4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/smp_lock.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
51 static sector_t max_block(struct block_device *bdev)
53 sector_t retval = ~((sector_t)0);
54 loff_t sz = i_size_read(bdev->bd_inode);
57 unsigned int size = block_size(bdev);
58 unsigned int sizebits = blksize_bits(size);
59 retval = (sz >> sizebits);
64 /* Kill _all_ buffers and pagecache , dirty or not.. */
65 static void kill_bdev(struct block_device *bdev)
67 if (bdev->bd_inode->i_mapping->nrpages == 0)
70 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
73 int set_blocksize(struct block_device *bdev, int size)
75 /* Size must be a power of two, and between 512 and PAGE_SIZE */
76 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
79 /* Size cannot be smaller than the size supported by the device */
80 if (size < bdev_logical_block_size(bdev))
83 /* Don't change the size if it is same as current */
84 if (bdev->bd_block_size != size) {
86 bdev->bd_block_size = size;
87 bdev->bd_inode->i_blkbits = blksize_bits(size);
93 EXPORT_SYMBOL(set_blocksize);
95 int sb_set_blocksize(struct super_block *sb, int size)
97 if (set_blocksize(sb->s_bdev, size))
99 /* If we get here, we know size is power of two
100 * and it's value is between 512 and PAGE_SIZE */
101 sb->s_blocksize = size;
102 sb->s_blocksize_bits = blksize_bits(size);
103 return sb->s_blocksize;
106 EXPORT_SYMBOL(sb_set_blocksize);
108 int sb_min_blocksize(struct super_block *sb, int size)
110 int minsize = bdev_logical_block_size(sb->s_bdev);
113 return sb_set_blocksize(sb, size);
116 EXPORT_SYMBOL(sb_min_blocksize);
119 blkdev_get_block(struct inode *inode, sector_t iblock,
120 struct buffer_head *bh, int create)
122 if (iblock >= max_block(I_BDEV(inode))) {
127 * for reads, we're just trying to fill a partial page.
128 * return a hole, they will have to call get_block again
129 * before they can fill it, and they will get -EIO at that
134 bh->b_bdev = I_BDEV(inode);
135 bh->b_blocknr = iblock;
136 set_buffer_mapped(bh);
141 blkdev_get_blocks(struct inode *inode, sector_t iblock,
142 struct buffer_head *bh, int create)
144 sector_t end_block = max_block(I_BDEV(inode));
145 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
147 if ((iblock + max_blocks) > end_block) {
148 max_blocks = end_block - iblock;
149 if ((long)max_blocks <= 0) {
151 return -EIO; /* write fully beyond EOF */
153 * It is a read which is fully beyond EOF. We return
154 * a !buffer_mapped buffer
160 bh->b_bdev = I_BDEV(inode);
161 bh->b_blocknr = iblock;
162 bh->b_size = max_blocks << inode->i_blkbits;
164 set_buffer_mapped(bh);
169 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
170 loff_t offset, unsigned long nr_segs)
172 struct file *file = iocb->ki_filp;
173 struct inode *inode = file->f_mapping->host;
175 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
176 iov, offset, nr_segs, blkdev_get_blocks, NULL);
179 int __sync_blockdev(struct block_device *bdev, int wait)
184 return filemap_flush(bdev->bd_inode->i_mapping);
185 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
189 * Write out and wait upon all the dirty data associated with a block
190 * device via its mapping. Does not take the superblock lock.
192 int sync_blockdev(struct block_device *bdev)
194 return __sync_blockdev(bdev, 1);
196 EXPORT_SYMBOL(sync_blockdev);
199 * Write out and wait upon all dirty data associated with this
200 * device. Filesystem data as well as the underlying block
201 * device. Takes the superblock lock.
203 int fsync_bdev(struct block_device *bdev)
205 struct super_block *sb = get_super(bdev);
207 int res = sync_filesystem(sb);
211 return sync_blockdev(bdev);
213 EXPORT_SYMBOL(fsync_bdev);
216 * freeze_bdev -- lock a filesystem and force it into a consistent state
217 * @bdev: blockdevice to lock
219 * If a superblock is found on this device, we take the s_umount semaphore
220 * on it to make sure nobody unmounts until the snapshot creation is done.
221 * The reference counter (bd_fsfreeze_count) guarantees that only the last
222 * unfreeze process can unfreeze the frozen filesystem actually when multiple
223 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
224 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
227 struct super_block *freeze_bdev(struct block_device *bdev)
229 struct super_block *sb;
232 mutex_lock(&bdev->bd_fsfreeze_mutex);
233 if (++bdev->bd_fsfreeze_count > 1) {
235 * We don't even need to grab a reference - the first call
236 * to freeze_bdev grab an active reference and only the last
237 * thaw_bdev drops it.
239 sb = get_super(bdev);
241 mutex_unlock(&bdev->bd_fsfreeze_mutex);
245 sb = get_active_super(bdev);
248 if (sb->s_flags & MS_RDONLY) {
249 sb->s_frozen = SB_FREEZE_TRANS;
250 up_write(&sb->s_umount);
251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
255 sb->s_frozen = SB_FREEZE_WRITE;
260 sb->s_frozen = SB_FREEZE_TRANS;
263 sync_blockdev(sb->s_bdev);
265 if (sb->s_op->freeze_fs) {
266 error = sb->s_op->freeze_fs(sb);
269 "VFS:Filesystem freeze failed\n");
270 sb->s_frozen = SB_UNFROZEN;
271 deactivate_locked_super(sb);
272 bdev->bd_fsfreeze_count--;
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
274 return ERR_PTR(error);
277 up_write(&sb->s_umount);
281 mutex_unlock(&bdev->bd_fsfreeze_mutex);
282 return sb; /* thaw_bdev releases s->s_umount */
284 EXPORT_SYMBOL(freeze_bdev);
287 * thaw_bdev -- unlock filesystem
288 * @bdev: blockdevice to unlock
289 * @sb: associated superblock
291 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
293 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
297 mutex_lock(&bdev->bd_fsfreeze_mutex);
298 if (!bdev->bd_fsfreeze_count)
302 if (--bdev->bd_fsfreeze_count > 0)
308 BUG_ON(sb->s_bdev != bdev);
309 down_write(&sb->s_umount);
310 if (sb->s_flags & MS_RDONLY)
313 if (sb->s_op->unfreeze_fs) {
314 error = sb->s_op->unfreeze_fs(sb);
317 "VFS:Filesystem thaw failed\n");
318 sb->s_frozen = SB_FREEZE_TRANS;
319 bdev->bd_fsfreeze_count++;
320 mutex_unlock(&bdev->bd_fsfreeze_mutex);
326 sb->s_frozen = SB_UNFROZEN;
328 wake_up(&sb->s_wait_unfrozen);
331 deactivate_locked_super(sb);
333 mutex_unlock(&bdev->bd_fsfreeze_mutex);
336 EXPORT_SYMBOL(thaw_bdev);
338 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
340 return block_write_full_page(page, blkdev_get_block, wbc);
343 static int blkdev_readpage(struct file * file, struct page * page)
345 return block_read_full_page(page, blkdev_get_block);
348 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
353 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
357 static int blkdev_write_end(struct file *file, struct address_space *mapping,
358 loff_t pos, unsigned len, unsigned copied,
359 struct page *page, void *fsdata)
362 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
365 page_cache_release(page);
372 * for a block special file file->f_path.dentry->d_inode->i_size is zero
373 * so we compute the size by hand (just as in block_read/write above)
375 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
377 struct inode *bd_inode = file->f_mapping->host;
381 mutex_lock(&bd_inode->i_mutex);
382 size = i_size_read(bd_inode);
389 offset += file->f_pos;
392 if (offset >= 0 && offset <= size) {
393 if (offset != file->f_pos) {
394 file->f_pos = offset;
398 mutex_unlock(&bd_inode->i_mutex);
403 * Filp is never NULL; the only case when ->fsync() is called with
404 * NULL first argument is nfsd_sync_dir() and that's not a directory.
407 int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync)
409 struct inode *bd_inode = filp->f_mapping->host;
410 struct block_device *bdev = I_BDEV(bd_inode);
414 * There is no need to serialise calls to blkdev_issue_flush with
415 * i_mutex and doing so causes performance issues with concurrent
416 * O_SYNC writers to a block device.
418 mutex_unlock(&bd_inode->i_mutex);
420 error = blkdev_issue_flush(bdev, NULL);
421 if (error == -EOPNOTSUPP)
424 mutex_lock(&bd_inode->i_mutex);
428 EXPORT_SYMBOL(blkdev_fsync);
434 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
435 static struct kmem_cache * bdev_cachep __read_mostly;
437 static struct inode *bdev_alloc_inode(struct super_block *sb)
439 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
442 return &ei->vfs_inode;
445 static void bdev_destroy_inode(struct inode *inode)
447 struct bdev_inode *bdi = BDEV_I(inode);
449 kmem_cache_free(bdev_cachep, bdi);
452 static void init_once(void *foo)
454 struct bdev_inode *ei = (struct bdev_inode *) foo;
455 struct block_device *bdev = &ei->bdev;
457 memset(bdev, 0, sizeof(*bdev));
458 mutex_init(&bdev->bd_mutex);
459 INIT_LIST_HEAD(&bdev->bd_inodes);
460 INIT_LIST_HEAD(&bdev->bd_list);
462 INIT_LIST_HEAD(&bdev->bd_holder_list);
464 inode_init_once(&ei->vfs_inode);
465 /* Initialize mutex for freeze. */
466 mutex_init(&bdev->bd_fsfreeze_mutex);
469 static inline void __bd_forget(struct inode *inode)
471 list_del_init(&inode->i_devices);
472 inode->i_bdev = NULL;
473 inode->i_mapping = &inode->i_data;
476 static void bdev_clear_inode(struct inode *inode)
478 struct block_device *bdev = &BDEV_I(inode)->bdev;
480 spin_lock(&bdev_lock);
481 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
482 __bd_forget(list_entry(p, struct inode, i_devices));
484 list_del_init(&bdev->bd_list);
485 spin_unlock(&bdev_lock);
488 static const struct super_operations bdev_sops = {
489 .statfs = simple_statfs,
490 .alloc_inode = bdev_alloc_inode,
491 .destroy_inode = bdev_destroy_inode,
492 .drop_inode = generic_delete_inode,
493 .clear_inode = bdev_clear_inode,
496 static int bd_get_sb(struct file_system_type *fs_type,
497 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
499 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
502 static struct file_system_type bd_type = {
505 .kill_sb = kill_anon_super,
508 struct super_block *blockdev_superblock __read_mostly;
510 void __init bdev_cache_init(void)
513 struct vfsmount *bd_mnt;
515 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
516 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
517 SLAB_MEM_SPREAD|SLAB_PANIC),
519 err = register_filesystem(&bd_type);
521 panic("Cannot register bdev pseudo-fs");
522 bd_mnt = kern_mount(&bd_type);
524 panic("Cannot create bdev pseudo-fs");
526 * This vfsmount structure is only used to obtain the
527 * blockdev_superblock, so tell kmemleak not to report it.
529 kmemleak_not_leak(bd_mnt);
530 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
534 * Most likely _very_ bad one - but then it's hardly critical for small
535 * /dev and can be fixed when somebody will need really large one.
536 * Keep in mind that it will be fed through icache hash function too.
538 static inline unsigned long hash(dev_t dev)
540 return MAJOR(dev)+MINOR(dev);
543 static int bdev_test(struct inode *inode, void *data)
545 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
548 static int bdev_set(struct inode *inode, void *data)
550 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
554 static LIST_HEAD(all_bdevs);
556 struct block_device *bdget(dev_t dev)
558 struct block_device *bdev;
561 inode = iget5_locked(blockdev_superblock, hash(dev),
562 bdev_test, bdev_set, &dev);
567 bdev = &BDEV_I(inode)->bdev;
569 if (inode->i_state & I_NEW) {
570 bdev->bd_contains = NULL;
571 bdev->bd_inode = inode;
572 bdev->bd_block_size = (1 << inode->i_blkbits);
573 bdev->bd_part_count = 0;
574 bdev->bd_invalidated = 0;
575 inode->i_mode = S_IFBLK;
577 inode->i_bdev = bdev;
578 inode->i_data.a_ops = &def_blk_aops;
579 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
580 inode->i_data.backing_dev_info = &default_backing_dev_info;
581 spin_lock(&bdev_lock);
582 list_add(&bdev->bd_list, &all_bdevs);
583 spin_unlock(&bdev_lock);
584 unlock_new_inode(inode);
589 EXPORT_SYMBOL(bdget);
592 * bdgrab -- Grab a reference to an already referenced block device
593 * @bdev: Block device to grab a reference to.
595 struct block_device *bdgrab(struct block_device *bdev)
597 atomic_inc(&bdev->bd_inode->i_count);
601 long nr_blockdev_pages(void)
603 struct block_device *bdev;
605 spin_lock(&bdev_lock);
606 list_for_each_entry(bdev, &all_bdevs, bd_list) {
607 ret += bdev->bd_inode->i_mapping->nrpages;
609 spin_unlock(&bdev_lock);
613 void bdput(struct block_device *bdev)
615 iput(bdev->bd_inode);
618 EXPORT_SYMBOL(bdput);
620 static struct block_device *bd_acquire(struct inode *inode)
622 struct block_device *bdev;
624 spin_lock(&bdev_lock);
625 bdev = inode->i_bdev;
627 atomic_inc(&bdev->bd_inode->i_count);
628 spin_unlock(&bdev_lock);
631 spin_unlock(&bdev_lock);
633 bdev = bdget(inode->i_rdev);
635 spin_lock(&bdev_lock);
636 if (!inode->i_bdev) {
638 * We take an additional bd_inode->i_count for inode,
639 * and it's released in clear_inode() of inode.
640 * So, we can access it via ->i_mapping always
643 atomic_inc(&bdev->bd_inode->i_count);
644 inode->i_bdev = bdev;
645 inode->i_mapping = bdev->bd_inode->i_mapping;
646 list_add(&inode->i_devices, &bdev->bd_inodes);
648 spin_unlock(&bdev_lock);
653 /* Call when you free inode */
655 void bd_forget(struct inode *inode)
657 struct block_device *bdev = NULL;
659 spin_lock(&bdev_lock);
661 if (!sb_is_blkdev_sb(inode->i_sb))
662 bdev = inode->i_bdev;
665 spin_unlock(&bdev_lock);
668 iput(bdev->bd_inode);
671 int bd_claim(struct block_device *bdev, void *holder)
674 spin_lock(&bdev_lock);
676 /* first decide result */
677 if (bdev->bd_holder == holder)
678 res = 0; /* already a holder */
679 else if (bdev->bd_holder != NULL)
680 res = -EBUSY; /* held by someone else */
681 else if (bdev->bd_contains == bdev)
682 res = 0; /* is a whole device which isn't held */
684 else if (bdev->bd_contains->bd_holder == bd_claim)
685 res = 0; /* is a partition of a device that is being partitioned */
686 else if (bdev->bd_contains->bd_holder != NULL)
687 res = -EBUSY; /* is a partition of a held device */
689 res = 0; /* is a partition of an un-held device */
691 /* now impose change */
693 /* note that for a whole device bd_holders
694 * will be incremented twice, and bd_holder will
695 * be set to bd_claim before being set to holder
697 bdev->bd_contains->bd_holders ++;
698 bdev->bd_contains->bd_holder = bd_claim;
700 bdev->bd_holder = holder;
702 spin_unlock(&bdev_lock);
706 EXPORT_SYMBOL(bd_claim);
708 void bd_release(struct block_device *bdev)
710 spin_lock(&bdev_lock);
711 if (!--bdev->bd_contains->bd_holders)
712 bdev->bd_contains->bd_holder = NULL;
713 if (!--bdev->bd_holders)
714 bdev->bd_holder = NULL;
715 spin_unlock(&bdev_lock);
718 EXPORT_SYMBOL(bd_release);
722 * Functions for bd_claim_by_kobject / bd_release_from_kobject
724 * If a kobject is passed to bd_claim_by_kobject()
725 * and the kobject has a parent directory,
726 * following symlinks are created:
727 * o from the kobject to the claimed bdev
728 * o from "holders" directory of the bdev to the parent of the kobject
729 * bd_release_from_kobject() removes these symlinks.
732 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
733 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
734 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
735 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
738 static int add_symlink(struct kobject *from, struct kobject *to)
742 return sysfs_create_link(from, to, kobject_name(to));
745 static void del_symlink(struct kobject *from, struct kobject *to)
749 sysfs_remove_link(from, kobject_name(to));
753 * 'struct bd_holder' contains pointers to kobjects symlinked by
754 * bd_claim_by_kobject.
755 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
758 struct list_head list; /* chain of holders of the bdev */
759 int count; /* references from the holder */
760 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
761 struct kobject *hdev; /* e.g. "/block/dm-0" */
762 struct kobject *hdir; /* e.g. "/block/sda/holders" */
763 struct kobject *sdev; /* e.g. "/block/sda" */
767 * Get references of related kobjects at once.
768 * Returns 1 on success. 0 on failure.
770 * Should call bd_holder_release_dirs() after successful use.
772 static int bd_holder_grab_dirs(struct block_device *bdev,
773 struct bd_holder *bo)
778 bo->sdir = kobject_get(bo->sdir);
782 bo->hdev = kobject_get(bo->sdir->parent);
786 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
790 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
797 kobject_put(bo->sdev);
799 kobject_put(bo->hdev);
801 kobject_put(bo->sdir);
806 /* Put references of related kobjects at once. */
807 static void bd_holder_release_dirs(struct bd_holder *bo)
809 kobject_put(bo->hdir);
810 kobject_put(bo->sdev);
811 kobject_put(bo->hdev);
812 kobject_put(bo->sdir);
815 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
817 struct bd_holder *bo;
819 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
829 static void free_bd_holder(struct bd_holder *bo)
835 * find_bd_holder - find matching struct bd_holder from the block device
837 * @bdev: struct block device to be searched
838 * @bo: target struct bd_holder
840 * Returns matching entry with @bo in @bdev->bd_holder_list.
841 * If found, increment the reference count and return the pointer.
842 * If not found, returns NULL.
844 static struct bd_holder *find_bd_holder(struct block_device *bdev,
845 struct bd_holder *bo)
847 struct bd_holder *tmp;
849 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
850 if (tmp->sdir == bo->sdir) {
859 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
861 * @bdev: block device to be bd_claimed
862 * @bo: preallocated and initialized by alloc_bd_holder()
864 * Add @bo to @bdev->bd_holder_list, create symlinks.
866 * Returns 0 if symlinks are created.
867 * Returns -ve if something fails.
869 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
876 if (!bd_holder_grab_dirs(bdev, bo))
879 err = add_symlink(bo->sdir, bo->sdev);
883 err = add_symlink(bo->hdir, bo->hdev);
885 del_symlink(bo->sdir, bo->sdev);
889 list_add_tail(&bo->list, &bdev->bd_holder_list);
894 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
896 * @bdev: block device to be bd_claimed
897 * @kobj: holder's kobject
899 * If there is matching entry with @kobj in @bdev->bd_holder_list
900 * and no other bd_claim() from the same kobject,
901 * remove the struct bd_holder from the list, delete symlinks for it.
903 * Returns a pointer to the struct bd_holder when it's removed from the list
904 * and ready to be freed.
905 * Returns NULL if matching claim isn't found or there is other bd_claim()
906 * by the same kobject.
908 static struct bd_holder *del_bd_holder(struct block_device *bdev,
909 struct kobject *kobj)
911 struct bd_holder *bo;
913 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
914 if (bo->sdir == kobj) {
916 BUG_ON(bo->count < 0);
919 del_symlink(bo->sdir, bo->sdev);
920 del_symlink(bo->hdir, bo->hdev);
921 bd_holder_release_dirs(bo);
932 * bd_claim_by_kobject - bd_claim() with additional kobject signature
934 * @bdev: block device to be claimed
935 * @holder: holder's signature
936 * @kobj: holder's kobject
938 * Do bd_claim() and if it succeeds, create sysfs symlinks between
939 * the bdev and the holder's kobject.
940 * Use bd_release_from_kobject() when relesing the claimed bdev.
942 * Returns 0 on success. (same as bd_claim())
943 * Returns errno on failure.
945 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
946 struct kobject *kobj)
949 struct bd_holder *bo, *found;
954 bo = alloc_bd_holder(kobj);
958 mutex_lock(&bdev->bd_mutex);
960 err = bd_claim(bdev, holder);
964 found = find_bd_holder(bdev, bo);
968 err = add_bd_holder(bdev, bo);
974 mutex_unlock(&bdev->bd_mutex);
980 * bd_release_from_kobject - bd_release() with additional kobject signature
982 * @bdev: block device to be released
983 * @kobj: holder's kobject
985 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
987 static void bd_release_from_kobject(struct block_device *bdev,
988 struct kobject *kobj)
993 mutex_lock(&bdev->bd_mutex);
995 free_bd_holder(del_bd_holder(bdev, kobj));
996 mutex_unlock(&bdev->bd_mutex);
1000 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1002 * @bdev: block device to be claimed
1003 * @holder: holder's signature
1004 * @disk: holder's gendisk
1006 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1008 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1009 struct gendisk *disk)
1011 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1013 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1016 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1018 * @bdev: block device to be claimed
1019 * @disk: holder's gendisk
1021 * Call bd_release_from_kobject() and put @disk->slave_dir.
1023 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1025 bd_release_from_kobject(bdev, disk->slave_dir);
1026 kobject_put(disk->slave_dir);
1028 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1032 * Tries to open block device by device number. Use it ONLY if you
1033 * really do not have anything better - i.e. when you are behind a
1034 * truly sucky interface and all you are given is a device number. _Never_
1035 * to be used for internal purposes. If you ever need it - reconsider
1038 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1040 struct block_device *bdev = bdget(dev);
1043 err = blkdev_get(bdev, mode);
1044 return err ? ERR_PTR(err) : bdev;
1047 EXPORT_SYMBOL(open_by_devnum);
1050 * flush_disk - invalidates all buffer-cache entries on a disk
1052 * @bdev: struct block device to be flushed
1054 * Invalidates all buffer-cache entries on a disk. It should be called
1055 * when a disk has been changed -- either by a media change or online
1058 static void flush_disk(struct block_device *bdev)
1060 if (__invalidate_device(bdev)) {
1061 char name[BDEVNAME_SIZE] = "";
1064 disk_name(bdev->bd_disk, 0, name);
1065 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1066 "resized disk %s\n", name);
1071 if (disk_partitionable(bdev->bd_disk))
1072 bdev->bd_invalidated = 1;
1076 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1077 * @disk: struct gendisk to check
1078 * @bdev: struct bdev to adjust.
1080 * This routine checks to see if the bdev size does not match the disk size
1081 * and adjusts it if it differs.
1083 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1085 loff_t disk_size, bdev_size;
1087 disk_size = (loff_t)get_capacity(disk) << 9;
1088 bdev_size = i_size_read(bdev->bd_inode);
1089 if (disk_size != bdev_size) {
1090 char name[BDEVNAME_SIZE];
1092 disk_name(disk, 0, name);
1094 "%s: detected capacity change from %lld to %lld\n",
1095 name, bdev_size, disk_size);
1096 i_size_write(bdev->bd_inode, disk_size);
1100 EXPORT_SYMBOL(check_disk_size_change);
1103 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1104 * @disk: struct gendisk to be revalidated
1106 * This routine is a wrapper for lower-level driver's revalidate_disk
1107 * call-backs. It is used to do common pre and post operations needed
1108 * for all revalidate_disk operations.
1110 int revalidate_disk(struct gendisk *disk)
1112 struct block_device *bdev;
1115 if (disk->fops->revalidate_disk)
1116 ret = disk->fops->revalidate_disk(disk);
1118 bdev = bdget_disk(disk, 0);
1122 mutex_lock(&bdev->bd_mutex);
1123 check_disk_size_change(disk, bdev);
1124 mutex_unlock(&bdev->bd_mutex);
1128 EXPORT_SYMBOL(revalidate_disk);
1131 * This routine checks whether a removable media has been changed,
1132 * and invalidates all buffer-cache-entries in that case. This
1133 * is a relatively slow routine, so we have to try to minimize using
1134 * it. Thus it is called only upon a 'mount' or 'open'. This
1135 * is the best way of combining speed and utility, I think.
1136 * People changing diskettes in the middle of an operation deserve
1139 int check_disk_change(struct block_device *bdev)
1141 struct gendisk *disk = bdev->bd_disk;
1142 const struct block_device_operations *bdops = disk->fops;
1144 if (!bdops->media_changed)
1146 if (!bdops->media_changed(bdev->bd_disk))
1150 if (bdops->revalidate_disk)
1151 bdops->revalidate_disk(bdev->bd_disk);
1155 EXPORT_SYMBOL(check_disk_change);
1157 void bd_set_size(struct block_device *bdev, loff_t size)
1159 unsigned bsize = bdev_logical_block_size(bdev);
1161 bdev->bd_inode->i_size = size;
1162 while (bsize < PAGE_CACHE_SIZE) {
1167 bdev->bd_block_size = bsize;
1168 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1170 EXPORT_SYMBOL(bd_set_size);
1172 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1177 * mutex_lock(part->bd_mutex)
1178 * mutex_lock_nested(whole->bd_mutex, 1)
1181 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1183 struct gendisk *disk;
1188 if (mode & FMODE_READ)
1190 if (mode & FMODE_WRITE)
1193 * hooks: /n/, see "layering violations".
1195 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1205 disk = get_gendisk(bdev->bd_dev, &partno);
1207 goto out_unlock_kernel;
1209 mutex_lock_nested(&bdev->bd_mutex, for_part);
1210 if (!bdev->bd_openers) {
1211 bdev->bd_disk = disk;
1212 bdev->bd_contains = bdev;
1214 struct backing_dev_info *bdi;
1217 bdev->bd_part = disk_get_part(disk, partno);
1221 if (disk->fops->open) {
1222 ret = disk->fops->open(bdev, mode);
1223 if (ret == -ERESTARTSYS) {
1224 /* Lost a race with 'disk' being
1225 * deleted, try again.
1228 disk_put_part(bdev->bd_part);
1229 bdev->bd_part = NULL;
1230 module_put(disk->fops->owner);
1232 bdev->bd_disk = NULL;
1233 mutex_unlock(&bdev->bd_mutex);
1239 if (!bdev->bd_openers) {
1240 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1241 bdi = blk_get_backing_dev_info(bdev);
1243 bdi = &default_backing_dev_info;
1244 bdev->bd_inode->i_data.backing_dev_info = bdi;
1246 if (bdev->bd_invalidated)
1247 rescan_partitions(disk, bdev);
1249 struct block_device *whole;
1250 whole = bdget_disk(disk, 0);
1255 ret = __blkdev_get(whole, mode, 1);
1258 bdev->bd_contains = whole;
1259 bdev->bd_inode->i_data.backing_dev_info =
1260 whole->bd_inode->i_data.backing_dev_info;
1261 bdev->bd_part = disk_get_part(disk, partno);
1262 if (!(disk->flags & GENHD_FL_UP) ||
1263 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1267 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1270 module_put(disk->fops->owner);
1273 if (bdev->bd_contains == bdev) {
1274 if (bdev->bd_disk->fops->open) {
1275 ret = bdev->bd_disk->fops->open(bdev, mode);
1277 goto out_unlock_bdev;
1279 if (bdev->bd_invalidated)
1280 rescan_partitions(bdev->bd_disk, bdev);
1285 bdev->bd_part_count++;
1286 mutex_unlock(&bdev->bd_mutex);
1291 disk_put_part(bdev->bd_part);
1292 bdev->bd_disk = NULL;
1293 bdev->bd_part = NULL;
1294 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1295 if (bdev != bdev->bd_contains)
1296 __blkdev_put(bdev->bd_contains, mode, 1);
1297 bdev->bd_contains = NULL;
1299 mutex_unlock(&bdev->bd_mutex);
1304 module_put(disk->fops->owner);
1311 int blkdev_get(struct block_device *bdev, fmode_t mode)
1313 return __blkdev_get(bdev, mode, 0);
1315 EXPORT_SYMBOL(blkdev_get);
1317 static int blkdev_open(struct inode * inode, struct file * filp)
1319 struct block_device *bdev;
1323 * Preserve backwards compatibility and allow large file access
1324 * even if userspace doesn't ask for it explicitly. Some mkfs
1325 * binary needs it. We might want to drop this workaround
1326 * during an unstable branch.
1328 filp->f_flags |= O_LARGEFILE;
1330 if (filp->f_flags & O_NDELAY)
1331 filp->f_mode |= FMODE_NDELAY;
1332 if (filp->f_flags & O_EXCL)
1333 filp->f_mode |= FMODE_EXCL;
1334 if ((filp->f_flags & O_ACCMODE) == 3)
1335 filp->f_mode |= FMODE_WRITE_IOCTL;
1337 bdev = bd_acquire(inode);
1341 filp->f_mapping = bdev->bd_inode->i_mapping;
1343 res = blkdev_get(bdev, filp->f_mode);
1347 if (filp->f_mode & FMODE_EXCL) {
1348 res = bd_claim(bdev, filp);
1350 goto out_blkdev_put;
1356 blkdev_put(bdev, filp->f_mode);
1360 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1363 struct gendisk *disk = bdev->bd_disk;
1364 struct block_device *victim = NULL;
1366 mutex_lock_nested(&bdev->bd_mutex, for_part);
1369 bdev->bd_part_count--;
1371 if (!--bdev->bd_openers) {
1372 sync_blockdev(bdev);
1375 if (bdev->bd_contains == bdev) {
1376 if (disk->fops->release)
1377 ret = disk->fops->release(disk, mode);
1379 if (!bdev->bd_openers) {
1380 struct module *owner = disk->fops->owner;
1384 disk_put_part(bdev->bd_part);
1385 bdev->bd_part = NULL;
1386 bdev->bd_disk = NULL;
1387 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1388 if (bdev != bdev->bd_contains)
1389 victim = bdev->bd_contains;
1390 bdev->bd_contains = NULL;
1393 mutex_unlock(&bdev->bd_mutex);
1396 __blkdev_put(victim, mode, 1);
1400 int blkdev_put(struct block_device *bdev, fmode_t mode)
1402 return __blkdev_put(bdev, mode, 0);
1404 EXPORT_SYMBOL(blkdev_put);
1406 static int blkdev_close(struct inode * inode, struct file * filp)
1408 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1409 if (bdev->bd_holder == filp)
1411 return blkdev_put(bdev, filp->f_mode);
1414 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1416 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1417 fmode_t mode = file->f_mode;
1420 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1421 * to updated it before every ioctl.
1423 if (file->f_flags & O_NDELAY)
1424 mode |= FMODE_NDELAY;
1426 mode &= ~FMODE_NDELAY;
1428 return blkdev_ioctl(bdev, mode, cmd, arg);
1432 * Write data to the block device. Only intended for the block device itself
1433 * and the raw driver which basically is a fake block device.
1435 * Does not take i_mutex for the write and thus is not for general purpose
1438 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1439 unsigned long nr_segs, loff_t pos)
1441 struct file *file = iocb->ki_filp;
1444 BUG_ON(iocb->ki_pos != pos);
1446 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1447 if (ret > 0 || ret == -EIOCBQUEUED) {
1450 err = generic_write_sync(file, pos, ret);
1451 if (err < 0 && ret > 0)
1456 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1459 * Try to release a page associated with block device when the system
1460 * is under memory pressure.
1462 static int blkdev_releasepage(struct page *page, gfp_t wait)
1464 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1466 if (super && super->s_op->bdev_try_to_free_page)
1467 return super->s_op->bdev_try_to_free_page(super, page, wait);
1469 return try_to_free_buffers(page);
1472 static const struct address_space_operations def_blk_aops = {
1473 .readpage = blkdev_readpage,
1474 .writepage = blkdev_writepage,
1475 .sync_page = block_sync_page,
1476 .write_begin = blkdev_write_begin,
1477 .write_end = blkdev_write_end,
1478 .writepages = generic_writepages,
1479 .releasepage = blkdev_releasepage,
1480 .direct_IO = blkdev_direct_IO,
1483 const struct file_operations def_blk_fops = {
1484 .open = blkdev_open,
1485 .release = blkdev_close,
1486 .llseek = block_llseek,
1487 .read = do_sync_read,
1488 .write = do_sync_write,
1489 .aio_read = generic_file_aio_read,
1490 .aio_write = blkdev_aio_write,
1491 .mmap = generic_file_mmap,
1492 .fsync = blkdev_fsync,
1493 .unlocked_ioctl = block_ioctl,
1494 #ifdef CONFIG_COMPAT
1495 .compat_ioctl = compat_blkdev_ioctl,
1497 .splice_read = generic_file_splice_read,
1498 .splice_write = generic_file_splice_write,
1501 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1504 mm_segment_t old_fs = get_fs();
1506 res = blkdev_ioctl(bdev, 0, cmd, arg);
1511 EXPORT_SYMBOL(ioctl_by_bdev);
1514 * lookup_bdev - lookup a struct block_device by name
1515 * @pathname: special file representing the block device
1517 * Get a reference to the blockdevice at @pathname in the current
1518 * namespace if possible and return it. Return ERR_PTR(error)
1521 struct block_device *lookup_bdev(const char *pathname)
1523 struct block_device *bdev;
1524 struct inode *inode;
1528 if (!pathname || !*pathname)
1529 return ERR_PTR(-EINVAL);
1531 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1533 return ERR_PTR(error);
1535 inode = path.dentry->d_inode;
1537 if (!S_ISBLK(inode->i_mode))
1540 if (path.mnt->mnt_flags & MNT_NODEV)
1543 bdev = bd_acquire(inode);
1550 bdev = ERR_PTR(error);
1553 EXPORT_SYMBOL(lookup_bdev);
1556 * open_bdev_exclusive - open a block device by name and set it up for use
1558 * @path: special file representing the block device
1559 * @mode: FMODE_... combination to pass be used
1560 * @holder: owner for exclusion
1562 * Open the blockdevice described by the special file at @path, claim it
1565 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1567 struct block_device *bdev;
1570 bdev = lookup_bdev(path);
1574 error = blkdev_get(bdev, mode);
1576 return ERR_PTR(error);
1578 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1580 error = bd_claim(bdev, holder);
1587 blkdev_put(bdev, mode);
1588 return ERR_PTR(error);
1591 EXPORT_SYMBOL(open_bdev_exclusive);
1594 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1596 * @bdev: blockdevice to close
1597 * @mode: mode, must match that used to open.
1599 * This is the counterpart to open_bdev_exclusive().
1601 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1604 blkdev_put(bdev, mode);
1607 EXPORT_SYMBOL(close_bdev_exclusive);
1609 int __invalidate_device(struct block_device *bdev)
1611 struct super_block *sb = get_super(bdev);
1616 * no need to lock the super, get_super holds the
1617 * read mutex so the filesystem cannot go away
1618 * under us (->put_super runs with the write lock
1621 shrink_dcache_sb(sb);
1622 res = invalidate_inodes(sb);
1625 invalidate_bdev(bdev);
1628 EXPORT_SYMBOL(__invalidate_device);