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/syscalls.h>
36 #include <linux/vfs.h>
37 #include <linux/writeback.h> /* for the emergency remount stuff */
38 #include <linux/idr.h>
39 #include <linux/kobject.h>
40 #include <linux/devpts_fs.h>
41 #include <linux/proc_fs.h>
42 #include <asm/uaccess.h>
45 void get_filesystem(struct file_system_type *fs);
46 void put_filesystem(struct file_system_type *fs);
47 struct file_system_type *get_fs_type(const char *name);
49 LIST_HEAD(super_blocks);
50 DEFINE_SPINLOCK(sb_lock);
53 * alloc_super - create new superblock
55 * Allocates and initializes a new &struct super_block. alloc_super()
56 * returns a pointer new superblock or %NULL if allocation had failed.
58 static struct super_block *alloc_super(void)
60 struct super_block *s = kmalloc(sizeof(struct super_block), GFP_USER);
61 static struct super_operations default_op;
64 memset(s, 0, sizeof(struct super_block));
65 if (security_sb_alloc(s)) {
70 INIT_LIST_HEAD(&s->s_dirty);
71 INIT_LIST_HEAD(&s->s_io);
72 INIT_LIST_HEAD(&s->s_files);
73 INIT_LIST_HEAD(&s->s_instances);
74 INIT_HLIST_HEAD(&s->s_anon);
75 INIT_LIST_HEAD(&s->s_inodes);
76 init_rwsem(&s->s_umount);
77 sema_init(&s->s_lock, 1);
78 down_write(&s->s_umount);
80 atomic_set(&s->s_active, 1);
81 sema_init(&s->s_vfs_rename_sem,1);
82 sema_init(&s->s_dquot.dqio_sem, 1);
83 sema_init(&s->s_dquot.dqonoff_sem, 1);
84 init_rwsem(&s->s_dquot.dqptr_sem);
85 init_waitqueue_head(&s->s_wait_unfrozen);
86 s->s_maxbytes = MAX_NON_LFS;
87 s->dq_op = sb_dquot_ops;
88 s->s_qcop = sb_quotactl_ops;
89 s->s_op = &default_op;
90 s->s_time_gran = 1000000000;
97 * destroy_super - frees a superblock
98 * @s: superblock to free
100 * Frees a superblock.
102 static inline void destroy_super(struct super_block *s)
108 /* Superblock refcounting */
111 * Drop a superblock's refcount. Returns non-zero if the superblock was
112 * destroyed. The caller must hold sb_lock.
114 int __put_super(struct super_block *sb)
118 if (!--sb->s_count) {
126 * Drop a superblock's refcount.
127 * Returns non-zero if the superblock is about to be destroyed and
128 * at least is already removed from super_blocks list, so if we are
129 * making a loop through super blocks then we need to restart.
130 * The caller must hold sb_lock.
132 int __put_super_and_need_restart(struct super_block *sb)
134 /* check for race with generic_shutdown_super() */
135 if (list_empty(&sb->s_list)) {
136 /* super block is removed, need to restart... */
140 /* can't be the last, since s_list is still in use */
142 BUG_ON(sb->s_count == 0);
147 * put_super - drop a temporary reference to superblock
148 * @s: superblock in question
150 * Drops a temporary reference, frees superblock if there's no
153 static void put_super(struct super_block *sb)
157 spin_unlock(&sb_lock);
162 * deactivate_super - drop an active reference to superblock
163 * @s: superblock to deactivate
165 * Drops an active reference to superblock, acquiring a temprory one if
166 * there is no active references left. In that case we lock superblock,
167 * tell fs driver to shut it down and drop the temporary reference we
170 void deactivate_super(struct super_block *s)
172 struct file_system_type *fs = s->s_type;
173 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
174 s->s_count -= S_BIAS-1;
175 spin_unlock(&sb_lock);
176 down_write(&s->s_umount);
183 EXPORT_SYMBOL(deactivate_super);
186 * grab_super - acquire an active reference
187 * @s: reference we are trying to make active
189 * Tries to acquire an active reference. grab_super() is used when we
190 * had just found a superblock in super_blocks or fs_type->fs_supers
191 * and want to turn it into a full-blown active reference. grab_super()
192 * is called with sb_lock held and drops it. Returns 1 in case of
193 * success, 0 if we had failed (superblock contents was already dead or
194 * dying when grab_super() had been called).
196 static int grab_super(struct super_block *s)
199 spin_unlock(&sb_lock);
200 down_write(&s->s_umount);
203 if (s->s_count > S_BIAS) {
204 atomic_inc(&s->s_active);
206 spin_unlock(&sb_lock);
209 spin_unlock(&sb_lock);
211 up_write(&s->s_umount);
218 * generic_shutdown_super - common helper for ->kill_sb()
219 * @sb: superblock to kill
221 * generic_shutdown_super() does all fs-independent work on superblock
222 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
223 * that need destruction out of superblock, call generic_shutdown_super()
224 * and release aforementioned objects. Note: dentries and inodes _are_
225 * taken care of and do not need specific handling.
227 void generic_shutdown_super(struct super_block *sb)
229 struct dentry *root = sb->s_root;
230 struct super_operations *sop = sb->s_op;
234 shrink_dcache_parent(root);
235 shrink_dcache_anon(&sb->s_anon);
239 sb->s_flags &= ~MS_ACTIVE;
240 /* bad name - it should be evict_inodes() */
241 invalidate_inodes(sb);
244 if (sop->write_super && sb->s_dirt)
245 sop->write_super(sb);
249 /* Forget any remaining inodes */
250 if (invalidate_inodes(sb)) {
251 printk("VFS: Busy inodes after unmount. "
252 "Self-destruct in 5 seconds. Have a nice day...\n");
259 /* should be initialized for __put_super_and_need_restart() */
260 list_del_init(&sb->s_list);
261 list_del(&sb->s_instances);
262 spin_unlock(&sb_lock);
263 up_write(&sb->s_umount);
266 EXPORT_SYMBOL(generic_shutdown_super);
269 * sget - find or create a superblock
270 * @type: filesystem type superblock should belong to
271 * @test: comparison callback
272 * @set: setup callback
273 * @data: argument to each of them
275 struct super_block *sget(struct file_system_type *type,
276 int (*test)(struct super_block *,void *),
277 int (*set)(struct super_block *,void *),
280 struct super_block *s = NULL;
286 if (test) list_for_each(p, &type->fs_supers) {
287 struct super_block *old;
288 old = list_entry(p, struct super_block, s_instances);
289 if (!test(old, data))
291 if (!grab_super(old))
298 spin_unlock(&sb_lock);
301 return ERR_PTR(-ENOMEM);
307 spin_unlock(&sb_lock);
312 strlcpy(s->s_id, type->name, sizeof(s->s_id));
313 list_add_tail(&s->s_list, &super_blocks);
314 list_add(&s->s_instances, &type->fs_supers);
315 spin_unlock(&sb_lock);
316 get_filesystem(type);
322 void drop_super(struct super_block *sb)
324 up_read(&sb->s_umount);
328 EXPORT_SYMBOL(drop_super);
330 static inline void write_super(struct super_block *sb)
333 if (sb->s_root && sb->s_dirt)
334 if (sb->s_op->write_super)
335 sb->s_op->write_super(sb);
340 * Note: check the dirty flag before waiting, so we don't
341 * hold up the sync while mounting a device. (The newly
342 * mounted device won't need syncing.)
344 void sync_supers(void)
346 struct super_block * sb;
349 sb = sb_entry(super_blocks.next);
350 while (sb != sb_entry(&super_blocks))
353 spin_unlock(&sb_lock);
354 down_read(&sb->s_umount);
359 sb = sb_entry(sb->s_list.next);
360 spin_unlock(&sb_lock);
364 * Call the ->sync_fs super_op against all filesytems which are r/w and
365 * which implement it.
367 * This operation is careful to avoid the livelock which could easily happen
368 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
369 * is used only here. We set it against all filesystems and then clear it as
370 * we sync them. So redirtied filesystems are skipped.
372 * But if process A is currently running sync_filesytems and then process B
373 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
374 * flags again, which will cause process A to resync everything. Fix that with
377 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
379 void sync_filesystems(int wait)
381 struct super_block *sb;
382 static DECLARE_MUTEX(mutex);
384 down(&mutex); /* Could be down_interruptible */
386 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
387 sb = sb_entry(sb->s_list.next)) {
388 if (!sb->s_op->sync_fs)
390 if (sb->s_flags & MS_RDONLY)
392 sb->s_need_sync_fs = 1;
394 spin_unlock(&sb_lock);
398 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
399 sb = sb_entry(sb->s_list.next)) {
400 if (!sb->s_need_sync_fs)
402 sb->s_need_sync_fs = 0;
403 if (sb->s_flags & MS_RDONLY)
404 continue; /* hm. Was remounted r/o meanwhile */
406 spin_unlock(&sb_lock);
407 down_read(&sb->s_umount);
408 if (sb->s_root && (wait || sb->s_dirt))
409 sb->s_op->sync_fs(sb, wait);
413 spin_unlock(&sb_lock);
418 * get_super - get the superblock of a device
419 * @bdev: device to get the superblock for
421 * Scans the superblock list and finds the superblock of the file system
422 * mounted on the device given. %NULL is returned if no match is found.
425 struct super_block * get_super(struct block_device *bdev)
432 list_for_each(p, &super_blocks) {
433 struct super_block *s = sb_entry(p);
434 if (s->s_bdev == bdev) {
436 spin_unlock(&sb_lock);
437 down_read(&s->s_umount);
444 spin_unlock(&sb_lock);
448 EXPORT_SYMBOL(get_super);
450 struct super_block * user_get_super(dev_t dev)
456 list_for_each(p, &super_blocks) {
457 struct super_block *s = sb_entry(p);
458 if (s->s_dev == dev) {
460 spin_unlock(&sb_lock);
461 down_read(&s->s_umount);
468 spin_unlock(&sb_lock);
472 EXPORT_SYMBOL(user_get_super);
474 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
476 struct super_block *s;
481 s = user_get_super(new_decode_dev(dev));
484 err = vfs_statfs(s, &sbuf);
489 memset(&tmp,0,sizeof(struct ustat));
490 tmp.f_tfree = sbuf.f_bfree;
491 tmp.f_tinode = sbuf.f_ffree;
493 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
500 * @sb: superblock in question
502 * All files are marked read/only. We don't care about pending
503 * delete files so this should be used in 'force' mode only
506 static void mark_files_ro(struct super_block *sb)
511 list_for_each_entry(f, &sb->s_files, f_list) {
512 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
513 f->f_mode &= ~FMODE_WRITE;
519 * do_remount_sb - asks filesystem to change mount options.
520 * @sb: superblock in question
521 * @flags: numeric part of options
522 * @data: the rest of options
523 * @force: whether or not to force the change
525 * Alters the mount options of a mounted file system.
527 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
531 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
533 if (flags & MS_RDONLY)
535 shrink_dcache_sb(sb);
538 /* If we are remounting RDONLY and current sb is read/write,
539 make sure there are no rw files opened */
540 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
543 else if (!fs_may_remount_ro(sb))
547 if (sb->s_op->remount_fs) {
549 retval = sb->s_op->remount_fs(sb, &flags, data);
554 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
558 static void do_emergency_remount(unsigned long foo)
560 struct super_block *sb;
563 list_for_each_entry(sb, &super_blocks, s_list) {
565 spin_unlock(&sb_lock);
566 down_read(&sb->s_umount);
567 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
569 * ->remount_fs needs lock_kernel().
571 * What lock protects sb->s_flags??
574 do_remount_sb(sb, MS_RDONLY, NULL, 1);
580 spin_unlock(&sb_lock);
581 printk("Emergency Remount complete\n");
584 void emergency_remount(void)
586 pdflush_operation(do_emergency_remount, 0);
590 * Unnamed block devices are dummy devices used by virtual
591 * filesystems which don't use real block-devices. -- jrs
594 static struct idr unnamed_dev_idr;
595 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
597 int set_anon_super(struct super_block *s, void *data)
603 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
605 spin_lock(&unnamed_dev_lock);
606 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
607 spin_unlock(&unnamed_dev_lock);
608 if (error == -EAGAIN)
609 /* We raced and lost with another CPU. */
614 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
615 spin_lock(&unnamed_dev_lock);
616 idr_remove(&unnamed_dev_idr, dev);
617 spin_unlock(&unnamed_dev_lock);
620 s->s_dev = MKDEV(0, dev & MINORMASK);
624 EXPORT_SYMBOL(set_anon_super);
626 void kill_anon_super(struct super_block *sb)
628 int slot = MINOR(sb->s_dev);
630 generic_shutdown_super(sb);
631 spin_lock(&unnamed_dev_lock);
632 idr_remove(&unnamed_dev_idr, slot);
633 spin_unlock(&unnamed_dev_lock);
636 EXPORT_SYMBOL(kill_anon_super);
638 void __init unnamed_dev_init(void)
640 idr_init(&unnamed_dev_idr);
643 void kill_litter_super(struct super_block *sb)
646 d_genocide(sb->s_root);
650 EXPORT_SYMBOL(kill_litter_super);
652 static int set_bdev_super(struct super_block *s, void *data)
655 s->s_dev = s->s_bdev->bd_dev;
659 static int test_bdev_super(struct super_block *s, void *data)
661 return (void *)s->s_bdev == data;
664 static void bdev_uevent(struct block_device *bdev, enum kobject_action action)
668 kobject_uevent(&bdev->bd_part->kobj, action, NULL);
670 kobject_uevent(&bdev->bd_disk->kobj, action, NULL);
674 struct super_block *get_sb_bdev(struct file_system_type *fs_type,
675 int flags, const char *dev_name, void *data,
676 int (*fill_super)(struct super_block *, void *, int))
678 struct block_device *bdev;
679 struct super_block *s;
682 bdev = open_bdev_excl(dev_name, flags, fs_type);
684 return (struct super_block *)bdev;
687 * once the super is inserted into the list by sget, s_umount
688 * will protect the lockfs code from trying to start a snapshot
689 * while we are mounting
691 down(&bdev->bd_mount_sem);
692 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
693 up(&bdev->bd_mount_sem);
698 if ((flags ^ s->s_flags) & MS_RDONLY) {
699 up_write(&s->s_umount);
705 char b[BDEVNAME_SIZE];
708 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
709 s->s_old_blocksize = block_size(bdev);
710 sb_set_blocksize(s, s->s_old_blocksize);
711 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
713 up_write(&s->s_umount);
717 s->s_flags |= MS_ACTIVE;
718 bdev_uevent(bdev, KOBJ_MOUNT);
725 close_bdev_excl(bdev);
729 EXPORT_SYMBOL(get_sb_bdev);
731 void kill_block_super(struct super_block *sb)
733 struct block_device *bdev = sb->s_bdev;
735 bdev_uevent(bdev, KOBJ_UMOUNT);
736 generic_shutdown_super(sb);
737 set_blocksize(bdev, sb->s_old_blocksize);
738 close_bdev_excl(bdev);
741 EXPORT_SYMBOL(kill_block_super);
743 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
744 int flags, void *data,
745 int (*fill_super)(struct super_block *, void *, int))
748 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
755 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
757 up_write(&s->s_umount);
759 return ERR_PTR(error);
761 s->s_flags |= MS_ACTIVE;
765 EXPORT_SYMBOL(get_sb_nodev);
767 static int compare_single(struct super_block *s, void *p)
772 struct super_block *get_sb_single(struct file_system_type *fs_type,
773 int flags, void *data,
774 int (*fill_super)(struct super_block *, void *, int))
776 struct super_block *s;
779 s = sget(fs_type, compare_single, set_anon_super, NULL);
784 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
786 up_write(&s->s_umount);
788 return ERR_PTR(error);
790 s->s_flags |= MS_ACTIVE;
792 do_remount_sb(s, flags, data, 0);
796 EXPORT_SYMBOL(get_sb_single);
799 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
801 struct file_system_type *type = get_fs_type(fstype);
802 struct super_block *sb;
803 struct vfsmount *mnt;
805 char *secdata = NULL;
808 return ERR_PTR(-ENODEV);
810 sb = ERR_PTR(-EPERM);
811 if ((type->fs_flags & FS_BINARY_MOUNTDATA) &&
812 !capable(CAP_SYS_ADMIN) && !vx_ccaps(VXC_BINARY_MOUNT))
815 sb = ERR_PTR(-ENOMEM);
816 mnt = alloc_vfsmnt(name);
821 secdata = alloc_secdata();
823 sb = ERR_PTR(-ENOMEM);
827 error = security_sb_copy_data(type, data, secdata);
830 goto out_free_secdata;
834 sb = type->get_sb(type, flags, name, data);
836 goto out_free_secdata;
839 if (!capable(CAP_SYS_ADMIN) && !sb->s_bdev &&
840 (sb->s_magic != PROC_SUPER_MAGIC) &&
841 (sb->s_magic != DEVPTS_SUPER_MAGIC))
844 error = security_sb_kern_mount(sb, secdata);
848 mnt->mnt_root = dget(sb->s_root);
849 mnt->mnt_mountpoint = sb->s_root;
850 mnt->mnt_parent = mnt;
851 mnt->mnt_namespace = current->namespace;
852 up_write(&sb->s_umount);
853 put_filesystem(type);
856 up_write(&sb->s_umount);
857 deactivate_super(sb);
860 free_secdata(secdata);
864 put_filesystem(type);
865 return (struct vfsmount *)sb;
868 EXPORT_SYMBOL_GPL(do_kern_mount);
870 struct vfsmount *kern_mount(struct file_system_type *type)
872 return do_kern_mount(type->name, 0, type->name, NULL);
875 EXPORT_SYMBOL(kern_mount);