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/device_cgroup.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/pagevec.h>
21 #include <linux/writeback.h>
22 #include <linux/mpage.h>
23 #include <linux/mount.h>
24 #include <linux/uio.h>
25 #include <linux/namei.h>
26 #include <linux/log2.h>
27 #include <linux/kmemleak.h>
28 #include <asm/uaccess.h>
32 struct block_device bdev;
33 struct inode vfs_inode;
36 static const struct address_space_operations def_blk_aops;
38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
40 return container_of(inode, struct bdev_inode, vfs_inode);
43 inline struct block_device *I_BDEV(struct inode *inode)
45 return &BDEV_I(inode)->bdev;
48 EXPORT_SYMBOL(I_BDEV);
51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
52 * we need to move it onto the dirty list of @dst so that the inode is always
55 static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
58 spin_lock(&inode_lock);
59 inode->i_data.backing_dev_info = dst;
60 if (inode->i_state & I_DIRTY)
61 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
62 spin_unlock(&inode_lock);
65 static sector_t max_block(struct block_device *bdev)
67 sector_t retval = ~((sector_t)0);
68 loff_t sz = i_size_read(bdev->bd_inode);
71 unsigned int size = block_size(bdev);
72 unsigned int sizebits = blksize_bits(size);
73 retval = (sz >> sizebits);
78 /* Kill _all_ buffers and pagecache , dirty or not.. */
79 static void kill_bdev(struct block_device *bdev)
81 if (bdev->bd_inode->i_mapping->nrpages == 0)
84 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
87 int set_blocksize(struct block_device *bdev, int size)
89 /* Size must be a power of two, and between 512 and PAGE_SIZE */
90 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
93 /* Size cannot be smaller than the size supported by the device */
94 if (size < bdev_logical_block_size(bdev))
97 /* Don't change the size if it is same as current */
98 if (bdev->bd_block_size != size) {
100 bdev->bd_block_size = size;
101 bdev->bd_inode->i_blkbits = blksize_bits(size);
107 EXPORT_SYMBOL(set_blocksize);
109 int sb_set_blocksize(struct super_block *sb, int size)
111 if (set_blocksize(sb->s_bdev, size))
113 /* If we get here, we know size is power of two
114 * and it's value is between 512 and PAGE_SIZE */
115 sb->s_blocksize = size;
116 sb->s_blocksize_bits = blksize_bits(size);
117 return sb->s_blocksize;
120 EXPORT_SYMBOL(sb_set_blocksize);
122 int sb_min_blocksize(struct super_block *sb, int size)
124 int minsize = bdev_logical_block_size(sb->s_bdev);
127 return sb_set_blocksize(sb, size);
130 EXPORT_SYMBOL(sb_min_blocksize);
133 blkdev_get_block(struct inode *inode, sector_t iblock,
134 struct buffer_head *bh, int create)
136 if (iblock >= max_block(I_BDEV(inode))) {
141 * for reads, we're just trying to fill a partial page.
142 * return a hole, they will have to call get_block again
143 * before they can fill it, and they will get -EIO at that
148 bh->b_bdev = I_BDEV(inode);
149 bh->b_blocknr = iblock;
150 set_buffer_mapped(bh);
155 blkdev_get_blocks(struct inode *inode, sector_t iblock,
156 struct buffer_head *bh, int create)
158 sector_t end_block = max_block(I_BDEV(inode));
159 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
161 if ((iblock + max_blocks) > end_block) {
162 max_blocks = end_block - iblock;
163 if ((long)max_blocks <= 0) {
165 return -EIO; /* write fully beyond EOF */
167 * It is a read which is fully beyond EOF. We return
168 * a !buffer_mapped buffer
174 bh->b_bdev = I_BDEV(inode);
175 bh->b_blocknr = iblock;
176 bh->b_size = max_blocks << inode->i_blkbits;
178 set_buffer_mapped(bh);
183 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
184 loff_t offset, unsigned long nr_segs)
186 struct file *file = iocb->ki_filp;
187 struct inode *inode = file->f_mapping->host;
189 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
190 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
193 int __sync_blockdev(struct block_device *bdev, int wait)
198 return filemap_flush(bdev->bd_inode->i_mapping);
199 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
203 * Write out and wait upon all the dirty data associated with a block
204 * device via its mapping. Does not take the superblock lock.
206 int sync_blockdev(struct block_device *bdev)
208 return __sync_blockdev(bdev, 1);
210 EXPORT_SYMBOL(sync_blockdev);
213 * Write out and wait upon all dirty data associated with this
214 * device. Filesystem data as well as the underlying block
215 * device. Takes the superblock lock.
217 int fsync_bdev(struct block_device *bdev)
219 struct super_block *sb = get_super(bdev);
221 int res = sync_filesystem(sb);
225 return sync_blockdev(bdev);
227 EXPORT_SYMBOL(fsync_bdev);
230 * freeze_bdev -- lock a filesystem and force it into a consistent state
231 * @bdev: blockdevice to lock
233 * If a superblock is found on this device, we take the s_umount semaphore
234 * on it to make sure nobody unmounts until the snapshot creation is done.
235 * The reference counter (bd_fsfreeze_count) guarantees that only the last
236 * unfreeze process can unfreeze the frozen filesystem actually when multiple
237 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
238 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
241 struct super_block *freeze_bdev(struct block_device *bdev)
243 struct super_block *sb;
246 mutex_lock(&bdev->bd_fsfreeze_mutex);
247 if (++bdev->bd_fsfreeze_count > 1) {
249 * We don't even need to grab a reference - the first call
250 * to freeze_bdev grab an active reference and only the last
251 * thaw_bdev drops it.
253 sb = get_super(bdev);
255 mutex_unlock(&bdev->bd_fsfreeze_mutex);
259 sb = get_active_super(bdev);
262 error = freeze_super(sb);
264 deactivate_super(sb);
265 bdev->bd_fsfreeze_count--;
266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
267 return ERR_PTR(error);
269 deactivate_super(sb);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273 return sb; /* thaw_bdev releases s->s_umount */
275 EXPORT_SYMBOL(freeze_bdev);
278 * thaw_bdev -- unlock filesystem
279 * @bdev: blockdevice to unlock
280 * @sb: associated superblock
282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
288 mutex_lock(&bdev->bd_fsfreeze_mutex);
289 if (!bdev->bd_fsfreeze_count)
293 if (--bdev->bd_fsfreeze_count > 0)
299 error = thaw_super(sb);
301 bdev->bd_fsfreeze_count++;
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
306 mutex_unlock(&bdev->bd_fsfreeze_mutex);
309 EXPORT_SYMBOL(thaw_bdev);
311 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
313 return block_write_full_page(page, blkdev_get_block, wbc);
316 static int blkdev_readpage(struct file * file, struct page * page)
318 return block_read_full_page(page, blkdev_get_block);
321 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
322 loff_t pos, unsigned len, unsigned flags,
323 struct page **pagep, void **fsdata)
325 return block_write_begin(mapping, pos, len, flags, pagep,
329 static int blkdev_write_end(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned copied,
331 struct page *page, void *fsdata)
334 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
337 page_cache_release(page);
344 * for a block special file file->f_path.dentry->d_inode->i_size is zero
345 * so we compute the size by hand (just as in block_read/write above)
347 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
349 struct inode *bd_inode = file->f_mapping->host;
353 mutex_lock(&bd_inode->i_mutex);
354 size = i_size_read(bd_inode);
361 offset += file->f_pos;
364 if (offset >= 0 && offset <= size) {
365 if (offset != file->f_pos) {
366 file->f_pos = offset;
370 mutex_unlock(&bd_inode->i_mutex);
374 int blkdev_fsync(struct file *filp, int datasync)
376 struct inode *bd_inode = filp->f_mapping->host;
377 struct block_device *bdev = I_BDEV(bd_inode);
381 * There is no need to serialise calls to blkdev_issue_flush with
382 * i_mutex and doing so causes performance issues with concurrent
383 * O_SYNC writers to a block device.
385 mutex_unlock(&bd_inode->i_mutex);
387 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
388 if (error == -EOPNOTSUPP)
391 mutex_lock(&bd_inode->i_mutex);
395 EXPORT_SYMBOL(blkdev_fsync);
401 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
402 static struct kmem_cache * bdev_cachep __read_mostly;
404 static struct inode *bdev_alloc_inode(struct super_block *sb)
406 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
409 return &ei->vfs_inode;
412 static void bdev_destroy_inode(struct inode *inode)
414 struct bdev_inode *bdi = BDEV_I(inode);
416 kmem_cache_free(bdev_cachep, bdi);
419 static void init_once(void *foo)
421 struct bdev_inode *ei = (struct bdev_inode *) foo;
422 struct block_device *bdev = &ei->bdev;
424 memset(bdev, 0, sizeof(*bdev));
425 mutex_init(&bdev->bd_mutex);
426 INIT_LIST_HEAD(&bdev->bd_inodes);
427 INIT_LIST_HEAD(&bdev->bd_list);
429 INIT_LIST_HEAD(&bdev->bd_holder_list);
431 inode_init_once(&ei->vfs_inode);
432 /* Initialize mutex for freeze. */
433 mutex_init(&bdev->bd_fsfreeze_mutex);
436 static inline void __bd_forget(struct inode *inode)
438 list_del_init(&inode->i_devices);
439 inode->i_bdev = NULL;
440 inode->i_mapping = &inode->i_data;
443 static void bdev_evict_inode(struct inode *inode)
445 struct block_device *bdev = &BDEV_I(inode)->bdev;
447 truncate_inode_pages(&inode->i_data, 0);
448 invalidate_inode_buffers(inode); /* is it needed here? */
449 end_writeback(inode);
450 spin_lock(&bdev_lock);
451 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
452 __bd_forget(list_entry(p, struct inode, i_devices));
454 list_del_init(&bdev->bd_list);
455 spin_unlock(&bdev_lock);
458 static const struct super_operations bdev_sops = {
459 .statfs = simple_statfs,
460 .alloc_inode = bdev_alloc_inode,
461 .destroy_inode = bdev_destroy_inode,
462 .drop_inode = generic_delete_inode,
463 .evict_inode = bdev_evict_inode,
466 static struct dentry *bd_mount(struct file_system_type *fs_type,
467 int flags, const char *dev_name, void *data)
469 return mount_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
472 static struct file_system_type bd_type = {
475 .kill_sb = kill_anon_super,
478 struct super_block *blockdev_superblock __read_mostly;
480 void __init bdev_cache_init(void)
483 struct vfsmount *bd_mnt;
485 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
486 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
487 SLAB_MEM_SPREAD|SLAB_PANIC),
489 err = register_filesystem(&bd_type);
491 panic("Cannot register bdev pseudo-fs");
492 bd_mnt = kern_mount(&bd_type);
494 panic("Cannot create bdev pseudo-fs");
496 * This vfsmount structure is only used to obtain the
497 * blockdev_superblock, so tell kmemleak not to report it.
499 kmemleak_not_leak(bd_mnt);
500 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
504 * Most likely _very_ bad one - but then it's hardly critical for small
505 * /dev and can be fixed when somebody will need really large one.
506 * Keep in mind that it will be fed through icache hash function too.
508 static inline unsigned long hash(dev_t dev)
510 return MAJOR(dev)+MINOR(dev);
513 static int bdev_test(struct inode *inode, void *data)
515 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
518 static int bdev_set(struct inode *inode, void *data)
520 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
524 static LIST_HEAD(all_bdevs);
526 struct block_device *bdget(dev_t dev)
528 struct block_device *bdev;
531 inode = iget5_locked(blockdev_superblock, hash(dev),
532 bdev_test, bdev_set, &dev);
537 bdev = &BDEV_I(inode)->bdev;
539 if (inode->i_state & I_NEW) {
540 bdev->bd_contains = NULL;
541 bdev->bd_inode = inode;
542 bdev->bd_block_size = (1 << inode->i_blkbits);
543 bdev->bd_part_count = 0;
544 bdev->bd_invalidated = 0;
545 inode->i_mode = S_IFBLK;
547 inode->i_bdev = bdev;
548 inode->i_data.a_ops = &def_blk_aops;
549 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
550 inode->i_data.backing_dev_info = &default_backing_dev_info;
551 spin_lock(&bdev_lock);
552 list_add(&bdev->bd_list, &all_bdevs);
553 spin_unlock(&bdev_lock);
554 unlock_new_inode(inode);
559 EXPORT_SYMBOL(bdget);
562 * bdgrab -- Grab a reference to an already referenced block device
563 * @bdev: Block device to grab a reference to.
565 struct block_device *bdgrab(struct block_device *bdev)
567 ihold(bdev->bd_inode);
571 long nr_blockdev_pages(void)
573 struct block_device *bdev;
575 spin_lock(&bdev_lock);
576 list_for_each_entry(bdev, &all_bdevs, bd_list) {
577 ret += bdev->bd_inode->i_mapping->nrpages;
579 spin_unlock(&bdev_lock);
583 void bdput(struct block_device *bdev)
585 iput(bdev->bd_inode);
588 EXPORT_SYMBOL(bdput);
590 static struct block_device *bd_acquire(struct inode *inode)
592 struct block_device *bdev;
594 spin_lock(&bdev_lock);
595 bdev = inode->i_bdev;
597 ihold(bdev->bd_inode);
598 spin_unlock(&bdev_lock);
601 spin_unlock(&bdev_lock);
603 bdev = bdget(inode->i_rdev);
605 spin_lock(&bdev_lock);
606 if (!inode->i_bdev) {
608 * We take an additional reference to bd_inode,
609 * and it's released in clear_inode() of inode.
610 * So, we can access it via ->i_mapping always
613 ihold(bdev->bd_inode);
614 inode->i_bdev = bdev;
615 inode->i_mapping = bdev->bd_inode->i_mapping;
616 list_add(&inode->i_devices, &bdev->bd_inodes);
618 spin_unlock(&bdev_lock);
623 /* Call when you free inode */
625 void bd_forget(struct inode *inode)
627 struct block_device *bdev = NULL;
629 spin_lock(&bdev_lock);
631 if (!sb_is_blkdev_sb(inode->i_sb))
632 bdev = inode->i_bdev;
635 spin_unlock(&bdev_lock);
638 iput(bdev->bd_inode);
642 * bd_may_claim - test whether a block device can be claimed
643 * @bdev: block device of interest
644 * @whole: whole block device containing @bdev, may equal @bdev
645 * @holder: holder trying to claim @bdev
647 * Test whther @bdev can be claimed by @holder.
650 * spin_lock(&bdev_lock).
653 * %true if @bdev can be claimed, %false otherwise.
655 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
658 if (bdev->bd_holder == holder)
659 return true; /* already a holder */
660 else if (bdev->bd_holder != NULL)
661 return false; /* held by someone else */
662 else if (bdev->bd_contains == bdev)
663 return true; /* is a whole device which isn't held */
665 else if (whole->bd_holder == bd_claim)
666 return true; /* is a partition of a device that is being partitioned */
667 else if (whole->bd_holder != NULL)
668 return false; /* is a partition of a held device */
670 return true; /* is a partition of an un-held device */
674 * bd_prepare_to_claim - prepare to claim a block device
675 * @bdev: block device of interest
676 * @whole: the whole device containing @bdev, may equal @bdev
677 * @holder: holder trying to claim @bdev
679 * Prepare to claim @bdev. This function fails if @bdev is already
680 * claimed by another holder and waits if another claiming is in
681 * progress. This function doesn't actually claim. On successful
682 * return, the caller has ownership of bd_claiming and bd_holder[s].
685 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
689 * 0 if @bdev can be claimed, -EBUSY otherwise.
691 static int bd_prepare_to_claim(struct block_device *bdev,
692 struct block_device *whole, void *holder)
695 /* if someone else claimed, fail */
696 if (!bd_may_claim(bdev, whole, holder))
699 /* if claiming is already in progress, wait for it to finish */
700 if (whole->bd_claiming) {
701 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
704 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
705 spin_unlock(&bdev_lock);
707 finish_wait(wq, &wait);
708 spin_lock(&bdev_lock);
717 * bd_start_claiming - start claiming a block device
718 * @bdev: block device of interest
719 * @holder: holder trying to claim @bdev
721 * @bdev is about to be opened exclusively. Check @bdev can be opened
722 * exclusively and mark that an exclusive open is in progress. Each
723 * successful call to this function must be matched with a call to
724 * either bd_finish_claiming() or bd_abort_claiming() (which do not
727 * This function is used to gain exclusive access to the block device
728 * without actually causing other exclusive open attempts to fail. It
729 * should be used when the open sequence itself requires exclusive
730 * access but may subsequently fail.
736 * Pointer to the block device containing @bdev on success, ERR_PTR()
739 static struct block_device *bd_start_claiming(struct block_device *bdev,
742 struct gendisk *disk;
743 struct block_device *whole;
749 * @bdev might not have been initialized properly yet, look up
750 * and grab the outer block device the hard way.
752 disk = get_gendisk(bdev->bd_dev, &partno);
754 return ERR_PTR(-ENXIO);
756 whole = bdget_disk(disk, 0);
757 module_put(disk->fops->owner);
760 return ERR_PTR(-ENOMEM);
762 /* prepare to claim, if successful, mark claiming in progress */
763 spin_lock(&bdev_lock);
765 err = bd_prepare_to_claim(bdev, whole, holder);
767 whole->bd_claiming = holder;
768 spin_unlock(&bdev_lock);
771 spin_unlock(&bdev_lock);
777 /* releases bdev_lock */
778 static void __bd_abort_claiming(struct block_device *whole, void *holder)
780 BUG_ON(whole->bd_claiming != holder);
781 whole->bd_claiming = NULL;
782 wake_up_bit(&whole->bd_claiming, 0);
784 spin_unlock(&bdev_lock);
789 * bd_abort_claiming - abort claiming a block device
790 * @whole: whole block device returned by bd_start_claiming()
791 * @holder: holder trying to claim @bdev
793 * Abort a claiming block started by bd_start_claiming(). Note that
794 * @whole is not the block device to be claimed but the whole device
795 * returned by bd_start_claiming().
798 * Grabs and releases bdev_lock.
800 static void bd_abort_claiming(struct block_device *whole, void *holder)
802 spin_lock(&bdev_lock);
803 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
806 /* increment holders when we have a legitimate claim. requires bdev_lock */
807 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
810 /* note that for a whole device bd_holders
811 * will be incremented twice, and bd_holder will
812 * be set to bd_claim before being set to holder
815 whole->bd_holder = bd_claim;
817 bdev->bd_holder = holder;
821 * bd_finish_claiming - finish claiming a block device
822 * @bdev: block device of interest (passed to bd_start_claiming())
823 * @whole: whole block device returned by bd_start_claiming()
824 * @holder: holder trying to claim @bdev
826 * Finish a claiming block started by bd_start_claiming().
829 * Grabs and releases bdev_lock.
831 static void bd_finish_claiming(struct block_device *bdev,
832 struct block_device *whole, void *holder)
834 spin_lock(&bdev_lock);
835 BUG_ON(!bd_may_claim(bdev, whole, holder));
836 __bd_claim(bdev, whole, holder);
837 __bd_abort_claiming(whole, holder); /* not actually an abort */
841 * bd_claim - claim a block device
842 * @bdev: block device to claim
843 * @holder: holder trying to claim @bdev
845 * Try to claim @bdev which must have been opened successfully.
851 * 0 if successful, -EBUSY if @bdev is already claimed.
853 int bd_claim(struct block_device *bdev, void *holder)
855 struct block_device *whole = bdev->bd_contains;
860 spin_lock(&bdev_lock);
861 res = bd_prepare_to_claim(bdev, whole, holder);
863 __bd_claim(bdev, whole, holder);
864 spin_unlock(&bdev_lock);
868 EXPORT_SYMBOL(bd_claim);
870 void bd_release(struct block_device *bdev)
872 spin_lock(&bdev_lock);
873 if (!--bdev->bd_contains->bd_holders)
874 bdev->bd_contains->bd_holder = NULL;
875 if (!--bdev->bd_holders)
876 bdev->bd_holder = NULL;
877 spin_unlock(&bdev_lock);
880 EXPORT_SYMBOL(bd_release);
884 * Functions for bd_claim_by_kobject / bd_release_from_kobject
886 * If a kobject is passed to bd_claim_by_kobject()
887 * and the kobject has a parent directory,
888 * following symlinks are created:
889 * o from the kobject to the claimed bdev
890 * o from "holders" directory of the bdev to the parent of the kobject
891 * bd_release_from_kobject() removes these symlinks.
894 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
895 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
896 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
897 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
900 static int add_symlink(struct kobject *from, struct kobject *to)
904 return sysfs_create_link(from, to, kobject_name(to));
907 static void del_symlink(struct kobject *from, struct kobject *to)
911 sysfs_remove_link(from, kobject_name(to));
915 * 'struct bd_holder' contains pointers to kobjects symlinked by
916 * bd_claim_by_kobject.
917 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
920 struct list_head list; /* chain of holders of the bdev */
921 int count; /* references from the holder */
922 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
923 struct kobject *hdev; /* e.g. "/block/dm-0" */
924 struct kobject *hdir; /* e.g. "/block/sda/holders" */
925 struct kobject *sdev; /* e.g. "/block/sda" */
929 * Get references of related kobjects at once.
930 * Returns 1 on success. 0 on failure.
932 * Should call bd_holder_release_dirs() after successful use.
934 static int bd_holder_grab_dirs(struct block_device *bdev,
935 struct bd_holder *bo)
940 bo->sdir = kobject_get(bo->sdir);
944 bo->hdev = kobject_get(bo->sdir->parent);
948 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
952 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
959 kobject_put(bo->sdev);
961 kobject_put(bo->hdev);
963 kobject_put(bo->sdir);
968 /* Put references of related kobjects at once. */
969 static void bd_holder_release_dirs(struct bd_holder *bo)
971 kobject_put(bo->hdir);
972 kobject_put(bo->sdev);
973 kobject_put(bo->hdev);
974 kobject_put(bo->sdir);
977 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
979 struct bd_holder *bo;
981 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
991 static void free_bd_holder(struct bd_holder *bo)
997 * find_bd_holder - find matching struct bd_holder from the block device
999 * @bdev: struct block device to be searched
1000 * @bo: target struct bd_holder
1002 * Returns matching entry with @bo in @bdev->bd_holder_list.
1003 * If found, increment the reference count and return the pointer.
1004 * If not found, returns NULL.
1006 static struct bd_holder *find_bd_holder(struct block_device *bdev,
1007 struct bd_holder *bo)
1009 struct bd_holder *tmp;
1011 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
1012 if (tmp->sdir == bo->sdir) {
1021 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1023 * @bdev: block device to be bd_claimed
1024 * @bo: preallocated and initialized by alloc_bd_holder()
1026 * Add @bo to @bdev->bd_holder_list, create symlinks.
1028 * Returns 0 if symlinks are created.
1029 * Returns -ve if something fails.
1031 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1038 if (!bd_holder_grab_dirs(bdev, bo))
1041 err = add_symlink(bo->sdir, bo->sdev);
1045 err = add_symlink(bo->hdir, bo->hdev);
1047 del_symlink(bo->sdir, bo->sdev);
1051 list_add_tail(&bo->list, &bdev->bd_holder_list);
1056 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1058 * @bdev: block device to be bd_claimed
1059 * @kobj: holder's kobject
1061 * If there is matching entry with @kobj in @bdev->bd_holder_list
1062 * and no other bd_claim() from the same kobject,
1063 * remove the struct bd_holder from the list, delete symlinks for it.
1065 * Returns a pointer to the struct bd_holder when it's removed from the list
1066 * and ready to be freed.
1067 * Returns NULL if matching claim isn't found or there is other bd_claim()
1068 * by the same kobject.
1070 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1071 struct kobject *kobj)
1073 struct bd_holder *bo;
1075 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1076 if (bo->sdir == kobj) {
1078 BUG_ON(bo->count < 0);
1080 list_del(&bo->list);
1081 del_symlink(bo->sdir, bo->sdev);
1082 del_symlink(bo->hdir, bo->hdev);
1083 bd_holder_release_dirs(bo);
1094 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1096 * @bdev: block device to be claimed
1097 * @holder: holder's signature
1098 * @kobj: holder's kobject
1100 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1101 * the bdev and the holder's kobject.
1102 * Use bd_release_from_kobject() when relesing the claimed bdev.
1104 * Returns 0 on success. (same as bd_claim())
1105 * Returns errno on failure.
1107 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1108 struct kobject *kobj)
1111 struct bd_holder *bo, *found;
1116 bo = alloc_bd_holder(kobj);
1120 mutex_lock(&bdev->bd_mutex);
1122 err = bd_claim(bdev, holder);
1126 found = find_bd_holder(bdev, bo);
1130 err = add_bd_holder(bdev, bo);
1136 mutex_unlock(&bdev->bd_mutex);
1142 * bd_release_from_kobject - bd_release() with additional kobject signature
1144 * @bdev: block device to be released
1145 * @kobj: holder's kobject
1147 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1149 static void bd_release_from_kobject(struct block_device *bdev,
1150 struct kobject *kobj)
1155 mutex_lock(&bdev->bd_mutex);
1157 free_bd_holder(del_bd_holder(bdev, kobj));
1158 mutex_unlock(&bdev->bd_mutex);
1162 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1164 * @bdev: block device to be claimed
1165 * @holder: holder's signature
1166 * @disk: holder's gendisk
1168 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1170 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1171 struct gendisk *disk)
1173 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1175 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1178 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1180 * @bdev: block device to be claimed
1181 * @disk: holder's gendisk
1183 * Call bd_release_from_kobject() and put @disk->slave_dir.
1185 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1187 bd_release_from_kobject(bdev, disk->slave_dir);
1188 kobject_put(disk->slave_dir);
1190 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1194 * Tries to open block device by device number. Use it ONLY if you
1195 * really do not have anything better - i.e. when you are behind a
1196 * truly sucky interface and all you are given is a device number. _Never_
1197 * to be used for internal purposes. If you ever need it - reconsider
1200 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1202 struct block_device *bdev = bdget(dev);
1205 err = blkdev_get(bdev, mode);
1206 return err ? ERR_PTR(err) : bdev;
1209 EXPORT_SYMBOL(open_by_devnum);
1212 * flush_disk - invalidates all buffer-cache entries on a disk
1214 * @bdev: struct block device to be flushed
1216 * Invalidates all buffer-cache entries on a disk. It should be called
1217 * when a disk has been changed -- either by a media change or online
1220 static void flush_disk(struct block_device *bdev)
1222 if (__invalidate_device(bdev)) {
1223 char name[BDEVNAME_SIZE] = "";
1226 disk_name(bdev->bd_disk, 0, name);
1227 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1228 "resized disk %s\n", name);
1233 if (disk_partitionable(bdev->bd_disk))
1234 bdev->bd_invalidated = 1;
1238 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1239 * @disk: struct gendisk to check
1240 * @bdev: struct bdev to adjust.
1242 * This routine checks to see if the bdev size does not match the disk size
1243 * and adjusts it if it differs.
1245 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1247 loff_t disk_size, bdev_size;
1249 disk_size = (loff_t)get_capacity(disk) << 9;
1250 bdev_size = i_size_read(bdev->bd_inode);
1251 if (disk_size != bdev_size) {
1252 char name[BDEVNAME_SIZE];
1254 disk_name(disk, 0, name);
1256 "%s: detected capacity change from %lld to %lld\n",
1257 name, bdev_size, disk_size);
1258 i_size_write(bdev->bd_inode, disk_size);
1262 EXPORT_SYMBOL(check_disk_size_change);
1265 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1266 * @disk: struct gendisk to be revalidated
1268 * This routine is a wrapper for lower-level driver's revalidate_disk
1269 * call-backs. It is used to do common pre and post operations needed
1270 * for all revalidate_disk operations.
1272 int revalidate_disk(struct gendisk *disk)
1274 struct block_device *bdev;
1277 if (disk->fops->revalidate_disk)
1278 ret = disk->fops->revalidate_disk(disk);
1280 bdev = bdget_disk(disk, 0);
1284 mutex_lock(&bdev->bd_mutex);
1285 check_disk_size_change(disk, bdev);
1286 mutex_unlock(&bdev->bd_mutex);
1290 EXPORT_SYMBOL(revalidate_disk);
1293 * This routine checks whether a removable media has been changed,
1294 * and invalidates all buffer-cache-entries in that case. This
1295 * is a relatively slow routine, so we have to try to minimize using
1296 * it. Thus it is called only upon a 'mount' or 'open'. This
1297 * is the best way of combining speed and utility, I think.
1298 * People changing diskettes in the middle of an operation deserve
1301 int check_disk_change(struct block_device *bdev)
1303 struct gendisk *disk = bdev->bd_disk;
1304 const struct block_device_operations *bdops = disk->fops;
1306 if (!bdops->media_changed)
1308 if (!bdops->media_changed(bdev->bd_disk))
1312 if (bdops->revalidate_disk)
1313 bdops->revalidate_disk(bdev->bd_disk);
1317 EXPORT_SYMBOL(check_disk_change);
1319 void bd_set_size(struct block_device *bdev, loff_t size)
1321 unsigned bsize = bdev_logical_block_size(bdev);
1323 bdev->bd_inode->i_size = size;
1324 while (bsize < PAGE_CACHE_SIZE) {
1329 bdev->bd_block_size = bsize;
1330 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1332 EXPORT_SYMBOL(bd_set_size);
1334 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1339 * mutex_lock(part->bd_mutex)
1340 * mutex_lock_nested(whole->bd_mutex, 1)
1343 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1345 struct gendisk *disk;
1350 if (mode & FMODE_READ)
1352 if (mode & FMODE_WRITE)
1355 * hooks: /n/, see "layering violations".
1358 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1368 disk = get_gendisk(bdev->bd_dev, &partno);
1372 mutex_lock_nested(&bdev->bd_mutex, for_part);
1373 if (!bdev->bd_openers) {
1374 bdev->bd_disk = disk;
1375 bdev->bd_contains = bdev;
1377 struct backing_dev_info *bdi;
1380 bdev->bd_part = disk_get_part(disk, partno);
1384 if (disk->fops->open) {
1385 ret = disk->fops->open(bdev, mode);
1386 if (ret == -ERESTARTSYS) {
1387 /* Lost a race with 'disk' being
1388 * deleted, try again.
1391 disk_put_part(bdev->bd_part);
1392 bdev->bd_part = NULL;
1393 module_put(disk->fops->owner);
1395 bdev->bd_disk = NULL;
1396 mutex_unlock(&bdev->bd_mutex);
1402 if (!bdev->bd_openers) {
1403 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1404 bdi = blk_get_backing_dev_info(bdev);
1406 bdi = &default_backing_dev_info;
1407 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1409 if (bdev->bd_invalidated)
1410 rescan_partitions(disk, bdev);
1412 struct block_device *whole;
1413 whole = bdget_disk(disk, 0);
1418 ret = __blkdev_get(whole, mode, 1);
1421 bdev->bd_contains = whole;
1422 bdev_inode_switch_bdi(bdev->bd_inode,
1423 whole->bd_inode->i_data.backing_dev_info);
1424 bdev->bd_part = disk_get_part(disk, partno);
1425 if (!(disk->flags & GENHD_FL_UP) ||
1426 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1430 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1433 module_put(disk->fops->owner);
1436 if (bdev->bd_contains == bdev) {
1437 if (bdev->bd_disk->fops->open) {
1438 ret = bdev->bd_disk->fops->open(bdev, mode);
1440 goto out_unlock_bdev;
1442 if (bdev->bd_invalidated)
1443 rescan_partitions(bdev->bd_disk, bdev);
1448 bdev->bd_part_count++;
1449 mutex_unlock(&bdev->bd_mutex);
1453 disk_put_part(bdev->bd_part);
1454 bdev->bd_disk = NULL;
1455 bdev->bd_part = NULL;
1456 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1457 if (bdev != bdev->bd_contains)
1458 __blkdev_put(bdev->bd_contains, mode, 1);
1459 bdev->bd_contains = NULL;
1461 mutex_unlock(&bdev->bd_mutex);
1464 module_put(disk->fops->owner);
1471 int blkdev_get(struct block_device *bdev, fmode_t mode)
1473 return __blkdev_get(bdev, mode, 0);
1475 EXPORT_SYMBOL(blkdev_get);
1477 static int blkdev_open(struct inode * inode, struct file * filp)
1479 struct block_device *whole = NULL;
1480 struct block_device *bdev;
1484 * Preserve backwards compatibility and allow large file access
1485 * even if userspace doesn't ask for it explicitly. Some mkfs
1486 * binary needs it. We might want to drop this workaround
1487 * during an unstable branch.
1489 filp->f_flags |= O_LARGEFILE;
1491 if (filp->f_flags & O_NDELAY)
1492 filp->f_mode |= FMODE_NDELAY;
1493 if (filp->f_flags & O_EXCL)
1494 filp->f_mode |= FMODE_EXCL;
1495 if ((filp->f_flags & O_ACCMODE) == 3)
1496 filp->f_mode |= FMODE_WRITE_IOCTL;
1498 bdev = bd_acquire(inode);
1502 if (filp->f_mode & FMODE_EXCL) {
1503 whole = bd_start_claiming(bdev, filp);
1504 if (IS_ERR(whole)) {
1506 return PTR_ERR(whole);
1510 filp->f_mapping = bdev->bd_inode->i_mapping;
1512 res = blkdev_get(bdev, filp->f_mode);
1516 bd_finish_claiming(bdev, whole, filp);
1518 bd_abort_claiming(whole, filp);
1524 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1527 struct gendisk *disk = bdev->bd_disk;
1528 struct block_device *victim = NULL;
1530 mutex_lock_nested(&bdev->bd_mutex, for_part);
1532 bdev->bd_part_count--;
1534 if (!--bdev->bd_openers) {
1535 sync_blockdev(bdev);
1538 if (bdev->bd_contains == bdev) {
1539 if (disk->fops->release)
1540 ret = disk->fops->release(disk, mode);
1542 if (!bdev->bd_openers) {
1543 struct module *owner = disk->fops->owner;
1547 disk_put_part(bdev->bd_part);
1548 bdev->bd_part = NULL;
1549 bdev->bd_disk = NULL;
1550 bdev_inode_switch_bdi(bdev->bd_inode,
1551 &default_backing_dev_info);
1552 if (bdev != bdev->bd_contains)
1553 victim = bdev->bd_contains;
1554 bdev->bd_contains = NULL;
1556 mutex_unlock(&bdev->bd_mutex);
1559 __blkdev_put(victim, mode, 1);
1563 int blkdev_put(struct block_device *bdev, fmode_t mode)
1565 return __blkdev_put(bdev, mode, 0);
1567 EXPORT_SYMBOL(blkdev_put);
1569 static int blkdev_close(struct inode * inode, struct file * filp)
1571 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1572 if (bdev->bd_holder == filp)
1574 return blkdev_put(bdev, filp->f_mode);
1577 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1579 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1580 fmode_t mode = file->f_mode;
1583 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1584 * to updated it before every ioctl.
1586 if (file->f_flags & O_NDELAY)
1587 mode |= FMODE_NDELAY;
1589 mode &= ~FMODE_NDELAY;
1591 return blkdev_ioctl(bdev, mode, cmd, arg);
1595 * Write data to the block device. Only intended for the block device itself
1596 * and the raw driver which basically is a fake block device.
1598 * Does not take i_mutex for the write and thus is not for general purpose
1601 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1602 unsigned long nr_segs, loff_t pos)
1604 struct file *file = iocb->ki_filp;
1607 BUG_ON(iocb->ki_pos != pos);
1609 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1610 if (ret > 0 || ret == -EIOCBQUEUED) {
1613 err = generic_write_sync(file, pos, ret);
1614 if (err < 0 && ret > 0)
1619 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1622 * Try to release a page associated with block device when the system
1623 * is under memory pressure.
1625 static int blkdev_releasepage(struct page *page, gfp_t wait)
1627 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1629 if (super && super->s_op->bdev_try_to_free_page)
1630 return super->s_op->bdev_try_to_free_page(super, page, wait);
1632 return try_to_free_buffers(page);
1635 static const struct address_space_operations def_blk_aops = {
1636 .readpage = blkdev_readpage,
1637 .writepage = blkdev_writepage,
1638 .sync_page = block_sync_page,
1639 .write_begin = blkdev_write_begin,
1640 .write_end = blkdev_write_end,
1641 .writepages = generic_writepages,
1642 .releasepage = blkdev_releasepage,
1643 .direct_IO = blkdev_direct_IO,
1646 const struct file_operations def_blk_fops = {
1647 .open = blkdev_open,
1648 .release = blkdev_close,
1649 .llseek = block_llseek,
1650 .read = do_sync_read,
1651 .write = do_sync_write,
1652 .aio_read = generic_file_aio_read,
1653 .aio_write = blkdev_aio_write,
1654 .mmap = generic_file_mmap,
1655 .fsync = blkdev_fsync,
1656 .unlocked_ioctl = block_ioctl,
1657 #ifdef CONFIG_COMPAT
1658 .compat_ioctl = compat_blkdev_ioctl,
1660 .splice_read = generic_file_splice_read,
1661 .splice_write = generic_file_splice_write,
1664 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1667 mm_segment_t old_fs = get_fs();
1669 res = blkdev_ioctl(bdev, 0, cmd, arg);
1674 EXPORT_SYMBOL(ioctl_by_bdev);
1677 * lookup_bdev - lookup a struct block_device by name
1678 * @pathname: special file representing the block device
1680 * Get a reference to the blockdevice at @pathname in the current
1681 * namespace if possible and return it. Return ERR_PTR(error)
1684 struct block_device *lookup_bdev(const char *pathname)
1686 struct block_device *bdev;
1687 struct inode *inode;
1691 if (!pathname || !*pathname)
1692 return ERR_PTR(-EINVAL);
1694 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1696 return ERR_PTR(error);
1698 inode = path.dentry->d_inode;
1700 if (!S_ISBLK(inode->i_mode))
1703 if (path.mnt->mnt_flags & MNT_NODEV)
1706 bdev = bd_acquire(inode);
1713 bdev = ERR_PTR(error);
1716 EXPORT_SYMBOL(lookup_bdev);
1719 * open_bdev_exclusive - open a block device by name and set it up for use
1721 * @path: special file representing the block device
1722 * @mode: FMODE_... combination to pass be used
1723 * @holder: owner for exclusion
1725 * Open the blockdevice described by the special file at @path, claim it
1728 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1730 struct block_device *bdev, *whole;
1733 bdev = lookup_bdev(path);
1737 whole = bd_start_claiming(bdev, holder);
1738 if (IS_ERR(whole)) {
1743 error = blkdev_get(bdev, mode);
1745 goto out_abort_claiming;
1748 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1749 goto out_blkdev_put;
1751 bd_finish_claiming(bdev, whole, holder);
1755 blkdev_put(bdev, mode);
1757 bd_abort_claiming(whole, holder);
1758 return ERR_PTR(error);
1761 EXPORT_SYMBOL(open_bdev_exclusive);
1764 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1766 * @bdev: blockdevice to close
1767 * @mode: mode, must match that used to open.
1769 * This is the counterpart to open_bdev_exclusive().
1771 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1774 blkdev_put(bdev, mode);
1777 EXPORT_SYMBOL(close_bdev_exclusive);
1779 int __invalidate_device(struct block_device *bdev)
1781 struct super_block *sb = get_super(bdev);
1786 * no need to lock the super, get_super holds the
1787 * read mutex so the filesystem cannot go away
1788 * under us (->put_super runs with the write lock
1791 shrink_dcache_sb(sb);
1792 res = invalidate_inodes(sb);
1795 invalidate_bdev(bdev);
1798 EXPORT_SYMBOL(__invalidate_device);