4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/smp_lock.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/buffer_head.h>
20 #include <linux/writeback.h>
21 #include <linux/mpage.h>
22 #include <linux/mount.h>
23 #include <linux/uio.h>
24 #include <linux/namei.h>
25 #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 int bd_get_sb(struct file_system_type *fs_type,
304 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
306 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
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;
418 spin_lock(&bdev_lock);
419 bdev = inode->i_bdev;
421 atomic_inc(&bdev->bd_inode->i_count);
422 spin_unlock(&bdev_lock);
425 spin_unlock(&bdev_lock);
427 bdev = bdget(inode->i_rdev);
429 spin_lock(&bdev_lock);
430 if (!inode->i_bdev) {
432 * We take an additional bd_inode->i_count for inode,
433 * and it's released in clear_inode() of inode.
434 * So, we can access it via ->i_mapping always
437 atomic_inc(&bdev->bd_inode->i_count);
438 inode->i_bdev = bdev;
439 inode->i_mapping = bdev->bd_inode->i_mapping;
440 list_add(&inode->i_devices, &bdev->bd_inodes);
442 spin_unlock(&bdev_lock);
447 /* Call when you free inode */
449 void bd_forget(struct inode *inode)
451 struct block_device *bdev = NULL;
453 spin_lock(&bdev_lock);
455 if (inode->i_sb != blockdev_superblock)
456 bdev = inode->i_bdev;
459 spin_unlock(&bdev_lock);
462 iput(bdev->bd_inode);
465 int bd_claim(struct block_device *bdev, void *holder)
468 spin_lock(&bdev_lock);
470 /* first decide result */
471 if (bdev->bd_holder == holder)
472 res = 0; /* already a holder */
473 else if (bdev->bd_holder != NULL)
474 res = -EBUSY; /* held by someone else */
475 else if (bdev->bd_contains == bdev)
476 res = 0; /* is a whole device which isn't held */
478 else if (bdev->bd_contains->bd_holder == bd_claim)
479 res = 0; /* is a partition of a device that is being partitioned */
480 else if (bdev->bd_contains->bd_holder != NULL)
481 res = -EBUSY; /* is a partition of a held device */
483 res = 0; /* is a partition of an un-held device */
485 /* now impose change */
487 /* note that for a whole device bd_holders
488 * will be incremented twice, and bd_holder will
489 * be set to bd_claim before being set to holder
491 bdev->bd_contains->bd_holders ++;
492 bdev->bd_contains->bd_holder = bd_claim;
494 bdev->bd_holder = holder;
496 spin_unlock(&bdev_lock);
500 EXPORT_SYMBOL(bd_claim);
502 void bd_release(struct block_device *bdev)
504 spin_lock(&bdev_lock);
505 if (!--bdev->bd_contains->bd_holders)
506 bdev->bd_contains->bd_holder = NULL;
507 if (!--bdev->bd_holders)
508 bdev->bd_holder = NULL;
509 spin_unlock(&bdev_lock);
512 EXPORT_SYMBOL(bd_release);
516 * Functions for bd_claim_by_kobject / bd_release_from_kobject
518 * If a kobject is passed to bd_claim_by_kobject()
519 * and the kobject has a parent directory,
520 * following symlinks are created:
521 * o from the kobject to the claimed bdev
522 * o from "holders" directory of the bdev to the parent of the kobject
523 * bd_release_from_kobject() removes these symlinks.
526 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
527 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
528 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
529 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
532 static struct kobject *bdev_get_kobj(struct block_device *bdev)
534 if (bdev->bd_contains != bdev)
535 return kobject_get(&bdev->bd_part->kobj);
537 return kobject_get(&bdev->bd_disk->kobj);
540 static struct kobject *bdev_get_holder(struct block_device *bdev)
542 if (bdev->bd_contains != bdev)
543 return kobject_get(bdev->bd_part->holder_dir);
545 return kobject_get(bdev->bd_disk->holder_dir);
548 static int add_symlink(struct kobject *from, struct kobject *to)
552 return sysfs_create_link(from, to, kobject_name(to));
555 static void del_symlink(struct kobject *from, struct kobject *to)
559 sysfs_remove_link(from, kobject_name(to));
563 * 'struct bd_holder' contains pointers to kobjects symlinked by
564 * bd_claim_by_kobject.
565 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
568 struct list_head list; /* chain of holders of the bdev */
569 int count; /* references from the holder */
570 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
571 struct kobject *hdev; /* e.g. "/block/dm-0" */
572 struct kobject *hdir; /* e.g. "/block/sda/holders" */
573 struct kobject *sdev; /* e.g. "/block/sda" */
577 * Get references of related kobjects at once.
578 * Returns 1 on success. 0 on failure.
580 * Should call bd_holder_release_dirs() after successful use.
582 static int bd_holder_grab_dirs(struct block_device *bdev,
583 struct bd_holder *bo)
588 bo->sdir = kobject_get(bo->sdir);
592 bo->hdev = kobject_get(bo->sdir->parent);
596 bo->sdev = bdev_get_kobj(bdev);
600 bo->hdir = bdev_get_holder(bdev);
607 kobject_put(bo->sdev);
609 kobject_put(bo->hdev);
611 kobject_put(bo->sdir);
616 /* Put references of related kobjects at once. */
617 static void bd_holder_release_dirs(struct bd_holder *bo)
619 kobject_put(bo->hdir);
620 kobject_put(bo->sdev);
621 kobject_put(bo->hdev);
622 kobject_put(bo->sdir);
625 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
627 struct bd_holder *bo;
629 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
639 static void free_bd_holder(struct bd_holder *bo)
645 * find_bd_holder - find matching struct bd_holder from the block device
647 * @bdev: struct block device to be searched
648 * @bo: target struct bd_holder
650 * Returns matching entry with @bo in @bdev->bd_holder_list.
651 * If found, increment the reference count and return the pointer.
652 * If not found, returns NULL.
654 static struct bd_holder *find_bd_holder(struct block_device *bdev,
655 struct bd_holder *bo)
657 struct bd_holder *tmp;
659 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
660 if (tmp->sdir == bo->sdir) {
669 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
671 * @bdev: block device to be bd_claimed
672 * @bo: preallocated and initialized by alloc_bd_holder()
674 * Add @bo to @bdev->bd_holder_list, create symlinks.
676 * Returns 0 if symlinks are created.
677 * Returns -ve if something fails.
679 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
686 if (!bd_holder_grab_dirs(bdev, bo))
689 ret = add_symlink(bo->sdir, bo->sdev);
691 ret = add_symlink(bo->hdir, bo->hdev);
693 del_symlink(bo->sdir, bo->sdev);
696 list_add_tail(&bo->list, &bdev->bd_holder_list);
701 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
703 * @bdev: block device to be bd_claimed
704 * @kobj: holder's kobject
706 * If there is matching entry with @kobj in @bdev->bd_holder_list
707 * and no other bd_claim() from the same kobject,
708 * remove the struct bd_holder from the list, delete symlinks for it.
710 * Returns a pointer to the struct bd_holder when it's removed from the list
711 * and ready to be freed.
712 * Returns NULL if matching claim isn't found or there is other bd_claim()
713 * by the same kobject.
715 static struct bd_holder *del_bd_holder(struct block_device *bdev,
716 struct kobject *kobj)
718 struct bd_holder *bo;
720 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
721 if (bo->sdir == kobj) {
723 BUG_ON(bo->count < 0);
726 del_symlink(bo->sdir, bo->sdev);
727 del_symlink(bo->hdir, bo->hdev);
728 bd_holder_release_dirs(bo);
739 * bd_claim_by_kobject - bd_claim() with additional kobject signature
741 * @bdev: block device to be claimed
742 * @holder: holder's signature
743 * @kobj: holder's kobject
745 * Do bd_claim() and if it succeeds, create sysfs symlinks between
746 * the bdev and the holder's kobject.
747 * Use bd_release_from_kobject() when relesing the claimed bdev.
749 * Returns 0 on success. (same as bd_claim())
750 * Returns errno on failure.
752 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
753 struct kobject *kobj)
756 struct bd_holder *bo, *found;
761 bo = alloc_bd_holder(kobj);
765 mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
766 res = bd_claim(bdev, holder);
768 found = find_bd_holder(bdev, bo);
770 res = add_bd_holder(bdev, bo);
778 mutex_unlock(&bdev->bd_mutex);
784 * bd_release_from_kobject - bd_release() with additional kobject signature
786 * @bdev: block device to be released
787 * @kobj: holder's kobject
789 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
791 static void bd_release_from_kobject(struct block_device *bdev,
792 struct kobject *kobj)
794 struct bd_holder *bo;
799 mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
801 if ((bo = del_bd_holder(bdev, kobj)))
803 mutex_unlock(&bdev->bd_mutex);
807 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
809 * @bdev: block device to be claimed
810 * @holder: holder's signature
811 * @disk: holder's gendisk
813 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
815 int bd_claim_by_disk(struct block_device *bdev, void *holder,
816 struct gendisk *disk)
818 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
820 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
823 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
825 * @bdev: block device to be claimed
826 * @disk: holder's gendisk
828 * Call bd_release_from_kobject() and put @disk->slave_dir.
830 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
832 bd_release_from_kobject(bdev, disk->slave_dir);
833 kobject_put(disk->slave_dir);
835 EXPORT_SYMBOL_GPL(bd_release_from_disk);
839 * Tries to open block device by device number. Use it ONLY if you
840 * really do not have anything better - i.e. when you are behind a
841 * truly sucky interface and all you are given is a device number. _Never_
842 * to be used for internal purposes. If you ever need it - reconsider
845 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
847 struct block_device *bdev = bdget(dev);
849 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
851 err = blkdev_get(bdev, mode, flags);
852 return err ? ERR_PTR(err) : bdev;
855 EXPORT_SYMBOL(open_by_devnum);
858 blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags);
860 struct block_device *open_partition_by_devnum(dev_t dev, unsigned mode)
862 struct block_device *bdev = bdget(dev);
864 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
866 err = blkdev_get_partition(bdev, mode, flags);
867 return err ? ERR_PTR(err) : bdev;
870 EXPORT_SYMBOL(open_partition_by_devnum);
874 * This routine checks whether a removable media has been changed,
875 * and invalidates all buffer-cache-entries in that case. This
876 * is a relatively slow routine, so we have to try to minimize using
877 * it. Thus it is called only upon a 'mount' or 'open'. This
878 * is the best way of combining speed and utility, I think.
879 * People changing diskettes in the middle of an operation deserve
882 int check_disk_change(struct block_device *bdev)
884 struct gendisk *disk = bdev->bd_disk;
885 struct block_device_operations * bdops = disk->fops;
887 if (!bdops->media_changed)
889 if (!bdops->media_changed(bdev->bd_disk))
892 if (__invalidate_device(bdev))
893 printk("VFS: busy inodes on changed media.\n");
895 if (bdops->revalidate_disk)
896 bdops->revalidate_disk(bdev->bd_disk);
897 if (bdev->bd_disk->minors > 1)
898 bdev->bd_invalidated = 1;
902 EXPORT_SYMBOL(check_disk_change);
904 void bd_set_size(struct block_device *bdev, loff_t size)
906 unsigned bsize = bdev_hardsect_size(bdev);
908 bdev->bd_inode->i_size = size;
909 while (bsize < PAGE_CACHE_SIZE) {
914 bdev->bd_block_size = bsize;
915 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
917 EXPORT_SYMBOL(bd_set_size);
919 static int __blkdev_put(struct block_device *bdev, unsigned int subclass)
922 struct inode *bd_inode = bdev->bd_inode;
923 struct gendisk *disk = bdev->bd_disk;
925 mutex_lock_nested(&bdev->bd_mutex, subclass);
927 if (!--bdev->bd_openers) {
931 if (bdev->bd_contains == bdev) {
932 if (disk->fops->release)
933 ret = disk->fops->release(bd_inode, NULL);
935 mutex_lock_nested(&bdev->bd_contains->bd_mutex,
937 bdev->bd_contains->bd_part_count--;
938 mutex_unlock(&bdev->bd_contains->bd_mutex);
940 if (!bdev->bd_openers) {
941 struct module *owner = disk->fops->owner;
946 if (bdev->bd_contains != bdev) {
947 kobject_put(&bdev->bd_part->kobj);
948 bdev->bd_part = NULL;
950 bdev->bd_disk = NULL;
951 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
952 if (bdev != bdev->bd_contains)
953 __blkdev_put(bdev->bd_contains, subclass + 1);
954 bdev->bd_contains = NULL;
957 mutex_unlock(&bdev->bd_mutex);
962 int blkdev_put(struct block_device *bdev)
964 return __blkdev_put(bdev, BD_MUTEX_NORMAL);
966 EXPORT_SYMBOL(blkdev_put);
968 int blkdev_put_partition(struct block_device *bdev)
970 return __blkdev_put(bdev, BD_MUTEX_PARTITION);
972 EXPORT_SYMBOL(blkdev_put_partition);
975 blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags);
978 do_open(struct block_device *bdev, struct file *file, unsigned int subclass)
980 struct module *owner = NULL;
981 struct gendisk *disk;
985 file->f_mapping = bdev->bd_inode->i_mapping;
987 disk = get_gendisk(bdev->bd_dev, &part);
993 owner = disk->fops->owner;
995 mutex_lock_nested(&bdev->bd_mutex, subclass);
997 if (!bdev->bd_openers) {
998 bdev->bd_disk = disk;
999 bdev->bd_contains = bdev;
1001 struct backing_dev_info *bdi;
1002 if (disk->fops->open) {
1003 ret = disk->fops->open(bdev->bd_inode, file);
1007 if (!bdev->bd_openers) {
1008 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1009 bdi = blk_get_backing_dev_info(bdev);
1011 bdi = &default_backing_dev_info;
1012 bdev->bd_inode->i_data.backing_dev_info = bdi;
1014 if (bdev->bd_invalidated)
1015 rescan_partitions(disk, bdev);
1017 struct hd_struct *p;
1018 struct block_device *whole;
1019 whole = bdget_disk(disk, 0);
1023 ret = blkdev_get_whole(whole, file->f_mode, file->f_flags);
1026 bdev->bd_contains = whole;
1027 mutex_lock_nested(&whole->bd_mutex, BD_MUTEX_WHOLE);
1028 whole->bd_part_count++;
1029 p = disk->part[part - 1];
1030 bdev->bd_inode->i_data.backing_dev_info =
1031 whole->bd_inode->i_data.backing_dev_info;
1032 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
1033 whole->bd_part_count--;
1034 mutex_unlock(&whole->bd_mutex);
1038 kobject_get(&p->kobj);
1040 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
1041 mutex_unlock(&whole->bd_mutex);
1046 if (bdev->bd_contains == bdev) {
1047 if (bdev->bd_disk->fops->open) {
1048 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1052 if (bdev->bd_invalidated)
1053 rescan_partitions(bdev->bd_disk, bdev);
1055 mutex_lock_nested(&bdev->bd_contains->bd_mutex,
1057 bdev->bd_contains->bd_part_count++;
1058 mutex_unlock(&bdev->bd_contains->bd_mutex);
1062 mutex_unlock(&bdev->bd_mutex);
1067 bdev->bd_disk = NULL;
1068 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1069 if (bdev != bdev->bd_contains)
1070 __blkdev_put(bdev->bd_contains, BD_MUTEX_WHOLE);
1071 bdev->bd_contains = NULL;
1075 mutex_unlock(&bdev->bd_mutex);
1082 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1085 * This crockload is due to bad choice of ->open() type.
1087 * For now, block device ->open() routine must _not_
1088 * examine anything in 'inode' argument except ->i_rdev.
1090 struct file fake_file = {};
1091 struct dentry fake_dentry = {};
1092 fake_file.f_mode = mode;
1093 fake_file.f_flags = flags;
1094 fake_file.f_dentry = &fake_dentry;
1095 fake_dentry.d_inode = bdev->bd_inode;
1097 return do_open(bdev, &fake_file, BD_MUTEX_NORMAL);
1100 EXPORT_SYMBOL(blkdev_get);
1103 blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags)
1106 * This crockload is due to bad choice of ->open() type.
1108 * For now, block device ->open() routine must _not_
1109 * examine anything in 'inode' argument except ->i_rdev.
1111 struct file fake_file = {};
1112 struct dentry fake_dentry = {};
1113 fake_file.f_mode = mode;
1114 fake_file.f_flags = flags;
1115 fake_file.f_dentry = &fake_dentry;
1116 fake_dentry.d_inode = bdev->bd_inode;
1118 return do_open(bdev, &fake_file, BD_MUTEX_WHOLE);
1122 blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags)
1125 * This crockload is due to bad choice of ->open() type.
1127 * For now, block device ->open() routine must _not_
1128 * examine anything in 'inode' argument except ->i_rdev.
1130 struct file fake_file = {};
1131 struct dentry fake_dentry = {};
1132 fake_file.f_mode = mode;
1133 fake_file.f_flags = flags;
1134 fake_file.f_dentry = &fake_dentry;
1135 fake_dentry.d_inode = bdev->bd_inode;
1137 return do_open(bdev, &fake_file, BD_MUTEX_PARTITION);
1140 static int blkdev_open(struct inode * inode, struct file * filp)
1142 struct block_device *bdev;
1146 * Preserve backwards compatibility and allow large file access
1147 * even if userspace doesn't ask for it explicitly. Some mkfs
1148 * binary needs it. We might want to drop this workaround
1149 * during an unstable branch.
1151 filp->f_flags |= O_LARGEFILE;
1153 bdev = bd_acquire(inode);
1157 res = do_open(bdev, filp, BD_MUTEX_NORMAL);
1161 if (!(filp->f_flags & O_EXCL) )
1164 if (!(res = bd_claim(bdev, filp)))
1171 static int blkdev_close(struct inode * inode, struct file * filp)
1173 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1174 if (bdev->bd_holder == filp)
1176 return blkdev_put(bdev);
1179 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1181 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1184 const struct address_space_operations def_blk_aops = {
1185 .readpage = blkdev_readpage,
1186 .writepage = blkdev_writepage,
1187 .sync_page = block_sync_page,
1188 .prepare_write = blkdev_prepare_write,
1189 .commit_write = blkdev_commit_write,
1190 .writepages = generic_writepages,
1191 .direct_IO = blkdev_direct_IO,
1194 const struct file_operations def_blk_fops = {
1195 .open = blkdev_open,
1196 .release = blkdev_close,
1197 .llseek = block_llseek,
1198 .read = do_sync_read,
1199 .write = do_sync_write,
1200 .aio_read = generic_file_aio_read,
1201 .aio_write = generic_file_aio_write_nolock,
1202 .mmap = generic_file_mmap,
1203 .fsync = block_fsync,
1204 .unlocked_ioctl = block_ioctl,
1205 #ifdef CONFIG_COMPAT
1206 .compat_ioctl = compat_blkdev_ioctl,
1208 .sendfile = generic_file_sendfile,
1209 .splice_read = generic_file_splice_read,
1210 .splice_write = generic_file_splice_write,
1213 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1216 mm_segment_t old_fs = get_fs();
1218 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1223 EXPORT_SYMBOL(ioctl_by_bdev);
1226 * lookup_bdev - lookup a struct block_device by name
1228 * @path: special file representing the block device
1230 * Get a reference to the blockdevice at @path in the current
1231 * namespace if possible and return it. Return ERR_PTR(error)
1234 struct block_device *lookup_bdev(const char *path)
1236 struct block_device *bdev;
1237 struct inode *inode;
1238 struct nameidata nd;
1241 if (!path || !*path)
1242 return ERR_PTR(-EINVAL);
1244 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1246 return ERR_PTR(error);
1248 inode = nd.dentry->d_inode;
1250 if (!S_ISBLK(inode->i_mode))
1253 if (nd.mnt->mnt_flags & MNT_NODEV)
1256 bdev = bd_acquire(inode);
1263 bdev = ERR_PTR(error);
1268 * open_bdev_excl - open a block device by name and set it up for use
1270 * @path: special file representing the block device
1271 * @flags: %MS_RDONLY for opening read-only
1272 * @holder: owner for exclusion
1274 * Open the blockdevice described by the special file at @path, claim it
1277 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1279 struct block_device *bdev;
1280 mode_t mode = FMODE_READ;
1283 bdev = lookup_bdev(path);
1287 if (!(flags & MS_RDONLY))
1288 mode |= FMODE_WRITE;
1289 error = blkdev_get(bdev, mode, 0);
1291 return ERR_PTR(error);
1293 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1295 error = bd_claim(bdev, holder);
1303 return ERR_PTR(error);
1306 EXPORT_SYMBOL(open_bdev_excl);
1309 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1311 * @bdev: blockdevice to close
1313 * This is the counterpart to open_bdev_excl().
1315 void close_bdev_excl(struct block_device *bdev)
1321 EXPORT_SYMBOL(close_bdev_excl);
1323 int __invalidate_device(struct block_device *bdev)
1325 struct super_block *sb = get_super(bdev);
1330 * no need to lock the super, get_super holds the
1331 * read mutex so the filesystem cannot go away
1332 * under us (->put_super runs with the write lock
1335 shrink_dcache_sb(sb);
1336 res = invalidate_inodes(sb);
1339 invalidate_bdev(bdev, 0);
1342 EXPORT_SYMBOL(__invalidate_device);