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 if (set_blocksize(sb->s_bdev, size))
91 /* If we get here, we know size is power of two
92 * and it's value is between 512 and PAGE_SIZE */
93 sb->s_blocksize = size;
94 sb->s_blocksize_bits = blksize_bits(size);
95 return sb->s_blocksize;
98 EXPORT_SYMBOL(sb_set_blocksize);
100 int sb_min_blocksize(struct super_block *sb, int size)
102 int minsize = bdev_hardsect_size(sb->s_bdev);
105 return sb_set_blocksize(sb, size);
108 EXPORT_SYMBOL(sb_min_blocksize);
111 blkdev_get_block(struct inode *inode, sector_t iblock,
112 struct buffer_head *bh, int create)
114 if (iblock >= max_block(I_BDEV(inode))) {
119 * for reads, we're just trying to fill a partial page.
120 * return a hole, they will have to call get_block again
121 * before they can fill it, and they will get -EIO at that
126 bh->b_bdev = I_BDEV(inode);
127 bh->b_blocknr = iblock;
128 set_buffer_mapped(bh);
133 blkdev_get_blocks(struct inode *inode, sector_t iblock,
134 struct buffer_head *bh, int create)
136 sector_t end_block = max_block(I_BDEV(inode));
137 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
139 if ((iblock + max_blocks) > end_block) {
140 max_blocks = end_block - iblock;
141 if ((long)max_blocks <= 0) {
143 return -EIO; /* write fully beyond EOF */
145 * It is a read which is fully beyond EOF. We return
146 * a !buffer_mapped buffer
152 bh->b_bdev = I_BDEV(inode);
153 bh->b_blocknr = iblock;
154 bh->b_size = max_blocks << inode->i_blkbits;
156 set_buffer_mapped(bh);
161 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
162 loff_t offset, unsigned long nr_segs)
164 struct file *file = iocb->ki_filp;
165 struct inode *inode = file->f_mapping->host;
167 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
168 iov, offset, nr_segs, blkdev_get_blocks, NULL);
171 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
173 return block_write_full_page(page, blkdev_get_block, wbc);
176 static int blkdev_readpage(struct file * file, struct page * page)
178 return block_read_full_page(page, blkdev_get_block);
181 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
183 return block_prepare_write(page, from, to, blkdev_get_block);
186 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
188 return block_commit_write(page, from, to);
193 * for a block special file file->f_dentry->d_inode->i_size is zero
194 * so we compute the size by hand (just as in block_read/write above)
196 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
198 struct inode *bd_inode = file->f_mapping->host;
202 mutex_lock(&bd_inode->i_mutex);
203 size = i_size_read(bd_inode);
210 offset += file->f_pos;
213 if (offset >= 0 && offset <= size) {
214 if (offset != file->f_pos) {
215 file->f_pos = offset;
219 mutex_unlock(&bd_inode->i_mutex);
224 * Filp is never NULL; the only case when ->fsync() is called with
225 * NULL first argument is nfsd_sync_dir() and that's not a directory.
228 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
230 return sync_blockdev(I_BDEV(filp->f_mapping->host));
237 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
238 static kmem_cache_t * bdev_cachep __read_mostly;
240 static struct inode *bdev_alloc_inode(struct super_block *sb)
242 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
245 return &ei->vfs_inode;
248 static void bdev_destroy_inode(struct inode *inode)
250 struct bdev_inode *bdi = BDEV_I(inode);
252 bdi->bdev.bd_inode_backing_dev_info = NULL;
253 kmem_cache_free(bdev_cachep, bdi);
256 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
258 struct bdev_inode *ei = (struct bdev_inode *) foo;
259 struct block_device *bdev = &ei->bdev;
261 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
262 SLAB_CTOR_CONSTRUCTOR)
264 memset(bdev, 0, sizeof(*bdev));
265 mutex_init(&bdev->bd_mutex);
266 mutex_init(&bdev->bd_mount_mutex);
267 INIT_LIST_HEAD(&bdev->bd_inodes);
268 INIT_LIST_HEAD(&bdev->bd_list);
270 INIT_LIST_HEAD(&bdev->bd_holder_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 __read_mostly;
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|
323 SLAB_MEM_SPREAD|SLAB_PANIC),
325 err = register_filesystem(&bd_type);
327 panic("Cannot register bdev pseudo-fs");
328 bd_mnt = kern_mount(&bd_type);
329 err = PTR_ERR(bd_mnt);
331 panic("Cannot create bdev pseudo-fs");
332 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
336 * Most likely _very_ bad one - but then it's hardly critical for small
337 * /dev and can be fixed when somebody will need really large one.
338 * Keep in mind that it will be fed through icache hash function too.
340 static inline unsigned long hash(dev_t dev)
342 return MAJOR(dev)+MINOR(dev);
345 static int bdev_test(struct inode *inode, void *data)
347 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
350 static int bdev_set(struct inode *inode, void *data)
352 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
356 static LIST_HEAD(all_bdevs);
358 struct block_device *bdget(dev_t dev)
360 struct block_device *bdev;
363 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
364 bdev_test, bdev_set, &dev);
369 bdev = &BDEV_I(inode)->bdev;
371 if (inode->i_state & I_NEW) {
372 bdev->bd_contains = NULL;
373 bdev->bd_inode = inode;
374 bdev->bd_block_size = (1 << inode->i_blkbits);
375 bdev->bd_part_count = 0;
376 bdev->bd_invalidated = 0;
377 inode->i_mode = S_IFBLK;
379 inode->i_bdev = bdev;
380 inode->i_data.a_ops = &def_blk_aops;
381 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
382 inode->i_data.backing_dev_info = &default_backing_dev_info;
383 spin_lock(&bdev_lock);
384 list_add(&bdev->bd_list, &all_bdevs);
385 spin_unlock(&bdev_lock);
386 unlock_new_inode(inode);
391 EXPORT_SYMBOL(bdget);
393 long nr_blockdev_pages(void)
397 spin_lock(&bdev_lock);
398 list_for_each(p, &all_bdevs) {
399 struct block_device *bdev;
400 bdev = list_entry(p, struct block_device, bd_list);
401 ret += bdev->bd_inode->i_mapping->nrpages;
403 spin_unlock(&bdev_lock);
407 void bdput(struct block_device *bdev)
409 iput(bdev->bd_inode);
412 EXPORT_SYMBOL(bdput);
414 static struct block_device *bd_acquire(struct inode *inode)
416 struct block_device *bdev;
417 spin_lock(&bdev_lock);
418 bdev = inode->i_bdev;
419 if (bdev && igrab(bdev->bd_inode)) {
420 spin_unlock(&bdev_lock);
423 spin_unlock(&bdev_lock);
424 bdev = bdget(inode->i_rdev);
426 spin_lock(&bdev_lock);
429 inode->i_bdev = bdev;
430 inode->i_mapping = bdev->bd_inode->i_mapping;
431 list_add(&inode->i_devices, &bdev->bd_inodes);
432 spin_unlock(&bdev_lock);
437 /* Call when you free inode */
439 void bd_forget(struct inode *inode)
441 spin_lock(&bdev_lock);
444 spin_unlock(&bdev_lock);
447 int bd_claim(struct block_device *bdev, void *holder)
450 spin_lock(&bdev_lock);
452 /* first decide result */
453 if (bdev->bd_holder == holder)
454 res = 0; /* already a holder */
455 else if (bdev->bd_holder != NULL)
456 res = -EBUSY; /* held by someone else */
457 else if (bdev->bd_contains == bdev)
458 res = 0; /* is a whole device which isn't held */
460 else if (bdev->bd_contains->bd_holder == bd_claim)
461 res = 0; /* is a partition of a device that is being partitioned */
462 else if (bdev->bd_contains->bd_holder != NULL)
463 res = -EBUSY; /* is a partition of a held device */
465 res = 0; /* is a partition of an un-held device */
467 /* now impose change */
469 /* note that for a whole device bd_holders
470 * will be incremented twice, and bd_holder will
471 * be set to bd_claim before being set to holder
473 bdev->bd_contains->bd_holders ++;
474 bdev->bd_contains->bd_holder = bd_claim;
476 bdev->bd_holder = holder;
478 spin_unlock(&bdev_lock);
482 EXPORT_SYMBOL(bd_claim);
484 void bd_release(struct block_device *bdev)
486 spin_lock(&bdev_lock);
487 if (!--bdev->bd_contains->bd_holders)
488 bdev->bd_contains->bd_holder = NULL;
489 if (!--bdev->bd_holders)
490 bdev->bd_holder = NULL;
491 spin_unlock(&bdev_lock);
494 EXPORT_SYMBOL(bd_release);
498 * Functions for bd_claim_by_kobject / bd_release_from_kobject
500 * If a kobject is passed to bd_claim_by_kobject()
501 * and the kobject has a parent directory,
502 * following symlinks are created:
503 * o from the kobject to the claimed bdev
504 * o from "holders" directory of the bdev to the parent of the kobject
505 * bd_release_from_kobject() removes these symlinks.
508 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
509 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
510 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
511 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
514 static struct kobject *bdev_get_kobj(struct block_device *bdev)
516 if (bdev->bd_contains != bdev)
517 return kobject_get(&bdev->bd_part->kobj);
519 return kobject_get(&bdev->bd_disk->kobj);
522 static struct kobject *bdev_get_holder(struct block_device *bdev)
524 if (bdev->bd_contains != bdev)
525 return kobject_get(bdev->bd_part->holder_dir);
527 return kobject_get(bdev->bd_disk->holder_dir);
530 static void add_symlink(struct kobject *from, struct kobject *to)
534 sysfs_create_link(from, to, kobject_name(to));
537 static void del_symlink(struct kobject *from, struct kobject *to)
541 sysfs_remove_link(from, kobject_name(to));
545 * 'struct bd_holder' contains pointers to kobjects symlinked by
546 * bd_claim_by_kobject.
547 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
550 struct list_head list; /* chain of holders of the bdev */
551 int count; /* references from the holder */
552 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
553 struct kobject *hdev; /* e.g. "/block/dm-0" */
554 struct kobject *hdir; /* e.g. "/block/sda/holders" */
555 struct kobject *sdev; /* e.g. "/block/sda" */
559 * Get references of related kobjects at once.
560 * Returns 1 on success. 0 on failure.
562 * Should call bd_holder_release_dirs() after successful use.
564 static int bd_holder_grab_dirs(struct block_device *bdev,
565 struct bd_holder *bo)
570 bo->sdir = kobject_get(bo->sdir);
574 bo->hdev = kobject_get(bo->sdir->parent);
578 bo->sdev = bdev_get_kobj(bdev);
582 bo->hdir = bdev_get_holder(bdev);
589 kobject_put(bo->sdev);
591 kobject_put(bo->hdev);
593 kobject_put(bo->sdir);
598 /* Put references of related kobjects at once. */
599 static void bd_holder_release_dirs(struct bd_holder *bo)
601 kobject_put(bo->hdir);
602 kobject_put(bo->sdev);
603 kobject_put(bo->hdev);
604 kobject_put(bo->sdir);
607 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
609 struct bd_holder *bo;
611 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
621 static void free_bd_holder(struct bd_holder *bo)
627 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
629 * @bdev: block device to be bd_claimed
630 * @bo: preallocated and initialized by alloc_bd_holder()
632 * If there is no matching entry with @bo in @bdev->bd_holder_list,
633 * add @bo to the list, create symlinks.
635 * Returns 1 if @bo was added to the list.
636 * Returns 0 if @bo wasn't used by any reason and should be freed.
638 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
640 struct bd_holder *tmp;
645 list_for_each_entry(tmp, &bdev->bd_holder_list, list) {
646 if (tmp->sdir == bo->sdir) {
652 if (!bd_holder_grab_dirs(bdev, bo))
655 add_symlink(bo->sdir, bo->sdev);
656 add_symlink(bo->hdir, bo->hdev);
657 list_add_tail(&bo->list, &bdev->bd_holder_list);
662 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
664 * @bdev: block device to be bd_claimed
665 * @kobj: holder's kobject
667 * If there is matching entry with @kobj in @bdev->bd_holder_list
668 * and no other bd_claim() from the same kobject,
669 * remove the struct bd_holder from the list, delete symlinks for it.
671 * Returns a pointer to the struct bd_holder when it's removed from the list
672 * and ready to be freed.
673 * Returns NULL if matching claim isn't found or there is other bd_claim()
674 * by the same kobject.
676 static struct bd_holder *del_bd_holder(struct block_device *bdev,
677 struct kobject *kobj)
679 struct bd_holder *bo;
681 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
682 if (bo->sdir == kobj) {
684 BUG_ON(bo->count < 0);
687 del_symlink(bo->sdir, bo->sdev);
688 del_symlink(bo->hdir, bo->hdev);
689 bd_holder_release_dirs(bo);
700 * bd_claim_by_kobject - bd_claim() with additional kobject signature
702 * @bdev: block device to be claimed
703 * @holder: holder's signature
704 * @kobj: holder's kobject
706 * Do bd_claim() and if it succeeds, create sysfs symlinks between
707 * the bdev and the holder's kobject.
708 * Use bd_release_from_kobject() when relesing the claimed bdev.
710 * Returns 0 on success. (same as bd_claim())
711 * Returns errno on failure.
713 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
714 struct kobject *kobj)
717 struct bd_holder *bo;
722 bo = alloc_bd_holder(kobj);
726 mutex_lock(&bdev->bd_mutex);
727 res = bd_claim(bdev, holder);
728 if (res || !add_bd_holder(bdev, bo))
730 mutex_unlock(&bdev->bd_mutex);
736 * bd_release_from_kobject - bd_release() with additional kobject signature
738 * @bdev: block device to be released
739 * @kobj: holder's kobject
741 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
743 static void bd_release_from_kobject(struct block_device *bdev,
744 struct kobject *kobj)
746 struct bd_holder *bo;
751 mutex_lock(&bdev->bd_mutex);
753 if ((bo = del_bd_holder(bdev, kobj)))
755 mutex_unlock(&bdev->bd_mutex);
759 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
761 * @bdev: block device to be claimed
762 * @holder: holder's signature
763 * @disk: holder's gendisk
765 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
767 int bd_claim_by_disk(struct block_device *bdev, void *holder,
768 struct gendisk *disk)
770 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
772 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
775 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
777 * @bdev: block device to be claimed
778 * @disk: holder's gendisk
780 * Call bd_release_from_kobject() and put @disk->slave_dir.
782 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
784 bd_release_from_kobject(bdev, disk->slave_dir);
785 kobject_put(disk->slave_dir);
787 EXPORT_SYMBOL_GPL(bd_release_from_disk);
791 * Tries to open block device by device number. Use it ONLY if you
792 * really do not have anything better - i.e. when you are behind a
793 * truly sucky interface and all you are given is a device number. _Never_
794 * to be used for internal purposes. If you ever need it - reconsider
797 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
799 struct block_device *bdev = bdget(dev);
801 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
803 err = blkdev_get(bdev, mode, flags);
804 return err ? ERR_PTR(err) : bdev;
807 EXPORT_SYMBOL(open_by_devnum);
810 * This routine checks whether a removable media has been changed,
811 * and invalidates all buffer-cache-entries in that case. This
812 * is a relatively slow routine, so we have to try to minimize using
813 * it. Thus it is called only upon a 'mount' or 'open'. This
814 * is the best way of combining speed and utility, I think.
815 * People changing diskettes in the middle of an operation deserve
818 int check_disk_change(struct block_device *bdev)
820 struct gendisk *disk = bdev->bd_disk;
821 struct block_device_operations * bdops = disk->fops;
823 if (!bdops->media_changed)
825 if (!bdops->media_changed(bdev->bd_disk))
828 if (__invalidate_device(bdev))
829 printk("VFS: busy inodes on changed media.\n");
831 if (bdops->revalidate_disk)
832 bdops->revalidate_disk(bdev->bd_disk);
833 if (bdev->bd_disk->minors > 1)
834 bdev->bd_invalidated = 1;
838 EXPORT_SYMBOL(check_disk_change);
840 void bd_set_size(struct block_device *bdev, loff_t size)
842 unsigned bsize = bdev_hardsect_size(bdev);
844 bdev->bd_inode->i_size = size;
845 while (bsize < PAGE_CACHE_SIZE) {
850 bdev->bd_block_size = bsize;
851 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
853 EXPORT_SYMBOL(bd_set_size);
855 static int do_open(struct block_device *bdev, struct file *file)
857 struct module *owner = NULL;
858 struct gendisk *disk;
862 file->f_mapping = bdev->bd_inode->i_mapping;
864 disk = get_gendisk(bdev->bd_dev, &part);
870 owner = disk->fops->owner;
872 mutex_lock(&bdev->bd_mutex);
873 if (!bdev->bd_openers) {
874 bdev->bd_disk = disk;
875 bdev->bd_contains = bdev;
877 struct backing_dev_info *bdi;
878 if (disk->fops->open) {
879 ret = disk->fops->open(bdev->bd_inode, file);
883 if (!bdev->bd_openers) {
884 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
885 bdi = blk_get_backing_dev_info(bdev);
887 bdi = &default_backing_dev_info;
888 bdev->bd_inode->i_data.backing_dev_info = bdi;
890 if (bdev->bd_invalidated)
891 rescan_partitions(disk, bdev);
894 struct block_device *whole;
895 whole = bdget_disk(disk, 0);
899 ret = blkdev_get(whole, file->f_mode, file->f_flags);
902 bdev->bd_contains = whole;
903 mutex_lock(&whole->bd_mutex);
904 whole->bd_part_count++;
905 p = disk->part[part - 1];
906 bdev->bd_inode->i_data.backing_dev_info =
907 whole->bd_inode->i_data.backing_dev_info;
908 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
909 whole->bd_part_count--;
910 mutex_unlock(&whole->bd_mutex);
914 kobject_get(&p->kobj);
916 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
917 mutex_unlock(&whole->bd_mutex);
922 if (bdev->bd_contains == bdev) {
923 if (bdev->bd_disk->fops->open) {
924 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
928 if (bdev->bd_invalidated)
929 rescan_partitions(bdev->bd_disk, bdev);
931 mutex_lock(&bdev->bd_contains->bd_mutex);
932 bdev->bd_contains->bd_part_count++;
933 mutex_unlock(&bdev->bd_contains->bd_mutex);
937 mutex_unlock(&bdev->bd_mutex);
942 bdev->bd_disk = NULL;
943 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
944 if (bdev != bdev->bd_contains)
945 blkdev_put(bdev->bd_contains);
946 bdev->bd_contains = NULL;
950 mutex_unlock(&bdev->bd_mutex);
957 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
960 * This crockload is due to bad choice of ->open() type.
962 * For now, block device ->open() routine must _not_
963 * examine anything in 'inode' argument except ->i_rdev.
965 struct file fake_file = {};
966 struct dentry fake_dentry = {};
967 fake_file.f_mode = mode;
968 fake_file.f_flags = flags;
969 fake_file.f_dentry = &fake_dentry;
970 fake_dentry.d_inode = bdev->bd_inode;
972 return do_open(bdev, &fake_file);
975 EXPORT_SYMBOL(blkdev_get);
977 static int blkdev_open(struct inode * inode, struct file * filp)
979 struct block_device *bdev;
983 * Preserve backwards compatibility and allow large file access
984 * even if userspace doesn't ask for it explicitly. Some mkfs
985 * binary needs it. We might want to drop this workaround
986 * during an unstable branch.
988 filp->f_flags |= O_LARGEFILE;
990 bdev = bd_acquire(inode);
992 res = do_open(bdev, filp);
996 if (!(filp->f_flags & O_EXCL) )
999 if (!(res = bd_claim(bdev, filp)))
1006 int blkdev_put(struct block_device *bdev)
1009 struct inode *bd_inode = bdev->bd_inode;
1010 struct gendisk *disk = bdev->bd_disk;
1012 mutex_lock(&bdev->bd_mutex);
1014 if (!--bdev->bd_openers) {
1015 sync_blockdev(bdev);
1018 if (bdev->bd_contains == bdev) {
1019 if (disk->fops->release)
1020 ret = disk->fops->release(bd_inode, NULL);
1022 mutex_lock(&bdev->bd_contains->bd_mutex);
1023 bdev->bd_contains->bd_part_count--;
1024 mutex_unlock(&bdev->bd_contains->bd_mutex);
1026 if (!bdev->bd_openers) {
1027 struct module *owner = disk->fops->owner;
1032 if (bdev->bd_contains != bdev) {
1033 kobject_put(&bdev->bd_part->kobj);
1034 bdev->bd_part = NULL;
1036 bdev->bd_disk = NULL;
1037 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1038 if (bdev != bdev->bd_contains) {
1039 blkdev_put(bdev->bd_contains);
1041 bdev->bd_contains = NULL;
1044 mutex_unlock(&bdev->bd_mutex);
1049 EXPORT_SYMBOL(blkdev_put);
1051 static int blkdev_close(struct inode * inode, struct file * filp)
1053 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1054 if (bdev->bd_holder == filp)
1056 return blkdev_put(bdev);
1059 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
1060 size_t count, loff_t *ppos)
1062 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1064 return generic_file_write_nolock(file, &local_iov, 1, ppos);
1067 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
1068 size_t count, loff_t pos)
1070 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1072 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
1075 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1077 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1080 struct address_space_operations def_blk_aops = {
1081 .readpage = blkdev_readpage,
1082 .writepage = blkdev_writepage,
1083 .sync_page = block_sync_page,
1084 .prepare_write = blkdev_prepare_write,
1085 .commit_write = blkdev_commit_write,
1086 .writepages = generic_writepages,
1087 .direct_IO = blkdev_direct_IO,
1090 const struct file_operations def_blk_fops = {
1091 .open = blkdev_open,
1092 .release = blkdev_close,
1093 .llseek = block_llseek,
1094 .read = generic_file_read,
1095 .write = blkdev_file_write,
1096 .aio_read = generic_file_aio_read,
1097 .aio_write = blkdev_file_aio_write,
1098 .mmap = generic_file_mmap,
1099 .fsync = block_fsync,
1100 .unlocked_ioctl = block_ioctl,
1101 #ifdef CONFIG_COMPAT
1102 .compat_ioctl = compat_blkdev_ioctl,
1104 .readv = generic_file_readv,
1105 .writev = generic_file_write_nolock,
1106 .sendfile = generic_file_sendfile,
1107 .splice_read = generic_file_splice_read,
1108 .splice_write = generic_file_splice_write,
1111 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1114 mm_segment_t old_fs = get_fs();
1116 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1121 EXPORT_SYMBOL(ioctl_by_bdev);
1124 * lookup_bdev - lookup a struct block_device by name
1126 * @path: special file representing the block device
1128 * Get a reference to the blockdevice at @path in the current
1129 * namespace if possible and return it. Return ERR_PTR(error)
1132 struct block_device *lookup_bdev(const char *path)
1134 struct block_device *bdev;
1135 struct inode *inode;
1136 struct nameidata nd;
1139 if (!path || !*path)
1140 return ERR_PTR(-EINVAL);
1142 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1144 return ERR_PTR(error);
1146 inode = nd.dentry->d_inode;
1148 if (!S_ISBLK(inode->i_mode))
1151 if (nd.mnt->mnt_flags & MNT_NODEV)
1154 bdev = bd_acquire(inode);
1161 bdev = ERR_PTR(error);
1166 * open_bdev_excl - open a block device by name and set it up for use
1168 * @path: special file representing the block device
1169 * @flags: %MS_RDONLY for opening read-only
1170 * @holder: owner for exclusion
1172 * Open the blockdevice described by the special file at @path, claim it
1175 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1177 struct block_device *bdev;
1178 mode_t mode = FMODE_READ;
1181 bdev = lookup_bdev(path);
1185 if (!(flags & MS_RDONLY))
1186 mode |= FMODE_WRITE;
1187 error = blkdev_get(bdev, mode, 0);
1189 return ERR_PTR(error);
1191 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1193 error = bd_claim(bdev, holder);
1201 return ERR_PTR(error);
1204 EXPORT_SYMBOL(open_bdev_excl);
1207 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1209 * @bdev: blockdevice to close
1211 * This is the counterpart to open_bdev_excl().
1213 void close_bdev_excl(struct block_device *bdev)
1219 EXPORT_SYMBOL(close_bdev_excl);