4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/buffer_head.h>
20 #include <linux/pagevec.h>
21 #include <linux/writeback.h>
22 #include <linux/mpage.h>
23 #include <linux/mount.h>
24 #include <linux/uio.h>
25 #include <linux/namei.h>
26 #include <linux/log2.h>
27 #include <linux/kmemleak.h>
28 #include <asm/uaccess.h>
32 struct block_device bdev;
33 struct inode vfs_inode;
36 static const struct address_space_operations def_blk_aops;
38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
40 return container_of(inode, struct bdev_inode, vfs_inode);
43 inline struct block_device *I_BDEV(struct inode *inode)
45 return &BDEV_I(inode)->bdev;
48 EXPORT_SYMBOL(I_BDEV);
51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
52 * we need to move it onto the dirty list of @dst so that the inode is always
55 static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
58 spin_lock(&inode_lock);
59 inode->i_data.backing_dev_info = dst;
60 if (inode->i_state & I_DIRTY)
61 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
62 spin_unlock(&inode_lock);
65 static sector_t max_block(struct block_device *bdev)
67 sector_t retval = ~((sector_t)0);
68 loff_t sz = i_size_read(bdev->bd_inode);
71 unsigned int size = block_size(bdev);
72 unsigned int sizebits = blksize_bits(size);
73 retval = (sz >> sizebits);
78 /* Kill _all_ buffers and pagecache , dirty or not.. */
79 static void kill_bdev(struct block_device *bdev)
81 if (bdev->bd_inode->i_mapping->nrpages == 0)
84 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
87 int set_blocksize(struct block_device *bdev, int size)
89 /* Size must be a power of two, and between 512 and PAGE_SIZE */
90 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
93 /* Size cannot be smaller than the size supported by the device */
94 if (size < bdev_logical_block_size(bdev))
97 /* Don't change the size if it is same as current */
98 if (bdev->bd_block_size != size) {
100 bdev->bd_block_size = size;
101 bdev->bd_inode->i_blkbits = blksize_bits(size);
107 EXPORT_SYMBOL(set_blocksize);
109 int sb_set_blocksize(struct super_block *sb, int size)
111 if (set_blocksize(sb->s_bdev, size))
113 /* If we get here, we know size is power of two
114 * and it's value is between 512 and PAGE_SIZE */
115 sb->s_blocksize = size;
116 sb->s_blocksize_bits = blksize_bits(size);
117 return sb->s_blocksize;
120 EXPORT_SYMBOL(sb_set_blocksize);
122 int sb_min_blocksize(struct super_block *sb, int size)
124 int minsize = bdev_logical_block_size(sb->s_bdev);
127 return sb_set_blocksize(sb, size);
130 EXPORT_SYMBOL(sb_min_blocksize);
133 blkdev_get_block(struct inode *inode, sector_t iblock,
134 struct buffer_head *bh, int create)
136 if (iblock >= max_block(I_BDEV(inode))) {
141 * for reads, we're just trying to fill a partial page.
142 * return a hole, they will have to call get_block again
143 * before they can fill it, and they will get -EIO at that
148 bh->b_bdev = I_BDEV(inode);
149 bh->b_blocknr = iblock;
150 set_buffer_mapped(bh);
155 blkdev_get_blocks(struct inode *inode, sector_t iblock,
156 struct buffer_head *bh, int create)
158 sector_t end_block = max_block(I_BDEV(inode));
159 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
161 if ((iblock + max_blocks) > end_block) {
162 max_blocks = end_block - iblock;
163 if ((long)max_blocks <= 0) {
165 return -EIO; /* write fully beyond EOF */
167 * It is a read which is fully beyond EOF. We return
168 * a !buffer_mapped buffer
174 bh->b_bdev = I_BDEV(inode);
175 bh->b_blocknr = iblock;
176 bh->b_size = max_blocks << inode->i_blkbits;
178 set_buffer_mapped(bh);
183 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
184 loff_t offset, unsigned long nr_segs)
186 struct file *file = iocb->ki_filp;
187 struct inode *inode = file->f_mapping->host;
189 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
190 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
193 int __sync_blockdev(struct block_device *bdev, int wait)
198 return filemap_flush(bdev->bd_inode->i_mapping);
199 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
203 * Write out and wait upon all the dirty data associated with a block
204 * device via its mapping. Does not take the superblock lock.
206 int sync_blockdev(struct block_device *bdev)
208 return __sync_blockdev(bdev, 1);
210 EXPORT_SYMBOL(sync_blockdev);
213 * Write out and wait upon all dirty data associated with this
214 * device. Filesystem data as well as the underlying block
215 * device. Takes the superblock lock.
217 int fsync_bdev(struct block_device *bdev)
219 struct super_block *sb = get_super(bdev);
221 int res = sync_filesystem(sb);
225 return sync_blockdev(bdev);
227 EXPORT_SYMBOL(fsync_bdev);
230 * freeze_bdev -- lock a filesystem and force it into a consistent state
231 * @bdev: blockdevice to lock
233 * If a superblock is found on this device, we take the s_umount semaphore
234 * on it to make sure nobody unmounts until the snapshot creation is done.
235 * The reference counter (bd_fsfreeze_count) guarantees that only the last
236 * unfreeze process can unfreeze the frozen filesystem actually when multiple
237 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
238 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
241 struct super_block *freeze_bdev(struct block_device *bdev)
243 struct super_block *sb;
246 mutex_lock(&bdev->bd_fsfreeze_mutex);
247 if (++bdev->bd_fsfreeze_count > 1) {
249 * We don't even need to grab a reference - the first call
250 * to freeze_bdev grab an active reference and only the last
251 * thaw_bdev drops it.
253 sb = get_super(bdev);
255 mutex_unlock(&bdev->bd_fsfreeze_mutex);
259 sb = get_active_super(bdev);
262 error = freeze_super(sb);
264 deactivate_super(sb);
265 bdev->bd_fsfreeze_count--;
266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
267 return ERR_PTR(error);
269 deactivate_super(sb);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273 return sb; /* thaw_bdev releases s->s_umount */
275 EXPORT_SYMBOL(freeze_bdev);
278 * thaw_bdev -- unlock filesystem
279 * @bdev: blockdevice to unlock
280 * @sb: associated superblock
282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
288 mutex_lock(&bdev->bd_fsfreeze_mutex);
289 if (!bdev->bd_fsfreeze_count)
293 if (--bdev->bd_fsfreeze_count > 0)
299 error = thaw_super(sb);
301 bdev->bd_fsfreeze_count++;
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
306 mutex_unlock(&bdev->bd_fsfreeze_mutex);
309 EXPORT_SYMBOL(thaw_bdev);
311 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
313 return block_write_full_page(page, blkdev_get_block, wbc);
316 static int blkdev_readpage(struct file * file, struct page * page)
318 return block_read_full_page(page, blkdev_get_block);
321 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
322 loff_t pos, unsigned len, unsigned flags,
323 struct page **pagep, void **fsdata)
325 return block_write_begin(mapping, pos, len, flags, pagep,
329 static int blkdev_write_end(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned copied,
331 struct page *page, void *fsdata)
334 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
337 page_cache_release(page);
344 * for a block special file file->f_path.dentry->d_inode->i_size is zero
345 * so we compute the size by hand (just as in block_read/write above)
347 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
349 struct inode *bd_inode = file->f_mapping->host;
353 mutex_lock(&bd_inode->i_mutex);
354 size = i_size_read(bd_inode);
361 offset += file->f_pos;
364 if (offset >= 0 && offset <= size) {
365 if (offset != file->f_pos) {
366 file->f_pos = offset;
370 mutex_unlock(&bd_inode->i_mutex);
374 int blkdev_fsync(struct file *filp, int datasync)
376 struct inode *bd_inode = filp->f_mapping->host;
377 struct block_device *bdev = I_BDEV(bd_inode);
381 * There is no need to serialise calls to blkdev_issue_flush with
382 * i_mutex and doing so causes performance issues with concurrent
383 * O_SYNC writers to a block device.
385 mutex_unlock(&bd_inode->i_mutex);
387 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
388 if (error == -EOPNOTSUPP)
391 mutex_lock(&bd_inode->i_mutex);
395 EXPORT_SYMBOL(blkdev_fsync);
401 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
402 static struct kmem_cache * bdev_cachep __read_mostly;
404 static struct inode *bdev_alloc_inode(struct super_block *sb)
406 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
409 return &ei->vfs_inode;
412 static void bdev_i_callback(struct rcu_head *head)
414 struct inode *inode = container_of(head, struct inode, i_rcu);
415 struct bdev_inode *bdi = BDEV_I(inode);
417 INIT_LIST_HEAD(&inode->i_dentry);
418 kmem_cache_free(bdev_cachep, bdi);
421 static void bdev_destroy_inode(struct inode *inode)
423 call_rcu(&inode->i_rcu, bdev_i_callback);
426 static void init_once(void *foo)
428 struct bdev_inode *ei = (struct bdev_inode *) foo;
429 struct block_device *bdev = &ei->bdev;
431 memset(bdev, 0, sizeof(*bdev));
432 mutex_init(&bdev->bd_mutex);
433 INIT_LIST_HEAD(&bdev->bd_inodes);
434 INIT_LIST_HEAD(&bdev->bd_list);
436 INIT_LIST_HEAD(&bdev->bd_holder_disks);
438 inode_init_once(&ei->vfs_inode);
439 /* Initialize mutex for freeze. */
440 mutex_init(&bdev->bd_fsfreeze_mutex);
443 static inline void __bd_forget(struct inode *inode)
445 list_del_init(&inode->i_devices);
446 inode->i_bdev = NULL;
447 inode->i_mapping = &inode->i_data;
450 static void bdev_evict_inode(struct inode *inode)
452 struct block_device *bdev = &BDEV_I(inode)->bdev;
454 truncate_inode_pages(&inode->i_data, 0);
455 invalidate_inode_buffers(inode); /* is it needed here? */
456 end_writeback(inode);
457 spin_lock(&bdev_lock);
458 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
459 __bd_forget(list_entry(p, struct inode, i_devices));
461 list_del_init(&bdev->bd_list);
462 spin_unlock(&bdev_lock);
465 static const struct super_operations bdev_sops = {
466 .statfs = simple_statfs,
467 .alloc_inode = bdev_alloc_inode,
468 .destroy_inode = bdev_destroy_inode,
469 .drop_inode = generic_delete_inode,
470 .evict_inode = bdev_evict_inode,
473 static struct dentry *bd_mount(struct file_system_type *fs_type,
474 int flags, const char *dev_name, void *data)
476 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
479 static struct file_system_type bd_type = {
482 .kill_sb = kill_anon_super,
485 struct super_block *blockdev_superblock __read_mostly;
487 void __init bdev_cache_init(void)
490 struct vfsmount *bd_mnt;
492 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
493 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
494 SLAB_MEM_SPREAD|SLAB_PANIC),
496 err = register_filesystem(&bd_type);
498 panic("Cannot register bdev pseudo-fs");
499 bd_mnt = kern_mount(&bd_type);
501 panic("Cannot create bdev pseudo-fs");
503 * This vfsmount structure is only used to obtain the
504 * blockdev_superblock, so tell kmemleak not to report it.
506 kmemleak_not_leak(bd_mnt);
507 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
511 * Most likely _very_ bad one - but then it's hardly critical for small
512 * /dev and can be fixed when somebody will need really large one.
513 * Keep in mind that it will be fed through icache hash function too.
515 static inline unsigned long hash(dev_t dev)
517 return MAJOR(dev)+MINOR(dev);
520 static int bdev_test(struct inode *inode, void *data)
522 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
525 static int bdev_set(struct inode *inode, void *data)
527 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
531 static LIST_HEAD(all_bdevs);
533 struct block_device *bdget(dev_t dev)
535 struct block_device *bdev;
538 inode = iget5_locked(blockdev_superblock, hash(dev),
539 bdev_test, bdev_set, &dev);
544 bdev = &BDEV_I(inode)->bdev;
546 if (inode->i_state & I_NEW) {
547 bdev->bd_contains = NULL;
548 bdev->bd_inode = inode;
549 bdev->bd_block_size = (1 << inode->i_blkbits);
550 bdev->bd_part_count = 0;
551 bdev->bd_invalidated = 0;
552 inode->i_mode = S_IFBLK;
554 inode->i_bdev = bdev;
555 inode->i_data.a_ops = &def_blk_aops;
556 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
557 inode->i_data.backing_dev_info = &default_backing_dev_info;
558 spin_lock(&bdev_lock);
559 list_add(&bdev->bd_list, &all_bdevs);
560 spin_unlock(&bdev_lock);
561 unlock_new_inode(inode);
566 EXPORT_SYMBOL(bdget);
569 * bdgrab -- Grab a reference to an already referenced block device
570 * @bdev: Block device to grab a reference to.
572 struct block_device *bdgrab(struct block_device *bdev)
574 ihold(bdev->bd_inode);
578 long nr_blockdev_pages(void)
580 struct block_device *bdev;
582 spin_lock(&bdev_lock);
583 list_for_each_entry(bdev, &all_bdevs, bd_list) {
584 ret += bdev->bd_inode->i_mapping->nrpages;
586 spin_unlock(&bdev_lock);
590 void bdput(struct block_device *bdev)
592 iput(bdev->bd_inode);
595 EXPORT_SYMBOL(bdput);
597 static struct block_device *bd_acquire(struct inode *inode)
599 struct block_device *bdev;
601 spin_lock(&bdev_lock);
602 bdev = inode->i_bdev;
604 ihold(bdev->bd_inode);
605 spin_unlock(&bdev_lock);
608 spin_unlock(&bdev_lock);
610 bdev = bdget(inode->i_rdev);
612 spin_lock(&bdev_lock);
613 if (!inode->i_bdev) {
615 * We take an additional reference to bd_inode,
616 * and it's released in clear_inode() of inode.
617 * So, we can access it via ->i_mapping always
620 ihold(bdev->bd_inode);
621 inode->i_bdev = bdev;
622 inode->i_mapping = bdev->bd_inode->i_mapping;
623 list_add(&inode->i_devices, &bdev->bd_inodes);
625 spin_unlock(&bdev_lock);
630 /* Call when you free inode */
632 void bd_forget(struct inode *inode)
634 struct block_device *bdev = NULL;
636 spin_lock(&bdev_lock);
638 if (!sb_is_blkdev_sb(inode->i_sb))
639 bdev = inode->i_bdev;
642 spin_unlock(&bdev_lock);
645 iput(bdev->bd_inode);
649 * bd_may_claim - test whether a block device can be claimed
650 * @bdev: block device of interest
651 * @whole: whole block device containing @bdev, may equal @bdev
652 * @holder: holder trying to claim @bdev
654 * Test whther @bdev can be claimed by @holder.
657 * spin_lock(&bdev_lock).
660 * %true if @bdev can be claimed, %false otherwise.
662 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
665 if (bdev->bd_holder == holder)
666 return true; /* already a holder */
667 else if (bdev->bd_holder != NULL)
668 return false; /* held by someone else */
669 else if (bdev->bd_contains == bdev)
670 return true; /* is a whole device which isn't held */
672 else if (whole->bd_holder == bd_may_claim)
673 return true; /* is a partition of a device that is being partitioned */
674 else if (whole->bd_holder != NULL)
675 return false; /* is a partition of a held device */
677 return true; /* is a partition of an un-held device */
681 * bd_prepare_to_claim - prepare to claim a block device
682 * @bdev: block device of interest
683 * @whole: the whole device containing @bdev, may equal @bdev
684 * @holder: holder trying to claim @bdev
686 * Prepare to claim @bdev. This function fails if @bdev is already
687 * claimed by another holder and waits if another claiming is in
688 * progress. This function doesn't actually claim. On successful
689 * return, the caller has ownership of bd_claiming and bd_holder[s].
692 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
696 * 0 if @bdev can be claimed, -EBUSY otherwise.
698 static int bd_prepare_to_claim(struct block_device *bdev,
699 struct block_device *whole, void *holder)
702 /* if someone else claimed, fail */
703 if (!bd_may_claim(bdev, whole, holder))
706 /* if claiming is already in progress, wait for it to finish */
707 if (whole->bd_claiming) {
708 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
711 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
712 spin_unlock(&bdev_lock);
714 finish_wait(wq, &wait);
715 spin_lock(&bdev_lock);
724 * bd_start_claiming - start claiming a block device
725 * @bdev: block device of interest
726 * @holder: holder trying to claim @bdev
728 * @bdev is about to be opened exclusively. Check @bdev can be opened
729 * exclusively and mark that an exclusive open is in progress. Each
730 * successful call to this function must be matched with a call to
731 * either bd_finish_claiming() or bd_abort_claiming() (which do not
734 * This function is used to gain exclusive access to the block device
735 * without actually causing other exclusive open attempts to fail. It
736 * should be used when the open sequence itself requires exclusive
737 * access but may subsequently fail.
743 * Pointer to the block device containing @bdev on success, ERR_PTR()
746 static struct block_device *bd_start_claiming(struct block_device *bdev,
749 struct gendisk *disk;
750 struct block_device *whole;
756 * @bdev might not have been initialized properly yet, look up
757 * and grab the outer block device the hard way.
759 disk = get_gendisk(bdev->bd_dev, &partno);
761 return ERR_PTR(-ENXIO);
763 whole = bdget_disk(disk, 0);
764 module_put(disk->fops->owner);
767 return ERR_PTR(-ENOMEM);
769 /* prepare to claim, if successful, mark claiming in progress */
770 spin_lock(&bdev_lock);
772 err = bd_prepare_to_claim(bdev, whole, holder);
774 whole->bd_claiming = holder;
775 spin_unlock(&bdev_lock);
778 spin_unlock(&bdev_lock);
785 struct bd_holder_disk {
786 struct list_head list;
787 struct gendisk *disk;
791 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
792 struct gendisk *disk)
794 struct bd_holder_disk *holder;
796 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
797 if (holder->disk == disk)
802 static int add_symlink(struct kobject *from, struct kobject *to)
804 return sysfs_create_link(from, to, kobject_name(to));
807 static void del_symlink(struct kobject *from, struct kobject *to)
809 sysfs_remove_link(from, kobject_name(to));
813 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
814 * @bdev: the claimed slave bdev
815 * @disk: the holding disk
817 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
819 * This functions creates the following sysfs symlinks.
821 * - from "slaves" directory of the holder @disk to the claimed @bdev
822 * - from "holders" directory of the @bdev to the holder @disk
824 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
825 * passed to bd_link_disk_holder(), then:
827 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
828 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
830 * The caller must have claimed @bdev before calling this function and
831 * ensure that both @bdev and @disk are valid during the creation and
832 * lifetime of these symlinks.
838 * 0 on success, -errno on failure.
840 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
842 struct bd_holder_disk *holder;
845 mutex_lock(&bdev->bd_mutex);
847 WARN_ON_ONCE(!bdev->bd_holder);
849 /* FIXME: remove the following once add_disk() handles errors */
850 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
853 holder = bd_find_holder_disk(bdev, disk);
859 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
865 INIT_LIST_HEAD(&holder->list);
869 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
873 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
877 * bdev could be deleted beneath us which would implicitly destroy
878 * the holder directory. Hold on to it.
880 kobject_get(bdev->bd_part->holder_dir);
882 list_add(&holder->list, &bdev->bd_holder_disks);
886 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
890 mutex_unlock(&bdev->bd_mutex);
893 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
896 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
897 * @bdev: the calimed slave bdev
898 * @disk: the holding disk
900 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
905 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
907 struct bd_holder_disk *holder;
909 mutex_lock(&bdev->bd_mutex);
911 holder = bd_find_holder_disk(bdev, disk);
913 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
914 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
915 del_symlink(bdev->bd_part->holder_dir,
916 &disk_to_dev(disk)->kobj);
917 kobject_put(bdev->bd_part->holder_dir);
918 list_del_init(&holder->list);
922 mutex_unlock(&bdev->bd_mutex);
924 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
928 * flush_disk - invalidates all buffer-cache entries on a disk
930 * @bdev: struct block device to be flushed
931 * @kill_dirty: flag to guide handling of dirty inodes
933 * Invalidates all buffer-cache entries on a disk. It should be called
934 * when a disk has been changed -- either by a media change or online
937 static void flush_disk(struct block_device *bdev, bool kill_dirty)
939 if (__invalidate_device(bdev, kill_dirty)) {
940 char name[BDEVNAME_SIZE] = "";
943 disk_name(bdev->bd_disk, 0, name);
944 printk(KERN_WARNING "VFS: busy inodes on changed media or "
945 "resized disk %s\n", name);
950 if (disk_partitionable(bdev->bd_disk))
951 bdev->bd_invalidated = 1;
955 * check_disk_size_change - checks for disk size change and adjusts bdev size.
956 * @disk: struct gendisk to check
957 * @bdev: struct bdev to adjust.
959 * This routine checks to see if the bdev size does not match the disk size
960 * and adjusts it if it differs.
962 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
964 loff_t disk_size, bdev_size;
966 disk_size = (loff_t)get_capacity(disk) << 9;
967 bdev_size = i_size_read(bdev->bd_inode);
968 if (disk_size != bdev_size) {
969 char name[BDEVNAME_SIZE];
971 disk_name(disk, 0, name);
973 "%s: detected capacity change from %lld to %lld\n",
974 name, bdev_size, disk_size);
975 i_size_write(bdev->bd_inode, disk_size);
976 flush_disk(bdev, false);
979 EXPORT_SYMBOL(check_disk_size_change);
982 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
983 * @disk: struct gendisk to be revalidated
985 * This routine is a wrapper for lower-level driver's revalidate_disk
986 * call-backs. It is used to do common pre and post operations needed
987 * for all revalidate_disk operations.
989 int revalidate_disk(struct gendisk *disk)
991 struct block_device *bdev;
994 if (disk->fops->revalidate_disk)
995 ret = disk->fops->revalidate_disk(disk);
997 bdev = bdget_disk(disk, 0);
1001 mutex_lock(&bdev->bd_mutex);
1002 check_disk_size_change(disk, bdev);
1003 mutex_unlock(&bdev->bd_mutex);
1007 EXPORT_SYMBOL(revalidate_disk);
1010 * This routine checks whether a removable media has been changed,
1011 * and invalidates all buffer-cache-entries in that case. This
1012 * is a relatively slow routine, so we have to try to minimize using
1013 * it. Thus it is called only upon a 'mount' or 'open'. This
1014 * is the best way of combining speed and utility, I think.
1015 * People changing diskettes in the middle of an operation deserve
1018 int check_disk_change(struct block_device *bdev)
1020 struct gendisk *disk = bdev->bd_disk;
1021 const struct block_device_operations *bdops = disk->fops;
1022 unsigned int events;
1024 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1025 DISK_EVENT_EJECT_REQUEST);
1026 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1029 flush_disk(bdev, true);
1030 if (bdops->revalidate_disk)
1031 bdops->revalidate_disk(bdev->bd_disk);
1035 EXPORT_SYMBOL(check_disk_change);
1037 void bd_set_size(struct block_device *bdev, loff_t size)
1039 unsigned bsize = bdev_logical_block_size(bdev);
1041 bdev->bd_inode->i_size = size;
1042 while (bsize < PAGE_CACHE_SIZE) {
1047 bdev->bd_block_size = bsize;
1048 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1050 EXPORT_SYMBOL(bd_set_size);
1052 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1057 * mutex_lock(part->bd_mutex)
1058 * mutex_lock_nested(whole->bd_mutex, 1)
1061 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1063 struct gendisk *disk;
1068 if (mode & FMODE_READ)
1070 if (mode & FMODE_WRITE)
1073 * hooks: /n/, see "layering violations".
1076 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1086 disk = get_gendisk(bdev->bd_dev, &partno);
1090 mutex_lock_nested(&bdev->bd_mutex, for_part);
1091 if (!bdev->bd_openers) {
1092 bdev->bd_disk = disk;
1093 bdev->bd_contains = bdev;
1095 struct backing_dev_info *bdi;
1098 bdev->bd_part = disk_get_part(disk, partno);
1102 if (disk->fops->open) {
1103 ret = disk->fops->open(bdev, mode);
1104 if (ret == -ERESTARTSYS) {
1105 /* Lost a race with 'disk' being
1106 * deleted, try again.
1109 disk_put_part(bdev->bd_part);
1110 bdev->bd_part = NULL;
1111 module_put(disk->fops->owner);
1113 bdev->bd_disk = NULL;
1114 mutex_unlock(&bdev->bd_mutex);
1120 if (!bdev->bd_openers) {
1121 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1122 bdi = blk_get_backing_dev_info(bdev);
1124 bdi = &default_backing_dev_info;
1125 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1127 if (bdev->bd_invalidated)
1128 rescan_partitions(disk, bdev);
1130 struct block_device *whole;
1131 whole = bdget_disk(disk, 0);
1136 ret = __blkdev_get(whole, mode, 1);
1139 bdev->bd_contains = whole;
1140 bdev_inode_switch_bdi(bdev->bd_inode,
1141 whole->bd_inode->i_data.backing_dev_info);
1142 bdev->bd_part = disk_get_part(disk, partno);
1143 if (!(disk->flags & GENHD_FL_UP) ||
1144 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1148 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1151 module_put(disk->fops->owner);
1154 if (bdev->bd_contains == bdev) {
1155 if (bdev->bd_disk->fops->open) {
1156 ret = bdev->bd_disk->fops->open(bdev, mode);
1158 goto out_unlock_bdev;
1160 if (bdev->bd_invalidated)
1161 rescan_partitions(bdev->bd_disk, bdev);
1166 bdev->bd_part_count++;
1167 mutex_unlock(&bdev->bd_mutex);
1171 disk_put_part(bdev->bd_part);
1172 bdev->bd_disk = NULL;
1173 bdev->bd_part = NULL;
1174 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1175 if (bdev != bdev->bd_contains)
1176 __blkdev_put(bdev->bd_contains, mode, 1);
1177 bdev->bd_contains = NULL;
1179 mutex_unlock(&bdev->bd_mutex);
1182 module_put(disk->fops->owner);
1190 * blkdev_get - open a block device
1191 * @bdev: block_device to open
1192 * @mode: FMODE_* mask
1193 * @holder: exclusive holder identifier
1195 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1196 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1197 * @holder is invalid. Exclusive opens may nest for the same @holder.
1199 * On success, the reference count of @bdev is unchanged. On failure,
1206 * 0 on success, -errno on failure.
1208 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1210 struct block_device *whole = NULL;
1213 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1215 if ((mode & FMODE_EXCL) && holder) {
1216 whole = bd_start_claiming(bdev, holder);
1217 if (IS_ERR(whole)) {
1219 return PTR_ERR(whole);
1223 res = __blkdev_get(bdev, mode, 0);
1226 /* finish claiming */
1227 mutex_lock(&bdev->bd_mutex);
1228 spin_lock(&bdev_lock);
1231 BUG_ON(!bd_may_claim(bdev, whole, holder));
1233 * Note that for a whole device bd_holders
1234 * will be incremented twice, and bd_holder
1235 * will be set to bd_may_claim before being
1238 whole->bd_holders++;
1239 whole->bd_holder = bd_may_claim;
1241 bdev->bd_holder = holder;
1244 /* tell others that we're done */
1245 BUG_ON(whole->bd_claiming != holder);
1246 whole->bd_claiming = NULL;
1247 wake_up_bit(&whole->bd_claiming, 0);
1249 spin_unlock(&bdev_lock);
1252 * Block event polling for write claims. Any write
1253 * holder makes the write_holder state stick until all
1254 * are released. This is good enough and tracking
1255 * individual writeable reference is too fragile given
1256 * the way @mode is used in blkdev_get/put().
1258 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
1259 bdev->bd_write_holder = true;
1260 disk_block_events(bdev->bd_disk);
1263 mutex_unlock(&bdev->bd_mutex);
1269 EXPORT_SYMBOL(blkdev_get);
1272 * blkdev_get_by_path - open a block device by name
1273 * @path: path to the block device to open
1274 * @mode: FMODE_* mask
1275 * @holder: exclusive holder identifier
1277 * Open the blockdevice described by the device file at @path. @mode
1278 * and @holder are identical to blkdev_get().
1280 * On success, the returned block_device has reference count of one.
1286 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1288 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1291 struct block_device *bdev;
1294 bdev = lookup_bdev(path);
1298 err = blkdev_get(bdev, mode, holder);
1300 return ERR_PTR(err);
1302 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1303 blkdev_put(bdev, mode);
1304 return ERR_PTR(-EACCES);
1309 EXPORT_SYMBOL(blkdev_get_by_path);
1312 * blkdev_get_by_dev - open a block device by device number
1313 * @dev: device number of block device to open
1314 * @mode: FMODE_* mask
1315 * @holder: exclusive holder identifier
1317 * Open the blockdevice described by device number @dev. @mode and
1318 * @holder are identical to blkdev_get().
1320 * Use it ONLY if you really do not have anything better - i.e. when
1321 * you are behind a truly sucky interface and all you are given is a
1322 * device number. _Never_ to be used for internal purposes. If you
1323 * ever need it - reconsider your API.
1325 * On success, the returned block_device has reference count of one.
1331 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1333 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1335 struct block_device *bdev;
1340 return ERR_PTR(-ENOMEM);
1342 err = blkdev_get(bdev, mode, holder);
1344 return ERR_PTR(err);
1348 EXPORT_SYMBOL(blkdev_get_by_dev);
1350 static int blkdev_open(struct inode * inode, struct file * filp)
1352 struct block_device *bdev;
1355 * Preserve backwards compatibility and allow large file access
1356 * even if userspace doesn't ask for it explicitly. Some mkfs
1357 * binary needs it. We might want to drop this workaround
1358 * during an unstable branch.
1360 filp->f_flags |= O_LARGEFILE;
1362 if (filp->f_flags & O_NDELAY)
1363 filp->f_mode |= FMODE_NDELAY;
1364 if (filp->f_flags & O_EXCL)
1365 filp->f_mode |= FMODE_EXCL;
1366 if ((filp->f_flags & O_ACCMODE) == 3)
1367 filp->f_mode |= FMODE_WRITE_IOCTL;
1369 bdev = bd_acquire(inode);
1373 filp->f_mapping = bdev->bd_inode->i_mapping;
1375 return blkdev_get(bdev, filp->f_mode, filp);
1378 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1381 struct gendisk *disk = bdev->bd_disk;
1382 struct block_device *victim = NULL;
1384 mutex_lock_nested(&bdev->bd_mutex, for_part);
1386 bdev->bd_part_count--;
1388 if (!--bdev->bd_openers) {
1389 WARN_ON_ONCE(bdev->bd_holders);
1390 sync_blockdev(bdev);
1393 if (bdev->bd_contains == bdev) {
1394 if (disk->fops->release)
1395 ret = disk->fops->release(disk, mode);
1397 if (!bdev->bd_openers) {
1398 struct module *owner = disk->fops->owner;
1402 disk_put_part(bdev->bd_part);
1403 bdev->bd_part = NULL;
1404 bdev->bd_disk = NULL;
1405 bdev_inode_switch_bdi(bdev->bd_inode,
1406 &default_backing_dev_info);
1407 if (bdev != bdev->bd_contains)
1408 victim = bdev->bd_contains;
1409 bdev->bd_contains = NULL;
1411 mutex_unlock(&bdev->bd_mutex);
1414 __blkdev_put(victim, mode, 1);
1418 int blkdev_put(struct block_device *bdev, fmode_t mode)
1420 if (mode & FMODE_EXCL) {
1424 * Release a claim on the device. The holder fields
1425 * are protected with bdev_lock. bd_mutex is to
1426 * synchronize disk_holder unlinking.
1428 mutex_lock(&bdev->bd_mutex);
1429 spin_lock(&bdev_lock);
1431 WARN_ON_ONCE(--bdev->bd_holders < 0);
1432 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1434 /* bd_contains might point to self, check in a separate step */
1435 if ((bdev_free = !bdev->bd_holders))
1436 bdev->bd_holder = NULL;
1437 if (!bdev->bd_contains->bd_holders)
1438 bdev->bd_contains->bd_holder = NULL;
1440 spin_unlock(&bdev_lock);
1443 * If this was the last claim, remove holder link and
1444 * unblock evpoll if it was a write holder.
1447 if (bdev->bd_write_holder) {
1448 disk_unblock_events(bdev->bd_disk);
1449 bdev->bd_write_holder = false;
1451 disk_check_events(bdev->bd_disk);
1454 mutex_unlock(&bdev->bd_mutex);
1456 disk_check_events(bdev->bd_disk);
1458 return __blkdev_put(bdev, mode, 0);
1460 EXPORT_SYMBOL(blkdev_put);
1462 static int blkdev_close(struct inode * inode, struct file * filp)
1464 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1466 return blkdev_put(bdev, filp->f_mode);
1469 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1471 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1472 fmode_t mode = file->f_mode;
1475 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1476 * to updated it before every ioctl.
1478 if (file->f_flags & O_NDELAY)
1479 mode |= FMODE_NDELAY;
1481 mode &= ~FMODE_NDELAY;
1483 return blkdev_ioctl(bdev, mode, cmd, arg);
1487 * Write data to the block device. Only intended for the block device itself
1488 * and the raw driver which basically is a fake block device.
1490 * Does not take i_mutex for the write and thus is not for general purpose
1493 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1494 unsigned long nr_segs, loff_t pos)
1496 struct file *file = iocb->ki_filp;
1499 BUG_ON(iocb->ki_pos != pos);
1501 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1502 if (ret > 0 || ret == -EIOCBQUEUED) {
1505 err = generic_write_sync(file, pos, ret);
1506 if (err < 0 && ret > 0)
1511 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1514 * Try to release a page associated with block device when the system
1515 * is under memory pressure.
1517 static int blkdev_releasepage(struct page *page, gfp_t wait)
1519 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1521 if (super && super->s_op->bdev_try_to_free_page)
1522 return super->s_op->bdev_try_to_free_page(super, page, wait);
1524 return try_to_free_buffers(page);
1527 static const struct address_space_operations def_blk_aops = {
1528 .readpage = blkdev_readpage,
1529 .writepage = blkdev_writepage,
1530 .sync_page = block_sync_page,
1531 .write_begin = blkdev_write_begin,
1532 .write_end = blkdev_write_end,
1533 .writepages = generic_writepages,
1534 .releasepage = blkdev_releasepage,
1535 .direct_IO = blkdev_direct_IO,
1538 const struct file_operations def_blk_fops = {
1539 .open = blkdev_open,
1540 .release = blkdev_close,
1541 .llseek = block_llseek,
1542 .read = do_sync_read,
1543 .write = do_sync_write,
1544 .aio_read = generic_file_aio_read,
1545 .aio_write = blkdev_aio_write,
1546 .mmap = generic_file_mmap,
1547 .fsync = blkdev_fsync,
1548 .unlocked_ioctl = block_ioctl,
1549 #ifdef CONFIG_COMPAT
1550 .compat_ioctl = compat_blkdev_ioctl,
1552 .splice_read = generic_file_splice_read,
1553 .splice_write = generic_file_splice_write,
1556 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1559 mm_segment_t old_fs = get_fs();
1561 res = blkdev_ioctl(bdev, 0, cmd, arg);
1566 EXPORT_SYMBOL(ioctl_by_bdev);
1569 * lookup_bdev - lookup a struct block_device by name
1570 * @pathname: special file representing the block device
1572 * Get a reference to the blockdevice at @pathname in the current
1573 * namespace if possible and return it. Return ERR_PTR(error)
1576 struct block_device *lookup_bdev(const char *pathname)
1578 struct block_device *bdev;
1579 struct inode *inode;
1583 if (!pathname || !*pathname)
1584 return ERR_PTR(-EINVAL);
1586 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1588 return ERR_PTR(error);
1590 inode = path.dentry->d_inode;
1592 if (!S_ISBLK(inode->i_mode))
1595 if (path.mnt->mnt_flags & MNT_NODEV)
1598 bdev = bd_acquire(inode);
1605 bdev = ERR_PTR(error);
1608 EXPORT_SYMBOL(lookup_bdev);
1610 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1612 struct super_block *sb = get_super(bdev);
1617 * no need to lock the super, get_super holds the
1618 * read mutex so the filesystem cannot go away
1619 * under us (->put_super runs with the write lock
1622 shrink_dcache_sb(sb);
1623 res = invalidate_inodes(sb, kill_dirty);
1626 invalidate_bdev(bdev);
1629 EXPORT_SYMBOL(__invalidate_device);