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 <linux/async.h>
42 #include <asm/uaccess.h>
46 LIST_HEAD(super_blocks);
47 DEFINE_SPINLOCK(sb_lock);
50 * alloc_super - create new superblock
51 * @type: filesystem type superblock should belong to
53 * Allocates and initializes a new &struct super_block. alloc_super()
54 * returns a pointer new superblock or %NULL if allocation had failed.
56 static struct super_block *alloc_super(struct file_system_type *type)
58 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
59 static struct super_operations default_op;
62 if (security_sb_alloc(s)) {
67 INIT_LIST_HEAD(&s->s_dirty);
68 INIT_LIST_HEAD(&s->s_io);
69 INIT_LIST_HEAD(&s->s_more_io);
70 INIT_LIST_HEAD(&s->s_files);
71 INIT_LIST_HEAD(&s->s_instances);
72 INIT_HLIST_HEAD(&s->s_anon);
73 INIT_LIST_HEAD(&s->s_inodes);
74 INIT_LIST_HEAD(&s->s_dentry_lru);
75 INIT_LIST_HEAD(&s->s_async_list);
76 init_rwsem(&s->s_umount);
77 mutex_init(&s->s_lock);
78 lockdep_set_class(&s->s_umount, &type->s_umount_key);
80 * The locking rules for s_lock are up to the
81 * filesystem. For example ext3fs has different
82 * lock ordering than usbfs:
84 lockdep_set_class(&s->s_lock, &type->s_lock_key);
86 * sget() can have s_umount recursion.
88 * When it cannot find a suitable sb, it allocates a new
89 * one (this one), and tries again to find a suitable old
92 * In case that succeeds, it will acquire the s_umount
93 * lock of the old one. Since these are clearly distrinct
94 * locks, and this object isn't exposed yet, there's no
97 * Annotate this by putting this lock in a different
100 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
102 atomic_set(&s->s_active, 1);
103 mutex_init(&s->s_vfs_rename_mutex);
104 mutex_init(&s->s_dquot.dqio_mutex);
105 mutex_init(&s->s_dquot.dqonoff_mutex);
106 init_rwsem(&s->s_dquot.dqptr_sem);
107 init_waitqueue_head(&s->s_wait_unfrozen);
108 s->s_maxbytes = MAX_NON_LFS;
109 s->dq_op = sb_dquot_ops;
110 s->s_qcop = sb_quotactl_ops;
111 s->s_op = &default_op;
112 s->s_time_gran = 1000000000;
119 * destroy_super - frees a superblock
120 * @s: superblock to free
122 * Frees a superblock.
124 static inline void destroy_super(struct super_block *s)
132 /* Superblock refcounting */
135 * Drop a superblock's refcount. Returns non-zero if the superblock was
136 * destroyed. The caller must hold sb_lock.
138 static int __put_super(struct super_block *sb)
142 if (!--sb->s_count) {
150 * Drop a superblock's refcount.
151 * Returns non-zero if the superblock is about to be destroyed and
152 * at least is already removed from super_blocks list, so if we are
153 * making a loop through super blocks then we need to restart.
154 * The caller must hold sb_lock.
156 int __put_super_and_need_restart(struct super_block *sb)
158 /* check for race with generic_shutdown_super() */
159 if (list_empty(&sb->s_list)) {
160 /* super block is removed, need to restart... */
164 /* can't be the last, since s_list is still in use */
166 BUG_ON(sb->s_count == 0);
171 * put_super - drop a temporary reference to superblock
172 * @sb: superblock in question
174 * Drops a temporary reference, frees superblock if there's no
177 static void put_super(struct super_block *sb)
181 spin_unlock(&sb_lock);
186 * deactivate_super - drop an active reference to superblock
187 * @s: superblock to deactivate
189 * Drops an active reference to superblock, acquiring a temprory one if
190 * there is no active references left. In that case we lock superblock,
191 * tell fs driver to shut it down and drop the temporary reference we
194 void deactivate_super(struct super_block *s)
196 struct file_system_type *fs = s->s_type;
197 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
198 s->s_count -= S_BIAS-1;
199 spin_unlock(&sb_lock);
201 down_write(&s->s_umount);
208 EXPORT_SYMBOL(deactivate_super);
211 * grab_super - acquire an active reference
212 * @s: reference we are trying to make active
214 * Tries to acquire an active reference. grab_super() is used when we
215 * had just found a superblock in super_blocks or fs_type->fs_supers
216 * and want to turn it into a full-blown active reference. grab_super()
217 * is called with sb_lock held and drops it. Returns 1 in case of
218 * success, 0 if we had failed (superblock contents was already dead or
219 * dying when grab_super() had been called).
221 static int grab_super(struct super_block *s) __releases(sb_lock)
224 spin_unlock(&sb_lock);
225 down_write(&s->s_umount);
228 if (s->s_count > S_BIAS) {
229 atomic_inc(&s->s_active);
231 spin_unlock(&sb_lock);
234 spin_unlock(&sb_lock);
236 up_write(&s->s_umount);
243 * Superblock locking. We really ought to get rid of these two.
245 void lock_super(struct super_block * sb)
248 mutex_lock(&sb->s_lock);
251 void unlock_super(struct super_block * sb)
254 mutex_unlock(&sb->s_lock);
257 EXPORT_SYMBOL(lock_super);
258 EXPORT_SYMBOL(unlock_super);
261 * Write out and wait upon all dirty data associated with this
262 * superblock. Filesystem data as well as the underlying block
263 * device. Takes the superblock lock. Requires a second blkdev
264 * flush by the caller to complete the operation.
266 void __fsync_super(struct super_block *sb)
268 sync_inodes_sb(sb, 0);
271 if (sb->s_dirt && sb->s_op->write_super)
272 sb->s_op->write_super(sb);
274 if (sb->s_op->sync_fs)
275 sb->s_op->sync_fs(sb, 1);
276 sync_blockdev(sb->s_bdev);
277 sync_inodes_sb(sb, 1);
281 * Write out and wait upon all dirty data associated with this
282 * superblock. Filesystem data as well as the underlying block
283 * device. Takes the superblock lock.
285 int fsync_super(struct super_block *sb)
288 return sync_blockdev(sb->s_bdev);
292 * generic_shutdown_super - common helper for ->kill_sb()
293 * @sb: superblock to kill
295 * generic_shutdown_super() does all fs-independent work on superblock
296 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
297 * that need destruction out of superblock, call generic_shutdown_super()
298 * and release aforementioned objects. Note: dentries and inodes _are_
299 * taken care of and do not need specific handling.
301 * Upon calling this function, the filesystem may no longer alter or
302 * rearrange the set of dentries belonging to this super_block, nor may it
303 * change the attachments of dentries to inodes.
305 void generic_shutdown_super(struct super_block *sb)
307 const struct super_operations *sop = sb->s_op;
311 shrink_dcache_for_umount(sb);
314 sb->s_flags &= ~MS_ACTIVE;
317 * wait for asynchronous fs operations to finish before going further
319 async_synchronize_full_domain(&sb->s_async_list);
321 /* bad name - it should be evict_inodes() */
322 invalidate_inodes(sb);
325 if (sop->write_super && sb->s_dirt)
326 sop->write_super(sb);
330 /* Forget any remaining inodes */
331 if (invalidate_inodes(sb)) {
332 printk("VFS: Busy inodes after unmount of %s. "
333 "Self-destruct in 5 seconds. Have a nice day...\n",
341 /* should be initialized for __put_super_and_need_restart() */
342 list_del_init(&sb->s_list);
343 list_del(&sb->s_instances);
344 spin_unlock(&sb_lock);
345 up_write(&sb->s_umount);
348 EXPORT_SYMBOL(generic_shutdown_super);
351 * sget - find or create a superblock
352 * @type: filesystem type superblock should belong to
353 * @test: comparison callback
354 * @set: setup callback
355 * @data: argument to each of them
357 struct super_block *sget(struct file_system_type *type,
358 int (*test)(struct super_block *,void *),
359 int (*set)(struct super_block *,void *),
362 struct super_block *s = NULL;
363 struct super_block *old;
369 list_for_each_entry(old, &type->fs_supers, s_instances) {
370 if (!test(old, data))
372 if (!grab_super(old))
375 up_write(&s->s_umount);
382 spin_unlock(&sb_lock);
383 s = alloc_super(type);
385 return ERR_PTR(-ENOMEM);
391 spin_unlock(&sb_lock);
392 up_write(&s->s_umount);
397 strlcpy(s->s_id, type->name, sizeof(s->s_id));
398 list_add_tail(&s->s_list, &super_blocks);
399 list_add(&s->s_instances, &type->fs_supers);
400 spin_unlock(&sb_lock);
401 get_filesystem(type);
407 void drop_super(struct super_block *sb)
409 up_read(&sb->s_umount);
413 EXPORT_SYMBOL(drop_super);
415 static inline void write_super(struct super_block *sb)
418 if (sb->s_root && sb->s_dirt)
419 if (sb->s_op->write_super)
420 sb->s_op->write_super(sb);
425 * Note: check the dirty flag before waiting, so we don't
426 * hold up the sync while mounting a device. (The newly
427 * mounted device won't need syncing.)
429 void sync_supers(void)
431 struct super_block *sb;
435 list_for_each_entry(sb, &super_blocks, s_list) {
438 spin_unlock(&sb_lock);
439 down_read(&sb->s_umount);
441 up_read(&sb->s_umount);
443 if (__put_super_and_need_restart(sb))
447 spin_unlock(&sb_lock);
451 * Call the ->sync_fs super_op against all filesystems which are r/w and
452 * which implement it.
454 * This operation is careful to avoid the livelock which could easily happen
455 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
456 * is used only here. We set it against all filesystems and then clear it as
457 * we sync them. So redirtied filesystems are skipped.
459 * But if process A is currently running sync_filesystems and then process B
460 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
461 * flags again, which will cause process A to resync everything. Fix that with
464 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
466 void sync_filesystems(int wait)
468 struct super_block *sb;
469 static DEFINE_MUTEX(mutex);
471 mutex_lock(&mutex); /* Could be down_interruptible */
473 list_for_each_entry(sb, &super_blocks, s_list) {
474 if (!sb->s_op->sync_fs)
476 if (sb->s_flags & MS_RDONLY)
478 sb->s_need_sync_fs = 1;
482 list_for_each_entry(sb, &super_blocks, s_list) {
483 if (!sb->s_need_sync_fs)
485 sb->s_need_sync_fs = 0;
486 if (sb->s_flags & MS_RDONLY)
487 continue; /* hm. Was remounted r/o meanwhile */
489 spin_unlock(&sb_lock);
490 down_read(&sb->s_umount);
491 async_synchronize_full_domain(&sb->s_async_list);
492 if (sb->s_root && (wait || sb->s_dirt))
493 sb->s_op->sync_fs(sb, wait);
494 up_read(&sb->s_umount);
495 /* restart only when sb is no longer on the list */
497 if (__put_super_and_need_restart(sb))
500 spin_unlock(&sb_lock);
501 mutex_unlock(&mutex);
505 * get_super - get the superblock of a device
506 * @bdev: device to get the superblock for
508 * Scans the superblock list and finds the superblock of the file system
509 * mounted on the device given. %NULL is returned if no match is found.
512 struct super_block * get_super(struct block_device *bdev)
514 struct super_block *sb;
521 list_for_each_entry(sb, &super_blocks, s_list) {
522 if (sb->s_bdev == bdev) {
524 spin_unlock(&sb_lock);
525 down_read(&sb->s_umount);
528 up_read(&sb->s_umount);
529 /* restart only when sb is no longer on the list */
531 if (__put_super_and_need_restart(sb))
535 spin_unlock(&sb_lock);
539 EXPORT_SYMBOL(get_super);
541 struct super_block * user_get_super(dev_t dev)
543 struct super_block *sb;
547 list_for_each_entry(sb, &super_blocks, s_list) {
548 if (sb->s_dev == dev) {
550 spin_unlock(&sb_lock);
551 down_read(&sb->s_umount);
554 up_read(&sb->s_umount);
555 /* restart only when sb is no longer on the list */
557 if (__put_super_and_need_restart(sb))
561 spin_unlock(&sb_lock);
565 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
567 struct super_block *s;
572 s = user_get_super(new_decode_dev(dev));
575 err = vfs_statfs(s->s_root, &sbuf);
580 memset(&tmp,0,sizeof(struct ustat));
581 tmp.f_tfree = sbuf.f_bfree;
582 tmp.f_tinode = sbuf.f_ffree;
584 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
590 * mark_files_ro - mark all files read-only
591 * @sb: superblock in question
593 * All files are marked read-only. We don't care about pending
594 * delete files so this should be used in 'force' mode only.
597 static void mark_files_ro(struct super_block *sb)
603 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
604 struct vfsmount *mnt;
605 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
609 if (!(f->f_mode & FMODE_WRITE))
611 f->f_mode &= ~FMODE_WRITE;
612 if (file_check_writeable(f) != 0)
614 file_release_write(f);
615 mnt = mntget(f->f_path.mnt);
618 * This can sleep, so we can't hold
619 * the file_list_lock() spinlock.
629 * do_remount_sb - asks filesystem to change mount options.
630 * @sb: superblock in question
631 * @flags: numeric part of options
632 * @data: the rest of options
633 * @force: whether or not to force the change
635 * Alters the mount options of a mounted file system.
637 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
643 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
646 if (flags & MS_RDONLY)
648 shrink_dcache_sb(sb);
651 /* If we are remounting RDONLY and current sb is read/write,
652 make sure there are no rw files opened */
653 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
656 else if (!fs_may_remount_ro(sb))
658 retval = vfs_dq_off(sb, 1);
659 if (retval < 0 && retval != -ENOSYS)
662 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
664 if (sb->s_op->remount_fs) {
666 retval = sb->s_op->remount_fs(sb, &flags, data);
671 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
673 vfs_dq_quota_on_remount(sb);
677 static void do_emergency_remount(struct work_struct *work)
679 struct super_block *sb;
682 list_for_each_entry(sb, &super_blocks, s_list) {
684 spin_unlock(&sb_lock);
685 down_read(&sb->s_umount);
686 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
688 * ->remount_fs needs lock_kernel().
690 * What lock protects sb->s_flags??
693 do_remount_sb(sb, MS_RDONLY, NULL, 1);
699 spin_unlock(&sb_lock);
701 printk("Emergency Remount complete\n");
704 void emergency_remount(void)
706 struct work_struct *work;
708 work = kmalloc(sizeof(*work), GFP_ATOMIC);
710 INIT_WORK(work, do_emergency_remount);
716 * Unnamed block devices are dummy devices used by virtual
717 * filesystems which don't use real block-devices. -- jrs
720 static DEFINE_IDA(unnamed_dev_ida);
721 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
723 int set_anon_super(struct super_block *s, void *data)
729 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
731 spin_lock(&unnamed_dev_lock);
732 error = ida_get_new(&unnamed_dev_ida, &dev);
733 spin_unlock(&unnamed_dev_lock);
734 if (error == -EAGAIN)
735 /* We raced and lost with another CPU. */
740 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
741 spin_lock(&unnamed_dev_lock);
742 ida_remove(&unnamed_dev_ida, dev);
743 spin_unlock(&unnamed_dev_lock);
746 s->s_dev = MKDEV(0, dev & MINORMASK);
750 EXPORT_SYMBOL(set_anon_super);
752 void kill_anon_super(struct super_block *sb)
754 int slot = MINOR(sb->s_dev);
756 generic_shutdown_super(sb);
757 spin_lock(&unnamed_dev_lock);
758 ida_remove(&unnamed_dev_ida, slot);
759 spin_unlock(&unnamed_dev_lock);
762 EXPORT_SYMBOL(kill_anon_super);
764 void kill_litter_super(struct super_block *sb)
767 d_genocide(sb->s_root);
771 EXPORT_SYMBOL(kill_litter_super);
774 static int set_bdev_super(struct super_block *s, void *data)
777 s->s_dev = s->s_bdev->bd_dev;
781 static int test_bdev_super(struct super_block *s, void *data)
783 return (void *)s->s_bdev == data;
786 int get_sb_bdev(struct file_system_type *fs_type,
787 int flags, const char *dev_name, void *data,
788 int (*fill_super)(struct super_block *, void *, int),
789 struct vfsmount *mnt)
791 struct block_device *bdev;
792 struct super_block *s;
793 fmode_t mode = FMODE_READ;
796 if (!(flags & MS_RDONLY))
799 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
801 return PTR_ERR(bdev);
804 * once the super is inserted into the list by sget, s_umount
805 * will protect the lockfs code from trying to start a snapshot
806 * while we are mounting
808 down(&bdev->bd_mount_sem);
809 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
810 up(&bdev->bd_mount_sem);
815 if ((flags ^ s->s_flags) & MS_RDONLY) {
816 up_write(&s->s_umount);
822 close_bdev_exclusive(bdev, mode);
824 char b[BDEVNAME_SIZE];
828 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
829 sb_set_blocksize(s, block_size(bdev));
830 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
832 up_write(&s->s_umount);
837 s->s_flags |= MS_ACTIVE;
841 simple_set_mnt(mnt, s);
847 close_bdev_exclusive(bdev, mode);
852 EXPORT_SYMBOL(get_sb_bdev);
854 void kill_block_super(struct super_block *sb)
856 struct block_device *bdev = sb->s_bdev;
857 fmode_t mode = sb->s_mode;
860 generic_shutdown_super(sb);
862 close_bdev_exclusive(bdev, mode);
865 EXPORT_SYMBOL(kill_block_super);
868 int get_sb_nodev(struct file_system_type *fs_type,
869 int flags, void *data,
870 int (*fill_super)(struct super_block *, void *, int),
871 struct vfsmount *mnt)
874 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
881 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
883 up_write(&s->s_umount);
887 s->s_flags |= MS_ACTIVE;
888 simple_set_mnt(mnt, s);
892 EXPORT_SYMBOL(get_sb_nodev);
894 static int compare_single(struct super_block *s, void *p)
899 int get_sb_single(struct file_system_type *fs_type,
900 int flags, void *data,
901 int (*fill_super)(struct super_block *, void *, int),
902 struct vfsmount *mnt)
904 struct super_block *s;
907 s = sget(fs_type, compare_single, set_anon_super, NULL);
912 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
914 up_write(&s->s_umount);
918 s->s_flags |= MS_ACTIVE;
920 do_remount_sb(s, flags, data, 0);
921 simple_set_mnt(mnt, s);
925 EXPORT_SYMBOL(get_sb_single);
928 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
930 struct vfsmount *mnt;
931 char *secdata = NULL;
935 return ERR_PTR(-ENODEV);
938 mnt = alloc_vfsmnt(name);
942 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
943 secdata = alloc_secdata();
947 error = security_sb_copy_data(data, secdata);
949 goto out_free_secdata;
952 error = type->get_sb(type, flags, name, data, mnt);
954 goto out_free_secdata;
955 BUG_ON(!mnt->mnt_sb);
957 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
961 mnt->mnt_mountpoint = mnt->mnt_root;
962 mnt->mnt_parent = mnt;
963 up_write(&mnt->mnt_sb->s_umount);
964 free_secdata(secdata);
968 up_write(&mnt->mnt_sb->s_umount);
969 deactivate_super(mnt->mnt_sb);
971 free_secdata(secdata);
975 return ERR_PTR(error);
978 EXPORT_SYMBOL_GPL(vfs_kern_mount);
980 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
983 const char *subtype = strchr(fstype, '.');
992 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
994 if (!mnt->mnt_sb->s_subtype)
1000 return ERR_PTR(err);
1004 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1006 struct file_system_type *type = get_fs_type(fstype);
1007 struct vfsmount *mnt;
1009 return ERR_PTR(-ENODEV);
1010 mnt = vfs_kern_mount(type, flags, name, data);
1011 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1012 !mnt->mnt_sb->s_subtype)
1013 mnt = fs_set_subtype(mnt, fstype);
1014 put_filesystem(type);
1017 EXPORT_SYMBOL_GPL(do_kern_mount);
1019 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
1021 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
1024 EXPORT_SYMBOL_GPL(kern_mount_data);