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)
311 return block_write_begin(mapping, pos, len, flags, pagep,
315 static int blkdev_write_end(struct file *file, struct address_space *mapping,
316 loff_t pos, unsigned len, unsigned copied,
317 struct page *page, void *fsdata)
320 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
323 page_cache_release(page);
330 * for a block special file file->f_path.dentry->d_inode->i_size is zero
331 * so we compute the size by hand (just as in block_read/write above)
333 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
335 struct inode *bd_inode = file->f_mapping->host;
339 mutex_lock(&bd_inode->i_mutex);
340 size = i_size_read(bd_inode);
347 offset += file->f_pos;
350 if (offset >= 0 && offset <= size) {
351 if (offset != file->f_pos) {
352 file->f_pos = offset;
356 mutex_unlock(&bd_inode->i_mutex);
360 int blkdev_fsync(struct file *filp, int datasync)
362 struct inode *bd_inode = filp->f_mapping->host;
363 struct block_device *bdev = I_BDEV(bd_inode);
367 * There is no need to serialise calls to blkdev_issue_flush with
368 * i_mutex and doing so causes performance issues with concurrent
369 * O_SYNC writers to a block device.
371 mutex_unlock(&bd_inode->i_mutex);
373 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL, BLKDEV_IFL_WAIT);
374 if (error == -EOPNOTSUPP)
377 mutex_lock(&bd_inode->i_mutex);
381 EXPORT_SYMBOL(blkdev_fsync);
387 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
388 static struct kmem_cache * bdev_cachep __read_mostly;
390 static struct inode *bdev_alloc_inode(struct super_block *sb)
392 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
395 return &ei->vfs_inode;
398 static void bdev_destroy_inode(struct inode *inode)
400 struct bdev_inode *bdi = BDEV_I(inode);
402 kmem_cache_free(bdev_cachep, bdi);
405 static void init_once(void *foo)
407 struct bdev_inode *ei = (struct bdev_inode *) foo;
408 struct block_device *bdev = &ei->bdev;
410 memset(bdev, 0, sizeof(*bdev));
411 mutex_init(&bdev->bd_mutex);
412 INIT_LIST_HEAD(&bdev->bd_inodes);
413 INIT_LIST_HEAD(&bdev->bd_list);
415 INIT_LIST_HEAD(&bdev->bd_holder_list);
417 inode_init_once(&ei->vfs_inode);
418 /* Initialize mutex for freeze. */
419 mutex_init(&bdev->bd_fsfreeze_mutex);
422 static inline void __bd_forget(struct inode *inode)
424 list_del_init(&inode->i_devices);
425 inode->i_bdev = NULL;
426 inode->i_mapping = &inode->i_data;
429 static void bdev_evict_inode(struct inode *inode)
431 struct block_device *bdev = &BDEV_I(inode)->bdev;
433 truncate_inode_pages(&inode->i_data, 0);
434 invalidate_inode_buffers(inode); /* is it needed here? */
435 end_writeback(inode);
436 spin_lock(&bdev_lock);
437 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
438 __bd_forget(list_entry(p, struct inode, i_devices));
440 list_del_init(&bdev->bd_list);
441 spin_unlock(&bdev_lock);
444 static const struct super_operations bdev_sops = {
445 .statfs = simple_statfs,
446 .alloc_inode = bdev_alloc_inode,
447 .destroy_inode = bdev_destroy_inode,
448 .drop_inode = generic_delete_inode,
449 .evict_inode = bdev_evict_inode,
452 static int bd_get_sb(struct file_system_type *fs_type,
453 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
455 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
458 static struct file_system_type bd_type = {
461 .kill_sb = kill_anon_super,
464 struct super_block *blockdev_superblock __read_mostly;
466 void __init bdev_cache_init(void)
469 struct vfsmount *bd_mnt;
471 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
472 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
473 SLAB_MEM_SPREAD|SLAB_PANIC),
475 err = register_filesystem(&bd_type);
477 panic("Cannot register bdev pseudo-fs");
478 bd_mnt = kern_mount(&bd_type);
480 panic("Cannot create bdev pseudo-fs");
482 * This vfsmount structure is only used to obtain the
483 * blockdev_superblock, so tell kmemleak not to report it.
485 kmemleak_not_leak(bd_mnt);
486 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
490 * Most likely _very_ bad one - but then it's hardly critical for small
491 * /dev and can be fixed when somebody will need really large one.
492 * Keep in mind that it will be fed through icache hash function too.
494 static inline unsigned long hash(dev_t dev)
496 return MAJOR(dev)+MINOR(dev);
499 static int bdev_test(struct inode *inode, void *data)
501 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
504 static int bdev_set(struct inode *inode, void *data)
506 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
510 static LIST_HEAD(all_bdevs);
512 struct block_device *bdget(dev_t dev)
514 struct block_device *bdev;
517 inode = iget5_locked(blockdev_superblock, hash(dev),
518 bdev_test, bdev_set, &dev);
523 bdev = &BDEV_I(inode)->bdev;
525 if (inode->i_state & I_NEW) {
526 bdev->bd_contains = NULL;
527 bdev->bd_inode = inode;
528 bdev->bd_block_size = (1 << inode->i_blkbits);
529 bdev->bd_part_count = 0;
530 bdev->bd_invalidated = 0;
531 inode->i_mode = S_IFBLK;
533 inode->i_bdev = bdev;
534 inode->i_data.a_ops = &def_blk_aops;
535 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
536 inode->i_data.backing_dev_info = &default_backing_dev_info;
537 spin_lock(&bdev_lock);
538 list_add(&bdev->bd_list, &all_bdevs);
539 spin_unlock(&bdev_lock);
540 unlock_new_inode(inode);
545 EXPORT_SYMBOL(bdget);
548 * bdgrab -- Grab a reference to an already referenced block device
549 * @bdev: Block device to grab a reference to.
551 struct block_device *bdgrab(struct block_device *bdev)
553 atomic_inc(&bdev->bd_inode->i_count);
557 long nr_blockdev_pages(void)
559 struct block_device *bdev;
561 spin_lock(&bdev_lock);
562 list_for_each_entry(bdev, &all_bdevs, bd_list) {
563 ret += bdev->bd_inode->i_mapping->nrpages;
565 spin_unlock(&bdev_lock);
569 void bdput(struct block_device *bdev)
571 iput(bdev->bd_inode);
574 EXPORT_SYMBOL(bdput);
576 static struct block_device *bd_acquire(struct inode *inode)
578 struct block_device *bdev;
580 spin_lock(&bdev_lock);
581 bdev = inode->i_bdev;
583 atomic_inc(&bdev->bd_inode->i_count);
584 spin_unlock(&bdev_lock);
587 spin_unlock(&bdev_lock);
589 bdev = bdget(inode->i_rdev);
591 spin_lock(&bdev_lock);
592 if (!inode->i_bdev) {
594 * We take an additional bd_inode->i_count for inode,
595 * and it's released in clear_inode() of inode.
596 * So, we can access it via ->i_mapping always
599 atomic_inc(&bdev->bd_inode->i_count);
600 inode->i_bdev = bdev;
601 inode->i_mapping = bdev->bd_inode->i_mapping;
602 list_add(&inode->i_devices, &bdev->bd_inodes);
604 spin_unlock(&bdev_lock);
609 /* Call when you free inode */
611 void bd_forget(struct inode *inode)
613 struct block_device *bdev = NULL;
615 spin_lock(&bdev_lock);
617 if (!sb_is_blkdev_sb(inode->i_sb))
618 bdev = inode->i_bdev;
621 spin_unlock(&bdev_lock);
624 iput(bdev->bd_inode);
628 * bd_may_claim - test whether a block device can be claimed
629 * @bdev: block device of interest
630 * @whole: whole block device containing @bdev, may equal @bdev
631 * @holder: holder trying to claim @bdev
633 * Test whther @bdev can be claimed by @holder.
636 * spin_lock(&bdev_lock).
639 * %true if @bdev can be claimed, %false otherwise.
641 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
644 if (bdev->bd_holder == holder)
645 return true; /* already a holder */
646 else if (bdev->bd_holder != NULL)
647 return false; /* held by someone else */
648 else if (bdev->bd_contains == bdev)
649 return true; /* is a whole device which isn't held */
651 else if (whole->bd_holder == bd_claim)
652 return true; /* is a partition of a device that is being partitioned */
653 else if (whole->bd_holder != NULL)
654 return false; /* is a partition of a held device */
656 return true; /* is a partition of an un-held device */
660 * bd_prepare_to_claim - prepare to claim a block device
661 * @bdev: block device of interest
662 * @whole: the whole device containing @bdev, may equal @bdev
663 * @holder: holder trying to claim @bdev
665 * Prepare to claim @bdev. This function fails if @bdev is already
666 * claimed by another holder and waits if another claiming is in
667 * progress. This function doesn't actually claim. On successful
668 * return, the caller has ownership of bd_claiming and bd_holder[s].
671 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
675 * 0 if @bdev can be claimed, -EBUSY otherwise.
677 static int bd_prepare_to_claim(struct block_device *bdev,
678 struct block_device *whole, void *holder)
681 /* if someone else claimed, fail */
682 if (!bd_may_claim(bdev, whole, holder))
685 /* if claiming is already in progress, wait for it to finish */
686 if (whole->bd_claiming) {
687 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
690 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
691 spin_unlock(&bdev_lock);
693 finish_wait(wq, &wait);
694 spin_lock(&bdev_lock);
703 * bd_start_claiming - start claiming a block device
704 * @bdev: block device of interest
705 * @holder: holder trying to claim @bdev
707 * @bdev is about to be opened exclusively. Check @bdev can be opened
708 * exclusively and mark that an exclusive open is in progress. Each
709 * successful call to this function must be matched with a call to
710 * either bd_finish_claiming() or bd_abort_claiming() (which do not
713 * This function is used to gain exclusive access to the block device
714 * without actually causing other exclusive open attempts to fail. It
715 * should be used when the open sequence itself requires exclusive
716 * access but may subsequently fail.
722 * Pointer to the block device containing @bdev on success, ERR_PTR()
725 static struct block_device *bd_start_claiming(struct block_device *bdev,
728 struct gendisk *disk;
729 struct block_device *whole;
735 * @bdev might not have been initialized properly yet, look up
736 * and grab the outer block device the hard way.
738 disk = get_gendisk(bdev->bd_dev, &partno);
740 return ERR_PTR(-ENXIO);
742 whole = bdget_disk(disk, 0);
743 module_put(disk->fops->owner);
746 return ERR_PTR(-ENOMEM);
748 /* prepare to claim, if successful, mark claiming in progress */
749 spin_lock(&bdev_lock);
751 err = bd_prepare_to_claim(bdev, whole, holder);
753 whole->bd_claiming = holder;
754 spin_unlock(&bdev_lock);
757 spin_unlock(&bdev_lock);
763 /* releases bdev_lock */
764 static void __bd_abort_claiming(struct block_device *whole, void *holder)
766 BUG_ON(whole->bd_claiming != holder);
767 whole->bd_claiming = NULL;
768 wake_up_bit(&whole->bd_claiming, 0);
770 spin_unlock(&bdev_lock);
775 * bd_abort_claiming - abort claiming a block device
776 * @whole: whole block device returned by bd_start_claiming()
777 * @holder: holder trying to claim @bdev
779 * Abort a claiming block started by bd_start_claiming(). Note that
780 * @whole is not the block device to be claimed but the whole device
781 * returned by bd_start_claiming().
784 * Grabs and releases bdev_lock.
786 static void bd_abort_claiming(struct block_device *whole, void *holder)
788 spin_lock(&bdev_lock);
789 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
792 /* increment holders when we have a legitimate claim. requires bdev_lock */
793 static void __bd_claim(struct block_device *bdev, struct block_device *whole,
796 /* note that for a whole device bd_holders
797 * will be incremented twice, and bd_holder will
798 * be set to bd_claim before being set to holder
801 whole->bd_holder = bd_claim;
803 bdev->bd_holder = holder;
807 * bd_finish_claiming - finish claiming a block device
808 * @bdev: block device of interest (passed to bd_start_claiming())
809 * @whole: whole block device returned by bd_start_claiming()
810 * @holder: holder trying to claim @bdev
812 * Finish a claiming block started by bd_start_claiming().
815 * Grabs and releases bdev_lock.
817 static void bd_finish_claiming(struct block_device *bdev,
818 struct block_device *whole, void *holder)
820 spin_lock(&bdev_lock);
821 BUG_ON(!bd_may_claim(bdev, whole, holder));
822 __bd_claim(bdev, whole, holder);
823 __bd_abort_claiming(whole, holder); /* not actually an abort */
827 * bd_claim - claim a block device
828 * @bdev: block device to claim
829 * @holder: holder trying to claim @bdev
831 * Try to claim @bdev which must have been opened successfully.
837 * 0 if successful, -EBUSY if @bdev is already claimed.
839 int bd_claim(struct block_device *bdev, void *holder)
841 struct block_device *whole = bdev->bd_contains;
846 spin_lock(&bdev_lock);
847 res = bd_prepare_to_claim(bdev, whole, holder);
849 __bd_claim(bdev, whole, holder);
850 spin_unlock(&bdev_lock);
854 EXPORT_SYMBOL(bd_claim);
856 void bd_release(struct block_device *bdev)
858 spin_lock(&bdev_lock);
859 if (!--bdev->bd_contains->bd_holders)
860 bdev->bd_contains->bd_holder = NULL;
861 if (!--bdev->bd_holders)
862 bdev->bd_holder = NULL;
863 spin_unlock(&bdev_lock);
866 EXPORT_SYMBOL(bd_release);
870 * Functions for bd_claim_by_kobject / bd_release_from_kobject
872 * If a kobject is passed to bd_claim_by_kobject()
873 * and the kobject has a parent directory,
874 * following symlinks are created:
875 * o from the kobject to the claimed bdev
876 * o from "holders" directory of the bdev to the parent of the kobject
877 * bd_release_from_kobject() removes these symlinks.
880 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
881 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
882 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
883 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
886 static int add_symlink(struct kobject *from, struct kobject *to)
890 return sysfs_create_link(from, to, kobject_name(to));
893 static void del_symlink(struct kobject *from, struct kobject *to)
897 sysfs_remove_link(from, kobject_name(to));
901 * 'struct bd_holder' contains pointers to kobjects symlinked by
902 * bd_claim_by_kobject.
903 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
906 struct list_head list; /* chain of holders of the bdev */
907 int count; /* references from the holder */
908 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
909 struct kobject *hdev; /* e.g. "/block/dm-0" */
910 struct kobject *hdir; /* e.g. "/block/sda/holders" */
911 struct kobject *sdev; /* e.g. "/block/sda" */
915 * Get references of related kobjects at once.
916 * Returns 1 on success. 0 on failure.
918 * Should call bd_holder_release_dirs() after successful use.
920 static int bd_holder_grab_dirs(struct block_device *bdev,
921 struct bd_holder *bo)
926 bo->sdir = kobject_get(bo->sdir);
930 bo->hdev = kobject_get(bo->sdir->parent);
934 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
938 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
945 kobject_put(bo->sdev);
947 kobject_put(bo->hdev);
949 kobject_put(bo->sdir);
954 /* Put references of related kobjects at once. */
955 static void bd_holder_release_dirs(struct bd_holder *bo)
957 kobject_put(bo->hdir);
958 kobject_put(bo->sdev);
959 kobject_put(bo->hdev);
960 kobject_put(bo->sdir);
963 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
965 struct bd_holder *bo;
967 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
977 static void free_bd_holder(struct bd_holder *bo)
983 * find_bd_holder - find matching struct bd_holder from the block device
985 * @bdev: struct block device to be searched
986 * @bo: target struct bd_holder
988 * Returns matching entry with @bo in @bdev->bd_holder_list.
989 * If found, increment the reference count and return the pointer.
990 * If not found, returns NULL.
992 static struct bd_holder *find_bd_holder(struct block_device *bdev,
993 struct bd_holder *bo)
995 struct bd_holder *tmp;
997 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
998 if (tmp->sdir == bo->sdir) {
1007 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1009 * @bdev: block device to be bd_claimed
1010 * @bo: preallocated and initialized by alloc_bd_holder()
1012 * Add @bo to @bdev->bd_holder_list, create symlinks.
1014 * Returns 0 if symlinks are created.
1015 * Returns -ve if something fails.
1017 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1024 if (!bd_holder_grab_dirs(bdev, bo))
1027 err = add_symlink(bo->sdir, bo->sdev);
1031 err = add_symlink(bo->hdir, bo->hdev);
1033 del_symlink(bo->sdir, bo->sdev);
1037 list_add_tail(&bo->list, &bdev->bd_holder_list);
1042 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1044 * @bdev: block device to be bd_claimed
1045 * @kobj: holder's kobject
1047 * If there is matching entry with @kobj in @bdev->bd_holder_list
1048 * and no other bd_claim() from the same kobject,
1049 * remove the struct bd_holder from the list, delete symlinks for it.
1051 * Returns a pointer to the struct bd_holder when it's removed from the list
1052 * and ready to be freed.
1053 * Returns NULL if matching claim isn't found or there is other bd_claim()
1054 * by the same kobject.
1056 static struct bd_holder *del_bd_holder(struct block_device *bdev,
1057 struct kobject *kobj)
1059 struct bd_holder *bo;
1061 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1062 if (bo->sdir == kobj) {
1064 BUG_ON(bo->count < 0);
1066 list_del(&bo->list);
1067 del_symlink(bo->sdir, bo->sdev);
1068 del_symlink(bo->hdir, bo->hdev);
1069 bd_holder_release_dirs(bo);
1080 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1082 * @bdev: block device to be claimed
1083 * @holder: holder's signature
1084 * @kobj: holder's kobject
1086 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1087 * the bdev and the holder's kobject.
1088 * Use bd_release_from_kobject() when relesing the claimed bdev.
1090 * Returns 0 on success. (same as bd_claim())
1091 * Returns errno on failure.
1093 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1094 struct kobject *kobj)
1097 struct bd_holder *bo, *found;
1102 bo = alloc_bd_holder(kobj);
1106 mutex_lock(&bdev->bd_mutex);
1108 err = bd_claim(bdev, holder);
1112 found = find_bd_holder(bdev, bo);
1116 err = add_bd_holder(bdev, bo);
1122 mutex_unlock(&bdev->bd_mutex);
1128 * bd_release_from_kobject - bd_release() with additional kobject signature
1130 * @bdev: block device to be released
1131 * @kobj: holder's kobject
1133 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1135 static void bd_release_from_kobject(struct block_device *bdev,
1136 struct kobject *kobj)
1141 mutex_lock(&bdev->bd_mutex);
1143 free_bd_holder(del_bd_holder(bdev, kobj));
1144 mutex_unlock(&bdev->bd_mutex);
1148 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1150 * @bdev: block device to be claimed
1151 * @holder: holder's signature
1152 * @disk: holder's gendisk
1154 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1156 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1157 struct gendisk *disk)
1159 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1161 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1164 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1166 * @bdev: block device to be claimed
1167 * @disk: holder's gendisk
1169 * Call bd_release_from_kobject() and put @disk->slave_dir.
1171 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1173 bd_release_from_kobject(bdev, disk->slave_dir);
1174 kobject_put(disk->slave_dir);
1176 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1180 * Tries to open block device by device number. Use it ONLY if you
1181 * really do not have anything better - i.e. when you are behind a
1182 * truly sucky interface and all you are given is a device number. _Never_
1183 * to be used for internal purposes. If you ever need it - reconsider
1186 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1188 struct block_device *bdev = bdget(dev);
1191 err = blkdev_get(bdev, mode);
1192 return err ? ERR_PTR(err) : bdev;
1195 EXPORT_SYMBOL(open_by_devnum);
1198 * flush_disk - invalidates all buffer-cache entries on a disk
1200 * @bdev: struct block device to be flushed
1202 * Invalidates all buffer-cache entries on a disk. It should be called
1203 * when a disk has been changed -- either by a media change or online
1206 static void flush_disk(struct block_device *bdev)
1208 if (__invalidate_device(bdev)) {
1209 char name[BDEVNAME_SIZE] = "";
1212 disk_name(bdev->bd_disk, 0, name);
1213 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1214 "resized disk %s\n", name);
1219 if (disk_partitionable(bdev->bd_disk))
1220 bdev->bd_invalidated = 1;
1224 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1225 * @disk: struct gendisk to check
1226 * @bdev: struct bdev to adjust.
1228 * This routine checks to see if the bdev size does not match the disk size
1229 * and adjusts it if it differs.
1231 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1233 loff_t disk_size, bdev_size;
1235 disk_size = (loff_t)get_capacity(disk) << 9;
1236 bdev_size = i_size_read(bdev->bd_inode);
1237 if (disk_size != bdev_size) {
1238 char name[BDEVNAME_SIZE];
1240 disk_name(disk, 0, name);
1242 "%s: detected capacity change from %lld to %lld\n",
1243 name, bdev_size, disk_size);
1244 i_size_write(bdev->bd_inode, disk_size);
1248 EXPORT_SYMBOL(check_disk_size_change);
1251 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1252 * @disk: struct gendisk to be revalidated
1254 * This routine is a wrapper for lower-level driver's revalidate_disk
1255 * call-backs. It is used to do common pre and post operations needed
1256 * for all revalidate_disk operations.
1258 int revalidate_disk(struct gendisk *disk)
1260 struct block_device *bdev;
1263 if (disk->fops->revalidate_disk)
1264 ret = disk->fops->revalidate_disk(disk);
1266 bdev = bdget_disk(disk, 0);
1270 mutex_lock(&bdev->bd_mutex);
1271 check_disk_size_change(disk, bdev);
1272 mutex_unlock(&bdev->bd_mutex);
1276 EXPORT_SYMBOL(revalidate_disk);
1279 * This routine checks whether a removable media has been changed,
1280 * and invalidates all buffer-cache-entries in that case. This
1281 * is a relatively slow routine, so we have to try to minimize using
1282 * it. Thus it is called only upon a 'mount' or 'open'. This
1283 * is the best way of combining speed and utility, I think.
1284 * People changing diskettes in the middle of an operation deserve
1287 int check_disk_change(struct block_device *bdev)
1289 struct gendisk *disk = bdev->bd_disk;
1290 const struct block_device_operations *bdops = disk->fops;
1292 if (!bdops->media_changed)
1294 if (!bdops->media_changed(bdev->bd_disk))
1298 if (bdops->revalidate_disk)
1299 bdops->revalidate_disk(bdev->bd_disk);
1303 EXPORT_SYMBOL(check_disk_change);
1305 void bd_set_size(struct block_device *bdev, loff_t size)
1307 unsigned bsize = bdev_logical_block_size(bdev);
1309 bdev->bd_inode->i_size = size;
1310 while (bsize < PAGE_CACHE_SIZE) {
1315 bdev->bd_block_size = bsize;
1316 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1318 EXPORT_SYMBOL(bd_set_size);
1320 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1325 * mutex_lock(part->bd_mutex)
1326 * mutex_lock_nested(whole->bd_mutex, 1)
1329 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1331 struct gendisk *disk;
1336 if (mode & FMODE_READ)
1338 if (mode & FMODE_WRITE)
1341 * hooks: /n/, see "layering violations".
1344 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1354 disk = get_gendisk(bdev->bd_dev, &partno);
1358 mutex_lock_nested(&bdev->bd_mutex, for_part);
1359 if (!bdev->bd_openers) {
1360 bdev->bd_disk = disk;
1361 bdev->bd_contains = bdev;
1363 struct backing_dev_info *bdi;
1366 bdev->bd_part = disk_get_part(disk, partno);
1370 if (disk->fops->open) {
1371 ret = disk->fops->open(bdev, mode);
1372 if (ret == -ERESTARTSYS) {
1373 /* Lost a race with 'disk' being
1374 * deleted, try again.
1377 disk_put_part(bdev->bd_part);
1378 bdev->bd_part = NULL;
1379 module_put(disk->fops->owner);
1381 bdev->bd_disk = NULL;
1382 mutex_unlock(&bdev->bd_mutex);
1388 if (!bdev->bd_openers) {
1389 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1390 bdi = blk_get_backing_dev_info(bdev);
1392 bdi = &default_backing_dev_info;
1393 bdev->bd_inode->i_data.backing_dev_info = bdi;
1395 if (bdev->bd_invalidated)
1396 rescan_partitions(disk, bdev);
1398 struct block_device *whole;
1399 whole = bdget_disk(disk, 0);
1404 ret = __blkdev_get(whole, mode, 1);
1407 bdev->bd_contains = whole;
1408 bdev->bd_inode->i_data.backing_dev_info =
1409 whole->bd_inode->i_data.backing_dev_info;
1410 bdev->bd_part = disk_get_part(disk, partno);
1411 if (!(disk->flags & GENHD_FL_UP) ||
1412 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1416 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1419 module_put(disk->fops->owner);
1422 if (bdev->bd_contains == bdev) {
1423 if (bdev->bd_disk->fops->open) {
1424 ret = bdev->bd_disk->fops->open(bdev, mode);
1426 goto out_unlock_bdev;
1428 if (bdev->bd_invalidated)
1429 rescan_partitions(bdev->bd_disk, bdev);
1434 bdev->bd_part_count++;
1435 mutex_unlock(&bdev->bd_mutex);
1439 disk_put_part(bdev->bd_part);
1440 bdev->bd_disk = NULL;
1441 bdev->bd_part = NULL;
1442 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1443 if (bdev != bdev->bd_contains)
1444 __blkdev_put(bdev->bd_contains, mode, 1);
1445 bdev->bd_contains = NULL;
1447 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);
1518 bdev->bd_part_count--;
1520 if (!--bdev->bd_openers) {
1521 sync_blockdev(bdev);
1524 if (bdev->bd_contains == bdev) {
1525 if (disk->fops->release)
1526 ret = disk->fops->release(disk, mode);
1528 if (!bdev->bd_openers) {
1529 struct module *owner = disk->fops->owner;
1533 disk_put_part(bdev->bd_part);
1534 bdev->bd_part = NULL;
1535 bdev->bd_disk = NULL;
1536 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1537 if (bdev != bdev->bd_contains)
1538 victim = bdev->bd_contains;
1539 bdev->bd_contains = NULL;
1541 mutex_unlock(&bdev->bd_mutex);
1544 __blkdev_put(victim, mode, 1);
1548 int blkdev_put(struct block_device *bdev, fmode_t mode)
1550 return __blkdev_put(bdev, mode, 0);
1552 EXPORT_SYMBOL(blkdev_put);
1554 static int blkdev_close(struct inode * inode, struct file * filp)
1556 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1557 if (bdev->bd_holder == filp)
1559 return blkdev_put(bdev, filp->f_mode);
1562 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1564 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1565 fmode_t mode = file->f_mode;
1568 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1569 * to updated it before every ioctl.
1571 if (file->f_flags & O_NDELAY)
1572 mode |= FMODE_NDELAY;
1574 mode &= ~FMODE_NDELAY;
1576 return blkdev_ioctl(bdev, mode, cmd, arg);
1580 * Write data to the block device. Only intended for the block device itself
1581 * and the raw driver which basically is a fake block device.
1583 * Does not take i_mutex for the write and thus is not for general purpose
1586 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1587 unsigned long nr_segs, loff_t pos)
1589 struct file *file = iocb->ki_filp;
1592 BUG_ON(iocb->ki_pos != pos);
1594 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1595 if (ret > 0 || ret == -EIOCBQUEUED) {
1598 err = generic_write_sync(file, pos, ret);
1599 if (err < 0 && ret > 0)
1604 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1607 * Try to release a page associated with block device when the system
1608 * is under memory pressure.
1610 static int blkdev_releasepage(struct page *page, gfp_t wait)
1612 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1614 if (super && super->s_op->bdev_try_to_free_page)
1615 return super->s_op->bdev_try_to_free_page(super, page, wait);
1617 return try_to_free_buffers(page);
1620 static const struct address_space_operations def_blk_aops = {
1621 .readpage = blkdev_readpage,
1622 .writepage = blkdev_writepage,
1623 .sync_page = block_sync_page,
1624 .write_begin = blkdev_write_begin,
1625 .write_end = blkdev_write_end,
1626 .writepages = generic_writepages,
1627 .releasepage = blkdev_releasepage,
1628 .direct_IO = blkdev_direct_IO,
1631 const struct file_operations def_blk_fops = {
1632 .open = blkdev_open,
1633 .release = blkdev_close,
1634 .llseek = block_llseek,
1635 .read = do_sync_read,
1636 .write = do_sync_write,
1637 .aio_read = generic_file_aio_read,
1638 .aio_write = blkdev_aio_write,
1639 .mmap = generic_file_mmap,
1640 .fsync = blkdev_fsync,
1641 .unlocked_ioctl = block_ioctl,
1642 #ifdef CONFIG_COMPAT
1643 .compat_ioctl = compat_blkdev_ioctl,
1645 .splice_read = generic_file_splice_read,
1646 .splice_write = generic_file_splice_write,
1649 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1652 mm_segment_t old_fs = get_fs();
1654 res = blkdev_ioctl(bdev, 0, cmd, arg);
1659 EXPORT_SYMBOL(ioctl_by_bdev);
1662 * lookup_bdev - lookup a struct block_device by name
1663 * @pathname: special file representing the block device
1665 * Get a reference to the blockdevice at @pathname in the current
1666 * namespace if possible and return it. Return ERR_PTR(error)
1669 struct block_device *lookup_bdev(const char *pathname)
1671 struct block_device *bdev;
1672 struct inode *inode;
1676 if (!pathname || !*pathname)
1677 return ERR_PTR(-EINVAL);
1679 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1681 return ERR_PTR(error);
1683 inode = path.dentry->d_inode;
1685 if (!S_ISBLK(inode->i_mode))
1688 if (path.mnt->mnt_flags & MNT_NODEV)
1691 bdev = bd_acquire(inode);
1698 bdev = ERR_PTR(error);
1701 EXPORT_SYMBOL(lookup_bdev);
1704 * open_bdev_exclusive - open a block device by name and set it up for use
1706 * @path: special file representing the block device
1707 * @mode: FMODE_... combination to pass be used
1708 * @holder: owner for exclusion
1710 * Open the blockdevice described by the special file at @path, claim it
1713 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1715 struct block_device *bdev, *whole;
1718 bdev = lookup_bdev(path);
1722 whole = bd_start_claiming(bdev, holder);
1723 if (IS_ERR(whole)) {
1728 error = blkdev_get(bdev, mode);
1730 goto out_abort_claiming;
1733 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1734 goto out_blkdev_put;
1736 bd_finish_claiming(bdev, whole, holder);
1740 blkdev_put(bdev, mode);
1742 bd_abort_claiming(whole, holder);
1743 return ERR_PTR(error);
1746 EXPORT_SYMBOL(open_bdev_exclusive);
1749 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1751 * @bdev: blockdevice to close
1752 * @mode: mode, must match that used to open.
1754 * This is the counterpart to open_bdev_exclusive().
1756 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1759 blkdev_put(bdev, mode);
1762 EXPORT_SYMBOL(close_bdev_exclusive);
1764 int __invalidate_device(struct block_device *bdev)
1766 struct super_block *sb = get_super(bdev);
1771 * no need to lock the super, get_super holds the
1772 * read mutex so the filesystem cannot go away
1773 * under us (->put_super runs with the write lock
1776 shrink_dcache_sb(sb);
1777 res = invalidate_inodes(sb);
1780 invalidate_bdev(bdev);
1783 EXPORT_SYMBOL(__invalidate_device);