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);
175 down_write(&s->s_umount);
182 EXPORT_SYMBOL(deactivate_super);
185 * grab_super - acquire an active reference
186 * @s: reference we are trying to make active
188 * Tries to acquire an active reference. grab_super() is used when we
189 * had just found a superblock in super_blocks or fs_type->fs_supers
190 * and want to turn it into a full-blown active reference. grab_super()
191 * is called with sb_lock held and drops it. Returns 1 in case of
192 * success, 0 if we had failed (superblock contents was already dead or
193 * dying when grab_super() had been called).
195 static int grab_super(struct super_block *s)
198 spin_unlock(&sb_lock);
199 down_write(&s->s_umount);
202 if (s->s_count > S_BIAS) {
203 atomic_inc(&s->s_active);
205 spin_unlock(&sb_lock);
208 spin_unlock(&sb_lock);
210 up_write(&s->s_umount);
217 * generic_shutdown_super - common helper for ->kill_sb()
218 * @sb: superblock to kill
220 * generic_shutdown_super() does all fs-independent work on superblock
221 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
222 * that need destruction out of superblock, call generic_shutdown_super()
223 * and release aforementioned objects. Note: dentries and inodes _are_
224 * taken care of and do not need specific handling.
226 void generic_shutdown_super(struct super_block *sb)
228 struct dentry *root = sb->s_root;
229 struct super_operations *sop = sb->s_op;
233 shrink_dcache_parent(root);
234 shrink_dcache_anon(&sb->s_anon);
238 sb->s_flags &= ~MS_ACTIVE;
239 /* bad name - it should be evict_inodes() */
240 invalidate_inodes(sb);
243 if (sop->write_super && sb->s_dirt)
244 sop->write_super(sb);
248 /* Forget any remaining inodes */
249 if (invalidate_inodes(sb)) {
250 printk("VFS: Busy inodes after unmount. "
251 "Self-destruct in 5 seconds. Have a nice day...\n");
258 /* should be initialized for __put_super_and_need_restart() */
259 list_del_init(&sb->s_list);
260 list_del(&sb->s_instances);
261 spin_unlock(&sb_lock);
262 up_write(&sb->s_umount);
265 EXPORT_SYMBOL(generic_shutdown_super);
268 * sget - find or create a superblock
269 * @type: filesystem type superblock should belong to
270 * @test: comparison callback
271 * @set: setup callback
272 * @data: argument to each of them
274 struct super_block *sget(struct file_system_type *type,
275 int (*test)(struct super_block *,void *),
276 int (*set)(struct super_block *,void *),
279 struct super_block *s = NULL;
285 if (test) list_for_each(p, &type->fs_supers) {
286 struct super_block *old;
287 old = list_entry(p, struct super_block, s_instances);
288 if (!test(old, data))
290 if (!grab_super(old))
297 spin_unlock(&sb_lock);
300 return ERR_PTR(-ENOMEM);
306 spin_unlock(&sb_lock);
311 strlcpy(s->s_id, type->name, sizeof(s->s_id));
312 list_add_tail(&s->s_list, &super_blocks);
313 list_add(&s->s_instances, &type->fs_supers);
314 spin_unlock(&sb_lock);
315 get_filesystem(type);
321 void drop_super(struct super_block *sb)
323 up_read(&sb->s_umount);
327 EXPORT_SYMBOL(drop_super);
329 static inline void write_super(struct super_block *sb)
332 if (sb->s_root && sb->s_dirt)
333 if (sb->s_op->write_super)
334 sb->s_op->write_super(sb);
339 * Note: check the dirty flag before waiting, so we don't
340 * hold up the sync while mounting a device. (The newly
341 * mounted device won't need syncing.)
343 void sync_supers(void)
345 struct super_block *sb;
349 list_for_each_entry(sb, &super_blocks, s_list) {
352 spin_unlock(&sb_lock);
353 down_read(&sb->s_umount);
355 up_read(&sb->s_umount);
357 if (__put_super_and_need_restart(sb))
361 spin_unlock(&sb_lock);
365 * Call the ->sync_fs super_op against all filesytems which are r/w and
366 * which implement it.
368 * This operation is careful to avoid the livelock which could easily happen
369 * if two or more filesystems are being continuously dirtied. s_need_sync_fs
370 * is used only here. We set it against all filesystems and then clear it as
371 * we sync them. So redirtied filesystems are skipped.
373 * But if process A is currently running sync_filesytems and then process B
374 * calls sync_filesystems as well, process B will set all the s_need_sync_fs
375 * flags again, which will cause process A to resync everything. Fix that with
378 * (Fabian) Avoid sync_fs with clean fs & wait mode 0
380 void sync_filesystems(int wait)
382 struct super_block *sb;
383 static DECLARE_MUTEX(mutex);
385 down(&mutex); /* Could be down_interruptible */
387 list_for_each_entry(sb, &super_blocks, s_list) {
388 if (!sb->s_op->sync_fs)
390 if (sb->s_flags & MS_RDONLY)
392 sb->s_need_sync_fs = 1;
396 list_for_each_entry(sb, &super_blocks, s_list) {
397 if (!sb->s_need_sync_fs)
399 sb->s_need_sync_fs = 0;
400 if (sb->s_flags & MS_RDONLY)
401 continue; /* hm. Was remounted r/o meanwhile */
403 spin_unlock(&sb_lock);
404 down_read(&sb->s_umount);
405 if (sb->s_root && (wait || sb->s_dirt))
406 sb->s_op->sync_fs(sb, wait);
407 up_read(&sb->s_umount);
408 /* restart only when sb is no longer on the list */
410 if (__put_super_and_need_restart(sb))
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)
427 struct super_block *sb;
434 list_for_each_entry(sb, &super_blocks, s_list) {
435 if (sb->s_bdev == bdev) {
437 spin_unlock(&sb_lock);
438 down_read(&sb->s_umount);
441 up_read(&sb->s_umount);
442 /* restart only when sb is no longer on the list */
444 if (__put_super_and_need_restart(sb))
448 spin_unlock(&sb_lock);
452 EXPORT_SYMBOL(get_super);
454 struct super_block * user_get_super(dev_t dev)
456 struct super_block *sb;
460 list_for_each_entry(sb, &super_blocks, s_list) {
461 if (sb->s_dev == dev) {
463 spin_unlock(&sb_lock);
464 down_read(&sb->s_umount);
467 up_read(&sb->s_umount);
468 /* restart only when sb is no longer on the list */
470 if (__put_super_and_need_restart(sb))
474 spin_unlock(&sb_lock);
478 asmlinkage long sys_ustat(unsigned dev, struct ustat __user * ubuf)
480 struct super_block *s;
485 s = user_get_super(new_decode_dev(dev));
488 err = vfs_statfs(s, &sbuf);
493 memset(&tmp,0,sizeof(struct ustat));
494 tmp.f_tfree = sbuf.f_bfree;
495 tmp.f_tinode = sbuf.f_ffree;
497 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
504 * @sb: superblock in question
506 * All files are marked read/only. We don't care about pending
507 * delete files so this should be used in 'force' mode only
510 static void mark_files_ro(struct super_block *sb)
515 list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
516 if (S_ISREG(f->f_dentry->d_inode->i_mode) && file_count(f))
517 f->f_mode &= ~FMODE_WRITE;
523 * do_remount_sb - asks filesystem to change mount options.
524 * @sb: superblock in question
525 * @flags: numeric part of options
526 * @data: the rest of options
527 * @force: whether or not to force the change
529 * Alters the mount options of a mounted file system.
531 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
535 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
537 if (flags & MS_RDONLY)
539 shrink_dcache_sb(sb);
542 /* If we are remounting RDONLY and current sb is read/write,
543 make sure there are no rw files opened */
544 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
547 else if (!fs_may_remount_ro(sb))
551 if (sb->s_op->remount_fs) {
553 retval = sb->s_op->remount_fs(sb, &flags, data);
558 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
562 static void do_emergency_remount(unsigned long foo)
564 struct super_block *sb;
567 list_for_each_entry(sb, &super_blocks, s_list) {
569 spin_unlock(&sb_lock);
570 down_read(&sb->s_umount);
571 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
573 * ->remount_fs needs lock_kernel().
575 * What lock protects sb->s_flags??
578 do_remount_sb(sb, MS_RDONLY, NULL, 1);
584 spin_unlock(&sb_lock);
585 printk("Emergency Remount complete\n");
588 void emergency_remount(void)
590 pdflush_operation(do_emergency_remount, 0);
594 * Unnamed block devices are dummy devices used by virtual
595 * filesystems which don't use real block-devices. -- jrs
598 static struct idr unnamed_dev_idr;
599 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
601 int set_anon_super(struct super_block *s, void *data)
607 if (idr_pre_get(&unnamed_dev_idr, GFP_ATOMIC) == 0)
609 spin_lock(&unnamed_dev_lock);
610 error = idr_get_new(&unnamed_dev_idr, NULL, &dev);
611 spin_unlock(&unnamed_dev_lock);
612 if (error == -EAGAIN)
613 /* We raced and lost with another CPU. */
618 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
619 spin_lock(&unnamed_dev_lock);
620 idr_remove(&unnamed_dev_idr, dev);
621 spin_unlock(&unnamed_dev_lock);
624 s->s_dev = MKDEV(0, dev & MINORMASK);
628 EXPORT_SYMBOL(set_anon_super);
630 void kill_anon_super(struct super_block *sb)
632 int slot = MINOR(sb->s_dev);
634 generic_shutdown_super(sb);
635 spin_lock(&unnamed_dev_lock);
636 idr_remove(&unnamed_dev_idr, slot);
637 spin_unlock(&unnamed_dev_lock);
640 EXPORT_SYMBOL(kill_anon_super);
642 void __init unnamed_dev_init(void)
644 idr_init(&unnamed_dev_idr);
647 void kill_litter_super(struct super_block *sb)
650 d_genocide(sb->s_root);
654 EXPORT_SYMBOL(kill_litter_super);
656 static int set_bdev_super(struct super_block *s, void *data)
659 s->s_dev = s->s_bdev->bd_dev;
663 static int test_bdev_super(struct super_block *s, void *data)
665 return (void *)s->s_bdev == data;
668 struct super_block *get_sb_bdev(struct file_system_type *fs_type,
669 int flags, const char *dev_name, void *data,
670 int (*fill_super)(struct super_block *, void *, int))
672 struct block_device *bdev;
673 struct super_block *s;
676 bdev = open_bdev_excl(dev_name, flags, fs_type);
678 return (struct super_block *)bdev;
681 * once the super is inserted into the list by sget, s_umount
682 * will protect the lockfs code from trying to start a snapshot
683 * while we are mounting
685 down(&bdev->bd_mount_sem);
686 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
687 up(&bdev->bd_mount_sem);
692 if ((flags ^ s->s_flags) & MS_RDONLY) {
693 up_write(&s->s_umount);
699 char b[BDEVNAME_SIZE];
702 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
703 sb_set_blocksize(s, block_size(bdev));
704 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
706 up_write(&s->s_umount);
710 s->s_flags |= MS_ACTIVE;
716 close_bdev_excl(bdev);
720 EXPORT_SYMBOL(get_sb_bdev);
722 void kill_block_super(struct super_block *sb)
724 struct block_device *bdev = sb->s_bdev;
726 generic_shutdown_super(sb);
728 close_bdev_excl(bdev);
731 EXPORT_SYMBOL(kill_block_super);
733 struct super_block *get_sb_nodev(struct file_system_type *fs_type,
734 int flags, void *data,
735 int (*fill_super)(struct super_block *, void *, int))
738 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
745 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
747 up_write(&s->s_umount);
749 return ERR_PTR(error);
751 s->s_flags |= MS_ACTIVE;
755 EXPORT_SYMBOL(get_sb_nodev);
757 static int compare_single(struct super_block *s, void *p)
762 struct super_block *get_sb_single(struct file_system_type *fs_type,
763 int flags, void *data,
764 int (*fill_super)(struct super_block *, void *, int))
766 struct super_block *s;
769 s = sget(fs_type, compare_single, set_anon_super, NULL);
774 error = fill_super(s, data, flags & MS_VERBOSE ? 1 : 0);
776 up_write(&s->s_umount);
778 return ERR_PTR(error);
780 s->s_flags |= MS_ACTIVE;
782 do_remount_sb(s, flags, data, 0);
786 EXPORT_SYMBOL(get_sb_single);
789 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
791 struct file_system_type *type = get_fs_type(fstype);
792 struct super_block *sb = ERR_PTR(-ENOMEM);
793 struct vfsmount *mnt;
795 char *secdata = NULL;
798 return ERR_PTR(-ENODEV);
800 mnt = alloc_vfsmnt(name);
805 secdata = alloc_secdata();
807 sb = ERR_PTR(-ENOMEM);
811 error = security_sb_copy_data(type, data, secdata);
814 goto out_free_secdata;
818 sb = type->get_sb(type, flags, name, data);
820 goto out_free_secdata;
821 error = security_sb_kern_mount(sb, secdata);
825 mnt->mnt_root = dget(sb->s_root);
826 mnt->mnt_mountpoint = sb->s_root;
827 mnt->mnt_parent = mnt;
828 up_write(&sb->s_umount);
829 free_secdata(secdata);
830 put_filesystem(type);
833 up_write(&sb->s_umount);
834 deactivate_super(sb);
837 free_secdata(secdata);
841 put_filesystem(type);
842 return (struct vfsmount *)sb;
845 EXPORT_SYMBOL_GPL(do_kern_mount);
847 struct vfsmount *kern_mount(struct file_system_type *type)
849 return do_kern_mount(type->name, 0, type->name, NULL);
852 EXPORT_SYMBOL(kern_mount);