4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/config.h>
9 #include <linux/init.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/devfs_fs_kernel.h>
16 #include <linux/smp_lock.h>
17 #include <linux/highmem.h>
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/buffer_head.h>
22 #include <linux/mpage.h>
23 #include <linux/mount.h>
24 #include <linux/uio.h>
25 #include <linux/namei.h>
26 #include <asm/uaccess.h>
29 struct block_device bdev;
30 struct inode vfs_inode;
33 static inline struct bdev_inode *BDEV_I(struct inode *inode)
35 return container_of(inode, struct bdev_inode, vfs_inode);
38 inline struct block_device *I_BDEV(struct inode *inode)
40 return &BDEV_I(inode)->bdev;
43 EXPORT_SYMBOL(I_BDEV);
45 static sector_t max_block(struct block_device *bdev)
47 sector_t retval = ~((sector_t)0);
48 loff_t sz = i_size_read(bdev->bd_inode);
51 unsigned int size = block_size(bdev);
52 unsigned int sizebits = blksize_bits(size);
53 retval = (sz >> sizebits);
58 /* Kill _all_ buffers, dirty or not.. */
59 static void kill_bdev(struct block_device *bdev)
61 invalidate_bdev(bdev, 1);
62 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
65 int set_blocksize(struct block_device *bdev, int size)
69 /* Size must be a power of two, and between 512 and PAGE_SIZE */
70 if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
73 /* Size cannot be smaller than the size supported by the device */
74 if (size < bdev_hardsect_size(bdev))
77 oldsize = bdev->bd_block_size;
81 /* Ok, we're actually changing the blocksize.. */
83 bdev->bd_block_size = size;
84 bdev->bd_inode->i_blkbits = blksize_bits(size);
89 EXPORT_SYMBOL(set_blocksize);
91 int sb_set_blocksize(struct super_block *sb, int size)
94 if (set_blocksize(sb->s_bdev, size) < 0)
96 sb->s_blocksize = size;
97 for (bits = 9, size >>= 9; size >>= 1; bits++)
99 sb->s_blocksize_bits = bits;
100 return sb->s_blocksize;
103 EXPORT_SYMBOL(sb_set_blocksize);
105 int sb_min_blocksize(struct super_block *sb, int size)
107 int minsize = bdev_hardsect_size(sb->s_bdev);
110 return sb_set_blocksize(sb, size);
113 EXPORT_SYMBOL(sb_min_blocksize);
116 blkdev_get_block(struct inode *inode, sector_t iblock,
117 struct buffer_head *bh, int create)
119 if (iblock >= max_block(I_BDEV(inode))) {
124 * for reads, we're just trying to fill a partial page.
125 * return a hole, they will have to call get_block again
126 * before they can fill it, and they will get -EIO at that
131 bh->b_bdev = I_BDEV(inode);
132 bh->b_blocknr = iblock;
133 set_buffer_mapped(bh);
138 blkdev_get_blocks(struct inode *inode, sector_t iblock,
139 unsigned long max_blocks, struct buffer_head *bh, int create)
141 if ((iblock + max_blocks) > max_block(I_BDEV(inode)))
144 bh->b_bdev = I_BDEV(inode);
145 bh->b_blocknr = iblock;
146 bh->b_size = max_blocks << inode->i_blkbits;
147 set_buffer_mapped(bh);
152 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
153 loff_t offset, unsigned long nr_segs)
155 struct file *file = iocb->ki_filp;
156 struct inode *inode = file->f_mapping->host;
158 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
159 iov, offset, nr_segs, blkdev_get_blocks, NULL);
162 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
164 return block_write_full_page(page, blkdev_get_block, wbc);
167 static int blkdev_readpage(struct file * file, struct page * page)
169 return block_read_full_page(page, blkdev_get_block);
172 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
174 return block_prepare_write(page, from, to, blkdev_get_block);
177 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
179 return block_commit_write(page, from, to);
184 * for a block special file file->f_dentry->d_inode->i_size is zero
185 * so we compute the size by hand (just as in block_read/write above)
187 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
189 struct inode *bd_inode = file->f_mapping->host;
193 down(&bd_inode->i_sem);
194 size = i_size_read(bd_inode);
201 offset += file->f_pos;
204 if (offset >= 0 && offset <= size) {
205 if (offset != file->f_pos) {
206 file->f_pos = offset;
210 up(&bd_inode->i_sem);
215 * Filp is never NULL; the only case when ->fsync() is called with
216 * NULL first argument is nfsd_sync_dir() and that's not a directory.
219 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
221 return sync_blockdev(I_BDEV(filp->f_mapping->host));
228 static spinlock_t bdev_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
229 static kmem_cache_t * bdev_cachep;
231 static struct inode *bdev_alloc_inode(struct super_block *sb)
233 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
236 return &ei->vfs_inode;
239 static void bdev_destroy_inode(struct inode *inode)
241 struct bdev_inode *bdi = BDEV_I(inode);
243 bdi->bdev.bd_inode_backing_dev_info = NULL;
244 kmem_cache_free(bdev_cachep, bdi);
247 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
249 struct bdev_inode *ei = (struct bdev_inode *) foo;
250 struct block_device *bdev = &ei->bdev;
252 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
253 SLAB_CTOR_CONSTRUCTOR)
255 memset(bdev, 0, sizeof(*bdev));
256 sema_init(&bdev->bd_sem, 1);
257 sema_init(&bdev->bd_mount_sem, 1);
258 INIT_LIST_HEAD(&bdev->bd_inodes);
259 INIT_LIST_HEAD(&bdev->bd_list);
260 inode_init_once(&ei->vfs_inode);
264 static inline void __bd_forget(struct inode *inode)
266 list_del_init(&inode->i_devices);
267 inode->i_bdev = NULL;
268 inode->i_mapping = &inode->i_data;
271 static void bdev_clear_inode(struct inode *inode)
273 struct block_device *bdev = &BDEV_I(inode)->bdev;
275 spin_lock(&bdev_lock);
276 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
277 __bd_forget(list_entry(p, struct inode, i_devices));
279 list_del_init(&bdev->bd_list);
280 spin_unlock(&bdev_lock);
283 static struct super_operations bdev_sops = {
284 .statfs = simple_statfs,
285 .alloc_inode = bdev_alloc_inode,
286 .destroy_inode = bdev_destroy_inode,
287 .drop_inode = generic_delete_inode,
288 .clear_inode = bdev_clear_inode,
291 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
292 int flags, const char *dev_name, void *data)
294 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
297 static struct file_system_type bd_type = {
300 .kill_sb = kill_anon_super,
303 static struct vfsmount *bd_mnt;
304 struct super_block *blockdev_superblock;
306 void __init bdev_cache_init(void)
309 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
310 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
312 err = register_filesystem(&bd_type);
314 panic("Cannot register bdev pseudo-fs");
315 bd_mnt = kern_mount(&bd_type);
316 err = PTR_ERR(bd_mnt);
318 panic("Cannot create bdev pseudo-fs");
319 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
323 * Most likely _very_ bad one - but then it's hardly critical for small
324 * /dev and can be fixed when somebody will need really large one.
325 * Keep in mind that it will be fed through icache hash function too.
327 static inline unsigned long hash(dev_t dev)
329 return MAJOR(dev)+MINOR(dev);
332 static int bdev_test(struct inode *inode, void *data)
334 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
337 static int bdev_set(struct inode *inode, void *data)
339 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
343 static LIST_HEAD(all_bdevs);
345 struct block_device *bdget(dev_t dev)
347 struct block_device *bdev;
350 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
351 bdev_test, bdev_set, &dev);
356 bdev = &BDEV_I(inode)->bdev;
358 if (inode->i_state & I_NEW) {
359 bdev->bd_contains = NULL;
360 bdev->bd_inode = inode;
361 bdev->bd_block_size = (1 << inode->i_blkbits);
362 bdev->bd_part_count = 0;
363 bdev->bd_invalidated = 0;
364 inode->i_mode = S_IFBLK;
366 inode->i_bdev = bdev;
367 inode->i_data.a_ops = &def_blk_aops;
368 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
369 inode->i_data.backing_dev_info = &default_backing_dev_info;
370 spin_lock(&bdev_lock);
371 list_add(&bdev->bd_list, &all_bdevs);
372 spin_unlock(&bdev_lock);
373 unlock_new_inode(inode);
378 EXPORT_SYMBOL(bdget);
380 long nr_blockdev_pages(void)
384 spin_lock(&bdev_lock);
385 list_for_each(p, &all_bdevs) {
386 struct block_device *bdev;
387 bdev = list_entry(p, struct block_device, bd_list);
388 ret += bdev->bd_inode->i_mapping->nrpages;
390 spin_unlock(&bdev_lock);
394 void bdput(struct block_device *bdev)
396 iput(bdev->bd_inode);
399 EXPORT_SYMBOL(bdput);
401 static struct block_device *bd_acquire(struct inode *inode)
403 struct block_device *bdev;
404 spin_lock(&bdev_lock);
405 bdev = inode->i_bdev;
406 if (bdev && igrab(bdev->bd_inode)) {
407 spin_unlock(&bdev_lock);
410 spin_unlock(&bdev_lock);
411 bdev = bdget(inode->i_rdev);
413 spin_lock(&bdev_lock);
416 inode->i_bdev = bdev;
417 inode->i_mapping = bdev->bd_inode->i_mapping;
418 list_add(&inode->i_devices, &bdev->bd_inodes);
419 spin_unlock(&bdev_lock);
424 /* Call when you free inode */
426 void bd_forget(struct inode *inode)
428 spin_lock(&bdev_lock);
431 spin_unlock(&bdev_lock);
434 int bd_claim(struct block_device *bdev, void *holder)
437 spin_lock(&bdev_lock);
439 /* first decide result */
440 if (bdev->bd_holder == holder)
441 res = 0; /* already a holder */
442 else if (bdev->bd_holder != NULL)
443 res = -EBUSY; /* held by someone else */
444 else if (bdev->bd_contains == bdev)
445 res = 0; /* is a whole device which isn't held */
447 else if (bdev->bd_contains->bd_holder == bd_claim)
448 res = 0; /* is a partition of a device that is being partitioned */
449 else if (bdev->bd_contains->bd_holder != NULL)
450 res = -EBUSY; /* is a partition of a held device */
452 res = 0; /* is a partition of an un-held device */
454 /* now impose change */
456 /* note that for a whole device bd_holders
457 * will be incremented twice, and bd_holder will
458 * be set to bd_claim before being set to holder
460 bdev->bd_contains->bd_holders ++;
461 bdev->bd_contains->bd_holder = bd_claim;
463 bdev->bd_holder = holder;
465 spin_unlock(&bdev_lock);
469 EXPORT_SYMBOL(bd_claim);
471 void bd_release(struct block_device *bdev)
473 spin_lock(&bdev_lock);
474 if (!--bdev->bd_contains->bd_holders)
475 bdev->bd_contains->bd_holder = NULL;
476 if (!--bdev->bd_holders)
477 bdev->bd_holder = NULL;
478 spin_unlock(&bdev_lock);
481 EXPORT_SYMBOL(bd_release);
484 * Tries to open block device by device number. Use it ONLY if you
485 * really do not have anything better - i.e. when you are behind a
486 * truly sucky interface and all you are given is a device number. _Never_
487 * to be used for internal purposes. If you ever need it - reconsider
490 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
492 struct block_device *bdev = bdget(dev);
494 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
496 err = blkdev_get(bdev, mode, flags);
497 return err ? ERR_PTR(err) : bdev;
500 EXPORT_SYMBOL(open_by_devnum);
503 * This routine checks whether a removable media has been changed,
504 * and invalidates all buffer-cache-entries in that case. This
505 * is a relatively slow routine, so we have to try to minimize using
506 * it. Thus it is called only upon a 'mount' or 'open'. This
507 * is the best way of combining speed and utility, I think.
508 * People changing diskettes in the middle of an operation deserve
511 int check_disk_change(struct block_device *bdev)
513 struct gendisk *disk = bdev->bd_disk;
514 struct block_device_operations * bdops = disk->fops;
516 if (!bdops->media_changed)
518 if (!bdops->media_changed(bdev->bd_disk))
521 if (__invalidate_device(bdev, 0))
522 printk("VFS: busy inodes on changed media.\n");
524 if (bdops->revalidate_disk)
525 bdops->revalidate_disk(bdev->bd_disk);
526 if (bdev->bd_disk->minors > 1)
527 bdev->bd_invalidated = 1;
531 EXPORT_SYMBOL(check_disk_change);
533 void bd_set_size(struct block_device *bdev, loff_t size)
535 unsigned bsize = bdev_hardsect_size(bdev);
537 bdev->bd_inode->i_size = size;
538 while (bsize < PAGE_CACHE_SIZE) {
543 bdev->bd_block_size = bsize;
544 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
546 EXPORT_SYMBOL(bd_set_size);
548 static int do_open(struct block_device *bdev, struct file *file)
550 struct module *owner = NULL;
551 struct gendisk *disk;
555 file->f_mapping = bdev->bd_inode->i_mapping;
557 disk = get_gendisk(bdev->bd_dev, &part);
563 owner = disk->fops->owner;
566 if (!bdev->bd_openers) {
567 bdev->bd_disk = disk;
568 bdev->bd_contains = bdev;
570 struct backing_dev_info *bdi;
571 if (disk->fops->open) {
572 ret = disk->fops->open(bdev->bd_inode, file);
576 if (!bdev->bd_openers) {
577 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
578 bdi = blk_get_backing_dev_info(bdev);
580 bdi = &default_backing_dev_info;
581 bdev->bd_inode->i_data.backing_dev_info = bdi;
583 if (bdev->bd_invalidated)
584 rescan_partitions(disk, bdev);
587 struct block_device *whole;
588 whole = bdget_disk(disk, 0);
592 ret = blkdev_get(whole, file->f_mode, file->f_flags);
595 bdev->bd_contains = whole;
596 down(&whole->bd_sem);
597 whole->bd_part_count++;
598 p = disk->part[part - 1];
599 bdev->bd_inode->i_data.backing_dev_info =
600 whole->bd_inode->i_data.backing_dev_info;
601 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
602 whole->bd_part_count--;
607 kobject_get(&p->kobj);
609 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
615 if (bdev->bd_contains == bdev) {
616 if (bdev->bd_disk->fops->open) {
617 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
621 if (bdev->bd_invalidated)
622 rescan_partitions(bdev->bd_disk, bdev);
624 down(&bdev->bd_contains->bd_sem);
625 bdev->bd_contains->bd_part_count++;
626 up(&bdev->bd_contains->bd_sem);
635 bdev->bd_disk = NULL;
636 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
637 if (bdev != bdev->bd_contains)
638 blkdev_put(bdev->bd_contains);
639 bdev->bd_contains = NULL;
650 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
653 * This crockload is due to bad choice of ->open() type.
655 * For now, block device ->open() routine must _not_
656 * examine anything in 'inode' argument except ->i_rdev.
658 struct file fake_file = {};
659 struct dentry fake_dentry = {};
660 fake_file.f_mode = mode;
661 fake_file.f_flags = flags;
662 fake_file.f_dentry = &fake_dentry;
663 fake_dentry.d_inode = bdev->bd_inode;
665 return do_open(bdev, &fake_file);
668 EXPORT_SYMBOL(blkdev_get);
670 int blkdev_open(struct inode * inode, struct file * filp)
672 struct block_device *bdev;
676 * Preserve backwards compatibility and allow large file access
677 * even if userspace doesn't ask for it explicitly. Some mkfs
678 * binary needs it. We might want to drop this workaround
679 * during an unstable branch.
681 filp->f_flags |= O_LARGEFILE;
683 bdev = bd_acquire(inode);
685 res = do_open(bdev, filp);
689 if (!(filp->f_flags & O_EXCL) )
692 if (!(res = bd_claim(bdev, filp)))
699 EXPORT_SYMBOL(blkdev_open);
701 int blkdev_put(struct block_device *bdev)
704 struct inode *bd_inode = bdev->bd_inode;
705 struct gendisk *disk = bdev->bd_disk;
709 if (!--bdev->bd_openers) {
713 if (bdev->bd_contains == bdev) {
714 if (disk->fops->release)
715 ret = disk->fops->release(bd_inode, NULL);
717 down(&bdev->bd_contains->bd_sem);
718 bdev->bd_contains->bd_part_count--;
719 up(&bdev->bd_contains->bd_sem);
721 if (!bdev->bd_openers) {
722 struct module *owner = disk->fops->owner;
727 if (bdev->bd_contains != bdev) {
728 kobject_put(&bdev->bd_part->kobj);
729 bdev->bd_part = NULL;
731 bdev->bd_disk = NULL;
732 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
733 if (bdev != bdev->bd_contains) {
734 blkdev_put(bdev->bd_contains);
736 bdev->bd_contains = NULL;
744 EXPORT_SYMBOL(blkdev_put);
746 static int blkdev_close(struct inode * inode, struct file * filp)
748 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
749 if (bdev->bd_holder == filp)
751 return blkdev_put(bdev);
754 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
755 size_t count, loff_t *ppos)
757 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
759 return generic_file_write_nolock(file, &local_iov, 1, ppos);
762 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
763 size_t count, loff_t pos)
765 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
767 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
770 static int block_ioctl(struct inode *inode, struct file *file, unsigned cmd,
773 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
776 struct address_space_operations def_blk_aops = {
777 .readpage = blkdev_readpage,
778 .writepage = blkdev_writepage,
779 .sync_page = block_sync_page,
780 .prepare_write = blkdev_prepare_write,
781 .commit_write = blkdev_commit_write,
782 .writepages = generic_writepages,
783 .direct_IO = blkdev_direct_IO,
786 struct file_operations def_blk_fops = {
788 .release = blkdev_close,
789 .llseek = block_llseek,
790 .read = generic_file_read,
791 .write = blkdev_file_write,
792 .aio_read = generic_file_aio_read,
793 .aio_write = blkdev_file_aio_write,
794 .mmap = generic_file_mmap,
795 .fsync = block_fsync,
796 .ioctl = block_ioctl,
797 .readv = generic_file_readv,
798 .writev = generic_file_write_nolock,
799 .sendfile = generic_file_sendfile,
802 EXPORT_SYMBOL(def_blk_fops);
804 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
807 mm_segment_t old_fs = get_fs();
809 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
814 EXPORT_SYMBOL(ioctl_by_bdev);
817 * lookup_bdev - lookup a struct block_device by name
819 * @path: special file representing the block device
821 * Get a reference to the blockdevice at @path in the current
822 * namespace if possible and return it. Return ERR_PTR(error)
825 struct block_device *lookup_bdev(const char *path)
827 struct block_device *bdev;
833 return ERR_PTR(-EINVAL);
835 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
837 return ERR_PTR(error);
839 inode = nd.dentry->d_inode;
841 if (!S_ISBLK(inode->i_mode))
844 if (nd.mnt->mnt_flags & MNT_NODEV)
847 bdev = bd_acquire(inode);
854 bdev = ERR_PTR(error);
859 * open_bdev_excl - open a block device by name and set it up for use
861 * @path: special file representing the block device
862 * @flags: %MS_RDONLY for opening read-only
863 * @holder: owner for exclusion
865 * Open the blockdevice described by the special file at @path, claim it
868 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
870 struct block_device *bdev;
871 mode_t mode = FMODE_READ;
874 bdev = lookup_bdev(path);
878 if (!(flags & MS_RDONLY))
880 error = blkdev_get(bdev, mode, 0);
882 return ERR_PTR(error);
884 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
886 error = bd_claim(bdev, holder);
894 return ERR_PTR(error);
897 EXPORT_SYMBOL(open_bdev_excl);
900 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
902 * @bdev: blockdevice to close
904 * This is the counterpart to open_bdev_excl().
906 void close_bdev_excl(struct block_device *bdev)
912 EXPORT_SYMBOL(close_bdev_excl);