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);
207 EXPORT_SYMBOL(fsync_bdev);
210 * freeze_bdev -- lock a filesystem and force it into a consistent state
211 * @bdev: blockdevice to lock
213 * This takes the block device bd_mount_sem to make sure no new mounts
214 * happen on bdev until thaw_bdev() is called.
215 * If a superblock is found on this device, we take the s_umount semaphore
216 * on it to make sure nobody unmounts until the snapshot creation is done.
217 * The reference counter (bd_fsfreeze_count) guarantees that only the last
218 * unfreeze process can unfreeze the frozen filesystem actually when multiple
219 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
220 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
223 struct super_block *freeze_bdev(struct block_device *bdev)
225 struct super_block *sb;
228 mutex_lock(&bdev->bd_fsfreeze_mutex);
229 if (bdev->bd_fsfreeze_count > 0) {
230 bdev->bd_fsfreeze_count++;
231 sb = get_super(bdev);
232 mutex_unlock(&bdev->bd_fsfreeze_mutex);
235 bdev->bd_fsfreeze_count++;
237 down(&bdev->bd_mount_sem);
238 sb = get_super(bdev);
239 if (sb && !(sb->s_flags & MS_RDONLY)) {
240 sb->s_frozen = SB_FREEZE_WRITE;
245 sb->s_frozen = SB_FREEZE_TRANS;
248 sync_blockdev(sb->s_bdev);
250 if (sb->s_op->freeze_fs) {
251 error = sb->s_op->freeze_fs(sb);
254 "VFS:Filesystem freeze failed\n");
255 sb->s_frozen = SB_UNFROZEN;
257 up(&bdev->bd_mount_sem);
258 bdev->bd_fsfreeze_count--;
259 mutex_unlock(&bdev->bd_fsfreeze_mutex);
260 return ERR_PTR(error);
266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
268 return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */
270 EXPORT_SYMBOL(freeze_bdev);
273 * thaw_bdev -- unlock filesystem
274 * @bdev: blockdevice to unlock
275 * @sb: associated superblock
277 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
279 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
283 mutex_lock(&bdev->bd_fsfreeze_mutex);
284 if (!bdev->bd_fsfreeze_count) {
285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
289 bdev->bd_fsfreeze_count--;
290 if (bdev->bd_fsfreeze_count > 0) {
293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
298 BUG_ON(sb->s_bdev != bdev);
299 if (!(sb->s_flags & MS_RDONLY)) {
300 if (sb->s_op->unfreeze_fs) {
301 error = sb->s_op->unfreeze_fs(sb);
304 "VFS:Filesystem thaw failed\n");
305 sb->s_frozen = SB_FREEZE_TRANS;
306 bdev->bd_fsfreeze_count++;
307 mutex_unlock(&bdev->bd_fsfreeze_mutex);
311 sb->s_frozen = SB_UNFROZEN;
313 wake_up(&sb->s_wait_unfrozen);
318 up(&bdev->bd_mount_sem);
319 mutex_unlock(&bdev->bd_fsfreeze_mutex);
322 EXPORT_SYMBOL(thaw_bdev);
324 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
326 return block_write_full_page(page, blkdev_get_block, wbc);
329 static int blkdev_readpage(struct file * file, struct page * page)
331 return block_read_full_page(page, blkdev_get_block);
334 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
335 loff_t pos, unsigned len, unsigned flags,
336 struct page **pagep, void **fsdata)
339 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
343 static int blkdev_write_end(struct file *file, struct address_space *mapping,
344 loff_t pos, unsigned len, unsigned copied,
345 struct page *page, void *fsdata)
348 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
351 page_cache_release(page);
358 * for a block special file file->f_path.dentry->d_inode->i_size is zero
359 * so we compute the size by hand (just as in block_read/write above)
361 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
363 struct inode *bd_inode = file->f_mapping->host;
367 mutex_lock(&bd_inode->i_mutex);
368 size = i_size_read(bd_inode);
375 offset += file->f_pos;
378 if (offset >= 0 && offset <= size) {
379 if (offset != file->f_pos) {
380 file->f_pos = offset;
384 mutex_unlock(&bd_inode->i_mutex);
389 * Filp is never NULL; the only case when ->fsync() is called with
390 * NULL first argument is nfsd_sync_dir() and that's not a directory.
393 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
395 return sync_blockdev(I_BDEV(filp->f_mapping->host));
402 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
403 static struct kmem_cache * bdev_cachep __read_mostly;
405 static struct inode *bdev_alloc_inode(struct super_block *sb)
407 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
410 return &ei->vfs_inode;
413 static void bdev_destroy_inode(struct inode *inode)
415 struct bdev_inode *bdi = BDEV_I(inode);
417 bdi->bdev.bd_inode_backing_dev_info = NULL;
418 kmem_cache_free(bdev_cachep, bdi);
421 static void init_once(void *foo)
423 struct bdev_inode *ei = (struct bdev_inode *) foo;
424 struct block_device *bdev = &ei->bdev;
426 memset(bdev, 0, sizeof(*bdev));
427 mutex_init(&bdev->bd_mutex);
428 sema_init(&bdev->bd_mount_sem, 1);
429 INIT_LIST_HEAD(&bdev->bd_inodes);
430 INIT_LIST_HEAD(&bdev->bd_list);
432 INIT_LIST_HEAD(&bdev->bd_holder_list);
434 inode_init_once(&ei->vfs_inode);
435 /* Initialize mutex for freeze. */
436 mutex_init(&bdev->bd_fsfreeze_mutex);
439 static inline void __bd_forget(struct inode *inode)
441 list_del_init(&inode->i_devices);
442 inode->i_bdev = NULL;
443 inode->i_mapping = &inode->i_data;
446 static void bdev_clear_inode(struct inode *inode)
448 struct block_device *bdev = &BDEV_I(inode)->bdev;
450 spin_lock(&bdev_lock);
451 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
452 __bd_forget(list_entry(p, struct inode, i_devices));
454 list_del_init(&bdev->bd_list);
455 spin_unlock(&bdev_lock);
458 static const struct super_operations bdev_sops = {
459 .statfs = simple_statfs,
460 .alloc_inode = bdev_alloc_inode,
461 .destroy_inode = bdev_destroy_inode,
462 .drop_inode = generic_delete_inode,
463 .clear_inode = bdev_clear_inode,
466 static int bd_get_sb(struct file_system_type *fs_type,
467 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
469 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
472 static struct file_system_type bd_type = {
475 .kill_sb = kill_anon_super,
478 struct super_block *blockdev_superblock __read_mostly;
480 void __init bdev_cache_init(void)
483 struct vfsmount *bd_mnt;
485 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
486 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
487 SLAB_MEM_SPREAD|SLAB_PANIC),
489 err = register_filesystem(&bd_type);
491 panic("Cannot register bdev pseudo-fs");
492 bd_mnt = kern_mount(&bd_type);
494 panic("Cannot create bdev pseudo-fs");
495 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
499 * Most likely _very_ bad one - but then it's hardly critical for small
500 * /dev and can be fixed when somebody will need really large one.
501 * Keep in mind that it will be fed through icache hash function too.
503 static inline unsigned long hash(dev_t dev)
505 return MAJOR(dev)+MINOR(dev);
508 static int bdev_test(struct inode *inode, void *data)
510 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
513 static int bdev_set(struct inode *inode, void *data)
515 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
519 static LIST_HEAD(all_bdevs);
521 struct block_device *bdget(dev_t dev)
523 struct block_device *bdev;
526 inode = iget5_locked(blockdev_superblock, hash(dev),
527 bdev_test, bdev_set, &dev);
532 bdev = &BDEV_I(inode)->bdev;
534 if (inode->i_state & I_NEW) {
535 bdev->bd_contains = NULL;
536 bdev->bd_inode = inode;
537 bdev->bd_block_size = (1 << inode->i_blkbits);
538 bdev->bd_part_count = 0;
539 bdev->bd_invalidated = 0;
540 inode->i_mode = S_IFBLK;
542 inode->i_bdev = bdev;
543 inode->i_data.a_ops = &def_blk_aops;
544 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
545 inode->i_data.backing_dev_info = &default_backing_dev_info;
546 spin_lock(&bdev_lock);
547 list_add(&bdev->bd_list, &all_bdevs);
548 spin_unlock(&bdev_lock);
549 unlock_new_inode(inode);
554 EXPORT_SYMBOL(bdget);
556 long nr_blockdev_pages(void)
558 struct block_device *bdev;
560 spin_lock(&bdev_lock);
561 list_for_each_entry(bdev, &all_bdevs, bd_list) {
562 ret += bdev->bd_inode->i_mapping->nrpages;
564 spin_unlock(&bdev_lock);
568 void bdput(struct block_device *bdev)
570 iput(bdev->bd_inode);
573 EXPORT_SYMBOL(bdput);
575 static struct block_device *bd_acquire(struct inode *inode)
577 struct block_device *bdev;
579 spin_lock(&bdev_lock);
580 bdev = inode->i_bdev;
582 atomic_inc(&bdev->bd_inode->i_count);
583 spin_unlock(&bdev_lock);
586 spin_unlock(&bdev_lock);
588 bdev = bdget(inode->i_rdev);
590 spin_lock(&bdev_lock);
591 if (!inode->i_bdev) {
593 * We take an additional bd_inode->i_count for inode,
594 * and it's released in clear_inode() of inode.
595 * So, we can access it via ->i_mapping always
598 atomic_inc(&bdev->bd_inode->i_count);
599 inode->i_bdev = bdev;
600 inode->i_mapping = bdev->bd_inode->i_mapping;
601 list_add(&inode->i_devices, &bdev->bd_inodes);
603 spin_unlock(&bdev_lock);
608 /* Call when you free inode */
610 void bd_forget(struct inode *inode)
612 struct block_device *bdev = NULL;
614 spin_lock(&bdev_lock);
616 if (!sb_is_blkdev_sb(inode->i_sb))
617 bdev = inode->i_bdev;
620 spin_unlock(&bdev_lock);
623 iput(bdev->bd_inode);
626 int bd_claim(struct block_device *bdev, void *holder)
629 spin_lock(&bdev_lock);
631 /* first decide result */
632 if (bdev->bd_holder == holder)
633 res = 0; /* already a holder */
634 else if (bdev->bd_holder != NULL)
635 res = -EBUSY; /* held by someone else */
636 else if (bdev->bd_contains == bdev)
637 res = 0; /* is a whole device which isn't held */
639 else if (bdev->bd_contains->bd_holder == bd_claim)
640 res = 0; /* is a partition of a device that is being partitioned */
641 else if (bdev->bd_contains->bd_holder != NULL)
642 res = -EBUSY; /* is a partition of a held device */
644 res = 0; /* is a partition of an un-held device */
646 /* now impose change */
648 /* note that for a whole device bd_holders
649 * will be incremented twice, and bd_holder will
650 * be set to bd_claim before being set to holder
652 bdev->bd_contains->bd_holders ++;
653 bdev->bd_contains->bd_holder = bd_claim;
655 bdev->bd_holder = holder;
657 spin_unlock(&bdev_lock);
661 EXPORT_SYMBOL(bd_claim);
663 void bd_release(struct block_device *bdev)
665 spin_lock(&bdev_lock);
666 if (!--bdev->bd_contains->bd_holders)
667 bdev->bd_contains->bd_holder = NULL;
668 if (!--bdev->bd_holders)
669 bdev->bd_holder = NULL;
670 spin_unlock(&bdev_lock);
673 EXPORT_SYMBOL(bd_release);
677 * Functions for bd_claim_by_kobject / bd_release_from_kobject
679 * If a kobject is passed to bd_claim_by_kobject()
680 * and the kobject has a parent directory,
681 * following symlinks are created:
682 * o from the kobject to the claimed bdev
683 * o from "holders" directory of the bdev to the parent of the kobject
684 * bd_release_from_kobject() removes these symlinks.
687 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
688 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
689 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
690 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
693 static int add_symlink(struct kobject *from, struct kobject *to)
697 return sysfs_create_link(from, to, kobject_name(to));
700 static void del_symlink(struct kobject *from, struct kobject *to)
704 sysfs_remove_link(from, kobject_name(to));
708 * 'struct bd_holder' contains pointers to kobjects symlinked by
709 * bd_claim_by_kobject.
710 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
713 struct list_head list; /* chain of holders of the bdev */
714 int count; /* references from the holder */
715 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
716 struct kobject *hdev; /* e.g. "/block/dm-0" */
717 struct kobject *hdir; /* e.g. "/block/sda/holders" */
718 struct kobject *sdev; /* e.g. "/block/sda" */
722 * Get references of related kobjects at once.
723 * Returns 1 on success. 0 on failure.
725 * Should call bd_holder_release_dirs() after successful use.
727 static int bd_holder_grab_dirs(struct block_device *bdev,
728 struct bd_holder *bo)
733 bo->sdir = kobject_get(bo->sdir);
737 bo->hdev = kobject_get(bo->sdir->parent);
741 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
745 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
752 kobject_put(bo->sdev);
754 kobject_put(bo->hdev);
756 kobject_put(bo->sdir);
761 /* Put references of related kobjects at once. */
762 static void bd_holder_release_dirs(struct bd_holder *bo)
764 kobject_put(bo->hdir);
765 kobject_put(bo->sdev);
766 kobject_put(bo->hdev);
767 kobject_put(bo->sdir);
770 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
772 struct bd_holder *bo;
774 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
784 static void free_bd_holder(struct bd_holder *bo)
790 * find_bd_holder - find matching struct bd_holder from the block device
792 * @bdev: struct block device to be searched
793 * @bo: target struct bd_holder
795 * Returns matching entry with @bo in @bdev->bd_holder_list.
796 * If found, increment the reference count and return the pointer.
797 * If not found, returns NULL.
799 static struct bd_holder *find_bd_holder(struct block_device *bdev,
800 struct bd_holder *bo)
802 struct bd_holder *tmp;
804 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
805 if (tmp->sdir == bo->sdir) {
814 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
816 * @bdev: block device to be bd_claimed
817 * @bo: preallocated and initialized by alloc_bd_holder()
819 * Add @bo to @bdev->bd_holder_list, create symlinks.
821 * Returns 0 if symlinks are created.
822 * Returns -ve if something fails.
824 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
831 if (!bd_holder_grab_dirs(bdev, bo))
834 err = add_symlink(bo->sdir, bo->sdev);
838 err = add_symlink(bo->hdir, bo->hdev);
840 del_symlink(bo->sdir, bo->sdev);
844 list_add_tail(&bo->list, &bdev->bd_holder_list);
849 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
851 * @bdev: block device to be bd_claimed
852 * @kobj: holder's kobject
854 * If there is matching entry with @kobj in @bdev->bd_holder_list
855 * and no other bd_claim() from the same kobject,
856 * remove the struct bd_holder from the list, delete symlinks for it.
858 * Returns a pointer to the struct bd_holder when it's removed from the list
859 * and ready to be freed.
860 * Returns NULL if matching claim isn't found or there is other bd_claim()
861 * by the same kobject.
863 static struct bd_holder *del_bd_holder(struct block_device *bdev,
864 struct kobject *kobj)
866 struct bd_holder *bo;
868 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
869 if (bo->sdir == kobj) {
871 BUG_ON(bo->count < 0);
874 del_symlink(bo->sdir, bo->sdev);
875 del_symlink(bo->hdir, bo->hdev);
876 bd_holder_release_dirs(bo);
887 * bd_claim_by_kobject - bd_claim() with additional kobject signature
889 * @bdev: block device to be claimed
890 * @holder: holder's signature
891 * @kobj: holder's kobject
893 * Do bd_claim() and if it succeeds, create sysfs symlinks between
894 * the bdev and the holder's kobject.
895 * Use bd_release_from_kobject() when relesing the claimed bdev.
897 * Returns 0 on success. (same as bd_claim())
898 * Returns errno on failure.
900 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
901 struct kobject *kobj)
904 struct bd_holder *bo, *found;
909 bo = alloc_bd_holder(kobj);
913 mutex_lock(&bdev->bd_mutex);
915 err = bd_claim(bdev, holder);
919 found = find_bd_holder(bdev, bo);
923 err = add_bd_holder(bdev, bo);
929 mutex_unlock(&bdev->bd_mutex);
935 * bd_release_from_kobject - bd_release() with additional kobject signature
937 * @bdev: block device to be released
938 * @kobj: holder's kobject
940 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
942 static void bd_release_from_kobject(struct block_device *bdev,
943 struct kobject *kobj)
948 mutex_lock(&bdev->bd_mutex);
950 free_bd_holder(del_bd_holder(bdev, kobj));
951 mutex_unlock(&bdev->bd_mutex);
955 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
957 * @bdev: block device to be claimed
958 * @holder: holder's signature
959 * @disk: holder's gendisk
961 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
963 int bd_claim_by_disk(struct block_device *bdev, void *holder,
964 struct gendisk *disk)
966 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
968 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
971 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
973 * @bdev: block device to be claimed
974 * @disk: holder's gendisk
976 * Call bd_release_from_kobject() and put @disk->slave_dir.
978 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
980 bd_release_from_kobject(bdev, disk->slave_dir);
981 kobject_put(disk->slave_dir);
983 EXPORT_SYMBOL_GPL(bd_release_from_disk);
987 * Tries to open block device by device number. Use it ONLY if you
988 * really do not have anything better - i.e. when you are behind a
989 * truly sucky interface and all you are given is a device number. _Never_
990 * to be used for internal purposes. If you ever need it - reconsider
993 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
995 struct block_device *bdev = bdget(dev);
998 err = blkdev_get(bdev, mode);
999 return err ? ERR_PTR(err) : bdev;
1002 EXPORT_SYMBOL(open_by_devnum);
1005 * flush_disk - invalidates all buffer-cache entries on a disk
1007 * @bdev: struct block device to be flushed
1009 * Invalidates all buffer-cache entries on a disk. It should be called
1010 * when a disk has been changed -- either by a media change or online
1013 static void flush_disk(struct block_device *bdev)
1015 if (__invalidate_device(bdev)) {
1016 char name[BDEVNAME_SIZE] = "";
1019 disk_name(bdev->bd_disk, 0, name);
1020 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1021 "resized disk %s\n", name);
1026 if (disk_partitionable(bdev->bd_disk))
1027 bdev->bd_invalidated = 1;
1031 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1032 * @disk: struct gendisk to check
1033 * @bdev: struct bdev to adjust.
1035 * This routine checks to see if the bdev size does not match the disk size
1036 * and adjusts it if it differs.
1038 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1040 loff_t disk_size, bdev_size;
1042 disk_size = (loff_t)get_capacity(disk) << 9;
1043 bdev_size = i_size_read(bdev->bd_inode);
1044 if (disk_size != bdev_size) {
1045 char name[BDEVNAME_SIZE];
1047 disk_name(disk, 0, name);
1049 "%s: detected capacity change from %lld to %lld\n",
1050 name, bdev_size, disk_size);
1051 i_size_write(bdev->bd_inode, disk_size);
1055 EXPORT_SYMBOL(check_disk_size_change);
1058 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1059 * @disk: struct gendisk to be revalidated
1061 * This routine is a wrapper for lower-level driver's revalidate_disk
1062 * call-backs. It is used to do common pre and post operations needed
1063 * for all revalidate_disk operations.
1065 int revalidate_disk(struct gendisk *disk)
1067 struct block_device *bdev;
1070 if (disk->fops->revalidate_disk)
1071 ret = disk->fops->revalidate_disk(disk);
1073 bdev = bdget_disk(disk, 0);
1077 mutex_lock(&bdev->bd_mutex);
1078 check_disk_size_change(disk, bdev);
1079 mutex_unlock(&bdev->bd_mutex);
1083 EXPORT_SYMBOL(revalidate_disk);
1086 * This routine checks whether a removable media has been changed,
1087 * and invalidates all buffer-cache-entries in that case. This
1088 * is a relatively slow routine, so we have to try to minimize using
1089 * it. Thus it is called only upon a 'mount' or 'open'. This
1090 * is the best way of combining speed and utility, I think.
1091 * People changing diskettes in the middle of an operation deserve
1094 int check_disk_change(struct block_device *bdev)
1096 struct gendisk *disk = bdev->bd_disk;
1097 struct block_device_operations * bdops = disk->fops;
1099 if (!bdops->media_changed)
1101 if (!bdops->media_changed(bdev->bd_disk))
1105 if (bdops->revalidate_disk)
1106 bdops->revalidate_disk(bdev->bd_disk);
1110 EXPORT_SYMBOL(check_disk_change);
1112 void bd_set_size(struct block_device *bdev, loff_t size)
1114 unsigned bsize = bdev_hardsect_size(bdev);
1116 bdev->bd_inode->i_size = size;
1117 while (bsize < PAGE_CACHE_SIZE) {
1122 bdev->bd_block_size = bsize;
1123 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1125 EXPORT_SYMBOL(bd_set_size);
1127 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1132 * mutex_lock(part->bd_mutex)
1133 * mutex_lock_nested(whole->bd_mutex, 1)
1136 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1138 struct gendisk *disk;
1143 if (mode & FMODE_READ)
1145 if (mode & FMODE_WRITE)
1148 * hooks: /n/, see "layering violations".
1150 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1160 disk = get_gendisk(bdev->bd_dev, &partno);
1162 goto out_unlock_kernel;
1164 mutex_lock_nested(&bdev->bd_mutex, for_part);
1165 if (!bdev->bd_openers) {
1166 bdev->bd_disk = disk;
1167 bdev->bd_contains = bdev;
1169 struct backing_dev_info *bdi;
1172 bdev->bd_part = disk_get_part(disk, partno);
1176 if (disk->fops->open) {
1177 ret = disk->fops->open(bdev, mode);
1178 if (ret == -ERESTARTSYS) {
1179 /* Lost a race with 'disk' being
1180 * deleted, try again.
1183 disk_put_part(bdev->bd_part);
1184 bdev->bd_part = NULL;
1185 module_put(disk->fops->owner);
1187 bdev->bd_disk = NULL;
1188 mutex_unlock(&bdev->bd_mutex);
1194 if (!bdev->bd_openers) {
1195 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1196 bdi = blk_get_backing_dev_info(bdev);
1198 bdi = &default_backing_dev_info;
1199 bdev->bd_inode->i_data.backing_dev_info = bdi;
1201 if (bdev->bd_invalidated)
1202 rescan_partitions(disk, bdev);
1204 struct block_device *whole;
1205 whole = bdget_disk(disk, 0);
1210 ret = __blkdev_get(whole, mode, 1);
1213 bdev->bd_contains = whole;
1214 bdev->bd_inode->i_data.backing_dev_info =
1215 whole->bd_inode->i_data.backing_dev_info;
1216 bdev->bd_part = disk_get_part(disk, partno);
1217 if (!(disk->flags & GENHD_FL_UP) ||
1218 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1222 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1226 module_put(disk->fops->owner);
1228 if (bdev->bd_contains == bdev) {
1229 if (bdev->bd_disk->fops->open) {
1230 ret = bdev->bd_disk->fops->open(bdev, mode);
1232 goto out_unlock_bdev;
1234 if (bdev->bd_invalidated)
1235 rescan_partitions(bdev->bd_disk, bdev);
1240 bdev->bd_part_count++;
1241 mutex_unlock(&bdev->bd_mutex);
1246 disk_put_part(bdev->bd_part);
1247 bdev->bd_disk = NULL;
1248 bdev->bd_part = NULL;
1249 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1250 if (bdev != bdev->bd_contains)
1251 __blkdev_put(bdev->bd_contains, mode, 1);
1252 bdev->bd_contains = NULL;
1254 mutex_unlock(&bdev->bd_mutex);
1259 module_put(disk->fops->owner);
1266 int blkdev_get(struct block_device *bdev, fmode_t mode)
1268 return __blkdev_get(bdev, mode, 0);
1270 EXPORT_SYMBOL(blkdev_get);
1272 static int blkdev_open(struct inode * inode, struct file * filp)
1274 struct block_device *bdev;
1278 * Preserve backwards compatibility and allow large file access
1279 * even if userspace doesn't ask for it explicitly. Some mkfs
1280 * binary needs it. We might want to drop this workaround
1281 * during an unstable branch.
1283 filp->f_flags |= O_LARGEFILE;
1285 if (filp->f_flags & O_NDELAY)
1286 filp->f_mode |= FMODE_NDELAY;
1287 if (filp->f_flags & O_EXCL)
1288 filp->f_mode |= FMODE_EXCL;
1289 if ((filp->f_flags & O_ACCMODE) == 3)
1290 filp->f_mode |= FMODE_WRITE_IOCTL;
1292 bdev = bd_acquire(inode);
1296 filp->f_mapping = bdev->bd_inode->i_mapping;
1298 res = blkdev_get(bdev, filp->f_mode);
1302 if (filp->f_mode & FMODE_EXCL) {
1303 res = bd_claim(bdev, filp);
1305 goto out_blkdev_put;
1311 blkdev_put(bdev, filp->f_mode);
1315 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1318 struct gendisk *disk = bdev->bd_disk;
1319 struct block_device *victim = NULL;
1321 mutex_lock_nested(&bdev->bd_mutex, for_part);
1324 bdev->bd_part_count--;
1326 if (!--bdev->bd_openers) {
1327 sync_blockdev(bdev);
1330 if (bdev->bd_contains == bdev) {
1331 if (disk->fops->release)
1332 ret = disk->fops->release(disk, mode);
1334 if (!bdev->bd_openers) {
1335 struct module *owner = disk->fops->owner;
1339 disk_put_part(bdev->bd_part);
1340 bdev->bd_part = NULL;
1341 bdev->bd_disk = NULL;
1342 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1343 if (bdev != bdev->bd_contains)
1344 victim = bdev->bd_contains;
1345 bdev->bd_contains = NULL;
1348 mutex_unlock(&bdev->bd_mutex);
1351 __blkdev_put(victim, mode, 1);
1355 int blkdev_put(struct block_device *bdev, fmode_t mode)
1357 return __blkdev_put(bdev, mode, 0);
1359 EXPORT_SYMBOL(blkdev_put);
1361 static int blkdev_close(struct inode * inode, struct file * filp)
1363 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1364 if (bdev->bd_holder == filp)
1366 return blkdev_put(bdev, filp->f_mode);
1369 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1371 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1372 fmode_t mode = file->f_mode;
1375 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1376 * to updated it before every ioctl.
1378 if (file->f_flags & O_NDELAY)
1379 mode |= FMODE_NDELAY;
1381 mode &= ~FMODE_NDELAY;
1383 return blkdev_ioctl(bdev, mode, cmd, arg);
1387 * Try to release a page associated with block device when the system
1388 * is under memory pressure.
1390 static int blkdev_releasepage(struct page *page, gfp_t wait)
1392 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1394 if (super && super->s_op->bdev_try_to_free_page)
1395 return super->s_op->bdev_try_to_free_page(super, page, wait);
1397 return try_to_free_buffers(page);
1400 static const struct address_space_operations def_blk_aops = {
1401 .readpage = blkdev_readpage,
1402 .writepage = blkdev_writepage,
1403 .sync_page = block_sync_page,
1404 .write_begin = blkdev_write_begin,
1405 .write_end = blkdev_write_end,
1406 .writepages = generic_writepages,
1407 .releasepage = blkdev_releasepage,
1408 .direct_IO = blkdev_direct_IO,
1411 const struct file_operations def_blk_fops = {
1412 .open = blkdev_open,
1413 .release = blkdev_close,
1414 .llseek = block_llseek,
1415 .read = do_sync_read,
1416 .write = do_sync_write,
1417 .aio_read = generic_file_aio_read,
1418 .aio_write = generic_file_aio_write_nolock,
1419 .mmap = generic_file_mmap,
1420 .fsync = block_fsync,
1421 .unlocked_ioctl = block_ioctl,
1422 #ifdef CONFIG_COMPAT
1423 .compat_ioctl = compat_blkdev_ioctl,
1425 .splice_read = generic_file_splice_read,
1426 .splice_write = generic_file_splice_write,
1429 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1432 mm_segment_t old_fs = get_fs();
1434 res = blkdev_ioctl(bdev, 0, cmd, arg);
1439 EXPORT_SYMBOL(ioctl_by_bdev);
1442 * lookup_bdev - lookup a struct block_device by name
1443 * @pathname: special file representing the block device
1445 * Get a reference to the blockdevice at @pathname in the current
1446 * namespace if possible and return it. Return ERR_PTR(error)
1449 struct block_device *lookup_bdev(const char *pathname)
1451 struct block_device *bdev;
1452 struct inode *inode;
1456 if (!pathname || !*pathname)
1457 return ERR_PTR(-EINVAL);
1459 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1461 return ERR_PTR(error);
1463 inode = path.dentry->d_inode;
1465 if (!S_ISBLK(inode->i_mode))
1468 if (path.mnt->mnt_flags & MNT_NODEV)
1471 bdev = bd_acquire(inode);
1478 bdev = ERR_PTR(error);
1481 EXPORT_SYMBOL(lookup_bdev);
1484 * open_bdev_exclusive - open a block device by name and set it up for use
1486 * @path: special file representing the block device
1487 * @mode: FMODE_... combination to pass be used
1488 * @holder: owner for exclusion
1490 * Open the blockdevice described by the special file at @path, claim it
1493 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1495 struct block_device *bdev;
1498 bdev = lookup_bdev(path);
1502 error = blkdev_get(bdev, mode);
1504 return ERR_PTR(error);
1506 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1508 error = bd_claim(bdev, holder);
1515 blkdev_put(bdev, mode);
1516 return ERR_PTR(error);
1519 EXPORT_SYMBOL(open_bdev_exclusive);
1522 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1524 * @bdev: blockdevice to close
1525 * @mode: mode, must match that used to open.
1527 * This is the counterpart to open_bdev_exclusive().
1529 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1532 blkdev_put(bdev, mode);
1535 EXPORT_SYMBOL(close_bdev_exclusive);
1537 int __invalidate_device(struct block_device *bdev)
1539 struct super_block *sb = get_super(bdev);
1544 * no need to lock the super, get_super holds the
1545 * read mutex so the filesystem cannot go away
1546 * under us (->put_super runs with the write lock
1549 shrink_dcache_sb(sb);
1550 res = invalidate_inodes(sb);
1553 invalidate_bdev(bdev);
1556 EXPORT_SYMBOL(__invalidate_device);