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;
47 EXPORT_SYMBOL(I_BDEV);
50 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
51 * need to move it onto the dirty list of @dst so that the inode is always on
54 static void bdev_inode_switch_bdi(struct inode *inode,
55 struct backing_dev_info *dst)
57 struct backing_dev_info *old = inode->i_data.backing_dev_info;
59 if (unlikely(dst == old)) /* deadlock avoidance */
61 bdi_lock_two(&old->wb, &dst->wb);
62 spin_lock(&inode->i_lock);
63 inode->i_data.backing_dev_info = dst;
64 if (inode->i_state & I_DIRTY)
65 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
66 spin_unlock(&inode->i_lock);
67 spin_unlock(&old->wb.list_lock);
68 spin_unlock(&dst->wb.list_lock);
71 static sector_t max_block(struct block_device *bdev)
73 sector_t retval = ~((sector_t)0);
74 loff_t sz = i_size_read(bdev->bd_inode);
77 unsigned int size = block_size(bdev);
78 unsigned int sizebits = blksize_bits(size);
79 retval = (sz >> sizebits);
84 /* Kill _all_ buffers and pagecache , dirty or not.. */
85 static void kill_bdev(struct block_device *bdev)
87 if (bdev->bd_inode->i_mapping->nrpages == 0)
90 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
93 int set_blocksize(struct block_device *bdev, int size)
95 /* Size must be a power of two, and between 512 and PAGE_SIZE */
96 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
99 /* Size cannot be smaller than the size supported by the device */
100 if (size < bdev_logical_block_size(bdev))
103 /* Don't change the size if it is same as current */
104 if (bdev->bd_block_size != size) {
106 bdev->bd_block_size = size;
107 bdev->bd_inode->i_blkbits = blksize_bits(size);
113 EXPORT_SYMBOL(set_blocksize);
115 int sb_set_blocksize(struct super_block *sb, int size)
117 if (set_blocksize(sb->s_bdev, size))
119 /* If we get here, we know size is power of two
120 * and it's value is between 512 and PAGE_SIZE */
121 sb->s_blocksize = size;
122 sb->s_blocksize_bits = blksize_bits(size);
123 return sb->s_blocksize;
126 EXPORT_SYMBOL(sb_set_blocksize);
128 int sb_min_blocksize(struct super_block *sb, int size)
130 int minsize = bdev_logical_block_size(sb->s_bdev);
133 return sb_set_blocksize(sb, size);
136 EXPORT_SYMBOL(sb_min_blocksize);
139 blkdev_get_block(struct inode *inode, sector_t iblock,
140 struct buffer_head *bh, int create)
142 if (iblock >= max_block(I_BDEV(inode))) {
147 * for reads, we're just trying to fill a partial page.
148 * return a hole, they will have to call get_block again
149 * before they can fill it, and they will get -EIO at that
154 bh->b_bdev = I_BDEV(inode);
155 bh->b_blocknr = iblock;
156 set_buffer_mapped(bh);
161 blkdev_get_blocks(struct inode *inode, sector_t iblock,
162 struct buffer_head *bh, int create)
164 sector_t end_block = max_block(I_BDEV(inode));
165 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
167 if ((iblock + max_blocks) > end_block) {
168 max_blocks = end_block - iblock;
169 if ((long)max_blocks <= 0) {
171 return -EIO; /* write fully beyond EOF */
173 * It is a read which is fully beyond EOF. We return
174 * a !buffer_mapped buffer
180 bh->b_bdev = I_BDEV(inode);
181 bh->b_blocknr = iblock;
182 bh->b_size = max_blocks << inode->i_blkbits;
184 set_buffer_mapped(bh);
189 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
190 loff_t offset, unsigned long nr_segs)
192 struct file *file = iocb->ki_filp;
193 struct inode *inode = file->f_mapping->host;
195 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
196 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
199 int __sync_blockdev(struct block_device *bdev, int wait)
204 return filemap_flush(bdev->bd_inode->i_mapping);
205 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
209 * Write out and wait upon all the dirty data associated with a block
210 * device via its mapping. Does not take the superblock lock.
212 int sync_blockdev(struct block_device *bdev)
214 return __sync_blockdev(bdev, 1);
216 EXPORT_SYMBOL(sync_blockdev);
219 * Write out and wait upon all dirty data associated with this
220 * device. Filesystem data as well as the underlying block
221 * device. Takes the superblock lock.
223 int fsync_bdev(struct block_device *bdev)
225 struct super_block *sb = get_super(bdev);
227 int res = sync_filesystem(sb);
231 return sync_blockdev(bdev);
233 EXPORT_SYMBOL(fsync_bdev);
236 * freeze_bdev -- lock a filesystem and force it into a consistent state
237 * @bdev: blockdevice to lock
239 * If a superblock is found on this device, we take the s_umount semaphore
240 * on it to make sure nobody unmounts until the snapshot creation is done.
241 * The reference counter (bd_fsfreeze_count) guarantees that only the last
242 * unfreeze process can unfreeze the frozen filesystem actually when multiple
243 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
244 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
247 struct super_block *freeze_bdev(struct block_device *bdev)
249 struct super_block *sb;
252 mutex_lock(&bdev->bd_fsfreeze_mutex);
253 if (++bdev->bd_fsfreeze_count > 1) {
255 * We don't even need to grab a reference - the first call
256 * to freeze_bdev grab an active reference and only the last
257 * thaw_bdev drops it.
259 sb = get_super(bdev);
261 mutex_unlock(&bdev->bd_fsfreeze_mutex);
265 sb = get_active_super(bdev);
268 error = freeze_super(sb);
270 deactivate_super(sb);
271 bdev->bd_fsfreeze_count--;
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273 return ERR_PTR(error);
275 deactivate_super(sb);
278 mutex_unlock(&bdev->bd_fsfreeze_mutex);
279 return sb; /* thaw_bdev releases s->s_umount */
281 EXPORT_SYMBOL(freeze_bdev);
284 * thaw_bdev -- unlock filesystem
285 * @bdev: blockdevice to unlock
286 * @sb: associated superblock
288 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
290 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
294 mutex_lock(&bdev->bd_fsfreeze_mutex);
295 if (!bdev->bd_fsfreeze_count)
299 if (--bdev->bd_fsfreeze_count > 0)
305 error = thaw_super(sb);
307 bdev->bd_fsfreeze_count++;
308 mutex_unlock(&bdev->bd_fsfreeze_mutex);
312 mutex_unlock(&bdev->bd_fsfreeze_mutex);
315 EXPORT_SYMBOL(thaw_bdev);
317 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
319 return block_write_full_page(page, blkdev_get_block, wbc);
322 static int blkdev_readpage(struct file * file, struct page * page)
324 return block_read_full_page(page, blkdev_get_block);
327 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
328 loff_t pos, unsigned len, unsigned flags,
329 struct page **pagep, void **fsdata)
331 return block_write_begin(mapping, pos, len, flags, pagep,
335 static int blkdev_write_end(struct file *file, struct address_space *mapping,
336 loff_t pos, unsigned len, unsigned copied,
337 struct page *page, void *fsdata)
340 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
343 page_cache_release(page);
350 * for a block special file file->f_path.dentry->d_inode->i_size is zero
351 * so we compute the size by hand (just as in block_read/write above)
353 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
355 struct inode *bd_inode = file->f_mapping->host;
359 mutex_lock(&bd_inode->i_mutex);
360 size = i_size_read(bd_inode);
368 offset += file->f_pos;
374 if (offset >= 0 && offset <= size) {
375 if (offset != file->f_pos) {
376 file->f_pos = offset;
381 mutex_unlock(&bd_inode->i_mutex);
385 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
387 struct inode *bd_inode = filp->f_mapping->host;
388 struct block_device *bdev = I_BDEV(bd_inode);
391 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
396 * There is no need to serialise calls to blkdev_issue_flush with
397 * i_mutex and doing so causes performance issues with concurrent
398 * O_SYNC writers to a block device.
400 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
401 if (error == -EOPNOTSUPP)
406 EXPORT_SYMBOL(blkdev_fsync);
412 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
413 static struct kmem_cache * bdev_cachep __read_mostly;
415 static struct inode *bdev_alloc_inode(struct super_block *sb)
417 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
420 return &ei->vfs_inode;
423 static void bdev_i_callback(struct rcu_head *head)
425 struct inode *inode = container_of(head, struct inode, i_rcu);
426 struct bdev_inode *bdi = BDEV_I(inode);
428 INIT_LIST_HEAD(&inode->i_dentry);
429 kmem_cache_free(bdev_cachep, bdi);
432 static void bdev_destroy_inode(struct inode *inode)
434 call_rcu(&inode->i_rcu, bdev_i_callback);
437 static void init_once(void *foo)
439 struct bdev_inode *ei = (struct bdev_inode *) foo;
440 struct block_device *bdev = &ei->bdev;
442 memset(bdev, 0, sizeof(*bdev));
443 mutex_init(&bdev->bd_mutex);
444 INIT_LIST_HEAD(&bdev->bd_inodes);
445 INIT_LIST_HEAD(&bdev->bd_list);
447 INIT_LIST_HEAD(&bdev->bd_holder_disks);
449 inode_init_once(&ei->vfs_inode);
450 /* Initialize mutex for freeze. */
451 mutex_init(&bdev->bd_fsfreeze_mutex);
454 static inline void __bd_forget(struct inode *inode)
456 list_del_init(&inode->i_devices);
457 inode->i_bdev = NULL;
458 inode->i_mapping = &inode->i_data;
461 static void bdev_evict_inode(struct inode *inode)
463 struct block_device *bdev = &BDEV_I(inode)->bdev;
465 truncate_inode_pages(&inode->i_data, 0);
466 invalidate_inode_buffers(inode); /* is it needed here? */
467 end_writeback(inode);
468 spin_lock(&bdev_lock);
469 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
470 __bd_forget(list_entry(p, struct inode, i_devices));
472 list_del_init(&bdev->bd_list);
473 spin_unlock(&bdev_lock);
476 static const struct super_operations bdev_sops = {
477 .statfs = simple_statfs,
478 .alloc_inode = bdev_alloc_inode,
479 .destroy_inode = bdev_destroy_inode,
480 .drop_inode = generic_delete_inode,
481 .evict_inode = bdev_evict_inode,
484 static struct dentry *bd_mount(struct file_system_type *fs_type,
485 int flags, const char *dev_name, void *data)
487 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
490 static struct file_system_type bd_type = {
493 .kill_sb = kill_anon_super,
496 static struct super_block *blockdev_superblock __read_mostly;
498 void __init bdev_cache_init(void)
501 struct vfsmount *bd_mnt;
503 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
504 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
505 SLAB_MEM_SPREAD|SLAB_PANIC),
507 err = register_filesystem(&bd_type);
509 panic("Cannot register bdev pseudo-fs");
510 bd_mnt = kern_mount(&bd_type);
512 panic("Cannot create bdev pseudo-fs");
514 * This vfsmount structure is only used to obtain the
515 * blockdev_superblock, so tell kmemleak not to report it.
517 kmemleak_not_leak(bd_mnt);
518 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
522 * Most likely _very_ bad one - but then it's hardly critical for small
523 * /dev and can be fixed when somebody will need really large one.
524 * Keep in mind that it will be fed through icache hash function too.
526 static inline unsigned long hash(dev_t dev)
528 return MAJOR(dev)+MINOR(dev);
531 static int bdev_test(struct inode *inode, void *data)
533 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
536 static int bdev_set(struct inode *inode, void *data)
538 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
542 static LIST_HEAD(all_bdevs);
544 struct block_device *bdget(dev_t dev)
546 struct block_device *bdev;
549 inode = iget5_locked(blockdev_superblock, hash(dev),
550 bdev_test, bdev_set, &dev);
555 bdev = &BDEV_I(inode)->bdev;
557 if (inode->i_state & I_NEW) {
558 bdev->bd_contains = NULL;
559 bdev->bd_super = NULL;
560 bdev->bd_inode = inode;
561 bdev->bd_block_size = (1 << inode->i_blkbits);
562 bdev->bd_part_count = 0;
563 bdev->bd_invalidated = 0;
564 inode->i_mode = S_IFBLK;
566 inode->i_bdev = bdev;
567 inode->i_data.a_ops = &def_blk_aops;
568 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
569 inode->i_data.backing_dev_info = &default_backing_dev_info;
570 spin_lock(&bdev_lock);
571 list_add(&bdev->bd_list, &all_bdevs);
572 spin_unlock(&bdev_lock);
573 unlock_new_inode(inode);
578 EXPORT_SYMBOL(bdget);
581 * bdgrab -- Grab a reference to an already referenced block device
582 * @bdev: Block device to grab a reference to.
584 struct block_device *bdgrab(struct block_device *bdev)
586 ihold(bdev->bd_inode);
590 long nr_blockdev_pages(void)
592 struct block_device *bdev;
594 spin_lock(&bdev_lock);
595 list_for_each_entry(bdev, &all_bdevs, bd_list) {
596 ret += bdev->bd_inode->i_mapping->nrpages;
598 spin_unlock(&bdev_lock);
602 void bdput(struct block_device *bdev)
604 iput(bdev->bd_inode);
607 EXPORT_SYMBOL(bdput);
609 static struct block_device *bd_acquire(struct inode *inode)
611 struct block_device *bdev;
613 spin_lock(&bdev_lock);
614 bdev = inode->i_bdev;
616 ihold(bdev->bd_inode);
617 spin_unlock(&bdev_lock);
620 spin_unlock(&bdev_lock);
622 bdev = bdget(inode->i_rdev);
624 spin_lock(&bdev_lock);
625 if (!inode->i_bdev) {
627 * We take an additional reference to bd_inode,
628 * and it's released in clear_inode() of inode.
629 * So, we can access it via ->i_mapping always
632 ihold(bdev->bd_inode);
633 inode->i_bdev = bdev;
634 inode->i_mapping = bdev->bd_inode->i_mapping;
635 list_add(&inode->i_devices, &bdev->bd_inodes);
637 spin_unlock(&bdev_lock);
642 static inline int sb_is_blkdev_sb(struct super_block *sb)
644 return sb == blockdev_superblock;
647 /* Call when you free inode */
649 void bd_forget(struct inode *inode)
651 struct block_device *bdev = NULL;
653 spin_lock(&bdev_lock);
655 if (!sb_is_blkdev_sb(inode->i_sb))
656 bdev = inode->i_bdev;
659 spin_unlock(&bdev_lock);
662 iput(bdev->bd_inode);
666 * bd_may_claim - test whether a block device can be claimed
667 * @bdev: block device of interest
668 * @whole: whole block device containing @bdev, may equal @bdev
669 * @holder: holder trying to claim @bdev
671 * Test whether @bdev can be claimed by @holder.
674 * spin_lock(&bdev_lock).
677 * %true if @bdev can be claimed, %false otherwise.
679 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
682 if (bdev->bd_holder == holder)
683 return true; /* already a holder */
684 else if (bdev->bd_holder != NULL)
685 return false; /* held by someone else */
686 else if (bdev->bd_contains == bdev)
687 return true; /* is a whole device which isn't held */
689 else if (whole->bd_holder == bd_may_claim)
690 return true; /* is a partition of a device that is being partitioned */
691 else if (whole->bd_holder != NULL)
692 return false; /* is a partition of a held device */
694 return true; /* is a partition of an un-held device */
698 * bd_prepare_to_claim - prepare to claim a block device
699 * @bdev: block device of interest
700 * @whole: the whole device containing @bdev, may equal @bdev
701 * @holder: holder trying to claim @bdev
703 * Prepare to claim @bdev. This function fails if @bdev is already
704 * claimed by another holder and waits if another claiming is in
705 * progress. This function doesn't actually claim. On successful
706 * return, the caller has ownership of bd_claiming and bd_holder[s].
709 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
713 * 0 if @bdev can be claimed, -EBUSY otherwise.
715 static int bd_prepare_to_claim(struct block_device *bdev,
716 struct block_device *whole, void *holder)
719 /* if someone else claimed, fail */
720 if (!bd_may_claim(bdev, whole, holder))
723 /* if claiming is already in progress, wait for it to finish */
724 if (whole->bd_claiming) {
725 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
728 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
729 spin_unlock(&bdev_lock);
731 finish_wait(wq, &wait);
732 spin_lock(&bdev_lock);
741 * bd_start_claiming - start claiming a block device
742 * @bdev: block device of interest
743 * @holder: holder trying to claim @bdev
745 * @bdev is about to be opened exclusively. Check @bdev can be opened
746 * exclusively and mark that an exclusive open is in progress. Each
747 * successful call to this function must be matched with a call to
748 * either bd_finish_claiming() or bd_abort_claiming() (which do not
751 * This function is used to gain exclusive access to the block device
752 * without actually causing other exclusive open attempts to fail. It
753 * should be used when the open sequence itself requires exclusive
754 * access but may subsequently fail.
760 * Pointer to the block device containing @bdev on success, ERR_PTR()
763 static struct block_device *bd_start_claiming(struct block_device *bdev,
766 struct gendisk *disk;
767 struct block_device *whole;
773 * @bdev might not have been initialized properly yet, look up
774 * and grab the outer block device the hard way.
776 disk = get_gendisk(bdev->bd_dev, &partno);
778 return ERR_PTR(-ENXIO);
781 * Normally, @bdev should equal what's returned from bdget_disk()
782 * if partno is 0; however, some drivers (floppy) use multiple
783 * bdev's for the same physical device and @bdev may be one of the
784 * aliases. Keep @bdev if partno is 0. This means claimer
785 * tracking is broken for those devices but it has always been that
789 whole = bdget_disk(disk, 0);
791 whole = bdgrab(bdev);
793 module_put(disk->fops->owner);
796 return ERR_PTR(-ENOMEM);
798 /* prepare to claim, if successful, mark claiming in progress */
799 spin_lock(&bdev_lock);
801 err = bd_prepare_to_claim(bdev, whole, holder);
803 whole->bd_claiming = holder;
804 spin_unlock(&bdev_lock);
807 spin_unlock(&bdev_lock);
814 struct bd_holder_disk {
815 struct list_head list;
816 struct gendisk *disk;
820 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
821 struct gendisk *disk)
823 struct bd_holder_disk *holder;
825 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
826 if (holder->disk == disk)
831 static int add_symlink(struct kobject *from, struct kobject *to)
833 return sysfs_create_link(from, to, kobject_name(to));
836 static void del_symlink(struct kobject *from, struct kobject *to)
838 sysfs_remove_link(from, kobject_name(to));
842 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
843 * @bdev: the claimed slave bdev
844 * @disk: the holding disk
846 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
848 * This functions creates the following sysfs symlinks.
850 * - from "slaves" directory of the holder @disk to the claimed @bdev
851 * - from "holders" directory of the @bdev to the holder @disk
853 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
854 * passed to bd_link_disk_holder(), then:
856 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
857 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
859 * The caller must have claimed @bdev before calling this function and
860 * ensure that both @bdev and @disk are valid during the creation and
861 * lifetime of these symlinks.
867 * 0 on success, -errno on failure.
869 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
871 struct bd_holder_disk *holder;
874 mutex_lock(&bdev->bd_mutex);
876 WARN_ON_ONCE(!bdev->bd_holder);
878 /* FIXME: remove the following once add_disk() handles errors */
879 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
882 holder = bd_find_holder_disk(bdev, disk);
888 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
894 INIT_LIST_HEAD(&holder->list);
898 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
902 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
906 * bdev could be deleted beneath us which would implicitly destroy
907 * the holder directory. Hold on to it.
909 kobject_get(bdev->bd_part->holder_dir);
911 list_add(&holder->list, &bdev->bd_holder_disks);
915 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
919 mutex_unlock(&bdev->bd_mutex);
922 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
925 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
926 * @bdev: the calimed slave bdev
927 * @disk: the holding disk
929 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
934 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
936 struct bd_holder_disk *holder;
938 mutex_lock(&bdev->bd_mutex);
940 holder = bd_find_holder_disk(bdev, disk);
942 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
943 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
944 del_symlink(bdev->bd_part->holder_dir,
945 &disk_to_dev(disk)->kobj);
946 kobject_put(bdev->bd_part->holder_dir);
947 list_del_init(&holder->list);
951 mutex_unlock(&bdev->bd_mutex);
953 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
957 * flush_disk - invalidates all buffer-cache entries on a disk
959 * @bdev: struct block device to be flushed
960 * @kill_dirty: flag to guide handling of dirty inodes
962 * Invalidates all buffer-cache entries on a disk. It should be called
963 * when a disk has been changed -- either by a media change or online
966 static void flush_disk(struct block_device *bdev, bool kill_dirty)
968 if (__invalidate_device(bdev, kill_dirty)) {
969 char name[BDEVNAME_SIZE] = "";
972 disk_name(bdev->bd_disk, 0, name);
973 printk(KERN_WARNING "VFS: busy inodes on changed media or "
974 "resized disk %s\n", name);
979 if (disk_part_scan_enabled(bdev->bd_disk))
980 bdev->bd_invalidated = 1;
984 * check_disk_size_change - checks for disk size change and adjusts bdev size.
985 * @disk: struct gendisk to check
986 * @bdev: struct bdev to adjust.
988 * This routine checks to see if the bdev size does not match the disk size
989 * and adjusts it if it differs.
991 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
993 loff_t disk_size, bdev_size;
995 disk_size = (loff_t)get_capacity(disk) << 9;
996 bdev_size = i_size_read(bdev->bd_inode);
997 if (disk_size != bdev_size) {
998 char name[BDEVNAME_SIZE];
1000 disk_name(disk, 0, name);
1002 "%s: detected capacity change from %lld to %lld\n",
1003 name, bdev_size, disk_size);
1004 i_size_write(bdev->bd_inode, disk_size);
1005 flush_disk(bdev, false);
1008 EXPORT_SYMBOL(check_disk_size_change);
1011 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1012 * @disk: struct gendisk to be revalidated
1014 * This routine is a wrapper for lower-level driver's revalidate_disk
1015 * call-backs. It is used to do common pre and post operations needed
1016 * for all revalidate_disk operations.
1018 int revalidate_disk(struct gendisk *disk)
1020 struct block_device *bdev;
1023 if (disk->fops->revalidate_disk)
1024 ret = disk->fops->revalidate_disk(disk);
1026 bdev = bdget_disk(disk, 0);
1030 mutex_lock(&bdev->bd_mutex);
1031 check_disk_size_change(disk, bdev);
1032 mutex_unlock(&bdev->bd_mutex);
1036 EXPORT_SYMBOL(revalidate_disk);
1039 * This routine checks whether a removable media has been changed,
1040 * and invalidates all buffer-cache-entries in that case. This
1041 * is a relatively slow routine, so we have to try to minimize using
1042 * it. Thus it is called only upon a 'mount' or 'open'. This
1043 * is the best way of combining speed and utility, I think.
1044 * People changing diskettes in the middle of an operation deserve
1047 int check_disk_change(struct block_device *bdev)
1049 struct gendisk *disk = bdev->bd_disk;
1050 const struct block_device_operations *bdops = disk->fops;
1051 unsigned int events;
1053 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1054 DISK_EVENT_EJECT_REQUEST);
1055 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1058 flush_disk(bdev, true);
1059 if (bdops->revalidate_disk)
1060 bdops->revalidate_disk(bdev->bd_disk);
1064 EXPORT_SYMBOL(check_disk_change);
1066 void bd_set_size(struct block_device *bdev, loff_t size)
1068 unsigned bsize = bdev_logical_block_size(bdev);
1070 bdev->bd_inode->i_size = size;
1071 while (bsize < PAGE_CACHE_SIZE) {
1076 bdev->bd_block_size = bsize;
1077 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1079 EXPORT_SYMBOL(bd_set_size);
1081 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1086 * mutex_lock(part->bd_mutex)
1087 * mutex_lock_nested(whole->bd_mutex, 1)
1090 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1092 struct gendisk *disk;
1093 struct module *owner;
1098 if (mode & FMODE_READ)
1100 if (mode & FMODE_WRITE)
1103 * hooks: /n/, see "layering violations".
1106 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1116 disk = get_gendisk(bdev->bd_dev, &partno);
1119 owner = disk->fops->owner;
1121 disk_block_events(disk);
1122 mutex_lock_nested(&bdev->bd_mutex, for_part);
1123 if (!bdev->bd_openers) {
1124 bdev->bd_disk = disk;
1125 bdev->bd_contains = bdev;
1127 struct backing_dev_info *bdi;
1130 bdev->bd_part = disk_get_part(disk, partno);
1135 if (disk->fops->open) {
1136 ret = disk->fops->open(bdev, mode);
1137 if (ret == -ERESTARTSYS) {
1138 /* Lost a race with 'disk' being
1139 * deleted, try again.
1142 disk_put_part(bdev->bd_part);
1143 bdev->bd_part = NULL;
1144 bdev->bd_disk = NULL;
1145 mutex_unlock(&bdev->bd_mutex);
1146 disk_unblock_events(disk);
1153 if (!ret && !bdev->bd_openers) {
1154 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1155 bdi = blk_get_backing_dev_info(bdev);
1157 bdi = &default_backing_dev_info;
1158 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1162 * If the device is invalidated, rescan partition
1163 * if open succeeded or failed with -ENOMEDIUM.
1164 * The latter is necessary to prevent ghost
1165 * partitions on a removed medium.
1167 if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1168 rescan_partitions(disk, bdev);
1172 struct block_device *whole;
1173 whole = bdget_disk(disk, 0);
1178 ret = __blkdev_get(whole, mode, 1);
1181 bdev->bd_contains = whole;
1182 bdev_inode_switch_bdi(bdev->bd_inode,
1183 whole->bd_inode->i_data.backing_dev_info);
1184 bdev->bd_part = disk_get_part(disk, partno);
1185 if (!(disk->flags & GENHD_FL_UP) ||
1186 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1190 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1193 if (bdev->bd_contains == bdev) {
1195 if (bdev->bd_disk->fops->open)
1196 ret = bdev->bd_disk->fops->open(bdev, mode);
1197 /* the same as first opener case, read comment there */
1198 if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1199 rescan_partitions(bdev->bd_disk, bdev);
1201 goto out_unlock_bdev;
1203 /* only one opener holds refs to the module and disk */
1209 bdev->bd_part_count++;
1210 mutex_unlock(&bdev->bd_mutex);
1211 disk_unblock_events(disk);
1215 disk_put_part(bdev->bd_part);
1216 bdev->bd_disk = NULL;
1217 bdev->bd_part = NULL;
1218 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1219 if (bdev != bdev->bd_contains)
1220 __blkdev_put(bdev->bd_contains, mode, 1);
1221 bdev->bd_contains = NULL;
1223 mutex_unlock(&bdev->bd_mutex);
1224 disk_unblock_events(disk);
1234 * blkdev_get - open a block device
1235 * @bdev: block_device to open
1236 * @mode: FMODE_* mask
1237 * @holder: exclusive holder identifier
1239 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1240 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1241 * @holder is invalid. Exclusive opens may nest for the same @holder.
1243 * On success, the reference count of @bdev is unchanged. On failure,
1250 * 0 on success, -errno on failure.
1252 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1254 struct block_device *whole = NULL;
1257 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1259 if ((mode & FMODE_EXCL) && holder) {
1260 whole = bd_start_claiming(bdev, holder);
1261 if (IS_ERR(whole)) {
1263 return PTR_ERR(whole);
1267 res = __blkdev_get(bdev, mode, 0);
1270 struct gendisk *disk = whole->bd_disk;
1272 /* finish claiming */
1273 mutex_lock(&bdev->bd_mutex);
1274 spin_lock(&bdev_lock);
1277 BUG_ON(!bd_may_claim(bdev, whole, holder));
1279 * Note that for a whole device bd_holders
1280 * will be incremented twice, and bd_holder
1281 * will be set to bd_may_claim before being
1284 whole->bd_holders++;
1285 whole->bd_holder = bd_may_claim;
1287 bdev->bd_holder = holder;
1290 /* tell others that we're done */
1291 BUG_ON(whole->bd_claiming != holder);
1292 whole->bd_claiming = NULL;
1293 wake_up_bit(&whole->bd_claiming, 0);
1295 spin_unlock(&bdev_lock);
1298 * Block event polling for write claims if requested. Any
1299 * write holder makes the write_holder state stick until
1300 * all are released. This is good enough and tracking
1301 * individual writeable reference is too fragile given the
1302 * way @mode is used in blkdev_get/put().
1304 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1305 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1306 bdev->bd_write_holder = true;
1307 disk_block_events(disk);
1310 mutex_unlock(&bdev->bd_mutex);
1316 EXPORT_SYMBOL(blkdev_get);
1319 * blkdev_get_by_path - open a block device by name
1320 * @path: path to the block device to open
1321 * @mode: FMODE_* mask
1322 * @holder: exclusive holder identifier
1324 * Open the blockdevice described by the device file at @path. @mode
1325 * and @holder are identical to blkdev_get().
1327 * On success, the returned block_device has reference count of one.
1333 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1335 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1338 struct block_device *bdev;
1341 bdev = lookup_bdev(path);
1345 err = blkdev_get(bdev, mode, holder);
1347 return ERR_PTR(err);
1349 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1350 blkdev_put(bdev, mode);
1351 return ERR_PTR(-EACCES);
1356 EXPORT_SYMBOL(blkdev_get_by_path);
1359 * blkdev_get_by_dev - open a block device by device number
1360 * @dev: device number of block device to open
1361 * @mode: FMODE_* mask
1362 * @holder: exclusive holder identifier
1364 * Open the blockdevice described by device number @dev. @mode and
1365 * @holder are identical to blkdev_get().
1367 * Use it ONLY if you really do not have anything better - i.e. when
1368 * you are behind a truly sucky interface and all you are given is a
1369 * device number. _Never_ to be used for internal purposes. If you
1370 * ever need it - reconsider your API.
1372 * On success, the returned block_device has reference count of one.
1378 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1380 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1382 struct block_device *bdev;
1387 return ERR_PTR(-ENOMEM);
1389 err = blkdev_get(bdev, mode, holder);
1391 return ERR_PTR(err);
1395 EXPORT_SYMBOL(blkdev_get_by_dev);
1397 static int blkdev_open(struct inode * inode, struct file * filp)
1399 struct block_device *bdev;
1402 * Preserve backwards compatibility and allow large file access
1403 * even if userspace doesn't ask for it explicitly. Some mkfs
1404 * binary needs it. We might want to drop this workaround
1405 * during an unstable branch.
1407 filp->f_flags |= O_LARGEFILE;
1409 if (filp->f_flags & O_NDELAY)
1410 filp->f_mode |= FMODE_NDELAY;
1411 if (filp->f_flags & O_EXCL)
1412 filp->f_mode |= FMODE_EXCL;
1413 if ((filp->f_flags & O_ACCMODE) == 3)
1414 filp->f_mode |= FMODE_WRITE_IOCTL;
1416 bdev = bd_acquire(inode);
1420 filp->f_mapping = bdev->bd_inode->i_mapping;
1422 return blkdev_get(bdev, filp->f_mode, filp);
1425 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1428 struct gendisk *disk = bdev->bd_disk;
1429 struct block_device *victim = NULL;
1431 mutex_lock_nested(&bdev->bd_mutex, for_part);
1433 bdev->bd_part_count--;
1435 if (!--bdev->bd_openers) {
1436 WARN_ON_ONCE(bdev->bd_holders);
1437 sync_blockdev(bdev);
1439 /* ->release can cause the old bdi to disappear,
1440 * so must switch it out first
1442 bdev_inode_switch_bdi(bdev->bd_inode,
1443 &default_backing_dev_info);
1445 if (bdev->bd_contains == bdev) {
1446 if (disk->fops->release)
1447 ret = disk->fops->release(disk, mode);
1449 if (!bdev->bd_openers) {
1450 struct module *owner = disk->fops->owner;
1452 disk_put_part(bdev->bd_part);
1453 bdev->bd_part = NULL;
1454 bdev->bd_disk = NULL;
1455 if (bdev != bdev->bd_contains)
1456 victim = bdev->bd_contains;
1457 bdev->bd_contains = NULL;
1462 mutex_unlock(&bdev->bd_mutex);
1465 __blkdev_put(victim, mode, 1);
1469 int blkdev_put(struct block_device *bdev, fmode_t mode)
1471 mutex_lock(&bdev->bd_mutex);
1473 if (mode & FMODE_EXCL) {
1477 * Release a claim on the device. The holder fields
1478 * are protected with bdev_lock. bd_mutex is to
1479 * synchronize disk_holder unlinking.
1481 spin_lock(&bdev_lock);
1483 WARN_ON_ONCE(--bdev->bd_holders < 0);
1484 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1486 /* bd_contains might point to self, check in a separate step */
1487 if ((bdev_free = !bdev->bd_holders))
1488 bdev->bd_holder = NULL;
1489 if (!bdev->bd_contains->bd_holders)
1490 bdev->bd_contains->bd_holder = NULL;
1492 spin_unlock(&bdev_lock);
1495 * If this was the last claim, remove holder link and
1496 * unblock evpoll if it was a write holder.
1498 if (bdev_free && bdev->bd_write_holder) {
1499 disk_unblock_events(bdev->bd_disk);
1500 bdev->bd_write_holder = false;
1505 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1506 * event. This is to ensure detection of media removal commanded
1507 * from userland - e.g. eject(1).
1509 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1511 mutex_unlock(&bdev->bd_mutex);
1513 return __blkdev_put(bdev, mode, 0);
1515 EXPORT_SYMBOL(blkdev_put);
1517 static int blkdev_close(struct inode * inode, struct file * filp)
1519 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1521 return blkdev_put(bdev, filp->f_mode);
1524 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1526 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1527 fmode_t mode = file->f_mode;
1530 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1531 * to updated it before every ioctl.
1533 if (file->f_flags & O_NDELAY)
1534 mode |= FMODE_NDELAY;
1536 mode &= ~FMODE_NDELAY;
1538 return blkdev_ioctl(bdev, mode, cmd, arg);
1542 * Write data to the block device. Only intended for the block device itself
1543 * and the raw driver which basically is a fake block device.
1545 * Does not take i_mutex for the write and thus is not for general purpose
1548 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1549 unsigned long nr_segs, loff_t pos)
1551 struct file *file = iocb->ki_filp;
1554 BUG_ON(iocb->ki_pos != pos);
1556 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1557 if (ret > 0 || ret == -EIOCBQUEUED) {
1560 err = generic_write_sync(file, pos, ret);
1561 if (err < 0 && ret > 0)
1566 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1569 * Try to release a page associated with block device when the system
1570 * is under memory pressure.
1572 static int blkdev_releasepage(struct page *page, gfp_t wait)
1574 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1576 if (super && super->s_op->bdev_try_to_free_page)
1577 return super->s_op->bdev_try_to_free_page(super, page, wait);
1579 return try_to_free_buffers(page);
1582 static const struct address_space_operations def_blk_aops = {
1583 .readpage = blkdev_readpage,
1584 .writepage = blkdev_writepage,
1585 .write_begin = blkdev_write_begin,
1586 .write_end = blkdev_write_end,
1587 .writepages = generic_writepages,
1588 .releasepage = blkdev_releasepage,
1589 .direct_IO = blkdev_direct_IO,
1592 const struct file_operations def_blk_fops = {
1593 .open = blkdev_open,
1594 .release = blkdev_close,
1595 .llseek = block_llseek,
1596 .read = do_sync_read,
1597 .write = do_sync_write,
1598 .aio_read = generic_file_aio_read,
1599 .aio_write = blkdev_aio_write,
1600 .mmap = generic_file_mmap,
1601 .fsync = blkdev_fsync,
1602 .unlocked_ioctl = block_ioctl,
1603 #ifdef CONFIG_COMPAT
1604 .compat_ioctl = compat_blkdev_ioctl,
1606 .splice_read = generic_file_splice_read,
1607 .splice_write = generic_file_splice_write,
1610 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1613 mm_segment_t old_fs = get_fs();
1615 res = blkdev_ioctl(bdev, 0, cmd, arg);
1620 EXPORT_SYMBOL(ioctl_by_bdev);
1623 * lookup_bdev - lookup a struct block_device by name
1624 * @pathname: special file representing the block device
1626 * Get a reference to the blockdevice at @pathname in the current
1627 * namespace if possible and return it. Return ERR_PTR(error)
1630 struct block_device *lookup_bdev(const char *pathname)
1632 struct block_device *bdev;
1633 struct inode *inode;
1637 if (!pathname || !*pathname)
1638 return ERR_PTR(-EINVAL);
1640 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1642 return ERR_PTR(error);
1644 inode = path.dentry->d_inode;
1646 if (!S_ISBLK(inode->i_mode))
1649 if (path.mnt->mnt_flags & MNT_NODEV)
1652 bdev = bd_acquire(inode);
1659 bdev = ERR_PTR(error);
1662 EXPORT_SYMBOL(lookup_bdev);
1664 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1666 struct super_block *sb = get_super(bdev);
1671 * no need to lock the super, get_super holds the
1672 * read mutex so the filesystem cannot go away
1673 * under us (->put_super runs with the write lock
1676 shrink_dcache_sb(sb);
1677 res = invalidate_inodes(sb, kill_dirty);
1680 invalidate_bdev(bdev);
1683 EXPORT_SYMBOL(__invalidate_device);