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 * put_super - drop a temporary reference to superblock
123 * @s: superblock in question
125 * Drops a temporary reference, frees superblock if there's no
128 static void put_super(struct super_block *sb)
132 spin_unlock(&sb_lock);
137 * deactivate_super - drop an active reference to superblock
138 * @s: superblock to deactivate
140 * Drops an active reference to superblock, acquiring a temprory one if
141 * there is no active references left. In that case we lock superblock,
142 * tell fs driver to shut it down and drop the temporary reference we
145 void deactivate_super(struct super_block *s)
147 struct file_system_type *fs = s->s_type;
148 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
149 s->s_count -= S_BIAS-1;
150 spin_unlock(&sb_lock);
151 down_write(&s->s_umount);
158 EXPORT_SYMBOL(deactivate_super);
161 * grab_super - acquire an active reference
162 * @s: reference we are trying to make active
164 * Tries to acquire an active reference. grab_super() is used when we
165 * had just found a superblock in super_blocks or fs_type->fs_supers
166 * and want to turn it into a full-blown active reference. grab_super()
167 * is called with sb_lock held and drops it. Returns 1 in case of
168 * success, 0 if we had failed (superblock contents was already dead or
169 * dying when grab_super() had been called).
171 static int grab_super(struct super_block *s)
174 spin_unlock(&sb_lock);
175 down_write(&s->s_umount);
178 if (s->s_count > S_BIAS) {
179 atomic_inc(&s->s_active);
181 spin_unlock(&sb_lock);
184 spin_unlock(&sb_lock);
186 up_write(&s->s_umount);
193 * generic_shutdown_super - common helper for ->kill_sb()
194 * @sb: superblock to kill
196 * generic_shutdown_super() does all fs-independent work on superblock
197 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
198 * that need destruction out of superblock, call generic_shutdown_super()
199 * and release aforementioned objects. Note: dentries and inodes _are_
200 * taken care of and do not need specific handling.
202 void generic_shutdown_super(struct super_block *sb)
204 struct dentry *root = sb->s_root;
205 struct super_operations *sop = sb->s_op;
209 shrink_dcache_parent(root);
210 shrink_dcache_anon(&sb->s_anon);
215 sb->s_flags &= ~MS_ACTIVE;
216 /* bad name - it should be evict_inodes() */
217 invalidate_inodes(sb);
219 if (sop->write_super && sb->s_dirt)
220 sop->write_super(sb);
224 /* Forget any remaining inodes */
225 if (invalidate_inodes(sb)) {
226 printk("VFS: Busy inodes after unmount. "
227 "Self-destruct in 5 seconds. Have a nice day...\n");
234 list_del(&sb->s_list);
235 list_del(&sb->s_instances);
236 spin_unlock(&sb_lock);
237 up_write(&sb->s_umount);
240 EXPORT_SYMBOL(generic_shutdown_super);
243 * sget - find or create a superblock
244 * @type: filesystem type superblock should belong to
245 * @test: comparison callback
246 * @set: setup callback
247 * @data: argument to each of them
249 struct super_block *sget(struct file_system_type *type,
250 int (*test)(struct super_block *,void *),
251 int (*set)(struct super_block *,void *),
254 struct super_block *s = NULL;
260 if (test) list_for_each(p, &type->fs_supers) {
261 struct super_block *old;
262 old = list_entry(p, struct super_block, s_instances);
263 if (!test(old, data))
265 if (!grab_super(old))
272 spin_unlock(&sb_lock);
275 return ERR_PTR(-ENOMEM);
281 spin_unlock(&sb_lock);
286 strlcpy(s->s_id, type->name, sizeof(s->s_id));
287 list_add(&s->s_list, super_blocks.prev);
288 list_add(&s->s_instances, &type->fs_supers);
289 spin_unlock(&sb_lock);
290 get_filesystem(type);
296 void drop_super(struct super_block *sb)
298 up_read(&sb->s_umount);
302 EXPORT_SYMBOL(drop_super);
304 static inline void write_super(struct super_block *sb)
307 if (sb->s_root && sb->s_dirt)
308 if (sb->s_op->write_super)
309 sb->s_op->write_super(sb);
314 * Note: check the dirty flag before waiting, so we don't
315 * hold up the sync while mounting a device. (The newly
316 * mounted device won't need syncing.)
318 void sync_supers(void)
320 struct super_block * sb;
323 sb = sb_entry(super_blocks.next);
324 while (sb != sb_entry(&super_blocks))
327 spin_unlock(&sb_lock);
328 down_read(&sb->s_umount);
333 sb = sb_entry(sb->s_list.next);
334 spin_unlock(&sb_lock);
338 * Call the ->sync_fs super_op against all filesytems which are r/w and
339 * which implement it.
341 * This operation is careful to avoid the livelock which could easily happen
342 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
343 * is used only here. We set it against all filesystems and then clear it as
344 * we sync them. So redirtied filesystems are skipped.
346 * But if process A is currently running sync_filesytems and then process B
347 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
348 * flags again, which will cause process A to resync everything. Fix that with
351 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
353 void sync_filesystems(int wait)
355 struct super_block *sb;
356 static DECLARE_MUTEX(mutex);
358 down(&mutex); /* Could be down_interruptible */
360 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
361 sb = sb_entry(sb->s_list.next)) {
362 if (!sb->s_op->sync_fs)
364 if (sb->s_flags & MS_RDONLY)
366 sb->s_need_sync_fs = 1;
368 spin_unlock(&sb_lock);
372 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
373 sb = sb_entry(sb->s_list.next)) {
374 if (!sb->s_need_sync_fs)
376 sb->s_need_sync_fs = 0;
377 if (sb->s_flags & MS_RDONLY)
378 continue; /* hm. Was remounted r/o meanwhile */
380 spin_unlock(&sb_lock);
381 down_read(&sb->s_umount);
382 if (sb->s_root && (wait || sb->s_dirt))
383 sb->s_op->sync_fs(sb, wait);
387 spin_unlock(&sb_lock);
392 * get_super - get the superblock of a device
393 * @bdev: device to get the superblock for
395 * Scans the superblock list and finds the superblock of the file system
396 * mounted on the device given. %NULL is returned if no match is found.
399 struct super_block * get_super(struct block_device *bdev)
406 list_for_each(p, &super_blocks) {
407 struct super_block *s = sb_entry(p);
408 if (s->s_bdev == bdev) {
410 spin_unlock(&sb_lock);
411 down_read(&s->s_umount);
418 spin_unlock(&sb_lock);
422 EXPORT_SYMBOL(get_super);
424 struct super_block * user_get_super(dev_t dev)
430 list_for_each(p, &super_blocks) {
431 struct super_block *s = sb_entry(p);
432 if (s->s_dev == dev) {
434 spin_unlock(&sb_lock);
435 down_read(&s->s_umount);
442 spin_unlock(&sb_lock);
446 EXPORT_SYMBOL(user_get_super);
448 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
450 struct super_block *s;
455 s = user_get_super(new_decode_dev(dev));
458 err = vfs_statfs(s, &sbuf);
463 memset(&tmp,0,sizeof(struct ustat));
464 tmp.f_tfree = sbuf.f_bfree;
465 tmp.f_tinode = sbuf.f_ffree;
467 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
474 * @sb: superblock in question
476 * All files are marked read/only. We don't care about pending
477 * delete files so this should be used in 'force' mode only
480 static void mark_files_ro(struct super_block *sb)
485 list_for_each_entry(f, &sb->s_files, f_list) {
486 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
487 f->f_mode &= ~FMODE_WRITE;
493 * do_remount_sb - asks filesystem to change mount options.
494 * @sb: superblock in question
495 * @flags: numeric part of options
496 * @data: the rest of options
497 * @force: whether or not to force the change
499 * Alters the mount options of a mounted file system.
501 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
505 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
507 if (flags & MS_RDONLY)
509 shrink_dcache_sb(sb);
512 /* If we are remounting RDONLY and current sb is read/write,
513 make sure there are no rw files opened */
514 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
517 else if (!fs_may_remount_ro(sb))
521 if (sb->s_op->remount_fs) {
523 retval = sb->s_op->remount_fs(sb, &flags, data);
528 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
532 static void do_emergency_remount(unsigned long foo)
534 struct super_block *sb;
537 list_for_each_entry(sb, &super_blocks, s_list) {
539 spin_unlock(&sb_lock);
540 down_read(&sb->s_umount);
541 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
543 * ->remount_fs needs lock_kernel().
545 * What lock protects sb->s_flags??
548 do_remount_sb(sb, MS_RDONLY, NULL, 1);
554 spin_unlock(&sb_lock);
555 printk("Emergency Remount complete\n");
558 void emergency_remount(void)
560 pdflush_operation(do_emergency_remount, 0);
564 * Unnamed block devices are dummy devices used by virtual
565 * filesystems which don't use real block-devices. -- jrs
568 static struct idr unnamed_dev_idr;
569 static spinlock_t unnamed_dev_lock = SPIN_LOCK_UNLOCKED;/* protects the above */
571 int set_anon_super(struct super_block *s, void *data)
577 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
579 spin_lock(&unnamed_dev_lock);
580 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
581 spin_unlock(&unnamed_dev_lock);
582 if (error == -EAGAIN)
583 /* We raced and lost with another CPU. */
588 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
589 spin_lock(&unnamed_dev_lock);
590 idr_remove(&unnamed_dev_idr, dev);
591 spin_unlock(&unnamed_dev_lock);
594 s->s_dev = MKDEV(0, dev & MINORMASK);
598 EXPORT_SYMBOL(set_anon_super);
600 void kill_anon_super(struct super_block *sb)
602 int slot = MINOR(sb->s_dev);
604 generic_shutdown_super(sb);
605 spin_lock(&unnamed_dev_lock);
606 idr_remove(&unnamed_dev_idr, slot);
607 spin_unlock(&unnamed_dev_lock);
610 EXPORT_SYMBOL(kill_anon_super);
612 void __init unnamed_dev_init(void)
614 idr_init(&unnamed_dev_idr);
617 void kill_litter_super(struct super_block *sb)
620 d_genocide(sb->s_root);
624 EXPORT_SYMBOL(kill_litter_super);
626 static int set_bdev_super(struct super_block *s, void *data)
629 s->s_dev = s->s_bdev->bd_dev;
633 static int test_bdev_super(struct super_block *s, void *data)
635 return (void *)s->s_bdev == data;
638 struct super_block *get_sb_bdev(struct file_system_type *fs_type,
639 int flags, const char *dev_name, void *data,
640 int (*fill_super)(struct super_block *, void *, int))
642 struct block_device *bdev;
643 struct super_block *s;
646 bdev = open_bdev_excl(dev_name, flags, fs_type);
648 return (struct super_block *)bdev;
651 * once the super is inserted into the list by sget, s_umount
652 * will protect the lockfs code from trying to start a snapshot
653 * while we are mounting
655 down(&bdev->bd_mount_sem);
656 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
657 up(&bdev->bd_mount_sem);
662 if ((flags ^ s->s_flags) & MS_RDONLY) {
663 up_write(&s->s_umount);
669 char b[BDEVNAME_SIZE];
672 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
673 s->s_old_blocksize = block_size(bdev);
674 sb_set_blocksize(s, s->s_old_blocksize);
675 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
677 up_write(&s->s_umount);
681 s->s_flags |= MS_ACTIVE;
687 close_bdev_excl(bdev);
691 EXPORT_SYMBOL(get_sb_bdev);
693 void kill_block_super(struct super_block *sb)
695 struct block_device *bdev = sb->s_bdev;
696 generic_shutdown_super(sb);
697 set_blocksize(bdev, sb->s_old_blocksize);
698 close_bdev_excl(bdev);
701 EXPORT_SYMBOL(kill_block_super);
703 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
704 int flags, void *data,
705 int (*fill_super)(struct super_block *, void *, int))
708 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
715 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
717 up_write(&s->s_umount);
719 return ERR_PTR(error);
721 s->s_flags |= MS_ACTIVE;
725 EXPORT_SYMBOL(get_sb_nodev);
727 static int compare_single(struct super_block *s, void *p)
732 struct super_block *get_sb_single(struct file_system_type *fs_type,
733 int flags, void *data,
734 int (*fill_super)(struct super_block *, void *, int))
736 struct super_block *s;
739 s = sget(fs_type, compare_single, set_anon_super, NULL);
744 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
746 up_write(&s->s_umount);
748 return ERR_PTR(error);
750 s->s_flags |= MS_ACTIVE;
752 do_remount_sb(s, flags, data, 0);
756 EXPORT_SYMBOL(get_sb_single);
759 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
761 struct file_system_type *type = get_fs_type(fstype);
762 struct super_block *sb = ERR_PTR(-ENOMEM);
763 struct vfsmount *mnt;
765 char *secdata = NULL;
768 return ERR_PTR(-ENODEV);
770 mnt = alloc_vfsmnt(name);
775 secdata = alloc_secdata();
777 sb = ERR_PTR(-ENOMEM);
781 error = security_sb_copy_data(type, data, secdata);
784 goto out_free_secdata;
788 sb = type->get_sb(type, flags, name, data);
790 goto out_free_secdata;
793 if (!capable(CAP_SYS_ADMIN) && !sb->s_bdev &&
794 (sb->s_magic != PROC_SUPER_MAGIC) &&
795 (sb->s_magic != DEVPTS_SUPER_MAGIC))
798 error = security_sb_kern_mount(sb, secdata);
802 mnt->mnt_root = dget(sb->s_root);
803 mnt->mnt_mountpoint = sb->s_root;
804 mnt->mnt_parent = mnt;
805 mnt->mnt_namespace = current->namespace;
806 up_write(&sb->s_umount);
807 put_filesystem(type);
810 up_write(&sb->s_umount);
811 deactivate_super(sb);
814 free_secdata(secdata);
818 put_filesystem(type);
819 return (struct vfsmount *)sb;
822 EXPORT_SYMBOL_GPL(do_kern_mount);
824 struct vfsmount *kern_mount(struct file_system_type *type)
826 return do_kern_mount(type->name, 0, type->name, NULL);
829 EXPORT_SYMBOL(kern_mount);