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/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
52 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
53 * we need to move it onto the dirty list of @dst so that the inode is always
56 static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
59 spin_lock(&inode_lock);
60 inode->i_data.backing_dev_info = dst;
61 if (inode->i_state & I_DIRTY)
62 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
63 spin_unlock(&inode_lock);
66 static sector_t max_block(struct block_device *bdev)
68 sector_t retval = ~((sector_t)0);
69 loff_t sz = i_size_read(bdev->bd_inode);
72 unsigned int size = block_size(bdev);
73 unsigned int sizebits = blksize_bits(size);
74 retval = (sz >> sizebits);
79 /* Kill _all_ buffers and pagecache , dirty or not.. */
80 static void kill_bdev(struct block_device *bdev)
82 if (bdev->bd_inode->i_mapping->nrpages == 0)
85 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
88 int set_blocksize(struct block_device *bdev, int size)
90 /* Size must be a power of two, and between 512 and PAGE_SIZE */
91 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
94 /* Size cannot be smaller than the size supported by the device */
95 if (size < bdev_logical_block_size(bdev))
98 /* Don't change the size if it is same as current */
99 if (bdev->bd_block_size != size) {
101 bdev->bd_block_size = size;
102 bdev->bd_inode->i_blkbits = blksize_bits(size);
108 EXPORT_SYMBOL(set_blocksize);
110 int sb_set_blocksize(struct super_block *sb, int size)
112 if (set_blocksize(sb->s_bdev, size))
114 /* If we get here, we know size is power of two
115 * and it's value is between 512 and PAGE_SIZE */
116 sb->s_blocksize = size;
117 sb->s_blocksize_bits = blksize_bits(size);
118 return sb->s_blocksize;
121 EXPORT_SYMBOL(sb_set_blocksize);
123 int sb_min_blocksize(struct super_block *sb, int size)
125 int minsize = bdev_logical_block_size(sb->s_bdev);
128 return sb_set_blocksize(sb, size);
131 EXPORT_SYMBOL(sb_min_blocksize);
134 blkdev_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh, int create)
137 if (iblock >= max_block(I_BDEV(inode))) {
142 * for reads, we're just trying to fill a partial page.
143 * return a hole, they will have to call get_block again
144 * before they can fill it, and they will get -EIO at that
149 bh->b_bdev = I_BDEV(inode);
150 bh->b_blocknr = iblock;
151 set_buffer_mapped(bh);
156 blkdev_get_blocks(struct inode *inode, sector_t iblock,
157 struct buffer_head *bh, int create)
159 sector_t end_block = max_block(I_BDEV(inode));
160 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
162 if ((iblock + max_blocks) > end_block) {
163 max_blocks = end_block - iblock;
164 if ((long)max_blocks <= 0) {
166 return -EIO; /* write fully beyond EOF */
168 * It is a read which is fully beyond EOF. We return
169 * a !buffer_mapped buffer
175 bh->b_bdev = I_BDEV(inode);
176 bh->b_blocknr = iblock;
177 bh->b_size = max_blocks << inode->i_blkbits;
179 set_buffer_mapped(bh);
184 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
185 loff_t offset, unsigned long nr_segs)
187 struct file *file = iocb->ki_filp;
188 struct inode *inode = file->f_mapping->host;
190 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
191 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
194 int __sync_blockdev(struct block_device *bdev, int wait)
199 return filemap_flush(bdev->bd_inode->i_mapping);
200 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
204 * Write out and wait upon all the dirty data associated with a block
205 * device via its mapping. Does not take the superblock lock.
207 int sync_blockdev(struct block_device *bdev)
209 return __sync_blockdev(bdev, 1);
211 EXPORT_SYMBOL(sync_blockdev);
214 * Write out and wait upon all dirty data associated with this
215 * device. Filesystem data as well as the underlying block
216 * device. Takes the superblock lock.
218 int fsync_bdev(struct block_device *bdev)
220 struct super_block *sb = get_super(bdev);
222 int res = sync_filesystem(sb);
226 return sync_blockdev(bdev);
228 EXPORT_SYMBOL(fsync_bdev);
231 * freeze_bdev -- lock a filesystem and force it into a consistent state
232 * @bdev: blockdevice to lock
234 * If a superblock is found on this device, we take the s_umount semaphore
235 * on it to make sure nobody unmounts until the snapshot creation is done.
236 * The reference counter (bd_fsfreeze_count) guarantees that only the last
237 * unfreeze process can unfreeze the frozen filesystem actually when multiple
238 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
239 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
242 struct super_block *freeze_bdev(struct block_device *bdev)
244 struct super_block *sb;
247 mutex_lock(&bdev->bd_fsfreeze_mutex);
248 if (++bdev->bd_fsfreeze_count > 1) {
250 * We don't even need to grab a reference - the first call
251 * to freeze_bdev grab an active reference and only the last
252 * thaw_bdev drops it.
254 sb = get_super(bdev);
256 mutex_unlock(&bdev->bd_fsfreeze_mutex);
260 sb = get_active_super(bdev);
263 error = freeze_super(sb);
265 deactivate_super(sb);
266 bdev->bd_fsfreeze_count--;
267 mutex_unlock(&bdev->bd_fsfreeze_mutex);
268 return ERR_PTR(error);
270 deactivate_super(sb);
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
274 return sb; /* thaw_bdev releases s->s_umount */
276 EXPORT_SYMBOL(freeze_bdev);
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
285 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
289 mutex_lock(&bdev->bd_fsfreeze_mutex);
290 if (!bdev->bd_fsfreeze_count)
294 if (--bdev->bd_fsfreeze_count > 0)
300 error = thaw_super(sb);
302 bdev->bd_fsfreeze_count++;
303 mutex_unlock(&bdev->bd_fsfreeze_mutex);
307 mutex_unlock(&bdev->bd_fsfreeze_mutex);
310 EXPORT_SYMBOL(thaw_bdev);
312 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
314 return block_write_full_page(page, blkdev_get_block, wbc);
317 static int blkdev_readpage(struct file * file, struct page * page)
319 return block_read_full_page(page, blkdev_get_block);
322 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned flags,
324 struct page **pagep, void **fsdata)
326 return block_write_begin(mapping, pos, len, flags, pagep,
330 static int blkdev_write_end(struct file *file, struct address_space *mapping,
331 loff_t pos, unsigned len, unsigned copied,
332 struct page *page, void *fsdata)
335 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
338 page_cache_release(page);
345 * for a block special file file->f_path.dentry->d_inode->i_size is zero
346 * so we compute the size by hand (just as in block_read/write above)
348 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
350 struct inode *bd_inode = file->f_mapping->host;
354 mutex_lock(&bd_inode->i_mutex);
355 size = i_size_read(bd_inode);
362 offset += file->f_pos;
365 if (offset >= 0 && offset <= size) {
366 if (offset != file->f_pos) {
367 file->f_pos = offset;
371 mutex_unlock(&bd_inode->i_mutex);
375 int blkdev_fsync(struct file *filp, int datasync)
377 struct inode *bd_inode = filp->f_mapping->host;
378 struct block_device *bdev = I_BDEV(bd_inode);
382 * There is no need to serialise calls to blkdev_issue_flush with
383 * i_mutex and doing so causes performance issues with concurrent
384 * O_SYNC writers to a block device.
386 mutex_unlock(&bd_inode->i_mutex);
388 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
389 if (error == -EOPNOTSUPP)
392 mutex_lock(&bd_inode->i_mutex);
396 EXPORT_SYMBOL(blkdev_fsync);
402 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
403 static struct kmem_cache * bdev_cachep __read_mostly;
405 static struct inode *bdev_alloc_inode(struct super_block *sb)
407 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
410 return &ei->vfs_inode;
413 static void bdev_destroy_inode(struct inode *inode)
415 struct bdev_inode *bdi = BDEV_I(inode);
417 kmem_cache_free(bdev_cachep, bdi);
420 static void init_once(void *foo)
422 struct bdev_inode *ei = (struct bdev_inode *) foo;
423 struct block_device *bdev = &ei->bdev;
425 memset(bdev, 0, sizeof(*bdev));
426 mutex_init(&bdev->bd_mutex);
427 INIT_LIST_HEAD(&bdev->bd_inodes);
428 INIT_LIST_HEAD(&bdev->bd_list);
430 INIT_LIST_HEAD(&bdev->bd_holder_list);
432 inode_init_once(&ei->vfs_inode);
433 /* Initialize mutex for freeze. */
434 mutex_init(&bdev->bd_fsfreeze_mutex);
437 static inline void __bd_forget(struct inode *inode)
439 list_del_init(&inode->i_devices);
440 inode->i_bdev = NULL;
441 inode->i_mapping = &inode->i_data;
444 static void bdev_evict_inode(struct inode *inode)
446 struct block_device *bdev = &BDEV_I(inode)->bdev;
448 truncate_inode_pages(&inode->i_data, 0);
449 invalidate_inode_buffers(inode); /* is it needed here? */
450 end_writeback(inode);
451 spin_lock(&bdev_lock);
452 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
453 __bd_forget(list_entry(p, struct inode, i_devices));
455 list_del_init(&bdev->bd_list);
456 spin_unlock(&bdev_lock);
459 static const struct super_operations bdev_sops = {
460 .statfs = simple_statfs,
461 .alloc_inode = bdev_alloc_inode,
462 .destroy_inode = bdev_destroy_inode,
463 .drop_inode = generic_delete_inode,
464 .evict_inode = bdev_evict_inode,
467 static int bd_get_sb(struct file_system_type *fs_type,
468 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
470 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
473 static struct file_system_type bd_type = {
476 .kill_sb = kill_anon_super,
479 struct super_block *blockdev_superblock __read_mostly;
481 void __init bdev_cache_init(void)
484 struct vfsmount *bd_mnt;
486 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
487 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
488 SLAB_MEM_SPREAD|SLAB_PANIC),
490 err = register_filesystem(&bd_type);
492 panic("Cannot register bdev pseudo-fs");
493 bd_mnt = kern_mount(&bd_type);
495 panic("Cannot create bdev pseudo-fs");
497 * This vfsmount structure is only used to obtain the
498 * blockdev_superblock, so tell kmemleak not to report it.
500 kmemleak_not_leak(bd_mnt);
501 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
505 * Most likely _very_ bad one - but then it's hardly critical for small
506 * /dev and can be fixed when somebody will need really large one.
507 * Keep in mind that it will be fed through icache hash function too.
509 static inline unsigned long hash(dev_t dev)
511 return MAJOR(dev)+MINOR(dev);
514 static int bdev_test(struct inode *inode, void *data)
516 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
519 static int bdev_set(struct inode *inode, void *data)
521 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
525 static LIST_HEAD(all_bdevs);
527 struct block_device *bdget(dev_t dev)
529 struct block_device *bdev;
532 inode = iget5_locked(blockdev_superblock, hash(dev),
533 bdev_test, bdev_set, &dev);
538 bdev = &BDEV_I(inode)->bdev;
540 if (inode->i_state & I_NEW) {
541 bdev->bd_contains = NULL;
542 bdev->bd_inode = inode;
543 bdev->bd_block_size = (1 << inode->i_blkbits);
544 bdev->bd_part_count = 0;
545 bdev->bd_invalidated = 0;
546 inode->i_mode = S_IFBLK;
548 inode->i_bdev = bdev;
549 inode->i_data.a_ops = &def_blk_aops;
550 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
551 inode->i_data.backing_dev_info = &default_backing_dev_info;
552 spin_lock(&bdev_lock);
553 list_add(&bdev->bd_list, &all_bdevs);
554 spin_unlock(&bdev_lock);
555 unlock_new_inode(inode);
560 EXPORT_SYMBOL(bdget);
563 * bdgrab -- Grab a reference to an already referenced block device
564 * @bdev: Block device to grab a reference to.
566 struct block_device *bdgrab(struct block_device *bdev)
568 ihold(bdev->bd_inode);
572 long nr_blockdev_pages(void)
574 struct block_device *bdev;
576 spin_lock(&bdev_lock);
577 list_for_each_entry(bdev, &all_bdevs, bd_list) {
578 ret += bdev->bd_inode->i_mapping->nrpages;
580 spin_unlock(&bdev_lock);
584 void bdput(struct block_device *bdev)
586 iput(bdev->bd_inode);
589 EXPORT_SYMBOL(bdput);
591 static struct block_device *bd_acquire(struct inode *inode)
593 struct block_device *bdev;
595 spin_lock(&bdev_lock);
596 bdev = inode->i_bdev;
598 ihold(bdev->bd_inode);
599 spin_unlock(&bdev_lock);
602 spin_unlock(&bdev_lock);
604 bdev = bdget(inode->i_rdev);
606 spin_lock(&bdev_lock);
607 if (!inode->i_bdev) {
609 * We take an additional reference to bd_inode,
610 * and it's released in clear_inode() of inode.
611 * So, we can access it via ->i_mapping always
614 ihold(bdev->bd_inode);
615 inode->i_bdev = bdev;
616 inode->i_mapping = bdev->bd_inode->i_mapping;
617 list_add(&inode->i_devices, &bdev->bd_inodes);
619 spin_unlock(&bdev_lock);
624 /* Call when you free inode */
626 void bd_forget(struct inode *inode)
628 struct block_device *bdev = NULL;
630 spin_lock(&bdev_lock);
632 if (!sb_is_blkdev_sb(inode->i_sb))
633 bdev = inode->i_bdev;
636 spin_unlock(&bdev_lock);
639 iput(bdev->bd_inode);
643 * bd_may_claim - test whether a block device can be claimed
644 * @bdev: block device of interest
645 * @whole: whole block device containing @bdev, may equal @bdev
646 * @holder: holder trying to claim @bdev
648 * Test whther @bdev can be claimed by @holder.
651 * spin_lock(&bdev_lock).
654 * %true if @bdev can be claimed, %false otherwise.
656 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
659 if (bdev->bd_holder == holder)
660 return true; /* already a holder */
661 else if (bdev->bd_holder != NULL)
662 return false; /* held by someone else */
663 else if (bdev->bd_contains == bdev)
664 return true; /* is a whole device which isn't held */
666 else if (whole->bd_holder == bd_claim)
667 return true; /* is a partition of a device that is being partitioned */
668 else if (whole->bd_holder != NULL)
669 return false; /* is a partition of a held device */
671 return true; /* is a partition of an un-held device */
675 * bd_prepare_to_claim - prepare to claim a block device
676 * @bdev: block device of interest
677 * @whole: the whole device containing @bdev, may equal @bdev
678 * @holder: holder trying to claim @bdev
680 * Prepare to claim @bdev. This function fails if @bdev is already
681 * claimed by another holder and waits if another claiming is in
682 * progress. This function doesn't actually claim. On successful
683 * return, the caller has ownership of bd_claiming and bd_holder[s].
686 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
690 * 0 if @bdev can be claimed, -EBUSY otherwise.
692 static int bd_prepare_to_claim(struct block_device *bdev,
693 struct block_device *whole, void *holder)
696 /* if someone else claimed, fail */
697 if (!bd_may_claim(bdev, whole, holder))
700 /* if claiming is already in progress, wait for it to finish */
701 if (whole->bd_claiming) {
702 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
705 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
706 spin_unlock(&bdev_lock);
708 finish_wait(wq, &wait);
709 spin_lock(&bdev_lock);
718 * bd_start_claiming - start claiming a block device
719 * @bdev: block device of interest
720 * @holder: holder trying to claim @bdev
722 * @bdev is about to be opened exclusively. Check @bdev can be opened
723 * exclusively and mark that an exclusive open is in progress. Each
724 * successful call to this function must be matched with a call to
725 * either bd_finish_claiming() or bd_abort_claiming() (which do not
728 * This function is used to gain exclusive access to the block device
729 * without actually causing other exclusive open attempts to fail. It
730 * should be used when the open sequence itself requires exclusive
731 * access but may subsequently fail.
737 * Pointer to the block device containing @bdev on success, ERR_PTR()
740 static struct block_device *bd_start_claiming(struct block_device *bdev,
743 struct gendisk *disk;
744 struct block_device *whole;
750 * @bdev might not have been initialized properly yet, look up
751 * and grab the outer block device the hard way.
753 disk = get_gendisk(bdev->bd_dev, &partno);
755 return ERR_PTR(-ENXIO);
757 whole = bdget_disk(disk, 0);
758 module_put(disk->fops->owner);
761 return ERR_PTR(-ENOMEM);
763 /* prepare to claim, if successful, mark claiming in progress */
764 spin_lock(&bdev_lock);
766 err = bd_prepare_to_claim(bdev, whole, holder);
768 whole->bd_claiming = holder;
769 spin_unlock(&bdev_lock);
772 spin_unlock(&bdev_lock);
778 /* releases bdev_lock */
779 static void __bd_abort_claiming(struct block_device *whole, void *holder)
781 BUG_ON(whole->bd_claiming != holder);
782 whole->bd_claiming = NULL;
783 wake_up_bit(&whole->bd_claiming, 0);
785 spin_unlock(&bdev_lock);
790 * bd_abort_claiming - abort claiming a block device
791 * @whole: whole block device returned by bd_start_claiming()
792 * @holder: holder trying to claim @bdev
794 * Abort a claiming block started by bd_start_claiming(). Note that
795 * @whole is not the block device to be claimed but the whole device
796 * returned by bd_start_claiming().
799 * Grabs and releases bdev_lock.
801 static void bd_abort_claiming(struct block_device *whole, void *holder)
803 spin_lock(&bdev_lock);
804 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
807 /* increment holders when we have a legitimate claim. requires bdev_lock */
808 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
811 /* note that for a whole device bd_holders
812 * will be incremented twice, and bd_holder will
813 * be set to bd_claim before being set to holder
816 whole->bd_holder = bd_claim;
818 bdev->bd_holder = holder;
822 * bd_finish_claiming - finish claiming a block device
823 * @bdev: block device of interest (passed to bd_start_claiming())
824 * @whole: whole block device returned by bd_start_claiming()
825 * @holder: holder trying to claim @bdev
827 * Finish a claiming block started by bd_start_claiming().
830 * Grabs and releases bdev_lock.
832 static void bd_finish_claiming(struct block_device *bdev,
833 struct block_device *whole, void *holder)
835 spin_lock(&bdev_lock);
836 BUG_ON(!bd_may_claim(bdev, whole, holder));
837 __bd_claim(bdev, whole, holder);
838 __bd_abort_claiming(whole, holder); /* not actually an abort */
842 * bd_claim - claim a block device
843 * @bdev: block device to claim
844 * @holder: holder trying to claim @bdev
846 * Try to claim @bdev which must have been opened successfully.
852 * 0 if successful, -EBUSY if @bdev is already claimed.
854 int bd_claim(struct block_device *bdev, void *holder)
856 struct block_device *whole = bdev->bd_contains;
861 spin_lock(&bdev_lock);
862 res = bd_prepare_to_claim(bdev, whole, holder);
864 __bd_claim(bdev, whole, holder);
865 spin_unlock(&bdev_lock);
869 EXPORT_SYMBOL(bd_claim);
871 void bd_release(struct block_device *bdev)
873 spin_lock(&bdev_lock);
874 if (!--bdev->bd_contains->bd_holders)
875 bdev->bd_contains->bd_holder = NULL;
876 if (!--bdev->bd_holders)
877 bdev->bd_holder = NULL;
878 spin_unlock(&bdev_lock);
881 EXPORT_SYMBOL(bd_release);
885 * Functions for bd_claim_by_kobject / bd_release_from_kobject
887 * If a kobject is passed to bd_claim_by_kobject()
888 * and the kobject has a parent directory,
889 * following symlinks are created:
890 * o from the kobject to the claimed bdev
891 * o from "holders" directory of the bdev to the parent of the kobject
892 * bd_release_from_kobject() removes these symlinks.
895 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
896 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
897 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
898 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
901 static int add_symlink(struct kobject *from, struct kobject *to)
905 return sysfs_create_link(from, to, kobject_name(to));
908 static void del_symlink(struct kobject *from, struct kobject *to)
912 sysfs_remove_link(from, kobject_name(to));
916 * 'struct bd_holder' contains pointers to kobjects symlinked by
917 * bd_claim_by_kobject.
918 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
921 struct list_head list; /* chain of holders of the bdev */
922 int count; /* references from the holder */
923 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
924 struct kobject *hdev; /* e.g. "/block/dm-0" */
925 struct kobject *hdir; /* e.g. "/block/sda/holders" */
926 struct kobject *sdev; /* e.g. "/block/sda" */
930 * Get references of related kobjects at once.
931 * Returns 1 on success. 0 on failure.
933 * Should call bd_holder_release_dirs() after successful use.
935 static int bd_holder_grab_dirs(struct block_device *bdev,
936 struct bd_holder *bo)
941 bo->sdir = kobject_get(bo->sdir);
945 bo->hdev = kobject_get(bo->sdir->parent);
949 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
953 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
960 kobject_put(bo->sdev);
962 kobject_put(bo->hdev);
964 kobject_put(bo->sdir);
969 /* Put references of related kobjects at once. */
970 static void bd_holder_release_dirs(struct bd_holder *bo)
972 kobject_put(bo->hdir);
973 kobject_put(bo->sdev);
974 kobject_put(bo->hdev);
975 kobject_put(bo->sdir);
978 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
980 struct bd_holder *bo;
982 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
992 static void free_bd_holder(struct bd_holder *bo)
998 * find_bd_holder - find matching struct bd_holder from the block device
1000 * @bdev: struct block device to be searched
1001 * @bo: target struct bd_holder
1003 * Returns matching entry with @bo in @bdev->bd_holder_list.
1004 * If found, increment the reference count and return the pointer.
1005 * If not found, returns NULL.
1007 static struct bd_holder *find_bd_holder(struct block_device *bdev,
1008 struct bd_holder *bo)
1010 struct bd_holder *tmp;
1012 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
1013 if (tmp->sdir == bo->sdir) {
1022 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1024 * @bdev: block device to be bd_claimed
1025 * @bo: preallocated and initialized by alloc_bd_holder()
1027 * Add @bo to @bdev->bd_holder_list, create symlinks.
1029 * Returns 0 if symlinks are created.
1030 * Returns -ve if something fails.
1032 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1039 if (!bd_holder_grab_dirs(bdev, bo))
1042 err = add_symlink(bo->sdir, bo->sdev);
1046 err = add_symlink(bo->hdir, bo->hdev);
1048 del_symlink(bo->sdir, bo->sdev);
1052 list_add_tail(&bo->list, &bdev->bd_holder_list);
1057 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1059 * @bdev: block device to be bd_claimed
1060 * @kobj: holder's kobject
1062 * If there is matching entry with @kobj in @bdev->bd_holder_list
1063 * and no other bd_claim() from the same kobject,
1064 * remove the struct bd_holder from the list, delete symlinks for it.
1066 * Returns a pointer to the struct bd_holder when it's removed from the list
1067 * and ready to be freed.
1068 * Returns NULL if matching claim isn't found or there is other bd_claim()
1069 * by the same kobject.
1071 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1072 struct kobject *kobj)
1074 struct bd_holder *bo;
1076 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1077 if (bo->sdir == kobj) {
1079 BUG_ON(bo->count < 0);
1081 list_del(&bo->list);
1082 del_symlink(bo->sdir, bo->sdev);
1083 del_symlink(bo->hdir, bo->hdev);
1084 bd_holder_release_dirs(bo);
1095 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1097 * @bdev: block device to be claimed
1098 * @holder: holder's signature
1099 * @kobj: holder's kobject
1101 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1102 * the bdev and the holder's kobject.
1103 * Use bd_release_from_kobject() when relesing the claimed bdev.
1105 * Returns 0 on success. (same as bd_claim())
1106 * Returns errno on failure.
1108 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1109 struct kobject *kobj)
1112 struct bd_holder *bo, *found;
1117 bo = alloc_bd_holder(kobj);
1121 mutex_lock(&bdev->bd_mutex);
1123 err = bd_claim(bdev, holder);
1127 found = find_bd_holder(bdev, bo);
1131 err = add_bd_holder(bdev, bo);
1137 mutex_unlock(&bdev->bd_mutex);
1143 * bd_release_from_kobject - bd_release() with additional kobject signature
1145 * @bdev: block device to be released
1146 * @kobj: holder's kobject
1148 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1150 static void bd_release_from_kobject(struct block_device *bdev,
1151 struct kobject *kobj)
1156 mutex_lock(&bdev->bd_mutex);
1158 free_bd_holder(del_bd_holder(bdev, kobj));
1159 mutex_unlock(&bdev->bd_mutex);
1163 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1165 * @bdev: block device to be claimed
1166 * @holder: holder's signature
1167 * @disk: holder's gendisk
1169 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1171 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1172 struct gendisk *disk)
1174 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1176 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1179 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1181 * @bdev: block device to be claimed
1182 * @disk: holder's gendisk
1184 * Call bd_release_from_kobject() and put @disk->slave_dir.
1186 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1188 bd_release_from_kobject(bdev, disk->slave_dir);
1189 kobject_put(disk->slave_dir);
1191 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1195 * Tries to open block device by device number. Use it ONLY if you
1196 * really do not have anything better - i.e. when you are behind a
1197 * truly sucky interface and all you are given is a device number. _Never_
1198 * to be used for internal purposes. If you ever need it - reconsider
1201 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1203 struct block_device *bdev = bdget(dev);
1206 err = blkdev_get(bdev, mode);
1207 return err ? ERR_PTR(err) : bdev;
1210 EXPORT_SYMBOL(open_by_devnum);
1213 * flush_disk - invalidates all buffer-cache entries on a disk
1215 * @bdev: struct block device to be flushed
1217 * Invalidates all buffer-cache entries on a disk. It should be called
1218 * when a disk has been changed -- either by a media change or online
1221 static void flush_disk(struct block_device *bdev)
1223 if (__invalidate_device(bdev)) {
1224 char name[BDEVNAME_SIZE] = "";
1227 disk_name(bdev->bd_disk, 0, name);
1228 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1229 "resized disk %s\n", name);
1234 if (disk_partitionable(bdev->bd_disk))
1235 bdev->bd_invalidated = 1;
1239 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1240 * @disk: struct gendisk to check
1241 * @bdev: struct bdev to adjust.
1243 * This routine checks to see if the bdev size does not match the disk size
1244 * and adjusts it if it differs.
1246 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1248 loff_t disk_size, bdev_size;
1250 disk_size = (loff_t)get_capacity(disk) << 9;
1251 bdev_size = i_size_read(bdev->bd_inode);
1252 if (disk_size != bdev_size) {
1253 char name[BDEVNAME_SIZE];
1255 disk_name(disk, 0, name);
1257 "%s: detected capacity change from %lld to %lld\n",
1258 name, bdev_size, disk_size);
1259 i_size_write(bdev->bd_inode, disk_size);
1263 EXPORT_SYMBOL(check_disk_size_change);
1266 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1267 * @disk: struct gendisk to be revalidated
1269 * This routine is a wrapper for lower-level driver's revalidate_disk
1270 * call-backs. It is used to do common pre and post operations needed
1271 * for all revalidate_disk operations.
1273 int revalidate_disk(struct gendisk *disk)
1275 struct block_device *bdev;
1278 if (disk->fops->revalidate_disk)
1279 ret = disk->fops->revalidate_disk(disk);
1281 bdev = bdget_disk(disk, 0);
1285 mutex_lock(&bdev->bd_mutex);
1286 check_disk_size_change(disk, bdev);
1287 mutex_unlock(&bdev->bd_mutex);
1291 EXPORT_SYMBOL(revalidate_disk);
1294 * This routine checks whether a removable media has been changed,
1295 * and invalidates all buffer-cache-entries in that case. This
1296 * is a relatively slow routine, so we have to try to minimize using
1297 * it. Thus it is called only upon a 'mount' or 'open'. This
1298 * is the best way of combining speed and utility, I think.
1299 * People changing diskettes in the middle of an operation deserve
1302 int check_disk_change(struct block_device *bdev)
1304 struct gendisk *disk = bdev->bd_disk;
1305 const struct block_device_operations *bdops = disk->fops;
1307 if (!bdops->media_changed)
1309 if (!bdops->media_changed(bdev->bd_disk))
1313 if (bdops->revalidate_disk)
1314 bdops->revalidate_disk(bdev->bd_disk);
1318 EXPORT_SYMBOL(check_disk_change);
1320 void bd_set_size(struct block_device *bdev, loff_t size)
1322 unsigned bsize = bdev_logical_block_size(bdev);
1324 bdev->bd_inode->i_size = size;
1325 while (bsize < PAGE_CACHE_SIZE) {
1330 bdev->bd_block_size = bsize;
1331 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1333 EXPORT_SYMBOL(bd_set_size);
1335 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1340 * mutex_lock(part->bd_mutex)
1341 * mutex_lock_nested(whole->bd_mutex, 1)
1344 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1346 struct gendisk *disk;
1351 if (mode & FMODE_READ)
1353 if (mode & FMODE_WRITE)
1356 * hooks: /n/, see "layering violations".
1359 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1369 disk = get_gendisk(bdev->bd_dev, &partno);
1373 mutex_lock_nested(&bdev->bd_mutex, for_part);
1374 if (!bdev->bd_openers) {
1375 bdev->bd_disk = disk;
1376 bdev->bd_contains = bdev;
1378 struct backing_dev_info *bdi;
1381 bdev->bd_part = disk_get_part(disk, partno);
1385 if (disk->fops->open) {
1386 ret = disk->fops->open(bdev, mode);
1387 if (ret == -ERESTARTSYS) {
1388 /* Lost a race with 'disk' being
1389 * deleted, try again.
1392 disk_put_part(bdev->bd_part);
1393 bdev->bd_part = NULL;
1394 module_put(disk->fops->owner);
1396 bdev->bd_disk = NULL;
1397 mutex_unlock(&bdev->bd_mutex);
1403 if (!bdev->bd_openers) {
1404 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1405 bdi = blk_get_backing_dev_info(bdev);
1407 bdi = &default_backing_dev_info;
1408 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1410 if (bdev->bd_invalidated)
1411 rescan_partitions(disk, bdev);
1413 struct block_device *whole;
1414 whole = bdget_disk(disk, 0);
1419 ret = __blkdev_get(whole, mode, 1);
1422 bdev->bd_contains = whole;
1423 bdev_inode_switch_bdi(bdev->bd_inode,
1424 whole->bd_inode->i_data.backing_dev_info);
1425 bdev->bd_part = disk_get_part(disk, partno);
1426 if (!(disk->flags & GENHD_FL_UP) ||
1427 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1431 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1434 module_put(disk->fops->owner);
1437 if (bdev->bd_contains == bdev) {
1438 if (bdev->bd_disk->fops->open) {
1439 ret = bdev->bd_disk->fops->open(bdev, mode);
1441 goto out_unlock_bdev;
1443 if (bdev->bd_invalidated)
1444 rescan_partitions(bdev->bd_disk, bdev);
1449 bdev->bd_part_count++;
1450 mutex_unlock(&bdev->bd_mutex);
1454 disk_put_part(bdev->bd_part);
1455 bdev->bd_disk = NULL;
1456 bdev->bd_part = NULL;
1457 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1458 if (bdev != bdev->bd_contains)
1459 __blkdev_put(bdev->bd_contains, mode, 1);
1460 bdev->bd_contains = NULL;
1462 mutex_unlock(&bdev->bd_mutex);
1465 module_put(disk->fops->owner);
1472 int blkdev_get(struct block_device *bdev, fmode_t mode)
1474 return __blkdev_get(bdev, mode, 0);
1476 EXPORT_SYMBOL(blkdev_get);
1478 static int blkdev_open(struct inode * inode, struct file * filp)
1480 struct block_device *whole = NULL;
1481 struct block_device *bdev;
1485 * Preserve backwards compatibility and allow large file access
1486 * even if userspace doesn't ask for it explicitly. Some mkfs
1487 * binary needs it. We might want to drop this workaround
1488 * during an unstable branch.
1490 filp->f_flags |= O_LARGEFILE;
1492 if (filp->f_flags & O_NDELAY)
1493 filp->f_mode |= FMODE_NDELAY;
1494 if (filp->f_flags & O_EXCL)
1495 filp->f_mode |= FMODE_EXCL;
1496 if ((filp->f_flags & O_ACCMODE) == 3)
1497 filp->f_mode |= FMODE_WRITE_IOCTL;
1499 bdev = bd_acquire(inode);
1503 if (filp->f_mode & FMODE_EXCL) {
1504 whole = bd_start_claiming(bdev, filp);
1505 if (IS_ERR(whole)) {
1507 return PTR_ERR(whole);
1511 filp->f_mapping = bdev->bd_inode->i_mapping;
1513 res = blkdev_get(bdev, filp->f_mode);
1517 bd_finish_claiming(bdev, whole, filp);
1519 bd_abort_claiming(whole, filp);
1525 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1528 struct gendisk *disk = bdev->bd_disk;
1529 struct block_device *victim = NULL;
1531 mutex_lock_nested(&bdev->bd_mutex, for_part);
1533 bdev->bd_part_count--;
1535 if (!--bdev->bd_openers) {
1536 sync_blockdev(bdev);
1539 if (bdev->bd_contains == bdev) {
1540 if (disk->fops->release)
1541 ret = disk->fops->release(disk, mode);
1543 if (!bdev->bd_openers) {
1544 struct module *owner = disk->fops->owner;
1548 disk_put_part(bdev->bd_part);
1549 bdev->bd_part = NULL;
1550 bdev->bd_disk = NULL;
1551 bdev_inode_switch_bdi(bdev->bd_inode,
1552 &default_backing_dev_info);
1553 if (bdev != bdev->bd_contains)
1554 victim = bdev->bd_contains;
1555 bdev->bd_contains = NULL;
1557 mutex_unlock(&bdev->bd_mutex);
1560 __blkdev_put(victim, mode, 1);
1564 int blkdev_put(struct block_device *bdev, fmode_t mode)
1566 return __blkdev_put(bdev, mode, 0);
1568 EXPORT_SYMBOL(blkdev_put);
1570 static int blkdev_close(struct inode * inode, struct file * filp)
1572 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1573 if (bdev->bd_holder == filp)
1575 return blkdev_put(bdev, filp->f_mode);
1578 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1580 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1581 fmode_t mode = file->f_mode;
1584 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1585 * to updated it before every ioctl.
1587 if (file->f_flags & O_NDELAY)
1588 mode |= FMODE_NDELAY;
1590 mode &= ~FMODE_NDELAY;
1592 return blkdev_ioctl(bdev, mode, cmd, arg);
1596 * Write data to the block device. Only intended for the block device itself
1597 * and the raw driver which basically is a fake block device.
1599 * Does not take i_mutex for the write and thus is not for general purpose
1602 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1603 unsigned long nr_segs, loff_t pos)
1605 struct file *file = iocb->ki_filp;
1608 BUG_ON(iocb->ki_pos != pos);
1610 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1611 if (ret > 0 || ret == -EIOCBQUEUED) {
1614 err = generic_write_sync(file, pos, ret);
1615 if (err < 0 && ret > 0)
1620 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1623 * Try to release a page associated with block device when the system
1624 * is under memory pressure.
1626 static int blkdev_releasepage(struct page *page, gfp_t wait)
1628 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1630 if (super && super->s_op->bdev_try_to_free_page)
1631 return super->s_op->bdev_try_to_free_page(super, page, wait);
1633 return try_to_free_buffers(page);
1636 static const struct address_space_operations def_blk_aops = {
1637 .readpage = blkdev_readpage,
1638 .writepage = blkdev_writepage,
1639 .sync_page = block_sync_page,
1640 .write_begin = blkdev_write_begin,
1641 .write_end = blkdev_write_end,
1642 .writepages = generic_writepages,
1643 .releasepage = blkdev_releasepage,
1644 .direct_IO = blkdev_direct_IO,
1647 const struct file_operations def_blk_fops = {
1648 .open = blkdev_open,
1649 .release = blkdev_close,
1650 .llseek = block_llseek,
1651 .read = do_sync_read,
1652 .write = do_sync_write,
1653 .aio_read = generic_file_aio_read,
1654 .aio_write = blkdev_aio_write,
1655 .mmap = generic_file_mmap,
1656 .fsync = blkdev_fsync,
1657 .unlocked_ioctl = block_ioctl,
1658 #ifdef CONFIG_COMPAT
1659 .compat_ioctl = compat_blkdev_ioctl,
1661 .splice_read = generic_file_splice_read,
1662 .splice_write = generic_file_splice_write,
1665 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1668 mm_segment_t old_fs = get_fs();
1670 res = blkdev_ioctl(bdev, 0, cmd, arg);
1675 EXPORT_SYMBOL(ioctl_by_bdev);
1678 * lookup_bdev - lookup a struct block_device by name
1679 * @pathname: special file representing the block device
1681 * Get a reference to the blockdevice at @pathname in the current
1682 * namespace if possible and return it. Return ERR_PTR(error)
1685 struct block_device *lookup_bdev(const char *pathname)
1687 struct block_device *bdev;
1688 struct inode *inode;
1692 if (!pathname || !*pathname)
1693 return ERR_PTR(-EINVAL);
1695 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1697 return ERR_PTR(error);
1699 inode = path.dentry->d_inode;
1701 if (!S_ISBLK(inode->i_mode))
1704 if (path.mnt->mnt_flags & MNT_NODEV)
1707 bdev = bd_acquire(inode);
1714 bdev = ERR_PTR(error);
1717 EXPORT_SYMBOL(lookup_bdev);
1720 * open_bdev_exclusive - open a block device by name and set it up for use
1722 * @path: special file representing the block device
1723 * @mode: FMODE_... combination to pass be used
1724 * @holder: owner for exclusion
1726 * Open the blockdevice described by the special file at @path, claim it
1729 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1731 struct block_device *bdev, *whole;
1734 bdev = lookup_bdev(path);
1738 whole = bd_start_claiming(bdev, holder);
1739 if (IS_ERR(whole)) {
1744 error = blkdev_get(bdev, mode);
1746 goto out_abort_claiming;
1749 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1750 goto out_blkdev_put;
1752 bd_finish_claiming(bdev, whole, holder);
1756 blkdev_put(bdev, mode);
1758 bd_abort_claiming(whole, holder);
1759 return ERR_PTR(error);
1762 EXPORT_SYMBOL(open_bdev_exclusive);
1765 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1767 * @bdev: blockdevice to close
1768 * @mode: mode, must match that used to open.
1770 * This is the counterpart to open_bdev_exclusive().
1772 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1775 blkdev_put(bdev, mode);
1778 EXPORT_SYMBOL(close_bdev_exclusive);
1780 int __invalidate_device(struct block_device *bdev)
1782 struct super_block *sb = get_super(bdev);
1787 * no need to lock the super, get_super holds the
1788 * read mutex so the filesystem cannot go away
1789 * under us (->put_super runs with the write lock
1792 shrink_dcache_sb(sb);
1793 res = invalidate_inodes(sb);
1796 invalidate_bdev(bdev);
1799 EXPORT_SYMBOL(__invalidate_device);