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(rw, iocb, inode, I_BDEV(inode), iov, offset,
176 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
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 error = freeze_super(sb);
250 deactivate_super(sb);
251 bdev->bd_fsfreeze_count--;
252 mutex_unlock(&bdev->bd_fsfreeze_mutex);
253 return ERR_PTR(error);
255 deactivate_super(sb);
258 mutex_unlock(&bdev->bd_fsfreeze_mutex);
259 return sb; /* thaw_bdev releases s->s_umount */
261 EXPORT_SYMBOL(freeze_bdev);
264 * thaw_bdev -- unlock filesystem
265 * @bdev: blockdevice to unlock
266 * @sb: associated superblock
268 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
270 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
274 mutex_lock(&bdev->bd_fsfreeze_mutex);
275 if (!bdev->bd_fsfreeze_count)
279 if (--bdev->bd_fsfreeze_count > 0)
285 error = thaw_super(sb);
287 bdev->bd_fsfreeze_count++;
288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
295 EXPORT_SYMBOL(thaw_bdev);
297 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
299 return block_write_full_page(page, blkdev_get_block, wbc);
302 static int blkdev_readpage(struct file * file, struct page * page)
304 return block_read_full_page(page, blkdev_get_block);
307 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned flags,
309 struct page **pagep, void **fsdata)
312 return block_write_begin_newtrunc(file, mapping, pos, len, flags,
313 pagep, fsdata, blkdev_get_block);
316 static int blkdev_write_end(struct file *file, struct address_space *mapping,
317 loff_t pos, unsigned len, unsigned copied,
318 struct page *page, void *fsdata)
321 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
324 page_cache_release(page);
331 * for a block special file file->f_path.dentry->d_inode->i_size is zero
332 * so we compute the size by hand (just as in block_read/write above)
334 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
336 struct inode *bd_inode = file->f_mapping->host;
340 mutex_lock(&bd_inode->i_mutex);
341 size = i_size_read(bd_inode);
348 offset += file->f_pos;
351 if (offset >= 0 && offset <= size) {
352 if (offset != file->f_pos) {
353 file->f_pos = offset;
357 mutex_unlock(&bd_inode->i_mutex);
361 int blkdev_fsync(struct file *filp, int datasync)
363 struct inode *bd_inode = filp->f_mapping->host;
364 struct block_device *bdev = I_BDEV(bd_inode);
368 * There is no need to serialise calls to blkdev_issue_flush with
369 * i_mutex and doing so causes performance issues with concurrent
370 * O_SYNC writers to a block device.
372 mutex_unlock(&bd_inode->i_mutex);
374 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL, BLKDEV_IFL_WAIT);
375 if (error == -EOPNOTSUPP)
378 mutex_lock(&bd_inode->i_mutex);
382 EXPORT_SYMBOL(blkdev_fsync);
388 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
389 static struct kmem_cache * bdev_cachep __read_mostly;
391 static struct inode *bdev_alloc_inode(struct super_block *sb)
393 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
396 return &ei->vfs_inode;
399 static void bdev_destroy_inode(struct inode *inode)
401 struct bdev_inode *bdi = BDEV_I(inode);
403 kmem_cache_free(bdev_cachep, bdi);
406 static void init_once(void *foo)
408 struct bdev_inode *ei = (struct bdev_inode *) foo;
409 struct block_device *bdev = &ei->bdev;
411 memset(bdev, 0, sizeof(*bdev));
412 mutex_init(&bdev->bd_mutex);
413 INIT_LIST_HEAD(&bdev->bd_inodes);
414 INIT_LIST_HEAD(&bdev->bd_list);
416 INIT_LIST_HEAD(&bdev->bd_holder_list);
418 inode_init_once(&ei->vfs_inode);
419 /* Initialize mutex for freeze. */
420 mutex_init(&bdev->bd_fsfreeze_mutex);
423 static inline void __bd_forget(struct inode *inode)
425 list_del_init(&inode->i_devices);
426 inode->i_bdev = NULL;
427 inode->i_mapping = &inode->i_data;
430 static void bdev_clear_inode(struct inode *inode)
432 struct block_device *bdev = &BDEV_I(inode)->bdev;
434 spin_lock(&bdev_lock);
435 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
436 __bd_forget(list_entry(p, struct inode, i_devices));
438 list_del_init(&bdev->bd_list);
439 spin_unlock(&bdev_lock);
442 static const struct super_operations bdev_sops = {
443 .statfs = simple_statfs,
444 .alloc_inode = bdev_alloc_inode,
445 .destroy_inode = bdev_destroy_inode,
446 .drop_inode = generic_delete_inode,
447 .clear_inode = bdev_clear_inode,
450 static int bd_get_sb(struct file_system_type *fs_type,
451 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
453 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
456 static struct file_system_type bd_type = {
459 .kill_sb = kill_anon_super,
462 struct super_block *blockdev_superblock __read_mostly;
464 void __init bdev_cache_init(void)
467 struct vfsmount *bd_mnt;
469 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
470 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
471 SLAB_MEM_SPREAD|SLAB_PANIC),
473 err = register_filesystem(&bd_type);
475 panic("Cannot register bdev pseudo-fs");
476 bd_mnt = kern_mount(&bd_type);
478 panic("Cannot create bdev pseudo-fs");
480 * This vfsmount structure is only used to obtain the
481 * blockdev_superblock, so tell kmemleak not to report it.
483 kmemleak_not_leak(bd_mnt);
484 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
488 * Most likely _very_ bad one - but then it's hardly critical for small
489 * /dev and can be fixed when somebody will need really large one.
490 * Keep in mind that it will be fed through icache hash function too.
492 static inline unsigned long hash(dev_t dev)
494 return MAJOR(dev)+MINOR(dev);
497 static int bdev_test(struct inode *inode, void *data)
499 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
502 static int bdev_set(struct inode *inode, void *data)
504 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
508 static LIST_HEAD(all_bdevs);
510 struct block_device *bdget(dev_t dev)
512 struct block_device *bdev;
515 inode = iget5_locked(blockdev_superblock, hash(dev),
516 bdev_test, bdev_set, &dev);
521 bdev = &BDEV_I(inode)->bdev;
523 if (inode->i_state & I_NEW) {
524 bdev->bd_contains = NULL;
525 bdev->bd_inode = inode;
526 bdev->bd_block_size = (1 << inode->i_blkbits);
527 bdev->bd_part_count = 0;
528 bdev->bd_invalidated = 0;
529 inode->i_mode = S_IFBLK;
531 inode->i_bdev = bdev;
532 inode->i_data.a_ops = &def_blk_aops;
533 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
534 inode->i_data.backing_dev_info = &default_backing_dev_info;
535 spin_lock(&bdev_lock);
536 list_add(&bdev->bd_list, &all_bdevs);
537 spin_unlock(&bdev_lock);
538 unlock_new_inode(inode);
543 EXPORT_SYMBOL(bdget);
546 * bdgrab -- Grab a reference to an already referenced block device
547 * @bdev: Block device to grab a reference to.
549 struct block_device *bdgrab(struct block_device *bdev)
551 atomic_inc(&bdev->bd_inode->i_count);
555 long nr_blockdev_pages(void)
557 struct block_device *bdev;
559 spin_lock(&bdev_lock);
560 list_for_each_entry(bdev, &all_bdevs, bd_list) {
561 ret += bdev->bd_inode->i_mapping->nrpages;
563 spin_unlock(&bdev_lock);
567 void bdput(struct block_device *bdev)
569 iput(bdev->bd_inode);
572 EXPORT_SYMBOL(bdput);
574 static struct block_device *bd_acquire(struct inode *inode)
576 struct block_device *bdev;
578 spin_lock(&bdev_lock);
579 bdev = inode->i_bdev;
581 atomic_inc(&bdev->bd_inode->i_count);
582 spin_unlock(&bdev_lock);
585 spin_unlock(&bdev_lock);
587 bdev = bdget(inode->i_rdev);
589 spin_lock(&bdev_lock);
590 if (!inode->i_bdev) {
592 * We take an additional bd_inode->i_count for inode,
593 * and it's released in clear_inode() of inode.
594 * So, we can access it via ->i_mapping always
597 atomic_inc(&bdev->bd_inode->i_count);
598 inode->i_bdev = bdev;
599 inode->i_mapping = bdev->bd_inode->i_mapping;
600 list_add(&inode->i_devices, &bdev->bd_inodes);
602 spin_unlock(&bdev_lock);
607 /* Call when you free inode */
609 void bd_forget(struct inode *inode)
611 struct block_device *bdev = NULL;
613 spin_lock(&bdev_lock);
615 if (!sb_is_blkdev_sb(inode->i_sb))
616 bdev = inode->i_bdev;
619 spin_unlock(&bdev_lock);
622 iput(bdev->bd_inode);
626 * bd_may_claim - test whether a block device can be claimed
627 * @bdev: block device of interest
628 * @whole: whole block device containing @bdev, may equal @bdev
629 * @holder: holder trying to claim @bdev
631 * Test whther @bdev can be claimed by @holder.
634 * spin_lock(&bdev_lock).
637 * %true if @bdev can be claimed, %false otherwise.
639 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
642 if (bdev->bd_holder == holder)
643 return true; /* already a holder */
644 else if (bdev->bd_holder != NULL)
645 return false; /* held by someone else */
646 else if (bdev->bd_contains == bdev)
647 return true; /* is a whole device which isn't held */
649 else if (whole->bd_holder == bd_claim)
650 return true; /* is a partition of a device that is being partitioned */
651 else if (whole->bd_holder != NULL)
652 return false; /* is a partition of a held device */
654 return true; /* is a partition of an un-held device */
658 * bd_prepare_to_claim - prepare to claim a block device
659 * @bdev: block device of interest
660 * @whole: the whole device containing @bdev, may equal @bdev
661 * @holder: holder trying to claim @bdev
663 * Prepare to claim @bdev. This function fails if @bdev is already
664 * claimed by another holder and waits if another claiming is in
665 * progress. This function doesn't actually claim. On successful
666 * return, the caller has ownership of bd_claiming and bd_holder[s].
669 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
673 * 0 if @bdev can be claimed, -EBUSY otherwise.
675 static int bd_prepare_to_claim(struct block_device *bdev,
676 struct block_device *whole, void *holder)
679 /* if someone else claimed, fail */
680 if (!bd_may_claim(bdev, whole, holder))
683 /* if someone else is claiming, wait for it to finish */
684 if (whole->bd_claiming && whole->bd_claiming != holder) {
685 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
688 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
689 spin_unlock(&bdev_lock);
691 finish_wait(wq, &wait);
692 spin_lock(&bdev_lock);
701 * bd_start_claiming - start claiming a block device
702 * @bdev: block device of interest
703 * @holder: holder trying to claim @bdev
705 * @bdev is about to be opened exclusively. Check @bdev can be opened
706 * exclusively and mark that an exclusive open is in progress. Each
707 * successful call to this function must be matched with a call to
708 * either bd_finish_claiming() or bd_abort_claiming() (which do not
711 * This function is used to gain exclusive access to the block device
712 * without actually causing other exclusive open attempts to fail. It
713 * should be used when the open sequence itself requires exclusive
714 * access but may subsequently fail.
720 * Pointer to the block device containing @bdev on success, ERR_PTR()
723 static struct block_device *bd_start_claiming(struct block_device *bdev,
726 struct gendisk *disk;
727 struct block_device *whole;
733 * @bdev might not have been initialized properly yet, look up
734 * and grab the outer block device the hard way.
736 disk = get_gendisk(bdev->bd_dev, &partno);
738 return ERR_PTR(-ENXIO);
740 whole = bdget_disk(disk, 0);
741 module_put(disk->fops->owner);
744 return ERR_PTR(-ENOMEM);
746 /* prepare to claim, if successful, mark claiming in progress */
747 spin_lock(&bdev_lock);
749 err = bd_prepare_to_claim(bdev, whole, holder);
751 whole->bd_claiming = holder;
752 spin_unlock(&bdev_lock);
755 spin_unlock(&bdev_lock);
761 /* releases bdev_lock */
762 static void __bd_abort_claiming(struct block_device *whole, void *holder)
764 BUG_ON(whole->bd_claiming != holder);
765 whole->bd_claiming = NULL;
766 wake_up_bit(&whole->bd_claiming, 0);
768 spin_unlock(&bdev_lock);
773 * bd_abort_claiming - abort claiming a block device
774 * @whole: whole block device returned by bd_start_claiming()
775 * @holder: holder trying to claim @bdev
777 * Abort a claiming block started by bd_start_claiming(). Note that
778 * @whole is not the block device to be claimed but the whole device
779 * returned by bd_start_claiming().
782 * Grabs and releases bdev_lock.
784 static void bd_abort_claiming(struct block_device *whole, void *holder)
786 spin_lock(&bdev_lock);
787 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
790 /* increment holders when we have a legitimate claim. requires bdev_lock */
791 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
794 /* note that for a whole device bd_holders
795 * will be incremented twice, and bd_holder will
796 * be set to bd_claim before being set to holder
799 whole->bd_holder = bd_claim;
801 bdev->bd_holder = holder;
805 * bd_finish_claiming - finish claiming a block device
806 * @bdev: block device of interest (passed to bd_start_claiming())
807 * @whole: whole block device returned by bd_start_claiming()
808 * @holder: holder trying to claim @bdev
810 * Finish a claiming block started by bd_start_claiming().
813 * Grabs and releases bdev_lock.
815 static void bd_finish_claiming(struct block_device *bdev,
816 struct block_device *whole, void *holder)
818 spin_lock(&bdev_lock);
819 BUG_ON(!bd_may_claim(bdev, whole, holder));
820 __bd_claim(bdev, whole, holder);
821 __bd_abort_claiming(whole, holder); /* not actually an abort */
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.
835 * 0 if successful, -EBUSY if @bdev is already claimed.
837 int bd_claim(struct block_device *bdev, void *holder)
839 struct block_device *whole = bdev->bd_contains;
844 spin_lock(&bdev_lock);
845 res = bd_prepare_to_claim(bdev, whole, holder);
847 __bd_claim(bdev, whole, holder);
848 spin_unlock(&bdev_lock);
852 EXPORT_SYMBOL(bd_claim);
854 void bd_release(struct block_device *bdev)
856 spin_lock(&bdev_lock);
857 if (!--bdev->bd_contains->bd_holders)
858 bdev->bd_contains->bd_holder = NULL;
859 if (!--bdev->bd_holders)
860 bdev->bd_holder = NULL;
861 spin_unlock(&bdev_lock);
864 EXPORT_SYMBOL(bd_release);
868 * Functions for bd_claim_by_kobject / bd_release_from_kobject
870 * If a kobject is passed to bd_claim_by_kobject()
871 * and the kobject has a parent directory,
872 * following symlinks are created:
873 * o from the kobject to the claimed bdev
874 * o from "holders" directory of the bdev to the parent of the kobject
875 * bd_release_from_kobject() removes these symlinks.
878 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
879 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
880 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
881 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
884 static int add_symlink(struct kobject *from, struct kobject *to)
888 return sysfs_create_link(from, to, kobject_name(to));
891 static void del_symlink(struct kobject *from, struct kobject *to)
895 sysfs_remove_link(from, kobject_name(to));
899 * 'struct bd_holder' contains pointers to kobjects symlinked by
900 * bd_claim_by_kobject.
901 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
904 struct list_head list; /* chain of holders of the bdev */
905 int count; /* references from the holder */
906 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
907 struct kobject *hdev; /* e.g. "/block/dm-0" */
908 struct kobject *hdir; /* e.g. "/block/sda/holders" */
909 struct kobject *sdev; /* e.g. "/block/sda" */
913 * Get references of related kobjects at once.
914 * Returns 1 on success. 0 on failure.
916 * Should call bd_holder_release_dirs() after successful use.
918 static int bd_holder_grab_dirs(struct block_device *bdev,
919 struct bd_holder *bo)
924 bo->sdir = kobject_get(bo->sdir);
928 bo->hdev = kobject_get(bo->sdir->parent);
932 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
936 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
943 kobject_put(bo->sdev);
945 kobject_put(bo->hdev);
947 kobject_put(bo->sdir);
952 /* Put references of related kobjects at once. */
953 static void bd_holder_release_dirs(struct bd_holder *bo)
955 kobject_put(bo->hdir);
956 kobject_put(bo->sdev);
957 kobject_put(bo->hdev);
958 kobject_put(bo->sdir);
961 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
963 struct bd_holder *bo;
965 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
975 static void free_bd_holder(struct bd_holder *bo)
981 * find_bd_holder - find matching struct bd_holder from the block device
983 * @bdev: struct block device to be searched
984 * @bo: target struct bd_holder
986 * Returns matching entry with @bo in @bdev->bd_holder_list.
987 * If found, increment the reference count and return the pointer.
988 * If not found, returns NULL.
990 static struct bd_holder *find_bd_holder(struct block_device *bdev,
991 struct bd_holder *bo)
993 struct bd_holder *tmp;
995 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
996 if (tmp->sdir == bo->sdir) {
1005 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1007 * @bdev: block device to be bd_claimed
1008 * @bo: preallocated and initialized by alloc_bd_holder()
1010 * Add @bo to @bdev->bd_holder_list, create symlinks.
1012 * Returns 0 if symlinks are created.
1013 * Returns -ve if something fails.
1015 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1022 if (!bd_holder_grab_dirs(bdev, bo))
1025 err = add_symlink(bo->sdir, bo->sdev);
1029 err = add_symlink(bo->hdir, bo->hdev);
1031 del_symlink(bo->sdir, bo->sdev);
1035 list_add_tail(&bo->list, &bdev->bd_holder_list);
1040 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1042 * @bdev: block device to be bd_claimed
1043 * @kobj: holder's kobject
1045 * If there is matching entry with @kobj in @bdev->bd_holder_list
1046 * and no other bd_claim() from the same kobject,
1047 * remove the struct bd_holder from the list, delete symlinks for it.
1049 * Returns a pointer to the struct bd_holder when it's removed from the list
1050 * and ready to be freed.
1051 * Returns NULL if matching claim isn't found or there is other bd_claim()
1052 * by the same kobject.
1054 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1055 struct kobject *kobj)
1057 struct bd_holder *bo;
1059 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1060 if (bo->sdir == kobj) {
1062 BUG_ON(bo->count < 0);
1064 list_del(&bo->list);
1065 del_symlink(bo->sdir, bo->sdev);
1066 del_symlink(bo->hdir, bo->hdev);
1067 bd_holder_release_dirs(bo);
1078 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1080 * @bdev: block device to be claimed
1081 * @holder: holder's signature
1082 * @kobj: holder's kobject
1084 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1085 * the bdev and the holder's kobject.
1086 * Use bd_release_from_kobject() when relesing the claimed bdev.
1088 * Returns 0 on success. (same as bd_claim())
1089 * Returns errno on failure.
1091 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1092 struct kobject *kobj)
1095 struct bd_holder *bo, *found;
1100 bo = alloc_bd_holder(kobj);
1104 mutex_lock(&bdev->bd_mutex);
1106 err = bd_claim(bdev, holder);
1110 found = find_bd_holder(bdev, bo);
1114 err = add_bd_holder(bdev, bo);
1120 mutex_unlock(&bdev->bd_mutex);
1126 * bd_release_from_kobject - bd_release() with additional kobject signature
1128 * @bdev: block device to be released
1129 * @kobj: holder's kobject
1131 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1133 static void bd_release_from_kobject(struct block_device *bdev,
1134 struct kobject *kobj)
1139 mutex_lock(&bdev->bd_mutex);
1141 free_bd_holder(del_bd_holder(bdev, kobj));
1142 mutex_unlock(&bdev->bd_mutex);
1146 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1148 * @bdev: block device to be claimed
1149 * @holder: holder's signature
1150 * @disk: holder's gendisk
1152 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1154 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1155 struct gendisk *disk)
1157 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1159 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1162 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1164 * @bdev: block device to be claimed
1165 * @disk: holder's gendisk
1167 * Call bd_release_from_kobject() and put @disk->slave_dir.
1169 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1171 bd_release_from_kobject(bdev, disk->slave_dir);
1172 kobject_put(disk->slave_dir);
1174 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1178 * Tries to open block device by device number. Use it ONLY if you
1179 * really do not have anything better - i.e. when you are behind a
1180 * truly sucky interface and all you are given is a device number. _Never_
1181 * to be used for internal purposes. If you ever need it - reconsider
1184 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1186 struct block_device *bdev = bdget(dev);
1189 err = blkdev_get(bdev, mode);
1190 return err ? ERR_PTR(err) : bdev;
1193 EXPORT_SYMBOL(open_by_devnum);
1196 * flush_disk - invalidates all buffer-cache entries on a disk
1198 * @bdev: struct block device to be flushed
1200 * Invalidates all buffer-cache entries on a disk. It should be called
1201 * when a disk has been changed -- either by a media change or online
1204 static void flush_disk(struct block_device *bdev)
1206 if (__invalidate_device(bdev)) {
1207 char name[BDEVNAME_SIZE] = "";
1210 disk_name(bdev->bd_disk, 0, name);
1211 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1212 "resized disk %s\n", name);
1217 if (disk_partitionable(bdev->bd_disk))
1218 bdev->bd_invalidated = 1;
1222 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1223 * @disk: struct gendisk to check
1224 * @bdev: struct bdev to adjust.
1226 * This routine checks to see if the bdev size does not match the disk size
1227 * and adjusts it if it differs.
1229 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1231 loff_t disk_size, bdev_size;
1233 disk_size = (loff_t)get_capacity(disk) << 9;
1234 bdev_size = i_size_read(bdev->bd_inode);
1235 if (disk_size != bdev_size) {
1236 char name[BDEVNAME_SIZE];
1238 disk_name(disk, 0, name);
1240 "%s: detected capacity change from %lld to %lld\n",
1241 name, bdev_size, disk_size);
1242 i_size_write(bdev->bd_inode, disk_size);
1246 EXPORT_SYMBOL(check_disk_size_change);
1249 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1250 * @disk: struct gendisk to be revalidated
1252 * This routine is a wrapper for lower-level driver's revalidate_disk
1253 * call-backs. It is used to do common pre and post operations needed
1254 * for all revalidate_disk operations.
1256 int revalidate_disk(struct gendisk *disk)
1258 struct block_device *bdev;
1261 if (disk->fops->revalidate_disk)
1262 ret = disk->fops->revalidate_disk(disk);
1264 bdev = bdget_disk(disk, 0);
1268 mutex_lock(&bdev->bd_mutex);
1269 check_disk_size_change(disk, bdev);
1270 mutex_unlock(&bdev->bd_mutex);
1274 EXPORT_SYMBOL(revalidate_disk);
1277 * This routine checks whether a removable media has been changed,
1278 * and invalidates all buffer-cache-entries in that case. This
1279 * is a relatively slow routine, so we have to try to minimize using
1280 * it. Thus it is called only upon a 'mount' or 'open'. This
1281 * is the best way of combining speed and utility, I think.
1282 * People changing diskettes in the middle of an operation deserve
1285 int check_disk_change(struct block_device *bdev)
1287 struct gendisk *disk = bdev->bd_disk;
1288 const struct block_device_operations *bdops = disk->fops;
1290 if (!bdops->media_changed)
1292 if (!bdops->media_changed(bdev->bd_disk))
1296 if (bdops->revalidate_disk)
1297 bdops->revalidate_disk(bdev->bd_disk);
1301 EXPORT_SYMBOL(check_disk_change);
1303 void bd_set_size(struct block_device *bdev, loff_t size)
1305 unsigned bsize = bdev_logical_block_size(bdev);
1307 bdev->bd_inode->i_size = size;
1308 while (bsize < PAGE_CACHE_SIZE) {
1313 bdev->bd_block_size = bsize;
1314 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1316 EXPORT_SYMBOL(bd_set_size);
1318 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1323 * mutex_lock(part->bd_mutex)
1324 * mutex_lock_nested(whole->bd_mutex, 1)
1327 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1329 struct gendisk *disk;
1334 if (mode & FMODE_READ)
1336 if (mode & FMODE_WRITE)
1339 * hooks: /n/, see "layering violations".
1341 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1351 disk = get_gendisk(bdev->bd_dev, &partno);
1353 goto out_unlock_kernel;
1355 mutex_lock_nested(&bdev->bd_mutex, for_part);
1356 if (!bdev->bd_openers) {
1357 bdev->bd_disk = disk;
1358 bdev->bd_contains = bdev;
1360 struct backing_dev_info *bdi;
1363 bdev->bd_part = disk_get_part(disk, partno);
1367 if (disk->fops->open) {
1368 ret = disk->fops->open(bdev, mode);
1369 if (ret == -ERESTARTSYS) {
1370 /* Lost a race with 'disk' being
1371 * deleted, try again.
1374 disk_put_part(bdev->bd_part);
1375 bdev->bd_part = NULL;
1376 module_put(disk->fops->owner);
1378 bdev->bd_disk = NULL;
1379 mutex_unlock(&bdev->bd_mutex);
1385 if (!bdev->bd_openers) {
1386 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1387 bdi = blk_get_backing_dev_info(bdev);
1389 bdi = &default_backing_dev_info;
1390 bdev->bd_inode->i_data.backing_dev_info = bdi;
1392 if (bdev->bd_invalidated)
1393 rescan_partitions(disk, bdev);
1395 struct block_device *whole;
1396 whole = bdget_disk(disk, 0);
1401 ret = __blkdev_get(whole, mode, 1);
1404 bdev->bd_contains = whole;
1405 bdev->bd_inode->i_data.backing_dev_info =
1406 whole->bd_inode->i_data.backing_dev_info;
1407 bdev->bd_part = disk_get_part(disk, partno);
1408 if (!(disk->flags & GENHD_FL_UP) ||
1409 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1413 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1416 module_put(disk->fops->owner);
1419 if (bdev->bd_contains == bdev) {
1420 if (bdev->bd_disk->fops->open) {
1421 ret = bdev->bd_disk->fops->open(bdev, mode);
1423 goto out_unlock_bdev;
1425 if (bdev->bd_invalidated)
1426 rescan_partitions(bdev->bd_disk, bdev);
1431 bdev->bd_part_count++;
1432 mutex_unlock(&bdev->bd_mutex);
1437 disk_put_part(bdev->bd_part);
1438 bdev->bd_disk = NULL;
1439 bdev->bd_part = NULL;
1440 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1441 if (bdev != bdev->bd_contains)
1442 __blkdev_put(bdev->bd_contains, mode, 1);
1443 bdev->bd_contains = NULL;
1445 mutex_unlock(&bdev->bd_mutex);
1450 module_put(disk->fops->owner);
1457 int blkdev_get(struct block_device *bdev, fmode_t mode)
1459 return __blkdev_get(bdev, mode, 0);
1461 EXPORT_SYMBOL(blkdev_get);
1463 static int blkdev_open(struct inode * inode, struct file * filp)
1465 struct block_device *whole = NULL;
1466 struct block_device *bdev;
1470 * Preserve backwards compatibility and allow large file access
1471 * even if userspace doesn't ask for it explicitly. Some mkfs
1472 * binary needs it. We might want to drop this workaround
1473 * during an unstable branch.
1475 filp->f_flags |= O_LARGEFILE;
1477 if (filp->f_flags & O_NDELAY)
1478 filp->f_mode |= FMODE_NDELAY;
1479 if (filp->f_flags & O_EXCL)
1480 filp->f_mode |= FMODE_EXCL;
1481 if ((filp->f_flags & O_ACCMODE) == 3)
1482 filp->f_mode |= FMODE_WRITE_IOCTL;
1484 bdev = bd_acquire(inode);
1488 if (filp->f_mode & FMODE_EXCL) {
1489 whole = bd_start_claiming(bdev, filp);
1490 if (IS_ERR(whole)) {
1492 return PTR_ERR(whole);
1496 filp->f_mapping = bdev->bd_inode->i_mapping;
1498 res = blkdev_get(bdev, filp->f_mode);
1502 bd_finish_claiming(bdev, whole, filp);
1504 bd_abort_claiming(whole, filp);
1510 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1513 struct gendisk *disk = bdev->bd_disk;
1514 struct block_device *victim = NULL;
1516 mutex_lock_nested(&bdev->bd_mutex, for_part);
1519 bdev->bd_part_count--;
1521 if (!--bdev->bd_openers) {
1522 sync_blockdev(bdev);
1525 if (bdev->bd_contains == bdev) {
1526 if (disk->fops->release)
1527 ret = disk->fops->release(disk, mode);
1529 if (!bdev->bd_openers) {
1530 struct module *owner = disk->fops->owner;
1534 disk_put_part(bdev->bd_part);
1535 bdev->bd_part = NULL;
1536 bdev->bd_disk = NULL;
1537 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1538 if (bdev != bdev->bd_contains)
1539 victim = bdev->bd_contains;
1540 bdev->bd_contains = NULL;
1543 mutex_unlock(&bdev->bd_mutex);
1546 __blkdev_put(victim, mode, 1);
1550 int blkdev_put(struct block_device *bdev, fmode_t mode)
1552 return __blkdev_put(bdev, mode, 0);
1554 EXPORT_SYMBOL(blkdev_put);
1556 static int blkdev_close(struct inode * inode, struct file * filp)
1558 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1559 if (bdev->bd_holder == filp)
1561 return blkdev_put(bdev, filp->f_mode);
1564 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1566 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1567 fmode_t mode = file->f_mode;
1570 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1571 * to updated it before every ioctl.
1573 if (file->f_flags & O_NDELAY)
1574 mode |= FMODE_NDELAY;
1576 mode &= ~FMODE_NDELAY;
1578 return blkdev_ioctl(bdev, mode, cmd, arg);
1582 * Write data to the block device. Only intended for the block device itself
1583 * and the raw driver which basically is a fake block device.
1585 * Does not take i_mutex for the write and thus is not for general purpose
1588 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1589 unsigned long nr_segs, loff_t pos)
1591 struct file *file = iocb->ki_filp;
1594 BUG_ON(iocb->ki_pos != pos);
1596 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1597 if (ret > 0 || ret == -EIOCBQUEUED) {
1600 err = generic_write_sync(file, pos, ret);
1601 if (err < 0 && ret > 0)
1606 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1609 * Try to release a page associated with block device when the system
1610 * is under memory pressure.
1612 static int blkdev_releasepage(struct page *page, gfp_t wait)
1614 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1616 if (super && super->s_op->bdev_try_to_free_page)
1617 return super->s_op->bdev_try_to_free_page(super, page, wait);
1619 return try_to_free_buffers(page);
1622 static const struct address_space_operations def_blk_aops = {
1623 .readpage = blkdev_readpage,
1624 .writepage = blkdev_writepage,
1625 .sync_page = block_sync_page,
1626 .write_begin = blkdev_write_begin,
1627 .write_end = blkdev_write_end,
1628 .writepages = generic_writepages,
1629 .releasepage = blkdev_releasepage,
1630 .direct_IO = blkdev_direct_IO,
1633 const struct file_operations def_blk_fops = {
1634 .open = blkdev_open,
1635 .release = blkdev_close,
1636 .llseek = block_llseek,
1637 .read = do_sync_read,
1638 .write = do_sync_write,
1639 .aio_read = generic_file_aio_read,
1640 .aio_write = blkdev_aio_write,
1641 .mmap = generic_file_mmap,
1642 .fsync = blkdev_fsync,
1643 .unlocked_ioctl = block_ioctl,
1644 #ifdef CONFIG_COMPAT
1645 .compat_ioctl = compat_blkdev_ioctl,
1647 .splice_read = generic_file_splice_read,
1648 .splice_write = generic_file_splice_write,
1651 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1654 mm_segment_t old_fs = get_fs();
1656 res = blkdev_ioctl(bdev, 0, cmd, arg);
1661 EXPORT_SYMBOL(ioctl_by_bdev);
1664 * lookup_bdev - lookup a struct block_device by name
1665 * @pathname: special file representing the block device
1667 * Get a reference to the blockdevice at @pathname in the current
1668 * namespace if possible and return it. Return ERR_PTR(error)
1671 struct block_device *lookup_bdev(const char *pathname)
1673 struct block_device *bdev;
1674 struct inode *inode;
1678 if (!pathname || !*pathname)
1679 return ERR_PTR(-EINVAL);
1681 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1683 return ERR_PTR(error);
1685 inode = path.dentry->d_inode;
1687 if (!S_ISBLK(inode->i_mode))
1690 if (path.mnt->mnt_flags & MNT_NODEV)
1693 bdev = bd_acquire(inode);
1700 bdev = ERR_PTR(error);
1703 EXPORT_SYMBOL(lookup_bdev);
1706 * open_bdev_exclusive - open a block device by name and set it up for use
1708 * @path: special file representing the block device
1709 * @mode: FMODE_... combination to pass be used
1710 * @holder: owner for exclusion
1712 * Open the blockdevice described by the special file at @path, claim it
1715 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1717 struct block_device *bdev, *whole;
1720 bdev = lookup_bdev(path);
1724 whole = bd_start_claiming(bdev, holder);
1725 if (IS_ERR(whole)) {
1730 error = blkdev_get(bdev, mode);
1732 goto out_abort_claiming;
1735 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1736 goto out_blkdev_put;
1738 bd_finish_claiming(bdev, whole, holder);
1742 blkdev_put(bdev, mode);
1744 bd_abort_claiming(whole, holder);
1745 return ERR_PTR(error);
1748 EXPORT_SYMBOL(open_bdev_exclusive);
1751 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1753 * @bdev: blockdevice to close
1754 * @mode: mode, must match that used to open.
1756 * This is the counterpart to open_bdev_exclusive().
1758 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1761 blkdev_put(bdev, mode);
1764 EXPORT_SYMBOL(close_bdev_exclusive);
1766 int __invalidate_device(struct block_device *bdev)
1768 struct super_block *sb = get_super(bdev);
1773 * no need to lock the super, get_super holds the
1774 * read mutex so the filesystem cannot go away
1775 * under us (->put_super runs with the write lock
1778 shrink_dcache_sb(sb);
1779 res = invalidate_inodes(sb);
1782 invalidate_bdev(bdev);
1785 EXPORT_SYMBOL(__invalidate_device);