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_newtrunc(rw, iocb, inode,
176 I_BDEV(inode), iov, offset, nr_segs,
177 blkdev_get_blocks, NULL);
180 int __sync_blockdev(struct block_device *bdev, int wait)
185 return filemap_flush(bdev->bd_inode->i_mapping);
186 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
190 * Write out and wait upon all the dirty data associated with a block
191 * device via its mapping. Does not take the superblock lock.
193 int sync_blockdev(struct block_device *bdev)
195 return __sync_blockdev(bdev, 1);
197 EXPORT_SYMBOL(sync_blockdev);
200 * Write out and wait upon all dirty data associated with this
201 * device. Filesystem data as well as the underlying block
202 * device. Takes the superblock lock.
204 int fsync_bdev(struct block_device *bdev)
206 struct super_block *sb = get_super(bdev);
208 int res = sync_filesystem(sb);
212 return sync_blockdev(bdev);
214 EXPORT_SYMBOL(fsync_bdev);
217 * freeze_bdev -- lock a filesystem and force it into a consistent state
218 * @bdev: blockdevice to lock
220 * If a superblock is found on this device, we take the s_umount semaphore
221 * on it to make sure nobody unmounts until the snapshot creation is done.
222 * The reference counter (bd_fsfreeze_count) guarantees that only the last
223 * unfreeze process can unfreeze the frozen filesystem actually when multiple
224 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
225 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
228 struct super_block *freeze_bdev(struct block_device *bdev)
230 struct super_block *sb;
233 mutex_lock(&bdev->bd_fsfreeze_mutex);
234 if (++bdev->bd_fsfreeze_count > 1) {
236 * We don't even need to grab a reference - the first call
237 * to freeze_bdev grab an active reference and only the last
238 * thaw_bdev drops it.
240 sb = get_super(bdev);
242 mutex_unlock(&bdev->bd_fsfreeze_mutex);
246 sb = get_active_super(bdev);
249 error = freeze_super(sb);
251 deactivate_super(sb);
252 bdev->bd_fsfreeze_count--;
253 mutex_unlock(&bdev->bd_fsfreeze_mutex);
254 return ERR_PTR(error);
256 deactivate_super(sb);
259 mutex_unlock(&bdev->bd_fsfreeze_mutex);
260 return sb; /* thaw_bdev releases s->s_umount */
262 EXPORT_SYMBOL(freeze_bdev);
265 * thaw_bdev -- unlock filesystem
266 * @bdev: blockdevice to unlock
267 * @sb: associated superblock
269 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
271 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
275 mutex_lock(&bdev->bd_fsfreeze_mutex);
276 if (!bdev->bd_fsfreeze_count)
280 if (--bdev->bd_fsfreeze_count > 0)
286 error = thaw_super(sb);
288 bdev->bd_fsfreeze_count++;
289 mutex_unlock(&bdev->bd_fsfreeze_mutex);
293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
296 EXPORT_SYMBOL(thaw_bdev);
298 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
300 return block_write_full_page(page, blkdev_get_block, wbc);
303 static int blkdev_readpage(struct file * file, struct page * page)
305 return block_read_full_page(page, blkdev_get_block);
308 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
309 loff_t pos, unsigned len, unsigned flags,
310 struct page **pagep, void **fsdata)
313 return block_write_begin_newtrunc(file, mapping, pos, len, flags,
314 pagep, fsdata, blkdev_get_block);
317 static int blkdev_write_end(struct file *file, struct address_space *mapping,
318 loff_t pos, unsigned len, unsigned copied,
319 struct page *page, void *fsdata)
322 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
325 page_cache_release(page);
332 * for a block special file file->f_path.dentry->d_inode->i_size is zero
333 * so we compute the size by hand (just as in block_read/write above)
335 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
337 struct inode *bd_inode = file->f_mapping->host;
341 mutex_lock(&bd_inode->i_mutex);
342 size = i_size_read(bd_inode);
349 offset += file->f_pos;
352 if (offset >= 0 && offset <= size) {
353 if (offset != file->f_pos) {
354 file->f_pos = offset;
358 mutex_unlock(&bd_inode->i_mutex);
362 int blkdev_fsync(struct file *filp, int datasync)
364 struct inode *bd_inode = filp->f_mapping->host;
365 struct block_device *bdev = I_BDEV(bd_inode);
369 * There is no need to serialise calls to blkdev_issue_flush with
370 * i_mutex and doing so causes performance issues with concurrent
371 * O_SYNC writers to a block device.
373 mutex_unlock(&bd_inode->i_mutex);
375 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL, BLKDEV_IFL_WAIT);
376 if (error == -EOPNOTSUPP)
379 mutex_lock(&bd_inode->i_mutex);
383 EXPORT_SYMBOL(blkdev_fsync);
389 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
390 static struct kmem_cache * bdev_cachep __read_mostly;
392 static struct inode *bdev_alloc_inode(struct super_block *sb)
394 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
397 return &ei->vfs_inode;
400 static void bdev_destroy_inode(struct inode *inode)
402 struct bdev_inode *bdi = BDEV_I(inode);
404 kmem_cache_free(bdev_cachep, bdi);
407 static void init_once(void *foo)
409 struct bdev_inode *ei = (struct bdev_inode *) foo;
410 struct block_device *bdev = &ei->bdev;
412 memset(bdev, 0, sizeof(*bdev));
413 mutex_init(&bdev->bd_mutex);
414 INIT_LIST_HEAD(&bdev->bd_inodes);
415 INIT_LIST_HEAD(&bdev->bd_list);
417 INIT_LIST_HEAD(&bdev->bd_holder_list);
419 inode_init_once(&ei->vfs_inode);
420 /* Initialize mutex for freeze. */
421 mutex_init(&bdev->bd_fsfreeze_mutex);
424 static inline void __bd_forget(struct inode *inode)
426 list_del_init(&inode->i_devices);
427 inode->i_bdev = NULL;
428 inode->i_mapping = &inode->i_data;
431 static void bdev_clear_inode(struct inode *inode)
433 struct block_device *bdev = &BDEV_I(inode)->bdev;
435 spin_lock(&bdev_lock);
436 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
437 __bd_forget(list_entry(p, struct inode, i_devices));
439 list_del_init(&bdev->bd_list);
440 spin_unlock(&bdev_lock);
443 static const struct super_operations bdev_sops = {
444 .statfs = simple_statfs,
445 .alloc_inode = bdev_alloc_inode,
446 .destroy_inode = bdev_destroy_inode,
447 .drop_inode = generic_delete_inode,
448 .clear_inode = bdev_clear_inode,
451 static int bd_get_sb(struct file_system_type *fs_type,
452 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
454 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
457 static struct file_system_type bd_type = {
460 .kill_sb = kill_anon_super,
463 struct super_block *blockdev_superblock __read_mostly;
465 void __init bdev_cache_init(void)
468 struct vfsmount *bd_mnt;
470 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
471 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
472 SLAB_MEM_SPREAD|SLAB_PANIC),
474 err = register_filesystem(&bd_type);
476 panic("Cannot register bdev pseudo-fs");
477 bd_mnt = kern_mount(&bd_type);
479 panic("Cannot create bdev pseudo-fs");
481 * This vfsmount structure is only used to obtain the
482 * blockdev_superblock, so tell kmemleak not to report it.
484 kmemleak_not_leak(bd_mnt);
485 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
489 * Most likely _very_ bad one - but then it's hardly critical for small
490 * /dev and can be fixed when somebody will need really large one.
491 * Keep in mind that it will be fed through icache hash function too.
493 static inline unsigned long hash(dev_t dev)
495 return MAJOR(dev)+MINOR(dev);
498 static int bdev_test(struct inode *inode, void *data)
500 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
503 static int bdev_set(struct inode *inode, void *data)
505 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
509 static LIST_HEAD(all_bdevs);
511 struct block_device *bdget(dev_t dev)
513 struct block_device *bdev;
516 inode = iget5_locked(blockdev_superblock, hash(dev),
517 bdev_test, bdev_set, &dev);
522 bdev = &BDEV_I(inode)->bdev;
524 if (inode->i_state & I_NEW) {
525 bdev->bd_contains = NULL;
526 bdev->bd_inode = inode;
527 bdev->bd_block_size = (1 << inode->i_blkbits);
528 bdev->bd_part_count = 0;
529 bdev->bd_invalidated = 0;
530 inode->i_mode = S_IFBLK;
532 inode->i_bdev = bdev;
533 inode->i_data.a_ops = &def_blk_aops;
534 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
535 inode->i_data.backing_dev_info = &default_backing_dev_info;
536 spin_lock(&bdev_lock);
537 list_add(&bdev->bd_list, &all_bdevs);
538 spin_unlock(&bdev_lock);
539 unlock_new_inode(inode);
544 EXPORT_SYMBOL(bdget);
547 * bdgrab -- Grab a reference to an already referenced block device
548 * @bdev: Block device to grab a reference to.
550 struct block_device *bdgrab(struct block_device *bdev)
552 atomic_inc(&bdev->bd_inode->i_count);
556 long nr_blockdev_pages(void)
558 struct block_device *bdev;
560 spin_lock(&bdev_lock);
561 list_for_each_entry(bdev, &all_bdevs, bd_list) {
562 ret += bdev->bd_inode->i_mapping->nrpages;
564 spin_unlock(&bdev_lock);
568 void bdput(struct block_device *bdev)
570 iput(bdev->bd_inode);
573 EXPORT_SYMBOL(bdput);
575 static struct block_device *bd_acquire(struct inode *inode)
577 struct block_device *bdev;
579 spin_lock(&bdev_lock);
580 bdev = inode->i_bdev;
582 atomic_inc(&bdev->bd_inode->i_count);
583 spin_unlock(&bdev_lock);
586 spin_unlock(&bdev_lock);
588 bdev = bdget(inode->i_rdev);
590 spin_lock(&bdev_lock);
591 if (!inode->i_bdev) {
593 * We take an additional bd_inode->i_count for inode,
594 * and it's released in clear_inode() of inode.
595 * So, we can access it via ->i_mapping always
598 atomic_inc(&bdev->bd_inode->i_count);
599 inode->i_bdev = bdev;
600 inode->i_mapping = bdev->bd_inode->i_mapping;
601 list_add(&inode->i_devices, &bdev->bd_inodes);
603 spin_unlock(&bdev_lock);
608 /* Call when you free inode */
610 void bd_forget(struct inode *inode)
612 struct block_device *bdev = NULL;
614 spin_lock(&bdev_lock);
616 if (!sb_is_blkdev_sb(inode->i_sb))
617 bdev = inode->i_bdev;
620 spin_unlock(&bdev_lock);
623 iput(bdev->bd_inode);
627 * bd_may_claim - test whether a block device can be claimed
628 * @bdev: block device of interest
629 * @whole: whole block device containing @bdev, may equal @bdev
630 * @holder: holder trying to claim @bdev
632 * Test whther @bdev can be claimed by @holder.
635 * spin_lock(&bdev_lock).
638 * %true if @bdev can be claimed, %false otherwise.
640 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
643 if (bdev->bd_holder == holder)
644 return true; /* already a holder */
645 else if (bdev->bd_holder != NULL)
646 return false; /* held by someone else */
647 else if (bdev->bd_contains == bdev)
648 return true; /* is a whole device which isn't held */
650 else if (whole->bd_holder == bd_claim)
651 return true; /* is a partition of a device that is being partitioned */
652 else if (whole->bd_holder != NULL)
653 return false; /* is a partition of a held device */
655 return true; /* is a partition of an un-held device */
659 * bd_prepare_to_claim - prepare to claim a block device
660 * @bdev: block device of interest
661 * @whole: the whole device containing @bdev, may equal @bdev
662 * @holder: holder trying to claim @bdev
664 * Prepare to claim @bdev. This function fails if @bdev is already
665 * claimed by another holder and waits if another claiming is in
666 * progress. This function doesn't actually claim. On successful
667 * return, the caller has ownership of bd_claiming and bd_holder[s].
670 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
674 * 0 if @bdev can be claimed, -EBUSY otherwise.
676 static int bd_prepare_to_claim(struct block_device *bdev,
677 struct block_device *whole, void *holder)
680 /* if someone else claimed, fail */
681 if (!bd_may_claim(bdev, whole, holder))
684 /* if claiming is already in progress, wait for it to finish */
685 if (whole->bd_claiming) {
686 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
689 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
690 spin_unlock(&bdev_lock);
692 finish_wait(wq, &wait);
693 spin_lock(&bdev_lock);
702 * bd_start_claiming - start claiming a block device
703 * @bdev: block device of interest
704 * @holder: holder trying to claim @bdev
706 * @bdev is about to be opened exclusively. Check @bdev can be opened
707 * exclusively and mark that an exclusive open is in progress. Each
708 * successful call to this function must be matched with a call to
709 * either bd_finish_claiming() or bd_abort_claiming() (which do not
712 * This function is used to gain exclusive access to the block device
713 * without actually causing other exclusive open attempts to fail. It
714 * should be used when the open sequence itself requires exclusive
715 * access but may subsequently fail.
721 * Pointer to the block device containing @bdev on success, ERR_PTR()
724 static struct block_device *bd_start_claiming(struct block_device *bdev,
727 struct gendisk *disk;
728 struct block_device *whole;
734 * @bdev might not have been initialized properly yet, look up
735 * and grab the outer block device the hard way.
737 disk = get_gendisk(bdev->bd_dev, &partno);
739 return ERR_PTR(-ENXIO);
741 whole = bdget_disk(disk, 0);
742 module_put(disk->fops->owner);
745 return ERR_PTR(-ENOMEM);
747 /* prepare to claim, if successful, mark claiming in progress */
748 spin_lock(&bdev_lock);
750 err = bd_prepare_to_claim(bdev, whole, holder);
752 whole->bd_claiming = holder;
753 spin_unlock(&bdev_lock);
756 spin_unlock(&bdev_lock);
762 /* releases bdev_lock */
763 static void __bd_abort_claiming(struct block_device *whole, void *holder)
765 BUG_ON(whole->bd_claiming != holder);
766 whole->bd_claiming = NULL;
767 wake_up_bit(&whole->bd_claiming, 0);
769 spin_unlock(&bdev_lock);
774 * bd_abort_claiming - abort claiming a block device
775 * @whole: whole block device returned by bd_start_claiming()
776 * @holder: holder trying to claim @bdev
778 * Abort a claiming block started by bd_start_claiming(). Note that
779 * @whole is not the block device to be claimed but the whole device
780 * returned by bd_start_claiming().
783 * Grabs and releases bdev_lock.
785 static void bd_abort_claiming(struct block_device *whole, void *holder)
787 spin_lock(&bdev_lock);
788 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
791 /* increment holders when we have a legitimate claim. requires bdev_lock */
792 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
795 /* note that for a whole device bd_holders
796 * will be incremented twice, and bd_holder will
797 * be set to bd_claim before being set to holder
800 whole->bd_holder = bd_claim;
802 bdev->bd_holder = holder;
806 * bd_finish_claiming - finish claiming a block device
807 * @bdev: block device of interest (passed to bd_start_claiming())
808 * @whole: whole block device returned by bd_start_claiming()
809 * @holder: holder trying to claim @bdev
811 * Finish a claiming block started by bd_start_claiming().
814 * Grabs and releases bdev_lock.
816 static void bd_finish_claiming(struct block_device *bdev,
817 struct block_device *whole, void *holder)
819 spin_lock(&bdev_lock);
820 BUG_ON(!bd_may_claim(bdev, whole, holder));
821 __bd_claim(bdev, whole, holder);
822 __bd_abort_claiming(whole, holder); /* not actually an abort */
826 * bd_claim - claim a block device
827 * @bdev: block device to claim
828 * @holder: holder trying to claim @bdev
830 * Try to claim @bdev which must have been opened successfully.
836 * 0 if successful, -EBUSY if @bdev is already claimed.
838 int bd_claim(struct block_device *bdev, void *holder)
840 struct block_device *whole = bdev->bd_contains;
845 spin_lock(&bdev_lock);
846 res = bd_prepare_to_claim(bdev, whole, holder);
848 __bd_claim(bdev, whole, holder);
849 spin_unlock(&bdev_lock);
853 EXPORT_SYMBOL(bd_claim);
855 void bd_release(struct block_device *bdev)
857 spin_lock(&bdev_lock);
858 if (!--bdev->bd_contains->bd_holders)
859 bdev->bd_contains->bd_holder = NULL;
860 if (!--bdev->bd_holders)
861 bdev->bd_holder = NULL;
862 spin_unlock(&bdev_lock);
865 EXPORT_SYMBOL(bd_release);
869 * Functions for bd_claim_by_kobject / bd_release_from_kobject
871 * If a kobject is passed to bd_claim_by_kobject()
872 * and the kobject has a parent directory,
873 * following symlinks are created:
874 * o from the kobject to the claimed bdev
875 * o from "holders" directory of the bdev to the parent of the kobject
876 * bd_release_from_kobject() removes these symlinks.
879 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
880 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
881 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
882 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
885 static int add_symlink(struct kobject *from, struct kobject *to)
889 return sysfs_create_link(from, to, kobject_name(to));
892 static void del_symlink(struct kobject *from, struct kobject *to)
896 sysfs_remove_link(from, kobject_name(to));
900 * 'struct bd_holder' contains pointers to kobjects symlinked by
901 * bd_claim_by_kobject.
902 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
905 struct list_head list; /* chain of holders of the bdev */
906 int count; /* references from the holder */
907 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
908 struct kobject *hdev; /* e.g. "/block/dm-0" */
909 struct kobject *hdir; /* e.g. "/block/sda/holders" */
910 struct kobject *sdev; /* e.g. "/block/sda" */
914 * Get references of related kobjects at once.
915 * Returns 1 on success. 0 on failure.
917 * Should call bd_holder_release_dirs() after successful use.
919 static int bd_holder_grab_dirs(struct block_device *bdev,
920 struct bd_holder *bo)
925 bo->sdir = kobject_get(bo->sdir);
929 bo->hdev = kobject_get(bo->sdir->parent);
933 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
937 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
944 kobject_put(bo->sdev);
946 kobject_put(bo->hdev);
948 kobject_put(bo->sdir);
953 /* Put references of related kobjects at once. */
954 static void bd_holder_release_dirs(struct bd_holder *bo)
956 kobject_put(bo->hdir);
957 kobject_put(bo->sdev);
958 kobject_put(bo->hdev);
959 kobject_put(bo->sdir);
962 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
964 struct bd_holder *bo;
966 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
976 static void free_bd_holder(struct bd_holder *bo)
982 * find_bd_holder - find matching struct bd_holder from the block device
984 * @bdev: struct block device to be searched
985 * @bo: target struct bd_holder
987 * Returns matching entry with @bo in @bdev->bd_holder_list.
988 * If found, increment the reference count and return the pointer.
989 * If not found, returns NULL.
991 static struct bd_holder *find_bd_holder(struct block_device *bdev,
992 struct bd_holder *bo)
994 struct bd_holder *tmp;
996 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
997 if (tmp->sdir == bo->sdir) {
1006 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1008 * @bdev: block device to be bd_claimed
1009 * @bo: preallocated and initialized by alloc_bd_holder()
1011 * Add @bo to @bdev->bd_holder_list, create symlinks.
1013 * Returns 0 if symlinks are created.
1014 * Returns -ve if something fails.
1016 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1023 if (!bd_holder_grab_dirs(bdev, bo))
1026 err = add_symlink(bo->sdir, bo->sdev);
1030 err = add_symlink(bo->hdir, bo->hdev);
1032 del_symlink(bo->sdir, bo->sdev);
1036 list_add_tail(&bo->list, &bdev->bd_holder_list);
1041 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1043 * @bdev: block device to be bd_claimed
1044 * @kobj: holder's kobject
1046 * If there is matching entry with @kobj in @bdev->bd_holder_list
1047 * and no other bd_claim() from the same kobject,
1048 * remove the struct bd_holder from the list, delete symlinks for it.
1050 * Returns a pointer to the struct bd_holder when it's removed from the list
1051 * and ready to be freed.
1052 * Returns NULL if matching claim isn't found or there is other bd_claim()
1053 * by the same kobject.
1055 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1056 struct kobject *kobj)
1058 struct bd_holder *bo;
1060 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1061 if (bo->sdir == kobj) {
1063 BUG_ON(bo->count < 0);
1065 list_del(&bo->list);
1066 del_symlink(bo->sdir, bo->sdev);
1067 del_symlink(bo->hdir, bo->hdev);
1068 bd_holder_release_dirs(bo);
1079 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1081 * @bdev: block device to be claimed
1082 * @holder: holder's signature
1083 * @kobj: holder's kobject
1085 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1086 * the bdev and the holder's kobject.
1087 * Use bd_release_from_kobject() when relesing the claimed bdev.
1089 * Returns 0 on success. (same as bd_claim())
1090 * Returns errno on failure.
1092 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1093 struct kobject *kobj)
1096 struct bd_holder *bo, *found;
1101 bo = alloc_bd_holder(kobj);
1105 mutex_lock(&bdev->bd_mutex);
1107 err = bd_claim(bdev, holder);
1111 found = find_bd_holder(bdev, bo);
1115 err = add_bd_holder(bdev, bo);
1121 mutex_unlock(&bdev->bd_mutex);
1127 * bd_release_from_kobject - bd_release() with additional kobject signature
1129 * @bdev: block device to be released
1130 * @kobj: holder's kobject
1132 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1134 static void bd_release_from_kobject(struct block_device *bdev,
1135 struct kobject *kobj)
1140 mutex_lock(&bdev->bd_mutex);
1142 free_bd_holder(del_bd_holder(bdev, kobj));
1143 mutex_unlock(&bdev->bd_mutex);
1147 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1149 * @bdev: block device to be claimed
1150 * @holder: holder's signature
1151 * @disk: holder's gendisk
1153 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1155 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1156 struct gendisk *disk)
1158 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1160 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1163 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1165 * @bdev: block device to be claimed
1166 * @disk: holder's gendisk
1168 * Call bd_release_from_kobject() and put @disk->slave_dir.
1170 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1172 bd_release_from_kobject(bdev, disk->slave_dir);
1173 kobject_put(disk->slave_dir);
1175 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1179 * Tries to open block device by device number. Use it ONLY if you
1180 * really do not have anything better - i.e. when you are behind a
1181 * truly sucky interface and all you are given is a device number. _Never_
1182 * to be used for internal purposes. If you ever need it - reconsider
1185 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1187 struct block_device *bdev = bdget(dev);
1190 err = blkdev_get(bdev, mode);
1191 return err ? ERR_PTR(err) : bdev;
1194 EXPORT_SYMBOL(open_by_devnum);
1197 * flush_disk - invalidates all buffer-cache entries on a disk
1199 * @bdev: struct block device to be flushed
1201 * Invalidates all buffer-cache entries on a disk. It should be called
1202 * when a disk has been changed -- either by a media change or online
1205 static void flush_disk(struct block_device *bdev)
1207 if (__invalidate_device(bdev)) {
1208 char name[BDEVNAME_SIZE] = "";
1211 disk_name(bdev->bd_disk, 0, name);
1212 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1213 "resized disk %s\n", name);
1218 if (disk_partitionable(bdev->bd_disk))
1219 bdev->bd_invalidated = 1;
1223 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1224 * @disk: struct gendisk to check
1225 * @bdev: struct bdev to adjust.
1227 * This routine checks to see if the bdev size does not match the disk size
1228 * and adjusts it if it differs.
1230 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1232 loff_t disk_size, bdev_size;
1234 disk_size = (loff_t)get_capacity(disk) << 9;
1235 bdev_size = i_size_read(bdev->bd_inode);
1236 if (disk_size != bdev_size) {
1237 char name[BDEVNAME_SIZE];
1239 disk_name(disk, 0, name);
1241 "%s: detected capacity change from %lld to %lld\n",
1242 name, bdev_size, disk_size);
1243 i_size_write(bdev->bd_inode, disk_size);
1247 EXPORT_SYMBOL(check_disk_size_change);
1250 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1251 * @disk: struct gendisk to be revalidated
1253 * This routine is a wrapper for lower-level driver's revalidate_disk
1254 * call-backs. It is used to do common pre and post operations needed
1255 * for all revalidate_disk operations.
1257 int revalidate_disk(struct gendisk *disk)
1259 struct block_device *bdev;
1262 if (disk->fops->revalidate_disk)
1263 ret = disk->fops->revalidate_disk(disk);
1265 bdev = bdget_disk(disk, 0);
1269 mutex_lock(&bdev->bd_mutex);
1270 check_disk_size_change(disk, bdev);
1271 mutex_unlock(&bdev->bd_mutex);
1275 EXPORT_SYMBOL(revalidate_disk);
1278 * This routine checks whether a removable media has been changed,
1279 * and invalidates all buffer-cache-entries in that case. This
1280 * is a relatively slow routine, so we have to try to minimize using
1281 * it. Thus it is called only upon a 'mount' or 'open'. This
1282 * is the best way of combining speed and utility, I think.
1283 * People changing diskettes in the middle of an operation deserve
1286 int check_disk_change(struct block_device *bdev)
1288 struct gendisk *disk = bdev->bd_disk;
1289 const struct block_device_operations *bdops = disk->fops;
1291 if (!bdops->media_changed)
1293 if (!bdops->media_changed(bdev->bd_disk))
1297 if (bdops->revalidate_disk)
1298 bdops->revalidate_disk(bdev->bd_disk);
1302 EXPORT_SYMBOL(check_disk_change);
1304 void bd_set_size(struct block_device *bdev, loff_t size)
1306 unsigned bsize = bdev_logical_block_size(bdev);
1308 bdev->bd_inode->i_size = size;
1309 while (bsize < PAGE_CACHE_SIZE) {
1314 bdev->bd_block_size = bsize;
1315 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1317 EXPORT_SYMBOL(bd_set_size);
1319 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1324 * mutex_lock(part->bd_mutex)
1325 * mutex_lock_nested(whole->bd_mutex, 1)
1328 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1330 struct gendisk *disk;
1335 if (mode & FMODE_READ)
1337 if (mode & FMODE_WRITE)
1340 * hooks: /n/, see "layering violations".
1342 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1352 disk = get_gendisk(bdev->bd_dev, &partno);
1354 goto out_unlock_kernel;
1356 mutex_lock_nested(&bdev->bd_mutex, for_part);
1357 if (!bdev->bd_openers) {
1358 bdev->bd_disk = disk;
1359 bdev->bd_contains = bdev;
1361 struct backing_dev_info *bdi;
1364 bdev->bd_part = disk_get_part(disk, partno);
1368 if (disk->fops->open) {
1369 ret = disk->fops->open(bdev, mode);
1370 if (ret == -ERESTARTSYS) {
1371 /* Lost a race with 'disk' being
1372 * deleted, try again.
1375 disk_put_part(bdev->bd_part);
1376 bdev->bd_part = NULL;
1377 module_put(disk->fops->owner);
1379 bdev->bd_disk = NULL;
1380 mutex_unlock(&bdev->bd_mutex);
1386 if (!bdev->bd_openers) {
1387 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1388 bdi = blk_get_backing_dev_info(bdev);
1390 bdi = &default_backing_dev_info;
1391 bdev->bd_inode->i_data.backing_dev_info = bdi;
1393 if (bdev->bd_invalidated)
1394 rescan_partitions(disk, bdev);
1396 struct block_device *whole;
1397 whole = bdget_disk(disk, 0);
1402 ret = __blkdev_get(whole, mode, 1);
1405 bdev->bd_contains = whole;
1406 bdev->bd_inode->i_data.backing_dev_info =
1407 whole->bd_inode->i_data.backing_dev_info;
1408 bdev->bd_part = disk_get_part(disk, partno);
1409 if (!(disk->flags & GENHD_FL_UP) ||
1410 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1414 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1417 module_put(disk->fops->owner);
1420 if (bdev->bd_contains == bdev) {
1421 if (bdev->bd_disk->fops->open) {
1422 ret = bdev->bd_disk->fops->open(bdev, mode);
1424 goto out_unlock_bdev;
1426 if (bdev->bd_invalidated)
1427 rescan_partitions(bdev->bd_disk, bdev);
1432 bdev->bd_part_count++;
1433 mutex_unlock(&bdev->bd_mutex);
1438 disk_put_part(bdev->bd_part);
1439 bdev->bd_disk = NULL;
1440 bdev->bd_part = NULL;
1441 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1442 if (bdev != bdev->bd_contains)
1443 __blkdev_put(bdev->bd_contains, mode, 1);
1444 bdev->bd_contains = NULL;
1446 mutex_unlock(&bdev->bd_mutex);
1451 module_put(disk->fops->owner);
1458 int blkdev_get(struct block_device *bdev, fmode_t mode)
1460 return __blkdev_get(bdev, mode, 0);
1462 EXPORT_SYMBOL(blkdev_get);
1464 static int blkdev_open(struct inode * inode, struct file * filp)
1466 struct block_device *whole = NULL;
1467 struct block_device *bdev;
1471 * Preserve backwards compatibility and allow large file access
1472 * even if userspace doesn't ask for it explicitly. Some mkfs
1473 * binary needs it. We might want to drop this workaround
1474 * during an unstable branch.
1476 filp->f_flags |= O_LARGEFILE;
1478 if (filp->f_flags & O_NDELAY)
1479 filp->f_mode |= FMODE_NDELAY;
1480 if (filp->f_flags & O_EXCL)
1481 filp->f_mode |= FMODE_EXCL;
1482 if ((filp->f_flags & O_ACCMODE) == 3)
1483 filp->f_mode |= FMODE_WRITE_IOCTL;
1485 bdev = bd_acquire(inode);
1489 if (filp->f_mode & FMODE_EXCL) {
1490 whole = bd_start_claiming(bdev, filp);
1491 if (IS_ERR(whole)) {
1493 return PTR_ERR(whole);
1497 filp->f_mapping = bdev->bd_inode->i_mapping;
1499 res = blkdev_get(bdev, filp->f_mode);
1503 bd_finish_claiming(bdev, whole, filp);
1505 bd_abort_claiming(whole, filp);
1511 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1514 struct gendisk *disk = bdev->bd_disk;
1515 struct block_device *victim = NULL;
1517 mutex_lock_nested(&bdev->bd_mutex, for_part);
1520 bdev->bd_part_count--;
1522 if (!--bdev->bd_openers) {
1523 sync_blockdev(bdev);
1526 if (bdev->bd_contains == bdev) {
1527 if (disk->fops->release)
1528 ret = disk->fops->release(disk, mode);
1530 if (!bdev->bd_openers) {
1531 struct module *owner = disk->fops->owner;
1535 disk_put_part(bdev->bd_part);
1536 bdev->bd_part = NULL;
1537 bdev->bd_disk = NULL;
1538 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1539 if (bdev != bdev->bd_contains)
1540 victim = bdev->bd_contains;
1541 bdev->bd_contains = NULL;
1544 mutex_unlock(&bdev->bd_mutex);
1547 __blkdev_put(victim, mode, 1);
1551 int blkdev_put(struct block_device *bdev, fmode_t mode)
1553 return __blkdev_put(bdev, mode, 0);
1555 EXPORT_SYMBOL(blkdev_put);
1557 static int blkdev_close(struct inode * inode, struct file * filp)
1559 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1560 if (bdev->bd_holder == filp)
1562 return blkdev_put(bdev, filp->f_mode);
1565 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1567 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1568 fmode_t mode = file->f_mode;
1571 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1572 * to updated it before every ioctl.
1574 if (file->f_flags & O_NDELAY)
1575 mode |= FMODE_NDELAY;
1577 mode &= ~FMODE_NDELAY;
1579 return blkdev_ioctl(bdev, mode, cmd, arg);
1583 * Write data to the block device. Only intended for the block device itself
1584 * and the raw driver which basically is a fake block device.
1586 * Does not take i_mutex for the write and thus is not for general purpose
1589 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1590 unsigned long nr_segs, loff_t pos)
1592 struct file *file = iocb->ki_filp;
1595 BUG_ON(iocb->ki_pos != pos);
1597 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1598 if (ret > 0 || ret == -EIOCBQUEUED) {
1601 err = generic_write_sync(file, pos, ret);
1602 if (err < 0 && ret > 0)
1607 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1610 * Try to release a page associated with block device when the system
1611 * is under memory pressure.
1613 static int blkdev_releasepage(struct page *page, gfp_t wait)
1615 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1617 if (super && super->s_op->bdev_try_to_free_page)
1618 return super->s_op->bdev_try_to_free_page(super, page, wait);
1620 return try_to_free_buffers(page);
1623 static const struct address_space_operations def_blk_aops = {
1624 .readpage = blkdev_readpage,
1625 .writepage = blkdev_writepage,
1626 .sync_page = block_sync_page,
1627 .write_begin = blkdev_write_begin,
1628 .write_end = blkdev_write_end,
1629 .writepages = generic_writepages,
1630 .releasepage = blkdev_releasepage,
1631 .direct_IO = blkdev_direct_IO,
1634 const struct file_operations def_blk_fops = {
1635 .open = blkdev_open,
1636 .release = blkdev_close,
1637 .llseek = block_llseek,
1638 .read = do_sync_read,
1639 .write = do_sync_write,
1640 .aio_read = generic_file_aio_read,
1641 .aio_write = blkdev_aio_write,
1642 .mmap = generic_file_mmap,
1643 .fsync = blkdev_fsync,
1644 .unlocked_ioctl = block_ioctl,
1645 #ifdef CONFIG_COMPAT
1646 .compat_ioctl = compat_blkdev_ioctl,
1648 .splice_read = generic_file_splice_read,
1649 .splice_write = generic_file_splice_write,
1652 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1655 mm_segment_t old_fs = get_fs();
1657 res = blkdev_ioctl(bdev, 0, cmd, arg);
1662 EXPORT_SYMBOL(ioctl_by_bdev);
1665 * lookup_bdev - lookup a struct block_device by name
1666 * @pathname: special file representing the block device
1668 * Get a reference to the blockdevice at @pathname in the current
1669 * namespace if possible and return it. Return ERR_PTR(error)
1672 struct block_device *lookup_bdev(const char *pathname)
1674 struct block_device *bdev;
1675 struct inode *inode;
1679 if (!pathname || !*pathname)
1680 return ERR_PTR(-EINVAL);
1682 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1684 return ERR_PTR(error);
1686 inode = path.dentry->d_inode;
1688 if (!S_ISBLK(inode->i_mode))
1691 if (path.mnt->mnt_flags & MNT_NODEV)
1694 bdev = bd_acquire(inode);
1701 bdev = ERR_PTR(error);
1704 EXPORT_SYMBOL(lookup_bdev);
1707 * open_bdev_exclusive - open a block device by name and set it up for use
1709 * @path: special file representing the block device
1710 * @mode: FMODE_... combination to pass be used
1711 * @holder: owner for exclusion
1713 * Open the blockdevice described by the special file at @path, claim it
1716 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1718 struct block_device *bdev, *whole;
1721 bdev = lookup_bdev(path);
1725 whole = bd_start_claiming(bdev, holder);
1726 if (IS_ERR(whole)) {
1731 error = blkdev_get(bdev, mode);
1733 goto out_abort_claiming;
1736 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1737 goto out_blkdev_put;
1739 bd_finish_claiming(bdev, whole, holder);
1743 blkdev_put(bdev, mode);
1745 bd_abort_claiming(whole, holder);
1746 return ERR_PTR(error);
1749 EXPORT_SYMBOL(open_bdev_exclusive);
1752 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1754 * @bdev: blockdevice to close
1755 * @mode: mode, must match that used to open.
1757 * This is the counterpart to open_bdev_exclusive().
1759 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1762 blkdev_put(bdev, mode);
1765 EXPORT_SYMBOL(close_bdev_exclusive);
1767 int __invalidate_device(struct block_device *bdev)
1769 struct super_block *sb = get_super(bdev);
1774 * no need to lock the super, get_super holds the
1775 * read mutex so the filesystem cannot go away
1776 * under us (->put_super runs with the write lock
1779 shrink_dcache_sb(sb);
1780 res = invalidate_inodes(sb);
1783 invalidate_bdev(bdev);
1786 EXPORT_SYMBOL(__invalidate_device);