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 <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 struct super_operations default_op;
61 if (security_sb_alloc(s)) {
66 INIT_LIST_HEAD(&s->s_dirty);
67 INIT_LIST_HEAD(&s->s_io);
68 INIT_LIST_HEAD(&s->s_more_io);
69 INIT_LIST_HEAD(&s->s_files);
70 INIT_LIST_HEAD(&s->s_instances);
71 INIT_HLIST_HEAD(&s->s_anon);
72 INIT_LIST_HEAD(&s->s_inodes);
73 INIT_LIST_HEAD(&s->s_dentry_lru);
74 init_rwsem(&s->s_umount);
75 mutex_init(&s->s_lock);
76 lockdep_set_class(&s->s_umount, &type->s_umount_key);
78 * The locking rules for s_lock are up to the
79 * filesystem. For example ext3fs has different
80 * lock ordering than usbfs:
82 lockdep_set_class(&s->s_lock, &type->s_lock_key);
83 down_write(&s->s_umount);
85 atomic_set(&s->s_active, 1);
86 mutex_init(&s->s_vfs_rename_mutex);
87 mutex_init(&s->s_dquot.dqio_mutex);
88 mutex_init(&s->s_dquot.dqonoff_mutex);
89 init_rwsem(&s->s_dquot.dqptr_sem);
90 init_waitqueue_head(&s->s_wait_unfrozen);
91 s->s_maxbytes = MAX_NON_LFS;
92 s->dq_op = sb_dquot_ops;
93 s->s_qcop = sb_quotactl_ops;
94 s->s_op = &default_op;
95 s->s_time_gran = 1000000000;
102 * destroy_super - frees a superblock
103 * @s: superblock to free
105 * Frees a superblock.
107 static inline void destroy_super(struct super_block *s)
115 /* Superblock refcounting */
118 * Drop a superblock's refcount. Returns non-zero if the superblock was
119 * destroyed. The caller must hold sb_lock.
121 static int __put_super(struct super_block *sb)
125 if (!--sb->s_count) {
133 * Drop a superblock's refcount.
134 * Returns non-zero if the superblock is about to be destroyed and
135 * at least is already removed from super_blocks list, so if we are
136 * making a loop through super blocks then we need to restart.
137 * The caller must hold sb_lock.
139 int __put_super_and_need_restart(struct super_block *sb)
141 /* check for race with generic_shutdown_super() */
142 if (list_empty(&sb->s_list)) {
143 /* super block is removed, need to restart... */
147 /* can't be the last, since s_list is still in use */
149 BUG_ON(sb->s_count == 0);
154 * put_super - drop a temporary reference to superblock
155 * @sb: superblock in question
157 * Drops a temporary reference, frees superblock if there's no
160 static void put_super(struct super_block *sb)
164 spin_unlock(&sb_lock);
169 * deactivate_super - drop an active reference to superblock
170 * @s: superblock to deactivate
172 * Drops an active reference to superblock, acquiring a temprory one if
173 * there is no active references left. In that case we lock superblock,
174 * tell fs driver to shut it down and drop the temporary reference we
177 void deactivate_super(struct super_block *s)
179 struct file_system_type *fs = s->s_type;
180 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
181 s->s_count -= S_BIAS-1;
182 spin_unlock(&sb_lock);
184 down_write(&s->s_umount);
191 EXPORT_SYMBOL(deactivate_super);
194 * grab_super - acquire an active reference
195 * @s: reference we are trying to make active
197 * Tries to acquire an active reference. grab_super() is used when we
198 * had just found a superblock in super_blocks or fs_type->fs_supers
199 * and want to turn it into a full-blown active reference. grab_super()
200 * is called with sb_lock held and drops it. Returns 1 in case of
201 * success, 0 if we had failed (superblock contents was already dead or
202 * dying when grab_super() had been called).
204 static int grab_super(struct super_block *s) __releases(sb_lock)
207 spin_unlock(&sb_lock);
208 down_write(&s->s_umount);
211 if (s->s_count > S_BIAS) {
212 atomic_inc(&s->s_active);
214 spin_unlock(&sb_lock);
217 spin_unlock(&sb_lock);
219 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 * Write out and wait upon all dirty data associated with this
245 * superblock. Filesystem data as well as the underlying block
246 * device. Takes the superblock lock. Requires a second blkdev
247 * flush by the caller to complete the operation.
249 void __fsync_super(struct super_block *sb)
251 sync_inodes_sb(sb, 0);
254 if (sb->s_dirt && sb->s_op->write_super)
255 sb->s_op->write_super(sb);
257 if (sb->s_op->sync_fs)
258 sb->s_op->sync_fs(sb, 1);
259 sync_blockdev(sb->s_bdev);
260 sync_inodes_sb(sb, 1);
264 * Write out and wait upon all dirty data associated with this
265 * superblock. Filesystem data as well as the underlying block
266 * device. Takes the superblock lock.
268 int fsync_super(struct super_block *sb)
271 return sync_blockdev(sb->s_bdev);
275 * generic_shutdown_super - common helper for ->kill_sb()
276 * @sb: superblock to kill
278 * generic_shutdown_super() does all fs-independent work on superblock
279 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
280 * that need destruction out of superblock, call generic_shutdown_super()
281 * and release aforementioned objects. Note: dentries and inodes _are_
282 * taken care of and do not need specific handling.
284 * Upon calling this function, the filesystem may no longer alter or
285 * rearrange the set of dentries belonging to this super_block, nor may it
286 * change the attachments of dentries to inodes.
288 void generic_shutdown_super(struct super_block *sb)
290 const struct super_operations *sop = sb->s_op;
293 shrink_dcache_for_umount(sb);
296 sb->s_flags &= ~MS_ACTIVE;
297 /* bad name - it should be evict_inodes() */
298 invalidate_inodes(sb);
301 if (sop->write_super && sb->s_dirt)
302 sop->write_super(sb);
306 /* Forget any remaining inodes */
307 if (invalidate_inodes(sb)) {
308 printk("VFS: Busy inodes after unmount of %s. "
309 "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))
356 spin_unlock(&sb_lock);
357 s = alloc_super(type);
359 return ERR_PTR(-ENOMEM);
365 spin_unlock(&sb_lock);
370 strlcpy(s->s_id, type->name, sizeof(s->s_id));
371 list_add_tail(&s->s_list, &super_blocks);
372 list_add(&s->s_instances, &type->fs_supers);
373 spin_unlock(&sb_lock);
374 get_filesystem(type);
380 void drop_super(struct super_block *sb)
382 up_read(&sb->s_umount);
386 EXPORT_SYMBOL(drop_super);
388 static inline void write_super(struct super_block *sb)
391 if (sb->s_root && sb->s_dirt)
392 if (sb->s_op->write_super)
393 sb->s_op->write_super(sb);
398 * Note: check the dirty flag before waiting, so we don't
399 * hold up the sync while mounting a device. (The newly
400 * mounted device won't need syncing.)
402 void sync_supers(void)
404 struct super_block *sb;
408 list_for_each_entry(sb, &super_blocks, s_list) {
411 spin_unlock(&sb_lock);
412 down_read(&sb->s_umount);
414 up_read(&sb->s_umount);
416 if (__put_super_and_need_restart(sb))
420 spin_unlock(&sb_lock);
424 * Call the ->sync_fs super_op against all filesystems which are r/w and
425 * which implement it.
427 * This operation is careful to avoid the livelock which could easily happen
428 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
429 * is used only here. We set it against all filesystems and then clear it as
430 * we sync them. So redirtied filesystems are skipped.
432 * But if process A is currently running sync_filesystems and then process B
433 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
434 * flags again, which will cause process A to resync everything. Fix that with
437 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
439 void sync_filesystems(int wait)
441 struct super_block *sb;
442 static DEFINE_MUTEX(mutex);
444 mutex_lock(&mutex); /* Could be down_interruptible */
446 list_for_each_entry(sb, &super_blocks, s_list) {
447 if (!sb->s_op->sync_fs)
449 if (sb->s_flags & MS_RDONLY)
451 sb->s_need_sync_fs = 1;
455 list_for_each_entry(sb, &super_blocks, s_list) {
456 if (!sb->s_need_sync_fs)
458 sb->s_need_sync_fs = 0;
459 if (sb->s_flags & MS_RDONLY)
460 continue; /* hm. Was remounted r/o meanwhile */
462 spin_unlock(&sb_lock);
463 down_read(&sb->s_umount);
464 if (sb->s_root && (wait || sb->s_dirt))
465 sb->s_op->sync_fs(sb, wait);
466 up_read(&sb->s_umount);
467 /* restart only when sb is no longer on the list */
469 if (__put_super_and_need_restart(sb))
472 spin_unlock(&sb_lock);
473 mutex_unlock(&mutex);
477 * get_super - get the superblock of a device
478 * @bdev: device to get the superblock for
480 * Scans the superblock list and finds the superblock of the file system
481 * mounted on the device given. %NULL is returned if no match is found.
484 struct super_block * get_super(struct block_device *bdev)
486 struct super_block *sb;
493 list_for_each_entry(sb, &super_blocks, s_list) {
494 if (sb->s_bdev == bdev) {
496 spin_unlock(&sb_lock);
497 down_read(&sb->s_umount);
500 up_read(&sb->s_umount);
501 /* restart only when sb is no longer on the list */
503 if (__put_super_and_need_restart(sb))
507 spin_unlock(&sb_lock);
511 EXPORT_SYMBOL(get_super);
513 struct super_block * user_get_super(dev_t dev)
515 struct super_block *sb;
519 list_for_each_entry(sb, &super_blocks, s_list) {
520 if (sb->s_dev == dev) {
522 spin_unlock(&sb_lock);
523 down_read(&sb->s_umount);
526 up_read(&sb->s_umount);
527 /* restart only when sb is no longer on the list */
529 if (__put_super_and_need_restart(sb))
533 spin_unlock(&sb_lock);
537 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
539 struct super_block *s;
544 s = user_get_super(new_decode_dev(dev));
547 err = vfs_statfs(s->s_root, &sbuf);
552 memset(&tmp,0,sizeof(struct ustat));
553 tmp.f_tfree = sbuf.f_bfree;
554 tmp.f_tinode = sbuf.f_ffree;
556 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
562 * mark_files_ro - mark all files read-only
563 * @sb: superblock in question
565 * All files are marked read-only. We don't care about pending
566 * delete files so this should be used in 'force' mode only.
569 static void mark_files_ro(struct super_block *sb)
575 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
576 struct vfsmount *mnt;
577 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
581 if (!(f->f_mode & FMODE_WRITE))
583 f->f_mode &= ~FMODE_WRITE;
584 if (file_check_writeable(f) != 0)
586 file_release_write(f);
587 mnt = mntget(f->f_path.mnt);
590 * This can sleep, so we can't hold
591 * the file_list_lock() spinlock.
601 * do_remount_sb - asks filesystem to change mount options.
602 * @sb: superblock in question
603 * @flags: numeric part of options
604 * @data: the rest of options
605 * @force: whether or not to force the change
607 * Alters the mount options of a mounted file system.
609 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
615 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
618 if (flags & MS_RDONLY)
620 shrink_dcache_sb(sb);
623 /* If we are remounting RDONLY and current sb is read/write,
624 make sure there are no rw files opened */
625 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
628 else if (!fs_may_remount_ro(sb))
630 retval = DQUOT_OFF(sb, 1);
631 if (retval < 0 && retval != -ENOSYS)
634 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
636 if (sb->s_op->remount_fs) {
638 retval = sb->s_op->remount_fs(sb, &flags, data);
643 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
645 DQUOT_ON_REMOUNT(sb);
649 static void do_emergency_remount(unsigned long foo)
651 struct super_block *sb;
654 list_for_each_entry(sb, &super_blocks, s_list) {
656 spin_unlock(&sb_lock);
657 down_read(&sb->s_umount);
658 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
660 * ->remount_fs needs lock_kernel().
662 * What lock protects sb->s_flags??
665 do_remount_sb(sb, MS_RDONLY, NULL, 1);
671 spin_unlock(&sb_lock);
672 printk("Emergency Remount complete\n");
675 void emergency_remount(void)
677 pdflush_operation(do_emergency_remount, 0);
681 * Unnamed block devices are dummy devices used by virtual
682 * filesystems which don't use real block-devices. -- jrs
685 static DEFINE_IDA(unnamed_dev_ida);
686 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
688 int set_anon_super(struct super_block *s, void *data)
694 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
696 spin_lock(&unnamed_dev_lock);
697 error = ida_get_new(&unnamed_dev_ida, &dev);
698 spin_unlock(&unnamed_dev_lock);
699 if (error == -EAGAIN)
700 /* We raced and lost with another CPU. */
705 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
706 spin_lock(&unnamed_dev_lock);
707 ida_remove(&unnamed_dev_ida, dev);
708 spin_unlock(&unnamed_dev_lock);
711 s->s_dev = MKDEV(0, dev & MINORMASK);
715 EXPORT_SYMBOL(set_anon_super);
717 void kill_anon_super(struct super_block *sb)
719 int slot = MINOR(sb->s_dev);
721 generic_shutdown_super(sb);
722 spin_lock(&unnamed_dev_lock);
723 ida_remove(&unnamed_dev_ida, slot);
724 spin_unlock(&unnamed_dev_lock);
727 EXPORT_SYMBOL(kill_anon_super);
729 void kill_litter_super(struct super_block *sb)
732 d_genocide(sb->s_root);
736 EXPORT_SYMBOL(kill_litter_super);
739 static int set_bdev_super(struct super_block *s, void *data)
742 s->s_dev = s->s_bdev->bd_dev;
746 static int test_bdev_super(struct super_block *s, void *data)
748 return (void *)s->s_bdev == data;
751 int get_sb_bdev(struct file_system_type *fs_type,
752 int flags, const char *dev_name, void *data,
753 int (*fill_super)(struct super_block *, void *, int),
754 struct vfsmount *mnt)
756 struct block_device *bdev;
757 struct super_block *s;
758 fmode_t mode = FMODE_READ;
761 if (!(flags & MS_RDONLY))
764 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
766 return PTR_ERR(bdev);
769 * once the super is inserted into the list by sget, s_umount
770 * will protect the lockfs code from trying to start a snapshot
771 * while we are mounting
773 down(&bdev->bd_mount_sem);
774 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
775 up(&bdev->bd_mount_sem);
780 if ((flags ^ s->s_flags) & MS_RDONLY) {
781 up_write(&s->s_umount);
787 close_bdev_exclusive(bdev, mode);
789 char b[BDEVNAME_SIZE];
793 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
794 sb_set_blocksize(s, block_size(bdev));
795 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
797 up_write(&s->s_umount);
802 s->s_flags |= MS_ACTIVE;
805 return simple_set_mnt(mnt, s);
810 close_bdev_exclusive(bdev, mode);
815 EXPORT_SYMBOL(get_sb_bdev);
817 void kill_block_super(struct super_block *sb)
819 struct block_device *bdev = sb->s_bdev;
820 fmode_t mode = sb->s_mode;
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 up_write(&s->s_umount);
849 s->s_flags |= MS_ACTIVE;
850 return 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 up_write(&s->s_umount);
879 s->s_flags |= MS_ACTIVE;
881 do_remount_sb(s, flags, data, 0);
882 return simple_set_mnt(mnt, s);
885 EXPORT_SYMBOL(get_sb_single);
888 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
890 struct vfsmount *mnt;
891 char *secdata = NULL;
895 return ERR_PTR(-ENODEV);
898 mnt = alloc_vfsmnt(name);
902 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
903 secdata = alloc_secdata();
907 error = security_sb_copy_data(data, secdata);
909 goto out_free_secdata;
912 error = type->get_sb(type, flags, name, data, mnt);
914 goto out_free_secdata;
915 BUG_ON(!mnt->mnt_sb);
917 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
921 mnt->mnt_mountpoint = mnt->mnt_root;
922 mnt->mnt_parent = mnt;
923 up_write(&mnt->mnt_sb->s_umount);
924 free_secdata(secdata);
928 up_write(&mnt->mnt_sb->s_umount);
929 deactivate_super(mnt->mnt_sb);
931 free_secdata(secdata);
935 return ERR_PTR(error);
938 EXPORT_SYMBOL_GPL(vfs_kern_mount);
940 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
943 const char *subtype = strchr(fstype, '.');
952 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
954 if (!mnt->mnt_sb->s_subtype)
964 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
966 struct file_system_type *type = get_fs_type(fstype);
967 struct vfsmount *mnt;
969 return ERR_PTR(-ENODEV);
970 mnt = vfs_kern_mount(type, flags, name, data);
971 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
972 !mnt->mnt_sb->s_subtype)
973 mnt = fs_set_subtype(mnt, fstype);
974 put_filesystem(type);
977 EXPORT_SYMBOL_GPL(do_kern_mount);
979 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
981 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
984 EXPORT_SYMBOL_GPL(kern_mount_data);