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/writeback.h> /* for the emergency remount stuff */
35 #include <linux/idr.h>
36 #include <linux/kobject.h>
37 #include <linux/mutex.h>
38 #include <linux/file.h>
39 #include <linux/backing-dev.h>
40 #include <asm/uaccess.h>
44 LIST_HEAD(super_blocks);
45 DEFINE_SPINLOCK(sb_lock);
48 * alloc_super - create new superblock
49 * @type: filesystem type superblock should belong to
51 * Allocates and initializes a new &struct super_block. alloc_super()
52 * returns a pointer new superblock or %NULL if allocation had failed.
54 static struct super_block *alloc_super(struct file_system_type *type)
56 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
57 static const struct super_operations default_op;
60 if (security_sb_alloc(s)) {
65 INIT_LIST_HEAD(&s->s_files);
66 INIT_LIST_HEAD(&s->s_instances);
67 INIT_HLIST_HEAD(&s->s_anon);
68 INIT_LIST_HEAD(&s->s_inodes);
69 INIT_LIST_HEAD(&s->s_dentry_lru);
70 init_rwsem(&s->s_umount);
71 mutex_init(&s->s_lock);
72 lockdep_set_class(&s->s_umount, &type->s_umount_key);
74 * The locking rules for s_lock are up to the
75 * filesystem. For example ext3fs has different
76 * lock ordering than usbfs:
78 lockdep_set_class(&s->s_lock, &type->s_lock_key);
80 * sget() can have s_umount recursion.
82 * When it cannot find a suitable sb, it allocates a new
83 * one (this one), and tries again to find a suitable old
86 * In case that succeeds, it will acquire the s_umount
87 * lock of the old one. Since these are clearly distrinct
88 * locks, and this object isn't exposed yet, there's no
91 * Annotate this by putting this lock in a different
94 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
96 atomic_set(&s->s_active, 1);
97 mutex_init(&s->s_vfs_rename_mutex);
98 mutex_init(&s->s_dquot.dqio_mutex);
99 mutex_init(&s->s_dquot.dqonoff_mutex);
100 init_rwsem(&s->s_dquot.dqptr_sem);
101 init_waitqueue_head(&s->s_wait_unfrozen);
102 s->s_maxbytes = MAX_NON_LFS;
103 s->dq_op = sb_dquot_ops;
104 s->s_qcop = sb_quotactl_ops;
105 s->s_op = &default_op;
106 s->s_time_gran = 1000000000;
113 * destroy_super - frees a superblock
114 * @s: superblock to free
116 * Frees a superblock.
118 static inline void destroy_super(struct super_block *s)
126 /* Superblock refcounting */
129 * Drop a superblock's refcount. The caller must hold sb_lock.
131 void __put_super(struct super_block *sb)
133 if (!--sb->s_count) {
134 list_del_init(&sb->s_list);
140 * put_super - drop a temporary reference to superblock
141 * @sb: superblock in question
143 * Drops a temporary reference, frees superblock if there's no
146 void put_super(struct super_block *sb)
150 spin_unlock(&sb_lock);
155 * deactivate_locked_super - drop an active reference to superblock
156 * @s: superblock to deactivate
158 * Drops an active reference to superblock, converting it into a temprory
159 * one if there is no other active references left. In that case we
160 * tell fs driver to shut it down and drop the temporary reference we
163 * Caller holds exclusive lock on superblock; that lock is released.
165 void deactivate_locked_super(struct super_block *s)
167 struct file_system_type *fs = s->s_type;
168 if (atomic_dec_and_test(&s->s_active)) {
174 up_write(&s->s_umount);
178 EXPORT_SYMBOL(deactivate_locked_super);
181 * deactivate_super - drop an active reference to superblock
182 * @s: superblock to deactivate
184 * Variant of deactivate_locked_super(), except that superblock is *not*
185 * locked by caller. If we are going to drop the final active reference,
186 * lock will be acquired prior to that.
188 void deactivate_super(struct super_block *s)
190 if (!atomic_add_unless(&s->s_active, -1, 1)) {
191 down_write(&s->s_umount);
192 deactivate_locked_super(s);
196 EXPORT_SYMBOL(deactivate_super);
199 * grab_super - acquire an active reference
200 * @s: reference we are trying to make active
202 * Tries to acquire an active reference. grab_super() is used when we
203 * had just found a superblock in super_blocks or fs_type->fs_supers
204 * and want to turn it into a full-blown active reference. grab_super()
205 * is called with sb_lock held and drops it. Returns 1 in case of
206 * success, 0 if we had failed (superblock contents was already dead or
207 * dying when grab_super() had been called).
209 static int grab_super(struct super_block *s) __releases(sb_lock)
211 if (atomic_inc_not_zero(&s->s_active)) {
212 spin_unlock(&sb_lock);
215 /* it's going away */
217 spin_unlock(&sb_lock);
218 /* wait for it to die */
219 down_write(&s->s_umount);
220 up_write(&s->s_umount);
226 * Superblock locking. We really ought to get rid of these two.
228 void lock_super(struct super_block * sb)
231 mutex_lock(&sb->s_lock);
234 void unlock_super(struct super_block * sb)
237 mutex_unlock(&sb->s_lock);
240 EXPORT_SYMBOL(lock_super);
241 EXPORT_SYMBOL(unlock_super);
244 * generic_shutdown_super - common helper for ->kill_sb()
245 * @sb: superblock to kill
247 * generic_shutdown_super() does all fs-independent work on superblock
248 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
249 * that need destruction out of superblock, call generic_shutdown_super()
250 * and release aforementioned objects. Note: dentries and inodes _are_
251 * taken care of and do not need specific handling.
253 * Upon calling this function, the filesystem may no longer alter or
254 * rearrange the set of dentries belonging to this super_block, nor may it
255 * change the attachments of dentries to inodes.
257 void generic_shutdown_super(struct super_block *sb)
259 const struct super_operations *sop = sb->s_op;
263 shrink_dcache_for_umount(sb);
266 sb->s_flags &= ~MS_ACTIVE;
268 /* bad name - it should be evict_inodes() */
269 invalidate_inodes(sb);
274 /* Forget any remaining inodes */
275 if (invalidate_inodes(sb)) {
276 printk("VFS: Busy inodes after unmount of %s. "
277 "Self-destruct in 5 seconds. Have a nice day...\n",
283 /* should be initialized for __put_super_and_need_restart() */
284 list_del_init(&sb->s_instances);
285 spin_unlock(&sb_lock);
286 up_write(&sb->s_umount);
289 EXPORT_SYMBOL(generic_shutdown_super);
292 * sget - find or create a superblock
293 * @type: filesystem type superblock should belong to
294 * @test: comparison callback
295 * @set: setup callback
296 * @data: argument to each of them
298 struct super_block *sget(struct file_system_type *type,
299 int (*test)(struct super_block *,void *),
300 int (*set)(struct super_block *,void *),
303 struct super_block *s = NULL;
304 struct super_block *old;
310 list_for_each_entry(old, &type->fs_supers, s_instances) {
311 if (!test(old, data))
313 if (!grab_super(old))
316 up_write(&s->s_umount);
319 down_write(&old->s_umount);
324 spin_unlock(&sb_lock);
325 s = alloc_super(type);
327 return ERR_PTR(-ENOMEM);
333 spin_unlock(&sb_lock);
334 up_write(&s->s_umount);
339 strlcpy(s->s_id, type->name, sizeof(s->s_id));
340 list_add_tail(&s->s_list, &super_blocks);
341 list_add(&s->s_instances, &type->fs_supers);
342 spin_unlock(&sb_lock);
343 get_filesystem(type);
349 void drop_super(struct super_block *sb)
351 up_read(&sb->s_umount);
355 EXPORT_SYMBOL(drop_super);
358 * sync_supers - helper for periodic superblock writeback
360 * Call the write_super method if present on all dirty superblocks in
361 * the system. This is for the periodic writeback used by most older
362 * filesystems. For data integrity superblock writeback use
363 * sync_filesystems() instead.
365 * Note: check the dirty flag before waiting, so we don't
366 * hold up the sync while mounting a device. (The newly
367 * mounted device won't need syncing.)
369 void sync_supers(void)
371 struct super_block *sb, *n;
374 list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
375 if (list_empty(&sb->s_instances))
377 if (sb->s_op->write_super && sb->s_dirt) {
379 spin_unlock(&sb_lock);
381 down_read(&sb->s_umount);
382 if (sb->s_root && sb->s_dirt)
383 sb->s_op->write_super(sb);
384 up_read(&sb->s_umount);
390 spin_unlock(&sb_lock);
394 * iterate_supers - call function for all active superblocks
395 * @f: function to call
396 * @arg: argument to pass to it
398 * Scans the superblock list and calls given function, passing it
399 * locked superblock and given argument.
401 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
403 struct super_block *sb, *n;
406 list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
407 if (list_empty(&sb->s_instances))
410 spin_unlock(&sb_lock);
412 down_read(&sb->s_umount);
415 up_read(&sb->s_umount);
420 spin_unlock(&sb_lock);
424 * get_super - get the superblock of a device
425 * @bdev: device to get the superblock for
427 * Scans the superblock list and finds the superblock of the file system
428 * mounted on the device given. %NULL is returned if no match is found.
431 struct super_block *get_super(struct block_device *bdev)
433 struct super_block *sb;
440 list_for_each_entry(sb, &super_blocks, s_list) {
441 if (list_empty(&sb->s_instances))
443 if (sb->s_bdev == bdev) {
445 spin_unlock(&sb_lock);
446 down_read(&sb->s_umount);
450 up_read(&sb->s_umount);
451 /* nope, got unmounted */
457 spin_unlock(&sb_lock);
461 EXPORT_SYMBOL(get_super);
464 * get_active_super - get an active reference to the superblock of a device
465 * @bdev: device to get the superblock for
467 * Scans the superblock list and finds the superblock of the file system
468 * mounted on the device given. Returns the superblock with an active
469 * reference or %NULL if none was found.
471 struct super_block *get_active_super(struct block_device *bdev)
473 struct super_block *sb;
480 list_for_each_entry(sb, &super_blocks, s_list) {
481 if (list_empty(&sb->s_instances))
483 if (sb->s_bdev == bdev) {
484 if (grab_super(sb)) /* drops sb_lock */
490 spin_unlock(&sb_lock);
494 struct super_block *user_get_super(dev_t dev)
496 struct super_block *sb;
500 list_for_each_entry(sb, &super_blocks, s_list) {
501 if (list_empty(&sb->s_instances))
503 if (sb->s_dev == dev) {
505 spin_unlock(&sb_lock);
506 down_read(&sb->s_umount);
510 up_read(&sb->s_umount);
511 /* nope, got unmounted */
517 spin_unlock(&sb_lock);
522 * do_remount_sb - asks filesystem to change mount options.
523 * @sb: superblock in question
524 * @flags: numeric part of options
525 * @data: the rest of options
526 * @force: whether or not to force the change
528 * Alters the mount options of a mounted file system.
530 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
533 int remount_rw, remount_ro;
535 if (sb->s_frozen != SB_UNFROZEN)
539 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
543 if (flags & MS_RDONLY)
545 shrink_dcache_sb(sb);
548 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
549 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
551 /* If we are remounting RDONLY and current sb is read/write,
552 make sure there are no rw files opened */
556 else if (!fs_may_remount_ro(sb))
558 retval = vfs_dq_off(sb, 1);
559 if (retval < 0 && retval != -ENOSYS)
563 if (sb->s_op->remount_fs) {
564 retval = sb->s_op->remount_fs(sb, &flags, data);
568 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
570 vfs_dq_quota_on_remount(sb);
572 * Some filesystems modify their metadata via some other path than the
573 * bdev buffer cache (eg. use a private mapping, or directories in
574 * pagecache, etc). Also file data modifications go via their own
575 * mappings. So If we try to mount readonly then copy the filesystem
576 * from bdev, we could get stale data, so invalidate it to give a best
577 * effort at coherency.
579 if (remount_ro && sb->s_bdev)
580 invalidate_bdev(sb->s_bdev);
584 static void do_emergency_remount(struct work_struct *work)
586 struct super_block *sb, *n;
589 list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
590 if (list_empty(&sb->s_instances))
593 spin_unlock(&sb_lock);
594 down_write(&sb->s_umount);
595 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
597 * What lock protects sb->s_flags??
599 do_remount_sb(sb, MS_RDONLY, NULL, 1);
601 up_write(&sb->s_umount);
605 spin_unlock(&sb_lock);
607 printk("Emergency Remount complete\n");
610 void emergency_remount(void)
612 struct work_struct *work;
614 work = kmalloc(sizeof(*work), GFP_ATOMIC);
616 INIT_WORK(work, do_emergency_remount);
622 * Unnamed block devices are dummy devices used by virtual
623 * filesystems which don't use real block-devices. -- jrs
626 static DEFINE_IDA(unnamed_dev_ida);
627 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
628 static int unnamed_dev_start = 0; /* don't bother trying below it */
630 int set_anon_super(struct super_block *s, void *data)
636 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
638 spin_lock(&unnamed_dev_lock);
639 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
641 unnamed_dev_start = dev + 1;
642 spin_unlock(&unnamed_dev_lock);
643 if (error == -EAGAIN)
644 /* We raced and lost with another CPU. */
649 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
650 spin_lock(&unnamed_dev_lock);
651 ida_remove(&unnamed_dev_ida, dev);
652 if (unnamed_dev_start > dev)
653 unnamed_dev_start = dev;
654 spin_unlock(&unnamed_dev_lock);
657 s->s_dev = MKDEV(0, dev & MINORMASK);
658 s->s_bdi = &noop_backing_dev_info;
662 EXPORT_SYMBOL(set_anon_super);
664 void kill_anon_super(struct super_block *sb)
666 int slot = MINOR(sb->s_dev);
668 generic_shutdown_super(sb);
669 spin_lock(&unnamed_dev_lock);
670 ida_remove(&unnamed_dev_ida, slot);
671 if (slot < unnamed_dev_start)
672 unnamed_dev_start = slot;
673 spin_unlock(&unnamed_dev_lock);
676 EXPORT_SYMBOL(kill_anon_super);
678 void kill_litter_super(struct super_block *sb)
681 d_genocide(sb->s_root);
685 EXPORT_SYMBOL(kill_litter_super);
687 static int ns_test_super(struct super_block *sb, void *data)
689 return sb->s_fs_info == data;
692 static int ns_set_super(struct super_block *sb, void *data)
694 sb->s_fs_info = data;
695 return set_anon_super(sb, NULL);
698 int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
699 int (*fill_super)(struct super_block *, void *, int),
700 struct vfsmount *mnt)
702 struct super_block *sb;
704 sb = sget(fs_type, ns_test_super, ns_set_super, data);
711 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
713 deactivate_locked_super(sb);
717 sb->s_flags |= MS_ACTIVE;
720 simple_set_mnt(mnt, sb);
724 EXPORT_SYMBOL(get_sb_ns);
727 static int set_bdev_super(struct super_block *s, void *data)
730 s->s_dev = s->s_bdev->bd_dev;
733 * We set the bdi here to the queue backing, file systems can
734 * overwrite this in ->fill_super()
736 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
740 static int test_bdev_super(struct super_block *s, void *data)
742 return (void *)s->s_bdev == data;
745 int get_sb_bdev(struct file_system_type *fs_type,
746 int flags, const char *dev_name, void *data,
747 int (*fill_super)(struct super_block *, void *, int),
748 struct vfsmount *mnt)
750 struct block_device *bdev;
751 struct super_block *s;
752 fmode_t mode = FMODE_READ;
755 if (!(flags & MS_RDONLY))
758 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
760 return PTR_ERR(bdev);
763 * once the super is inserted into the list by sget, s_umount
764 * will protect the lockfs code from trying to start a snapshot
765 * while we are mounting
767 mutex_lock(&bdev->bd_fsfreeze_mutex);
768 if (bdev->bd_fsfreeze_count > 0) {
769 mutex_unlock(&bdev->bd_fsfreeze_mutex);
773 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
774 mutex_unlock(&bdev->bd_fsfreeze_mutex);
779 if ((flags ^ s->s_flags) & MS_RDONLY) {
780 deactivate_locked_super(s);
785 close_bdev_exclusive(bdev, mode);
787 char b[BDEVNAME_SIZE];
791 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
792 sb_set_blocksize(s, block_size(bdev));
793 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
795 deactivate_locked_super(s);
799 s->s_flags |= MS_ACTIVE;
803 simple_set_mnt(mnt, s);
809 close_bdev_exclusive(bdev, mode);
814 EXPORT_SYMBOL(get_sb_bdev);
816 void kill_block_super(struct super_block *sb)
818 struct block_device *bdev = sb->s_bdev;
819 fmode_t mode = sb->s_mode;
821 bdev->bd_super = NULL;
822 generic_shutdown_super(sb);
824 close_bdev_exclusive(bdev, mode);
827 EXPORT_SYMBOL(kill_block_super);
830 int get_sb_nodev(struct file_system_type *fs_type,
831 int flags, void *data,
832 int (*fill_super)(struct super_block *, void *, int),
833 struct vfsmount *mnt)
836 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
843 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
845 deactivate_locked_super(s);
848 s->s_flags |= MS_ACTIVE;
849 simple_set_mnt(mnt, s);
853 EXPORT_SYMBOL(get_sb_nodev);
855 static int compare_single(struct super_block *s, void *p)
860 int get_sb_single(struct file_system_type *fs_type,
861 int flags, void *data,
862 int (*fill_super)(struct super_block *, void *, int),
863 struct vfsmount *mnt)
865 struct super_block *s;
868 s = sget(fs_type, compare_single, set_anon_super, NULL);
873 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
875 deactivate_locked_super(s);
878 s->s_flags |= MS_ACTIVE;
880 do_remount_sb(s, flags, data, 0);
882 simple_set_mnt(mnt, s);
886 EXPORT_SYMBOL(get_sb_single);
889 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
891 struct vfsmount *mnt;
892 char *secdata = NULL;
896 return ERR_PTR(-ENODEV);
899 mnt = alloc_vfsmnt(name);
903 if (flags & MS_KERNMOUNT)
904 mnt->mnt_flags = MNT_INTERNAL;
906 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
907 secdata = alloc_secdata();
911 error = security_sb_copy_data(data, secdata);
913 goto out_free_secdata;
916 error = type->get_sb(type, flags, name, data, mnt);
918 goto out_free_secdata;
919 BUG_ON(!mnt->mnt_sb);
920 WARN_ON(!mnt->mnt_sb->s_bdi);
922 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
927 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
928 * but s_maxbytes was an unsigned long long for many releases. Throw
929 * this warning for a little while to try and catch filesystems that
930 * violate this rule. This warning should be either removed or
931 * converted to a BUG() in 2.6.34.
933 WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
934 "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
936 mnt->mnt_mountpoint = mnt->mnt_root;
937 mnt->mnt_parent = mnt;
938 up_write(&mnt->mnt_sb->s_umount);
939 free_secdata(secdata);
943 deactivate_locked_super(mnt->mnt_sb);
945 free_secdata(secdata);
949 return ERR_PTR(error);
952 EXPORT_SYMBOL_GPL(vfs_kern_mount);
954 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
957 const char *subtype = strchr(fstype, '.');
966 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
968 if (!mnt->mnt_sb->s_subtype)
978 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
980 struct file_system_type *type = get_fs_type(fstype);
981 struct vfsmount *mnt;
983 return ERR_PTR(-ENODEV);
984 mnt = vfs_kern_mount(type, flags, name, data);
985 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
986 !mnt->mnt_sb->s_subtype)
987 mnt = fs_set_subtype(mnt, fstype);
988 put_filesystem(type);
991 EXPORT_SYMBOL_GPL(do_kern_mount);
993 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
995 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
998 EXPORT_SYMBOL_GPL(kern_mount_data);