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/swap.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/cleancache.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
48 EXPORT_SYMBOL(I_BDEV);
51 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
52 * need to move it onto the dirty list of @dst so that the inode is always on
55 static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
58 struct backing_dev_info *old = inode->i_data.backing_dev_info;
60 if (unlikely(dst == old)) /* deadlock avoidance */
62 bdi_lock_two(&old->wb, &dst->wb);
63 spin_lock(&inode->i_lock);
64 inode->i_data.backing_dev_info = dst;
65 if (inode->i_state & I_DIRTY)
66 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
67 spin_unlock(&inode->i_lock);
68 spin_unlock(&old->wb.list_lock);
69 spin_unlock(&dst->wb.list_lock);
72 static sector_t max_block(struct block_device *bdev)
74 sector_t retval = ~((sector_t)0);
75 loff_t sz = i_size_read(bdev->bd_inode);
78 unsigned int size = block_size(bdev);
79 unsigned int sizebits = blksize_bits(size);
80 retval = (sz >> sizebits);
85 /* Kill _all_ buffers and pagecache , dirty or not.. */
86 void kill_bdev(struct block_device *bdev)
88 struct address_space *mapping = bdev->bd_inode->i_mapping;
90 if (mapping->nrpages == 0)
94 truncate_inode_pages(mapping, 0);
96 EXPORT_SYMBOL(kill_bdev);
98 /* Invalidate clean unused buffers and pagecache. */
99 void invalidate_bdev(struct block_device *bdev)
101 struct address_space *mapping = bdev->bd_inode->i_mapping;
103 if (mapping->nrpages == 0)
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping, 0, -1);
109 /* 99% of the time, we don't need to flush the cleancache on the bdev.
110 * But, for the strange corners, lets be cautious
112 cleancache_invalidate_inode(mapping);
114 EXPORT_SYMBOL(invalidate_bdev);
116 int set_blocksize(struct block_device *bdev, int size)
118 /* Size must be a power of two, and between 512 and PAGE_SIZE */
119 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
122 /* Size cannot be smaller than the size supported by the device */
123 if (size < bdev_logical_block_size(bdev))
126 /* Don't change the size if it is same as current */
127 if (bdev->bd_block_size != size) {
129 bdev->bd_block_size = size;
130 bdev->bd_inode->i_blkbits = blksize_bits(size);
136 EXPORT_SYMBOL(set_blocksize);
138 int sb_set_blocksize(struct super_block *sb, int size)
140 if (set_blocksize(sb->s_bdev, size))
142 /* If we get here, we know size is power of two
143 * and it's value is between 512 and PAGE_SIZE */
144 sb->s_blocksize = size;
145 sb->s_blocksize_bits = blksize_bits(size);
146 return sb->s_blocksize;
149 EXPORT_SYMBOL(sb_set_blocksize);
151 int sb_min_blocksize(struct super_block *sb, int size)
153 int minsize = bdev_logical_block_size(sb->s_bdev);
156 return sb_set_blocksize(sb, size);
159 EXPORT_SYMBOL(sb_min_blocksize);
162 blkdev_get_block(struct inode *inode, sector_t iblock,
163 struct buffer_head *bh, int create)
165 if (iblock >= max_block(I_BDEV(inode))) {
170 * for reads, we're just trying to fill a partial page.
171 * return a hole, they will have to call get_block again
172 * before they can fill it, and they will get -EIO at that
177 bh->b_bdev = I_BDEV(inode);
178 bh->b_blocknr = iblock;
179 set_buffer_mapped(bh);
184 blkdev_get_blocks(struct inode *inode, sector_t iblock,
185 struct buffer_head *bh, int create)
187 sector_t end_block = max_block(I_BDEV(inode));
188 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
190 if ((iblock + max_blocks) > end_block) {
191 max_blocks = end_block - iblock;
192 if ((long)max_blocks <= 0) {
194 return -EIO; /* write fully beyond EOF */
196 * It is a read which is fully beyond EOF. We return
197 * a !buffer_mapped buffer
203 bh->b_bdev = I_BDEV(inode);
204 bh->b_blocknr = iblock;
205 bh->b_size = max_blocks << inode->i_blkbits;
207 set_buffer_mapped(bh);
212 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
213 loff_t offset, unsigned long nr_segs)
215 struct file *file = iocb->ki_filp;
216 struct inode *inode = file->f_mapping->host;
218 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
219 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
222 int __sync_blockdev(struct block_device *bdev, int wait)
227 return filemap_flush(bdev->bd_inode->i_mapping);
228 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
232 * Write out and wait upon all the dirty data associated with a block
233 * device via its mapping. Does not take the superblock lock.
235 int sync_blockdev(struct block_device *bdev)
237 return __sync_blockdev(bdev, 1);
239 EXPORT_SYMBOL(sync_blockdev);
242 * Write out and wait upon all dirty data associated with this
243 * device. Filesystem data as well as the underlying block
244 * device. Takes the superblock lock.
246 int fsync_bdev(struct block_device *bdev)
248 struct super_block *sb = get_super(bdev);
250 int res = sync_filesystem(sb);
254 return sync_blockdev(bdev);
256 EXPORT_SYMBOL(fsync_bdev);
259 * freeze_bdev -- lock a filesystem and force it into a consistent state
260 * @bdev: blockdevice to lock
262 * If a superblock is found on this device, we take the s_umount semaphore
263 * on it to make sure nobody unmounts until the snapshot creation is done.
264 * The reference counter (bd_fsfreeze_count) guarantees that only the last
265 * unfreeze process can unfreeze the frozen filesystem actually when multiple
266 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
267 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
270 struct super_block *freeze_bdev(struct block_device *bdev)
272 struct super_block *sb;
275 mutex_lock(&bdev->bd_fsfreeze_mutex);
276 if (++bdev->bd_fsfreeze_count > 1) {
278 * We don't even need to grab a reference - the first call
279 * to freeze_bdev grab an active reference and only the last
280 * thaw_bdev drops it.
282 sb = get_super(bdev);
284 mutex_unlock(&bdev->bd_fsfreeze_mutex);
288 sb = get_active_super(bdev);
291 error = freeze_super(sb);
293 deactivate_super(sb);
294 bdev->bd_fsfreeze_count--;
295 mutex_unlock(&bdev->bd_fsfreeze_mutex);
296 return ERR_PTR(error);
298 deactivate_super(sb);
301 mutex_unlock(&bdev->bd_fsfreeze_mutex);
302 return sb; /* thaw_bdev releases s->s_umount */
304 EXPORT_SYMBOL(freeze_bdev);
307 * thaw_bdev -- unlock filesystem
308 * @bdev: blockdevice to unlock
309 * @sb: associated superblock
311 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
313 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
317 mutex_lock(&bdev->bd_fsfreeze_mutex);
318 if (!bdev->bd_fsfreeze_count)
322 if (--bdev->bd_fsfreeze_count > 0)
328 error = thaw_super(sb);
330 bdev->bd_fsfreeze_count++;
331 mutex_unlock(&bdev->bd_fsfreeze_mutex);
335 mutex_unlock(&bdev->bd_fsfreeze_mutex);
338 EXPORT_SYMBOL(thaw_bdev);
340 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
342 return block_write_full_page(page, blkdev_get_block, wbc);
345 static int blkdev_readpage(struct file * file, struct page * page)
347 return block_read_full_page(page, blkdev_get_block);
350 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
351 loff_t pos, unsigned len, unsigned flags,
352 struct page **pagep, void **fsdata)
354 return block_write_begin(mapping, pos, len, flags, pagep,
358 static int blkdev_write_end(struct file *file, struct address_space *mapping,
359 loff_t pos, unsigned len, unsigned copied,
360 struct page *page, void *fsdata)
363 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
366 page_cache_release(page);
373 * for a block special file file->f_path.dentry->d_inode->i_size is zero
374 * so we compute the size by hand (just as in block_read/write above)
376 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
378 struct inode *bd_inode = file->f_mapping->host;
382 mutex_lock(&bd_inode->i_mutex);
383 size = i_size_read(bd_inode);
391 offset += file->f_pos;
397 if (offset >= 0 && offset <= size) {
398 if (offset != file->f_pos) {
399 file->f_pos = offset;
404 mutex_unlock(&bd_inode->i_mutex);
408 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
410 struct inode *bd_inode = filp->f_mapping->host;
411 struct block_device *bdev = I_BDEV(bd_inode);
414 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
419 * There is no need to serialise calls to blkdev_issue_flush with
420 * i_mutex and doing so causes performance issues with concurrent
421 * O_SYNC writers to a block device.
423 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
424 if (error == -EOPNOTSUPP)
429 EXPORT_SYMBOL(blkdev_fsync);
435 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
436 static struct kmem_cache * bdev_cachep __read_mostly;
438 static struct inode *bdev_alloc_inode(struct super_block *sb)
440 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
443 return &ei->vfs_inode;
446 static void bdev_i_callback(struct rcu_head *head)
448 struct inode *inode = container_of(head, struct inode, i_rcu);
449 struct bdev_inode *bdi = BDEV_I(inode);
451 kmem_cache_free(bdev_cachep, bdi);
454 static void bdev_destroy_inode(struct inode *inode)
456 call_rcu(&inode->i_rcu, bdev_i_callback);
459 static void init_once(void *foo)
461 struct bdev_inode *ei = (struct bdev_inode *) foo;
462 struct block_device *bdev = &ei->bdev;
464 memset(bdev, 0, sizeof(*bdev));
465 mutex_init(&bdev->bd_mutex);
466 INIT_LIST_HEAD(&bdev->bd_inodes);
467 INIT_LIST_HEAD(&bdev->bd_list);
469 INIT_LIST_HEAD(&bdev->bd_holder_disks);
471 inode_init_once(&ei->vfs_inode);
472 /* Initialize mutex for freeze. */
473 mutex_init(&bdev->bd_fsfreeze_mutex);
476 static inline void __bd_forget(struct inode *inode)
478 list_del_init(&inode->i_devices);
479 inode->i_bdev = NULL;
480 inode->i_mapping = &inode->i_data;
483 static void bdev_evict_inode(struct inode *inode)
485 struct block_device *bdev = &BDEV_I(inode)->bdev;
487 truncate_inode_pages(&inode->i_data, 0);
488 invalidate_inode_buffers(inode); /* is it needed here? */
489 end_writeback(inode);
490 spin_lock(&bdev_lock);
491 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
492 __bd_forget(list_entry(p, struct inode, i_devices));
494 list_del_init(&bdev->bd_list);
495 spin_unlock(&bdev_lock);
498 static const struct super_operations bdev_sops = {
499 .statfs = simple_statfs,
500 .alloc_inode = bdev_alloc_inode,
501 .destroy_inode = bdev_destroy_inode,
502 .drop_inode = generic_delete_inode,
503 .evict_inode = bdev_evict_inode,
506 static struct dentry *bd_mount(struct file_system_type *fs_type,
507 int flags, const char *dev_name, void *data)
509 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
512 static struct file_system_type bd_type = {
515 .kill_sb = kill_anon_super,
518 static struct super_block *blockdev_superblock __read_mostly;
520 void __init bdev_cache_init(void)
523 static struct vfsmount *bd_mnt;
525 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
526 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
527 SLAB_MEM_SPREAD|SLAB_PANIC),
529 err = register_filesystem(&bd_type);
531 panic("Cannot register bdev pseudo-fs");
532 bd_mnt = kern_mount(&bd_type);
534 panic("Cannot create bdev pseudo-fs");
535 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
539 * Most likely _very_ bad one - but then it's hardly critical for small
540 * /dev and can be fixed when somebody will need really large one.
541 * Keep in mind that it will be fed through icache hash function too.
543 static inline unsigned long hash(dev_t dev)
545 return MAJOR(dev)+MINOR(dev);
548 static int bdev_test(struct inode *inode, void *data)
550 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
553 static int bdev_set(struct inode *inode, void *data)
555 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
559 static LIST_HEAD(all_bdevs);
561 struct block_device *bdget(dev_t dev)
563 struct block_device *bdev;
566 inode = iget5_locked(blockdev_superblock, hash(dev),
567 bdev_test, bdev_set, &dev);
572 bdev = &BDEV_I(inode)->bdev;
574 if (inode->i_state & I_NEW) {
575 bdev->bd_contains = NULL;
576 bdev->bd_super = NULL;
577 bdev->bd_inode = inode;
578 bdev->bd_block_size = (1 << inode->i_blkbits);
579 bdev->bd_part_count = 0;
580 bdev->bd_invalidated = 0;
581 inode->i_mode = S_IFBLK;
583 inode->i_bdev = bdev;
584 inode->i_data.a_ops = &def_blk_aops;
585 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
586 inode->i_data.backing_dev_info = &default_backing_dev_info;
587 spin_lock(&bdev_lock);
588 list_add(&bdev->bd_list, &all_bdevs);
589 spin_unlock(&bdev_lock);
590 unlock_new_inode(inode);
595 EXPORT_SYMBOL(bdget);
598 * bdgrab -- Grab a reference to an already referenced block device
599 * @bdev: Block device to grab a reference to.
601 struct block_device *bdgrab(struct block_device *bdev)
603 ihold(bdev->bd_inode);
607 long nr_blockdev_pages(void)
609 struct block_device *bdev;
611 spin_lock(&bdev_lock);
612 list_for_each_entry(bdev, &all_bdevs, bd_list) {
613 ret += bdev->bd_inode->i_mapping->nrpages;
615 spin_unlock(&bdev_lock);
619 void bdput(struct block_device *bdev)
621 iput(bdev->bd_inode);
624 EXPORT_SYMBOL(bdput);
626 static struct block_device *bd_acquire(struct inode *inode)
628 struct block_device *bdev;
630 spin_lock(&bdev_lock);
631 bdev = inode->i_bdev;
633 ihold(bdev->bd_inode);
634 spin_unlock(&bdev_lock);
637 spin_unlock(&bdev_lock);
639 bdev = bdget(inode->i_rdev);
641 spin_lock(&bdev_lock);
642 if (!inode->i_bdev) {
644 * We take an additional reference to bd_inode,
645 * and it's released in clear_inode() of inode.
646 * So, we can access it via ->i_mapping always
649 ihold(bdev->bd_inode);
650 inode->i_bdev = bdev;
651 inode->i_mapping = bdev->bd_inode->i_mapping;
652 list_add(&inode->i_devices, &bdev->bd_inodes);
654 spin_unlock(&bdev_lock);
659 static inline int sb_is_blkdev_sb(struct super_block *sb)
661 return sb == blockdev_superblock;
664 /* Call when you free inode */
666 void bd_forget(struct inode *inode)
668 struct block_device *bdev = NULL;
670 spin_lock(&bdev_lock);
672 if (!sb_is_blkdev_sb(inode->i_sb))
673 bdev = inode->i_bdev;
676 spin_unlock(&bdev_lock);
679 iput(bdev->bd_inode);
683 * bd_may_claim - test whether a block device can be claimed
684 * @bdev: block device of interest
685 * @whole: whole block device containing @bdev, may equal @bdev
686 * @holder: holder trying to claim @bdev
688 * Test whether @bdev can be claimed by @holder.
691 * spin_lock(&bdev_lock).
694 * %true if @bdev can be claimed, %false otherwise.
696 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
699 if (bdev->bd_holder == holder)
700 return true; /* already a holder */
701 else if (bdev->bd_holder != NULL)
702 return false; /* held by someone else */
703 else if (bdev->bd_contains == bdev)
704 return true; /* is a whole device which isn't held */
706 else if (whole->bd_holder == bd_may_claim)
707 return true; /* is a partition of a device that is being partitioned */
708 else if (whole->bd_holder != NULL)
709 return false; /* is a partition of a held device */
711 return true; /* is a partition of an un-held device */
715 * bd_prepare_to_claim - prepare to claim a block device
716 * @bdev: block device of interest
717 * @whole: the whole device containing @bdev, may equal @bdev
718 * @holder: holder trying to claim @bdev
720 * Prepare to claim @bdev. This function fails if @bdev is already
721 * claimed by another holder and waits if another claiming is in
722 * progress. This function doesn't actually claim. On successful
723 * return, the caller has ownership of bd_claiming and bd_holder[s].
726 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
730 * 0 if @bdev can be claimed, -EBUSY otherwise.
732 static int bd_prepare_to_claim(struct block_device *bdev,
733 struct block_device *whole, void *holder)
736 /* if someone else claimed, fail */
737 if (!bd_may_claim(bdev, whole, holder))
740 /* if claiming is already in progress, wait for it to finish */
741 if (whole->bd_claiming) {
742 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
745 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
746 spin_unlock(&bdev_lock);
748 finish_wait(wq, &wait);
749 spin_lock(&bdev_lock);
758 * bd_start_claiming - start claiming a block device
759 * @bdev: block device of interest
760 * @holder: holder trying to claim @bdev
762 * @bdev is about to be opened exclusively. Check @bdev can be opened
763 * exclusively and mark that an exclusive open is in progress. Each
764 * successful call to this function must be matched with a call to
765 * either bd_finish_claiming() or bd_abort_claiming() (which do not
768 * This function is used to gain exclusive access to the block device
769 * without actually causing other exclusive open attempts to fail. It
770 * should be used when the open sequence itself requires exclusive
771 * access but may subsequently fail.
777 * Pointer to the block device containing @bdev on success, ERR_PTR()
780 static struct block_device *bd_start_claiming(struct block_device *bdev,
783 struct gendisk *disk;
784 struct block_device *whole;
790 * @bdev might not have been initialized properly yet, look up
791 * and grab the outer block device the hard way.
793 disk = get_gendisk(bdev->bd_dev, &partno);
795 return ERR_PTR(-ENXIO);
798 * Normally, @bdev should equal what's returned from bdget_disk()
799 * if partno is 0; however, some drivers (floppy) use multiple
800 * bdev's for the same physical device and @bdev may be one of the
801 * aliases. Keep @bdev if partno is 0. This means claimer
802 * tracking is broken for those devices but it has always been that
806 whole = bdget_disk(disk, 0);
808 whole = bdgrab(bdev);
810 module_put(disk->fops->owner);
813 return ERR_PTR(-ENOMEM);
815 /* prepare to claim, if successful, mark claiming in progress */
816 spin_lock(&bdev_lock);
818 err = bd_prepare_to_claim(bdev, whole, holder);
820 whole->bd_claiming = holder;
821 spin_unlock(&bdev_lock);
824 spin_unlock(&bdev_lock);
831 struct bd_holder_disk {
832 struct list_head list;
833 struct gendisk *disk;
837 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
838 struct gendisk *disk)
840 struct bd_holder_disk *holder;
842 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
843 if (holder->disk == disk)
848 static int add_symlink(struct kobject *from, struct kobject *to)
850 return sysfs_create_link(from, to, kobject_name(to));
853 static void del_symlink(struct kobject *from, struct kobject *to)
855 sysfs_remove_link(from, kobject_name(to));
859 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
860 * @bdev: the claimed slave bdev
861 * @disk: the holding disk
863 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
865 * This functions creates the following sysfs symlinks.
867 * - from "slaves" directory of the holder @disk to the claimed @bdev
868 * - from "holders" directory of the @bdev to the holder @disk
870 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
871 * passed to bd_link_disk_holder(), then:
873 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
874 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
876 * The caller must have claimed @bdev before calling this function and
877 * ensure that both @bdev and @disk are valid during the creation and
878 * lifetime of these symlinks.
884 * 0 on success, -errno on failure.
886 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
888 struct bd_holder_disk *holder;
891 mutex_lock(&bdev->bd_mutex);
893 WARN_ON_ONCE(!bdev->bd_holder);
895 /* FIXME: remove the following once add_disk() handles errors */
896 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
899 holder = bd_find_holder_disk(bdev, disk);
905 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
911 INIT_LIST_HEAD(&holder->list);
915 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
919 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
923 * bdev could be deleted beneath us which would implicitly destroy
924 * the holder directory. Hold on to it.
926 kobject_get(bdev->bd_part->holder_dir);
928 list_add(&holder->list, &bdev->bd_holder_disks);
932 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
936 mutex_unlock(&bdev->bd_mutex);
939 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
942 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
943 * @bdev: the calimed slave bdev
944 * @disk: the holding disk
946 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
951 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
953 struct bd_holder_disk *holder;
955 mutex_lock(&bdev->bd_mutex);
957 holder = bd_find_holder_disk(bdev, disk);
959 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
960 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
961 del_symlink(bdev->bd_part->holder_dir,
962 &disk_to_dev(disk)->kobj);
963 kobject_put(bdev->bd_part->holder_dir);
964 list_del_init(&holder->list);
968 mutex_unlock(&bdev->bd_mutex);
970 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
974 * flush_disk - invalidates all buffer-cache entries on a disk
976 * @bdev: struct block device to be flushed
977 * @kill_dirty: flag to guide handling of dirty inodes
979 * Invalidates all buffer-cache entries on a disk. It should be called
980 * when a disk has been changed -- either by a media change or online
983 static void flush_disk(struct block_device *bdev, bool kill_dirty)
985 if (__invalidate_device(bdev, kill_dirty)) {
986 char name[BDEVNAME_SIZE] = "";
989 disk_name(bdev->bd_disk, 0, name);
990 printk(KERN_WARNING "VFS: busy inodes on changed media or "
991 "resized disk %s\n", name);
996 if (disk_part_scan_enabled(bdev->bd_disk))
997 bdev->bd_invalidated = 1;
1001 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1002 * @disk: struct gendisk to check
1003 * @bdev: struct bdev to adjust.
1005 * This routine checks to see if the bdev size does not match the disk size
1006 * and adjusts it if it differs.
1008 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1010 loff_t disk_size, bdev_size;
1012 disk_size = (loff_t)get_capacity(disk) << 9;
1013 bdev_size = i_size_read(bdev->bd_inode);
1014 if (disk_size != bdev_size) {
1015 char name[BDEVNAME_SIZE];
1017 disk_name(disk, 0, name);
1019 "%s: detected capacity change from %lld to %lld\n",
1020 name, bdev_size, disk_size);
1021 i_size_write(bdev->bd_inode, disk_size);
1022 flush_disk(bdev, false);
1025 EXPORT_SYMBOL(check_disk_size_change);
1028 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1029 * @disk: struct gendisk to be revalidated
1031 * This routine is a wrapper for lower-level driver's revalidate_disk
1032 * call-backs. It is used to do common pre and post operations needed
1033 * for all revalidate_disk operations.
1035 int revalidate_disk(struct gendisk *disk)
1037 struct block_device *bdev;
1040 if (disk->fops->revalidate_disk)
1041 ret = disk->fops->revalidate_disk(disk);
1043 bdev = bdget_disk(disk, 0);
1047 mutex_lock(&bdev->bd_mutex);
1048 check_disk_size_change(disk, bdev);
1049 mutex_unlock(&bdev->bd_mutex);
1053 EXPORT_SYMBOL(revalidate_disk);
1056 * This routine checks whether a removable media has been changed,
1057 * and invalidates all buffer-cache-entries in that case. This
1058 * is a relatively slow routine, so we have to try to minimize using
1059 * it. Thus it is called only upon a 'mount' or 'open'. This
1060 * is the best way of combining speed and utility, I think.
1061 * People changing diskettes in the middle of an operation deserve
1064 int check_disk_change(struct block_device *bdev)
1066 struct gendisk *disk = bdev->bd_disk;
1067 const struct block_device_operations *bdops = disk->fops;
1068 unsigned int events;
1070 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1071 DISK_EVENT_EJECT_REQUEST);
1072 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1075 flush_disk(bdev, true);
1076 if (bdops->revalidate_disk)
1077 bdops->revalidate_disk(bdev->bd_disk);
1081 EXPORT_SYMBOL(check_disk_change);
1083 void bd_set_size(struct block_device *bdev, loff_t size)
1085 unsigned bsize = bdev_logical_block_size(bdev);
1087 bdev->bd_inode->i_size = size;
1088 while (bsize < PAGE_CACHE_SIZE) {
1093 bdev->bd_block_size = bsize;
1094 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1096 EXPORT_SYMBOL(bd_set_size);
1098 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1103 * mutex_lock(part->bd_mutex)
1104 * mutex_lock_nested(whole->bd_mutex, 1)
1107 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1109 struct gendisk *disk;
1110 struct module *owner;
1115 if (mode & FMODE_READ)
1117 if (mode & FMODE_WRITE)
1120 * hooks: /n/, see "layering violations".
1123 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1133 disk = get_gendisk(bdev->bd_dev, &partno);
1136 owner = disk->fops->owner;
1138 disk_block_events(disk);
1139 mutex_lock_nested(&bdev->bd_mutex, for_part);
1140 if (!bdev->bd_openers) {
1141 bdev->bd_disk = disk;
1142 bdev->bd_queue = disk->queue;
1143 bdev->bd_contains = bdev;
1145 struct backing_dev_info *bdi;
1148 bdev->bd_part = disk_get_part(disk, partno);
1153 if (disk->fops->open) {
1154 ret = disk->fops->open(bdev, mode);
1155 if (ret == -ERESTARTSYS) {
1156 /* Lost a race with 'disk' being
1157 * deleted, try again.
1160 disk_put_part(bdev->bd_part);
1161 bdev->bd_part = NULL;
1162 bdev->bd_disk = NULL;
1163 bdev->bd_queue = NULL;
1164 mutex_unlock(&bdev->bd_mutex);
1165 disk_unblock_events(disk);
1172 if (!ret && !bdev->bd_openers) {
1173 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1174 bdi = blk_get_backing_dev_info(bdev);
1176 bdi = &default_backing_dev_info;
1177 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1181 * If the device is invalidated, rescan partition
1182 * if open succeeded or failed with -ENOMEDIUM.
1183 * The latter is necessary to prevent ghost
1184 * partitions on a removed medium.
1186 if (bdev->bd_invalidated) {
1188 rescan_partitions(disk, bdev);
1189 else if (ret == -ENOMEDIUM)
1190 invalidate_partitions(disk, bdev);
1195 struct block_device *whole;
1196 whole = bdget_disk(disk, 0);
1201 ret = __blkdev_get(whole, mode, 1);
1204 bdev->bd_contains = whole;
1205 bdev_inode_switch_bdi(bdev->bd_inode,
1206 whole->bd_inode->i_data.backing_dev_info);
1207 bdev->bd_part = disk_get_part(disk, partno);
1208 if (!(disk->flags & GENHD_FL_UP) ||
1209 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1213 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1216 if (bdev->bd_contains == bdev) {
1218 if (bdev->bd_disk->fops->open)
1219 ret = bdev->bd_disk->fops->open(bdev, mode);
1220 /* the same as first opener case, read comment there */
1221 if (bdev->bd_invalidated) {
1223 rescan_partitions(bdev->bd_disk, bdev);
1224 else if (ret == -ENOMEDIUM)
1225 invalidate_partitions(bdev->bd_disk, bdev);
1228 goto out_unlock_bdev;
1230 /* only one opener holds refs to the module and disk */
1236 bdev->bd_part_count++;
1237 mutex_unlock(&bdev->bd_mutex);
1238 disk_unblock_events(disk);
1242 disk_put_part(bdev->bd_part);
1243 bdev->bd_disk = NULL;
1244 bdev->bd_part = NULL;
1245 bdev->bd_queue = NULL;
1246 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1247 if (bdev != bdev->bd_contains)
1248 __blkdev_put(bdev->bd_contains, mode, 1);
1249 bdev->bd_contains = NULL;
1251 mutex_unlock(&bdev->bd_mutex);
1252 disk_unblock_events(disk);
1262 * blkdev_get - open a block device
1263 * @bdev: block_device to open
1264 * @mode: FMODE_* mask
1265 * @holder: exclusive holder identifier
1267 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1268 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1269 * @holder is invalid. Exclusive opens may nest for the same @holder.
1271 * On success, the reference count of @bdev is unchanged. On failure,
1278 * 0 on success, -errno on failure.
1280 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1282 struct block_device *whole = NULL;
1285 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1287 if ((mode & FMODE_EXCL) && holder) {
1288 whole = bd_start_claiming(bdev, holder);
1289 if (IS_ERR(whole)) {
1291 return PTR_ERR(whole);
1295 res = __blkdev_get(bdev, mode, 0);
1298 struct gendisk *disk = whole->bd_disk;
1300 /* finish claiming */
1301 mutex_lock(&bdev->bd_mutex);
1302 spin_lock(&bdev_lock);
1305 BUG_ON(!bd_may_claim(bdev, whole, holder));
1307 * Note that for a whole device bd_holders
1308 * will be incremented twice, and bd_holder
1309 * will be set to bd_may_claim before being
1312 whole->bd_holders++;
1313 whole->bd_holder = bd_may_claim;
1315 bdev->bd_holder = holder;
1318 /* tell others that we're done */
1319 BUG_ON(whole->bd_claiming != holder);
1320 whole->bd_claiming = NULL;
1321 wake_up_bit(&whole->bd_claiming, 0);
1323 spin_unlock(&bdev_lock);
1326 * Block event polling for write claims if requested. Any
1327 * write holder makes the write_holder state stick until
1328 * all are released. This is good enough and tracking
1329 * individual writeable reference is too fragile given the
1330 * way @mode is used in blkdev_get/put().
1332 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1333 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1334 bdev->bd_write_holder = true;
1335 disk_block_events(disk);
1338 mutex_unlock(&bdev->bd_mutex);
1344 EXPORT_SYMBOL(blkdev_get);
1347 * blkdev_get_by_path - open a block device by name
1348 * @path: path to the block device to open
1349 * @mode: FMODE_* mask
1350 * @holder: exclusive holder identifier
1352 * Open the blockdevice described by the device file at @path. @mode
1353 * and @holder are identical to blkdev_get().
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_path(const char *path, fmode_t mode,
1366 struct block_device *bdev;
1369 bdev = lookup_bdev(path);
1373 err = blkdev_get(bdev, mode, holder);
1375 return ERR_PTR(err);
1377 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1378 blkdev_put(bdev, mode);
1379 return ERR_PTR(-EACCES);
1384 EXPORT_SYMBOL(blkdev_get_by_path);
1387 * blkdev_get_by_dev - open a block device by device number
1388 * @dev: device number of block device to open
1389 * @mode: FMODE_* mask
1390 * @holder: exclusive holder identifier
1392 * Open the blockdevice described by device number @dev. @mode and
1393 * @holder are identical to blkdev_get().
1395 * Use it ONLY if you really do not have anything better - i.e. when
1396 * you are behind a truly sucky interface and all you are given is a
1397 * device number. _Never_ to be used for internal purposes. If you
1398 * ever need it - reconsider your API.
1400 * On success, the returned block_device has reference count of one.
1406 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1408 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1410 struct block_device *bdev;
1415 return ERR_PTR(-ENOMEM);
1417 err = blkdev_get(bdev, mode, holder);
1419 return ERR_PTR(err);
1423 EXPORT_SYMBOL(blkdev_get_by_dev);
1425 static int blkdev_open(struct inode * inode, struct file * filp)
1427 struct block_device *bdev;
1430 * Preserve backwards compatibility and allow large file access
1431 * even if userspace doesn't ask for it explicitly. Some mkfs
1432 * binary needs it. We might want to drop this workaround
1433 * during an unstable branch.
1435 filp->f_flags |= O_LARGEFILE;
1437 if (filp->f_flags & O_NDELAY)
1438 filp->f_mode |= FMODE_NDELAY;
1439 if (filp->f_flags & O_EXCL)
1440 filp->f_mode |= FMODE_EXCL;
1441 if ((filp->f_flags & O_ACCMODE) == 3)
1442 filp->f_mode |= FMODE_WRITE_IOCTL;
1444 bdev = bd_acquire(inode);
1448 filp->f_mapping = bdev->bd_inode->i_mapping;
1450 return blkdev_get(bdev, filp->f_mode, filp);
1453 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1456 struct gendisk *disk = bdev->bd_disk;
1457 struct block_device *victim = NULL;
1459 mutex_lock_nested(&bdev->bd_mutex, for_part);
1461 bdev->bd_part_count--;
1463 if (!--bdev->bd_openers) {
1464 WARN_ON_ONCE(bdev->bd_holders);
1465 sync_blockdev(bdev);
1467 /* ->release can cause the old bdi to disappear,
1468 * so must switch it out first
1470 bdev_inode_switch_bdi(bdev->bd_inode,
1471 &default_backing_dev_info);
1473 if (bdev->bd_contains == bdev) {
1474 if (disk->fops->release)
1475 ret = disk->fops->release(disk, mode);
1477 if (!bdev->bd_openers) {
1478 struct module *owner = disk->fops->owner;
1480 disk_put_part(bdev->bd_part);
1481 bdev->bd_part = NULL;
1482 bdev->bd_disk = NULL;
1483 if (bdev != bdev->bd_contains)
1484 victim = bdev->bd_contains;
1485 bdev->bd_contains = NULL;
1490 mutex_unlock(&bdev->bd_mutex);
1493 __blkdev_put(victim, mode, 1);
1497 int blkdev_put(struct block_device *bdev, fmode_t mode)
1499 mutex_lock(&bdev->bd_mutex);
1501 if (mode & FMODE_EXCL) {
1505 * Release a claim on the device. The holder fields
1506 * are protected with bdev_lock. bd_mutex is to
1507 * synchronize disk_holder unlinking.
1509 spin_lock(&bdev_lock);
1511 WARN_ON_ONCE(--bdev->bd_holders < 0);
1512 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1514 /* bd_contains might point to self, check in a separate step */
1515 if ((bdev_free = !bdev->bd_holders))
1516 bdev->bd_holder = NULL;
1517 if (!bdev->bd_contains->bd_holders)
1518 bdev->bd_contains->bd_holder = NULL;
1520 spin_unlock(&bdev_lock);
1523 * If this was the last claim, remove holder link and
1524 * unblock evpoll if it was a write holder.
1526 if (bdev_free && bdev->bd_write_holder) {
1527 disk_unblock_events(bdev->bd_disk);
1528 bdev->bd_write_holder = false;
1533 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1534 * event. This is to ensure detection of media removal commanded
1535 * from userland - e.g. eject(1).
1537 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1539 mutex_unlock(&bdev->bd_mutex);
1541 return __blkdev_put(bdev, mode, 0);
1543 EXPORT_SYMBOL(blkdev_put);
1545 static int blkdev_close(struct inode * inode, struct file * filp)
1547 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1549 return blkdev_put(bdev, filp->f_mode);
1552 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1554 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1555 fmode_t mode = file->f_mode;
1558 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1559 * to updated it before every ioctl.
1561 if (file->f_flags & O_NDELAY)
1562 mode |= FMODE_NDELAY;
1564 mode &= ~FMODE_NDELAY;
1566 return blkdev_ioctl(bdev, mode, cmd, arg);
1570 * Write data to the block device. Only intended for the block device itself
1571 * and the raw driver which basically is a fake block device.
1573 * Does not take i_mutex for the write and thus is not for general purpose
1576 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1577 unsigned long nr_segs, loff_t pos)
1579 struct file *file = iocb->ki_filp;
1582 BUG_ON(iocb->ki_pos != pos);
1584 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1585 if (ret > 0 || ret == -EIOCBQUEUED) {
1588 err = generic_write_sync(file, pos, ret);
1589 if (err < 0 && ret > 0)
1594 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1597 * Try to release a page associated with block device when the system
1598 * is under memory pressure.
1600 static int blkdev_releasepage(struct page *page, gfp_t wait)
1602 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1604 if (super && super->s_op->bdev_try_to_free_page)
1605 return super->s_op->bdev_try_to_free_page(super, page, wait);
1607 return try_to_free_buffers(page);
1610 static const struct address_space_operations def_blk_aops = {
1611 .readpage = blkdev_readpage,
1612 .writepage = blkdev_writepage,
1613 .write_begin = blkdev_write_begin,
1614 .write_end = blkdev_write_end,
1615 .writepages = generic_writepages,
1616 .releasepage = blkdev_releasepage,
1617 .direct_IO = blkdev_direct_IO,
1620 const struct file_operations def_blk_fops = {
1621 .open = blkdev_open,
1622 .release = blkdev_close,
1623 .llseek = block_llseek,
1624 .read = do_sync_read,
1625 .write = do_sync_write,
1626 .aio_read = generic_file_aio_read,
1627 .aio_write = blkdev_aio_write,
1628 .mmap = generic_file_mmap,
1629 .fsync = blkdev_fsync,
1630 .unlocked_ioctl = block_ioctl,
1631 #ifdef CONFIG_COMPAT
1632 .compat_ioctl = compat_blkdev_ioctl,
1634 .splice_read = generic_file_splice_read,
1635 .splice_write = generic_file_splice_write,
1638 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1641 mm_segment_t old_fs = get_fs();
1643 res = blkdev_ioctl(bdev, 0, cmd, arg);
1648 EXPORT_SYMBOL(ioctl_by_bdev);
1651 * lookup_bdev - lookup a struct block_device by name
1652 * @pathname: special file representing the block device
1654 * Get a reference to the blockdevice at @pathname in the current
1655 * namespace if possible and return it. Return ERR_PTR(error)
1658 struct block_device *lookup_bdev(const char *pathname)
1660 struct block_device *bdev;
1661 struct inode *inode;
1665 if (!pathname || !*pathname)
1666 return ERR_PTR(-EINVAL);
1668 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1670 return ERR_PTR(error);
1672 inode = path.dentry->d_inode;
1674 if (!S_ISBLK(inode->i_mode))
1677 if (path.mnt->mnt_flags & MNT_NODEV)
1680 bdev = bd_acquire(inode);
1687 bdev = ERR_PTR(error);
1690 EXPORT_SYMBOL(lookup_bdev);
1692 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1694 struct super_block *sb = get_super(bdev);
1699 * no need to lock the super, get_super holds the
1700 * read mutex so the filesystem cannot go away
1701 * under us (->put_super runs with the write lock
1704 shrink_dcache_sb(sb);
1705 res = invalidate_inodes(sb, kill_dirty);
1708 invalidate_bdev(bdev);
1711 EXPORT_SYMBOL(__invalidate_device);