s) __releases(sb_lock)
247 spin_unlock(&sb_lock);
248 down_write(&s->s_umount);
251 if (s->s_count > S_BIAS) {
252 atomic_inc(&s->s_active);
254 spin_unlock(&sb_lock);
257 spin_unlock(&sb_lock);
259 up_write(&s->s_umount);
266 * Superblock locking. We really ought to get rid of these two.
268 void lock_super(struct super_block * sb)
271 mutex_lock(&sb->s_lock);
274 void unlock_super(struct super_block * sb)
277 mutex_unlock(&sb->s_lock);
280 EXPORT_SYMBOL(lock_super);
281 EXPORT_SYMBOL(unlock_super);
284 * generic_shutdown_super - common helper for ->kill_sb()
285 * @sb: superblock to kill
287 * generic_shutdown_super() does all fs-independent work on superblock
288 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
289 * that need destruction out of superblock, call generic_shutdown_super()
290 * and release aforementioned objects. Note: dentries and inodes _are_
291 * taken care of and do not need specific handling.
293 * Upon calling this function, the filesystem may no longer alter or
294 * rearrange the set of dentries belonging to this super_block, nor may it
295 * change the attachments of dentries to inodes.
297 void generic_shutdown_super(struct super_block *sb)
299 const struct super_operations *sop = sb->s_op;
303 shrink_dcache_for_umount(sb);
306 sb->s_flags &= ~MS_ACTIVE;
308 /* bad name - it should be evict_inodes() */
309 invalidate_inodes(sb);
314 /* Forget any remaining inodes */
315 if (invalidate_inodes(sb)) {
316 printk("VFS: Busy inodes after unmount of %s. "
317 "Self-destruct in 5 seconds. Have a nice day...\n",
323 /* should be initialized for __put_super_and_need_restart() */
324 list_del_init(&sb->s_list);
325 list_del(&sb->s_instances);
326 spin_unlock(&sb_lock);
327 up_write(&sb->s_umount);
330 EXPORT_SYMBOL(generic_shutdown_super);
333 * sget - find or create a superblock
334 * @type: filesystem type superblock should belong to
335 * @test: comparison callback
336 * @set: setup callback
337 * @data: argument to each of them
339 struct super_block *sget(struct file_system_type *type,
340 int (*test)(struct super_block *,void *),
341 int (*set)(struct super_block *,void *),
344 struct super_block *s = NULL;
345 struct super_block *old;
351 list_for_each_entry(old, &type->fs_supers, s_instances) {
352 if (!test(old, data))
354 if (!grab_super(old))
357 up_write(&s->s_umount);
364 spin_unlock(&sb_lock);
365 s = alloc_super(type);
367 return ERR_PTR(-ENOMEM);
373 spin_unlock(&sb_lock);
374 up_write(&s->s_umount);
379 strlcpy(s->s_id, type->name, sizeof(s->s_id));
380 list_add_tail(&s->s_list, &super_blocks);
381 list_add(&s->s_instances, &type->fs_supers);
382 spin_unlock(&sb_lock);
383 get_filesystem(type);
389 void drop_super(struct super_block *sb)
391 up_read(&sb->s_umount);
395 EXPORT_SYMBOL(drop_super);
398 * sync_supers - helper for periodic superblock writeback
400 * Call the write_super method if present on all dirty superblocks in
401 * the system. This is for the periodic writeback used by most older
402 * filesystems. For data integrity superblock writeback use
403 * sync_filesystems() instead.
405 * Note: check the dirty flag before waiting, so we don't
406 * hold up the sync while mounting a device. (The newly
407 * mounted device won't need syncing.)
409 void sync_supers(void)
411 struct super_block *sb;
415 list_for_each_entry(sb, &super_blocks, s_list) {
416 if (sb->s_op->write_super && sb->s_dirt) {
418 spin_unlock(&sb_lock);
420 down_read(&sb->s_umount);
421 if (sb->s_root && sb->s_dirt)
422 sb->s_op->write_super(sb);
423 up_read(&sb->s_umount);
426 if (__put_super_and_need_restart(sb))
430 spin_unlock(&sb_lock);
434 * get_super - get the superblock of a device
435 * @bdev: device to get the superblock for
437 * Scans the superblock list and finds the superblock of the file system
438 * mounted on the device given. %NULL is returned if no match is found.
441 struct super_block * get_super(struct block_device *bdev)
443 struct super_block *sb;
450 list_for_each_entry(sb, &super_blocks, s_list) {
451 if (sb->s_bdev == bdev) {
453 spin_unlock(&sb_lock);
454 down_read(&sb->s_umount);
457 up_read(&sb->s_umount);
458 /* restart only when sb is no longer on the list */
460 if (__put_super_and_need_restart(sb))
464 spin_unlock(&sb_lock);
468 EXPORT_SYMBOL(get_super);
471 * get_active_super - get an active reference to the superblock of a device
472 * @bdev: device to get the superblock for
474 * Scans the superblock list and finds the superblock of the file system
475 * mounted on the device given. Returns the superblock with an active
476 * reference and s_umount held exclusively or %NULL if none was found.
478 struct super_block *get_active_super(struct block_device *bdev)
480 struct super_block *sb;
486 list_for_each_entry(sb, &super_blocks, s_list) {
487 if (sb->s_bdev != bdev)
490 if (grab_super(sb)) /* drops sb_lock */
495 spin_unlock(&sb_lock);
499 struct super_block * user_get_super(dev_t dev)
501 struct super_block *sb;
505 list_for_each_entry(sb, &super_blocks, s_list) {
506 if (sb->s_dev == dev) {
508 spin_unlock(&sb_lock);
509 down_read(&sb->s_umount);
512 up_read(&sb->s_umount);
513 /* restart only when sb is no longer on the list */
515 if (__put_super_and_need_restart(sb))
519 spin_unlock(&sb_lock);
523 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
525 struct super_block *s;
530 s = user_get_super(new_decode_dev(dev));
533 err = vfs_statfs(s->s_root, &sbuf);
538 memset(&tmp,0,sizeof(struct ustat));
539 tmp.f_tfree = sbuf.f_bfree;
540 tmp.f_tinode = sbuf.f_ffree;
542 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
548 * do_remount_sb - asks filesystem to change mount options.
549 * @sb: superblock in question
550 * @flags: numeric part of options
551 * @data: the rest of options
552 * @force: whether or not to force the change
554 * Alters the mount options of a mounted file system.
556 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
559 int remount_rw, remount_ro;
561 if (sb->s_frozen != SB_UNFROZEN)
565 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
569 if (flags & MS_RDONLY)
571 shrink_dcache_sb(sb);
574 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
575 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
577 /* If we are remounting RDONLY and current sb is read/write,
578 make sure there are no rw files opened */
582 else if (!fs_may_remount_ro(sb))
584 retval = vfs_dq_off(sb, 1);
585 if (retval < 0 && retval != -ENOSYS)
589 if (sb->s_op->remount_fs) {
590 retval = sb->s_op->remount_fs(sb, &flags, data);
594 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
596 vfs_dq_quota_on_remount(sb);
598 * Some filesystems modify their metadata via some other path than the
599 * bdev buffer cache (eg. use a private mapping, or directories in
600 * pagecache, etc). Also file data modifications go via their own
601 * mappings. So If we try to mount readonly then copy the filesystem
602 * from bdev, we could get stale data, so invalidate it to give a best
603 * effort at coherency.
605 if (remount_ro && sb->s_bdev)
606 invalidate_bdev(sb->s_bdev);
610 static void do_emergency_remount(struct work_struct *work)
612 struct super_block *sb;
615 list_for_each_entry(sb, &super_blocks, s_list) {
617 spin_unlock(&sb_lock);
618 down_write(&sb->s_umount);
619 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
621 * What lock protects sb->s_flags??
623 do_remount_sb(sb, MS_RDONLY, NULL, 1);
625 up_write(&sb->s_umount);
629 spin_unlock(&sb_lock);
631 printk("Emergency Remount complete\n");
634 void emergency_remount(void)
636 struct work_struct *work;
638 work = kmalloc(sizeof(*work), GFP_ATOMIC);
640 INIT_WORK(work, do_emergency_remount);
646 * Unnamed block devices are dummy devices used by virtual
647 * filesystems which don't use real block-devices. -- jrs
650 static DEFINE_IDA(unnamed_dev_ida);
651 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
652 static int unnamed_dev_start = 0; /* don't bother trying below it */
654 int set_anon_super(struct super_block *s, void *data)
660 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
662 spin_lock(&unnamed_dev_lock);
663 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
665 unnamed_dev_start = dev + 1;
666 spin_unlock(&unnamed_dev_lock);
667 if (error == -EAGAIN)
668 /* We raced and lost with another CPU. */
673 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
674 spin_lock(&unnamed_dev_lock);
675 ida_remove(&unnamed_dev_ida, dev);
676 if (unnamed_dev_start > dev)
677 unnamed_dev_start = dev;
678 spin_unlock(&unnamed_dev_lock);
681 s->s_dev = MKDEV(0, dev & MINORMASK);
682 s->s_bdi = &noop_backing_dev_info;
686 EXPORT_SYMBOL(set_anon_super);
688 void kill_anon_super(struct super_block *sb)
690 int slot = MINOR(sb->s_dev);
692 generic_shutdown_super(sb);
693 spin_lock(&unnamed_dev_lock);
694 ida_remove(&unnamed_dev_ida, slot);
695 if (slot < unnamed_dev_start)
696 unnamed_dev_start = slot;
697 spin_unlock(&unnamed_dev_lock);
700 EXPORT_SYMBOL(kill_anon_super);
702 void kill_litter_super(struct super_block *sb)
705 d_genocide(sb->s_root);
709 EXPORT_SYMBOL(kill_litter_super);
711 static int ns_test_super(struct super_block *sb, void *data)
713 return sb->s_fs_info == data;
716 static int ns_set_super(struct super_block *sb, void *data)
718 sb->s_fs_info = data;
719 return set_anon_super(sb, NULL);
722 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
723 int (*fill_super)(struct super_block *, void *, int),
724 struct vfsmount *mnt)
726 struct super_block *sb;
728 sb = sget(fs_type, ns_test_super, ns_set_super, data);
735 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
737 deactivate_locked_super(sb);
741 sb->s_flags |= MS_ACTIVE;
744 simple_set_mnt(mnt, sb);
748 EXPORT_SYMBOL(get_sb_ns);
751 static int set_bdev_super(struct super_block *s, void *data)
754 s->s_dev = s->s_bdev->bd_dev;
757 * We set the bdi here to the queue backing, file systems can
758 * overwrite this in ->fill_super()
760 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
764 static int test_bdev_super(struct super_block *s, void *data)
766 return (void *)s->s_bdev == data;
769 int get_sb_bdev(struct file_system_type *fs_type,
770 int flags, const char *dev_name, void *data,
771 int (*fill_super)(struct super_block *, void *, int),
772 struct vfsmount *mnt)
774 struct block_device *bdev;
775 struct super_block *s;
776 fmode_t mode = FMODE_READ;
779 if (!(flags & MS_RDONLY))
782 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
784 return PTR_ERR(bdev);
787 * once the super is inserted into the list by sget, s_umount
788 * will protect the lockfs code from trying to start a snapshot
789 * while we are mounting
791 mutex_lock(&bdev->bd_fsfreeze_mutex);
792 if (bdev->bd_fsfreeze_count > 0) {
793 mutex_unlock(&bdev->bd_fsfreeze_mutex);
797 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
798 mutex_unlock(&bdev->bd_fsfreeze_mutex);
803 if ((flags ^ s->s_flags) & MS_RDONLY) {
804 deactivate_locked_super(s);
809 close_bdev_exclusive(bdev, mode);
811 char b[BDEVNAME_SIZE];
815 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
816 sb_set_blocksize(s, block_size(bdev));
817 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
819 deactivate_locked_super(s);
823 s->s_flags |= MS_ACTIVE;
827 simple_set_mnt(mnt, s);
833 close_bdev_exclusive(bdev, mode);
838 EXPORT_SYMBOL(get_sb_bdev);
840 void kill_block_super(struct super_block *sb)
842 struct block_device *bdev = sb->s_bdev;
843 fmode_t mode = sb->s_mode;
845 bdev->bd_super = NULL;
846 generic_shutdown_super(sb);
848 close_bdev_exclusive(bdev, mode);
851 EXPORT_SYMBOL(kill_block_super);
854 int get_sb_nodev(struct file_system_type *fs_type,
855 int flags, void *data,
856 int (*fill_super)(struct super_block *, void *, int),
857 struct vfsmount *mnt)
860 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
867 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
869 deactivate_locked_super(s);
872 s->s_flags |= MS_ACTIVE;
873 simple_set_mnt(mnt, s);
877 EXPORT_SYMBOL(get_sb_nodev);
879 static int compare_single(struct super_block *s, void *p)
884 int get_sb_single(struct file_system_type *fs_type,
885 int flags, void *data,
886 int (*fill_super)(struct super_block *, void *, int),
887 struct vfsmount *mnt)
889 struct super_block *s;
892 s = sget(fs_type, compare_single, set_anon_super, NULL);
897 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
899 deactivate_locked_super(s);
902 s->s_flags |= MS_ACTIVE;
904 do_remount_sb(s, flags, data, 0);
906 simple_set_mnt(mnt, s);
910 EXPORT_SYMBOL(get_sb_single);
913 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
915 struct vfsmount *mnt;
916 char *secdata = NULL;
920 return ERR_PTR(-ENODEV);
923 mnt = alloc_vfsmnt(name);
927 if (flags & MS_KERNMOUNT)
928 mnt->mnt_flags = MNT_INTERNAL;
930 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
931 secdata = alloc_secdata();
935 error = security_sb_copy_data(data, secdata);
937 goto out_free_secdata;
940 error = type->get_sb(type, flags, name, data, mnt);
942 goto out_free_secdata;
943 BUG_ON(!mnt->mnt_sb);
944 WARN_ON(!mnt->mnt_sb->s_bdi);
946 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
951 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
952 * but s_maxbytes was an unsigned long long for many releases. Throw
953 * this warning for a little while to try and catch filesystems that
954 * violate this rule. This warning should be either removed or
955 * converted to a BUG() in 2.6.34.
957 WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
958 "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
960 mnt->mnt_mountpoint = mnt->mnt_root;
961 mnt->mnt_parent = mnt;
962 up_write(&mnt->mnt_sb->s_umount);
963 free_secdata(secdata);
967 deactivate_locked_super(mnt->mnt_sb);
969 free_secdata(secdata);
973 return ERR_PTR(error);
976 EXPORT_SYMBOL_GPL(vfs_kern_mount);
978 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
981 const char *subtype = strchr(fstype, '.');
990 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
992 if (!mnt->mnt_sb->s_subtype)
1002 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1004 struct file_system_type *type = get_fs_type(fstype);
1005 struct vfsmount *mnt;
1007 return ERR_PTR(-ENODEV);
1008 mnt = vfs_kern_mount(type, flags, name, data);
1009 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1010 !mnt->mnt_sb->s_subtype)
1011 mnt = fs_set_subtype(mnt, fstype);
1012 put_filesystem(type);
1015 EXPORT_SYMBOL_GPL(do_kern_mount);
1017 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1019 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1022 EXPORT_SYMBOL_GPL(kern_mount_data);