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/smp_lock.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
51 static sector_t max_block(struct block_device *bdev)
53 sector_t retval = ~((sector_t)0);
54 loff_t sz = i_size_read(bdev->bd_inode);
57 unsigned int size = block_size(bdev);
58 unsigned int sizebits = blksize_bits(size);
59 retval = (sz >> sizebits);
64 /* Kill _all_ buffers and pagecache , dirty or not.. */
65 static void kill_bdev(struct block_device *bdev)
67 if (bdev->bd_inode->i_mapping->nrpages == 0)
70 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
73 int set_blocksize(struct block_device *bdev, int size)
75 /* Size must be a power of two, and between 512 and PAGE_SIZE */
76 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
79 /* Size cannot be smaller than the size supported by the device */
80 if (size < bdev_logical_block_size(bdev))
83 /* Don't change the size if it is same as current */
84 if (bdev->bd_block_size != size) {
86 bdev->bd_block_size = size;
87 bdev->bd_inode->i_blkbits = blksize_bits(size);
93 EXPORT_SYMBOL(set_blocksize);
95 int sb_set_blocksize(struct super_block *sb, int size)
97 if (set_blocksize(sb->s_bdev, size))
99 /* If we get here, we know size is power of two
100 * and it's value is between 512 and PAGE_SIZE */
101 sb->s_blocksize = size;
102 sb->s_blocksize_bits = blksize_bits(size);
103 return sb->s_blocksize;
106 EXPORT_SYMBOL(sb_set_blocksize);
108 int sb_min_blocksize(struct super_block *sb, int size)
110 int minsize = bdev_logical_block_size(sb->s_bdev);
113 return sb_set_blocksize(sb, size);
116 EXPORT_SYMBOL(sb_min_blocksize);
119 blkdev_get_block(struct inode *inode, sector_t iblock,
120 struct buffer_head *bh, int create)
122 if (iblock >= max_block(I_BDEV(inode))) {
127 * for reads, we're just trying to fill a partial page.
128 * return a hole, they will have to call get_block again
129 * before they can fill it, and they will get -EIO at that
134 bh->b_bdev = I_BDEV(inode);
135 bh->b_blocknr = iblock;
136 set_buffer_mapped(bh);
141 blkdev_get_blocks(struct inode *inode, sector_t iblock,
142 struct buffer_head *bh, int create)
144 sector_t end_block = max_block(I_BDEV(inode));
145 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
147 if ((iblock + max_blocks) > end_block) {
148 max_blocks = end_block - iblock;
149 if ((long)max_blocks <= 0) {
151 return -EIO; /* write fully beyond EOF */
153 * It is a read which is fully beyond EOF. We return
154 * a !buffer_mapped buffer
160 bh->b_bdev = I_BDEV(inode);
161 bh->b_blocknr = iblock;
162 bh->b_size = max_blocks << inode->i_blkbits;
164 set_buffer_mapped(bh);
169 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
170 loff_t offset, unsigned long nr_segs)
172 struct file *file = iocb->ki_filp;
173 struct inode *inode = file->f_mapping->host;
175 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
176 iov, offset, nr_segs, blkdev_get_blocks, NULL);
179 int __sync_blockdev(struct block_device *bdev, int wait)
184 return filemap_flush(bdev->bd_inode->i_mapping);
185 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
189 * Write out and wait upon all the dirty data associated with a block
190 * device via its mapping. Does not take the superblock lock.
192 int sync_blockdev(struct block_device *bdev)
194 return __sync_blockdev(bdev, 1);
196 EXPORT_SYMBOL(sync_blockdev);
199 * Write out and wait upon all dirty data associated with this
200 * device. Filesystem data as well as the underlying block
201 * device. Takes the superblock lock.
203 int fsync_bdev(struct block_device *bdev)
205 struct super_block *sb = get_super(bdev);
207 int res = sync_filesystem(sb);
211 return sync_blockdev(bdev);
213 EXPORT_SYMBOL(fsync_bdev);
216 * freeze_bdev -- lock a filesystem and force it into a consistent state
217 * @bdev: blockdevice to lock
219 * If a superblock is found on this device, we take the s_umount semaphore
220 * on it to make sure nobody unmounts until the snapshot creation is done.
221 * The reference counter (bd_fsfreeze_count) guarantees that only the last
222 * unfreeze process can unfreeze the frozen filesystem actually when multiple
223 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
224 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
227 struct super_block *freeze_bdev(struct block_device *bdev)
229 struct super_block *sb;
232 mutex_lock(&bdev->bd_fsfreeze_mutex);
233 if (++bdev->bd_fsfreeze_count > 1) {
235 * We don't even need to grab a reference - the first call
236 * to freeze_bdev grab an active reference and only the last
237 * thaw_bdev drops it.
239 sb = get_super(bdev);
241 mutex_unlock(&bdev->bd_fsfreeze_mutex);
245 sb = get_active_super(bdev);
248 if (sb->s_flags & MS_RDONLY) {
249 sb->s_frozen = SB_FREEZE_TRANS;
250 up_write(&sb->s_umount);
251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
255 sb->s_frozen = SB_FREEZE_WRITE;
260 sb->s_frozen = SB_FREEZE_TRANS;
263 sync_blockdev(sb->s_bdev);
265 if (sb->s_op->freeze_fs) {
266 error = sb->s_op->freeze_fs(sb);
269 "VFS:Filesystem freeze failed\n");
270 sb->s_frozen = SB_UNFROZEN;
271 deactivate_locked_super(sb);
272 bdev->bd_fsfreeze_count--;
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
274 return ERR_PTR(error);
277 up_write(&sb->s_umount);
281 mutex_unlock(&bdev->bd_fsfreeze_mutex);
282 return sb; /* thaw_bdev releases s->s_umount */
284 EXPORT_SYMBOL(freeze_bdev);
287 * thaw_bdev -- unlock filesystem
288 * @bdev: blockdevice to unlock
289 * @sb: associated superblock
291 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
293 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
297 mutex_lock(&bdev->bd_fsfreeze_mutex);
298 if (!bdev->bd_fsfreeze_count)
302 if (--bdev->bd_fsfreeze_count > 0)
308 BUG_ON(sb->s_bdev != bdev);
309 down_write(&sb->s_umount);
310 if (sb->s_flags & MS_RDONLY)
313 if (sb->s_op->unfreeze_fs) {
314 error = sb->s_op->unfreeze_fs(sb);
317 "VFS:Filesystem thaw failed\n");
318 sb->s_frozen = SB_FREEZE_TRANS;
319 bdev->bd_fsfreeze_count++;
320 mutex_unlock(&bdev->bd_fsfreeze_mutex);
326 sb->s_frozen = SB_UNFROZEN;
328 wake_up(&sb->s_wait_unfrozen);
331 deactivate_locked_super(sb);
333 mutex_unlock(&bdev->bd_fsfreeze_mutex);
336 EXPORT_SYMBOL(thaw_bdev);
338 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
340 return block_write_full_page(page, blkdev_get_block, wbc);
343 static int blkdev_readpage(struct file * file, struct page * page)
345 return block_read_full_page(page, blkdev_get_block);
348 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
353 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
357 static int blkdev_write_end(struct file *file, struct address_space *mapping,
358 loff_t pos, unsigned len, unsigned copied,
359 struct page *page, void *fsdata)
362 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
365 page_cache_release(page);
372 * for a block special file file->f_path.dentry->d_inode->i_size is zero
373 * so we compute the size by hand (just as in block_read/write above)
375 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
377 struct inode *bd_inode = file->f_mapping->host;
381 mutex_lock(&bd_inode->i_mutex);
382 size = i_size_read(bd_inode);
389 offset += file->f_pos;
392 if (offset >= 0 && offset <= size) {
393 if (offset != file->f_pos) {
394 file->f_pos = offset;
398 mutex_unlock(&bd_inode->i_mutex);
403 * Filp is never NULL; the only case when ->fsync() is called with
404 * NULL first argument is nfsd_sync_dir() and that's not a directory.
407 int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync)
409 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
412 error = sync_blockdev(bdev);
416 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL,
418 if (error == -EOPNOTSUPP)
422 EXPORT_SYMBOL(blkdev_fsync);
428 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
429 static struct kmem_cache * bdev_cachep __read_mostly;
431 static struct inode *bdev_alloc_inode(struct super_block *sb)
433 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
436 return &ei->vfs_inode;
439 static void bdev_destroy_inode(struct inode *inode)
441 struct bdev_inode *bdi = BDEV_I(inode);
443 kmem_cache_free(bdev_cachep, bdi);
446 static void init_once(void *foo)
448 struct bdev_inode *ei = (struct bdev_inode *) foo;
449 struct block_device *bdev = &ei->bdev;
451 memset(bdev, 0, sizeof(*bdev));
452 mutex_init(&bdev->bd_mutex);
453 INIT_LIST_HEAD(&bdev->bd_inodes);
454 INIT_LIST_HEAD(&bdev->bd_list);
456 INIT_LIST_HEAD(&bdev->bd_holder_list);
458 inode_init_once(&ei->vfs_inode);
459 /* Initialize mutex for freeze. */
460 mutex_init(&bdev->bd_fsfreeze_mutex);
463 static inline void __bd_forget(struct inode *inode)
465 list_del_init(&inode->i_devices);
466 inode->i_bdev = NULL;
467 inode->i_mapping = &inode->i_data;
470 static void bdev_clear_inode(struct inode *inode)
472 struct block_device *bdev = &BDEV_I(inode)->bdev;
474 spin_lock(&bdev_lock);
475 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
476 __bd_forget(list_entry(p, struct inode, i_devices));
478 list_del_init(&bdev->bd_list);
479 spin_unlock(&bdev_lock);
482 static const struct super_operations bdev_sops = {
483 .statfs = simple_statfs,
484 .alloc_inode = bdev_alloc_inode,
485 .destroy_inode = bdev_destroy_inode,
486 .drop_inode = generic_delete_inode,
487 .clear_inode = bdev_clear_inode,
490 static int bd_get_sb(struct file_system_type *fs_type,
491 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
493 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
496 static struct file_system_type bd_type = {
499 .kill_sb = kill_anon_super,
502 struct super_block *blockdev_superblock __read_mostly;
504 void __init bdev_cache_init(void)
507 struct vfsmount *bd_mnt;
509 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
510 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
511 SLAB_MEM_SPREAD|SLAB_PANIC),
513 err = register_filesystem(&bd_type);
515 panic("Cannot register bdev pseudo-fs");
516 bd_mnt = kern_mount(&bd_type);
518 panic("Cannot create bdev pseudo-fs");
520 * This vfsmount structure is only used to obtain the
521 * blockdev_superblock, so tell kmemleak not to report it.
523 kmemleak_not_leak(bd_mnt);
524 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
528 * Most likely _very_ bad one - but then it's hardly critical for small
529 * /dev and can be fixed when somebody will need really large one.
530 * Keep in mind that it will be fed through icache hash function too.
532 static inline unsigned long hash(dev_t dev)
534 return MAJOR(dev)+MINOR(dev);
537 static int bdev_test(struct inode *inode, void *data)
539 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
542 static int bdev_set(struct inode *inode, void *data)
544 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
548 static LIST_HEAD(all_bdevs);
550 struct block_device *bdget(dev_t dev)
552 struct block_device *bdev;
555 inode = iget5_locked(blockdev_superblock, hash(dev),
556 bdev_test, bdev_set, &dev);
561 bdev = &BDEV_I(inode)->bdev;
563 if (inode->i_state & I_NEW) {
564 bdev->bd_contains = NULL;
565 bdev->bd_inode = inode;
566 bdev->bd_block_size = (1 << inode->i_blkbits);
567 bdev->bd_part_count = 0;
568 bdev->bd_invalidated = 0;
569 inode->i_mode = S_IFBLK;
571 inode->i_bdev = bdev;
572 inode->i_data.a_ops = &def_blk_aops;
573 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
574 inode->i_data.backing_dev_info = &default_backing_dev_info;
575 spin_lock(&bdev_lock);
576 list_add(&bdev->bd_list, &all_bdevs);
577 spin_unlock(&bdev_lock);
578 unlock_new_inode(inode);
583 EXPORT_SYMBOL(bdget);
586 * bdgrab -- Grab a reference to an already referenced block device
587 * @bdev: Block device to grab a reference to.
589 struct block_device *bdgrab(struct block_device *bdev)
591 atomic_inc(&bdev->bd_inode->i_count);
595 long nr_blockdev_pages(void)
597 struct block_device *bdev;
599 spin_lock(&bdev_lock);
600 list_for_each_entry(bdev, &all_bdevs, bd_list) {
601 ret += bdev->bd_inode->i_mapping->nrpages;
603 spin_unlock(&bdev_lock);
607 void bdput(struct block_device *bdev)
609 iput(bdev->bd_inode);
612 EXPORT_SYMBOL(bdput);
614 static struct block_device *bd_acquire(struct inode *inode)
616 struct block_device *bdev;
618 spin_lock(&bdev_lock);
619 bdev = inode->i_bdev;
621 atomic_inc(&bdev->bd_inode->i_count);
622 spin_unlock(&bdev_lock);
625 spin_unlock(&bdev_lock);
627 bdev = bdget(inode->i_rdev);
629 spin_lock(&bdev_lock);
630 if (!inode->i_bdev) {
632 * We take an additional bd_inode->i_count for inode,
633 * and it's released in clear_inode() of inode.
634 * So, we can access it via ->i_mapping always
637 atomic_inc(&bdev->bd_inode->i_count);
638 inode->i_bdev = bdev;
639 inode->i_mapping = bdev->bd_inode->i_mapping;
640 list_add(&inode->i_devices, &bdev->bd_inodes);
642 spin_unlock(&bdev_lock);
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 whther @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_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 someone else is claiming, wait for it to finish */
724 if (whole->bd_claiming && whole->bd_claiming != holder) {
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_claim() or bd_abort_claiming(). If this function
749 * succeeds, the matching bd_claim() is guaranteed to succeed.
755 * Pointer to the block device containing @bdev on success, ERR_PTR()
758 static struct block_device *bd_start_claiming(struct block_device *bdev,
761 struct gendisk *disk;
762 struct block_device *whole;
768 * @bdev might not have been initialized properly yet, look up
769 * and grab the outer block device the hard way.
771 disk = get_gendisk(bdev->bd_dev, &partno);
773 return ERR_PTR(-ENXIO);
775 whole = bdget_disk(disk, 0);
778 return ERR_PTR(-ENOMEM);
780 /* prepare to claim, if successful, mark claiming in progress */
781 spin_lock(&bdev_lock);
783 err = bd_prepare_to_claim(bdev, whole, holder);
785 whole->bd_claiming = holder;
786 spin_unlock(&bdev_lock);
789 spin_unlock(&bdev_lock);
795 /* releases bdev_lock */
796 static void __bd_abort_claiming(struct block_device *whole, void *holder)
798 BUG_ON(whole->bd_claiming != holder);
799 whole->bd_claiming = NULL;
800 wake_up_bit(&whole->bd_claiming, 0);
802 spin_unlock(&bdev_lock);
807 * bd_abort_claiming - abort claiming a block device
808 * @whole: whole block device returned by bd_start_claiming()
809 * @holder: holder trying to claim @bdev
811 * Abort a claiming block started by bd_start_claiming(). Note that
812 * @whole is not the block device to be claimed but the whole device
813 * returned by bd_start_claiming().
816 * Grabs and releases bdev_lock.
818 static void bd_abort_claiming(struct block_device *whole, void *holder)
820 spin_lock(&bdev_lock);
821 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
825 * bd_claim - claim a block device
826 * @bdev: block device to claim
827 * @holder: holder trying to claim @bdev
829 * Try to claim @bdev which must have been opened successfully. This
830 * function may be called with or without preceding
831 * blk_start_claiming(). In the former case, this function is always
832 * successful and terminates the claiming block.
838 * 0 if successful, -EBUSY if @bdev is already claimed.
840 int bd_claim(struct block_device *bdev, void *holder)
842 struct block_device *whole = bdev->bd_contains;
847 spin_lock(&bdev_lock);
849 res = bd_prepare_to_claim(bdev, whole, holder);
851 /* note that for a whole device bd_holders
852 * will be incremented twice, and bd_holder will
853 * be set to bd_claim before being set to holder
856 whole->bd_holder = bd_claim;
858 bdev->bd_holder = holder;
861 if (whole->bd_claiming)
862 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
864 spin_unlock(&bdev_lock);
868 EXPORT_SYMBOL(bd_claim);
870 void bd_release(struct block_device *bdev)
872 spin_lock(&bdev_lock);
873 if (!--bdev->bd_contains->bd_holders)
874 bdev->bd_contains->bd_holder = NULL;
875 if (!--bdev->bd_holders)
876 bdev->bd_holder = NULL;
877 spin_unlock(&bdev_lock);
880 EXPORT_SYMBOL(bd_release);
884 * Functions for bd_claim_by_kobject / bd_release_from_kobject
886 * If a kobject is passed to bd_claim_by_kobject()
887 * and the kobject has a parent directory,
888 * following symlinks are created:
889 * o from the kobject to the claimed bdev
890 * o from "holders" directory of the bdev to the parent of the kobject
891 * bd_release_from_kobject() removes these symlinks.
894 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
895 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
896 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
897 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
900 static int add_symlink(struct kobject *from, struct kobject *to)
904 return sysfs_create_link(from, to, kobject_name(to));
907 static void del_symlink(struct kobject *from, struct kobject *to)
911 sysfs_remove_link(from, kobject_name(to));
915 * 'struct bd_holder' contains pointers to kobjects symlinked by
916 * bd_claim_by_kobject.
917 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
920 struct list_head list; /* chain of holders of the bdev */
921 int count; /* references from the holder */
922 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
923 struct kobject *hdev; /* e.g. "/block/dm-0" */
924 struct kobject *hdir; /* e.g. "/block/sda/holders" */
925 struct kobject *sdev; /* e.g. "/block/sda" */
929 * Get references of related kobjects at once.
930 * Returns 1 on success. 0 on failure.
932 * Should call bd_holder_release_dirs() after successful use.
934 static int bd_holder_grab_dirs(struct block_device *bdev,
935 struct bd_holder *bo)
940 bo->sdir = kobject_get(bo->sdir);
944 bo->hdev = kobject_get(bo->sdir->parent);
948 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
952 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
959 kobject_put(bo->sdev);
961 kobject_put(bo->hdev);
963 kobject_put(bo->sdir);
968 /* Put references of related kobjects at once. */
969 static void bd_holder_release_dirs(struct bd_holder *bo)
971 kobject_put(bo->hdir);
972 kobject_put(bo->sdev);
973 kobject_put(bo->hdev);
974 kobject_put(bo->sdir);
977 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
979 struct bd_holder *bo;
981 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
991 static void free_bd_holder(struct bd_holder *bo)
997 * find_bd_holder - find matching struct bd_holder from the block device
999 * @bdev: struct block device to be searched
1000 * @bo: target struct bd_holder
1002 * Returns matching entry with @bo in @bdev->bd_holder_list.
1003 * If found, increment the reference count and return the pointer.
1004 * If not found, returns NULL.
1006 static struct bd_holder *find_bd_holder(struct block_device *bdev,
1007 struct bd_holder *bo)
1009 struct bd_holder *tmp;
1011 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
1012 if (tmp->sdir == bo->sdir) {
1021 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1023 * @bdev: block device to be bd_claimed
1024 * @bo: preallocated and initialized by alloc_bd_holder()
1026 * Add @bo to @bdev->bd_holder_list, create symlinks.
1028 * Returns 0 if symlinks are created.
1029 * Returns -ve if something fails.
1031 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1038 if (!bd_holder_grab_dirs(bdev, bo))
1041 err = add_symlink(bo->sdir, bo->sdev);
1045 err = add_symlink(bo->hdir, bo->hdev);
1047 del_symlink(bo->sdir, bo->sdev);
1051 list_add_tail(&bo->list, &bdev->bd_holder_list);
1056 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1058 * @bdev: block device to be bd_claimed
1059 * @kobj: holder's kobject
1061 * If there is matching entry with @kobj in @bdev->bd_holder_list
1062 * and no other bd_claim() from the same kobject,
1063 * remove the struct bd_holder from the list, delete symlinks for it.
1065 * Returns a pointer to the struct bd_holder when it's removed from the list
1066 * and ready to be freed.
1067 * Returns NULL if matching claim isn't found or there is other bd_claim()
1068 * by the same kobject.
1070 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1071 struct kobject *kobj)
1073 struct bd_holder *bo;
1075 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1076 if (bo->sdir == kobj) {
1078 BUG_ON(bo->count < 0);
1080 list_del(&bo->list);
1081 del_symlink(bo->sdir, bo->sdev);
1082 del_symlink(bo->hdir, bo->hdev);
1083 bd_holder_release_dirs(bo);
1094 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1096 * @bdev: block device to be claimed
1097 * @holder: holder's signature
1098 * @kobj: holder's kobject
1100 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1101 * the bdev and the holder's kobject.
1102 * Use bd_release_from_kobject() when relesing the claimed bdev.
1104 * Returns 0 on success. (same as bd_claim())
1105 * Returns errno on failure.
1107 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1108 struct kobject *kobj)
1111 struct bd_holder *bo, *found;
1116 bo = alloc_bd_holder(kobj);
1120 mutex_lock(&bdev->bd_mutex);
1122 err = bd_claim(bdev, holder);
1126 found = find_bd_holder(bdev, bo);
1130 err = add_bd_holder(bdev, bo);
1136 mutex_unlock(&bdev->bd_mutex);
1142 * bd_release_from_kobject - bd_release() with additional kobject signature
1144 * @bdev: block device to be released
1145 * @kobj: holder's kobject
1147 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1149 static void bd_release_from_kobject(struct block_device *bdev,
1150 struct kobject *kobj)
1155 mutex_lock(&bdev->bd_mutex);
1157 free_bd_holder(del_bd_holder(bdev, kobj));
1158 mutex_unlock(&bdev->bd_mutex);
1162 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1164 * @bdev: block device to be claimed
1165 * @holder: holder's signature
1166 * @disk: holder's gendisk
1168 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1170 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1171 struct gendisk *disk)
1173 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1175 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1178 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1180 * @bdev: block device to be claimed
1181 * @disk: holder's gendisk
1183 * Call bd_release_from_kobject() and put @disk->slave_dir.
1185 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1187 bd_release_from_kobject(bdev, disk->slave_dir);
1188 kobject_put(disk->slave_dir);
1190 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1194 * Tries to open block device by device number. Use it ONLY if you
1195 * really do not have anything better - i.e. when you are behind a
1196 * truly sucky interface and all you are given is a device number. _Never_
1197 * to be used for internal purposes. If you ever need it - reconsider
1200 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1202 struct block_device *bdev = bdget(dev);
1205 err = blkdev_get(bdev, mode);
1206 return err ? ERR_PTR(err) : bdev;
1209 EXPORT_SYMBOL(open_by_devnum);
1212 * flush_disk - invalidates all buffer-cache entries on a disk
1214 * @bdev: struct block device to be flushed
1216 * Invalidates all buffer-cache entries on a disk. It should be called
1217 * when a disk has been changed -- either by a media change or online
1220 static void flush_disk(struct block_device *bdev)
1222 if (__invalidate_device(bdev)) {
1223 char name[BDEVNAME_SIZE] = "";
1226 disk_name(bdev->bd_disk, 0, name);
1227 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1228 "resized disk %s\n", name);
1233 if (disk_partitionable(bdev->bd_disk))
1234 bdev->bd_invalidated = 1;
1238 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1239 * @disk: struct gendisk to check
1240 * @bdev: struct bdev to adjust.
1242 * This routine checks to see if the bdev size does not match the disk size
1243 * and adjusts it if it differs.
1245 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1247 loff_t disk_size, bdev_size;
1249 disk_size = (loff_t)get_capacity(disk) << 9;
1250 bdev_size = i_size_read(bdev->bd_inode);
1251 if (disk_size != bdev_size) {
1252 char name[BDEVNAME_SIZE];
1254 disk_name(disk, 0, name);
1256 "%s: detected capacity change from %lld to %lld\n",
1257 name, bdev_size, disk_size);
1258 i_size_write(bdev->bd_inode, disk_size);
1262 EXPORT_SYMBOL(check_disk_size_change);
1265 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1266 * @disk: struct gendisk to be revalidated
1268 * This routine is a wrapper for lower-level driver's revalidate_disk
1269 * call-backs. It is used to do common pre and post operations needed
1270 * for all revalidate_disk operations.
1272 int revalidate_disk(struct gendisk *disk)
1274 struct block_device *bdev;
1277 if (disk->fops->revalidate_disk)
1278 ret = disk->fops->revalidate_disk(disk);
1280 bdev = bdget_disk(disk, 0);
1284 mutex_lock(&bdev->bd_mutex);
1285 check_disk_size_change(disk, bdev);
1286 mutex_unlock(&bdev->bd_mutex);
1290 EXPORT_SYMBOL(revalidate_disk);
1293 * This routine checks whether a removable media has been changed,
1294 * and invalidates all buffer-cache-entries in that case. This
1295 * is a relatively slow routine, so we have to try to minimize using
1296 * it. Thus it is called only upon a 'mount' or 'open'. This
1297 * is the best way of combining speed and utility, I think.
1298 * People changing diskettes in the middle of an operation deserve
1301 int check_disk_change(struct block_device *bdev)
1303 struct gendisk *disk = bdev->bd_disk;
1304 const struct block_device_operations *bdops = disk->fops;
1306 if (!bdops->media_changed)
1308 if (!bdops->media_changed(bdev->bd_disk))
1312 if (bdops->revalidate_disk)
1313 bdops->revalidate_disk(bdev->bd_disk);
1317 EXPORT_SYMBOL(check_disk_change);
1319 void bd_set_size(struct block_device *bdev, loff_t size)
1321 unsigned bsize = bdev_logical_block_size(bdev);
1323 bdev->bd_inode->i_size = size;
1324 while (bsize < PAGE_CACHE_SIZE) {
1329 bdev->bd_block_size = bsize;
1330 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1332 EXPORT_SYMBOL(bd_set_size);
1334 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1339 * mutex_lock(part->bd_mutex)
1340 * mutex_lock_nested(whole->bd_mutex, 1)
1343 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1345 struct gendisk *disk;
1350 if (mode & FMODE_READ)
1352 if (mode & FMODE_WRITE)
1355 * hooks: /n/, see "layering violations".
1357 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1367 disk = get_gendisk(bdev->bd_dev, &partno);
1369 goto out_unlock_kernel;
1371 mutex_lock_nested(&bdev->bd_mutex, for_part);
1372 if (!bdev->bd_openers) {
1373 bdev->bd_disk = disk;
1374 bdev->bd_contains = bdev;
1376 struct backing_dev_info *bdi;
1379 bdev->bd_part = disk_get_part(disk, partno);
1383 if (disk->fops->open) {
1384 ret = disk->fops->open(bdev, mode);
1385 if (ret == -ERESTARTSYS) {
1386 /* Lost a race with 'disk' being
1387 * deleted, try again.
1390 disk_put_part(bdev->bd_part);
1391 bdev->bd_part = NULL;
1392 module_put(disk->fops->owner);
1394 bdev->bd_disk = NULL;
1395 mutex_unlock(&bdev->bd_mutex);
1401 if (!bdev->bd_openers) {
1402 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1403 bdi = blk_get_backing_dev_info(bdev);
1405 bdi = &default_backing_dev_info;
1406 bdev->bd_inode->i_data.backing_dev_info = bdi;
1408 if (bdev->bd_invalidated)
1409 rescan_partitions(disk, bdev);
1411 struct block_device *whole;
1412 whole = bdget_disk(disk, 0);
1417 ret = __blkdev_get(whole, mode, 1);
1420 bdev->bd_contains = whole;
1421 bdev->bd_inode->i_data.backing_dev_info =
1422 whole->bd_inode->i_data.backing_dev_info;
1423 bdev->bd_part = disk_get_part(disk, partno);
1424 if (!(disk->flags & GENHD_FL_UP) ||
1425 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1429 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1432 module_put(disk->fops->owner);
1435 if (bdev->bd_contains == bdev) {
1436 if (bdev->bd_disk->fops->open) {
1437 ret = bdev->bd_disk->fops->open(bdev, mode);
1439 goto out_unlock_bdev;
1441 if (bdev->bd_invalidated)
1442 rescan_partitions(bdev->bd_disk, bdev);
1447 bdev->bd_part_count++;
1448 mutex_unlock(&bdev->bd_mutex);
1453 disk_put_part(bdev->bd_part);
1454 bdev->bd_disk = NULL;
1455 bdev->bd_part = NULL;
1456 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1457 if (bdev != bdev->bd_contains)
1458 __blkdev_put(bdev->bd_contains, mode, 1);
1459 bdev->bd_contains = NULL;
1461 mutex_unlock(&bdev->bd_mutex);
1466 module_put(disk->fops->owner);
1473 int blkdev_get(struct block_device *bdev, fmode_t mode)
1475 return __blkdev_get(bdev, mode, 0);
1477 EXPORT_SYMBOL(blkdev_get);
1479 static int blkdev_open(struct inode * inode, struct file * filp)
1481 struct block_device *whole = NULL;
1482 struct block_device *bdev;
1486 * Preserve backwards compatibility and allow large file access
1487 * even if userspace doesn't ask for it explicitly. Some mkfs
1488 * binary needs it. We might want to drop this workaround
1489 * during an unstable branch.
1491 filp->f_flags |= O_LARGEFILE;
1493 if (filp->f_flags & O_NDELAY)
1494 filp->f_mode |= FMODE_NDELAY;
1495 if (filp->f_flags & O_EXCL)
1496 filp->f_mode |= FMODE_EXCL;
1497 if ((filp->f_flags & O_ACCMODE) == 3)
1498 filp->f_mode |= FMODE_WRITE_IOCTL;
1500 bdev = bd_acquire(inode);
1504 if (filp->f_mode & FMODE_EXCL) {
1505 whole = bd_start_claiming(bdev, filp);
1506 if (IS_ERR(whole)) {
1508 return PTR_ERR(whole);
1512 filp->f_mapping = bdev->bd_inode->i_mapping;
1514 res = blkdev_get(bdev, filp->f_mode);
1518 BUG_ON(bd_claim(bdev, filp) != 0);
1520 bd_abort_claiming(whole, filp);
1526 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1529 struct gendisk *disk = bdev->bd_disk;
1530 struct block_device *victim = NULL;
1532 mutex_lock_nested(&bdev->bd_mutex, for_part);
1535 bdev->bd_part_count--;
1537 if (!--bdev->bd_openers) {
1538 sync_blockdev(bdev);
1541 if (bdev->bd_contains == bdev) {
1542 if (disk->fops->release)
1543 ret = disk->fops->release(disk, mode);
1545 if (!bdev->bd_openers) {
1546 struct module *owner = disk->fops->owner;
1550 disk_put_part(bdev->bd_part);
1551 bdev->bd_part = NULL;
1552 bdev->bd_disk = NULL;
1553 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1554 if (bdev != bdev->bd_contains)
1555 victim = bdev->bd_contains;
1556 bdev->bd_contains = NULL;
1559 mutex_unlock(&bdev->bd_mutex);
1562 __blkdev_put(victim, mode, 1);
1566 int blkdev_put(struct block_device *bdev, fmode_t mode)
1568 return __blkdev_put(bdev, mode, 0);
1570 EXPORT_SYMBOL(blkdev_put);
1572 static int blkdev_close(struct inode * inode, struct file * filp)
1574 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1575 if (bdev->bd_holder == filp)
1577 return blkdev_put(bdev, filp->f_mode);
1580 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1582 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1583 fmode_t mode = file->f_mode;
1586 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1587 * to updated it before every ioctl.
1589 if (file->f_flags & O_NDELAY)
1590 mode |= FMODE_NDELAY;
1592 mode &= ~FMODE_NDELAY;
1594 return blkdev_ioctl(bdev, mode, cmd, arg);
1598 * Write data to the block device. Only intended for the block device itself
1599 * and the raw driver which basically is a fake block device.
1601 * Does not take i_mutex for the write and thus is not for general purpose
1604 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1605 unsigned long nr_segs, loff_t pos)
1607 struct file *file = iocb->ki_filp;
1610 BUG_ON(iocb->ki_pos != pos);
1612 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1613 if (ret > 0 || ret == -EIOCBQUEUED) {
1616 err = generic_write_sync(file, pos, ret);
1617 if (err < 0 && ret > 0)
1622 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1625 * Try to release a page associated with block device when the system
1626 * is under memory pressure.
1628 static int blkdev_releasepage(struct page *page, gfp_t wait)
1630 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1632 if (super && super->s_op->bdev_try_to_free_page)
1633 return super->s_op->bdev_try_to_free_page(super, page, wait);
1635 return try_to_free_buffers(page);
1638 static const struct address_space_operations def_blk_aops = {
1639 .readpage = blkdev_readpage,
1640 .writepage = blkdev_writepage,
1641 .sync_page = block_sync_page,
1642 .write_begin = blkdev_write_begin,
1643 .write_end = blkdev_write_end,
1644 .writepages = generic_writepages,
1645 .releasepage = blkdev_releasepage,
1646 .direct_IO = blkdev_direct_IO,
1649 const struct file_operations def_blk_fops = {
1650 .open = blkdev_open,
1651 .release = blkdev_close,
1652 .llseek = block_llseek,
1653 .read = do_sync_read,
1654 .write = do_sync_write,
1655 .aio_read = generic_file_aio_read,
1656 .aio_write = blkdev_aio_write,
1657 .mmap = generic_file_mmap,
1658 .fsync = blkdev_fsync,
1659 .unlocked_ioctl = block_ioctl,
1660 #ifdef CONFIG_COMPAT
1661 .compat_ioctl = compat_blkdev_ioctl,
1663 .splice_read = generic_file_splice_read,
1664 .splice_write = generic_file_splice_write,
1667 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1670 mm_segment_t old_fs = get_fs();
1672 res = blkdev_ioctl(bdev, 0, cmd, arg);
1677 EXPORT_SYMBOL(ioctl_by_bdev);
1680 * lookup_bdev - lookup a struct block_device by name
1681 * @pathname: special file representing the block device
1683 * Get a reference to the blockdevice at @pathname in the current
1684 * namespace if possible and return it. Return ERR_PTR(error)
1687 struct block_device *lookup_bdev(const char *pathname)
1689 struct block_device *bdev;
1690 struct inode *inode;
1694 if (!pathname || !*pathname)
1695 return ERR_PTR(-EINVAL);
1697 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1699 return ERR_PTR(error);
1701 inode = path.dentry->d_inode;
1703 if (!S_ISBLK(inode->i_mode))
1706 if (path.mnt->mnt_flags & MNT_NODEV)
1709 bdev = bd_acquire(inode);
1716 bdev = ERR_PTR(error);
1719 EXPORT_SYMBOL(lookup_bdev);
1722 * open_bdev_exclusive - open a block device by name and set it up for use
1724 * @path: special file representing the block device
1725 * @mode: FMODE_... combination to pass be used
1726 * @holder: owner for exclusion
1728 * Open the blockdevice described by the special file at @path, claim it
1731 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1733 struct block_device *bdev, *whole;
1736 bdev = lookup_bdev(path);
1740 whole = bd_start_claiming(bdev, holder);
1741 if (IS_ERR(whole)) {
1746 error = blkdev_get(bdev, mode);
1748 goto out_abort_claiming;
1751 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1752 goto out_blkdev_put;
1754 BUG_ON(bd_claim(bdev, holder) != 0);
1758 blkdev_put(bdev, mode);
1760 bd_abort_claiming(whole, holder);
1761 return ERR_PTR(error);
1764 EXPORT_SYMBOL(open_bdev_exclusive);
1767 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1769 * @bdev: blockdevice to close
1770 * @mode: mode, must match that used to open.
1772 * This is the counterpart to open_bdev_exclusive().
1774 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1777 blkdev_put(bdev, mode);
1780 EXPORT_SYMBOL(close_bdev_exclusive);
1782 int __invalidate_device(struct block_device *bdev)
1784 struct super_block *sb = get_super(bdev);
1789 * no need to lock the super, get_super holds the
1790 * read mutex so the filesystem cannot go away
1791 * under us (->put_super runs with the write lock
1794 shrink_dcache_sb(sb);
1795 res = invalidate_inodes(sb);
1798 invalidate_bdev(bdev);
1801 EXPORT_SYMBOL(__invalidate_device);