4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/smp_lock.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
51 static sector_t max_block(struct block_device *bdev)
53 sector_t retval = ~((sector_t)0);
54 loff_t sz = i_size_read(bdev->bd_inode);
57 unsigned int size = block_size(bdev);
58 unsigned int sizebits = blksize_bits(size);
59 retval = (sz >> sizebits);
64 /* Kill _all_ buffers and pagecache , dirty or not.. */
65 static void kill_bdev(struct block_device *bdev)
67 if (bdev->bd_inode->i_mapping->nrpages == 0)
70 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
73 int set_blocksize(struct block_device *bdev, int size)
75 /* Size must be a power of two, and between 512 and PAGE_SIZE */
76 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
79 /* Size cannot be smaller than the size supported by the device */
80 if (size < bdev_logical_block_size(bdev))
83 /* Don't change the size if it is same as current */
84 if (bdev->bd_block_size != size) {
86 bdev->bd_block_size = size;
87 bdev->bd_inode->i_blkbits = blksize_bits(size);
93 EXPORT_SYMBOL(set_blocksize);
95 int sb_set_blocksize(struct super_block *sb, int size)
97 if (set_blocksize(sb->s_bdev, size))
99 /* If we get here, we know size is power of two
100 * and it's value is between 512 and PAGE_SIZE */
101 sb->s_blocksize = size;
102 sb->s_blocksize_bits = blksize_bits(size);
103 return sb->s_blocksize;
106 EXPORT_SYMBOL(sb_set_blocksize);
108 int sb_min_blocksize(struct super_block *sb, int size)
110 int minsize = bdev_logical_block_size(sb->s_bdev);
113 return sb_set_blocksize(sb, size);
116 EXPORT_SYMBOL(sb_min_blocksize);
119 blkdev_get_block(struct inode *inode, sector_t iblock,
120 struct buffer_head *bh, int create)
122 if (iblock >= max_block(I_BDEV(inode))) {
127 * for reads, we're just trying to fill a partial page.
128 * return a hole, they will have to call get_block again
129 * before they can fill it, and they will get -EIO at that
134 bh->b_bdev = I_BDEV(inode);
135 bh->b_blocknr = iblock;
136 set_buffer_mapped(bh);
141 blkdev_get_blocks(struct inode *inode, sector_t iblock,
142 struct buffer_head *bh, int create)
144 sector_t end_block = max_block(I_BDEV(inode));
145 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
147 if ((iblock + max_blocks) > end_block) {
148 max_blocks = end_block - iblock;
149 if ((long)max_blocks <= 0) {
151 return -EIO; /* write fully beyond EOF */
153 * It is a read which is fully beyond EOF. We return
154 * a !buffer_mapped buffer
160 bh->b_bdev = I_BDEV(inode);
161 bh->b_blocknr = iblock;
162 bh->b_size = max_blocks << inode->i_blkbits;
164 set_buffer_mapped(bh);
169 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
170 loff_t offset, unsigned long nr_segs)
172 struct file *file = iocb->ki_filp;
173 struct inode *inode = file->f_mapping->host;
175 return blockdev_direct_IO_no_locking(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 down_write(&sb->s_umount);
249 if (sb->s_flags & MS_RDONLY) {
250 sb->s_frozen = SB_FREEZE_TRANS;
251 up_write(&sb->s_umount);
252 mutex_unlock(&bdev->bd_fsfreeze_mutex);
256 sb->s_frozen = SB_FREEZE_WRITE;
261 sb->s_frozen = SB_FREEZE_TRANS;
264 sync_blockdev(sb->s_bdev);
266 if (sb->s_op->freeze_fs) {
267 error = sb->s_op->freeze_fs(sb);
270 "VFS:Filesystem freeze failed\n");
271 sb->s_frozen = SB_UNFROZEN;
272 deactivate_locked_super(sb);
273 bdev->bd_fsfreeze_count--;
274 mutex_unlock(&bdev->bd_fsfreeze_mutex);
275 return ERR_PTR(error);
278 up_write(&sb->s_umount);
282 mutex_unlock(&bdev->bd_fsfreeze_mutex);
283 return sb; /* thaw_bdev releases s->s_umount */
285 EXPORT_SYMBOL(freeze_bdev);
288 * thaw_bdev -- unlock filesystem
289 * @bdev: blockdevice to unlock
290 * @sb: associated superblock
292 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
294 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
298 mutex_lock(&bdev->bd_fsfreeze_mutex);
299 if (!bdev->bd_fsfreeze_count)
303 if (--bdev->bd_fsfreeze_count > 0)
309 BUG_ON(sb->s_bdev != bdev);
310 down_write(&sb->s_umount);
311 if (sb->s_flags & MS_RDONLY)
314 if (sb->s_op->unfreeze_fs) {
315 error = sb->s_op->unfreeze_fs(sb);
318 "VFS:Filesystem thaw failed\n");
319 sb->s_frozen = SB_FREEZE_TRANS;
320 bdev->bd_fsfreeze_count++;
321 mutex_unlock(&bdev->bd_fsfreeze_mutex);
327 sb->s_frozen = SB_UNFROZEN;
329 wake_up(&sb->s_wait_unfrozen);
332 deactivate_locked_super(sb);
334 mutex_unlock(&bdev->bd_fsfreeze_mutex);
337 EXPORT_SYMBOL(thaw_bdev);
339 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
341 return block_write_full_page(page, blkdev_get_block, wbc);
344 static int blkdev_readpage(struct file * file, struct page * page)
346 return block_read_full_page(page, blkdev_get_block);
349 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
350 loff_t pos, unsigned len, unsigned flags,
351 struct page **pagep, void **fsdata)
354 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
358 static int blkdev_write_end(struct file *file, struct address_space *mapping,
359 loff_t pos, unsigned len, unsigned copied,
360 struct page *page, void *fsdata)
363 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
366 page_cache_release(page);
373 * for a block special file file->f_path.dentry->d_inode->i_size is zero
374 * so we compute the size by hand (just as in block_read/write above)
376 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
378 struct inode *bd_inode = file->f_mapping->host;
382 mutex_lock(&bd_inode->i_mutex);
383 size = i_size_read(bd_inode);
390 offset += file->f_pos;
393 if (offset >= 0 && offset <= size) {
394 if (offset != file->f_pos) {
395 file->f_pos = offset;
399 mutex_unlock(&bd_inode->i_mutex);
404 * Filp is never NULL; the only case when ->fsync() is called with
405 * NULL first argument is nfsd_sync_dir() and that's not a directory.
408 int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync)
410 struct inode *bd_inode = filp->f_mapping->host;
411 struct block_device *bdev = I_BDEV(bd_inode);
415 * There is no need to serialise calls to blkdev_issue_flush with
416 * i_mutex and doing so causes performance issues with concurrent
417 * O_SYNC writers to a block device.
419 mutex_unlock(&bd_inode->i_mutex);
421 error = blkdev_issue_flush(bdev, NULL);
422 if (error == -EOPNOTSUPP)
425 mutex_lock(&bd_inode->i_mutex);
429 EXPORT_SYMBOL(blkdev_fsync);
435 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
436 static struct kmem_cache * bdev_cachep __read_mostly;
438 static struct inode *bdev_alloc_inode(struct super_block *sb)
440 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
443 return &ei->vfs_inode;
446 static void bdev_destroy_inode(struct inode *inode)
448 struct bdev_inode *bdi = BDEV_I(inode);
450 kmem_cache_free(bdev_cachep, bdi);
453 static void init_once(void *foo)
455 struct bdev_inode *ei = (struct bdev_inode *) foo;
456 struct block_device *bdev = &ei->bdev;
458 memset(bdev, 0, sizeof(*bdev));
459 mutex_init(&bdev->bd_mutex);
460 INIT_LIST_HEAD(&bdev->bd_inodes);
461 INIT_LIST_HEAD(&bdev->bd_list);
463 INIT_LIST_HEAD(&bdev->bd_holder_list);
465 inode_init_once(&ei->vfs_inode);
466 /* Initialize mutex for freeze. */
467 mutex_init(&bdev->bd_fsfreeze_mutex);
470 static inline void __bd_forget(struct inode *inode)
472 list_del_init(&inode->i_devices);
473 inode->i_bdev = NULL;
474 inode->i_mapping = &inode->i_data;
477 static void bdev_clear_inode(struct inode *inode)
479 struct block_device *bdev = &BDEV_I(inode)->bdev;
481 spin_lock(&bdev_lock);
482 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
483 __bd_forget(list_entry(p, struct inode, i_devices));
485 list_del_init(&bdev->bd_list);
486 spin_unlock(&bdev_lock);
489 static const struct super_operations bdev_sops = {
490 .statfs = simple_statfs,
491 .alloc_inode = bdev_alloc_inode,
492 .destroy_inode = bdev_destroy_inode,
493 .drop_inode = generic_delete_inode,
494 .clear_inode = bdev_clear_inode,
497 static int bd_get_sb(struct file_system_type *fs_type,
498 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
500 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
503 static struct file_system_type bd_type = {
506 .kill_sb = kill_anon_super,
509 struct super_block *blockdev_superblock __read_mostly;
511 void __init bdev_cache_init(void)
514 struct vfsmount *bd_mnt;
516 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
517 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
518 SLAB_MEM_SPREAD|SLAB_PANIC),
520 err = register_filesystem(&bd_type);
522 panic("Cannot register bdev pseudo-fs");
523 bd_mnt = kern_mount(&bd_type);
525 panic("Cannot create bdev pseudo-fs");
527 * This vfsmount structure is only used to obtain the
528 * blockdev_superblock, so tell kmemleak not to report it.
530 kmemleak_not_leak(bd_mnt);
531 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
535 * Most likely _very_ bad one - but then it's hardly critical for small
536 * /dev and can be fixed when somebody will need really large one.
537 * Keep in mind that it will be fed through icache hash function too.
539 static inline unsigned long hash(dev_t dev)
541 return MAJOR(dev)+MINOR(dev);
544 static int bdev_test(struct inode *inode, void *data)
546 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
549 static int bdev_set(struct inode *inode, void *data)
551 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
555 static LIST_HEAD(all_bdevs);
557 struct block_device *bdget(dev_t dev)
559 struct block_device *bdev;
562 inode = iget5_locked(blockdev_superblock, hash(dev),
563 bdev_test, bdev_set, &dev);
568 bdev = &BDEV_I(inode)->bdev;
570 if (inode->i_state & I_NEW) {
571 bdev->bd_contains = NULL;
572 bdev->bd_inode = inode;
573 bdev->bd_block_size = (1 << inode->i_blkbits);
574 bdev->bd_part_count = 0;
575 bdev->bd_invalidated = 0;
576 inode->i_mode = S_IFBLK;
578 inode->i_bdev = bdev;
579 inode->i_data.a_ops = &def_blk_aops;
580 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
581 inode->i_data.backing_dev_info = &default_backing_dev_info;
582 spin_lock(&bdev_lock);
583 list_add(&bdev->bd_list, &all_bdevs);
584 spin_unlock(&bdev_lock);
585 unlock_new_inode(inode);
590 EXPORT_SYMBOL(bdget);
593 * bdgrab -- Grab a reference to an already referenced block device
594 * @bdev: Block device to grab a reference to.
596 struct block_device *bdgrab(struct block_device *bdev)
598 atomic_inc(&bdev->bd_inode->i_count);
602 long nr_blockdev_pages(void)
604 struct block_device *bdev;
606 spin_lock(&bdev_lock);
607 list_for_each_entry(bdev, &all_bdevs, bd_list) {
608 ret += bdev->bd_inode->i_mapping->nrpages;
610 spin_unlock(&bdev_lock);
614 void bdput(struct block_device *bdev)
616 iput(bdev->bd_inode);
619 EXPORT_SYMBOL(bdput);
621 static struct block_device *bd_acquire(struct inode *inode)
623 struct block_device *bdev;
625 spin_lock(&bdev_lock);
626 bdev = inode->i_bdev;
628 atomic_inc(&bdev->bd_inode->i_count);
629 spin_unlock(&bdev_lock);
632 spin_unlock(&bdev_lock);
634 bdev = bdget(inode->i_rdev);
636 spin_lock(&bdev_lock);
637 if (!inode->i_bdev) {
639 * We take an additional bd_inode->i_count for inode,
640 * and it's released in clear_inode() of inode.
641 * So, we can access it via ->i_mapping always
644 atomic_inc(&bdev->bd_inode->i_count);
645 inode->i_bdev = bdev;
646 inode->i_mapping = bdev->bd_inode->i_mapping;
647 list_add(&inode->i_devices, &bdev->bd_inodes);
649 spin_unlock(&bdev_lock);
654 /* Call when you free inode */
656 void bd_forget(struct inode *inode)
658 struct block_device *bdev = NULL;
660 spin_lock(&bdev_lock);
662 if (!sb_is_blkdev_sb(inode->i_sb))
663 bdev = inode->i_bdev;
666 spin_unlock(&bdev_lock);
669 iput(bdev->bd_inode);
672 int bd_claim(struct block_device *bdev, void *holder)
675 spin_lock(&bdev_lock);
677 /* first decide result */
678 if (bdev->bd_holder == holder)
679 res = 0; /* already a holder */
680 else if (bdev->bd_holder != NULL)
681 res = -EBUSY; /* held by someone else */
682 else if (bdev->bd_contains == bdev)
683 res = 0; /* is a whole device which isn't held */
685 else if (bdev->bd_contains->bd_holder == bd_claim)
686 res = 0; /* is a partition of a device that is being partitioned */
687 else if (bdev->bd_contains->bd_holder != NULL)
688 res = -EBUSY; /* is a partition of a held device */
690 res = 0; /* is a partition of an un-held device */
692 /* now impose change */
694 /* note that for a whole device bd_holders
695 * will be incremented twice, and bd_holder will
696 * be set to bd_claim before being set to holder
698 bdev->bd_contains->bd_holders ++;
699 bdev->bd_contains->bd_holder = bd_claim;
701 bdev->bd_holder = holder;
703 spin_unlock(&bdev_lock);
707 EXPORT_SYMBOL(bd_claim);
709 void bd_release(struct block_device *bdev)
711 spin_lock(&bdev_lock);
712 if (!--bdev->bd_contains->bd_holders)
713 bdev->bd_contains->bd_holder = NULL;
714 if (!--bdev->bd_holders)
715 bdev->bd_holder = NULL;
716 spin_unlock(&bdev_lock);
719 EXPORT_SYMBOL(bd_release);
723 * Functions for bd_claim_by_kobject / bd_release_from_kobject
725 * If a kobject is passed to bd_claim_by_kobject()
726 * and the kobject has a parent directory,
727 * following symlinks are created:
728 * o from the kobject to the claimed bdev
729 * o from "holders" directory of the bdev to the parent of the kobject
730 * bd_release_from_kobject() removes these symlinks.
733 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
734 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
735 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
736 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
739 static int add_symlink(struct kobject *from, struct kobject *to)
743 return sysfs_create_link(from, to, kobject_name(to));
746 static void del_symlink(struct kobject *from, struct kobject *to)
750 sysfs_remove_link(from, kobject_name(to));
754 * 'struct bd_holder' contains pointers to kobjects symlinked by
755 * bd_claim_by_kobject.
756 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
759 struct list_head list; /* chain of holders of the bdev */
760 int count; /* references from the holder */
761 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
762 struct kobject *hdev; /* e.g. "/block/dm-0" */
763 struct kobject *hdir; /* e.g. "/block/sda/holders" */
764 struct kobject *sdev; /* e.g. "/block/sda" */
768 * Get references of related kobjects at once.
769 * Returns 1 on success. 0 on failure.
771 * Should call bd_holder_release_dirs() after successful use.
773 static int bd_holder_grab_dirs(struct block_device *bdev,
774 struct bd_holder *bo)
779 bo->sdir = kobject_get(bo->sdir);
783 bo->hdev = kobject_get(bo->sdir->parent);
787 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
791 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
798 kobject_put(bo->sdev);
800 kobject_put(bo->hdev);
802 kobject_put(bo->sdir);
807 /* Put references of related kobjects at once. */
808 static void bd_holder_release_dirs(struct bd_holder *bo)
810 kobject_put(bo->hdir);
811 kobject_put(bo->sdev);
812 kobject_put(bo->hdev);
813 kobject_put(bo->sdir);
816 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
818 struct bd_holder *bo;
820 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
830 static void free_bd_holder(struct bd_holder *bo)
836 * find_bd_holder - find matching struct bd_holder from the block device
838 * @bdev: struct block device to be searched
839 * @bo: target struct bd_holder
841 * Returns matching entry with @bo in @bdev->bd_holder_list.
842 * If found, increment the reference count and return the pointer.
843 * If not found, returns NULL.
845 static struct bd_holder *find_bd_holder(struct block_device *bdev,
846 struct bd_holder *bo)
848 struct bd_holder *tmp;
850 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
851 if (tmp->sdir == bo->sdir) {
860 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
862 * @bdev: block device to be bd_claimed
863 * @bo: preallocated and initialized by alloc_bd_holder()
865 * Add @bo to @bdev->bd_holder_list, create symlinks.
867 * Returns 0 if symlinks are created.
868 * Returns -ve if something fails.
870 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
877 if (!bd_holder_grab_dirs(bdev, bo))
880 err = add_symlink(bo->sdir, bo->sdev);
884 err = add_symlink(bo->hdir, bo->hdev);
886 del_symlink(bo->sdir, bo->sdev);
890 list_add_tail(&bo->list, &bdev->bd_holder_list);
895 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
897 * @bdev: block device to be bd_claimed
898 * @kobj: holder's kobject
900 * If there is matching entry with @kobj in @bdev->bd_holder_list
901 * and no other bd_claim() from the same kobject,
902 * remove the struct bd_holder from the list, delete symlinks for it.
904 * Returns a pointer to the struct bd_holder when it's removed from the list
905 * and ready to be freed.
906 * Returns NULL if matching claim isn't found or there is other bd_claim()
907 * by the same kobject.
909 static struct bd_holder *del_bd_holder(struct block_device *bdev,
910 struct kobject *kobj)
912 struct bd_holder *bo;
914 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
915 if (bo->sdir == kobj) {
917 BUG_ON(bo->count < 0);
920 del_symlink(bo->sdir, bo->sdev);
921 del_symlink(bo->hdir, bo->hdev);
922 bd_holder_release_dirs(bo);
933 * bd_claim_by_kobject - bd_claim() with additional kobject signature
935 * @bdev: block device to be claimed
936 * @holder: holder's signature
937 * @kobj: holder's kobject
939 * Do bd_claim() and if it succeeds, create sysfs symlinks between
940 * the bdev and the holder's kobject.
941 * Use bd_release_from_kobject() when relesing the claimed bdev.
943 * Returns 0 on success. (same as bd_claim())
944 * Returns errno on failure.
946 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
947 struct kobject *kobj)
950 struct bd_holder *bo, *found;
955 bo = alloc_bd_holder(kobj);
959 mutex_lock(&bdev->bd_mutex);
961 err = bd_claim(bdev, holder);
965 found = find_bd_holder(bdev, bo);
969 err = add_bd_holder(bdev, bo);
975 mutex_unlock(&bdev->bd_mutex);
981 * bd_release_from_kobject - bd_release() with additional kobject signature
983 * @bdev: block device to be released
984 * @kobj: holder's kobject
986 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
988 static void bd_release_from_kobject(struct block_device *bdev,
989 struct kobject *kobj)
994 mutex_lock(&bdev->bd_mutex);
996 free_bd_holder(del_bd_holder(bdev, kobj));
997 mutex_unlock(&bdev->bd_mutex);
1001 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1003 * @bdev: block device to be claimed
1004 * @holder: holder's signature
1005 * @disk: holder's gendisk
1007 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1009 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1010 struct gendisk *disk)
1012 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1014 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1017 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1019 * @bdev: block device to be claimed
1020 * @disk: holder's gendisk
1022 * Call bd_release_from_kobject() and put @disk->slave_dir.
1024 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1026 bd_release_from_kobject(bdev, disk->slave_dir);
1027 kobject_put(disk->slave_dir);
1029 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1033 * Tries to open block device by device number. Use it ONLY if you
1034 * really do not have anything better - i.e. when you are behind a
1035 * truly sucky interface and all you are given is a device number. _Never_
1036 * to be used for internal purposes. If you ever need it - reconsider
1039 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1041 struct block_device *bdev = bdget(dev);
1044 err = blkdev_get(bdev, mode);
1045 return err ? ERR_PTR(err) : bdev;
1048 EXPORT_SYMBOL(open_by_devnum);
1051 * flush_disk - invalidates all buffer-cache entries on a disk
1053 * @bdev: struct block device to be flushed
1055 * Invalidates all buffer-cache entries on a disk. It should be called
1056 * when a disk has been changed -- either by a media change or online
1059 static void flush_disk(struct block_device *bdev)
1061 if (__invalidate_device(bdev)) {
1062 char name[BDEVNAME_SIZE] = "";
1065 disk_name(bdev->bd_disk, 0, name);
1066 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1067 "resized disk %s\n", name);
1072 if (disk_partitionable(bdev->bd_disk))
1073 bdev->bd_invalidated = 1;
1077 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1078 * @disk: struct gendisk to check
1079 * @bdev: struct bdev to adjust.
1081 * This routine checks to see if the bdev size does not match the disk size
1082 * and adjusts it if it differs.
1084 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1086 loff_t disk_size, bdev_size;
1088 disk_size = (loff_t)get_capacity(disk) << 9;
1089 bdev_size = i_size_read(bdev->bd_inode);
1090 if (disk_size != bdev_size) {
1091 char name[BDEVNAME_SIZE];
1093 disk_name(disk, 0, name);
1095 "%s: detected capacity change from %lld to %lld\n",
1096 name, bdev_size, disk_size);
1097 i_size_write(bdev->bd_inode, disk_size);
1101 EXPORT_SYMBOL(check_disk_size_change);
1104 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1105 * @disk: struct gendisk to be revalidated
1107 * This routine is a wrapper for lower-level driver's revalidate_disk
1108 * call-backs. It is used to do common pre and post operations needed
1109 * for all revalidate_disk operations.
1111 int revalidate_disk(struct gendisk *disk)
1113 struct block_device *bdev;
1116 if (disk->fops->revalidate_disk)
1117 ret = disk->fops->revalidate_disk(disk);
1119 bdev = bdget_disk(disk, 0);
1123 mutex_lock(&bdev->bd_mutex);
1124 check_disk_size_change(disk, bdev);
1125 mutex_unlock(&bdev->bd_mutex);
1129 EXPORT_SYMBOL(revalidate_disk);
1132 * This routine checks whether a removable media has been changed,
1133 * and invalidates all buffer-cache-entries in that case. This
1134 * is a relatively slow routine, so we have to try to minimize using
1135 * it. Thus it is called only upon a 'mount' or 'open'. This
1136 * is the best way of combining speed and utility, I think.
1137 * People changing diskettes in the middle of an operation deserve
1140 int check_disk_change(struct block_device *bdev)
1142 struct gendisk *disk = bdev->bd_disk;
1143 const struct block_device_operations *bdops = disk->fops;
1145 if (!bdops->media_changed)
1147 if (!bdops->media_changed(bdev->bd_disk))
1151 if (bdops->revalidate_disk)
1152 bdops->revalidate_disk(bdev->bd_disk);
1156 EXPORT_SYMBOL(check_disk_change);
1158 void bd_set_size(struct block_device *bdev, loff_t size)
1160 unsigned bsize = bdev_logical_block_size(bdev);
1162 bdev->bd_inode->i_size = size;
1163 while (bsize < PAGE_CACHE_SIZE) {
1168 bdev->bd_block_size = bsize;
1169 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1171 EXPORT_SYMBOL(bd_set_size);
1173 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1178 * mutex_lock(part->bd_mutex)
1179 * mutex_lock_nested(whole->bd_mutex, 1)
1182 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1184 struct gendisk *disk;
1189 if (mode & FMODE_READ)
1191 if (mode & FMODE_WRITE)
1194 * hooks: /n/, see "layering violations".
1196 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1206 disk = get_gendisk(bdev->bd_dev, &partno);
1208 goto out_unlock_kernel;
1210 mutex_lock_nested(&bdev->bd_mutex, for_part);
1211 if (!bdev->bd_openers) {
1212 bdev->bd_disk = disk;
1213 bdev->bd_contains = bdev;
1215 struct backing_dev_info *bdi;
1218 bdev->bd_part = disk_get_part(disk, partno);
1222 if (disk->fops->open) {
1223 ret = disk->fops->open(bdev, mode);
1224 if (ret == -ERESTARTSYS) {
1225 /* Lost a race with 'disk' being
1226 * deleted, try again.
1229 disk_put_part(bdev->bd_part);
1230 bdev->bd_part = NULL;
1231 module_put(disk->fops->owner);
1233 bdev->bd_disk = NULL;
1234 mutex_unlock(&bdev->bd_mutex);
1240 if (!bdev->bd_openers) {
1241 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1242 bdi = blk_get_backing_dev_info(bdev);
1244 bdi = &default_backing_dev_info;
1245 bdev->bd_inode->i_data.backing_dev_info = bdi;
1247 if (bdev->bd_invalidated)
1248 rescan_partitions(disk, bdev);
1250 struct block_device *whole;
1251 whole = bdget_disk(disk, 0);
1256 ret = __blkdev_get(whole, mode, 1);
1259 bdev->bd_contains = whole;
1260 bdev->bd_inode->i_data.backing_dev_info =
1261 whole->bd_inode->i_data.backing_dev_info;
1262 bdev->bd_part = disk_get_part(disk, partno);
1263 if (!(disk->flags & GENHD_FL_UP) ||
1264 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1268 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1271 module_put(disk->fops->owner);
1274 if (bdev->bd_contains == bdev) {
1275 if (bdev->bd_disk->fops->open) {
1276 ret = bdev->bd_disk->fops->open(bdev, mode);
1278 goto out_unlock_bdev;
1280 if (bdev->bd_invalidated)
1281 rescan_partitions(bdev->bd_disk, bdev);
1286 bdev->bd_part_count++;
1287 mutex_unlock(&bdev->bd_mutex);
1292 disk_put_part(bdev->bd_part);
1293 bdev->bd_disk = NULL;
1294 bdev->bd_part = NULL;
1295 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1296 if (bdev != bdev->bd_contains)
1297 __blkdev_put(bdev->bd_contains, mode, 1);
1298 bdev->bd_contains = NULL;
1300 mutex_unlock(&bdev->bd_mutex);
1305 module_put(disk->fops->owner);
1312 int blkdev_get(struct block_device *bdev, fmode_t mode)
1314 return __blkdev_get(bdev, mode, 0);
1316 EXPORT_SYMBOL(blkdev_get);
1318 static int blkdev_open(struct inode * inode, struct file * filp)
1320 struct block_device *bdev;
1324 * Preserve backwards compatibility and allow large file access
1325 * even if userspace doesn't ask for it explicitly. Some mkfs
1326 * binary needs it. We might want to drop this workaround
1327 * during an unstable branch.
1329 filp->f_flags |= O_LARGEFILE;
1331 if (filp->f_flags & O_NDELAY)
1332 filp->f_mode |= FMODE_NDELAY;
1333 if (filp->f_flags & O_EXCL)
1334 filp->f_mode |= FMODE_EXCL;
1335 if ((filp->f_flags & O_ACCMODE) == 3)
1336 filp->f_mode |= FMODE_WRITE_IOCTL;
1338 bdev = bd_acquire(inode);
1342 filp->f_mapping = bdev->bd_inode->i_mapping;
1344 res = blkdev_get(bdev, filp->f_mode);
1348 if (filp->f_mode & FMODE_EXCL) {
1349 res = bd_claim(bdev, filp);
1351 goto out_blkdev_put;
1357 blkdev_put(bdev, filp->f_mode);
1361 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1364 struct gendisk *disk = bdev->bd_disk;
1365 struct block_device *victim = NULL;
1367 mutex_lock_nested(&bdev->bd_mutex, for_part);
1370 bdev->bd_part_count--;
1372 if (!--bdev->bd_openers) {
1373 sync_blockdev(bdev);
1376 if (bdev->bd_contains == bdev) {
1377 if (disk->fops->release)
1378 ret = disk->fops->release(disk, mode);
1380 if (!bdev->bd_openers) {
1381 struct module *owner = disk->fops->owner;
1385 disk_put_part(bdev->bd_part);
1386 bdev->bd_part = NULL;
1387 bdev->bd_disk = NULL;
1388 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1389 if (bdev != bdev->bd_contains)
1390 victim = bdev->bd_contains;
1391 bdev->bd_contains = NULL;
1394 mutex_unlock(&bdev->bd_mutex);
1397 __blkdev_put(victim, mode, 1);
1401 int blkdev_put(struct block_device *bdev, fmode_t mode)
1403 return __blkdev_put(bdev, mode, 0);
1405 EXPORT_SYMBOL(blkdev_put);
1407 static int blkdev_close(struct inode * inode, struct file * filp)
1409 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1410 if (bdev->bd_holder == filp)
1412 return blkdev_put(bdev, filp->f_mode);
1415 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1417 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1418 fmode_t mode = file->f_mode;
1421 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1422 * to updated it before every ioctl.
1424 if (file->f_flags & O_NDELAY)
1425 mode |= FMODE_NDELAY;
1427 mode &= ~FMODE_NDELAY;
1429 return blkdev_ioctl(bdev, mode, cmd, arg);
1433 * Write data to the block device. Only intended for the block device itself
1434 * and the raw driver which basically is a fake block device.
1436 * Does not take i_mutex for the write and thus is not for general purpose
1439 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1440 unsigned long nr_segs, loff_t pos)
1442 struct file *file = iocb->ki_filp;
1445 BUG_ON(iocb->ki_pos != pos);
1447 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1448 if (ret > 0 || ret == -EIOCBQUEUED) {
1451 err = generic_write_sync(file, pos, ret);
1452 if (err < 0 && ret > 0)
1457 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1460 * Try to release a page associated with block device when the system
1461 * is under memory pressure.
1463 static int blkdev_releasepage(struct page *page, gfp_t wait)
1465 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1467 if (super && super->s_op->bdev_try_to_free_page)
1468 return super->s_op->bdev_try_to_free_page(super, page, wait);
1470 return try_to_free_buffers(page);
1473 static const struct address_space_operations def_blk_aops = {
1474 .readpage = blkdev_readpage,
1475 .writepage = blkdev_writepage,
1476 .sync_page = block_sync_page,
1477 .write_begin = blkdev_write_begin,
1478 .write_end = blkdev_write_end,
1479 .writepages = generic_writepages,
1480 .releasepage = blkdev_releasepage,
1481 .direct_IO = blkdev_direct_IO,
1484 const struct file_operations def_blk_fops = {
1485 .open = blkdev_open,
1486 .release = blkdev_close,
1487 .llseek = block_llseek,
1488 .read = do_sync_read,
1489 .write = do_sync_write,
1490 .aio_read = generic_file_aio_read,
1491 .aio_write = blkdev_aio_write,
1492 .mmap = generic_file_mmap,
1493 .fsync = blkdev_fsync,
1494 .unlocked_ioctl = block_ioctl,
1495 #ifdef CONFIG_COMPAT
1496 .compat_ioctl = compat_blkdev_ioctl,
1498 .splice_read = generic_file_splice_read,
1499 .splice_write = generic_file_splice_write,
1502 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1505 mm_segment_t old_fs = get_fs();
1507 res = blkdev_ioctl(bdev, 0, cmd, arg);
1512 EXPORT_SYMBOL(ioctl_by_bdev);
1515 * lookup_bdev - lookup a struct block_device by name
1516 * @pathname: special file representing the block device
1518 * Get a reference to the blockdevice at @pathname in the current
1519 * namespace if possible and return it. Return ERR_PTR(error)
1522 struct block_device *lookup_bdev(const char *pathname)
1524 struct block_device *bdev;
1525 struct inode *inode;
1529 if (!pathname || !*pathname)
1530 return ERR_PTR(-EINVAL);
1532 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1534 return ERR_PTR(error);
1536 inode = path.dentry->d_inode;
1538 if (!S_ISBLK(inode->i_mode))
1541 if (path.mnt->mnt_flags & MNT_NODEV)
1544 bdev = bd_acquire(inode);
1551 bdev = ERR_PTR(error);
1554 EXPORT_SYMBOL(lookup_bdev);
1557 * open_bdev_exclusive - open a block device by name and set it up for use
1559 * @path: special file representing the block device
1560 * @mode: FMODE_... combination to pass be used
1561 * @holder: owner for exclusion
1563 * Open the blockdevice described by the special file at @path, claim it
1566 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1568 struct block_device *bdev;
1571 bdev = lookup_bdev(path);
1575 error = blkdev_get(bdev, mode);
1577 return ERR_PTR(error);
1579 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1581 error = bd_claim(bdev, holder);
1588 blkdev_put(bdev, mode);
1589 return ERR_PTR(error);
1592 EXPORT_SYMBOL(open_bdev_exclusive);
1595 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1597 * @bdev: blockdevice to close
1598 * @mode: mode, must match that used to open.
1600 * This is the counterpart to open_bdev_exclusive().
1602 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1605 blkdev_put(bdev, mode);
1608 EXPORT_SYMBOL(close_bdev_exclusive);
1610 int __invalidate_device(struct block_device *bdev)
1612 struct super_block *sb = get_super(bdev);
1617 * no need to lock the super, get_super holds the
1618 * read mutex so the filesystem cannot go away
1619 * under us (->put_super runs with the write lock
1622 shrink_dcache_sb(sb);
1623 res = invalidate_inodes(sb);
1626 invalidate_bdev(bdev);
1629 EXPORT_SYMBOL(__invalidate_device);