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/devpts_fs.h>
40 #include <linux/proc_fs.h>
41 #include <linux/kobject.h>
42 #include <linux/devpts_fs.h>
43 #include <linux/proc_fs.h>
44 #include <asm/uaccess.h>
47 void get_filesystem(struct file_system_type *fs);
48 void put_filesystem(struct file_system_type *fs);
49 struct file_system_type *get_fs_type(const char *name);
51 LIST_HEAD(super_blocks);
52 DEFINE_SPINLOCK(sb_lock);
55 * alloc_super - create new superblock
57 * Allocates and initializes a new &struct super_block. alloc_super()
58 * returns a pointer new superblock or %NULL if allocation had failed.
60 static struct super_block *alloc_super(void)
62 struct super_block *s = kmalloc(sizeof(struct super_block), GFP_USER);
63 static struct super_operations default_op;
66 memset(s, 0, sizeof(struct super_block));
67 if (security_sb_alloc(s)) {
72 INIT_LIST_HEAD(&s->s_dirty);
73 INIT_LIST_HEAD(&s->s_io);
74 INIT_LIST_HEAD(&s->s_files);
75 INIT_LIST_HEAD(&s->s_instances);
76 INIT_HLIST_HEAD(&s->s_anon);
77 INIT_LIST_HEAD(&s->s_inodes);
78 init_rwsem(&s->s_umount);
79 sema_init(&s->s_lock, 1);
80 down_write(&s->s_umount);
82 atomic_set(&s->s_active, 1);
83 sema_init(&s->s_vfs_rename_sem,1);
84 sema_init(&s->s_dquot.dqio_sem, 1);
85 sema_init(&s->s_dquot.dqonoff_sem, 1);
86 init_rwsem(&s->s_dquot.dqptr_sem);
87 init_waitqueue_head(&s->s_wait_unfrozen);
88 s->s_maxbytes = MAX_NON_LFS;
89 s->dq_op = sb_dquot_ops;
90 s->s_qcop = sb_quotactl_ops;
91 s->s_op = &default_op;
92 s->s_time_gran = 1000000000;
99 * destroy_super - frees a superblock
100 * @s: superblock to free
102 * Frees a superblock.
104 static inline void destroy_super(struct super_block *s)
110 /* Superblock refcounting */
113 * Drop a superblock's refcount. Returns non-zero if the superblock was
114 * destroyed. The caller must hold sb_lock.
116 int __put_super(struct super_block *sb)
120 if (!--sb->s_count) {
128 * Drop a superblock's refcount.
129 * Returns non-zero if the superblock is about to be destroyed and
130 * at least is already removed from super_blocks list, so if we are
131 * making a loop through super blocks then we need to restart.
132 * The caller must hold sb_lock.
134 int __put_super_and_need_restart(struct super_block *sb)
136 /* check for race with generic_shutdown_super() */
137 if (list_empty(&sb->s_list)) {
138 /* super block is removed, need to restart... */
142 /* can't be the last, since s_list is still in use */
144 BUG_ON(sb->s_count == 0);
149 * put_super - drop a temporary reference to superblock
150 * @sb: superblock in question
152 * Drops a temporary reference, frees superblock if there's no
155 static void put_super(struct super_block *sb)
159 spin_unlock(&sb_lock);
164 * deactivate_super - drop an active reference to superblock
165 * @s: superblock to deactivate
167 * Drops an active reference to superblock, acquiring a temprory one if
168 * there is no active references left. In that case we lock superblock,
169 * tell fs driver to shut it down and drop the temporary reference we
172 void deactivate_super(struct super_block *s)
174 struct file_system_type *fs = s->s_type;
175 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
176 s->s_count -= S_BIAS-1;
177 spin_unlock(&sb_lock);
178 down_write(&s->s_umount);
185 EXPORT_SYMBOL(deactivate_super);
188 * grab_super - acquire an active reference
189 * @s: reference we are trying to make active
191 * Tries to acquire an active reference. grab_super() is used when we
192 * had just found a superblock in super_blocks or fs_type->fs_supers
193 * and want to turn it into a full-blown active reference. grab_super()
194 * is called with sb_lock held and drops it. Returns 1 in case of
195 * success, 0 if we had failed (superblock contents was already dead or
196 * dying when grab_super() had been called).
198 static int grab_super(struct super_block *s)
201 spin_unlock(&sb_lock);
202 down_write(&s->s_umount);
205 if (s->s_count > S_BIAS) {
206 atomic_inc(&s->s_active);
208 spin_unlock(&sb_lock);
211 spin_unlock(&sb_lock);
213 up_write(&s->s_umount);
220 * generic_shutdown_super - common helper for ->kill_sb()
221 * @sb: superblock to kill
223 * generic_shutdown_super() does all fs-independent work on superblock
224 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
225 * that need destruction out of superblock, call generic_shutdown_super()
226 * and release aforementioned objects. Note: dentries and inodes _are_
227 * taken care of and do not need specific handling.
229 void generic_shutdown_super(struct super_block *sb)
231 struct dentry *root = sb->s_root;
232 struct super_operations *sop = sb->s_op;
236 shrink_dcache_parent(root);
237 shrink_dcache_anon(&sb->s_anon);
241 sb->s_flags &= ~MS_ACTIVE;
242 /* bad name - it should be evict_inodes() */
243 invalidate_inodes(sb);
246 if (sop->write_super && sb->s_dirt)
247 sop->write_super(sb);
251 /* Forget any remaining inodes */
252 if (invalidate_inodes(sb)) {
253 printk("VFS: Busy inodes after unmount. "
254 "Self-destruct in 5 seconds. Have a nice day...\n");
261 /* should be initialized for __put_super_and_need_restart() */
262 list_del_init(&sb->s_list);
263 list_del(&sb->s_instances);
264 spin_unlock(&sb_lock);
265 up_write(&sb->s_umount);
268 EXPORT_SYMBOL(generic_shutdown_super);
271 * sget - find or create a superblock
272 * @type: filesystem type superblock should belong to
273 * @test: comparison callback
274 * @set: setup callback
275 * @data: argument to each of them
277 struct super_block *sget(struct file_system_type *type,
278 int (*test)(struct super_block *,void *),
279 int (*set)(struct super_block *,void *),
282 struct super_block *s = NULL;
288 if (test) list_for_each(p, &type->fs_supers) {
289 struct super_block *old;
290 old = list_entry(p, struct super_block, s_instances);
291 if (!test(old, data))
293 if (!grab_super(old))
300 spin_unlock(&sb_lock);
303 return ERR_PTR(-ENOMEM);
309 spin_unlock(&sb_lock);
314 strlcpy(s->s_id, type->name, sizeof(s->s_id));
315 list_add_tail(&s->s_list, &super_blocks);
316 list_add(&s->s_instances, &type->fs_supers);
317 spin_unlock(&sb_lock);
318 get_filesystem(type);
324 void drop_super(struct super_block *sb)
326 up_read(&sb->s_umount);
330 EXPORT_SYMBOL(drop_super);
332 static inline void write_super(struct super_block *sb)
335 if (sb->s_root && sb->s_dirt)
336 if (sb->s_op->write_super)
337 sb->s_op->write_super(sb);
342 * Note: check the dirty flag before waiting, so we don't
343 * hold up the sync while mounting a device. (The newly
344 * mounted device won't need syncing.)
346 void sync_supers(void)
348 struct super_block * sb;
351 sb = sb_entry(super_blocks.next);
352 while (sb != sb_entry(&super_blocks))
355 spin_unlock(&sb_lock);
356 down_read(&sb->s_umount);
361 sb = sb_entry(sb->s_list.next);
362 spin_unlock(&sb_lock);
366 * Call the ->sync_fs super_op against all filesytems which are r/w and
367 * which implement it.
369 * This operation is careful to avoid the livelock which could easily happen
370 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
371 * is used only here. We set it against all filesystems and then clear it as
372 * we sync them. So redirtied filesystems are skipped.
374 * But if process A is currently running sync_filesytems and then process B
375 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
376 * flags again, which will cause process A to resync everything. Fix that with
379 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
381 void sync_filesystems(int wait)
383 struct super_block *sb;
384 static DECLARE_MUTEX(mutex);
386 down(&mutex); /* Could be down_interruptible */
388 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
389 sb = sb_entry(sb->s_list.next)) {
390 if (!sb->s_op->sync_fs)
392 if (sb->s_flags & MS_RDONLY)
394 sb->s_need_sync_fs = 1;
396 spin_unlock(&sb_lock);
400 for (sb = sb_entry(super_blocks.next); sb != sb_entry(&super_blocks);
401 sb = sb_entry(sb->s_list.next)) {
402 if (!sb->s_need_sync_fs)
404 sb->s_need_sync_fs = 0;
405 if (sb->s_flags & MS_RDONLY)
406 continue; /* hm. Was remounted r/o meanwhile */
408 spin_unlock(&sb_lock);
409 down_read(&sb->s_umount);
410 if (sb->s_root && (wait || sb->s_dirt))
411 sb->s_op->sync_fs(sb, wait);
415 spin_unlock(&sb_lock);
420 * get_super - get the superblock of a device
421 * @bdev: device to get the superblock for
423 * Scans the superblock list and finds the superblock of the file system
424 * mounted on the device given. %NULL is returned if no match is found.
427 struct super_block * get_super(struct block_device *bdev)
434 list_for_each(p, &super_blocks) {
435 struct super_block *s = sb_entry(p);
436 if (s->s_bdev == bdev) {
438 spin_unlock(&sb_lock);
439 down_read(&s->s_umount);
446 spin_unlock(&sb_lock);
450 EXPORT_SYMBOL(get_super);
452 struct super_block * user_get_super(dev_t dev)
458 list_for_each(p, &super_blocks) {
459 struct super_block *s = sb_entry(p);
460 if (s->s_dev == dev) {
462 spin_unlock(&sb_lock);
463 down_read(&s->s_umount);
470 spin_unlock(&sb_lock);
474 EXPORT_SYMBOL(user_get_super);
476 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
478 struct super_block *s;
483 s = user_get_super(new_decode_dev(dev));
486 err = vfs_statfs(s, &sbuf);
491 memset(&tmp,0,sizeof(struct ustat));
492 tmp.f_tfree = sbuf.f_bfree;
493 tmp.f_tinode = sbuf.f_ffree;
495 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
502 * @sb: superblock in question
504 * All files are marked read/only. We don't care about pending
505 * delete files so this should be used in 'force' mode only
508 static void mark_files_ro(struct super_block *sb)
513 list_for_each_entry(f, &sb->s_files, f_list) {
514 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
515 f->f_mode &= ~FMODE_WRITE;
521 * do_remount_sb - asks filesystem to change mount options.
522 * @sb: superblock in question
523 * @flags: numeric part of options
524 * @data: the rest of options
525 * @force: whether or not to force the change
527 * Alters the mount options of a mounted file system.
529 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
533 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
535 if (flags & MS_RDONLY)
537 shrink_dcache_sb(sb);
540 /* If we are remounting RDONLY and current sb is read/write,
541 make sure there are no rw files opened */
542 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
545 else if (!fs_may_remount_ro(sb))
549 if (sb->s_op->remount_fs) {
551 retval = sb->s_op->remount_fs(sb, &flags, data);
556 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
560 static void do_emergency_remount(unsigned long foo)
562 struct super_block *sb;
565 list_for_each_entry(sb, &super_blocks, s_list) {
567 spin_unlock(&sb_lock);
568 down_read(&sb->s_umount);
569 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
571 * ->remount_fs needs lock_kernel().
573 * What lock protects sb->s_flags??
576 do_remount_sb(sb, MS_RDONLY, NULL, 1);
582 spin_unlock(&sb_lock);
583 printk("Emergency Remount complete\n");
586 void emergency_remount(void)
588 pdflush_operation(do_emergency_remount, 0);
592 * Unnamed block devices are dummy devices used by virtual
593 * filesystems which don't use real block-devices. -- jrs
596 static struct idr unnamed_dev_idr;
597 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
599 int set_anon_super(struct super_block *s, void *data)
605 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
607 spin_lock(&unnamed_dev_lock);
608 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
609 spin_unlock(&unnamed_dev_lock);
610 if (error == -EAGAIN)
611 /* We raced and lost with another CPU. */
616 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
617 spin_lock(&unnamed_dev_lock);
618 idr_remove(&unnamed_dev_idr, dev);
619 spin_unlock(&unnamed_dev_lock);
622 s->s_dev = MKDEV(0, dev & MINORMASK);
626 EXPORT_SYMBOL(set_anon_super);
628 void kill_anon_super(struct super_block *sb)
630 int slot = MINOR(sb->s_dev);
632 generic_shutdown_super(sb);
633 spin_lock(&unnamed_dev_lock);
634 idr_remove(&unnamed_dev_idr, slot);
635 spin_unlock(&unnamed_dev_lock);
638 EXPORT_SYMBOL(kill_anon_super);
640 void __init unnamed_dev_init(void)
642 idr_init(&unnamed_dev_idr);
645 void kill_litter_super(struct super_block *sb)
648 d_genocide(sb->s_root);
652 EXPORT_SYMBOL(kill_litter_super);
654 static int set_bdev_super(struct super_block *s, void *data)
657 s->s_dev = s->s_bdev->bd_dev;
661 static int test_bdev_super(struct super_block *s, void *data)
663 return (void *)s->s_bdev == data;
666 static void bdev_uevent(struct block_device *bdev, enum kobject_action action)
670 kobject_uevent(&bdev->bd_part->kobj, action, NULL);
672 kobject_uevent(&bdev->bd_disk->kobj, action, NULL);
676 struct super_block *get_sb_bdev(struct file_system_type *fs_type,
677 int flags, const char *dev_name, void *data,
678 int (*fill_super)(struct super_block *, void *, int))
680 struct block_device *bdev;
681 struct super_block *s;
684 bdev = open_bdev_excl(dev_name, flags, fs_type);
686 return (struct super_block *)bdev;
689 * once the super is inserted into the list by sget, s_umount
690 * will protect the lockfs code from trying to start a snapshot
691 * while we are mounting
693 down(&bdev->bd_mount_sem);
694 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
695 up(&bdev->bd_mount_sem);
700 if ((flags ^ s->s_flags) & MS_RDONLY) {
701 up_write(&s->s_umount);
707 char b[BDEVNAME_SIZE];
710 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
711 s->s_old_blocksize = block_size(bdev);
712 sb_set_blocksize(s, s->s_old_blocksize);
713 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
715 up_write(&s->s_umount);
719 s->s_flags |= MS_ACTIVE;
720 bdev_uevent(bdev, KOBJ_MOUNT);
727 close_bdev_excl(bdev);
731 EXPORT_SYMBOL(get_sb_bdev);
733 void kill_block_super(struct super_block *sb)
735 struct block_device *bdev = sb->s_bdev;
737 bdev_uevent(bdev, KOBJ_UMOUNT);
738 generic_shutdown_super(sb);
740 close_bdev_excl(bdev);
743 EXPORT_SYMBOL(kill_block_super);
745 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
746 int flags, void *data,
747 int (*fill_super)(struct super_block *, void *, int))
750 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
757 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
759 up_write(&s->s_umount);
761 return ERR_PTR(error);
763 s->s_flags |= MS_ACTIVE;
767 EXPORT_SYMBOL(get_sb_nodev);
769 static int compare_single(struct super_block *s, void *p)
774 struct super_block *get_sb_single(struct file_system_type *fs_type,
775 int flags, void *data,
776 int (*fill_super)(struct super_block *, void *, int))
778 struct super_block *s;
781 s = sget(fs_type, compare_single, set_anon_super, NULL);
786 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
788 up_write(&s->s_umount);
790 return ERR_PTR(error);
792 s->s_flags |= MS_ACTIVE;
794 do_remount_sb(s, flags, data, 0);
798 EXPORT_SYMBOL(get_sb_single);
801 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
803 struct file_system_type *type = get_fs_type(fstype);
804 struct super_block *sb;
805 struct vfsmount *mnt;
807 char *secdata = NULL;
810 return ERR_PTR(-ENODEV);
812 sb = ERR_PTR(-EPERM);
813 if ((type->fs_flags & FS_BINARY_MOUNTDATA) &&
814 !capable(CAP_SYS_ADMIN) && !vx_ccaps(VXC_BINARY_MOUNT))
817 sb = ERR_PTR(-ENOMEM);
818 mnt = alloc_vfsmnt(name);
823 secdata = alloc_secdata();
825 sb = ERR_PTR(-ENOMEM);
829 error = security_sb_copy_data(type, data, secdata);
832 goto out_free_secdata;
836 sb = type->get_sb(type, flags, name, data);
838 goto out_free_secdata;
841 if (!capable(CAP_SYS_ADMIN) && !sb->s_bdev &&
842 (sb->s_magic != PROC_SUPER_MAGIC) &&
843 (sb->s_magic != DEVPTS_SUPER_MAGIC))
846 error = security_sb_kern_mount(sb, secdata);
850 mnt->mnt_root = dget(sb->s_root);
851 mnt->mnt_mountpoint = sb->s_root;
852 mnt->mnt_parent = mnt;
853 mnt->mnt_namespace = current->namespace;
854 up_write(&sb->s_umount);
855 put_filesystem(type);
858 up_write(&sb->s_umount);
859 deactivate_super(sb);
862 free_secdata(secdata);
866 put_filesystem(type);
867 return (struct vfsmount *)sb;
870 EXPORT_SYMBOL_GPL(do_kern_mount);
872 struct vfsmount *kern_mount(struct file_system_type *type)
874 return do_kern_mount(type->name, 0, type->name, NULL);
877 EXPORT_SYMBOL(kern_mount);