4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/buffer_head.h>
20 #include <linux/pagevec.h>
21 #include <linux/writeback.h>
22 #include <linux/mpage.h>
23 #include <linux/mount.h>
24 #include <linux/uio.h>
25 #include <linux/namei.h>
26 #include <linux/log2.h>
27 #include <linux/kmemleak.h>
28 #include <asm/uaccess.h>
32 struct block_device bdev;
33 struct inode vfs_inode;
36 static const struct address_space_operations def_blk_aops;
38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
40 return container_of(inode, struct bdev_inode, vfs_inode);
43 inline struct block_device *I_BDEV(struct inode *inode)
45 return &BDEV_I(inode)->bdev;
48 EXPORT_SYMBOL(I_BDEV);
51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
52 * we need to move it onto the dirty list of @dst so that the inode is always
55 static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
58 spin_lock(&inode_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);
435 inode_init_once(&ei->vfs_inode);
436 /* Initialize mutex for freeze. */
437 mutex_init(&bdev->bd_fsfreeze_mutex);
440 static inline void __bd_forget(struct inode *inode)
442 list_del_init(&inode->i_devices);
443 inode->i_bdev = NULL;
444 inode->i_mapping = &inode->i_data;
447 static void bdev_evict_inode(struct inode *inode)
449 struct block_device *bdev = &BDEV_I(inode)->bdev;
451 truncate_inode_pages(&inode->i_data, 0);
452 invalidate_inode_buffers(inode); /* is it needed here? */
453 end_writeback(inode);
454 spin_lock(&bdev_lock);
455 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
456 __bd_forget(list_entry(p, struct inode, i_devices));
458 list_del_init(&bdev->bd_list);
459 spin_unlock(&bdev_lock);
462 static const struct super_operations bdev_sops = {
463 .statfs = simple_statfs,
464 .alloc_inode = bdev_alloc_inode,
465 .destroy_inode = bdev_destroy_inode,
466 .drop_inode = generic_delete_inode,
467 .evict_inode = bdev_evict_inode,
470 static struct dentry *bd_mount(struct file_system_type *fs_type,
471 int flags, const char *dev_name, void *data)
473 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
476 static struct file_system_type bd_type = {
479 .kill_sb = kill_anon_super,
482 struct super_block *blockdev_superblock __read_mostly;
484 void __init bdev_cache_init(void)
487 struct vfsmount *bd_mnt;
489 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
490 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
491 SLAB_MEM_SPREAD|SLAB_PANIC),
493 err = register_filesystem(&bd_type);
495 panic("Cannot register bdev pseudo-fs");
496 bd_mnt = kern_mount(&bd_type);
498 panic("Cannot create bdev pseudo-fs");
500 * This vfsmount structure is only used to obtain the
501 * blockdev_superblock, so tell kmemleak not to report it.
503 kmemleak_not_leak(bd_mnt);
504 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
508 * Most likely _very_ bad one - but then it's hardly critical for small
509 * /dev and can be fixed when somebody will need really large one.
510 * Keep in mind that it will be fed through icache hash function too.
512 static inline unsigned long hash(dev_t dev)
514 return MAJOR(dev)+MINOR(dev);
517 static int bdev_test(struct inode *inode, void *data)
519 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
522 static int bdev_set(struct inode *inode, void *data)
524 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
528 static LIST_HEAD(all_bdevs);
530 struct block_device *bdget(dev_t dev)
532 struct block_device *bdev;
535 inode = iget5_locked(blockdev_superblock, hash(dev),
536 bdev_test, bdev_set, &dev);
541 bdev = &BDEV_I(inode)->bdev;
543 if (inode->i_state & I_NEW) {
544 bdev->bd_contains = NULL;
545 bdev->bd_inode = inode;
546 bdev->bd_block_size = (1 << inode->i_blkbits);
547 bdev->bd_part_count = 0;
548 bdev->bd_invalidated = 0;
549 inode->i_mode = S_IFBLK;
551 inode->i_bdev = bdev;
552 inode->i_data.a_ops = &def_blk_aops;
553 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
554 inode->i_data.backing_dev_info = &default_backing_dev_info;
555 spin_lock(&bdev_lock);
556 list_add(&bdev->bd_list, &all_bdevs);
557 spin_unlock(&bdev_lock);
558 unlock_new_inode(inode);
563 EXPORT_SYMBOL(bdget);
566 * bdgrab -- Grab a reference to an already referenced block device
567 * @bdev: Block device to grab a reference to.
569 struct block_device *bdgrab(struct block_device *bdev)
571 ihold(bdev->bd_inode);
575 long nr_blockdev_pages(void)
577 struct block_device *bdev;
579 spin_lock(&bdev_lock);
580 list_for_each_entry(bdev, &all_bdevs, bd_list) {
581 ret += bdev->bd_inode->i_mapping->nrpages;
583 spin_unlock(&bdev_lock);
587 void bdput(struct block_device *bdev)
589 iput(bdev->bd_inode);
592 EXPORT_SYMBOL(bdput);
594 static struct block_device *bd_acquire(struct inode *inode)
596 struct block_device *bdev;
598 spin_lock(&bdev_lock);
599 bdev = inode->i_bdev;
601 ihold(bdev->bd_inode);
602 spin_unlock(&bdev_lock);
605 spin_unlock(&bdev_lock);
607 bdev = bdget(inode->i_rdev);
609 spin_lock(&bdev_lock);
610 if (!inode->i_bdev) {
612 * We take an additional reference to bd_inode,
613 * and it's released in clear_inode() of inode.
614 * So, we can access it via ->i_mapping always
617 ihold(bdev->bd_inode);
618 inode->i_bdev = bdev;
619 inode->i_mapping = bdev->bd_inode->i_mapping;
620 list_add(&inode->i_devices, &bdev->bd_inodes);
622 spin_unlock(&bdev_lock);
627 /* Call when you free inode */
629 void bd_forget(struct inode *inode)
631 struct block_device *bdev = NULL;
633 spin_lock(&bdev_lock);
635 if (!sb_is_blkdev_sb(inode->i_sb))
636 bdev = inode->i_bdev;
639 spin_unlock(&bdev_lock);
642 iput(bdev->bd_inode);
646 * bd_may_claim - test whether a block device can be claimed
647 * @bdev: block device of interest
648 * @whole: whole block device containing @bdev, may equal @bdev
649 * @holder: holder trying to claim @bdev
651 * Test whther @bdev can be claimed by @holder.
654 * spin_lock(&bdev_lock).
657 * %true if @bdev can be claimed, %false otherwise.
659 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
662 if (bdev->bd_holder == holder)
663 return true; /* already a holder */
664 else if (bdev->bd_holder != NULL)
665 return false; /* held by someone else */
666 else if (bdev->bd_contains == bdev)
667 return true; /* is a whole device which isn't held */
669 else if (whole->bd_holder == bd_may_claim)
670 return true; /* is a partition of a device that is being partitioned */
671 else if (whole->bd_holder != NULL)
672 return false; /* is a partition of a held device */
674 return true; /* is a partition of an un-held device */
678 * bd_prepare_to_claim - prepare to claim a block device
679 * @bdev: block device of interest
680 * @whole: the whole device containing @bdev, may equal @bdev
681 * @holder: holder trying to claim @bdev
683 * Prepare to claim @bdev. This function fails if @bdev is already
684 * claimed by another holder and waits if another claiming is in
685 * progress. This function doesn't actually claim. On successful
686 * return, the caller has ownership of bd_claiming and bd_holder[s].
689 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
693 * 0 if @bdev can be claimed, -EBUSY otherwise.
695 static int bd_prepare_to_claim(struct block_device *bdev,
696 struct block_device *whole, void *holder)
699 /* if someone else claimed, fail */
700 if (!bd_may_claim(bdev, whole, holder))
703 /* if claiming is already in progress, wait for it to finish */
704 if (whole->bd_claiming) {
705 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
708 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
709 spin_unlock(&bdev_lock);
711 finish_wait(wq, &wait);
712 spin_lock(&bdev_lock);
721 * bd_start_claiming - start claiming a block device
722 * @bdev: block device of interest
723 * @holder: holder trying to claim @bdev
725 * @bdev is about to be opened exclusively. Check @bdev can be opened
726 * exclusively and mark that an exclusive open is in progress. Each
727 * successful call to this function must be matched with a call to
728 * either bd_finish_claiming() or bd_abort_claiming() (which do not
731 * This function is used to gain exclusive access to the block device
732 * without actually causing other exclusive open attempts to fail. It
733 * should be used when the open sequence itself requires exclusive
734 * access but may subsequently fail.
740 * Pointer to the block device containing @bdev on success, ERR_PTR()
743 static struct block_device *bd_start_claiming(struct block_device *bdev,
746 struct gendisk *disk;
747 struct block_device *whole;
753 * @bdev might not have been initialized properly yet, look up
754 * and grab the outer block device the hard way.
756 disk = get_gendisk(bdev->bd_dev, &partno);
758 return ERR_PTR(-ENXIO);
760 whole = bdget_disk(disk, 0);
761 module_put(disk->fops->owner);
764 return ERR_PTR(-ENOMEM);
766 /* prepare to claim, if successful, mark claiming in progress */
767 spin_lock(&bdev_lock);
769 err = bd_prepare_to_claim(bdev, whole, holder);
771 whole->bd_claiming = holder;
772 spin_unlock(&bdev_lock);
775 spin_unlock(&bdev_lock);
782 static int add_symlink(struct kobject *from, struct kobject *to)
784 return sysfs_create_link(from, to, kobject_name(to));
787 static void del_symlink(struct kobject *from, struct kobject *to)
789 sysfs_remove_link(from, kobject_name(to));
793 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
794 * @bdev: the claimed slave bdev
795 * @disk: the holding disk
797 * This functions creates the following sysfs symlinks.
799 * - from "slaves" directory of the holder @disk to the claimed @bdev
800 * - from "holders" directory of the @bdev to the holder @disk
802 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
803 * passed to bd_link_disk_holder(), then:
805 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
806 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
808 * The caller must have claimed @bdev before calling this function and
809 * ensure that both @bdev and @disk are valid during the creation and
810 * lifetime of these symlinks.
816 * 0 on success, -errno on failure.
818 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
822 mutex_lock(&bdev->bd_mutex);
824 WARN_ON_ONCE(!bdev->bd_holder || bdev->bd_holder_disk);
826 /* FIXME: remove the following once add_disk() handles errors */
827 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
830 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
834 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
836 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
840 bdev->bd_holder_disk = disk;
842 mutex_unlock(&bdev->bd_mutex);
845 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
847 static void bd_unlink_disk_holder(struct block_device *bdev)
849 struct gendisk *disk = bdev->bd_holder_disk;
851 bdev->bd_holder_disk = NULL;
855 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
856 del_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
859 static inline void bd_unlink_disk_holder(struct block_device *bdev)
864 * flush_disk - invalidates all buffer-cache entries on a disk
866 * @bdev: struct block device to be flushed
868 * Invalidates all buffer-cache entries on a disk. It should be called
869 * when a disk has been changed -- either by a media change or online
872 static void flush_disk(struct block_device *bdev)
874 if (__invalidate_device(bdev)) {
875 char name[BDEVNAME_SIZE] = "";
878 disk_name(bdev->bd_disk, 0, name);
879 printk(KERN_WARNING "VFS: busy inodes on changed media or "
880 "resized disk %s\n", name);
885 if (disk_partitionable(bdev->bd_disk))
886 bdev->bd_invalidated = 1;
890 * check_disk_size_change - checks for disk size change and adjusts bdev size.
891 * @disk: struct gendisk to check
892 * @bdev: struct bdev to adjust.
894 * This routine checks to see if the bdev size does not match the disk size
895 * and adjusts it if it differs.
897 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
899 loff_t disk_size, bdev_size;
901 disk_size = (loff_t)get_capacity(disk) << 9;
902 bdev_size = i_size_read(bdev->bd_inode);
903 if (disk_size != bdev_size) {
904 char name[BDEVNAME_SIZE];
906 disk_name(disk, 0, name);
908 "%s: detected capacity change from %lld to %lld\n",
909 name, bdev_size, disk_size);
910 i_size_write(bdev->bd_inode, disk_size);
914 EXPORT_SYMBOL(check_disk_size_change);
917 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
918 * @disk: struct gendisk to be revalidated
920 * This routine is a wrapper for lower-level driver's revalidate_disk
921 * call-backs. It is used to do common pre and post operations needed
922 * for all revalidate_disk operations.
924 int revalidate_disk(struct gendisk *disk)
926 struct block_device *bdev;
929 if (disk->fops->revalidate_disk)
930 ret = disk->fops->revalidate_disk(disk);
932 bdev = bdget_disk(disk, 0);
936 mutex_lock(&bdev->bd_mutex);
937 check_disk_size_change(disk, bdev);
938 mutex_unlock(&bdev->bd_mutex);
942 EXPORT_SYMBOL(revalidate_disk);
945 * This routine checks whether a removable media has been changed,
946 * and invalidates all buffer-cache-entries in that case. This
947 * is a relatively slow routine, so we have to try to minimize using
948 * it. Thus it is called only upon a 'mount' or 'open'. This
949 * is the best way of combining speed and utility, I think.
950 * People changing diskettes in the middle of an operation deserve
953 int check_disk_change(struct block_device *bdev)
955 struct gendisk *disk = bdev->bd_disk;
956 const struct block_device_operations *bdops = disk->fops;
959 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
960 DISK_EVENT_EJECT_REQUEST);
961 if (!(events & DISK_EVENT_MEDIA_CHANGE))
965 if (bdops->revalidate_disk)
966 bdops->revalidate_disk(bdev->bd_disk);
970 EXPORT_SYMBOL(check_disk_change);
972 void bd_set_size(struct block_device *bdev, loff_t size)
974 unsigned bsize = bdev_logical_block_size(bdev);
976 bdev->bd_inode->i_size = size;
977 while (bsize < PAGE_CACHE_SIZE) {
982 bdev->bd_block_size = bsize;
983 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
985 EXPORT_SYMBOL(bd_set_size);
987 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
992 * mutex_lock(part->bd_mutex)
993 * mutex_lock_nested(whole->bd_mutex, 1)
996 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
998 struct gendisk *disk;
1003 if (mode & FMODE_READ)
1005 if (mode & FMODE_WRITE)
1008 * hooks: /n/, see "layering violations".
1011 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1021 disk = get_gendisk(bdev->bd_dev, &partno);
1025 mutex_lock_nested(&bdev->bd_mutex, for_part);
1026 if (!bdev->bd_openers) {
1027 bdev->bd_disk = disk;
1028 bdev->bd_contains = bdev;
1030 struct backing_dev_info *bdi;
1033 bdev->bd_part = disk_get_part(disk, partno);
1037 if (disk->fops->open) {
1038 ret = disk->fops->open(bdev, mode);
1039 if (ret == -ERESTARTSYS) {
1040 /* Lost a race with 'disk' being
1041 * deleted, try again.
1044 disk_put_part(bdev->bd_part);
1045 bdev->bd_part = NULL;
1046 module_put(disk->fops->owner);
1048 bdev->bd_disk = NULL;
1049 mutex_unlock(&bdev->bd_mutex);
1055 if (!bdev->bd_openers) {
1056 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1057 bdi = blk_get_backing_dev_info(bdev);
1059 bdi = &default_backing_dev_info;
1060 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1062 if (bdev->bd_invalidated)
1063 rescan_partitions(disk, bdev);
1065 struct block_device *whole;
1066 whole = bdget_disk(disk, 0);
1071 ret = __blkdev_get(whole, mode, 1);
1074 bdev->bd_contains = whole;
1075 bdev_inode_switch_bdi(bdev->bd_inode,
1076 whole->bd_inode->i_data.backing_dev_info);
1077 bdev->bd_part = disk_get_part(disk, partno);
1078 if (!(disk->flags & GENHD_FL_UP) ||
1079 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1083 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1086 module_put(disk->fops->owner);
1089 if (bdev->bd_contains == bdev) {
1090 if (bdev->bd_disk->fops->open) {
1091 ret = bdev->bd_disk->fops->open(bdev, mode);
1093 goto out_unlock_bdev;
1095 if (bdev->bd_invalidated)
1096 rescan_partitions(bdev->bd_disk, bdev);
1101 bdev->bd_part_count++;
1102 mutex_unlock(&bdev->bd_mutex);
1106 disk_put_part(bdev->bd_part);
1107 bdev->bd_disk = NULL;
1108 bdev->bd_part = NULL;
1109 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1110 if (bdev != bdev->bd_contains)
1111 __blkdev_put(bdev->bd_contains, mode, 1);
1112 bdev->bd_contains = NULL;
1114 mutex_unlock(&bdev->bd_mutex);
1117 module_put(disk->fops->owner);
1125 * blkdev_get - open a block device
1126 * @bdev: block_device to open
1127 * @mode: FMODE_* mask
1128 * @holder: exclusive holder identifier
1130 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1131 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1132 * @holder is invalid. Exclusive opens may nest for the same @holder.
1134 * On success, the reference count of @bdev is unchanged. On failure,
1141 * 0 on success, -errno on failure.
1143 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1145 struct block_device *whole = NULL;
1148 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1150 if ((mode & FMODE_EXCL) && holder) {
1151 whole = bd_start_claiming(bdev, holder);
1152 if (IS_ERR(whole)) {
1154 return PTR_ERR(whole);
1158 res = __blkdev_get(bdev, mode, 0);
1160 /* __blkdev_get() may alter read only status, check it afterwards */
1161 if (!res && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1162 __blkdev_put(bdev, mode, 0);
1167 /* finish claiming */
1168 mutex_lock(&bdev->bd_mutex);
1169 spin_lock(&bdev_lock);
1172 BUG_ON(!bd_may_claim(bdev, whole, holder));
1174 * Note that for a whole device bd_holders
1175 * will be incremented twice, and bd_holder
1176 * will be set to bd_may_claim before being
1179 whole->bd_holders++;
1180 whole->bd_holder = bd_may_claim;
1182 bdev->bd_holder = holder;
1185 /* tell others that we're done */
1186 BUG_ON(whole->bd_claiming != holder);
1187 whole->bd_claiming = NULL;
1188 wake_up_bit(&whole->bd_claiming, 0);
1190 spin_unlock(&bdev_lock);
1193 * Block event polling for write claims. Any write
1194 * holder makes the write_holder state stick until all
1195 * are released. This is good enough and tracking
1196 * individual writeable reference is too fragile given
1197 * the way @mode is used in blkdev_get/put().
1199 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
1200 bdev->bd_write_holder = true;
1201 disk_block_events(bdev->bd_disk);
1204 mutex_unlock(&bdev->bd_mutex);
1210 EXPORT_SYMBOL(blkdev_get);
1213 * blkdev_get_by_path - open a block device by name
1214 * @path: path to the block device to open
1215 * @mode: FMODE_* mask
1216 * @holder: exclusive holder identifier
1218 * Open the blockdevice described by the device file at @path. @mode
1219 * and @holder are identical to blkdev_get().
1221 * On success, the returned block_device has reference count of one.
1227 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1229 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1232 struct block_device *bdev;
1235 bdev = lookup_bdev(path);
1239 err = blkdev_get(bdev, mode, holder);
1241 return ERR_PTR(err);
1245 EXPORT_SYMBOL(blkdev_get_by_path);
1248 * blkdev_get_by_dev - open a block device by device number
1249 * @dev: device number of block device to open
1250 * @mode: FMODE_* mask
1251 * @holder: exclusive holder identifier
1253 * Open the blockdevice described by device number @dev. @mode and
1254 * @holder are identical to blkdev_get().
1256 * Use it ONLY if you really do not have anything better - i.e. when
1257 * you are behind a truly sucky interface and all you are given is a
1258 * device number. _Never_ to be used for internal purposes. If you
1259 * ever need it - reconsider your API.
1261 * On success, the returned block_device has reference count of one.
1267 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1269 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1271 struct block_device *bdev;
1276 return ERR_PTR(-ENOMEM);
1278 err = blkdev_get(bdev, mode, holder);
1280 return ERR_PTR(err);
1284 EXPORT_SYMBOL(blkdev_get_by_dev);
1286 static int blkdev_open(struct inode * inode, struct file * filp)
1288 struct block_device *bdev;
1291 * Preserve backwards compatibility and allow large file access
1292 * even if userspace doesn't ask for it explicitly. Some mkfs
1293 * binary needs it. We might want to drop this workaround
1294 * during an unstable branch.
1296 filp->f_flags |= O_LARGEFILE;
1298 if (filp->f_flags & O_NDELAY)
1299 filp->f_mode |= FMODE_NDELAY;
1300 if (filp->f_flags & O_EXCL)
1301 filp->f_mode |= FMODE_EXCL;
1302 if ((filp->f_flags & O_ACCMODE) == 3)
1303 filp->f_mode |= FMODE_WRITE_IOCTL;
1305 bdev = bd_acquire(inode);
1309 filp->f_mapping = bdev->bd_inode->i_mapping;
1311 return blkdev_get(bdev, filp->f_mode, filp);
1314 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1317 struct gendisk *disk = bdev->bd_disk;
1318 struct block_device *victim = NULL;
1320 mutex_lock_nested(&bdev->bd_mutex, for_part);
1322 bdev->bd_part_count--;
1324 if (!--bdev->bd_openers) {
1325 WARN_ON_ONCE(bdev->bd_holders);
1326 sync_blockdev(bdev);
1329 if (bdev->bd_contains == bdev) {
1330 if (disk->fops->release)
1331 ret = disk->fops->release(disk, mode);
1333 if (!bdev->bd_openers) {
1334 struct module *owner = disk->fops->owner;
1338 disk_put_part(bdev->bd_part);
1339 bdev->bd_part = NULL;
1340 bdev->bd_disk = NULL;
1341 bdev_inode_switch_bdi(bdev->bd_inode,
1342 &default_backing_dev_info);
1343 if (bdev != bdev->bd_contains)
1344 victim = bdev->bd_contains;
1345 bdev->bd_contains = NULL;
1347 mutex_unlock(&bdev->bd_mutex);
1350 __blkdev_put(victim, mode, 1);
1354 int blkdev_put(struct block_device *bdev, fmode_t mode)
1356 if (mode & FMODE_EXCL) {
1360 * Release a claim on the device. The holder fields
1361 * are protected with bdev_lock. bd_mutex is to
1362 * synchronize disk_holder unlinking.
1364 mutex_lock(&bdev->bd_mutex);
1365 spin_lock(&bdev_lock);
1367 WARN_ON_ONCE(--bdev->bd_holders < 0);
1368 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1370 /* bd_contains might point to self, check in a separate step */
1371 if ((bdev_free = !bdev->bd_holders))
1372 bdev->bd_holder = NULL;
1373 if (!bdev->bd_contains->bd_holders)
1374 bdev->bd_contains->bd_holder = NULL;
1376 spin_unlock(&bdev_lock);
1379 * If this was the last claim, remove holder link and
1380 * unblock evpoll if it was a write holder.
1383 bd_unlink_disk_holder(bdev);
1384 if (bdev->bd_write_holder) {
1385 disk_unblock_events(bdev->bd_disk);
1386 bdev->bd_write_holder = false;
1388 disk_check_events(bdev->bd_disk);
1391 mutex_unlock(&bdev->bd_mutex);
1393 disk_check_events(bdev->bd_disk);
1395 return __blkdev_put(bdev, mode, 0);
1397 EXPORT_SYMBOL(blkdev_put);
1399 static int blkdev_close(struct inode * inode, struct file * filp)
1401 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1403 return blkdev_put(bdev, filp->f_mode);
1406 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1408 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1409 fmode_t mode = file->f_mode;
1412 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1413 * to updated it before every ioctl.
1415 if (file->f_flags & O_NDELAY)
1416 mode |= FMODE_NDELAY;
1418 mode &= ~FMODE_NDELAY;
1420 return blkdev_ioctl(bdev, mode, cmd, arg);
1424 * Write data to the block device. Only intended for the block device itself
1425 * and the raw driver which basically is a fake block device.
1427 * Does not take i_mutex for the write and thus is not for general purpose
1430 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1431 unsigned long nr_segs, loff_t pos)
1433 struct file *file = iocb->ki_filp;
1436 BUG_ON(iocb->ki_pos != pos);
1438 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1439 if (ret > 0 || ret == -EIOCBQUEUED) {
1442 err = generic_write_sync(file, pos, ret);
1443 if (err < 0 && ret > 0)
1448 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1451 * Try to release a page associated with block device when the system
1452 * is under memory pressure.
1454 static int blkdev_releasepage(struct page *page, gfp_t wait)
1456 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1458 if (super && super->s_op->bdev_try_to_free_page)
1459 return super->s_op->bdev_try_to_free_page(super, page, wait);
1461 return try_to_free_buffers(page);
1464 static const struct address_space_operations def_blk_aops = {
1465 .readpage = blkdev_readpage,
1466 .writepage = blkdev_writepage,
1467 .sync_page = block_sync_page,
1468 .write_begin = blkdev_write_begin,
1469 .write_end = blkdev_write_end,
1470 .writepages = generic_writepages,
1471 .releasepage = blkdev_releasepage,
1472 .direct_IO = blkdev_direct_IO,
1475 const struct file_operations def_blk_fops = {
1476 .open = blkdev_open,
1477 .release = blkdev_close,
1478 .llseek = block_llseek,
1479 .read = do_sync_read,
1480 .write = do_sync_write,
1481 .aio_read = generic_file_aio_read,
1482 .aio_write = blkdev_aio_write,
1483 .mmap = generic_file_mmap,
1484 .fsync = blkdev_fsync,
1485 .unlocked_ioctl = block_ioctl,
1486 #ifdef CONFIG_COMPAT
1487 .compat_ioctl = compat_blkdev_ioctl,
1489 .splice_read = generic_file_splice_read,
1490 .splice_write = generic_file_splice_write,
1493 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1496 mm_segment_t old_fs = get_fs();
1498 res = blkdev_ioctl(bdev, 0, cmd, arg);
1503 EXPORT_SYMBOL(ioctl_by_bdev);
1506 * lookup_bdev - lookup a struct block_device by name
1507 * @pathname: special file representing the block device
1509 * Get a reference to the blockdevice at @pathname in the current
1510 * namespace if possible and return it. Return ERR_PTR(error)
1513 struct block_device *lookup_bdev(const char *pathname)
1515 struct block_device *bdev;
1516 struct inode *inode;
1520 if (!pathname || !*pathname)
1521 return ERR_PTR(-EINVAL);
1523 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1525 return ERR_PTR(error);
1527 inode = path.dentry->d_inode;
1529 if (!S_ISBLK(inode->i_mode))
1532 if (path.mnt->mnt_flags & MNT_NODEV)
1535 bdev = bd_acquire(inode);
1542 bdev = ERR_PTR(error);
1545 EXPORT_SYMBOL(lookup_bdev);
1547 int __invalidate_device(struct block_device *bdev)
1549 struct super_block *sb = get_super(bdev);
1554 * no need to lock the super, get_super holds the
1555 * read mutex so the filesystem cannot go away
1556 * under us (->put_super runs with the write lock
1559 shrink_dcache_sb(sb);
1560 res = invalidate_inodes(sb);
1563 invalidate_bdev(bdev);
1566 EXPORT_SYMBOL(__invalidate_device);