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 <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 DEFINE_SPINLOCK(sb_lock);
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_LIST_HEAD(&s->s_inodes);
74 init_rwsem(&s->s_umount);
75 sema_init(&s->s_lock, 1);
76 down_write(&s->s_umount);
78 atomic_set(&s->s_active, 1);
79 sema_init(&s->s_vfs_rename_sem,1);
80 sema_init(&s->s_dquot.dqio_sem, 1);
81 sema_init(&s->s_dquot.dqonoff_sem, 1);
82 init_rwsem(&s->s_dquot.dqptr_sem);
83 init_waitqueue_head(&s->s_wait_unfrozen);
84 s->s_maxbytes = MAX_NON_LFS;
85 s->dq_op = sb_dquot_ops;
86 s->s_qcop = sb_quotactl_ops;
87 s->s_op = &default_op;
88 s->s_time_gran = 1000000000;
95 * destroy_super - frees a superblock
96 * @s: superblock to free
100 static inline void destroy_super(struct super_block *s)
106 /* Superblock refcounting */
109 * Drop a superblock's refcount. Returns non-zero if the superblock was
110 * destroyed. The caller must hold sb_lock.
112 int __put_super(struct super_block *sb)
116 if (!--sb->s_count) {
124 * Drop a superblock's refcount.
125 * Returns non-zero if the superblock is about to be destroyed and
126 * at least is already removed from super_blocks list, so if we are
127 * making a loop through super blocks then we need to restart.
128 * The caller must hold sb_lock.
130 int __put_super_and_need_restart(struct super_block *sb)
132 /* check for race with generic_shutdown_super() */
133 if (list_empty(&sb->s_list)) {
134 /* super block is removed, need to restart... */
138 /* can't be the last, since s_list is still in use */
140 BUG_ON(sb->s_count == 0);
145 * put_super - drop a temporary reference to superblock
146 * @sb: superblock in question
148 * Drops a temporary reference, frees superblock if there's no
151 static void put_super(struct super_block *sb)
155 spin_unlock(&sb_lock);
160 * deactivate_super - drop an active reference to superblock
161 * @s: superblock to deactivate
163 * Drops an active reference to superblock, acquiring a temprory one if
164 * there is no active references left. In that case we lock superblock,
165 * tell fs driver to shut it down and drop the temporary reference we
168 void deactivate_super(struct super_block *s)
170 struct file_system_type *fs = s->s_type;
171 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
172 s->s_count -= S_BIAS-1;
173 spin_unlock(&sb_lock);
174 down_write(&s->s_umount);
181 EXPORT_SYMBOL(deactivate_super);
184 * grab_super - acquire an active reference
185 * @s: reference we are trying to make active
187 * Tries to acquire an active reference. grab_super() is used when we
188 * had just found a superblock in super_blocks or fs_type->fs_supers
189 * and want to turn it into a full-blown active reference. grab_super()
190 * is called with sb_lock held and drops it. Returns 1 in case of
191 * success, 0 if we had failed (superblock contents was already dead or
192 * dying when grab_super() had been called).
194 static int grab_super(struct super_block *s)
197 spin_unlock(&sb_lock);
198 down_write(&s->s_umount);
201 if (s->s_count > S_BIAS) {
202 atomic_inc(&s->s_active);
204 spin_unlock(&sb_lock);
207 spin_unlock(&sb_lock);
209 up_write(&s->s_umount);
216 * generic_shutdown_super - common helper for ->kill_sb()
217 * @sb: superblock to kill
219 * generic_shutdown_super() does all fs-independent work on superblock
220 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
221 * that need destruction out of superblock, call generic_shutdown_super()
222 * and release aforementioned objects. Note: dentries and inodes _are_
223 * taken care of and do not need specific handling.
225 void generic_shutdown_super(struct super_block *sb)
227 struct dentry *root = sb->s_root;
228 struct super_operations *sop = sb->s_op;
232 shrink_dcache_parent(root);
233 shrink_dcache_anon(&sb->s_anon);
237 sb->s_flags &= ~MS_ACTIVE;
238 /* bad name - it should be evict_inodes() */
239 invalidate_inodes(sb);
242 if (sop->write_super && sb->s_dirt)
243 sop->write_super(sb);
247 /* Forget any remaining inodes */
248 if (invalidate_inodes(sb)) {
249 printk("VFS: Busy inodes after unmount. "
250 "Self-destruct in 5 seconds. Have a nice day...\n");
257 /* should be initialized for __put_super_and_need_restart() */
258 list_del_init(&sb->s_list);
259 list_del(&sb->s_instances);
260 spin_unlock(&sb_lock);
261 up_write(&sb->s_umount);
264 EXPORT_SYMBOL(generic_shutdown_super);
267 * sget - find or create a superblock
268 * @type: filesystem type superblock should belong to
269 * @test: comparison callback
270 * @set: setup callback
271 * @data: argument to each of them
273 struct super_block *sget(struct file_system_type *type,
274 int (*test)(struct super_block *,void *),
275 int (*set)(struct super_block *,void *),
278 struct super_block *s = NULL;
284 if (test) list_for_each(p, &type->fs_supers) {
285 struct super_block *old;
286 old = list_entry(p, struct super_block, s_instances);
287 if (!test(old, data))
289 if (!grab_super(old))
296 spin_unlock(&sb_lock);
299 return ERR_PTR(-ENOMEM);
305 spin_unlock(&sb_lock);
310 strlcpy(s->s_id, type->name, sizeof(s->s_id));
311 list_add_tail(&s->s_list, &super_blocks);
312 list_add(&s->s_instances, &type->fs_supers);
313 spin_unlock(&sb_lock);
314 get_filesystem(type);
320 void drop_super(struct super_block *sb)
322 up_read(&sb->s_umount);
326 EXPORT_SYMBOL(drop_super);
328 static inline void write_super(struct super_block *sb)
331 if (sb->s_root && sb->s_dirt)
332 if (sb->s_op->write_super)
333 sb->s_op->write_super(sb);
338 * Note: check the dirty flag before waiting, so we don't
339 * hold up the sync while mounting a device. (The newly
340 * mounted device won't need syncing.)
342 void sync_supers(void)
344 struct super_block * sb;
347 sb = sb_entry(super_blocks.next);
348 while (sb != sb_entry(&super_blocks))
351 spin_unlock(&sb_lock);
352 down_read(&sb->s_umount);
357 sb = sb_entry(sb->s_list.next);
358 spin_unlock(&sb_lock);
362 * Call the ->sync_fs super_op against all filesytems which are r/w and
363 * which implement it.
365 * This operation is careful to avoid the livelock which could easily happen
366 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
367 * is used only here. We set it against all filesystems and then clear it as
368 * we sync them. So redirtied filesystems are skipped.
370 * But if process A is currently running sync_filesytems and then process B
371 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
372 * flags again, which will cause process A to resync everything. Fix that with
375 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
377 void sync_filesystems(int wait)
379 struct super_block *sb;
380 static DECLARE_MUTEX(mutex);
382 down(&mutex); /* Could be down_interruptible */
384 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
385 sb = sb_entry(sb->s_list.next)) {
386 if (!sb->s_op->sync_fs)
388 if (sb->s_flags & MS_RDONLY)
390 sb->s_need_sync_fs = 1;
392 spin_unlock(&sb_lock);
396 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
397 sb = sb_entry(sb->s_list.next)) {
398 if (!sb->s_need_sync_fs)
400 sb->s_need_sync_fs = 0;
401 if (sb->s_flags & MS_RDONLY)
402 continue; /* hm. Was remounted r/o meanwhile */
404 spin_unlock(&sb_lock);
405 down_read(&sb->s_umount);
406 if (sb->s_root && (wait || sb->s_dirt))
407 sb->s_op->sync_fs(sb, wait);
411 spin_unlock(&sb_lock);
416 * get_super - get the superblock of a device
417 * @bdev: device to get the superblock for
419 * Scans the superblock list and finds the superblock of the file system
420 * mounted on the device given. %NULL is returned if no match is found.
423 struct super_block * get_super(struct block_device *bdev)
430 list_for_each(p, &super_blocks) {
431 struct super_block *s = sb_entry(p);
432 if (s->s_bdev == bdev) {
434 spin_unlock(&sb_lock);
435 down_read(&s->s_umount);
442 spin_unlock(&sb_lock);
446 EXPORT_SYMBOL(get_super);
448 struct super_block * user_get_super(dev_t dev)
454 list_for_each(p, &super_blocks) {
455 struct super_block *s = sb_entry(p);
456 if (s->s_dev == dev) {
458 spin_unlock(&sb_lock);
459 down_read(&s->s_umount);
466 spin_unlock(&sb_lock);
470 EXPORT_SYMBOL(user_get_super);
472 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
474 struct super_block *s;
479 s = user_get_super(new_decode_dev(dev));
482 err = vfs_statfs(s, &sbuf);
487 memset(&tmp,0,sizeof(struct ustat));
488 tmp.f_tfree = sbuf.f_bfree;
489 tmp.f_tinode = sbuf.f_ffree;
491 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
498 * @sb: superblock in question
500 * All files are marked read/only. We don't care about pending
501 * delete files so this should be used in 'force' mode only
504 static void mark_files_ro(struct super_block *sb)
509 list_for_each_entry(f, &sb->s_files, f_list) {
510 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
511 f->f_mode &= ~FMODE_WRITE;
517 * do_remount_sb - asks filesystem to change mount options.
518 * @sb: superblock in question
519 * @flags: numeric part of options
520 * @data: the rest of options
521 * @force: whether or not to force the change
523 * Alters the mount options of a mounted file system.
525 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
529 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
531 if (flags & MS_RDONLY)
533 shrink_dcache_sb(sb);
536 /* If we are remounting RDONLY and current sb is read/write,
537 make sure there are no rw files opened */
538 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
541 else if (!fs_may_remount_ro(sb))
545 if (sb->s_op->remount_fs) {
547 retval = sb->s_op->remount_fs(sb, &flags, data);
552 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
556 static void do_emergency_remount(unsigned long foo)
558 struct super_block *sb;
561 list_for_each_entry(sb, &super_blocks, s_list) {
563 spin_unlock(&sb_lock);
564 down_read(&sb->s_umount);
565 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
567 * ->remount_fs needs lock_kernel().
569 * What lock protects sb->s_flags??
572 do_remount_sb(sb, MS_RDONLY, NULL, 1);
578 spin_unlock(&sb_lock);
579 printk("Emergency Remount complete\n");
582 void emergency_remount(void)
584 pdflush_operation(do_emergency_remount, 0);
588 * Unnamed block devices are dummy devices used by virtual
589 * filesystems which don't use real block-devices. -- jrs
592 static struct idr unnamed_dev_idr;
593 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
595 int set_anon_super(struct super_block *s, void *data)
601 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
603 spin_lock(&unnamed_dev_lock);
604 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
605 spin_unlock(&unnamed_dev_lock);
606 if (error == -EAGAIN)
607 /* We raced and lost with another CPU. */
612 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
613 spin_lock(&unnamed_dev_lock);
614 idr_remove(&unnamed_dev_idr, dev);
615 spin_unlock(&unnamed_dev_lock);
618 s->s_dev = MKDEV(0, dev & MINORMASK);
622 EXPORT_SYMBOL(set_anon_super);
624 void kill_anon_super(struct super_block *sb)
626 int slot = MINOR(sb->s_dev);
628 generic_shutdown_super(sb);
629 spin_lock(&unnamed_dev_lock);
630 idr_remove(&unnamed_dev_idr, slot);
631 spin_unlock(&unnamed_dev_lock);
634 EXPORT_SYMBOL(kill_anon_super);
636 void __init unnamed_dev_init(void)
638 idr_init(&unnamed_dev_idr);
641 void kill_litter_super(struct super_block *sb)
644 d_genocide(sb->s_root);
648 EXPORT_SYMBOL(kill_litter_super);
650 static int set_bdev_super(struct super_block *s, void *data)
653 s->s_dev = s->s_bdev->bd_dev;
657 static int test_bdev_super(struct super_block *s, void *data)
659 return (void *)s->s_bdev == data;
662 static void bdev_uevent(struct block_device *bdev, enum kobject_action action)
666 kobject_uevent(&bdev->bd_part->kobj, action, NULL);
668 kobject_uevent(&bdev->bd_disk->kobj, action, NULL);
672 struct super_block *get_sb_bdev(struct file_system_type *fs_type,
673 int flags, const char *dev_name, void *data,
674 int (*fill_super)(struct super_block *, void *, int))
676 struct block_device *bdev;
677 struct super_block *s;
680 bdev = open_bdev_excl(dev_name, flags, fs_type);
682 return (struct super_block *)bdev;
685 * once the super is inserted into the list by sget, s_umount
686 * will protect the lockfs code from trying to start a snapshot
687 * while we are mounting
689 down(&bdev->bd_mount_sem);
690 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
691 up(&bdev->bd_mount_sem);
696 if ((flags ^ s->s_flags) & MS_RDONLY) {
697 up_write(&s->s_umount);
703 char b[BDEVNAME_SIZE];
706 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
707 s->s_old_blocksize = block_size(bdev);
708 sb_set_blocksize(s, s->s_old_blocksize);
709 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
711 up_write(&s->s_umount);
715 s->s_flags |= MS_ACTIVE;
716 bdev_uevent(bdev, KOBJ_MOUNT);
723 close_bdev_excl(bdev);
727 EXPORT_SYMBOL(get_sb_bdev);
729 void kill_block_super(struct super_block *sb)
731 struct block_device *bdev = sb->s_bdev;
733 bdev_uevent(bdev, KOBJ_UMOUNT);
734 generic_shutdown_super(sb);
736 close_bdev_excl(bdev);
739 EXPORT_SYMBOL(kill_block_super);
741 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
742 int flags, void *data,
743 int (*fill_super)(struct super_block *, void *, int))
746 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
753 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
755 up_write(&s->s_umount);
757 return ERR_PTR(error);
759 s->s_flags |= MS_ACTIVE;
763 EXPORT_SYMBOL(get_sb_nodev);
765 static int compare_single(struct super_block *s, void *p)
770 struct super_block *get_sb_single(struct file_system_type *fs_type,
771 int flags, void *data,
772 int (*fill_super)(struct super_block *, void *, int))
774 struct super_block *s;
777 s = sget(fs_type, compare_single, set_anon_super, NULL);
782 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
784 up_write(&s->s_umount);
786 return ERR_PTR(error);
788 s->s_flags |= MS_ACTIVE;
790 do_remount_sb(s, flags, data, 0);
794 EXPORT_SYMBOL(get_sb_single);
797 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
799 struct file_system_type *type = get_fs_type(fstype);
800 struct super_block *sb = ERR_PTR(-ENOMEM);
801 struct vfsmount *mnt;
803 char *secdata = NULL;
806 return ERR_PTR(-ENODEV);
808 mnt = alloc_vfsmnt(name);
813 secdata = alloc_secdata();
815 sb = ERR_PTR(-ENOMEM);
819 error = security_sb_copy_data(type, data, secdata);
822 goto out_free_secdata;
826 sb = type->get_sb(type, flags, name, data);
828 goto out_free_secdata;
829 error = security_sb_kern_mount(sb, secdata);
833 mnt->mnt_root = dget(sb->s_root);
834 mnt->mnt_mountpoint = sb->s_root;
835 mnt->mnt_parent = mnt;
836 mnt->mnt_namespace = current->namespace;
837 up_write(&sb->s_umount);
838 put_filesystem(type);
841 up_write(&sb->s_umount);
842 deactivate_super(sb);
845 free_secdata(secdata);
849 put_filesystem(type);
850 return (struct vfsmount *)sb;
853 EXPORT_SYMBOL_GPL(do_kern_mount);
855 struct vfsmount *kern_mount(struct file_system_type *type)
857 return do_kern_mount(type->name, 0, type->name, NULL);
860 EXPORT_SYMBOL(kern_mount);