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)
67 /* Size must be a power of two, and between 512 and PAGE_SIZE */
68 if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
71 /* Size cannot be smaller than the size supported by the device */
72 if (size < bdev_hardsect_size(bdev))
75 /* Don't change the size if it is same as current */
76 if (bdev->bd_block_size != size) {
78 bdev->bd_block_size = size;
79 bdev->bd_inode->i_blkbits = blksize_bits(size);
85 EXPORT_SYMBOL(set_blocksize);
87 int sb_set_blocksize(struct super_block *sb, int size)
89 int bits = 9; /* 2^9 = 512 */
91 if (set_blocksize(sb->s_bdev, size))
93 /* If we get here, we know size is power of two
94 * and it's value is between 512 and PAGE_SIZE */
95 sb->s_blocksize = size;
96 for (size >>= 10; size; size >>= 1)
98 sb->s_blocksize_bits = bits;
99 return sb->s_blocksize;
102 EXPORT_SYMBOL(sb_set_blocksize);
104 int sb_min_blocksize(struct super_block *sb, int size)
106 int minsize = bdev_hardsect_size(sb->s_bdev);
109 return sb_set_blocksize(sb, size);
112 EXPORT_SYMBOL(sb_min_blocksize);
115 blkdev_get_block(struct inode *inode, sector_t iblock,
116 struct buffer_head *bh, int create)
118 if (iblock >= max_block(I_BDEV(inode))) {
123 * for reads, we're just trying to fill a partial page.
124 * return a hole, they will have to call get_block again
125 * before they can fill it, and they will get -EIO at that
130 bh->b_bdev = I_BDEV(inode);
131 bh->b_blocknr = iblock;
132 set_buffer_mapped(bh);
137 blkdev_get_blocks(struct inode *inode, sector_t iblock,
138 unsigned long max_blocks, struct buffer_head *bh, int create)
140 sector_t end_block = max_block(I_BDEV(inode));
142 if ((iblock + max_blocks) > end_block) {
143 max_blocks = end_block - iblock;
144 if ((long)max_blocks <= 0) {
146 return -EIO; /* write fully beyond EOF */
148 * It is a read which is fully beyond EOF. We return
149 * a !buffer_mapped buffer
155 bh->b_bdev = I_BDEV(inode);
156 bh->b_blocknr = iblock;
157 bh->b_size = max_blocks << inode->i_blkbits;
159 set_buffer_mapped(bh);
164 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
165 loff_t offset, unsigned long nr_segs)
167 struct file *file = iocb->ki_filp;
168 struct inode *inode = file->f_mapping->host;
170 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
171 iov, offset, nr_segs, blkdev_get_blocks, NULL);
174 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
176 return block_write_full_page(page, blkdev_get_block, wbc);
179 static int blkdev_readpage(struct file * file, struct page * page)
181 return block_read_full_page(page, blkdev_get_block);
184 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
186 return block_prepare_write(page, from, to, blkdev_get_block);
189 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
191 return block_commit_write(page, from, to);
196 * for a block special file file->f_dentry->d_inode->i_size is zero
197 * so we compute the size by hand (just as in block_read/write above)
199 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
201 struct inode *bd_inode = file->f_mapping->host;
205 down(&bd_inode->i_sem);
206 size = i_size_read(bd_inode);
213 offset += file->f_pos;
216 if (offset >= 0 && offset <= size) {
217 if (offset != file->f_pos) {
218 file->f_pos = offset;
222 up(&bd_inode->i_sem);
227 * Filp is never NULL; the only case when ->fsync() is called with
228 * NULL first argument is nfsd_sync_dir() and that's not a directory.
231 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
233 return sync_blockdev(I_BDEV(filp->f_mapping->host));
240 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
241 static kmem_cache_t * bdev_cachep;
243 static struct inode *bdev_alloc_inode(struct super_block *sb)
245 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
248 return &ei->vfs_inode;
251 static void bdev_destroy_inode(struct inode *inode)
253 struct bdev_inode *bdi = BDEV_I(inode);
255 bdi->bdev.bd_inode_backing_dev_info = NULL;
256 kmem_cache_free(bdev_cachep, bdi);
259 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
261 struct bdev_inode *ei = (struct bdev_inode *) foo;
262 struct block_device *bdev = &ei->bdev;
264 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
265 SLAB_CTOR_CONSTRUCTOR)
267 memset(bdev, 0, sizeof(*bdev));
268 sema_init(&bdev->bd_sem, 1);
269 sema_init(&bdev->bd_mount_sem, 1);
270 INIT_LIST_HEAD(&bdev->bd_inodes);
271 INIT_LIST_HEAD(&bdev->bd_list);
272 inode_init_once(&ei->vfs_inode);
276 static inline void __bd_forget(struct inode *inode)
278 list_del_init(&inode->i_devices);
279 inode->i_bdev = NULL;
280 inode->i_mapping = &inode->i_data;
283 static void bdev_clear_inode(struct inode *inode)
285 struct block_device *bdev = &BDEV_I(inode)->bdev;
287 spin_lock(&bdev_lock);
288 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
289 __bd_forget(list_entry(p, struct inode, i_devices));
291 list_del_init(&bdev->bd_list);
292 spin_unlock(&bdev_lock);
295 static struct super_operations bdev_sops = {
296 .statfs = simple_statfs,
297 .alloc_inode = bdev_alloc_inode,
298 .destroy_inode = bdev_destroy_inode,
299 .drop_inode = generic_delete_inode,
300 .clear_inode = bdev_clear_inode,
303 static struct super_block *bd_get_sb(struct file_system_type *fs_type,
304 int flags, const char *dev_name, void *data)
306 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
309 static struct file_system_type bd_type = {
312 .kill_sb = kill_anon_super,
315 static struct vfsmount *bd_mnt;
316 struct super_block *blockdev_superblock;
318 void __init bdev_cache_init(void)
321 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
322 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
324 err = register_filesystem(&bd_type);
326 panic("Cannot register bdev pseudo-fs");
327 bd_mnt = kern_mount(&bd_type);
328 err = PTR_ERR(bd_mnt);
330 panic("Cannot create bdev pseudo-fs");
331 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
335 * Most likely _very_ bad one - but then it's hardly critical for small
336 * /dev and can be fixed when somebody will need really large one.
337 * Keep in mind that it will be fed through icache hash function too.
339 static inline unsigned long hash(dev_t dev)
341 return MAJOR(dev)+MINOR(dev);
344 static int bdev_test(struct inode *inode, void *data)
346 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
349 static int bdev_set(struct inode *inode, void *data)
351 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
355 static LIST_HEAD(all_bdevs);
357 struct block_device *bdget(dev_t dev)
359 struct block_device *bdev;
362 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
363 bdev_test, bdev_set, &dev);
368 bdev = &BDEV_I(inode)->bdev;
370 if (inode->i_state & I_NEW) {
371 bdev->bd_contains = NULL;
372 bdev->bd_inode = inode;
373 bdev->bd_block_size = (1 << inode->i_blkbits);
374 bdev->bd_part_count = 0;
375 bdev->bd_invalidated = 0;
376 inode->i_mode = S_IFBLK;
378 inode->i_bdev = bdev;
379 inode->i_data.a_ops = &def_blk_aops;
380 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
381 inode->i_data.backing_dev_info = &default_backing_dev_info;
382 spin_lock(&bdev_lock);
383 list_add(&bdev->bd_list, &all_bdevs);
384 spin_unlock(&bdev_lock);
385 unlock_new_inode(inode);
390 EXPORT_SYMBOL(bdget);
392 long nr_blockdev_pages(void)
396 spin_lock(&bdev_lock);
397 list_for_each(p, &all_bdevs) {
398 struct block_device *bdev;
399 bdev = list_entry(p, struct block_device, bd_list);
400 ret += bdev->bd_inode->i_mapping->nrpages;
402 spin_unlock(&bdev_lock);
406 void bdput(struct block_device *bdev)
408 iput(bdev->bd_inode);
411 EXPORT_SYMBOL(bdput);
413 static struct block_device *bd_acquire(struct inode *inode)
415 struct block_device *bdev;
416 spin_lock(&bdev_lock);
417 bdev = inode->i_bdev;
418 if (bdev && igrab(bdev->bd_inode)) {
419 spin_unlock(&bdev_lock);
422 spin_unlock(&bdev_lock);
423 bdev = bdget(inode->i_rdev);
425 spin_lock(&bdev_lock);
428 inode->i_bdev = bdev;
429 inode->i_mapping = bdev->bd_inode->i_mapping;
430 list_add(&inode->i_devices, &bdev->bd_inodes);
431 spin_unlock(&bdev_lock);
436 /* Call when you free inode */
438 void bd_forget(struct inode *inode)
440 spin_lock(&bdev_lock);
443 spin_unlock(&bdev_lock);
446 int bd_claim(struct block_device *bdev, void *holder)
449 spin_lock(&bdev_lock);
451 /* first decide result */
452 if (bdev->bd_holder == holder)
453 res = 0; /* already a holder */
454 else if (bdev->bd_holder != NULL)
455 res = -EBUSY; /* held by someone else */
456 else if (bdev->bd_contains == bdev)
457 res = 0; /* is a whole device which isn't held */
459 else if (bdev->bd_contains->bd_holder == bd_claim)
460 res = 0; /* is a partition of a device that is being partitioned */
461 else if (bdev->bd_contains->bd_holder != NULL)
462 res = -EBUSY; /* is a partition of a held device */
464 res = 0; /* is a partition of an un-held device */
466 /* now impose change */
468 /* note that for a whole device bd_holders
469 * will be incremented twice, and bd_holder will
470 * be set to bd_claim before being set to holder
472 bdev->bd_contains->bd_holders ++;
473 bdev->bd_contains->bd_holder = bd_claim;
475 bdev->bd_holder = holder;
477 spin_unlock(&bdev_lock);
481 EXPORT_SYMBOL(bd_claim);
483 void bd_release(struct block_device *bdev)
485 spin_lock(&bdev_lock);
486 if (!--bdev->bd_contains->bd_holders)
487 bdev->bd_contains->bd_holder = NULL;
488 if (!--bdev->bd_holders)
489 bdev->bd_holder = NULL;
490 spin_unlock(&bdev_lock);
493 EXPORT_SYMBOL(bd_release);
496 * Tries to open block device by device number. Use it ONLY if you
497 * really do not have anything better - i.e. when you are behind a
498 * truly sucky interface and all you are given is a device number. _Never_
499 * to be used for internal purposes. If you ever need it - reconsider
502 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
504 struct block_device *bdev = bdget(dev);
506 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
508 err = blkdev_get(bdev, mode, flags);
509 return err ? ERR_PTR(err) : bdev;
512 EXPORT_SYMBOL(open_by_devnum);
515 * This routine checks whether a removable media has been changed,
516 * and invalidates all buffer-cache-entries in that case. This
517 * is a relatively slow routine, so we have to try to minimize using
518 * it. Thus it is called only upon a 'mount' or 'open'. This
519 * is the best way of combining speed and utility, I think.
520 * People changing diskettes in the middle of an operation deserve
523 int check_disk_change(struct block_device *bdev)
525 struct gendisk *disk = bdev->bd_disk;
526 struct block_device_operations * bdops = disk->fops;
528 if (!bdops->media_changed)
530 if (!bdops->media_changed(bdev->bd_disk))
533 if (__invalidate_device(bdev))
534 printk("VFS: busy inodes on changed media.\n");
536 if (bdops->revalidate_disk)
537 bdops->revalidate_disk(bdev->bd_disk);
538 if (bdev->bd_disk->minors > 1)
539 bdev->bd_invalidated = 1;
543 EXPORT_SYMBOL(check_disk_change);
545 void bd_set_size(struct block_device *bdev, loff_t size)
547 unsigned bsize = bdev_hardsect_size(bdev);
549 bdev->bd_inode->i_size = size;
550 while (bsize < PAGE_CACHE_SIZE) {
555 bdev->bd_block_size = bsize;
556 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
558 EXPORT_SYMBOL(bd_set_size);
560 static int do_open(struct block_device *bdev, struct file *file)
562 struct module *owner = NULL;
563 struct gendisk *disk;
567 file->f_mapping = bdev->bd_inode->i_mapping;
569 disk = get_gendisk(bdev->bd_dev, &part);
575 owner = disk->fops->owner;
578 if (!bdev->bd_openers) {
579 bdev->bd_disk = disk;
580 bdev->bd_contains = bdev;
582 struct backing_dev_info *bdi;
583 if (disk->fops->open) {
584 ret = disk->fops->open(bdev->bd_inode, file);
588 if (!bdev->bd_openers) {
589 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
590 bdi = blk_get_backing_dev_info(bdev);
592 bdi = &default_backing_dev_info;
593 bdev->bd_inode->i_data.backing_dev_info = bdi;
595 if (bdev->bd_invalidated)
596 rescan_partitions(disk, bdev);
599 struct block_device *whole;
600 whole = bdget_disk(disk, 0);
604 ret = blkdev_get(whole, file->f_mode, file->f_flags);
607 bdev->bd_contains = whole;
608 down(&whole->bd_sem);
609 whole->bd_part_count++;
610 p = disk->part[part - 1];
611 bdev->bd_inode->i_data.backing_dev_info =
612 whole->bd_inode->i_data.backing_dev_info;
613 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
614 whole->bd_part_count--;
619 kobject_get(&p->kobj);
621 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
627 if (bdev->bd_contains == bdev) {
628 if (bdev->bd_disk->fops->open) {
629 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
633 if (bdev->bd_invalidated)
634 rescan_partitions(bdev->bd_disk, bdev);
636 down(&bdev->bd_contains->bd_sem);
637 bdev->bd_contains->bd_part_count++;
638 up(&bdev->bd_contains->bd_sem);
647 bdev->bd_disk = NULL;
648 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
649 if (bdev != bdev->bd_contains)
650 blkdev_put(bdev->bd_contains);
651 bdev->bd_contains = NULL;
662 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
665 * This crockload is due to bad choice of ->open() type.
667 * For now, block device ->open() routine must _not_
668 * examine anything in 'inode' argument except ->i_rdev.
670 struct file fake_file = {};
671 struct dentry fake_dentry = {};
672 fake_file.f_mode = mode;
673 fake_file.f_flags = flags;
674 fake_file.f_dentry = &fake_dentry;
675 fake_dentry.d_inode = bdev->bd_inode;
677 return do_open(bdev, &fake_file);
680 EXPORT_SYMBOL(blkdev_get);
682 static int blkdev_open(struct inode * inode, struct file * filp)
684 struct block_device *bdev;
688 * Preserve backwards compatibility and allow large file access
689 * even if userspace doesn't ask for it explicitly. Some mkfs
690 * binary needs it. We might want to drop this workaround
691 * during an unstable branch.
693 filp->f_flags |= O_LARGEFILE;
695 bdev = bd_acquire(inode);
697 res = do_open(bdev, filp);
701 if (!(filp->f_flags & O_EXCL) )
704 if (!(res = bd_claim(bdev, filp)))
711 int blkdev_put(struct block_device *bdev)
714 struct inode *bd_inode = bdev->bd_inode;
715 struct gendisk *disk = bdev->bd_disk;
719 if (!--bdev->bd_openers) {
723 if (bdev->bd_contains == bdev) {
724 if (disk->fops->release)
725 ret = disk->fops->release(bd_inode, NULL);
727 down(&bdev->bd_contains->bd_sem);
728 bdev->bd_contains->bd_part_count--;
729 up(&bdev->bd_contains->bd_sem);
731 if (!bdev->bd_openers) {
732 struct module *owner = disk->fops->owner;
737 if (bdev->bd_contains != bdev) {
738 kobject_put(&bdev->bd_part->kobj);
739 bdev->bd_part = NULL;
741 bdev->bd_disk = NULL;
742 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
743 if (bdev != bdev->bd_contains) {
744 blkdev_put(bdev->bd_contains);
746 bdev->bd_contains = NULL;
754 EXPORT_SYMBOL(blkdev_put);
756 static int blkdev_close(struct inode * inode, struct file * filp)
758 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
759 if (bdev->bd_holder == filp)
761 return blkdev_put(bdev);
764 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
765 size_t count, loff_t *ppos)
767 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
769 return generic_file_write_nolock(file, &local_iov, 1, ppos);
772 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
773 size_t count, loff_t pos)
775 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
777 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
780 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
782 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
785 struct address_space_operations def_blk_aops = {
786 .readpage = blkdev_readpage,
787 .writepage = blkdev_writepage,
788 .sync_page = block_sync_page,
789 .prepare_write = blkdev_prepare_write,
790 .commit_write = blkdev_commit_write,
791 .writepages = generic_writepages,
792 .direct_IO = blkdev_direct_IO,
795 struct file_operations def_blk_fops = {
797 .release = blkdev_close,
798 .llseek = block_llseek,
799 .read = generic_file_read,
800 .write = blkdev_file_write,
801 .aio_read = generic_file_aio_read,
802 .aio_write = blkdev_file_aio_write,
803 .mmap = generic_file_mmap,
804 .fsync = block_fsync,
805 .unlocked_ioctl = block_ioctl,
807 .compat_ioctl = compat_blkdev_ioctl,
809 .readv = generic_file_readv,
810 .writev = generic_file_write_nolock,
811 .sendfile = generic_file_sendfile,
814 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
817 mm_segment_t old_fs = get_fs();
819 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
824 EXPORT_SYMBOL(ioctl_by_bdev);
827 * lookup_bdev - lookup a struct block_device by name
829 * @path: special file representing the block device
831 * Get a reference to the blockdevice at @path in the current
832 * namespace if possible and return it. Return ERR_PTR(error)
835 struct block_device *lookup_bdev(const char *path)
837 struct block_device *bdev;
843 return ERR_PTR(-EINVAL);
845 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
847 return ERR_PTR(error);
849 inode = nd.dentry->d_inode;
851 if (!S_ISBLK(inode->i_mode))
854 if (nd.mnt->mnt_flags & MNT_NODEV)
857 bdev = bd_acquire(inode);
864 bdev = ERR_PTR(error);
869 * open_bdev_excl - open a block device by name and set it up for use
871 * @path: special file representing the block device
872 * @flags: %MS_RDONLY for opening read-only
873 * @holder: owner for exclusion
875 * Open the blockdevice described by the special file at @path, claim it
878 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
880 struct block_device *bdev;
881 mode_t mode = FMODE_READ;
884 bdev = lookup_bdev(path);
888 if (!(flags & MS_RDONLY))
890 error = blkdev_get(bdev, mode, 0);
892 return ERR_PTR(error);
894 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
896 error = bd_claim(bdev, holder);
904 return ERR_PTR(error);
907 EXPORT_SYMBOL(open_bdev_excl);
910 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
912 * @bdev: blockdevice to close
914 * This is the counterpart to open_bdev_excl().
916 void close_bdev_excl(struct block_device *bdev)
922 EXPORT_SYMBOL(close_bdev_excl);