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/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
54 * @type: filesystem type superblock should belong to
56 * Allocates and initializes a new &struct super_block. alloc_super()
57 * returns a pointer new superblock or %NULL if allocation had failed.
59 static struct super_block *alloc_super(struct file_system_type *type)
61 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
62 static struct super_operations default_op;
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 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);
85 down_write(&s->s_umount);
87 atomic_set(&s->s_active, 1);
88 mutex_init(&s->s_vfs_rename_mutex);
89 mutex_init(&s->s_dquot.dqio_mutex);
90 mutex_init(&s->s_dquot.dqonoff_mutex);
91 init_rwsem(&s->s_dquot.dqptr_sem);
92 init_waitqueue_head(&s->s_wait_unfrozen);
93 s->s_maxbytes = MAX_NON_LFS;
94 s->dq_op = sb_dquot_ops;
95 s->s_qcop = sb_quotactl_ops;
96 s->s_op = &default_op;
97 s->s_time_gran = 1000000000;
104 * destroy_super - frees a superblock
105 * @s: superblock to free
107 * Frees a superblock.
109 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 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)
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 * generic_shutdown_super - common helper for ->kill_sb()
227 * @sb: superblock to kill
229 * generic_shutdown_super() does all fs-independent work on superblock
230 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
231 * that need destruction out of superblock, call generic_shutdown_super()
232 * and release aforementioned objects. Note: dentries and inodes _are_
233 * taken care of and do not need specific handling.
235 * Upon calling this function, the filesystem may no longer alter or
236 * rearrange the set of dentries belonging to this super_block, nor may it
237 * change the attachments of dentries to inodes.
239 void generic_shutdown_super(struct super_block *sb)
241 struct super_operations *sop = sb->s_op;
244 shrink_dcache_for_umount(sb);
247 sb->s_flags &= ~MS_ACTIVE;
248 /* bad name - it should be evict_inodes() */
249 invalidate_inodes(sb);
252 if (sop->write_super && sb->s_dirt)
253 sop->write_super(sb);
257 /* Forget any remaining inodes */
258 if (invalidate_inodes(sb)) {
259 printk("VFS: Busy inodes after unmount of %s. "
260 "Self-destruct in 5 seconds. Have a nice day...\n",
268 /* should be initialized for __put_super_and_need_restart() */
269 list_del_init(&sb->s_list);
270 list_del(&sb->s_instances);
271 spin_unlock(&sb_lock);
272 up_write(&sb->s_umount);
275 EXPORT_SYMBOL(generic_shutdown_super);
278 * sget - find or create a superblock
279 * @type: filesystem type superblock should belong to
280 * @test: comparison callback
281 * @set: setup callback
282 * @data: argument to each of them
284 struct super_block *sget(struct file_system_type *type,
285 int (*test)(struct super_block *,void *),
286 int (*set)(struct super_block *,void *),
289 struct super_block *s = NULL;
295 if (test) list_for_each(p, &type->fs_supers) {
296 struct super_block *old;
297 old = list_entry(p, struct super_block, s_instances);
298 if (!test(old, data))
300 if (!grab_super(old))
307 spin_unlock(&sb_lock);
308 s = alloc_super(type);
310 return ERR_PTR(-ENOMEM);
316 spin_unlock(&sb_lock);
321 strlcpy(s->s_id, type->name, sizeof(s->s_id));
322 list_add_tail(&s->s_list, &super_blocks);
323 list_add(&s->s_instances, &type->fs_supers);
324 spin_unlock(&sb_lock);
325 get_filesystem(type);
331 void drop_super(struct super_block *sb)
333 up_read(&sb->s_umount);
337 EXPORT_SYMBOL(drop_super);
339 static inline void write_super(struct super_block *sb)
342 if (sb->s_root && sb->s_dirt)
343 if (sb->s_op->write_super)
344 sb->s_op->write_super(sb);
349 * Note: check the dirty flag before waiting, so we don't
350 * hold up the sync while mounting a device. (The newly
351 * mounted device won't need syncing.)
353 void sync_supers(void)
355 struct super_block *sb;
359 list_for_each_entry(sb, &super_blocks, s_list) {
362 spin_unlock(&sb_lock);
363 down_read(&sb->s_umount);
365 up_read(&sb->s_umount);
367 if (__put_super_and_need_restart(sb))
371 spin_unlock(&sb_lock);
375 * Call the ->sync_fs super_op against all filesytems which are r/w and
376 * which implement it.
378 * This operation is careful to avoid the livelock which could easily happen
379 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
380 * is used only here. We set it against all filesystems and then clear it as
381 * we sync them. So redirtied filesystems are skipped.
383 * But if process A is currently running sync_filesytems and then process B
384 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
385 * flags again, which will cause process A to resync everything. Fix that with
388 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
390 void sync_filesystems(int wait)
392 struct super_block *sb;
393 static DEFINE_MUTEX(mutex);
395 mutex_lock(&mutex); /* Could be down_interruptible */
397 list_for_each_entry(sb, &super_blocks, s_list) {
398 if (!sb->s_op->sync_fs)
400 if (sb->s_flags & MS_RDONLY)
402 sb->s_need_sync_fs = 1;
406 list_for_each_entry(sb, &super_blocks, s_list) {
407 if (!sb->s_need_sync_fs)
409 sb->s_need_sync_fs = 0;
410 if (sb->s_flags & MS_RDONLY)
411 continue; /* hm. Was remounted r/o meanwhile */
413 spin_unlock(&sb_lock);
414 down_read(&sb->s_umount);
415 if (sb->s_root && (wait || sb->s_dirt))
416 sb->s_op->sync_fs(sb, wait);
417 up_read(&sb->s_umount);
418 /* restart only when sb is no longer on the list */
420 if (__put_super_and_need_restart(sb))
423 spin_unlock(&sb_lock);
424 mutex_unlock(&mutex);
428 * get_super - get the superblock of a device
429 * @bdev: device to get the superblock for
431 * Scans the superblock list and finds the superblock of the file system
432 * mounted on the device given. %NULL is returned if no match is found.
435 struct super_block * get_super(struct block_device *bdev)
437 struct super_block *sb;
444 list_for_each_entry(sb, &super_blocks, s_list) {
445 if (sb->s_bdev == bdev) {
447 spin_unlock(&sb_lock);
448 down_read(&sb->s_umount);
451 up_read(&sb->s_umount);
452 /* restart only when sb is no longer on the list */
454 if (__put_super_and_need_restart(sb))
458 spin_unlock(&sb_lock);
462 EXPORT_SYMBOL(get_super);
464 struct super_block * user_get_super(dev_t dev)
466 struct super_block *sb;
470 list_for_each_entry(sb, &super_blocks, s_list) {
471 if (sb->s_dev == dev) {
473 spin_unlock(&sb_lock);
474 down_read(&sb->s_umount);
477 up_read(&sb->s_umount);
478 /* restart only when sb is no longer on the list */
480 if (__put_super_and_need_restart(sb))
484 spin_unlock(&sb_lock);
488 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
490 struct super_block *s;
495 s = user_get_super(new_decode_dev(dev));
498 err = vfs_statfs(s->s_root, &sbuf);
503 memset(&tmp,0,sizeof(struct ustat));
504 tmp.f_tfree = sbuf.f_bfree;
505 tmp.f_tinode = sbuf.f_ffree;
507 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
514 * @sb: superblock in question
516 * All files are marked read/only. We don't care about pending
517 * delete files so this should be used in 'force' mode only
520 static void mark_files_ro(struct super_block *sb)
525 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
526 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
527 f->f_mode &= ~FMODE_WRITE;
533 * do_remount_sb - asks filesystem to change mount options.
534 * @sb: superblock in question
535 * @flags: numeric part of options
536 * @data: the rest of options
537 * @force: whether or not to force the change
539 * Alters the mount options of a mounted file system.
541 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
545 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
547 if (flags & MS_RDONLY)
549 shrink_dcache_sb(sb);
552 /* If we are remounting RDONLY and current sb is read/write,
553 make sure there are no rw files opened */
554 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
557 else if (!fs_may_remount_ro(sb))
561 if (sb->s_op->remount_fs) {
563 retval = sb->s_op->remount_fs(sb, &flags, data);
568 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
572 static void do_emergency_remount(unsigned long foo)
574 struct super_block *sb;
577 list_for_each_entry(sb, &super_blocks, s_list) {
579 spin_unlock(&sb_lock);
580 down_read(&sb->s_umount);
581 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
583 * ->remount_fs needs lock_kernel().
585 * What lock protects sb->s_flags??
588 do_remount_sb(sb, MS_RDONLY, NULL, 1);
594 spin_unlock(&sb_lock);
595 printk("Emergency Remount complete\n");
598 void emergency_remount(void)
600 pdflush_operation(do_emergency_remount, 0);
604 * Unnamed block devices are dummy devices used by virtual
605 * filesystems which don't use real block-devices. -- jrs
608 static struct idr unnamed_dev_idr;
609 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
611 int set_anon_super(struct super_block *s, void *data)
617 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
619 spin_lock(&unnamed_dev_lock);
620 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
621 spin_unlock(&unnamed_dev_lock);
622 if (error == -EAGAIN)
623 /* We raced and lost with another CPU. */
628 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
629 spin_lock(&unnamed_dev_lock);
630 idr_remove(&unnamed_dev_idr, dev);
631 spin_unlock(&unnamed_dev_lock);
634 s->s_dev = MKDEV(0, dev & MINORMASK);
638 EXPORT_SYMBOL(set_anon_super);
640 void kill_anon_super(struct super_block *sb)
642 int slot = MINOR(sb->s_dev);
644 generic_shutdown_super(sb);
645 spin_lock(&unnamed_dev_lock);
646 idr_remove(&unnamed_dev_idr, slot);
647 spin_unlock(&unnamed_dev_lock);
650 EXPORT_SYMBOL(kill_anon_super);
652 void __init unnamed_dev_init(void)
654 idr_init(&unnamed_dev_idr);
657 void kill_litter_super(struct super_block *sb)
660 d_genocide(sb->s_root);
664 EXPORT_SYMBOL(kill_litter_super);
666 static int set_bdev_super(struct super_block *s, void *data)
669 s->s_dev = s->s_bdev->bd_dev;
673 static int test_bdev_super(struct super_block *s, void *data)
675 return (void *)s->s_bdev == data;
678 static void bdev_uevent(struct block_device *bdev, enum kobject_action action)
682 kobject_uevent(&bdev->bd_part->kobj, action);
684 kobject_uevent(&bdev->bd_disk->kobj, action);
688 int get_sb_bdev(struct file_system_type *fs_type,
689 int flags, const char *dev_name, void *data,
690 int (*fill_super)(struct super_block *, void *, int),
691 struct vfsmount *mnt)
693 struct block_device *bdev;
694 struct super_block *s;
697 bdev = open_bdev_excl(dev_name, flags, fs_type);
699 return PTR_ERR(bdev);
702 * once the super is inserted into the list by sget, s_umount
703 * will protect the lockfs code from trying to start a snapshot
704 * while we are mounting
706 mutex_lock(&bdev->bd_mount_mutex);
707 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
708 mutex_unlock(&bdev->bd_mount_mutex);
713 if ((flags ^ s->s_flags) & MS_RDONLY) {
714 up_write(&s->s_umount);
720 close_bdev_excl(bdev);
722 char b[BDEVNAME_SIZE];
725 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
726 sb_set_blocksize(s, block_size(bdev));
727 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
729 up_write(&s->s_umount);
734 s->s_flags |= MS_ACTIVE;
735 bdev_uevent(bdev, KOBJ_MOUNT);
738 return simple_set_mnt(mnt, s);
743 close_bdev_excl(bdev);
748 EXPORT_SYMBOL(get_sb_bdev);
750 void kill_block_super(struct super_block *sb)
752 struct block_device *bdev = sb->s_bdev;
754 bdev_uevent(bdev, KOBJ_UMOUNT);
755 generic_shutdown_super(sb);
757 close_bdev_excl(bdev);
760 EXPORT_SYMBOL(kill_block_super);
762 int get_sb_nodev(struct file_system_type *fs_type,
763 int flags, void *data,
764 int (*fill_super)(struct super_block *, void *, int),
765 struct vfsmount *mnt)
768 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
775 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
777 up_write(&s->s_umount);
781 s->s_flags |= MS_ACTIVE;
782 return simple_set_mnt(mnt, s);
785 EXPORT_SYMBOL(get_sb_nodev);
787 static int compare_single(struct super_block *s, void *p)
792 int get_sb_single(struct file_system_type *fs_type,
793 int flags, void *data,
794 int (*fill_super)(struct super_block *, void *, int),
795 struct vfsmount *mnt)
797 struct super_block *s;
800 s = sget(fs_type, compare_single, set_anon_super, NULL);
805 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
807 up_write(&s->s_umount);
811 s->s_flags |= MS_ACTIVE;
813 do_remount_sb(s, flags, data, 0);
814 return simple_set_mnt(mnt, s);
817 EXPORT_SYMBOL(get_sb_single);
820 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
822 struct vfsmount *mnt;
823 struct super_block *sb;
824 char *secdata = NULL;
828 return ERR_PTR(-ENODEV);
831 mnt = alloc_vfsmnt(name);
836 secdata = alloc_secdata();
840 error = security_sb_copy_data(type, data, secdata);
842 goto out_free_secdata;
845 error = type->get_sb(type, flags, name, data, mnt);
847 goto out_free_secdata;
851 if (!vx_capable(CAP_SYS_ADMIN, VXC_BINARY_MOUNT) && !sb->s_bdev &&
852 (sb->s_magic != PROC_SUPER_MAGIC) &&
853 (sb->s_magic != DEVPTS_SUPER_MAGIC))
856 error = security_sb_kern_mount(sb, secdata);
860 mnt->mnt_mountpoint = mnt->mnt_root;
861 mnt->mnt_parent = mnt;
862 up_write(&mnt->mnt_sb->s_umount);
863 free_secdata(secdata);
867 up_write(&mnt->mnt_sb->s_umount);
868 deactivate_super(mnt->mnt_sb);
870 free_secdata(secdata);
874 return ERR_PTR(error);
877 EXPORT_SYMBOL_GPL(vfs_kern_mount);
880 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
882 struct file_system_type *type = get_fs_type(fstype);
883 struct vfsmount *mnt;
886 return ERR_PTR(-ENODEV);
888 mnt = ERR_PTR(-EPERM);
889 if ((type->fs_flags & FS_BINARY_MOUNTDATA) &&
890 !vx_capable(CAP_SYS_ADMIN, VXC_BINARY_MOUNT))
893 mnt = vfs_kern_mount(type, flags, name, data);
895 put_filesystem(type);
899 struct vfsmount *kern_mount(struct file_system_type *type)
901 return vfs_kern_mount(type, 0, type->name, NULL);
904 EXPORT_SYMBOL(kern_mount);