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 <asm/uaccess.h>
32 struct block_device bdev;
33 struct inode vfs_inode;
36 static const struct address_space_operations def_blk_aops;
38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
40 return container_of(inode, struct bdev_inode, vfs_inode);
43 inline struct block_device *I_BDEV(struct inode *inode)
45 return &BDEV_I(inode)->bdev;
48 EXPORT_SYMBOL(I_BDEV);
50 static sector_t max_block(struct block_device *bdev)
52 sector_t retval = ~((sector_t)0);
53 loff_t sz = i_size_read(bdev->bd_inode);
56 unsigned int size = block_size(bdev);
57 unsigned int sizebits = blksize_bits(size);
58 retval = (sz >> sizebits);
63 /* Kill _all_ buffers and pagecache , dirty or not.. */
64 static void kill_bdev(struct block_device *bdev)
66 if (bdev->bd_inode->i_mapping->nrpages == 0)
69 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
72 int set_blocksize(struct block_device *bdev, int size)
74 /* Size must be a power of two, and between 512 and PAGE_SIZE */
75 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
78 /* Size cannot be smaller than the size supported by the device */
79 if (size < bdev_hardsect_size(bdev))
82 /* Don't change the size if it is same as current */
83 if (bdev->bd_block_size != size) {
85 bdev->bd_block_size = size;
86 bdev->bd_inode->i_blkbits = blksize_bits(size);
92 EXPORT_SYMBOL(set_blocksize);
94 int sb_set_blocksize(struct super_block *sb, int size)
96 if (set_blocksize(sb->s_bdev, size))
98 /* If we get here, we know size is power of two
99 * and it's value is between 512 and PAGE_SIZE */
100 sb->s_blocksize = size;
101 sb->s_blocksize_bits = blksize_bits(size);
102 return sb->s_blocksize;
105 EXPORT_SYMBOL(sb_set_blocksize);
107 int sb_min_blocksize(struct super_block *sb, int size)
109 int minsize = bdev_hardsect_size(sb->s_bdev);
112 return sb_set_blocksize(sb, size);
115 EXPORT_SYMBOL(sb_min_blocksize);
118 blkdev_get_block(struct inode *inode, sector_t iblock,
119 struct buffer_head *bh, int create)
121 if (iblock >= max_block(I_BDEV(inode))) {
126 * for reads, we're just trying to fill a partial page.
127 * return a hole, they will have to call get_block again
128 * before they can fill it, and they will get -EIO at that
133 bh->b_bdev = I_BDEV(inode);
134 bh->b_blocknr = iblock;
135 set_buffer_mapped(bh);
140 blkdev_get_blocks(struct inode *inode, sector_t iblock,
141 struct buffer_head *bh, int create)
143 sector_t end_block = max_block(I_BDEV(inode));
144 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
146 if ((iblock + max_blocks) > end_block) {
147 max_blocks = end_block - iblock;
148 if ((long)max_blocks <= 0) {
150 return -EIO; /* write fully beyond EOF */
152 * It is a read which is fully beyond EOF. We return
153 * a !buffer_mapped buffer
159 bh->b_bdev = I_BDEV(inode);
160 bh->b_blocknr = iblock;
161 bh->b_size = max_blocks << inode->i_blkbits;
163 set_buffer_mapped(bh);
168 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
169 loff_t offset, unsigned long nr_segs)
171 struct file *file = iocb->ki_filp;
172 struct inode *inode = file->f_mapping->host;
174 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
175 iov, offset, nr_segs, blkdev_get_blocks, NULL);
179 * Write out and wait upon all the dirty data associated with a block
180 * device via its mapping. Does not take the superblock lock.
182 int sync_blockdev(struct block_device *bdev)
187 ret = filemap_write_and_wait(bdev->bd_inode->i_mapping);
190 EXPORT_SYMBOL(sync_blockdev);
193 * Write out and wait upon all dirty data associated with this
194 * device. Filesystem data as well as the underlying block
195 * device. Takes the superblock lock.
197 int fsync_bdev(struct block_device *bdev)
199 struct super_block *sb = get_super(bdev);
201 int res = fsync_super(sb);
205 return sync_blockdev(bdev);
209 * freeze_bdev -- lock a filesystem and force it into a consistent state
210 * @bdev: blockdevice to lock
212 * This takes the block device bd_mount_sem to make sure no new mounts
213 * happen on bdev until thaw_bdev() is called.
214 * If a superblock is found on this device, we take the s_umount semaphore
215 * on it to make sure nobody unmounts until the snapshot creation is done.
216 * The reference counter (bd_fsfreeze_count) guarantees that only the last
217 * unfreeze process can unfreeze the frozen filesystem actually when multiple
218 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
219 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
222 struct super_block *freeze_bdev(struct block_device *bdev)
224 struct super_block *sb;
227 mutex_lock(&bdev->bd_fsfreeze_mutex);
228 if (bdev->bd_fsfreeze_count > 0) {
229 bdev->bd_fsfreeze_count++;
230 sb = get_super(bdev);
231 mutex_unlock(&bdev->bd_fsfreeze_mutex);
234 bdev->bd_fsfreeze_count++;
236 down(&bdev->bd_mount_sem);
237 sb = get_super(bdev);
238 if (sb && !(sb->s_flags & MS_RDONLY)) {
239 sb->s_frozen = SB_FREEZE_WRITE;
244 sb->s_frozen = SB_FREEZE_TRANS;
247 sync_blockdev(sb->s_bdev);
249 if (sb->s_op->freeze_fs) {
250 error = sb->s_op->freeze_fs(sb);
253 "VFS:Filesystem freeze failed\n");
254 sb->s_frozen = SB_UNFROZEN;
256 up(&bdev->bd_mount_sem);
257 bdev->bd_fsfreeze_count--;
258 mutex_unlock(&bdev->bd_fsfreeze_mutex);
259 return ERR_PTR(error);
265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
267 return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */
269 EXPORT_SYMBOL(freeze_bdev);
272 * thaw_bdev -- unlock filesystem
273 * @bdev: blockdevice to unlock
274 * @sb: associated superblock
276 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
278 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
282 mutex_lock(&bdev->bd_fsfreeze_mutex);
283 if (!bdev->bd_fsfreeze_count) {
284 mutex_unlock(&bdev->bd_fsfreeze_mutex);
288 bdev->bd_fsfreeze_count--;
289 if (bdev->bd_fsfreeze_count > 0) {
292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
297 BUG_ON(sb->s_bdev != bdev);
298 if (!(sb->s_flags & MS_RDONLY)) {
299 if (sb->s_op->unfreeze_fs) {
300 error = sb->s_op->unfreeze_fs(sb);
303 "VFS:Filesystem thaw failed\n");
304 sb->s_frozen = SB_FREEZE_TRANS;
305 bdev->bd_fsfreeze_count++;
306 mutex_unlock(&bdev->bd_fsfreeze_mutex);
310 sb->s_frozen = SB_UNFROZEN;
312 wake_up(&sb->s_wait_unfrozen);
317 up(&bdev->bd_mount_sem);
318 mutex_unlock(&bdev->bd_fsfreeze_mutex);
321 EXPORT_SYMBOL(thaw_bdev);
323 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
325 return block_write_full_page(page, blkdev_get_block, wbc);
328 static int blkdev_readpage(struct file * file, struct page * page)
330 return block_read_full_page(page, blkdev_get_block);
333 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
334 loff_t pos, unsigned len, unsigned flags,
335 struct page **pagep, void **fsdata)
338 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
342 static int blkdev_write_end(struct file *file, struct address_space *mapping,
343 loff_t pos, unsigned len, unsigned copied,
344 struct page *page, void *fsdata)
347 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
350 page_cache_release(page);
357 * for a block special file file->f_path.dentry->d_inode->i_size is zero
358 * so we compute the size by hand (just as in block_read/write above)
360 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
362 struct inode *bd_inode = file->f_mapping->host;
366 mutex_lock(&bd_inode->i_mutex);
367 size = i_size_read(bd_inode);
374 offset += file->f_pos;
377 if (offset >= 0 && offset <= size) {
378 if (offset != file->f_pos) {
379 file->f_pos = offset;
383 mutex_unlock(&bd_inode->i_mutex);
388 * Filp is never NULL; the only case when ->fsync() is called with
389 * NULL first argument is nfsd_sync_dir() and that's not a directory.
392 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
394 return sync_blockdev(I_BDEV(filp->f_mapping->host));
401 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
402 static struct kmem_cache * bdev_cachep __read_mostly;
404 static struct inode *bdev_alloc_inode(struct super_block *sb)
406 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
409 return &ei->vfs_inode;
412 static void bdev_destroy_inode(struct inode *inode)
414 struct bdev_inode *bdi = BDEV_I(inode);
416 bdi->bdev.bd_inode_backing_dev_info = NULL;
417 kmem_cache_free(bdev_cachep, bdi);
420 static void init_once(void *foo)
422 struct bdev_inode *ei = (struct bdev_inode *) foo;
423 struct block_device *bdev = &ei->bdev;
425 memset(bdev, 0, sizeof(*bdev));
426 mutex_init(&bdev->bd_mutex);
427 sema_init(&bdev->bd_mount_sem, 1);
428 INIT_LIST_HEAD(&bdev->bd_inodes);
429 INIT_LIST_HEAD(&bdev->bd_list);
431 INIT_LIST_HEAD(&bdev->bd_holder_list);
433 inode_init_once(&ei->vfs_inode);
434 /* Initialize mutex for freeze. */
435 mutex_init(&bdev->bd_fsfreeze_mutex);
438 static inline void __bd_forget(struct inode *inode)
440 list_del_init(&inode->i_devices);
441 inode->i_bdev = NULL;
442 inode->i_mapping = &inode->i_data;
445 static void bdev_clear_inode(struct inode *inode)
447 struct block_device *bdev = &BDEV_I(inode)->bdev;
449 spin_lock(&bdev_lock);
450 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
451 __bd_forget(list_entry(p, struct inode, i_devices));
453 list_del_init(&bdev->bd_list);
454 spin_unlock(&bdev_lock);
457 static const struct super_operations bdev_sops = {
458 .statfs = simple_statfs,
459 .alloc_inode = bdev_alloc_inode,
460 .destroy_inode = bdev_destroy_inode,
461 .drop_inode = generic_delete_inode,
462 .clear_inode = bdev_clear_inode,
465 static int bd_get_sb(struct file_system_type *fs_type,
466 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
468 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
471 static struct file_system_type bd_type = {
474 .kill_sb = kill_anon_super,
477 struct super_block *blockdev_superblock __read_mostly;
479 void __init bdev_cache_init(void)
482 struct vfsmount *bd_mnt;
484 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
485 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
486 SLAB_MEM_SPREAD|SLAB_PANIC),
488 err = register_filesystem(&bd_type);
490 panic("Cannot register bdev pseudo-fs");
491 bd_mnt = kern_mount(&bd_type);
493 panic("Cannot create bdev pseudo-fs");
494 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
498 * Most likely _very_ bad one - but then it's hardly critical for small
499 * /dev and can be fixed when somebody will need really large one.
500 * Keep in mind that it will be fed through icache hash function too.
502 static inline unsigned long hash(dev_t dev)
504 return MAJOR(dev)+MINOR(dev);
507 static int bdev_test(struct inode *inode, void *data)
509 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
512 static int bdev_set(struct inode *inode, void *data)
514 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
518 static LIST_HEAD(all_bdevs);
520 struct block_device *bdget(dev_t dev)
522 struct block_device *bdev;
525 inode = iget5_locked(blockdev_superblock, hash(dev),
526 bdev_test, bdev_set, &dev);
531 bdev = &BDEV_I(inode)->bdev;
533 if (inode->i_state & I_NEW) {
534 bdev->bd_contains = NULL;
535 bdev->bd_inode = inode;
536 bdev->bd_block_size = (1 << inode->i_blkbits);
537 bdev->bd_part_count = 0;
538 bdev->bd_invalidated = 0;
539 inode->i_mode = S_IFBLK;
541 inode->i_bdev = bdev;
542 inode->i_data.a_ops = &def_blk_aops;
543 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
544 inode->i_data.backing_dev_info = &default_backing_dev_info;
545 spin_lock(&bdev_lock);
546 list_add(&bdev->bd_list, &all_bdevs);
547 spin_unlock(&bdev_lock);
548 unlock_new_inode(inode);
553 EXPORT_SYMBOL(bdget);
555 long nr_blockdev_pages(void)
557 struct block_device *bdev;
559 spin_lock(&bdev_lock);
560 list_for_each_entry(bdev, &all_bdevs, bd_list) {
561 ret += bdev->bd_inode->i_mapping->nrpages;
563 spin_unlock(&bdev_lock);
567 void bdput(struct block_device *bdev)
569 iput(bdev->bd_inode);
572 EXPORT_SYMBOL(bdput);
574 static struct block_device *bd_acquire(struct inode *inode)
576 struct block_device *bdev;
578 spin_lock(&bdev_lock);
579 bdev = inode->i_bdev;
581 atomic_inc(&bdev->bd_inode->i_count);
582 spin_unlock(&bdev_lock);
585 spin_unlock(&bdev_lock);
587 bdev = bdget(inode->i_rdev);
589 spin_lock(&bdev_lock);
590 if (!inode->i_bdev) {
592 * We take an additional bd_inode->i_count for inode,
593 * and it's released in clear_inode() of inode.
594 * So, we can access it via ->i_mapping always
597 atomic_inc(&bdev->bd_inode->i_count);
598 inode->i_bdev = bdev;
599 inode->i_mapping = bdev->bd_inode->i_mapping;
600 list_add(&inode->i_devices, &bdev->bd_inodes);
602 spin_unlock(&bdev_lock);
607 /* Call when you free inode */
609 void bd_forget(struct inode *inode)
611 struct block_device *bdev = NULL;
613 spin_lock(&bdev_lock);
615 if (!sb_is_blkdev_sb(inode->i_sb))
616 bdev = inode->i_bdev;
619 spin_unlock(&bdev_lock);
622 iput(bdev->bd_inode);
625 int bd_claim(struct block_device *bdev, void *holder)
628 spin_lock(&bdev_lock);
630 /* first decide result */
631 if (bdev->bd_holder == holder)
632 res = 0; /* already a holder */
633 else if (bdev->bd_holder != NULL)
634 res = -EBUSY; /* held by someone else */
635 else if (bdev->bd_contains == bdev)
636 res = 0; /* is a whole device which isn't held */
638 else if (bdev->bd_contains->bd_holder == bd_claim)
639 res = 0; /* is a partition of a device that is being partitioned */
640 else if (bdev->bd_contains->bd_holder != NULL)
641 res = -EBUSY; /* is a partition of a held device */
643 res = 0; /* is a partition of an un-held device */
645 /* now impose change */
647 /* note that for a whole device bd_holders
648 * will be incremented twice, and bd_holder will
649 * be set to bd_claim before being set to holder
651 bdev->bd_contains->bd_holders ++;
652 bdev->bd_contains->bd_holder = bd_claim;
654 bdev->bd_holder = holder;
656 spin_unlock(&bdev_lock);
660 EXPORT_SYMBOL(bd_claim);
662 void bd_release(struct block_device *bdev)
664 spin_lock(&bdev_lock);
665 if (!--bdev->bd_contains->bd_holders)
666 bdev->bd_contains->bd_holder = NULL;
667 if (!--bdev->bd_holders)
668 bdev->bd_holder = NULL;
669 spin_unlock(&bdev_lock);
672 EXPORT_SYMBOL(bd_release);
676 * Functions for bd_claim_by_kobject / bd_release_from_kobject
678 * If a kobject is passed to bd_claim_by_kobject()
679 * and the kobject has a parent directory,
680 * following symlinks are created:
681 * o from the kobject to the claimed bdev
682 * o from "holders" directory of the bdev to the parent of the kobject
683 * bd_release_from_kobject() removes these symlinks.
686 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
687 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
688 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
689 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
692 static int add_symlink(struct kobject *from, struct kobject *to)
696 return sysfs_create_link(from, to, kobject_name(to));
699 static void del_symlink(struct kobject *from, struct kobject *to)
703 sysfs_remove_link(from, kobject_name(to));
707 * 'struct bd_holder' contains pointers to kobjects symlinked by
708 * bd_claim_by_kobject.
709 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
712 struct list_head list; /* chain of holders of the bdev */
713 int count; /* references from the holder */
714 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
715 struct kobject *hdev; /* e.g. "/block/dm-0" */
716 struct kobject *hdir; /* e.g. "/block/sda/holders" */
717 struct kobject *sdev; /* e.g. "/block/sda" */
721 * Get references of related kobjects at once.
722 * Returns 1 on success. 0 on failure.
724 * Should call bd_holder_release_dirs() after successful use.
726 static int bd_holder_grab_dirs(struct block_device *bdev,
727 struct bd_holder *bo)
732 bo->sdir = kobject_get(bo->sdir);
736 bo->hdev = kobject_get(bo->sdir->parent);
740 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
744 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
751 kobject_put(bo->sdev);
753 kobject_put(bo->hdev);
755 kobject_put(bo->sdir);
760 /* Put references of related kobjects at once. */
761 static void bd_holder_release_dirs(struct bd_holder *bo)
763 kobject_put(bo->hdir);
764 kobject_put(bo->sdev);
765 kobject_put(bo->hdev);
766 kobject_put(bo->sdir);
769 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
771 struct bd_holder *bo;
773 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
783 static void free_bd_holder(struct bd_holder *bo)
789 * find_bd_holder - find matching struct bd_holder from the block device
791 * @bdev: struct block device to be searched
792 * @bo: target struct bd_holder
794 * Returns matching entry with @bo in @bdev->bd_holder_list.
795 * If found, increment the reference count and return the pointer.
796 * If not found, returns NULL.
798 static struct bd_holder *find_bd_holder(struct block_device *bdev,
799 struct bd_holder *bo)
801 struct bd_holder *tmp;
803 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
804 if (tmp->sdir == bo->sdir) {
813 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
815 * @bdev: block device to be bd_claimed
816 * @bo: preallocated and initialized by alloc_bd_holder()
818 * Add @bo to @bdev->bd_holder_list, create symlinks.
820 * Returns 0 if symlinks are created.
821 * Returns -ve if something fails.
823 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
830 if (!bd_holder_grab_dirs(bdev, bo))
833 err = add_symlink(bo->sdir, bo->sdev);
837 err = add_symlink(bo->hdir, bo->hdev);
839 del_symlink(bo->sdir, bo->sdev);
843 list_add_tail(&bo->list, &bdev->bd_holder_list);
848 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
850 * @bdev: block device to be bd_claimed
851 * @kobj: holder's kobject
853 * If there is matching entry with @kobj in @bdev->bd_holder_list
854 * and no other bd_claim() from the same kobject,
855 * remove the struct bd_holder from the list, delete symlinks for it.
857 * Returns a pointer to the struct bd_holder when it's removed from the list
858 * and ready to be freed.
859 * Returns NULL if matching claim isn't found or there is other bd_claim()
860 * by the same kobject.
862 static struct bd_holder *del_bd_holder(struct block_device *bdev,
863 struct kobject *kobj)
865 struct bd_holder *bo;
867 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
868 if (bo->sdir == kobj) {
870 BUG_ON(bo->count < 0);
873 del_symlink(bo->sdir, bo->sdev);
874 del_symlink(bo->hdir, bo->hdev);
875 bd_holder_release_dirs(bo);
886 * bd_claim_by_kobject - bd_claim() with additional kobject signature
888 * @bdev: block device to be claimed
889 * @holder: holder's signature
890 * @kobj: holder's kobject
892 * Do bd_claim() and if it succeeds, create sysfs symlinks between
893 * the bdev and the holder's kobject.
894 * Use bd_release_from_kobject() when relesing the claimed bdev.
896 * Returns 0 on success. (same as bd_claim())
897 * Returns errno on failure.
899 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
900 struct kobject *kobj)
903 struct bd_holder *bo, *found;
908 bo = alloc_bd_holder(kobj);
912 mutex_lock(&bdev->bd_mutex);
914 err = bd_claim(bdev, holder);
918 found = find_bd_holder(bdev, bo);
922 err = add_bd_holder(bdev, bo);
928 mutex_unlock(&bdev->bd_mutex);
934 * bd_release_from_kobject - bd_release() with additional kobject signature
936 * @bdev: block device to be released
937 * @kobj: holder's kobject
939 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
941 static void bd_release_from_kobject(struct block_device *bdev,
942 struct kobject *kobj)
947 mutex_lock(&bdev->bd_mutex);
949 free_bd_holder(del_bd_holder(bdev, kobj));
950 mutex_unlock(&bdev->bd_mutex);
954 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
956 * @bdev: block device to be claimed
957 * @holder: holder's signature
958 * @disk: holder's gendisk
960 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
962 int bd_claim_by_disk(struct block_device *bdev, void *holder,
963 struct gendisk *disk)
965 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
967 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
970 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
972 * @bdev: block device to be claimed
973 * @disk: holder's gendisk
975 * Call bd_release_from_kobject() and put @disk->slave_dir.
977 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
979 bd_release_from_kobject(bdev, disk->slave_dir);
980 kobject_put(disk->slave_dir);
982 EXPORT_SYMBOL_GPL(bd_release_from_disk);
986 * Tries to open block device by device number. Use it ONLY if you
987 * really do not have anything better - i.e. when you are behind a
988 * truly sucky interface and all you are given is a device number. _Never_
989 * to be used for internal purposes. If you ever need it - reconsider
992 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
994 struct block_device *bdev = bdget(dev);
997 err = blkdev_get(bdev, mode);
998 return err ? ERR_PTR(err) : bdev;
1001 EXPORT_SYMBOL(open_by_devnum);
1004 * flush_disk - invalidates all buffer-cache entries on a disk
1006 * @bdev: struct block device to be flushed
1008 * Invalidates all buffer-cache entries on a disk. It should be called
1009 * when a disk has been changed -- either by a media change or online
1012 static void flush_disk(struct block_device *bdev)
1014 if (__invalidate_device(bdev)) {
1015 char name[BDEVNAME_SIZE] = "";
1018 disk_name(bdev->bd_disk, 0, name);
1019 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1020 "resized disk %s\n", name);
1025 if (disk_partitionable(bdev->bd_disk))
1026 bdev->bd_invalidated = 1;
1030 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1031 * @disk: struct gendisk to check
1032 * @bdev: struct bdev to adjust.
1034 * This routine checks to see if the bdev size does not match the disk size
1035 * and adjusts it if it differs.
1037 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1039 loff_t disk_size, bdev_size;
1041 disk_size = (loff_t)get_capacity(disk) << 9;
1042 bdev_size = i_size_read(bdev->bd_inode);
1043 if (disk_size != bdev_size) {
1044 char name[BDEVNAME_SIZE];
1046 disk_name(disk, 0, name);
1048 "%s: detected capacity change from %lld to %lld\n",
1049 name, bdev_size, disk_size);
1050 i_size_write(bdev->bd_inode, disk_size);
1054 EXPORT_SYMBOL(check_disk_size_change);
1057 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1058 * @disk: struct gendisk to be revalidated
1060 * This routine is a wrapper for lower-level driver's revalidate_disk
1061 * call-backs. It is used to do common pre and post operations needed
1062 * for all revalidate_disk operations.
1064 int revalidate_disk(struct gendisk *disk)
1066 struct block_device *bdev;
1069 if (disk->fops->revalidate_disk)
1070 ret = disk->fops->revalidate_disk(disk);
1072 bdev = bdget_disk(disk, 0);
1076 mutex_lock(&bdev->bd_mutex);
1077 check_disk_size_change(disk, bdev);
1078 mutex_unlock(&bdev->bd_mutex);
1082 EXPORT_SYMBOL(revalidate_disk);
1085 * This routine checks whether a removable media has been changed,
1086 * and invalidates all buffer-cache-entries in that case. This
1087 * is a relatively slow routine, so we have to try to minimize using
1088 * it. Thus it is called only upon a 'mount' or 'open'. This
1089 * is the best way of combining speed and utility, I think.
1090 * People changing diskettes in the middle of an operation deserve
1093 int check_disk_change(struct block_device *bdev)
1095 struct gendisk *disk = bdev->bd_disk;
1096 struct block_device_operations * bdops = disk->fops;
1098 if (!bdops->media_changed)
1100 if (!bdops->media_changed(bdev->bd_disk))
1104 if (bdops->revalidate_disk)
1105 bdops->revalidate_disk(bdev->bd_disk);
1109 EXPORT_SYMBOL(check_disk_change);
1111 void bd_set_size(struct block_device *bdev, loff_t size)
1113 unsigned bsize = bdev_hardsect_size(bdev);
1115 bdev->bd_inode->i_size = size;
1116 while (bsize < PAGE_CACHE_SIZE) {
1121 bdev->bd_block_size = bsize;
1122 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1124 EXPORT_SYMBOL(bd_set_size);
1126 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1131 * mutex_lock(part->bd_mutex)
1132 * mutex_lock_nested(whole->bd_mutex, 1)
1135 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1137 struct gendisk *disk;
1142 if (mode & FMODE_READ)
1144 if (mode & FMODE_WRITE)
1147 * hooks: /n/, see "layering violations".
1149 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1159 disk = get_gendisk(bdev->bd_dev, &partno);
1161 goto out_unlock_kernel;
1163 mutex_lock_nested(&bdev->bd_mutex, for_part);
1164 if (!bdev->bd_openers) {
1165 bdev->bd_disk = disk;
1166 bdev->bd_contains = bdev;
1168 struct backing_dev_info *bdi;
1171 bdev->bd_part = disk_get_part(disk, partno);
1175 if (disk->fops->open) {
1176 ret = disk->fops->open(bdev, mode);
1177 if (ret == -ERESTARTSYS) {
1178 /* Lost a race with 'disk' being
1179 * deleted, try again.
1182 disk_put_part(bdev->bd_part);
1183 bdev->bd_part = NULL;
1184 module_put(disk->fops->owner);
1186 bdev->bd_disk = NULL;
1187 mutex_unlock(&bdev->bd_mutex);
1193 if (!bdev->bd_openers) {
1194 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1195 bdi = blk_get_backing_dev_info(bdev);
1197 bdi = &default_backing_dev_info;
1198 bdev->bd_inode->i_data.backing_dev_info = bdi;
1200 if (bdev->bd_invalidated)
1201 rescan_partitions(disk, bdev);
1203 struct block_device *whole;
1204 whole = bdget_disk(disk, 0);
1209 ret = __blkdev_get(whole, mode, 1);
1212 bdev->bd_contains = whole;
1213 bdev->bd_inode->i_data.backing_dev_info =
1214 whole->bd_inode->i_data.backing_dev_info;
1215 bdev->bd_part = disk_get_part(disk, partno);
1216 if (!(disk->flags & GENHD_FL_UP) ||
1217 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1221 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1225 module_put(disk->fops->owner);
1227 if (bdev->bd_contains == bdev) {
1228 if (bdev->bd_disk->fops->open) {
1229 ret = bdev->bd_disk->fops->open(bdev, mode);
1231 goto out_unlock_bdev;
1233 if (bdev->bd_invalidated)
1234 rescan_partitions(bdev->bd_disk, bdev);
1239 bdev->bd_part_count++;
1240 mutex_unlock(&bdev->bd_mutex);
1245 disk_put_part(bdev->bd_part);
1246 bdev->bd_disk = NULL;
1247 bdev->bd_part = NULL;
1248 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1249 if (bdev != bdev->bd_contains)
1250 __blkdev_put(bdev->bd_contains, mode, 1);
1251 bdev->bd_contains = NULL;
1253 mutex_unlock(&bdev->bd_mutex);
1258 module_put(disk->fops->owner);
1265 int blkdev_get(struct block_device *bdev, fmode_t mode)
1267 return __blkdev_get(bdev, mode, 0);
1269 EXPORT_SYMBOL(blkdev_get);
1271 static int blkdev_open(struct inode * inode, struct file * filp)
1273 struct block_device *bdev;
1277 * Preserve backwards compatibility and allow large file access
1278 * even if userspace doesn't ask for it explicitly. Some mkfs
1279 * binary needs it. We might want to drop this workaround
1280 * during an unstable branch.
1282 filp->f_flags |= O_LARGEFILE;
1284 if (filp->f_flags & O_NDELAY)
1285 filp->f_mode |= FMODE_NDELAY;
1286 if (filp->f_flags & O_EXCL)
1287 filp->f_mode |= FMODE_EXCL;
1288 if ((filp->f_flags & O_ACCMODE) == 3)
1289 filp->f_mode |= FMODE_WRITE_IOCTL;
1291 bdev = bd_acquire(inode);
1295 filp->f_mapping = bdev->bd_inode->i_mapping;
1297 res = blkdev_get(bdev, filp->f_mode);
1301 if (filp->f_mode & FMODE_EXCL) {
1302 res = bd_claim(bdev, filp);
1304 goto out_blkdev_put;
1310 blkdev_put(bdev, filp->f_mode);
1314 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1317 struct gendisk *disk = bdev->bd_disk;
1318 struct block_device *victim = NULL;
1320 mutex_lock_nested(&bdev->bd_mutex, for_part);
1323 bdev->bd_part_count--;
1325 if (!--bdev->bd_openers) {
1326 sync_blockdev(bdev);
1329 if (bdev->bd_contains == bdev) {
1330 if (disk->fops->release)
1331 ret = disk->fops->release(disk, mode);
1333 if (!bdev->bd_openers) {
1334 struct module *owner = disk->fops->owner;
1338 disk_put_part(bdev->bd_part);
1339 bdev->bd_part = NULL;
1340 bdev->bd_disk = NULL;
1341 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1342 if (bdev != bdev->bd_contains)
1343 victim = bdev->bd_contains;
1344 bdev->bd_contains = NULL;
1347 mutex_unlock(&bdev->bd_mutex);
1350 __blkdev_put(victim, mode, 1);
1354 int blkdev_put(struct block_device *bdev, fmode_t mode)
1356 return __blkdev_put(bdev, mode, 0);
1358 EXPORT_SYMBOL(blkdev_put);
1360 static int blkdev_close(struct inode * inode, struct file * filp)
1362 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1363 if (bdev->bd_holder == filp)
1365 return blkdev_put(bdev, filp->f_mode);
1368 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1370 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1371 fmode_t mode = file->f_mode;
1374 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1375 * to updated it before every ioctl.
1377 if (file->f_flags & O_NDELAY)
1378 mode |= FMODE_NDELAY;
1380 mode &= ~FMODE_NDELAY;
1382 return blkdev_ioctl(bdev, mode, cmd, arg);
1386 * Try to release a page associated with block device when the system
1387 * is under memory pressure.
1389 static int blkdev_releasepage(struct page *page, gfp_t wait)
1391 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1393 if (super && super->s_op->bdev_try_to_free_page)
1394 return super->s_op->bdev_try_to_free_page(super, page, wait);
1396 return try_to_free_buffers(page);
1399 static const struct address_space_operations def_blk_aops = {
1400 .readpage = blkdev_readpage,
1401 .writepage = blkdev_writepage,
1402 .sync_page = block_sync_page,
1403 .write_begin = blkdev_write_begin,
1404 .write_end = blkdev_write_end,
1405 .writepages = generic_writepages,
1406 .releasepage = blkdev_releasepage,
1407 .direct_IO = blkdev_direct_IO,
1410 const struct file_operations def_blk_fops = {
1411 .open = blkdev_open,
1412 .release = blkdev_close,
1413 .llseek = block_llseek,
1414 .read = do_sync_read,
1415 .write = do_sync_write,
1416 .aio_read = generic_file_aio_read,
1417 .aio_write = generic_file_aio_write_nolock,
1418 .mmap = generic_file_mmap,
1419 .fsync = block_fsync,
1420 .unlocked_ioctl = block_ioctl,
1421 #ifdef CONFIG_COMPAT
1422 .compat_ioctl = compat_blkdev_ioctl,
1424 .splice_read = generic_file_splice_read,
1425 .splice_write = generic_file_splice_write,
1428 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1431 mm_segment_t old_fs = get_fs();
1433 res = blkdev_ioctl(bdev, 0, cmd, arg);
1438 EXPORT_SYMBOL(ioctl_by_bdev);
1441 * lookup_bdev - lookup a struct block_device by name
1442 * @pathname: special file representing the block device
1444 * Get a reference to the blockdevice at @pathname in the current
1445 * namespace if possible and return it. Return ERR_PTR(error)
1448 struct block_device *lookup_bdev(const char *pathname)
1450 struct block_device *bdev;
1451 struct inode *inode;
1455 if (!pathname || !*pathname)
1456 return ERR_PTR(-EINVAL);
1458 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1460 return ERR_PTR(error);
1462 inode = path.dentry->d_inode;
1464 if (!S_ISBLK(inode->i_mode))
1467 if (path.mnt->mnt_flags & MNT_NODEV)
1470 bdev = bd_acquire(inode);
1477 bdev = ERR_PTR(error);
1480 EXPORT_SYMBOL(lookup_bdev);
1483 * open_bdev_exclusive - open a block device by name and set it up for use
1485 * @path: special file representing the block device
1486 * @mode: FMODE_... combination to pass be used
1487 * @holder: owner for exclusion
1489 * Open the blockdevice described by the special file at @path, claim it
1492 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1494 struct block_device *bdev;
1497 bdev = lookup_bdev(path);
1501 error = blkdev_get(bdev, mode);
1503 return ERR_PTR(error);
1505 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1507 error = bd_claim(bdev, holder);
1514 blkdev_put(bdev, mode);
1515 return ERR_PTR(error);
1518 EXPORT_SYMBOL(open_bdev_exclusive);
1521 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1523 * @bdev: blockdevice to close
1524 * @mode: mode, must match that used to open.
1526 * This is the counterpart to open_bdev_exclusive().
1528 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1531 blkdev_put(bdev, mode);
1534 EXPORT_SYMBOL(close_bdev_exclusive);
1536 int __invalidate_device(struct block_device *bdev)
1538 struct super_block *sb = get_super(bdev);
1543 * no need to lock the super, get_super holds the
1544 * read mutex so the filesystem cannot go away
1545 * under us (->put_super runs with the write lock
1548 shrink_dcache_sb(sb);
1549 res = invalidate_inodes(sb);
1552 invalidate_bdev(bdev);
1555 EXPORT_SYMBOL(__invalidate_device);