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/config.h>
24 #include <linux/module.h>
25 #include <linux/slab.h>
26 #include <linux/init.h>
27 #include <linux/smp_lock.h>
28 #include <linux/acct.h>
29 #include <linux/blkdev.h>
30 #include <linux/quotaops.h>
31 #include <linux/namei.h>
32 #include <linux/buffer_head.h> /* for fsync_super() */
33 #include <linux/mount.h>
34 #include <linux/security.h>
35 #include <linux/vfs.h>
36 #include <linux/writeback.h> /* for the emergency remount stuff */
37 #include <linux/idr.h>
38 #include <linux/devpts_fs.h>
39 #include <linux/proc_fs.h>
40 #include <asm/uaccess.h>
43 void get_filesystem(struct file_system_type *fs);
44 void put_filesystem(struct file_system_type *fs);
45 struct file_system_type *get_fs_type(const char *name);
47 LIST_HEAD(super_blocks);
48 spinlock_t sb_lock = SPIN_LOCK_UNLOCKED;
51 * alloc_super - create new superblock
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(void)
58 struct super_block *s = kmalloc(sizeof(struct super_block), GFP_USER);
59 static struct super_operations default_op;
62 memset(s, 0, sizeof(struct super_block));
63 if (security_sb_alloc(s)) {
68 INIT_LIST_HEAD(&s->s_dirty);
69 INIT_LIST_HEAD(&s->s_io);
70 INIT_LIST_HEAD(&s->s_files);
71 INIT_LIST_HEAD(&s->s_instances);
72 INIT_HLIST_HEAD(&s->s_anon);
73 init_rwsem(&s->s_umount);
74 sema_init(&s->s_lock, 1);
75 down_write(&s->s_umount);
77 atomic_set(&s->s_active, 1);
78 sema_init(&s->s_vfs_rename_sem,1);
79 sema_init(&s->s_dquot.dqio_sem, 1);
80 sema_init(&s->s_dquot.dqonoff_sem, 1);
81 init_rwsem(&s->s_dquot.dqptr_sem);
82 init_waitqueue_head(&s->s_wait_unfrozen);
83 s->s_maxbytes = MAX_NON_LFS;
84 s->dq_op = sb_dquot_ops;
85 s->s_qcop = sb_quotactl_ops;
86 s->s_op = &default_op;
93 * destroy_super - frees a superblock
94 * @s: superblock to free
98 static inline void destroy_super(struct super_block *s)
104 /* Superblock refcounting */
107 * Drop a superblock's refcount. Returns non-zero if the superblock was
108 * destroyed. The caller must hold sb_lock.
110 int __put_super(struct super_block *sb)
114 if (!--sb->s_count) {
122 * Drop a superblock's refcount.
123 * Returns non-zero if the superblock is about to be destroyed and
124 * at least is already removed from super_blocks list, so if we are
125 * making a loop through super blocks then we need to restart.
126 * The caller must hold sb_lock.
128 int __put_super_and_need_restart(struct super_block *sb)
130 /* check for race with generic_shutdown_super() */
131 if (list_empty(&sb->s_list)) {
132 /* super block is removed, need to restart... */
136 /* can't be the last, since s_list is still in use */
138 BUG_ON(sb->s_count == 0);
143 * put_super - drop a temporary reference to superblock
144 * @s: superblock in question
146 * Drops a temporary reference, frees superblock if there's no
149 static void put_super(struct super_block *sb)
153 spin_unlock(&sb_lock);
158 * deactivate_super - drop an active reference to superblock
159 * @s: superblock to deactivate
161 * Drops an active reference to superblock, acquiring a temprory one if
162 * there is no active references left. In that case we lock superblock,
163 * tell fs driver to shut it down and drop the temporary reference we
166 void deactivate_super(struct super_block *s)
168 struct file_system_type *fs = s->s_type;
169 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
170 s->s_count -= S_BIAS-1;
171 spin_unlock(&sb_lock);
172 down_write(&s->s_umount);
179 EXPORT_SYMBOL(deactivate_super);
182 * grab_super - acquire an active reference
183 * @s: reference we are trying to make active
185 * Tries to acquire an active reference. grab_super() is used when we
186 * had just found a superblock in super_blocks or fs_type->fs_supers
187 * and want to turn it into a full-blown active reference. grab_super()
188 * is called with sb_lock held and drops it. Returns 1 in case of
189 * success, 0 if we had failed (superblock contents was already dead or
190 * dying when grab_super() had been called).
192 static int grab_super(struct super_block *s)
195 spin_unlock(&sb_lock);
196 down_write(&s->s_umount);
199 if (s->s_count > S_BIAS) {
200 atomic_inc(&s->s_active);
202 spin_unlock(&sb_lock);
205 spin_unlock(&sb_lock);
207 up_write(&s->s_umount);
214 * generic_shutdown_super - common helper for ->kill_sb()
215 * @sb: superblock to kill
217 * generic_shutdown_super() does all fs-independent work on superblock
218 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
219 * that need destruction out of superblock, call generic_shutdown_super()
220 * and release aforementioned objects. Note: dentries and inodes _are_
221 * taken care of and do not need specific handling.
223 void generic_shutdown_super(struct super_block *sb)
225 struct dentry *root = sb->s_root;
226 struct super_operations *sop = sb->s_op;
230 shrink_dcache_parent(root);
231 shrink_dcache_anon(&sb->s_anon);
236 sb->s_flags &= ~MS_ACTIVE;
237 /* bad name - it should be evict_inodes() */
238 invalidate_inodes(sb);
240 if (sop->write_super && sb->s_dirt)
241 sop->write_super(sb);
245 /* Forget any remaining inodes */
246 if (invalidate_inodes(sb)) {
247 printk("VFS: Busy inodes after unmount. "
248 "Self-destruct in 5 seconds. Have a nice day...\n");
255 /* should be initialized for __put_super_and_need_restart() */
256 list_del_init(&sb->s_list);
257 list_del(&sb->s_instances);
258 spin_unlock(&sb_lock);
259 up_write(&sb->s_umount);
262 EXPORT_SYMBOL(generic_shutdown_super);
265 * sget - find or create a superblock
266 * @type: filesystem type superblock should belong to
267 * @test: comparison callback
268 * @set: setup callback
269 * @data: argument to each of them
271 struct super_block *sget(struct file_system_type *type,
272 int (*test)(struct super_block *,void *),
273 int (*set)(struct super_block *,void *),
276 struct super_block *s = NULL;
282 if (test) list_for_each(p, &type->fs_supers) {
283 struct super_block *old;
284 old = list_entry(p, struct super_block, s_instances);
285 if (!test(old, data))
287 if (!grab_super(old))
294 spin_unlock(&sb_lock);
297 return ERR_PTR(-ENOMEM);
303 spin_unlock(&sb_lock);
308 strlcpy(s->s_id, type->name, sizeof(s->s_id));
309 list_add_tail(&s->s_list, &super_blocks);
310 list_add(&s->s_instances, &type->fs_supers);
311 spin_unlock(&sb_lock);
312 get_filesystem(type);
318 void drop_super(struct super_block *sb)
320 up_read(&sb->s_umount);
324 EXPORT_SYMBOL(drop_super);
326 static inline void write_super(struct super_block *sb)
329 if (sb->s_root && sb->s_dirt)
330 if (sb->s_op->write_super)
331 sb->s_op->write_super(sb);
336 * Note: check the dirty flag before waiting, so we don't
337 * hold up the sync while mounting a device. (The newly
338 * mounted device won't need syncing.)
340 void sync_supers(void)
342 struct super_block * sb;
345 sb = sb_entry(super_blocks.next);
346 while (sb != sb_entry(&super_blocks))
349 spin_unlock(&sb_lock);
350 down_read(&sb->s_umount);
355 sb = sb_entry(sb->s_list.next);
356 spin_unlock(&sb_lock);
360 * Call the ->sync_fs super_op against all filesytems which are r/w and
361 * which implement it.
363 * This operation is careful to avoid the livelock which could easily happen
364 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
365 * is used only here. We set it against all filesystems and then clear it as
366 * we sync them. So redirtied filesystems are skipped.
368 * But if process A is currently running sync_filesytems and then process B
369 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
370 * flags again, which will cause process A to resync everything. Fix that with
373 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
375 void sync_filesystems(int wait)
377 struct super_block *sb;
378 static DECLARE_MUTEX(mutex);
380 down(&mutex); /* Could be down_interruptible */
382 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
383 sb = sb_entry(sb->s_list.next)) {
384 if (!sb->s_op->sync_fs)
386 if (sb->s_flags & MS_RDONLY)
388 sb->s_need_sync_fs = 1;
390 spin_unlock(&sb_lock);
394 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
395 sb = sb_entry(sb->s_list.next)) {
396 if (!sb->s_need_sync_fs)
398 sb->s_need_sync_fs = 0;
399 if (sb->s_flags & MS_RDONLY)
400 continue; /* hm. Was remounted r/o meanwhile */
402 spin_unlock(&sb_lock);
403 down_read(&sb->s_umount);
404 if (sb->s_root && (wait || sb->s_dirt))
405 sb->s_op->sync_fs(sb, wait);
409 spin_unlock(&sb_lock);
414 * get_super - get the superblock of a device
415 * @bdev: device to get the superblock for
417 * Scans the superblock list and finds the superblock of the file system
418 * mounted on the device given. %NULL is returned if no match is found.
421 struct super_block * get_super(struct block_device *bdev)
428 list_for_each(p, &super_blocks) {
429 struct super_block *s = sb_entry(p);
430 if (s->s_bdev == bdev) {
432 spin_unlock(&sb_lock);
433 down_read(&s->s_umount);
440 spin_unlock(&sb_lock);
444 EXPORT_SYMBOL(get_super);
446 struct super_block * user_get_super(dev_t dev)
452 list_for_each(p, &super_blocks) {
453 struct super_block *s = sb_entry(p);
454 if (s->s_dev == dev) {
456 spin_unlock(&sb_lock);
457 down_read(&s->s_umount);
464 spin_unlock(&sb_lock);
468 EXPORT_SYMBOL(user_get_super);
470 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
472 struct super_block *s;
477 s = user_get_super(new_decode_dev(dev));
480 err = vfs_statfs(s, &sbuf);
485 memset(&tmp,0,sizeof(struct ustat));
486 tmp.f_tfree = sbuf.f_bfree;
487 tmp.f_tinode = sbuf.f_ffree;
489 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
496 * @sb: superblock in question
498 * All files are marked read/only. We don't care about pending
499 * delete files so this should be used in 'force' mode only
502 static void mark_files_ro(struct super_block *sb)
507 list_for_each_entry(f, &sb->s_files, f_list) {
508 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
509 f->f_mode &= ~FMODE_WRITE;
515 * do_remount_sb - asks filesystem to change mount options.
516 * @sb: superblock in question
517 * @flags: numeric part of options
518 * @data: the rest of options
519 * @force: whether or not to force the change
521 * Alters the mount options of a mounted file system.
523 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
527 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
529 if (flags & MS_RDONLY)
531 shrink_dcache_sb(sb);
534 /* If we are remounting RDONLY and current sb is read/write,
535 make sure there are no rw files opened */
536 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
539 else if (!fs_may_remount_ro(sb))
543 if (sb->s_op->remount_fs) {
545 retval = sb->s_op->remount_fs(sb, &flags, data);
550 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
554 static void do_emergency_remount(unsigned long foo)
556 struct super_block *sb;
559 list_for_each_entry(sb, &super_blocks, s_list) {
561 spin_unlock(&sb_lock);
562 down_read(&sb->s_umount);
563 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
565 * ->remount_fs needs lock_kernel().
567 * What lock protects sb->s_flags??
570 do_remount_sb(sb, MS_RDONLY, NULL, 1);
576 spin_unlock(&sb_lock);
577 printk("Emergency Remount complete\n");
580 void emergency_remount(void)
582 pdflush_operation(do_emergency_remount, 0);
586 * Unnamed block devices are dummy devices used by virtual
587 * filesystems which don't use real block-devices. -- jrs
590 static struct idr unnamed_dev_idr;
591 static spinlock_t unnamed_dev_lock = SPIN_LOCK_UNLOCKED;/* protects the above */
593 int set_anon_super(struct super_block *s, void *data)
599 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
601 spin_lock(&unnamed_dev_lock);
602 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
603 spin_unlock(&unnamed_dev_lock);
604 if (error == -EAGAIN)
605 /* We raced and lost with another CPU. */
610 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
611 spin_lock(&unnamed_dev_lock);
612 idr_remove(&unnamed_dev_idr, dev);
613 spin_unlock(&unnamed_dev_lock);
616 s->s_dev = MKDEV(0, dev & MINORMASK);
620 EXPORT_SYMBOL(set_anon_super);
622 void kill_anon_super(struct super_block *sb)
624 int slot = MINOR(sb->s_dev);
626 generic_shutdown_super(sb);
627 spin_lock(&unnamed_dev_lock);
628 idr_remove(&unnamed_dev_idr, slot);
629 spin_unlock(&unnamed_dev_lock);
632 EXPORT_SYMBOL(kill_anon_super);
634 void __init unnamed_dev_init(void)
636 idr_init(&unnamed_dev_idr);
639 void kill_litter_super(struct super_block *sb)
642 d_genocide(sb->s_root);
646 EXPORT_SYMBOL(kill_litter_super);
648 static int set_bdev_super(struct super_block *s, void *data)
651 s->s_dev = s->s_bdev->bd_dev;
655 static int test_bdev_super(struct super_block *s, void *data)
657 return (void *)s->s_bdev == data;
660 struct super_block *get_sb_bdev(struct file_system_type *fs_type,
661 int flags, const char *dev_name, void *data,
662 int (*fill_super)(struct super_block *, void *, int))
664 struct block_device *bdev;
665 struct super_block *s;
668 bdev = open_bdev_excl(dev_name, flags, fs_type);
670 return (struct super_block *)bdev;
673 * once the super is inserted into the list by sget, s_umount
674 * will protect the lockfs code from trying to start a snapshot
675 * while we are mounting
677 down(&bdev->bd_mount_sem);
678 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
679 up(&bdev->bd_mount_sem);
684 if ((flags ^ s->s_flags) & MS_RDONLY) {
685 up_write(&s->s_umount);
691 char b[BDEVNAME_SIZE];
694 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
695 s->s_old_blocksize = block_size(bdev);
696 sb_set_blocksize(s, s->s_old_blocksize);
697 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
699 up_write(&s->s_umount);
703 s->s_flags |= MS_ACTIVE;
709 close_bdev_excl(bdev);
713 EXPORT_SYMBOL(get_sb_bdev);
715 void kill_block_super(struct super_block *sb)
717 struct block_device *bdev = sb->s_bdev;
718 generic_shutdown_super(sb);
719 set_blocksize(bdev, sb->s_old_blocksize);
720 close_bdev_excl(bdev);
723 EXPORT_SYMBOL(kill_block_super);
725 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
726 int flags, void *data,
727 int (*fill_super)(struct super_block *, void *, int))
730 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
737 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
739 up_write(&s->s_umount);
741 return ERR_PTR(error);
743 s->s_flags |= MS_ACTIVE;
747 EXPORT_SYMBOL(get_sb_nodev);
749 static int compare_single(struct super_block *s, void *p)
754 struct super_block *get_sb_single(struct file_system_type *fs_type,
755 int flags, void *data,
756 int (*fill_super)(struct super_block *, void *, int))
758 struct super_block *s;
761 s = sget(fs_type, compare_single, set_anon_super, NULL);
766 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
768 up_write(&s->s_umount);
770 return ERR_PTR(error);
772 s->s_flags |= MS_ACTIVE;
774 do_remount_sb(s, flags, data, 0);
778 EXPORT_SYMBOL(get_sb_single);
781 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
783 struct file_system_type *type = get_fs_type(fstype);
784 struct super_block *sb = ERR_PTR(-ENOMEM);
785 struct vfsmount *mnt;
787 char *secdata = NULL;
790 return ERR_PTR(-ENODEV);
792 mnt = alloc_vfsmnt(name);
797 secdata = alloc_secdata();
799 sb = ERR_PTR(-ENOMEM);
803 error = security_sb_copy_data(type, data, secdata);
806 goto out_free_secdata;
810 sb = type->get_sb(type, flags, name, data);
812 goto out_free_secdata;
815 if (!capable(CAP_SYS_ADMIN) && !sb->s_bdev &&
816 (sb->s_magic != PROC_SUPER_MAGIC) &&
817 (sb->s_magic != DEVPTS_SUPER_MAGIC))
820 error = security_sb_kern_mount(sb, secdata);
824 mnt->mnt_root = dget(sb->s_root);
825 mnt->mnt_mountpoint = sb->s_root;
826 mnt->mnt_parent = mnt;
827 mnt->mnt_namespace = current->namespace;
828 up_write(&sb->s_umount);
829 put_filesystem(type);
832 up_write(&sb->s_umount);
833 deactivate_super(sb);
836 free_secdata(secdata);
840 put_filesystem(type);
841 return (struct vfsmount *)sb;
844 EXPORT_SYMBOL_GPL(do_kern_mount);
846 struct vfsmount *kern_mount(struct file_system_type *type)
848 return do_kern_mount(type->name, 0, type->name, NULL);
851 EXPORT_SYMBOL(kern_mount);