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_wb_list_lock);
59 spin_lock(&inode->i_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->i_lock);
64 spin_unlock(&inode_wb_list_lock);
67 static sector_t max_block(struct block_device *bdev)
69 sector_t retval = ~((sector_t)0);
70 loff_t sz = i_size_read(bdev->bd_inode);
73 unsigned int size = block_size(bdev);
74 unsigned int sizebits = blksize_bits(size);
75 retval = (sz >> sizebits);
80 /* Kill _all_ buffers and pagecache , dirty or not.. */
81 static void kill_bdev(struct block_device *bdev)
83 if (bdev->bd_inode->i_mapping->nrpages == 0)
86 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
89 int set_blocksize(struct block_device *bdev, int size)
91 /* Size must be a power of two, and between 512 and PAGE_SIZE */
92 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
95 /* Size cannot be smaller than the size supported by the device */
96 if (size < bdev_logical_block_size(bdev))
99 /* Don't change the size if it is same as current */
100 if (bdev->bd_block_size != size) {
102 bdev->bd_block_size = size;
103 bdev->bd_inode->i_blkbits = blksize_bits(size);
109 EXPORT_SYMBOL(set_blocksize);
111 int sb_set_blocksize(struct super_block *sb, int size)
113 if (set_blocksize(sb->s_bdev, size))
115 /* If we get here, we know size is power of two
116 * and it's value is between 512 and PAGE_SIZE */
117 sb->s_blocksize = size;
118 sb->s_blocksize_bits = blksize_bits(size);
119 return sb->s_blocksize;
122 EXPORT_SYMBOL(sb_set_blocksize);
124 int sb_min_blocksize(struct super_block *sb, int size)
126 int minsize = bdev_logical_block_size(sb->s_bdev);
129 return sb_set_blocksize(sb, size);
132 EXPORT_SYMBOL(sb_min_blocksize);
135 blkdev_get_block(struct inode *inode, sector_t iblock,
136 struct buffer_head *bh, int create)
138 if (iblock >= max_block(I_BDEV(inode))) {
143 * for reads, we're just trying to fill a partial page.
144 * return a hole, they will have to call get_block again
145 * before they can fill it, and they will get -EIO at that
150 bh->b_bdev = I_BDEV(inode);
151 bh->b_blocknr = iblock;
152 set_buffer_mapped(bh);
157 blkdev_get_blocks(struct inode *inode, sector_t iblock,
158 struct buffer_head *bh, int create)
160 sector_t end_block = max_block(I_BDEV(inode));
161 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
163 if ((iblock + max_blocks) > end_block) {
164 max_blocks = end_block - iblock;
165 if ((long)max_blocks <= 0) {
167 return -EIO; /* write fully beyond EOF */
169 * It is a read which is fully beyond EOF. We return
170 * a !buffer_mapped buffer
176 bh->b_bdev = I_BDEV(inode);
177 bh->b_blocknr = iblock;
178 bh->b_size = max_blocks << inode->i_blkbits;
180 set_buffer_mapped(bh);
185 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
186 loff_t offset, unsigned long nr_segs)
188 struct file *file = iocb->ki_filp;
189 struct inode *inode = file->f_mapping->host;
191 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
192 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
195 int __sync_blockdev(struct block_device *bdev, int wait)
200 return filemap_flush(bdev->bd_inode->i_mapping);
201 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
205 * Write out and wait upon all the dirty data associated with a block
206 * device via its mapping. Does not take the superblock lock.
208 int sync_blockdev(struct block_device *bdev)
210 return __sync_blockdev(bdev, 1);
212 EXPORT_SYMBOL(sync_blockdev);
215 * Write out and wait upon all dirty data associated with this
216 * device. Filesystem data as well as the underlying block
217 * device. Takes the superblock lock.
219 int fsync_bdev(struct block_device *bdev)
221 struct super_block *sb = get_super(bdev);
223 int res = sync_filesystem(sb);
227 return sync_blockdev(bdev);
229 EXPORT_SYMBOL(fsync_bdev);
232 * freeze_bdev -- lock a filesystem and force it into a consistent state
233 * @bdev: blockdevice to lock
235 * If a superblock is found on this device, we take the s_umount semaphore
236 * on it to make sure nobody unmounts until the snapshot creation is done.
237 * The reference counter (bd_fsfreeze_count) guarantees that only the last
238 * unfreeze process can unfreeze the frozen filesystem actually when multiple
239 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
240 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
243 struct super_block *freeze_bdev(struct block_device *bdev)
245 struct super_block *sb;
248 mutex_lock(&bdev->bd_fsfreeze_mutex);
249 if (++bdev->bd_fsfreeze_count > 1) {
251 * We don't even need to grab a reference - the first call
252 * to freeze_bdev grab an active reference and only the last
253 * thaw_bdev drops it.
255 sb = get_super(bdev);
257 mutex_unlock(&bdev->bd_fsfreeze_mutex);
261 sb = get_active_super(bdev);
264 error = freeze_super(sb);
266 deactivate_super(sb);
267 bdev->bd_fsfreeze_count--;
268 mutex_unlock(&bdev->bd_fsfreeze_mutex);
269 return ERR_PTR(error);
271 deactivate_super(sb);
274 mutex_unlock(&bdev->bd_fsfreeze_mutex);
275 return sb; /* thaw_bdev releases s->s_umount */
277 EXPORT_SYMBOL(freeze_bdev);
280 * thaw_bdev -- unlock filesystem
281 * @bdev: blockdevice to unlock
282 * @sb: associated superblock
284 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
286 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
290 mutex_lock(&bdev->bd_fsfreeze_mutex);
291 if (!bdev->bd_fsfreeze_count)
295 if (--bdev->bd_fsfreeze_count > 0)
301 error = thaw_super(sb);
303 bdev->bd_fsfreeze_count++;
304 mutex_unlock(&bdev->bd_fsfreeze_mutex);
308 mutex_unlock(&bdev->bd_fsfreeze_mutex);
311 EXPORT_SYMBOL(thaw_bdev);
313 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
315 return block_write_full_page(page, blkdev_get_block, wbc);
318 static int blkdev_readpage(struct file * file, struct page * page)
320 return block_read_full_page(page, blkdev_get_block);
323 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
324 loff_t pos, unsigned len, unsigned flags,
325 struct page **pagep, void **fsdata)
327 return block_write_begin(mapping, pos, len, flags, pagep,
331 static int blkdev_write_end(struct file *file, struct address_space *mapping,
332 loff_t pos, unsigned len, unsigned copied,
333 struct page *page, void *fsdata)
336 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
339 page_cache_release(page);
346 * for a block special file file->f_path.dentry->d_inode->i_size is zero
347 * so we compute the size by hand (just as in block_read/write above)
349 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
351 struct inode *bd_inode = file->f_mapping->host;
355 mutex_lock(&bd_inode->i_mutex);
356 size = i_size_read(bd_inode);
363 offset += file->f_pos;
366 if (offset >= 0 && offset <= size) {
367 if (offset != file->f_pos) {
368 file->f_pos = offset;
372 mutex_unlock(&bd_inode->i_mutex);
376 int blkdev_fsync(struct file *filp, int datasync)
378 struct inode *bd_inode = filp->f_mapping->host;
379 struct block_device *bdev = I_BDEV(bd_inode);
383 * There is no need to serialise calls to blkdev_issue_flush with
384 * i_mutex and doing so causes performance issues with concurrent
385 * O_SYNC writers to a block device.
387 mutex_unlock(&bd_inode->i_mutex);
389 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
390 if (error == -EOPNOTSUPP)
393 mutex_lock(&bd_inode->i_mutex);
397 EXPORT_SYMBOL(blkdev_fsync);
403 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
404 static struct kmem_cache * bdev_cachep __read_mostly;
406 static struct inode *bdev_alloc_inode(struct super_block *sb)
408 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
411 return &ei->vfs_inode;
414 static void bdev_i_callback(struct rcu_head *head)
416 struct inode *inode = container_of(head, struct inode, i_rcu);
417 struct bdev_inode *bdi = BDEV_I(inode);
419 INIT_LIST_HEAD(&inode->i_dentry);
420 kmem_cache_free(bdev_cachep, bdi);
423 static void bdev_destroy_inode(struct inode *inode)
425 call_rcu(&inode->i_rcu, bdev_i_callback);
428 static void init_once(void *foo)
430 struct bdev_inode *ei = (struct bdev_inode *) foo;
431 struct block_device *bdev = &ei->bdev;
433 memset(bdev, 0, sizeof(*bdev));
434 mutex_init(&bdev->bd_mutex);
435 INIT_LIST_HEAD(&bdev->bd_inodes);
436 INIT_LIST_HEAD(&bdev->bd_list);
438 INIT_LIST_HEAD(&bdev->bd_holder_disks);
440 inode_init_once(&ei->vfs_inode);
441 /* Initialize mutex for freeze. */
442 mutex_init(&bdev->bd_fsfreeze_mutex);
445 static inline void __bd_forget(struct inode *inode)
447 list_del_init(&inode->i_devices);
448 inode->i_bdev = NULL;
449 inode->i_mapping = &inode->i_data;
452 static void bdev_evict_inode(struct inode *inode)
454 struct block_device *bdev = &BDEV_I(inode)->bdev;
456 truncate_inode_pages(&inode->i_data, 0);
457 invalidate_inode_buffers(inode); /* is it needed here? */
458 end_writeback(inode);
459 spin_lock(&bdev_lock);
460 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
461 __bd_forget(list_entry(p, struct inode, i_devices));
463 list_del_init(&bdev->bd_list);
464 spin_unlock(&bdev_lock);
467 static const struct super_operations bdev_sops = {
468 .statfs = simple_statfs,
469 .alloc_inode = bdev_alloc_inode,
470 .destroy_inode = bdev_destroy_inode,
471 .drop_inode = generic_delete_inode,
472 .evict_inode = bdev_evict_inode,
475 static struct dentry *bd_mount(struct file_system_type *fs_type,
476 int flags, const char *dev_name, void *data)
478 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
481 static struct file_system_type bd_type = {
484 .kill_sb = kill_anon_super,
487 struct super_block *blockdev_superblock __read_mostly;
489 void __init bdev_cache_init(void)
492 struct vfsmount *bd_mnt;
494 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
495 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
496 SLAB_MEM_SPREAD|SLAB_PANIC),
498 err = register_filesystem(&bd_type);
500 panic("Cannot register bdev pseudo-fs");
501 bd_mnt = kern_mount(&bd_type);
503 panic("Cannot create bdev pseudo-fs");
505 * This vfsmount structure is only used to obtain the
506 * blockdev_superblock, so tell kmemleak not to report it.
508 kmemleak_not_leak(bd_mnt);
509 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
513 * Most likely _very_ bad one - but then it's hardly critical for small
514 * /dev and can be fixed when somebody will need really large one.
515 * Keep in mind that it will be fed through icache hash function too.
517 static inline unsigned long hash(dev_t dev)
519 return MAJOR(dev)+MINOR(dev);
522 static int bdev_test(struct inode *inode, void *data)
524 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
527 static int bdev_set(struct inode *inode, void *data)
529 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
533 static LIST_HEAD(all_bdevs);
535 struct block_device *bdget(dev_t dev)
537 struct block_device *bdev;
540 inode = iget5_locked(blockdev_superblock, hash(dev),
541 bdev_test, bdev_set, &dev);
546 bdev = &BDEV_I(inode)->bdev;
548 if (inode->i_state & I_NEW) {
549 bdev->bd_contains = NULL;
550 bdev->bd_inode = inode;
551 bdev->bd_block_size = (1 << inode->i_blkbits);
552 bdev->bd_part_count = 0;
553 bdev->bd_invalidated = 0;
554 inode->i_mode = S_IFBLK;
556 inode->i_bdev = bdev;
557 inode->i_data.a_ops = &def_blk_aops;
558 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
559 inode->i_data.backing_dev_info = &default_backing_dev_info;
560 spin_lock(&bdev_lock);
561 list_add(&bdev->bd_list, &all_bdevs);
562 spin_unlock(&bdev_lock);
563 unlock_new_inode(inode);
568 EXPORT_SYMBOL(bdget);
571 * bdgrab -- Grab a reference to an already referenced block device
572 * @bdev: Block device to grab a reference to.
574 struct block_device *bdgrab(struct block_device *bdev)
576 ihold(bdev->bd_inode);
580 long nr_blockdev_pages(void)
582 struct block_device *bdev;
584 spin_lock(&bdev_lock);
585 list_for_each_entry(bdev, &all_bdevs, bd_list) {
586 ret += bdev->bd_inode->i_mapping->nrpages;
588 spin_unlock(&bdev_lock);
592 void bdput(struct block_device *bdev)
594 iput(bdev->bd_inode);
597 EXPORT_SYMBOL(bdput);
599 static struct block_device *bd_acquire(struct inode *inode)
601 struct block_device *bdev;
603 spin_lock(&bdev_lock);
604 bdev = inode->i_bdev;
606 ihold(bdev->bd_inode);
607 spin_unlock(&bdev_lock);
610 spin_unlock(&bdev_lock);
612 bdev = bdget(inode->i_rdev);
614 spin_lock(&bdev_lock);
615 if (!inode->i_bdev) {
617 * We take an additional reference to bd_inode,
618 * and it's released in clear_inode() of inode.
619 * So, we can access it via ->i_mapping always
622 ihold(bdev->bd_inode);
623 inode->i_bdev = bdev;
624 inode->i_mapping = bdev->bd_inode->i_mapping;
625 list_add(&inode->i_devices, &bdev->bd_inodes);
627 spin_unlock(&bdev_lock);
632 /* Call when you free inode */
634 void bd_forget(struct inode *inode)
636 struct block_device *bdev = NULL;
638 spin_lock(&bdev_lock);
640 if (!sb_is_blkdev_sb(inode->i_sb))
641 bdev = inode->i_bdev;
644 spin_unlock(&bdev_lock);
647 iput(bdev->bd_inode);
651 * bd_may_claim - test whether a block device can be claimed
652 * @bdev: block device of interest
653 * @whole: whole block device containing @bdev, may equal @bdev
654 * @holder: holder trying to claim @bdev
656 * Test whether @bdev can be claimed by @holder.
659 * spin_lock(&bdev_lock).
662 * %true if @bdev can be claimed, %false otherwise.
664 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
667 if (bdev->bd_holder == holder)
668 return true; /* already a holder */
669 else if (bdev->bd_holder != NULL)
670 return false; /* held by someone else */
671 else if (bdev->bd_contains == bdev)
672 return true; /* is a whole device which isn't held */
674 else if (whole->bd_holder == bd_may_claim)
675 return true; /* is a partition of a device that is being partitioned */
676 else if (whole->bd_holder != NULL)
677 return false; /* is a partition of a held device */
679 return true; /* is a partition of an un-held device */
683 * bd_prepare_to_claim - prepare to claim a block device
684 * @bdev: block device of interest
685 * @whole: the whole device containing @bdev, may equal @bdev
686 * @holder: holder trying to claim @bdev
688 * Prepare to claim @bdev. This function fails if @bdev is already
689 * claimed by another holder and waits if another claiming is in
690 * progress. This function doesn't actually claim. On successful
691 * return, the caller has ownership of bd_claiming and bd_holder[s].
694 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
698 * 0 if @bdev can be claimed, -EBUSY otherwise.
700 static int bd_prepare_to_claim(struct block_device *bdev,
701 struct block_device *whole, void *holder)
704 /* if someone else claimed, fail */
705 if (!bd_may_claim(bdev, whole, holder))
708 /* if claiming is already in progress, wait for it to finish */
709 if (whole->bd_claiming) {
710 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
713 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
714 spin_unlock(&bdev_lock);
716 finish_wait(wq, &wait);
717 spin_lock(&bdev_lock);
726 * bd_start_claiming - start claiming a block device
727 * @bdev: block device of interest
728 * @holder: holder trying to claim @bdev
730 * @bdev is about to be opened exclusively. Check @bdev can be opened
731 * exclusively and mark that an exclusive open is in progress. Each
732 * successful call to this function must be matched with a call to
733 * either bd_finish_claiming() or bd_abort_claiming() (which do not
736 * This function is used to gain exclusive access to the block device
737 * without actually causing other exclusive open attempts to fail. It
738 * should be used when the open sequence itself requires exclusive
739 * access but may subsequently fail.
745 * Pointer to the block device containing @bdev on success, ERR_PTR()
748 static struct block_device *bd_start_claiming(struct block_device *bdev,
751 struct gendisk *disk;
752 struct block_device *whole;
758 * @bdev might not have been initialized properly yet, look up
759 * and grab the outer block device the hard way.
761 disk = get_gendisk(bdev->bd_dev, &partno);
763 return ERR_PTR(-ENXIO);
766 * Normally, @bdev should equal what's returned from bdget_disk()
767 * if partno is 0; however, some drivers (floppy) use multiple
768 * bdev's for the same physical device and @bdev may be one of the
769 * aliases. Keep @bdev if partno is 0. This means claimer
770 * tracking is broken for those devices but it has always been that
774 whole = bdget_disk(disk, 0);
776 whole = bdgrab(bdev);
778 module_put(disk->fops->owner);
781 return ERR_PTR(-ENOMEM);
783 /* prepare to claim, if successful, mark claiming in progress */
784 spin_lock(&bdev_lock);
786 err = bd_prepare_to_claim(bdev, whole, holder);
788 whole->bd_claiming = holder;
789 spin_unlock(&bdev_lock);
792 spin_unlock(&bdev_lock);
799 struct bd_holder_disk {
800 struct list_head list;
801 struct gendisk *disk;
805 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
806 struct gendisk *disk)
808 struct bd_holder_disk *holder;
810 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
811 if (holder->disk == disk)
816 static int add_symlink(struct kobject *from, struct kobject *to)
818 return sysfs_create_link(from, to, kobject_name(to));
821 static void del_symlink(struct kobject *from, struct kobject *to)
823 sysfs_remove_link(from, kobject_name(to));
827 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
828 * @bdev: the claimed slave bdev
829 * @disk: the holding disk
831 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
833 * This functions creates the following sysfs symlinks.
835 * - from "slaves" directory of the holder @disk to the claimed @bdev
836 * - from "holders" directory of the @bdev to the holder @disk
838 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
839 * passed to bd_link_disk_holder(), then:
841 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
842 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
844 * The caller must have claimed @bdev before calling this function and
845 * ensure that both @bdev and @disk are valid during the creation and
846 * lifetime of these symlinks.
852 * 0 on success, -errno on failure.
854 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
856 struct bd_holder_disk *holder;
859 mutex_lock(&bdev->bd_mutex);
861 WARN_ON_ONCE(!bdev->bd_holder);
863 /* FIXME: remove the following once add_disk() handles errors */
864 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
867 holder = bd_find_holder_disk(bdev, disk);
873 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
879 INIT_LIST_HEAD(&holder->list);
883 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
887 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
891 * bdev could be deleted beneath us which would implicitly destroy
892 * the holder directory. Hold on to it.
894 kobject_get(bdev->bd_part->holder_dir);
896 list_add(&holder->list, &bdev->bd_holder_disks);
900 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
904 mutex_unlock(&bdev->bd_mutex);
907 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
910 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
911 * @bdev: the calimed slave bdev
912 * @disk: the holding disk
914 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
919 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
921 struct bd_holder_disk *holder;
923 mutex_lock(&bdev->bd_mutex);
925 holder = bd_find_holder_disk(bdev, disk);
927 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
928 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
929 del_symlink(bdev->bd_part->holder_dir,
930 &disk_to_dev(disk)->kobj);
931 kobject_put(bdev->bd_part->holder_dir);
932 list_del_init(&holder->list);
936 mutex_unlock(&bdev->bd_mutex);
938 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
942 * flush_disk - invalidates all buffer-cache entries on a disk
944 * @bdev: struct block device to be flushed
945 * @kill_dirty: flag to guide handling of dirty inodes
947 * Invalidates all buffer-cache entries on a disk. It should be called
948 * when a disk has been changed -- either by a media change or online
951 static void flush_disk(struct block_device *bdev, bool kill_dirty)
953 if (__invalidate_device(bdev, kill_dirty)) {
954 char name[BDEVNAME_SIZE] = "";
957 disk_name(bdev->bd_disk, 0, name);
958 printk(KERN_WARNING "VFS: busy inodes on changed media or "
959 "resized disk %s\n", name);
964 if (disk_partitionable(bdev->bd_disk))
965 bdev->bd_invalidated = 1;
969 * check_disk_size_change - checks for disk size change and adjusts bdev size.
970 * @disk: struct gendisk to check
971 * @bdev: struct bdev to adjust.
973 * This routine checks to see if the bdev size does not match the disk size
974 * and adjusts it if it differs.
976 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
978 loff_t disk_size, bdev_size;
980 disk_size = (loff_t)get_capacity(disk) << 9;
981 bdev_size = i_size_read(bdev->bd_inode);
982 if (disk_size != bdev_size) {
983 char name[BDEVNAME_SIZE];
985 disk_name(disk, 0, name);
987 "%s: detected capacity change from %lld to %lld\n",
988 name, bdev_size, disk_size);
989 i_size_write(bdev->bd_inode, disk_size);
990 flush_disk(bdev, false);
993 EXPORT_SYMBOL(check_disk_size_change);
996 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
997 * @disk: struct gendisk to be revalidated
999 * This routine is a wrapper for lower-level driver's revalidate_disk
1000 * call-backs. It is used to do common pre and post operations needed
1001 * for all revalidate_disk operations.
1003 int revalidate_disk(struct gendisk *disk)
1005 struct block_device *bdev;
1008 if (disk->fops->revalidate_disk)
1009 ret = disk->fops->revalidate_disk(disk);
1011 bdev = bdget_disk(disk, 0);
1015 mutex_lock(&bdev->bd_mutex);
1016 check_disk_size_change(disk, bdev);
1017 mutex_unlock(&bdev->bd_mutex);
1021 EXPORT_SYMBOL(revalidate_disk);
1024 * This routine checks whether a removable media has been changed,
1025 * and invalidates all buffer-cache-entries in that case. This
1026 * is a relatively slow routine, so we have to try to minimize using
1027 * it. Thus it is called only upon a 'mount' or 'open'. This
1028 * is the best way of combining speed and utility, I think.
1029 * People changing diskettes in the middle of an operation deserve
1032 int check_disk_change(struct block_device *bdev)
1034 struct gendisk *disk = bdev->bd_disk;
1035 const struct block_device_operations *bdops = disk->fops;
1036 unsigned int events;
1038 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1039 DISK_EVENT_EJECT_REQUEST);
1040 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1043 flush_disk(bdev, true);
1044 if (bdops->revalidate_disk)
1045 bdops->revalidate_disk(bdev->bd_disk);
1049 EXPORT_SYMBOL(check_disk_change);
1051 void bd_set_size(struct block_device *bdev, loff_t size)
1053 unsigned bsize = bdev_logical_block_size(bdev);
1055 bdev->bd_inode->i_size = size;
1056 while (bsize < PAGE_CACHE_SIZE) {
1061 bdev->bd_block_size = bsize;
1062 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1064 EXPORT_SYMBOL(bd_set_size);
1066 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1071 * mutex_lock(part->bd_mutex)
1072 * mutex_lock_nested(whole->bd_mutex, 1)
1075 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1077 struct gendisk *disk;
1082 if (mode & FMODE_READ)
1084 if (mode & FMODE_WRITE)
1087 * hooks: /n/, see "layering violations".
1090 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1100 disk = get_gendisk(bdev->bd_dev, &partno);
1104 disk_block_events(disk);
1105 mutex_lock_nested(&bdev->bd_mutex, for_part);
1106 if (!bdev->bd_openers) {
1107 bdev->bd_disk = disk;
1108 bdev->bd_contains = bdev;
1110 struct backing_dev_info *bdi;
1113 bdev->bd_part = disk_get_part(disk, partno);
1118 if (disk->fops->open) {
1119 ret = disk->fops->open(bdev, mode);
1120 if (ret == -ERESTARTSYS) {
1121 /* Lost a race with 'disk' being
1122 * deleted, try again.
1125 disk_put_part(bdev->bd_part);
1126 bdev->bd_part = NULL;
1127 bdev->bd_disk = NULL;
1128 mutex_unlock(&bdev->bd_mutex);
1129 disk_unblock_events(disk);
1130 module_put(disk->fops->owner);
1136 if (!ret && !bdev->bd_openers) {
1137 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1138 bdi = blk_get_backing_dev_info(bdev);
1140 bdi = &default_backing_dev_info;
1141 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1145 * If the device is invalidated, rescan partition
1146 * if open succeeded or failed with -ENOMEDIUM.
1147 * The latter is necessary to prevent ghost
1148 * partitions on a removed medium.
1150 if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1151 rescan_partitions(disk, bdev);
1155 struct block_device *whole;
1156 whole = bdget_disk(disk, 0);
1161 ret = __blkdev_get(whole, mode, 1);
1164 bdev->bd_contains = whole;
1165 bdev_inode_switch_bdi(bdev->bd_inode,
1166 whole->bd_inode->i_data.backing_dev_info);
1167 bdev->bd_part = disk_get_part(disk, partno);
1168 if (!(disk->flags & GENHD_FL_UP) ||
1169 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1173 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1176 if (bdev->bd_contains == bdev) {
1178 if (bdev->bd_disk->fops->open)
1179 ret = bdev->bd_disk->fops->open(bdev, mode);
1180 /* the same as first opener case, read comment there */
1181 if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1182 rescan_partitions(bdev->bd_disk, bdev);
1184 goto out_unlock_bdev;
1186 /* only one opener holds refs to the module and disk */
1187 module_put(disk->fops->owner);
1192 bdev->bd_part_count++;
1193 mutex_unlock(&bdev->bd_mutex);
1194 disk_unblock_events(disk);
1198 disk_put_part(bdev->bd_part);
1199 bdev->bd_disk = NULL;
1200 bdev->bd_part = NULL;
1201 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1202 if (bdev != bdev->bd_contains)
1203 __blkdev_put(bdev->bd_contains, mode, 1);
1204 bdev->bd_contains = NULL;
1206 mutex_unlock(&bdev->bd_mutex);
1207 disk_unblock_events(disk);
1208 module_put(disk->fops->owner);
1217 * blkdev_get - open a block device
1218 * @bdev: block_device to open
1219 * @mode: FMODE_* mask
1220 * @holder: exclusive holder identifier
1222 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1223 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1224 * @holder is invalid. Exclusive opens may nest for the same @holder.
1226 * On success, the reference count of @bdev is unchanged. On failure,
1233 * 0 on success, -errno on failure.
1235 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1237 struct block_device *whole = NULL;
1240 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1242 if ((mode & FMODE_EXCL) && holder) {
1243 whole = bd_start_claiming(bdev, holder);
1244 if (IS_ERR(whole)) {
1246 return PTR_ERR(whole);
1250 res = __blkdev_get(bdev, mode, 0);
1253 struct gendisk *disk = whole->bd_disk;
1255 /* finish claiming */
1256 mutex_lock(&bdev->bd_mutex);
1257 spin_lock(&bdev_lock);
1260 BUG_ON(!bd_may_claim(bdev, whole, holder));
1262 * Note that for a whole device bd_holders
1263 * will be incremented twice, and bd_holder
1264 * will be set to bd_may_claim before being
1267 whole->bd_holders++;
1268 whole->bd_holder = bd_may_claim;
1270 bdev->bd_holder = holder;
1273 /* tell others that we're done */
1274 BUG_ON(whole->bd_claiming != holder);
1275 whole->bd_claiming = NULL;
1276 wake_up_bit(&whole->bd_claiming, 0);
1278 spin_unlock(&bdev_lock);
1281 * Block event polling for write claims if requested. Any
1282 * write holder makes the write_holder state stick until
1283 * all are released. This is good enough and tracking
1284 * individual writeable reference is too fragile given the
1285 * way @mode is used in blkdev_get/put().
1287 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1288 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1289 bdev->bd_write_holder = true;
1290 disk_block_events(disk);
1293 mutex_unlock(&bdev->bd_mutex);
1299 EXPORT_SYMBOL(blkdev_get);
1302 * blkdev_get_by_path - open a block device by name
1303 * @path: path to the block device to open
1304 * @mode: FMODE_* mask
1305 * @holder: exclusive holder identifier
1307 * Open the blockdevice described by the device file at @path. @mode
1308 * and @holder are identical to blkdev_get().
1310 * On success, the returned block_device has reference count of one.
1316 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1318 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1321 struct block_device *bdev;
1324 bdev = lookup_bdev(path);
1328 err = blkdev_get(bdev, mode, holder);
1330 return ERR_PTR(err);
1332 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1333 blkdev_put(bdev, mode);
1334 return ERR_PTR(-EACCES);
1339 EXPORT_SYMBOL(blkdev_get_by_path);
1342 * blkdev_get_by_dev - open a block device by device number
1343 * @dev: device number of block device to open
1344 * @mode: FMODE_* mask
1345 * @holder: exclusive holder identifier
1347 * Open the blockdevice described by device number @dev. @mode and
1348 * @holder are identical to blkdev_get().
1350 * Use it ONLY if you really do not have anything better - i.e. when
1351 * you are behind a truly sucky interface and all you are given is a
1352 * device number. _Never_ to be used for internal purposes. If you
1353 * ever need it - reconsider your API.
1355 * On success, the returned block_device has reference count of one.
1361 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1363 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1365 struct block_device *bdev;
1370 return ERR_PTR(-ENOMEM);
1372 err = blkdev_get(bdev, mode, holder);
1374 return ERR_PTR(err);
1378 EXPORT_SYMBOL(blkdev_get_by_dev);
1380 static int blkdev_open(struct inode * inode, struct file * filp)
1382 struct block_device *bdev;
1385 * Preserve backwards compatibility and allow large file access
1386 * even if userspace doesn't ask for it explicitly. Some mkfs
1387 * binary needs it. We might want to drop this workaround
1388 * during an unstable branch.
1390 filp->f_flags |= O_LARGEFILE;
1392 if (filp->f_flags & O_NDELAY)
1393 filp->f_mode |= FMODE_NDELAY;
1394 if (filp->f_flags & O_EXCL)
1395 filp->f_mode |= FMODE_EXCL;
1396 if ((filp->f_flags & O_ACCMODE) == 3)
1397 filp->f_mode |= FMODE_WRITE_IOCTL;
1399 bdev = bd_acquire(inode);
1403 filp->f_mapping = bdev->bd_inode->i_mapping;
1405 return blkdev_get(bdev, filp->f_mode, filp);
1408 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1411 struct gendisk *disk = bdev->bd_disk;
1412 struct block_device *victim = NULL;
1414 mutex_lock_nested(&bdev->bd_mutex, for_part);
1416 bdev->bd_part_count--;
1418 if (!--bdev->bd_openers) {
1419 WARN_ON_ONCE(bdev->bd_holders);
1420 sync_blockdev(bdev);
1423 if (bdev->bd_contains == bdev) {
1424 if (disk->fops->release)
1425 ret = disk->fops->release(disk, mode);
1427 if (!bdev->bd_openers) {
1428 struct module *owner = disk->fops->owner;
1432 disk_put_part(bdev->bd_part);
1433 bdev->bd_part = NULL;
1434 bdev->bd_disk = NULL;
1435 bdev_inode_switch_bdi(bdev->bd_inode,
1436 &default_backing_dev_info);
1437 if (bdev != bdev->bd_contains)
1438 victim = bdev->bd_contains;
1439 bdev->bd_contains = NULL;
1441 mutex_unlock(&bdev->bd_mutex);
1444 __blkdev_put(victim, mode, 1);
1448 int blkdev_put(struct block_device *bdev, fmode_t mode)
1450 if (mode & FMODE_EXCL) {
1454 * Release a claim on the device. The holder fields
1455 * are protected with bdev_lock. bd_mutex is to
1456 * synchronize disk_holder unlinking.
1458 mutex_lock(&bdev->bd_mutex);
1459 spin_lock(&bdev_lock);
1461 WARN_ON_ONCE(--bdev->bd_holders < 0);
1462 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1464 /* bd_contains might point to self, check in a separate step */
1465 if ((bdev_free = !bdev->bd_holders))
1466 bdev->bd_holder = NULL;
1467 if (!bdev->bd_contains->bd_holders)
1468 bdev->bd_contains->bd_holder = NULL;
1470 spin_unlock(&bdev_lock);
1473 * If this was the last claim, remove holder link and
1474 * unblock evpoll if it was a write holder.
1477 if (bdev->bd_write_holder) {
1478 disk_unblock_events(bdev->bd_disk);
1479 disk_check_events(bdev->bd_disk);
1480 bdev->bd_write_holder = false;
1484 mutex_unlock(&bdev->bd_mutex);
1487 return __blkdev_put(bdev, mode, 0);
1489 EXPORT_SYMBOL(blkdev_put);
1491 static int blkdev_close(struct inode * inode, struct file * filp)
1493 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1495 return blkdev_put(bdev, filp->f_mode);
1498 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1500 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1501 fmode_t mode = file->f_mode;
1504 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1505 * to updated it before every ioctl.
1507 if (file->f_flags & O_NDELAY)
1508 mode |= FMODE_NDELAY;
1510 mode &= ~FMODE_NDELAY;
1512 return blkdev_ioctl(bdev, mode, cmd, arg);
1516 * Write data to the block device. Only intended for the block device itself
1517 * and the raw driver which basically is a fake block device.
1519 * Does not take i_mutex for the write and thus is not for general purpose
1522 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1523 unsigned long nr_segs, loff_t pos)
1525 struct file *file = iocb->ki_filp;
1528 BUG_ON(iocb->ki_pos != pos);
1530 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1531 if (ret > 0 || ret == -EIOCBQUEUED) {
1534 err = generic_write_sync(file, pos, ret);
1535 if (err < 0 && ret > 0)
1540 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1543 * Try to release a page associated with block device when the system
1544 * is under memory pressure.
1546 static int blkdev_releasepage(struct page *page, gfp_t wait)
1548 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1550 if (super && super->s_op->bdev_try_to_free_page)
1551 return super->s_op->bdev_try_to_free_page(super, page, wait);
1553 return try_to_free_buffers(page);
1556 static const struct address_space_operations def_blk_aops = {
1557 .readpage = blkdev_readpage,
1558 .writepage = blkdev_writepage,
1559 .write_begin = blkdev_write_begin,
1560 .write_end = blkdev_write_end,
1561 .writepages = generic_writepages,
1562 .releasepage = blkdev_releasepage,
1563 .direct_IO = blkdev_direct_IO,
1566 const struct file_operations def_blk_fops = {
1567 .open = blkdev_open,
1568 .release = blkdev_close,
1569 .llseek = block_llseek,
1570 .read = do_sync_read,
1571 .write = do_sync_write,
1572 .aio_read = generic_file_aio_read,
1573 .aio_write = blkdev_aio_write,
1574 .mmap = generic_file_mmap,
1575 .fsync = blkdev_fsync,
1576 .unlocked_ioctl = block_ioctl,
1577 #ifdef CONFIG_COMPAT
1578 .compat_ioctl = compat_blkdev_ioctl,
1580 .splice_read = generic_file_splice_read,
1581 .splice_write = generic_file_splice_write,
1584 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1587 mm_segment_t old_fs = get_fs();
1589 res = blkdev_ioctl(bdev, 0, cmd, arg);
1594 EXPORT_SYMBOL(ioctl_by_bdev);
1597 * lookup_bdev - lookup a struct block_device by name
1598 * @pathname: special file representing the block device
1600 * Get a reference to the blockdevice at @pathname in the current
1601 * namespace if possible and return it. Return ERR_PTR(error)
1604 struct block_device *lookup_bdev(const char *pathname)
1606 struct block_device *bdev;
1607 struct inode *inode;
1611 if (!pathname || !*pathname)
1612 return ERR_PTR(-EINVAL);
1614 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1616 return ERR_PTR(error);
1618 inode = path.dentry->d_inode;
1620 if (!S_ISBLK(inode->i_mode))
1623 if (path.mnt->mnt_flags & MNT_NODEV)
1626 bdev = bd_acquire(inode);
1633 bdev = ERR_PTR(error);
1636 EXPORT_SYMBOL(lookup_bdev);
1638 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1640 struct super_block *sb = get_super(bdev);
1645 * no need to lock the super, get_super holds the
1646 * read mutex so the filesystem cannot go away
1647 * under us (->put_super runs with the write lock
1650 shrink_dcache_sb(sb);
1651 res = invalidate_inodes(sb, kill_dirty);
1654 invalidate_bdev(bdev);
1657 EXPORT_SYMBOL(__invalidate_device);