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_i_callback(struct rcu_head *head)
414 struct inode *inode = container_of(head, struct inode, i_rcu);
415 struct bdev_inode *bdi = BDEV_I(inode);
417 INIT_LIST_HEAD(&inode->i_dentry);
418 kmem_cache_free(bdev_cachep, bdi);
421 static void bdev_destroy_inode(struct inode *inode)
423 call_rcu(&inode->i_rcu, bdev_i_callback);
426 static void init_once(void *foo)
428 struct bdev_inode *ei = (struct bdev_inode *) foo;
429 struct block_device *bdev = &ei->bdev;
431 memset(bdev, 0, sizeof(*bdev));
432 mutex_init(&bdev->bd_mutex);
433 INIT_LIST_HEAD(&bdev->bd_inodes);
434 INIT_LIST_HEAD(&bdev->bd_list);
436 INIT_LIST_HEAD(&bdev->bd_holder_list);
438 inode_init_once(&ei->vfs_inode);
439 /* Initialize mutex for freeze. */
440 mutex_init(&bdev->bd_fsfreeze_mutex);
443 static inline void __bd_forget(struct inode *inode)
445 list_del_init(&inode->i_devices);
446 inode->i_bdev = NULL;
447 inode->i_mapping = &inode->i_data;
450 static void bdev_evict_inode(struct inode *inode)
452 struct block_device *bdev = &BDEV_I(inode)->bdev;
454 truncate_inode_pages(&inode->i_data, 0);
455 invalidate_inode_buffers(inode); /* is it needed here? */
456 end_writeback(inode);
457 spin_lock(&bdev_lock);
458 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
459 __bd_forget(list_entry(p, struct inode, i_devices));
461 list_del_init(&bdev->bd_list);
462 spin_unlock(&bdev_lock);
465 static const struct super_operations bdev_sops = {
466 .statfs = simple_statfs,
467 .alloc_inode = bdev_alloc_inode,
468 .destroy_inode = bdev_destroy_inode,
469 .drop_inode = generic_delete_inode,
470 .evict_inode = bdev_evict_inode,
473 static struct dentry *bd_mount(struct file_system_type *fs_type,
474 int flags, const char *dev_name, void *data)
476 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
479 static struct file_system_type bd_type = {
482 .kill_sb = kill_anon_super,
485 struct super_block *blockdev_superblock __read_mostly;
487 void __init bdev_cache_init(void)
490 struct vfsmount *bd_mnt;
492 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
493 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
494 SLAB_MEM_SPREAD|SLAB_PANIC),
496 err = register_filesystem(&bd_type);
498 panic("Cannot register bdev pseudo-fs");
499 bd_mnt = kern_mount(&bd_type);
501 panic("Cannot create bdev pseudo-fs");
503 * This vfsmount structure is only used to obtain the
504 * blockdev_superblock, so tell kmemleak not to report it.
506 kmemleak_not_leak(bd_mnt);
507 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
511 * Most likely _very_ bad one - but then it's hardly critical for small
512 * /dev and can be fixed when somebody will need really large one.
513 * Keep in mind that it will be fed through icache hash function too.
515 static inline unsigned long hash(dev_t dev)
517 return MAJOR(dev)+MINOR(dev);
520 static int bdev_test(struct inode *inode, void *data)
522 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
525 static int bdev_set(struct inode *inode, void *data)
527 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
531 static LIST_HEAD(all_bdevs);
533 struct block_device *bdget(dev_t dev)
535 struct block_device *bdev;
538 inode = iget5_locked(blockdev_superblock, hash(dev),
539 bdev_test, bdev_set, &dev);
544 bdev = &BDEV_I(inode)->bdev;
546 if (inode->i_state & I_NEW) {
547 bdev->bd_contains = NULL;
548 bdev->bd_inode = inode;
549 bdev->bd_block_size = (1 << inode->i_blkbits);
550 bdev->bd_part_count = 0;
551 bdev->bd_invalidated = 0;
552 inode->i_mode = S_IFBLK;
554 inode->i_bdev = bdev;
555 inode->i_data.a_ops = &def_blk_aops;
556 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
557 inode->i_data.backing_dev_info = &default_backing_dev_info;
558 spin_lock(&bdev_lock);
559 list_add(&bdev->bd_list, &all_bdevs);
560 spin_unlock(&bdev_lock);
561 unlock_new_inode(inode);
566 EXPORT_SYMBOL(bdget);
569 * bdgrab -- Grab a reference to an already referenced block device
570 * @bdev: Block device to grab a reference to.
572 struct block_device *bdgrab(struct block_device *bdev)
574 ihold(bdev->bd_inode);
578 long nr_blockdev_pages(void)
580 struct block_device *bdev;
582 spin_lock(&bdev_lock);
583 list_for_each_entry(bdev, &all_bdevs, bd_list) {
584 ret += bdev->bd_inode->i_mapping->nrpages;
586 spin_unlock(&bdev_lock);
590 void bdput(struct block_device *bdev)
592 iput(bdev->bd_inode);
595 EXPORT_SYMBOL(bdput);
597 static struct block_device *bd_acquire(struct inode *inode)
599 struct block_device *bdev;
601 spin_lock(&bdev_lock);
602 bdev = inode->i_bdev;
604 ihold(bdev->bd_inode);
605 spin_unlock(&bdev_lock);
608 spin_unlock(&bdev_lock);
610 bdev = bdget(inode->i_rdev);
612 spin_lock(&bdev_lock);
613 if (!inode->i_bdev) {
615 * We take an additional reference to bd_inode,
616 * and it's released in clear_inode() of inode.
617 * So, we can access it via ->i_mapping always
620 ihold(bdev->bd_inode);
621 inode->i_bdev = bdev;
622 inode->i_mapping = bdev->bd_inode->i_mapping;
623 list_add(&inode->i_devices, &bdev->bd_inodes);
625 spin_unlock(&bdev_lock);
630 /* Call when you free inode */
632 void bd_forget(struct inode *inode)
634 struct block_device *bdev = NULL;
636 spin_lock(&bdev_lock);
638 if (!sb_is_blkdev_sb(inode->i_sb))
639 bdev = inode->i_bdev;
642 spin_unlock(&bdev_lock);
645 iput(bdev->bd_inode);
649 * bd_may_claim - test whether a block device can be claimed
650 * @bdev: block device of interest
651 * @whole: whole block device containing @bdev, may equal @bdev
652 * @holder: holder trying to claim @bdev
654 * Test whther @bdev can be claimed by @holder.
657 * spin_lock(&bdev_lock).
660 * %true if @bdev can be claimed, %false otherwise.
662 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
665 if (bdev->bd_holder == holder)
666 return true; /* already a holder */
667 else if (bdev->bd_holder != NULL)
668 return false; /* held by someone else */
669 else if (bdev->bd_contains == bdev)
670 return true; /* is a whole device which isn't held */
672 else if (whole->bd_holder == bd_claim)
673 return true; /* is a partition of a device that is being partitioned */
674 else if (whole->bd_holder != NULL)
675 return false; /* is a partition of a held device */
677 return true; /* is a partition of an un-held device */
681 * bd_prepare_to_claim - prepare to claim a block device
682 * @bdev: block device of interest
683 * @whole: the whole device containing @bdev, may equal @bdev
684 * @holder: holder trying to claim @bdev
686 * Prepare to claim @bdev. This function fails if @bdev is already
687 * claimed by another holder and waits if another claiming is in
688 * progress. This function doesn't actually claim. On successful
689 * return, the caller has ownership of bd_claiming and bd_holder[s].
692 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
696 * 0 if @bdev can be claimed, -EBUSY otherwise.
698 static int bd_prepare_to_claim(struct block_device *bdev,
699 struct block_device *whole, void *holder)
702 /* if someone else claimed, fail */
703 if (!bd_may_claim(bdev, whole, holder))
706 /* if claiming is already in progress, wait for it to finish */
707 if (whole->bd_claiming) {
708 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
711 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
712 spin_unlock(&bdev_lock);
714 finish_wait(wq, &wait);
715 spin_lock(&bdev_lock);
724 * bd_start_claiming - start claiming a block device
725 * @bdev: block device of interest
726 * @holder: holder trying to claim @bdev
728 * @bdev is about to be opened exclusively. Check @bdev can be opened
729 * exclusively and mark that an exclusive open is in progress. Each
730 * successful call to this function must be matched with a call to
731 * either bd_finish_claiming() or bd_abort_claiming() (which do not
734 * This function is used to gain exclusive access to the block device
735 * without actually causing other exclusive open attempts to fail. It
736 * should be used when the open sequence itself requires exclusive
737 * access but may subsequently fail.
743 * Pointer to the block device containing @bdev on success, ERR_PTR()
746 static struct block_device *bd_start_claiming(struct block_device *bdev,
749 struct gendisk *disk;
750 struct block_device *whole;
756 * @bdev might not have been initialized properly yet, look up
757 * and grab the outer block device the hard way.
759 disk = get_gendisk(bdev->bd_dev, &partno);
761 return ERR_PTR(-ENXIO);
763 whole = bdget_disk(disk, 0);
764 module_put(disk->fops->owner);
767 return ERR_PTR(-ENOMEM);
769 /* prepare to claim, if successful, mark claiming in progress */
770 spin_lock(&bdev_lock);
772 err = bd_prepare_to_claim(bdev, whole, holder);
774 whole->bd_claiming = holder;
775 spin_unlock(&bdev_lock);
778 spin_unlock(&bdev_lock);
784 /* releases bdev_lock */
785 static void __bd_abort_claiming(struct block_device *whole, void *holder)
787 BUG_ON(whole->bd_claiming != holder);
788 whole->bd_claiming = NULL;
789 wake_up_bit(&whole->bd_claiming, 0);
791 spin_unlock(&bdev_lock);
796 * bd_abort_claiming - abort claiming a block device
797 * @whole: whole block device returned by bd_start_claiming()
798 * @holder: holder trying to claim @bdev
800 * Abort a claiming block started by bd_start_claiming(). Note that
801 * @whole is not the block device to be claimed but the whole device
802 * returned by bd_start_claiming().
805 * Grabs and releases bdev_lock.
807 static void bd_abort_claiming(struct block_device *whole, void *holder)
809 spin_lock(&bdev_lock);
810 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
813 /* increment holders when we have a legitimate claim. requires bdev_lock */
814 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
817 /* note that for a whole device bd_holders
818 * will be incremented twice, and bd_holder will
819 * be set to bd_claim before being set to holder
822 whole->bd_holder = bd_claim;
824 bdev->bd_holder = holder;
828 * bd_finish_claiming - finish claiming a block device
829 * @bdev: block device of interest (passed to bd_start_claiming())
830 * @whole: whole block device returned by bd_start_claiming()
831 * @holder: holder trying to claim @bdev
833 * Finish a claiming block started by bd_start_claiming().
836 * Grabs and releases bdev_lock.
838 static void bd_finish_claiming(struct block_device *bdev,
839 struct block_device *whole, void *holder)
841 spin_lock(&bdev_lock);
842 BUG_ON(!bd_may_claim(bdev, whole, holder));
843 __bd_claim(bdev, whole, holder);
844 __bd_abort_claiming(whole, holder); /* not actually an abort */
848 * bd_claim - claim a block device
849 * @bdev: block device to claim
850 * @holder: holder trying to claim @bdev
852 * Try to claim @bdev which must have been opened successfully.
858 * 0 if successful, -EBUSY if @bdev is already claimed.
860 int bd_claim(struct block_device *bdev, void *holder)
862 struct block_device *whole = bdev->bd_contains;
867 spin_lock(&bdev_lock);
868 res = bd_prepare_to_claim(bdev, whole, holder);
870 __bd_claim(bdev, whole, holder);
871 spin_unlock(&bdev_lock);
875 EXPORT_SYMBOL(bd_claim);
877 void bd_release(struct block_device *bdev)
879 spin_lock(&bdev_lock);
880 if (!--bdev->bd_contains->bd_holders)
881 bdev->bd_contains->bd_holder = NULL;
882 if (!--bdev->bd_holders)
883 bdev->bd_holder = NULL;
884 spin_unlock(&bdev_lock);
887 EXPORT_SYMBOL(bd_release);
891 * Functions for bd_claim_by_kobject / bd_release_from_kobject
893 * If a kobject is passed to bd_claim_by_kobject()
894 * and the kobject has a parent directory,
895 * following symlinks are created:
896 * o from the kobject to the claimed bdev
897 * o from "holders" directory of the bdev to the parent of the kobject
898 * bd_release_from_kobject() removes these symlinks.
901 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
902 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
903 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
904 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
907 static int add_symlink(struct kobject *from, struct kobject *to)
911 return sysfs_create_link(from, to, kobject_name(to));
914 static void del_symlink(struct kobject *from, struct kobject *to)
918 sysfs_remove_link(from, kobject_name(to));
922 * 'struct bd_holder' contains pointers to kobjects symlinked by
923 * bd_claim_by_kobject.
924 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
927 struct list_head list; /* chain of holders of the bdev */
928 int count; /* references from the holder */
929 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
930 struct kobject *hdev; /* e.g. "/block/dm-0" */
931 struct kobject *hdir; /* e.g. "/block/sda/holders" */
932 struct kobject *sdev; /* e.g. "/block/sda" */
936 * Get references of related kobjects at once.
937 * Returns 1 on success. 0 on failure.
939 * Should call bd_holder_release_dirs() after successful use.
941 static int bd_holder_grab_dirs(struct block_device *bdev,
942 struct bd_holder *bo)
947 bo->sdir = kobject_get(bo->sdir);
951 bo->hdev = kobject_get(bo->sdir->parent);
955 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
959 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
966 kobject_put(bo->sdev);
968 kobject_put(bo->hdev);
970 kobject_put(bo->sdir);
975 /* Put references of related kobjects at once. */
976 static void bd_holder_release_dirs(struct bd_holder *bo)
978 kobject_put(bo->hdir);
979 kobject_put(bo->sdev);
980 kobject_put(bo->hdev);
981 kobject_put(bo->sdir);
984 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
986 struct bd_holder *bo;
988 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
998 static void free_bd_holder(struct bd_holder *bo)
1004 * find_bd_holder - find matching struct bd_holder from the block device
1006 * @bdev: struct block device to be searched
1007 * @bo: target struct bd_holder
1009 * Returns matching entry with @bo in @bdev->bd_holder_list.
1010 * If found, increment the reference count and return the pointer.
1011 * If not found, returns NULL.
1013 static struct bd_holder *find_bd_holder(struct block_device *bdev,
1014 struct bd_holder *bo)
1016 struct bd_holder *tmp;
1018 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
1019 if (tmp->sdir == bo->sdir) {
1028 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1030 * @bdev: block device to be bd_claimed
1031 * @bo: preallocated and initialized by alloc_bd_holder()
1033 * Add @bo to @bdev->bd_holder_list, create symlinks.
1035 * Returns 0 if symlinks are created.
1036 * Returns -ve if something fails.
1038 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1045 if (!bd_holder_grab_dirs(bdev, bo))
1048 err = add_symlink(bo->sdir, bo->sdev);
1052 err = add_symlink(bo->hdir, bo->hdev);
1054 del_symlink(bo->sdir, bo->sdev);
1058 list_add_tail(&bo->list, &bdev->bd_holder_list);
1063 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1065 * @bdev: block device to be bd_claimed
1066 * @kobj: holder's kobject
1068 * If there is matching entry with @kobj in @bdev->bd_holder_list
1069 * and no other bd_claim() from the same kobject,
1070 * remove the struct bd_holder from the list, delete symlinks for it.
1072 * Returns a pointer to the struct bd_holder when it's removed from the list
1073 * and ready to be freed.
1074 * Returns NULL if matching claim isn't found or there is other bd_claim()
1075 * by the same kobject.
1077 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1078 struct kobject *kobj)
1080 struct bd_holder *bo;
1082 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1083 if (bo->sdir == kobj) {
1085 BUG_ON(bo->count < 0);
1087 list_del(&bo->list);
1088 del_symlink(bo->sdir, bo->sdev);
1089 del_symlink(bo->hdir, bo->hdev);
1090 bd_holder_release_dirs(bo);
1101 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1103 * @bdev: block device to be claimed
1104 * @holder: holder's signature
1105 * @kobj: holder's kobject
1107 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1108 * the bdev and the holder's kobject.
1109 * Use bd_release_from_kobject() when relesing the claimed bdev.
1111 * Returns 0 on success. (same as bd_claim())
1112 * Returns errno on failure.
1114 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1115 struct kobject *kobj)
1118 struct bd_holder *bo, *found;
1123 bo = alloc_bd_holder(kobj);
1127 mutex_lock(&bdev->bd_mutex);
1129 err = bd_claim(bdev, holder);
1133 found = find_bd_holder(bdev, bo);
1137 err = add_bd_holder(bdev, bo);
1143 mutex_unlock(&bdev->bd_mutex);
1149 * bd_release_from_kobject - bd_release() with additional kobject signature
1151 * @bdev: block device to be released
1152 * @kobj: holder's kobject
1154 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1156 static void bd_release_from_kobject(struct block_device *bdev,
1157 struct kobject *kobj)
1162 mutex_lock(&bdev->bd_mutex);
1164 free_bd_holder(del_bd_holder(bdev, kobj));
1165 mutex_unlock(&bdev->bd_mutex);
1169 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1171 * @bdev: block device to be claimed
1172 * @holder: holder's signature
1173 * @disk: holder's gendisk
1175 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1177 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1178 struct gendisk *disk)
1180 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1182 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1185 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1187 * @bdev: block device to be claimed
1188 * @disk: holder's gendisk
1190 * Call bd_release_from_kobject() and put @disk->slave_dir.
1192 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1194 bd_release_from_kobject(bdev, disk->slave_dir);
1195 kobject_put(disk->slave_dir);
1197 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1201 * Tries to open block device by device number. Use it ONLY if you
1202 * really do not have anything better - i.e. when you are behind a
1203 * truly sucky interface and all you are given is a device number. _Never_
1204 * to be used for internal purposes. If you ever need it - reconsider
1207 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1209 struct block_device *bdev = bdget(dev);
1212 err = blkdev_get(bdev, mode);
1213 return err ? ERR_PTR(err) : bdev;
1216 EXPORT_SYMBOL(open_by_devnum);
1219 * flush_disk - invalidates all buffer-cache entries on a disk
1221 * @bdev: struct block device to be flushed
1223 * Invalidates all buffer-cache entries on a disk. It should be called
1224 * when a disk has been changed -- either by a media change or online
1227 static void flush_disk(struct block_device *bdev)
1229 if (__invalidate_device(bdev)) {
1230 char name[BDEVNAME_SIZE] = "";
1233 disk_name(bdev->bd_disk, 0, name);
1234 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1235 "resized disk %s\n", name);
1240 if (disk_partitionable(bdev->bd_disk))
1241 bdev->bd_invalidated = 1;
1245 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1246 * @disk: struct gendisk to check
1247 * @bdev: struct bdev to adjust.
1249 * This routine checks to see if the bdev size does not match the disk size
1250 * and adjusts it if it differs.
1252 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1254 loff_t disk_size, bdev_size;
1256 disk_size = (loff_t)get_capacity(disk) << 9;
1257 bdev_size = i_size_read(bdev->bd_inode);
1258 if (disk_size != bdev_size) {
1259 char name[BDEVNAME_SIZE];
1261 disk_name(disk, 0, name);
1263 "%s: detected capacity change from %lld to %lld\n",
1264 name, bdev_size, disk_size);
1265 i_size_write(bdev->bd_inode, disk_size);
1269 EXPORT_SYMBOL(check_disk_size_change);
1272 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1273 * @disk: struct gendisk to be revalidated
1275 * This routine is a wrapper for lower-level driver's revalidate_disk
1276 * call-backs. It is used to do common pre and post operations needed
1277 * for all revalidate_disk operations.
1279 int revalidate_disk(struct gendisk *disk)
1281 struct block_device *bdev;
1284 if (disk->fops->revalidate_disk)
1285 ret = disk->fops->revalidate_disk(disk);
1287 bdev = bdget_disk(disk, 0);
1291 mutex_lock(&bdev->bd_mutex);
1292 check_disk_size_change(disk, bdev);
1293 mutex_unlock(&bdev->bd_mutex);
1297 EXPORT_SYMBOL(revalidate_disk);
1300 * This routine checks whether a removable media has been changed,
1301 * and invalidates all buffer-cache-entries in that case. This
1302 * is a relatively slow routine, so we have to try to minimize using
1303 * it. Thus it is called only upon a 'mount' or 'open'. This
1304 * is the best way of combining speed and utility, I think.
1305 * People changing diskettes in the middle of an operation deserve
1308 int check_disk_change(struct block_device *bdev)
1310 struct gendisk *disk = bdev->bd_disk;
1311 const struct block_device_operations *bdops = disk->fops;
1313 if (!bdops->media_changed)
1315 if (!bdops->media_changed(bdev->bd_disk))
1319 if (bdops->revalidate_disk)
1320 bdops->revalidate_disk(bdev->bd_disk);
1324 EXPORT_SYMBOL(check_disk_change);
1326 void bd_set_size(struct block_device *bdev, loff_t size)
1328 unsigned bsize = bdev_logical_block_size(bdev);
1330 bdev->bd_inode->i_size = size;
1331 while (bsize < PAGE_CACHE_SIZE) {
1336 bdev->bd_block_size = bsize;
1337 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1339 EXPORT_SYMBOL(bd_set_size);
1341 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1346 * mutex_lock(part->bd_mutex)
1347 * mutex_lock_nested(whole->bd_mutex, 1)
1350 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1352 struct gendisk *disk;
1357 if (mode & FMODE_READ)
1359 if (mode & FMODE_WRITE)
1362 * hooks: /n/, see "layering violations".
1365 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1375 disk = get_gendisk(bdev->bd_dev, &partno);
1379 mutex_lock_nested(&bdev->bd_mutex, for_part);
1380 if (!bdev->bd_openers) {
1381 bdev->bd_disk = disk;
1382 bdev->bd_contains = bdev;
1384 struct backing_dev_info *bdi;
1387 bdev->bd_part = disk_get_part(disk, partno);
1391 if (disk->fops->open) {
1392 ret = disk->fops->open(bdev, mode);
1393 if (ret == -ERESTARTSYS) {
1394 /* Lost a race with 'disk' being
1395 * deleted, try again.
1398 disk_put_part(bdev->bd_part);
1399 bdev->bd_part = NULL;
1400 module_put(disk->fops->owner);
1402 bdev->bd_disk = NULL;
1403 mutex_unlock(&bdev->bd_mutex);
1409 if (!bdev->bd_openers) {
1410 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1411 bdi = blk_get_backing_dev_info(bdev);
1413 bdi = &default_backing_dev_info;
1414 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1416 if (bdev->bd_invalidated)
1417 rescan_partitions(disk, bdev);
1419 struct block_device *whole;
1420 whole = bdget_disk(disk, 0);
1425 ret = __blkdev_get(whole, mode, 1);
1428 bdev->bd_contains = whole;
1429 bdev_inode_switch_bdi(bdev->bd_inode,
1430 whole->bd_inode->i_data.backing_dev_info);
1431 bdev->bd_part = disk_get_part(disk, partno);
1432 if (!(disk->flags & GENHD_FL_UP) ||
1433 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1437 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1440 module_put(disk->fops->owner);
1443 if (bdev->bd_contains == bdev) {
1444 if (bdev->bd_disk->fops->open) {
1445 ret = bdev->bd_disk->fops->open(bdev, mode);
1447 goto out_unlock_bdev;
1449 if (bdev->bd_invalidated)
1450 rescan_partitions(bdev->bd_disk, bdev);
1455 bdev->bd_part_count++;
1456 mutex_unlock(&bdev->bd_mutex);
1460 disk_put_part(bdev->bd_part);
1461 bdev->bd_disk = NULL;
1462 bdev->bd_part = NULL;
1463 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1464 if (bdev != bdev->bd_contains)
1465 __blkdev_put(bdev->bd_contains, mode, 1);
1466 bdev->bd_contains = NULL;
1468 mutex_unlock(&bdev->bd_mutex);
1471 module_put(disk->fops->owner);
1478 int blkdev_get(struct block_device *bdev, fmode_t mode)
1480 return __blkdev_get(bdev, mode, 0);
1482 EXPORT_SYMBOL(blkdev_get);
1484 static int blkdev_open(struct inode * inode, struct file * filp)
1486 struct block_device *whole = NULL;
1487 struct block_device *bdev;
1491 * Preserve backwards compatibility and allow large file access
1492 * even if userspace doesn't ask for it explicitly. Some mkfs
1493 * binary needs it. We might want to drop this workaround
1494 * during an unstable branch.
1496 filp->f_flags |= O_LARGEFILE;
1498 if (filp->f_flags & O_NDELAY)
1499 filp->f_mode |= FMODE_NDELAY;
1500 if (filp->f_flags & O_EXCL)
1501 filp->f_mode |= FMODE_EXCL;
1502 if ((filp->f_flags & O_ACCMODE) == 3)
1503 filp->f_mode |= FMODE_WRITE_IOCTL;
1505 bdev = bd_acquire(inode);
1509 if (filp->f_mode & FMODE_EXCL) {
1510 whole = bd_start_claiming(bdev, filp);
1511 if (IS_ERR(whole)) {
1513 return PTR_ERR(whole);
1517 filp->f_mapping = bdev->bd_inode->i_mapping;
1519 res = blkdev_get(bdev, filp->f_mode);
1523 bd_finish_claiming(bdev, whole, filp);
1525 bd_abort_claiming(whole, filp);
1531 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1534 struct gendisk *disk = bdev->bd_disk;
1535 struct block_device *victim = NULL;
1537 mutex_lock_nested(&bdev->bd_mutex, for_part);
1539 bdev->bd_part_count--;
1541 if (!--bdev->bd_openers) {
1542 sync_blockdev(bdev);
1545 if (bdev->bd_contains == bdev) {
1546 if (disk->fops->release)
1547 ret = disk->fops->release(disk, mode);
1549 if (!bdev->bd_openers) {
1550 struct module *owner = disk->fops->owner;
1554 disk_put_part(bdev->bd_part);
1555 bdev->bd_part = NULL;
1556 bdev->bd_disk = NULL;
1557 bdev_inode_switch_bdi(bdev->bd_inode,
1558 &default_backing_dev_info);
1559 if (bdev != bdev->bd_contains)
1560 victim = bdev->bd_contains;
1561 bdev->bd_contains = NULL;
1563 mutex_unlock(&bdev->bd_mutex);
1566 __blkdev_put(victim, mode, 1);
1570 int blkdev_put(struct block_device *bdev, fmode_t mode)
1572 return __blkdev_put(bdev, mode, 0);
1574 EXPORT_SYMBOL(blkdev_put);
1576 static int blkdev_close(struct inode * inode, struct file * filp)
1578 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1579 if (bdev->bd_holder == filp)
1581 return blkdev_put(bdev, filp->f_mode);
1584 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1586 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1587 fmode_t mode = file->f_mode;
1590 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1591 * to updated it before every ioctl.
1593 if (file->f_flags & O_NDELAY)
1594 mode |= FMODE_NDELAY;
1596 mode &= ~FMODE_NDELAY;
1598 return blkdev_ioctl(bdev, mode, cmd, arg);
1602 * Write data to the block device. Only intended for the block device itself
1603 * and the raw driver which basically is a fake block device.
1605 * Does not take i_mutex for the write and thus is not for general purpose
1608 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1609 unsigned long nr_segs, loff_t pos)
1611 struct file *file = iocb->ki_filp;
1614 BUG_ON(iocb->ki_pos != pos);
1616 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1617 if (ret > 0 || ret == -EIOCBQUEUED) {
1620 err = generic_write_sync(file, pos, ret);
1621 if (err < 0 && ret > 0)
1626 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1629 * Try to release a page associated with block device when the system
1630 * is under memory pressure.
1632 static int blkdev_releasepage(struct page *page, gfp_t wait)
1634 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1636 if (super && super->s_op->bdev_try_to_free_page)
1637 return super->s_op->bdev_try_to_free_page(super, page, wait);
1639 return try_to_free_buffers(page);
1642 static const struct address_space_operations def_blk_aops = {
1643 .readpage = blkdev_readpage,
1644 .writepage = blkdev_writepage,
1645 .sync_page = block_sync_page,
1646 .write_begin = blkdev_write_begin,
1647 .write_end = blkdev_write_end,
1648 .writepages = generic_writepages,
1649 .releasepage = blkdev_releasepage,
1650 .direct_IO = blkdev_direct_IO,
1653 const struct file_operations def_blk_fops = {
1654 .open = blkdev_open,
1655 .release = blkdev_close,
1656 .llseek = block_llseek,
1657 .read = do_sync_read,
1658 .write = do_sync_write,
1659 .aio_read = generic_file_aio_read,
1660 .aio_write = blkdev_aio_write,
1661 .mmap = generic_file_mmap,
1662 .fsync = blkdev_fsync,
1663 .unlocked_ioctl = block_ioctl,
1664 #ifdef CONFIG_COMPAT
1665 .compat_ioctl = compat_blkdev_ioctl,
1667 .splice_read = generic_file_splice_read,
1668 .splice_write = generic_file_splice_write,
1671 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1674 mm_segment_t old_fs = get_fs();
1676 res = blkdev_ioctl(bdev, 0, cmd, arg);
1681 EXPORT_SYMBOL(ioctl_by_bdev);
1684 * lookup_bdev - lookup a struct block_device by name
1685 * @pathname: special file representing the block device
1687 * Get a reference to the blockdevice at @pathname in the current
1688 * namespace if possible and return it. Return ERR_PTR(error)
1691 struct block_device *lookup_bdev(const char *pathname)
1693 struct block_device *bdev;
1694 struct inode *inode;
1698 if (!pathname || !*pathname)
1699 return ERR_PTR(-EINVAL);
1701 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1703 return ERR_PTR(error);
1705 inode = path.dentry->d_inode;
1707 if (!S_ISBLK(inode->i_mode))
1710 if (path.mnt->mnt_flags & MNT_NODEV)
1713 bdev = bd_acquire(inode);
1720 bdev = ERR_PTR(error);
1723 EXPORT_SYMBOL(lookup_bdev);
1726 * open_bdev_exclusive - open a block device by name and set it up for use
1728 * @path: special file representing the block device
1729 * @mode: FMODE_... combination to pass be used
1730 * @holder: owner for exclusion
1732 * Open the blockdevice described by the special file at @path, claim it
1735 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1737 struct block_device *bdev, *whole;
1740 bdev = lookup_bdev(path);
1744 whole = bd_start_claiming(bdev, holder);
1745 if (IS_ERR(whole)) {
1750 error = blkdev_get(bdev, mode);
1752 goto out_abort_claiming;
1755 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1756 goto out_blkdev_put;
1758 bd_finish_claiming(bdev, whole, holder);
1762 blkdev_put(bdev, mode);
1764 bd_abort_claiming(whole, holder);
1765 return ERR_PTR(error);
1768 EXPORT_SYMBOL(open_bdev_exclusive);
1771 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1773 * @bdev: blockdevice to close
1774 * @mode: mode, must match that used to open.
1776 * This is the counterpart to open_bdev_exclusive().
1778 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1781 blkdev_put(bdev, mode);
1784 EXPORT_SYMBOL(close_bdev_exclusive);
1786 int __invalidate_device(struct block_device *bdev)
1788 struct super_block *sb = get_super(bdev);
1793 * no need to lock the super, get_super holds the
1794 * read mutex so the filesystem cannot go away
1795 * under us (->put_super runs with the write lock
1798 shrink_dcache_sb(sb);
1799 res = invalidate_inodes(sb);
1802 invalidate_bdev(bdev);
1805 EXPORT_SYMBOL(__invalidate_device);