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/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/pagevec.h>
23 #include <linux/writeback.h>
24 #include <linux/mpage.h>
25 #include <linux/mount.h>
26 #include <linux/uio.h>
27 #include <linux/namei.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30 #include <asm/uaccess.h>
34 struct block_device bdev;
35 struct inode vfs_inode;
38 static const struct address_space_operations def_blk_aops;
40 static inline struct bdev_inode *BDEV_I(struct inode *inode)
42 return container_of(inode, struct bdev_inode, vfs_inode);
45 inline struct block_device *I_BDEV(struct inode *inode)
47 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
52 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
53 * need to move it onto the dirty list of @dst so that the inode is always on
56 static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
59 struct backing_dev_info *old = inode->i_data.backing_dev_info;
61 if (unlikely(dst == old)) /* deadlock avoidance */
63 bdi_lock_two(&old->wb, &dst->wb);
64 spin_lock(&inode->i_lock);
65 inode->i_data.backing_dev_info = dst;
66 if (inode->i_state & I_DIRTY)
67 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
68 spin_unlock(&inode->i_lock);
69 spin_unlock(&old->wb.list_lock);
70 spin_unlock(&dst->wb.list_lock);
73 sector_t blkdev_max_block(struct block_device *bdev)
75 sector_t retval = ~((sector_t)0);
76 loff_t sz = i_size_read(bdev->bd_inode);
79 unsigned int size = block_size(bdev);
80 unsigned int sizebits = blksize_bits(size);
81 retval = (sz >> sizebits);
86 /* Kill _all_ buffers and pagecache , dirty or not.. */
87 void kill_bdev(struct block_device *bdev)
89 struct address_space *mapping = bdev->bd_inode->i_mapping;
91 if (mapping->nrpages == 0)
95 truncate_inode_pages(mapping, 0);
97 EXPORT_SYMBOL(kill_bdev);
99 /* Invalidate clean unused buffers and pagecache. */
100 void invalidate_bdev(struct block_device *bdev)
102 struct address_space *mapping = bdev->bd_inode->i_mapping;
104 if (mapping->nrpages == 0)
107 invalidate_bh_lrus();
108 lru_add_drain_all(); /* make sure all lru add caches are flushed */
109 invalidate_mapping_pages(mapping, 0, -1);
110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
113 cleancache_invalidate_inode(mapping);
115 EXPORT_SYMBOL(invalidate_bdev);
117 int set_blocksize(struct block_device *bdev, int size)
119 /* Size must be a power of two, and between 512 and PAGE_SIZE */
120 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
123 /* Size cannot be smaller than the size supported by the device */
124 if (size < bdev_logical_block_size(bdev))
127 /* Don't change the size if it is same as current */
128 if (bdev->bd_block_size != size) {
130 bdev->bd_block_size = size;
131 bdev->bd_inode->i_blkbits = blksize_bits(size);
137 EXPORT_SYMBOL(set_blocksize);
139 int sb_set_blocksize(struct super_block *sb, int size)
141 if (set_blocksize(sb->s_bdev, size))
143 /* If we get here, we know size is power of two
144 * and it's value is between 512 and PAGE_SIZE */
145 sb->s_blocksize = size;
146 sb->s_blocksize_bits = blksize_bits(size);
147 return sb->s_blocksize;
150 EXPORT_SYMBOL(sb_set_blocksize);
152 int sb_min_blocksize(struct super_block *sb, int size)
154 int minsize = bdev_logical_block_size(sb->s_bdev);
157 return sb_set_blocksize(sb, size);
160 EXPORT_SYMBOL(sb_min_blocksize);
163 blkdev_get_block(struct inode *inode, sector_t iblock,
164 struct buffer_head *bh, int create)
166 if (iblock >= blkdev_max_block(I_BDEV(inode))) {
171 * for reads, we're just trying to fill a partial page.
172 * return a hole, they will have to call get_block again
173 * before they can fill it, and they will get -EIO at that
178 bh->b_bdev = I_BDEV(inode);
179 bh->b_blocknr = iblock;
180 set_buffer_mapped(bh);
185 blkdev_get_blocks(struct inode *inode, sector_t iblock,
186 struct buffer_head *bh, int create)
188 sector_t end_block = blkdev_max_block(I_BDEV(inode));
189 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
191 if ((iblock + max_blocks) > end_block) {
192 max_blocks = end_block - iblock;
193 if ((long)max_blocks <= 0) {
195 return -EIO; /* write fully beyond EOF */
197 * It is a read which is fully beyond EOF. We return
198 * a !buffer_mapped buffer
204 bh->b_bdev = I_BDEV(inode);
205 bh->b_blocknr = iblock;
206 bh->b_size = max_blocks << inode->i_blkbits;
208 set_buffer_mapped(bh);
213 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
214 loff_t offset, unsigned long nr_segs)
216 struct file *file = iocb->ki_filp;
217 struct inode *inode = file->f_mapping->host;
219 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
220 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
223 int __sync_blockdev(struct block_device *bdev, int wait)
228 return filemap_flush(bdev->bd_inode->i_mapping);
229 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
233 * Write out and wait upon all the dirty data associated with a block
234 * device via its mapping. Does not take the superblock lock.
236 int sync_blockdev(struct block_device *bdev)
238 return __sync_blockdev(bdev, 1);
240 EXPORT_SYMBOL(sync_blockdev);
243 * Write out and wait upon all dirty data associated with this
244 * device. Filesystem data as well as the underlying block
245 * device. Takes the superblock lock.
247 int fsync_bdev(struct block_device *bdev)
249 struct super_block *sb = get_super(bdev);
251 int res = sync_filesystem(sb);
255 return sync_blockdev(bdev);
257 EXPORT_SYMBOL(fsync_bdev);
260 * freeze_bdev -- lock a filesystem and force it into a consistent state
261 * @bdev: blockdevice to lock
263 * If a superblock is found on this device, we take the s_umount semaphore
264 * on it to make sure nobody unmounts until the snapshot creation is done.
265 * The reference counter (bd_fsfreeze_count) guarantees that only the last
266 * unfreeze process can unfreeze the frozen filesystem actually when multiple
267 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
268 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
271 struct super_block *freeze_bdev(struct block_device *bdev)
273 struct super_block *sb;
276 mutex_lock(&bdev->bd_fsfreeze_mutex);
277 if (++bdev->bd_fsfreeze_count > 1) {
279 * We don't even need to grab a reference - the first call
280 * to freeze_bdev grab an active reference and only the last
281 * thaw_bdev drops it.
283 sb = get_super(bdev);
285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
289 sb = get_active_super(bdev);
292 error = freeze_super(sb);
294 deactivate_super(sb);
295 bdev->bd_fsfreeze_count--;
296 mutex_unlock(&bdev->bd_fsfreeze_mutex);
297 return ERR_PTR(error);
299 deactivate_super(sb);
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
303 return sb; /* thaw_bdev releases s->s_umount */
305 EXPORT_SYMBOL(freeze_bdev);
308 * thaw_bdev -- unlock filesystem
309 * @bdev: blockdevice to unlock
310 * @sb: associated superblock
312 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
314 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
318 mutex_lock(&bdev->bd_fsfreeze_mutex);
319 if (!bdev->bd_fsfreeze_count)
323 if (--bdev->bd_fsfreeze_count > 0)
329 error = thaw_super(sb);
331 bdev->bd_fsfreeze_count++;
332 mutex_unlock(&bdev->bd_fsfreeze_mutex);
336 mutex_unlock(&bdev->bd_fsfreeze_mutex);
339 EXPORT_SYMBOL(thaw_bdev);
341 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
343 return block_write_full_page(page, blkdev_get_block, wbc);
346 static int blkdev_readpage(struct file * file, struct page * page)
348 return block_read_full_page(page, blkdev_get_block);
351 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
352 loff_t pos, unsigned len, unsigned flags,
353 struct page **pagep, void **fsdata)
355 return block_write_begin(mapping, pos, len, flags, pagep,
359 static int blkdev_write_end(struct file *file, struct address_space *mapping,
360 loff_t pos, unsigned len, unsigned copied,
361 struct page *page, void *fsdata)
364 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
367 page_cache_release(page);
374 * for a block special file file->f_path.dentry->d_inode->i_size is zero
375 * so we compute the size by hand (just as in block_read/write above)
377 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
379 struct inode *bd_inode = file->f_mapping->host;
383 mutex_lock(&bd_inode->i_mutex);
384 size = i_size_read(bd_inode);
392 offset += file->f_pos;
398 if (offset >= 0 && offset <= size) {
399 if (offset != file->f_pos) {
400 file->f_pos = offset;
405 mutex_unlock(&bd_inode->i_mutex);
409 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
411 struct inode *bd_inode = filp->f_mapping->host;
412 struct block_device *bdev = I_BDEV(bd_inode);
415 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
420 * There is no need to serialise calls to blkdev_issue_flush with
421 * i_mutex and doing so causes performance issues with concurrent
422 * O_SYNC writers to a block device.
424 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
425 if (error == -EOPNOTSUPP)
430 EXPORT_SYMBOL(blkdev_fsync);
436 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
437 static struct kmem_cache * bdev_cachep __read_mostly;
439 static struct inode *bdev_alloc_inode(struct super_block *sb)
441 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
444 return &ei->vfs_inode;
447 static void bdev_i_callback(struct rcu_head *head)
449 struct inode *inode = container_of(head, struct inode, i_rcu);
450 struct bdev_inode *bdi = BDEV_I(inode);
452 kmem_cache_free(bdev_cachep, bdi);
455 static void bdev_destroy_inode(struct inode *inode)
457 call_rcu(&inode->i_rcu, bdev_i_callback);
460 static void init_once(void *foo)
462 struct bdev_inode *ei = (struct bdev_inode *) foo;
463 struct block_device *bdev = &ei->bdev;
465 memset(bdev, 0, sizeof(*bdev));
466 mutex_init(&bdev->bd_mutex);
467 INIT_LIST_HEAD(&bdev->bd_inodes);
468 INIT_LIST_HEAD(&bdev->bd_list);
470 INIT_LIST_HEAD(&bdev->bd_holder_disks);
472 inode_init_once(&ei->vfs_inode);
473 /* Initialize mutex for freeze. */
474 mutex_init(&bdev->bd_fsfreeze_mutex);
477 static inline void __bd_forget(struct inode *inode)
479 list_del_init(&inode->i_devices);
480 inode->i_bdev = NULL;
481 inode->i_mapping = &inode->i_data;
484 static void bdev_evict_inode(struct inode *inode)
486 struct block_device *bdev = &BDEV_I(inode)->bdev;
488 truncate_inode_pages(&inode->i_data, 0);
489 invalidate_inode_buffers(inode); /* is it needed here? */
491 spin_lock(&bdev_lock);
492 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
493 __bd_forget(list_entry(p, struct inode, i_devices));
495 list_del_init(&bdev->bd_list);
496 spin_unlock(&bdev_lock);
499 static const struct super_operations bdev_sops = {
500 .statfs = simple_statfs,
501 .alloc_inode = bdev_alloc_inode,
502 .destroy_inode = bdev_destroy_inode,
503 .drop_inode = generic_delete_inode,
504 .evict_inode = bdev_evict_inode,
507 static struct dentry *bd_mount(struct file_system_type *fs_type,
508 int flags, const char *dev_name, void *data)
510 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
513 static struct file_system_type bd_type = {
516 .kill_sb = kill_anon_super,
519 static struct super_block *blockdev_superblock __read_mostly;
521 void __init bdev_cache_init(void)
524 static struct vfsmount *bd_mnt;
526 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
527 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
528 SLAB_MEM_SPREAD|SLAB_PANIC),
530 err = register_filesystem(&bd_type);
532 panic("Cannot register bdev pseudo-fs");
533 bd_mnt = kern_mount(&bd_type);
535 panic("Cannot create bdev pseudo-fs");
536 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
540 * Most likely _very_ bad one - but then it's hardly critical for small
541 * /dev and can be fixed when somebody will need really large one.
542 * Keep in mind that it will be fed through icache hash function too.
544 static inline unsigned long hash(dev_t dev)
546 return MAJOR(dev)+MINOR(dev);
549 static int bdev_test(struct inode *inode, void *data)
551 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
554 static int bdev_set(struct inode *inode, void *data)
556 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
560 static LIST_HEAD(all_bdevs);
562 struct block_device *bdget(dev_t dev)
564 struct block_device *bdev;
567 inode = iget5_locked(blockdev_superblock, hash(dev),
568 bdev_test, bdev_set, &dev);
573 bdev = &BDEV_I(inode)->bdev;
575 if (inode->i_state & I_NEW) {
576 bdev->bd_contains = NULL;
577 bdev->bd_super = NULL;
578 bdev->bd_inode = inode;
579 bdev->bd_block_size = (1 << inode->i_blkbits);
580 bdev->bd_part_count = 0;
581 bdev->bd_invalidated = 0;
582 inode->i_mode = S_IFBLK;
584 inode->i_bdev = bdev;
585 inode->i_data.a_ops = &def_blk_aops;
586 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
587 inode->i_data.backing_dev_info = &default_backing_dev_info;
588 spin_lock(&bdev_lock);
589 list_add(&bdev->bd_list, &all_bdevs);
590 spin_unlock(&bdev_lock);
591 unlock_new_inode(inode);
596 EXPORT_SYMBOL(bdget);
599 * bdgrab -- Grab a reference to an already referenced block device
600 * @bdev: Block device to grab a reference to.
602 struct block_device *bdgrab(struct block_device *bdev)
604 ihold(bdev->bd_inode);
608 long nr_blockdev_pages(void)
610 struct block_device *bdev;
612 spin_lock(&bdev_lock);
613 list_for_each_entry(bdev, &all_bdevs, bd_list) {
614 ret += bdev->bd_inode->i_mapping->nrpages;
616 spin_unlock(&bdev_lock);
620 void bdput(struct block_device *bdev)
622 iput(bdev->bd_inode);
625 EXPORT_SYMBOL(bdput);
627 static struct block_device *bd_acquire(struct inode *inode)
629 struct block_device *bdev;
631 spin_lock(&bdev_lock);
632 bdev = inode->i_bdev;
634 ihold(bdev->bd_inode);
635 spin_unlock(&bdev_lock);
638 spin_unlock(&bdev_lock);
640 bdev = bdget(inode->i_rdev);
642 spin_lock(&bdev_lock);
643 if (!inode->i_bdev) {
645 * We take an additional reference to bd_inode,
646 * and it's released in clear_inode() of inode.
647 * So, we can access it via ->i_mapping always
650 ihold(bdev->bd_inode);
651 inode->i_bdev = bdev;
652 inode->i_mapping = bdev->bd_inode->i_mapping;
653 list_add(&inode->i_devices, &bdev->bd_inodes);
655 spin_unlock(&bdev_lock);
660 static inline int sb_is_blkdev_sb(struct super_block *sb)
662 return sb == blockdev_superblock;
665 /* Call when you free inode */
667 void bd_forget(struct inode *inode)
669 struct block_device *bdev = NULL;
671 spin_lock(&bdev_lock);
673 if (!sb_is_blkdev_sb(inode->i_sb))
674 bdev = inode->i_bdev;
677 spin_unlock(&bdev_lock);
680 iput(bdev->bd_inode);
684 * bd_may_claim - test whether a block device can be claimed
685 * @bdev: block device of interest
686 * @whole: whole block device containing @bdev, may equal @bdev
687 * @holder: holder trying to claim @bdev
689 * Test whether @bdev can be claimed by @holder.
692 * spin_lock(&bdev_lock).
695 * %true if @bdev can be claimed, %false otherwise.
697 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
700 if (bdev->bd_holder == holder)
701 return true; /* already a holder */
702 else if (bdev->bd_holder != NULL)
703 return false; /* held by someone else */
704 else if (bdev->bd_contains == bdev)
705 return true; /* is a whole device which isn't held */
707 else if (whole->bd_holder == bd_may_claim)
708 return true; /* is a partition of a device that is being partitioned */
709 else if (whole->bd_holder != NULL)
710 return false; /* is a partition of a held device */
712 return true; /* is a partition of an un-held device */
716 * bd_prepare_to_claim - prepare to claim a block device
717 * @bdev: block device of interest
718 * @whole: the whole device containing @bdev, may equal @bdev
719 * @holder: holder trying to claim @bdev
721 * Prepare to claim @bdev. This function fails if @bdev is already
722 * claimed by another holder and waits if another claiming is in
723 * progress. This function doesn't actually claim. On successful
724 * return, the caller has ownership of bd_claiming and bd_holder[s].
727 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
731 * 0 if @bdev can be claimed, -EBUSY otherwise.
733 static int bd_prepare_to_claim(struct block_device *bdev,
734 struct block_device *whole, void *holder)
737 /* if someone else claimed, fail */
738 if (!bd_may_claim(bdev, whole, holder))
741 /* if claiming is already in progress, wait for it to finish */
742 if (whole->bd_claiming) {
743 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
746 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
747 spin_unlock(&bdev_lock);
749 finish_wait(wq, &wait);
750 spin_lock(&bdev_lock);
759 * bd_start_claiming - start claiming a block device
760 * @bdev: block device of interest
761 * @holder: holder trying to claim @bdev
763 * @bdev is about to be opened exclusively. Check @bdev can be opened
764 * exclusively and mark that an exclusive open is in progress. Each
765 * successful call to this function must be matched with a call to
766 * either bd_finish_claiming() or bd_abort_claiming() (which do not
769 * This function is used to gain exclusive access to the block device
770 * without actually causing other exclusive open attempts to fail. It
771 * should be used when the open sequence itself requires exclusive
772 * access but may subsequently fail.
778 * Pointer to the block device containing @bdev on success, ERR_PTR()
781 static struct block_device *bd_start_claiming(struct block_device *bdev,
784 struct gendisk *disk;
785 struct block_device *whole;
791 * @bdev might not have been initialized properly yet, look up
792 * and grab the outer block device the hard way.
794 disk = get_gendisk(bdev->bd_dev, &partno);
796 return ERR_PTR(-ENXIO);
799 * Normally, @bdev should equal what's returned from bdget_disk()
800 * if partno is 0; however, some drivers (floppy) use multiple
801 * bdev's for the same physical device and @bdev may be one of the
802 * aliases. Keep @bdev if partno is 0. This means claimer
803 * tracking is broken for those devices but it has always been that
807 whole = bdget_disk(disk, 0);
809 whole = bdgrab(bdev);
811 module_put(disk->fops->owner);
814 return ERR_PTR(-ENOMEM);
816 /* prepare to claim, if successful, mark claiming in progress */
817 spin_lock(&bdev_lock);
819 err = bd_prepare_to_claim(bdev, whole, holder);
821 whole->bd_claiming = holder;
822 spin_unlock(&bdev_lock);
825 spin_unlock(&bdev_lock);
832 struct bd_holder_disk {
833 struct list_head list;
834 struct gendisk *disk;
838 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
839 struct gendisk *disk)
841 struct bd_holder_disk *holder;
843 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
844 if (holder->disk == disk)
849 static int add_symlink(struct kobject *from, struct kobject *to)
851 return sysfs_create_link(from, to, kobject_name(to));
854 static void del_symlink(struct kobject *from, struct kobject *to)
856 sysfs_remove_link(from, kobject_name(to));
860 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
861 * @bdev: the claimed slave bdev
862 * @disk: the holding disk
864 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
866 * This functions creates the following sysfs symlinks.
868 * - from "slaves" directory of the holder @disk to the claimed @bdev
869 * - from "holders" directory of the @bdev to the holder @disk
871 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
872 * passed to bd_link_disk_holder(), then:
874 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
875 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
877 * The caller must have claimed @bdev before calling this function and
878 * ensure that both @bdev and @disk are valid during the creation and
879 * lifetime of these symlinks.
885 * 0 on success, -errno on failure.
887 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
889 struct bd_holder_disk *holder;
892 mutex_lock(&bdev->bd_mutex);
894 WARN_ON_ONCE(!bdev->bd_holder);
896 /* FIXME: remove the following once add_disk() handles errors */
897 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
900 holder = bd_find_holder_disk(bdev, disk);
906 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
912 INIT_LIST_HEAD(&holder->list);
916 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
920 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
924 * bdev could be deleted beneath us which would implicitly destroy
925 * the holder directory. Hold on to it.
927 kobject_get(bdev->bd_part->holder_dir);
929 list_add(&holder->list, &bdev->bd_holder_disks);
933 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
937 mutex_unlock(&bdev->bd_mutex);
940 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
943 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
944 * @bdev: the calimed slave bdev
945 * @disk: the holding disk
947 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
952 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
954 struct bd_holder_disk *holder;
956 mutex_lock(&bdev->bd_mutex);
958 holder = bd_find_holder_disk(bdev, disk);
960 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
961 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
962 del_symlink(bdev->bd_part->holder_dir,
963 &disk_to_dev(disk)->kobj);
964 kobject_put(bdev->bd_part->holder_dir);
965 list_del_init(&holder->list);
969 mutex_unlock(&bdev->bd_mutex);
971 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
975 * flush_disk - invalidates all buffer-cache entries on a disk
977 * @bdev: struct block device to be flushed
978 * @kill_dirty: flag to guide handling of dirty inodes
980 * Invalidates all buffer-cache entries on a disk. It should be called
981 * when a disk has been changed -- either by a media change or online
984 static void flush_disk(struct block_device *bdev, bool kill_dirty)
986 if (__invalidate_device(bdev, kill_dirty)) {
987 char name[BDEVNAME_SIZE] = "";
990 disk_name(bdev->bd_disk, 0, name);
991 printk(KERN_WARNING "VFS: busy inodes on changed media or "
992 "resized disk %s\n", name);
997 if (disk_part_scan_enabled(bdev->bd_disk))
998 bdev->bd_invalidated = 1;
1002 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1003 * @disk: struct gendisk to check
1004 * @bdev: struct bdev to adjust.
1006 * This routine checks to see if the bdev size does not match the disk size
1007 * and adjusts it if it differs.
1009 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1011 loff_t disk_size, bdev_size;
1013 disk_size = (loff_t)get_capacity(disk) << 9;
1014 bdev_size = i_size_read(bdev->bd_inode);
1015 if (disk_size != bdev_size) {
1016 char name[BDEVNAME_SIZE];
1018 disk_name(disk, 0, name);
1020 "%s: detected capacity change from %lld to %lld\n",
1021 name, bdev_size, disk_size);
1022 i_size_write(bdev->bd_inode, disk_size);
1023 flush_disk(bdev, false);
1026 EXPORT_SYMBOL(check_disk_size_change);
1029 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1030 * @disk: struct gendisk to be revalidated
1032 * This routine is a wrapper for lower-level driver's revalidate_disk
1033 * call-backs. It is used to do common pre and post operations needed
1034 * for all revalidate_disk operations.
1036 int revalidate_disk(struct gendisk *disk)
1038 struct block_device *bdev;
1041 if (disk->fops->revalidate_disk)
1042 ret = disk->fops->revalidate_disk(disk);
1044 bdev = bdget_disk(disk, 0);
1048 mutex_lock(&bdev->bd_mutex);
1049 check_disk_size_change(disk, bdev);
1050 mutex_unlock(&bdev->bd_mutex);
1054 EXPORT_SYMBOL(revalidate_disk);
1057 * This routine checks whether a removable media has been changed,
1058 * and invalidates all buffer-cache-entries in that case. This
1059 * is a relatively slow routine, so we have to try to minimize using
1060 * it. Thus it is called only upon a 'mount' or 'open'. This
1061 * is the best way of combining speed and utility, I think.
1062 * People changing diskettes in the middle of an operation deserve
1065 int check_disk_change(struct block_device *bdev)
1067 struct gendisk *disk = bdev->bd_disk;
1068 const struct block_device_operations *bdops = disk->fops;
1069 unsigned int events;
1071 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1072 DISK_EVENT_EJECT_REQUEST);
1073 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1076 flush_disk(bdev, true);
1077 if (bdops->revalidate_disk)
1078 bdops->revalidate_disk(bdev->bd_disk);
1082 EXPORT_SYMBOL(check_disk_change);
1084 void bd_set_size(struct block_device *bdev, loff_t size)
1086 unsigned bsize = bdev_logical_block_size(bdev);
1088 bdev->bd_inode->i_size = size;
1089 while (bsize < PAGE_CACHE_SIZE) {
1094 bdev->bd_block_size = bsize;
1095 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1097 EXPORT_SYMBOL(bd_set_size);
1099 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1104 * mutex_lock(part->bd_mutex)
1105 * mutex_lock_nested(whole->bd_mutex, 1)
1108 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1110 struct gendisk *disk;
1111 struct module *owner;
1116 if (mode & FMODE_READ)
1118 if (mode & FMODE_WRITE)
1121 * hooks: /n/, see "layering violations".
1124 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1134 disk = get_gendisk(bdev->bd_dev, &partno);
1137 owner = disk->fops->owner;
1139 disk_block_events(disk);
1140 mutex_lock_nested(&bdev->bd_mutex, for_part);
1141 if (!bdev->bd_openers) {
1142 bdev->bd_disk = disk;
1143 bdev->bd_queue = disk->queue;
1144 bdev->bd_contains = bdev;
1146 struct backing_dev_info *bdi;
1149 bdev->bd_part = disk_get_part(disk, partno);
1154 if (disk->fops->open) {
1155 ret = disk->fops->open(bdev, mode);
1156 if (ret == -ERESTARTSYS) {
1157 /* Lost a race with 'disk' being
1158 * deleted, try again.
1161 disk_put_part(bdev->bd_part);
1162 bdev->bd_part = NULL;
1163 bdev->bd_disk = NULL;
1164 bdev->bd_queue = NULL;
1165 mutex_unlock(&bdev->bd_mutex);
1166 disk_unblock_events(disk);
1173 if (!ret && !bdev->bd_openers) {
1174 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1175 bdi = blk_get_backing_dev_info(bdev);
1177 bdi = &default_backing_dev_info;
1178 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1182 * If the device is invalidated, rescan partition
1183 * if open succeeded or failed with -ENOMEDIUM.
1184 * The latter is necessary to prevent ghost
1185 * partitions on a removed medium.
1187 if (bdev->bd_invalidated) {
1189 rescan_partitions(disk, bdev);
1190 else if (ret == -ENOMEDIUM)
1191 invalidate_partitions(disk, bdev);
1196 struct block_device *whole;
1197 whole = bdget_disk(disk, 0);
1202 ret = __blkdev_get(whole, mode, 1);
1205 bdev->bd_contains = whole;
1206 bdev_inode_switch_bdi(bdev->bd_inode,
1207 whole->bd_inode->i_data.backing_dev_info);
1208 bdev->bd_part = disk_get_part(disk, partno);
1209 if (!(disk->flags & GENHD_FL_UP) ||
1210 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1214 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1217 if (bdev->bd_contains == bdev) {
1219 if (bdev->bd_disk->fops->open)
1220 ret = bdev->bd_disk->fops->open(bdev, mode);
1221 /* the same as first opener case, read comment there */
1222 if (bdev->bd_invalidated) {
1224 rescan_partitions(bdev->bd_disk, bdev);
1225 else if (ret == -ENOMEDIUM)
1226 invalidate_partitions(bdev->bd_disk, bdev);
1229 goto out_unlock_bdev;
1231 /* only one opener holds refs to the module and disk */
1237 bdev->bd_part_count++;
1238 mutex_unlock(&bdev->bd_mutex);
1239 disk_unblock_events(disk);
1243 disk_put_part(bdev->bd_part);
1244 bdev->bd_disk = NULL;
1245 bdev->bd_part = NULL;
1246 bdev->bd_queue = NULL;
1247 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1248 if (bdev != bdev->bd_contains)
1249 __blkdev_put(bdev->bd_contains, mode, 1);
1250 bdev->bd_contains = NULL;
1252 mutex_unlock(&bdev->bd_mutex);
1253 disk_unblock_events(disk);
1263 * blkdev_get - open a block device
1264 * @bdev: block_device to open
1265 * @mode: FMODE_* mask
1266 * @holder: exclusive holder identifier
1268 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1269 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1270 * @holder is invalid. Exclusive opens may nest for the same @holder.
1272 * On success, the reference count of @bdev is unchanged. On failure,
1279 * 0 on success, -errno on failure.
1281 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1283 struct block_device *whole = NULL;
1286 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1288 if ((mode & FMODE_EXCL) && holder) {
1289 whole = bd_start_claiming(bdev, holder);
1290 if (IS_ERR(whole)) {
1292 return PTR_ERR(whole);
1296 res = __blkdev_get(bdev, mode, 0);
1299 struct gendisk *disk = whole->bd_disk;
1301 /* finish claiming */
1302 mutex_lock(&bdev->bd_mutex);
1303 spin_lock(&bdev_lock);
1306 BUG_ON(!bd_may_claim(bdev, whole, holder));
1308 * Note that for a whole device bd_holders
1309 * will be incremented twice, and bd_holder
1310 * will be set to bd_may_claim before being
1313 whole->bd_holders++;
1314 whole->bd_holder = bd_may_claim;
1316 bdev->bd_holder = holder;
1319 /* tell others that we're done */
1320 BUG_ON(whole->bd_claiming != holder);
1321 whole->bd_claiming = NULL;
1322 wake_up_bit(&whole->bd_claiming, 0);
1324 spin_unlock(&bdev_lock);
1327 * Block event polling for write claims if requested. Any
1328 * write holder makes the write_holder state stick until
1329 * all are released. This is good enough and tracking
1330 * individual writeable reference is too fragile given the
1331 * way @mode is used in blkdev_get/put().
1333 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1334 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1335 bdev->bd_write_holder = true;
1336 disk_block_events(disk);
1339 mutex_unlock(&bdev->bd_mutex);
1345 EXPORT_SYMBOL(blkdev_get);
1348 * blkdev_get_by_path - open a block device by name
1349 * @path: path to the block device to open
1350 * @mode: FMODE_* mask
1351 * @holder: exclusive holder identifier
1353 * Open the blockdevice described by the device file at @path. @mode
1354 * and @holder are identical to blkdev_get().
1356 * On success, the returned block_device has reference count of one.
1362 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1364 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1367 struct block_device *bdev;
1370 bdev = lookup_bdev(path);
1374 err = blkdev_get(bdev, mode, holder);
1376 return ERR_PTR(err);
1378 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1379 blkdev_put(bdev, mode);
1380 return ERR_PTR(-EACCES);
1385 EXPORT_SYMBOL(blkdev_get_by_path);
1388 * blkdev_get_by_dev - open a block device by device number
1389 * @dev: device number of block device to open
1390 * @mode: FMODE_* mask
1391 * @holder: exclusive holder identifier
1393 * Open the blockdevice described by device number @dev. @mode and
1394 * @holder are identical to blkdev_get().
1396 * Use it ONLY if you really do not have anything better - i.e. when
1397 * you are behind a truly sucky interface and all you are given is a
1398 * device number. _Never_ to be used for internal purposes. If you
1399 * ever need it - reconsider your API.
1401 * On success, the returned block_device has reference count of one.
1407 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1409 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1411 struct block_device *bdev;
1416 return ERR_PTR(-ENOMEM);
1418 err = blkdev_get(bdev, mode, holder);
1420 return ERR_PTR(err);
1424 EXPORT_SYMBOL(blkdev_get_by_dev);
1426 static int blkdev_open(struct inode * inode, struct file * filp)
1428 struct block_device *bdev;
1431 * Preserve backwards compatibility and allow large file access
1432 * even if userspace doesn't ask for it explicitly. Some mkfs
1433 * binary needs it. We might want to drop this workaround
1434 * during an unstable branch.
1436 filp->f_flags |= O_LARGEFILE;
1438 if (filp->f_flags & O_NDELAY)
1439 filp->f_mode |= FMODE_NDELAY;
1440 if (filp->f_flags & O_EXCL)
1441 filp->f_mode |= FMODE_EXCL;
1442 if ((filp->f_flags & O_ACCMODE) == 3)
1443 filp->f_mode |= FMODE_WRITE_IOCTL;
1445 bdev = bd_acquire(inode);
1449 filp->f_mapping = bdev->bd_inode->i_mapping;
1451 return blkdev_get(bdev, filp->f_mode, filp);
1454 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1457 struct gendisk *disk = bdev->bd_disk;
1458 struct block_device *victim = NULL;
1460 mutex_lock_nested(&bdev->bd_mutex, for_part);
1462 bdev->bd_part_count--;
1464 if (!--bdev->bd_openers) {
1465 WARN_ON_ONCE(bdev->bd_holders);
1466 sync_blockdev(bdev);
1468 /* ->release can cause the old bdi to disappear,
1469 * so must switch it out first
1471 bdev_inode_switch_bdi(bdev->bd_inode,
1472 &default_backing_dev_info);
1474 if (bdev->bd_contains == bdev) {
1475 if (disk->fops->release)
1476 ret = disk->fops->release(disk, mode);
1478 if (!bdev->bd_openers) {
1479 struct module *owner = disk->fops->owner;
1481 disk_put_part(bdev->bd_part);
1482 bdev->bd_part = NULL;
1483 bdev->bd_disk = NULL;
1484 if (bdev != bdev->bd_contains)
1485 victim = bdev->bd_contains;
1486 bdev->bd_contains = NULL;
1491 mutex_unlock(&bdev->bd_mutex);
1494 __blkdev_put(victim, mode, 1);
1498 int blkdev_put(struct block_device *bdev, fmode_t mode)
1500 mutex_lock(&bdev->bd_mutex);
1502 if (mode & FMODE_EXCL) {
1506 * Release a claim on the device. The holder fields
1507 * are protected with bdev_lock. bd_mutex is to
1508 * synchronize disk_holder unlinking.
1510 spin_lock(&bdev_lock);
1512 WARN_ON_ONCE(--bdev->bd_holders < 0);
1513 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1515 /* bd_contains might point to self, check in a separate step */
1516 if ((bdev_free = !bdev->bd_holders))
1517 bdev->bd_holder = NULL;
1518 if (!bdev->bd_contains->bd_holders)
1519 bdev->bd_contains->bd_holder = NULL;
1521 spin_unlock(&bdev_lock);
1524 * If this was the last claim, remove holder link and
1525 * unblock evpoll if it was a write holder.
1527 if (bdev_free && bdev->bd_write_holder) {
1528 disk_unblock_events(bdev->bd_disk);
1529 bdev->bd_write_holder = false;
1534 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1535 * event. This is to ensure detection of media removal commanded
1536 * from userland - e.g. eject(1).
1538 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1540 mutex_unlock(&bdev->bd_mutex);
1542 return __blkdev_put(bdev, mode, 0);
1544 EXPORT_SYMBOL(blkdev_put);
1546 static int blkdev_close(struct inode * inode, struct file * filp)
1548 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1550 return blkdev_put(bdev, filp->f_mode);
1553 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1555 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1556 fmode_t mode = file->f_mode;
1559 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1560 * to updated it before every ioctl.
1562 if (file->f_flags & O_NDELAY)
1563 mode |= FMODE_NDELAY;
1565 mode &= ~FMODE_NDELAY;
1567 return blkdev_ioctl(bdev, mode, cmd, arg);
1571 * Write data to the block device. Only intended for the block device itself
1572 * and the raw driver which basically is a fake block device.
1574 * Does not take i_mutex for the write and thus is not for general purpose
1577 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1578 unsigned long nr_segs, loff_t pos)
1580 struct file *file = iocb->ki_filp;
1583 BUG_ON(iocb->ki_pos != pos);
1585 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1586 if (ret > 0 || ret == -EIOCBQUEUED) {
1589 err = generic_write_sync(file, pos, ret);
1590 if (err < 0 && ret > 0)
1595 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1598 * Try to release a page associated with block device when the system
1599 * is under memory pressure.
1601 static int blkdev_releasepage(struct page *page, gfp_t wait)
1603 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1605 if (super && super->s_op->bdev_try_to_free_page)
1606 return super->s_op->bdev_try_to_free_page(super, page, wait);
1608 return try_to_free_buffers(page);
1611 static const struct address_space_operations def_blk_aops = {
1612 .readpage = blkdev_readpage,
1613 .writepage = blkdev_writepage,
1614 .write_begin = blkdev_write_begin,
1615 .write_end = blkdev_write_end,
1616 .writepages = generic_writepages,
1617 .releasepage = blkdev_releasepage,
1618 .direct_IO = blkdev_direct_IO,
1621 const struct file_operations def_blk_fops = {
1622 .open = blkdev_open,
1623 .release = blkdev_close,
1624 .llseek = block_llseek,
1625 .read = do_sync_read,
1626 .write = do_sync_write,
1627 .aio_read = generic_file_aio_read,
1628 .aio_write = blkdev_aio_write,
1629 .mmap = generic_file_mmap,
1630 .fsync = blkdev_fsync,
1631 .unlocked_ioctl = block_ioctl,
1632 #ifdef CONFIG_COMPAT
1633 .compat_ioctl = compat_blkdev_ioctl,
1635 .splice_read = generic_file_splice_read,
1636 .splice_write = generic_file_splice_write,
1639 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1642 mm_segment_t old_fs = get_fs();
1644 res = blkdev_ioctl(bdev, 0, cmd, arg);
1649 EXPORT_SYMBOL(ioctl_by_bdev);
1652 * lookup_bdev - lookup a struct block_device by name
1653 * @pathname: special file representing the block device
1655 * Get a reference to the blockdevice at @pathname in the current
1656 * namespace if possible and return it. Return ERR_PTR(error)
1659 struct block_device *lookup_bdev(const char *pathname)
1661 struct block_device *bdev;
1662 struct inode *inode;
1666 if (!pathname || !*pathname)
1667 return ERR_PTR(-EINVAL);
1669 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1671 return ERR_PTR(error);
1673 inode = path.dentry->d_inode;
1675 if (!S_ISBLK(inode->i_mode))
1678 if (path.mnt->mnt_flags & MNT_NODEV)
1681 bdev = bd_acquire(inode);
1688 bdev = ERR_PTR(error);
1691 EXPORT_SYMBOL(lookup_bdev);
1693 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1695 struct super_block *sb = get_super(bdev);
1700 * no need to lock the super, get_super holds the
1701 * read mutex so the filesystem cannot go away
1702 * under us (->put_super runs with the write lock
1705 shrink_dcache_sb(sb);
1706 res = invalidate_inodes(sb, kill_dirty);
1709 invalidate_bdev(bdev);
1712 EXPORT_SYMBOL(__invalidate_device);