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 if (sb->s_flags & MS_RDONLY) {
249 deactivate_locked_super(sb);
250 mutex_unlock(&bdev->bd_fsfreeze_mutex);
254 sb->s_frozen = SB_FREEZE_WRITE;
259 sb->s_frozen = SB_FREEZE_TRANS;
262 sync_blockdev(sb->s_bdev);
264 if (sb->s_op->freeze_fs) {
265 error = sb->s_op->freeze_fs(sb);
268 "VFS:Filesystem freeze failed\n");
269 sb->s_frozen = SB_UNFROZEN;
270 deactivate_locked_super(sb);
271 bdev->bd_fsfreeze_count--;
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273 return ERR_PTR(error);
276 up_write(&sb->s_umount);
280 mutex_unlock(&bdev->bd_fsfreeze_mutex);
281 return sb; /* thaw_bdev releases s->s_umount */
283 EXPORT_SYMBOL(freeze_bdev);
286 * thaw_bdev -- unlock filesystem
287 * @bdev: blockdevice to unlock
288 * @sb: associated superblock
290 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
292 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
296 mutex_lock(&bdev->bd_fsfreeze_mutex);
297 if (!bdev->bd_fsfreeze_count)
301 if (--bdev->bd_fsfreeze_count > 0)
307 BUG_ON(sb->s_bdev != bdev);
308 down_write(&sb->s_umount);
309 if (sb->s_flags & MS_RDONLY)
312 if (sb->s_op->unfreeze_fs) {
313 error = sb->s_op->unfreeze_fs(sb);
316 "VFS:Filesystem thaw failed\n");
317 sb->s_frozen = SB_FREEZE_TRANS;
318 bdev->bd_fsfreeze_count++;
319 mutex_unlock(&bdev->bd_fsfreeze_mutex);
324 sb->s_frozen = SB_UNFROZEN;
326 wake_up(&sb->s_wait_unfrozen);
330 deactivate_locked_super(sb);
332 mutex_unlock(&bdev->bd_fsfreeze_mutex);
335 EXPORT_SYMBOL(thaw_bdev);
337 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
339 return block_write_full_page(page, blkdev_get_block, wbc);
342 static int blkdev_readpage(struct file * file, struct page * page)
344 return block_read_full_page(page, blkdev_get_block);
347 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
348 loff_t pos, unsigned len, unsigned flags,
349 struct page **pagep, void **fsdata)
352 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
356 static int blkdev_write_end(struct file *file, struct address_space *mapping,
357 loff_t pos, unsigned len, unsigned copied,
358 struct page *page, void *fsdata)
361 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
364 page_cache_release(page);
371 * for a block special file file->f_path.dentry->d_inode->i_size is zero
372 * so we compute the size by hand (just as in block_read/write above)
374 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
376 struct inode *bd_inode = file->f_mapping->host;
380 mutex_lock(&bd_inode->i_mutex);
381 size = i_size_read(bd_inode);
388 offset += file->f_pos;
391 if (offset >= 0 && offset <= size) {
392 if (offset != file->f_pos) {
393 file->f_pos = offset;
397 mutex_unlock(&bd_inode->i_mutex);
402 * Filp is never NULL; the only case when ->fsync() is called with
403 * NULL first argument is nfsd_sync_dir() and that's not a directory.
406 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
408 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
411 error = sync_blockdev(bdev);
415 error = blkdev_issue_flush(bdev, NULL);
416 if (error == -EOPNOTSUPP)
425 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
426 static struct kmem_cache * bdev_cachep __read_mostly;
428 static struct inode *bdev_alloc_inode(struct super_block *sb)
430 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
433 return &ei->vfs_inode;
436 static void bdev_destroy_inode(struct inode *inode)
438 struct bdev_inode *bdi = BDEV_I(inode);
440 kmem_cache_free(bdev_cachep, bdi);
443 static void init_once(void *foo)
445 struct bdev_inode *ei = (struct bdev_inode *) foo;
446 struct block_device *bdev = &ei->bdev;
448 memset(bdev, 0, sizeof(*bdev));
449 mutex_init(&bdev->bd_mutex);
450 INIT_LIST_HEAD(&bdev->bd_inodes);
451 INIT_LIST_HEAD(&bdev->bd_list);
453 INIT_LIST_HEAD(&bdev->bd_holder_list);
455 inode_init_once(&ei->vfs_inode);
456 /* Initialize mutex for freeze. */
457 mutex_init(&bdev->bd_fsfreeze_mutex);
460 static inline void __bd_forget(struct inode *inode)
462 list_del_init(&inode->i_devices);
463 inode->i_bdev = NULL;
464 inode->i_mapping = &inode->i_data;
467 static void bdev_clear_inode(struct inode *inode)
469 struct block_device *bdev = &BDEV_I(inode)->bdev;
471 spin_lock(&bdev_lock);
472 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
473 __bd_forget(list_entry(p, struct inode, i_devices));
475 list_del_init(&bdev->bd_list);
476 spin_unlock(&bdev_lock);
479 static const struct super_operations bdev_sops = {
480 .statfs = simple_statfs,
481 .alloc_inode = bdev_alloc_inode,
482 .destroy_inode = bdev_destroy_inode,
483 .drop_inode = generic_delete_inode,
484 .clear_inode = bdev_clear_inode,
487 static int bd_get_sb(struct file_system_type *fs_type,
488 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
490 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
493 static struct file_system_type bd_type = {
496 .kill_sb = kill_anon_super,
499 struct super_block *blockdev_superblock __read_mostly;
501 void __init bdev_cache_init(void)
504 struct vfsmount *bd_mnt;
506 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
507 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
508 SLAB_MEM_SPREAD|SLAB_PANIC),
510 err = register_filesystem(&bd_type);
512 panic("Cannot register bdev pseudo-fs");
513 bd_mnt = kern_mount(&bd_type);
515 panic("Cannot create bdev pseudo-fs");
517 * This vfsmount structure is only used to obtain the
518 * blockdev_superblock, so tell kmemleak not to report it.
520 kmemleak_not_leak(bd_mnt);
521 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
525 * Most likely _very_ bad one - but then it's hardly critical for small
526 * /dev and can be fixed when somebody will need really large one.
527 * Keep in mind that it will be fed through icache hash function too.
529 static inline unsigned long hash(dev_t dev)
531 return MAJOR(dev)+MINOR(dev);
534 static int bdev_test(struct inode *inode, void *data)
536 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
539 static int bdev_set(struct inode *inode, void *data)
541 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
545 static LIST_HEAD(all_bdevs);
547 struct block_device *bdget(dev_t dev)
549 struct block_device *bdev;
552 inode = iget5_locked(blockdev_superblock, hash(dev),
553 bdev_test, bdev_set, &dev);
558 bdev = &BDEV_I(inode)->bdev;
560 if (inode->i_state & I_NEW) {
561 bdev->bd_contains = NULL;
562 bdev->bd_inode = inode;
563 bdev->bd_block_size = (1 << inode->i_blkbits);
564 bdev->bd_part_count = 0;
565 bdev->bd_invalidated = 0;
566 inode->i_mode = S_IFBLK;
568 inode->i_bdev = bdev;
569 inode->i_data.a_ops = &def_blk_aops;
570 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
571 inode->i_data.backing_dev_info = &default_backing_dev_info;
572 spin_lock(&bdev_lock);
573 list_add(&bdev->bd_list, &all_bdevs);
574 spin_unlock(&bdev_lock);
575 unlock_new_inode(inode);
580 EXPORT_SYMBOL(bdget);
583 * bdgrab -- Grab a reference to an already referenced block device
584 * @bdev: Block device to grab a reference to.
586 struct block_device *bdgrab(struct block_device *bdev)
588 atomic_inc(&bdev->bd_inode->i_count);
592 long nr_blockdev_pages(void)
594 struct block_device *bdev;
596 spin_lock(&bdev_lock);
597 list_for_each_entry(bdev, &all_bdevs, bd_list) {
598 ret += bdev->bd_inode->i_mapping->nrpages;
600 spin_unlock(&bdev_lock);
604 void bdput(struct block_device *bdev)
606 iput(bdev->bd_inode);
609 EXPORT_SYMBOL(bdput);
611 static struct block_device *bd_acquire(struct inode *inode)
613 struct block_device *bdev;
615 spin_lock(&bdev_lock);
616 bdev = inode->i_bdev;
618 atomic_inc(&bdev->bd_inode->i_count);
619 spin_unlock(&bdev_lock);
622 spin_unlock(&bdev_lock);
624 bdev = bdget(inode->i_rdev);
626 spin_lock(&bdev_lock);
627 if (!inode->i_bdev) {
629 * We take an additional bd_inode->i_count for inode,
630 * and it's released in clear_inode() of inode.
631 * So, we can access it via ->i_mapping always
634 atomic_inc(&bdev->bd_inode->i_count);
635 inode->i_bdev = bdev;
636 inode->i_mapping = bdev->bd_inode->i_mapping;
637 list_add(&inode->i_devices, &bdev->bd_inodes);
639 spin_unlock(&bdev_lock);
644 /* Call when you free inode */
646 void bd_forget(struct inode *inode)
648 struct block_device *bdev = NULL;
650 spin_lock(&bdev_lock);
652 if (!sb_is_blkdev_sb(inode->i_sb))
653 bdev = inode->i_bdev;
656 spin_unlock(&bdev_lock);
659 iput(bdev->bd_inode);
662 int bd_claim(struct block_device *bdev, void *holder)
665 spin_lock(&bdev_lock);
667 /* first decide result */
668 if (bdev->bd_holder == holder)
669 res = 0; /* already a holder */
670 else if (bdev->bd_holder != NULL)
671 res = -EBUSY; /* held by someone else */
672 else if (bdev->bd_contains == bdev)
673 res = 0; /* is a whole device which isn't held */
675 else if (bdev->bd_contains->bd_holder == bd_claim)
676 res = 0; /* is a partition of a device that is being partitioned */
677 else if (bdev->bd_contains->bd_holder != NULL)
678 res = -EBUSY; /* is a partition of a held device */
680 res = 0; /* is a partition of an un-held device */
682 /* now impose change */
684 /* note that for a whole device bd_holders
685 * will be incremented twice, and bd_holder will
686 * be set to bd_claim before being set to holder
688 bdev->bd_contains->bd_holders ++;
689 bdev->bd_contains->bd_holder = bd_claim;
691 bdev->bd_holder = holder;
693 spin_unlock(&bdev_lock);
697 EXPORT_SYMBOL(bd_claim);
699 void bd_release(struct block_device *bdev)
701 spin_lock(&bdev_lock);
702 if (!--bdev->bd_contains->bd_holders)
703 bdev->bd_contains->bd_holder = NULL;
704 if (!--bdev->bd_holders)
705 bdev->bd_holder = NULL;
706 spin_unlock(&bdev_lock);
709 EXPORT_SYMBOL(bd_release);
713 * Functions for bd_claim_by_kobject / bd_release_from_kobject
715 * If a kobject is passed to bd_claim_by_kobject()
716 * and the kobject has a parent directory,
717 * following symlinks are created:
718 * o from the kobject to the claimed bdev
719 * o from "holders" directory of the bdev to the parent of the kobject
720 * bd_release_from_kobject() removes these symlinks.
723 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
724 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
725 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
726 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
729 static int add_symlink(struct kobject *from, struct kobject *to)
733 return sysfs_create_link(from, to, kobject_name(to));
736 static void del_symlink(struct kobject *from, struct kobject *to)
740 sysfs_remove_link(from, kobject_name(to));
744 * 'struct bd_holder' contains pointers to kobjects symlinked by
745 * bd_claim_by_kobject.
746 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
749 struct list_head list; /* chain of holders of the bdev */
750 int count; /* references from the holder */
751 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
752 struct kobject *hdev; /* e.g. "/block/dm-0" */
753 struct kobject *hdir; /* e.g. "/block/sda/holders" */
754 struct kobject *sdev; /* e.g. "/block/sda" */
758 * Get references of related kobjects at once.
759 * Returns 1 on success. 0 on failure.
761 * Should call bd_holder_release_dirs() after successful use.
763 static int bd_holder_grab_dirs(struct block_device *bdev,
764 struct bd_holder *bo)
769 bo->sdir = kobject_get(bo->sdir);
773 bo->hdev = kobject_get(bo->sdir->parent);
777 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
781 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
788 kobject_put(bo->sdev);
790 kobject_put(bo->hdev);
792 kobject_put(bo->sdir);
797 /* Put references of related kobjects at once. */
798 static void bd_holder_release_dirs(struct bd_holder *bo)
800 kobject_put(bo->hdir);
801 kobject_put(bo->sdev);
802 kobject_put(bo->hdev);
803 kobject_put(bo->sdir);
806 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
808 struct bd_holder *bo;
810 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
820 static void free_bd_holder(struct bd_holder *bo)
826 * find_bd_holder - find matching struct bd_holder from the block device
828 * @bdev: struct block device to be searched
829 * @bo: target struct bd_holder
831 * Returns matching entry with @bo in @bdev->bd_holder_list.
832 * If found, increment the reference count and return the pointer.
833 * If not found, returns NULL.
835 static struct bd_holder *find_bd_holder(struct block_device *bdev,
836 struct bd_holder *bo)
838 struct bd_holder *tmp;
840 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
841 if (tmp->sdir == bo->sdir) {
850 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
852 * @bdev: block device to be bd_claimed
853 * @bo: preallocated and initialized by alloc_bd_holder()
855 * Add @bo to @bdev->bd_holder_list, create symlinks.
857 * Returns 0 if symlinks are created.
858 * Returns -ve if something fails.
860 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
867 if (!bd_holder_grab_dirs(bdev, bo))
870 err = add_symlink(bo->sdir, bo->sdev);
874 err = add_symlink(bo->hdir, bo->hdev);
876 del_symlink(bo->sdir, bo->sdev);
880 list_add_tail(&bo->list, &bdev->bd_holder_list);
885 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
887 * @bdev: block device to be bd_claimed
888 * @kobj: holder's kobject
890 * If there is matching entry with @kobj in @bdev->bd_holder_list
891 * and no other bd_claim() from the same kobject,
892 * remove the struct bd_holder from the list, delete symlinks for it.
894 * Returns a pointer to the struct bd_holder when it's removed from the list
895 * and ready to be freed.
896 * Returns NULL if matching claim isn't found or there is other bd_claim()
897 * by the same kobject.
899 static struct bd_holder *del_bd_holder(struct block_device *bdev,
900 struct kobject *kobj)
902 struct bd_holder *bo;
904 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
905 if (bo->sdir == kobj) {
907 BUG_ON(bo->count < 0);
910 del_symlink(bo->sdir, bo->sdev);
911 del_symlink(bo->hdir, bo->hdev);
912 bd_holder_release_dirs(bo);
923 * bd_claim_by_kobject - bd_claim() with additional kobject signature
925 * @bdev: block device to be claimed
926 * @holder: holder's signature
927 * @kobj: holder's kobject
929 * Do bd_claim() and if it succeeds, create sysfs symlinks between
930 * the bdev and the holder's kobject.
931 * Use bd_release_from_kobject() when relesing the claimed bdev.
933 * Returns 0 on success. (same as bd_claim())
934 * Returns errno on failure.
936 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
937 struct kobject *kobj)
940 struct bd_holder *bo, *found;
945 bo = alloc_bd_holder(kobj);
949 mutex_lock(&bdev->bd_mutex);
951 err = bd_claim(bdev, holder);
955 found = find_bd_holder(bdev, bo);
959 err = add_bd_holder(bdev, bo);
965 mutex_unlock(&bdev->bd_mutex);
971 * bd_release_from_kobject - bd_release() with additional kobject signature
973 * @bdev: block device to be released
974 * @kobj: holder's kobject
976 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
978 static void bd_release_from_kobject(struct block_device *bdev,
979 struct kobject *kobj)
984 mutex_lock(&bdev->bd_mutex);
986 free_bd_holder(del_bd_holder(bdev, kobj));
987 mutex_unlock(&bdev->bd_mutex);
991 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
993 * @bdev: block device to be claimed
994 * @holder: holder's signature
995 * @disk: holder's gendisk
997 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
999 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1000 struct gendisk *disk)
1002 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1004 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1007 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1009 * @bdev: block device to be claimed
1010 * @disk: holder's gendisk
1012 * Call bd_release_from_kobject() and put @disk->slave_dir.
1014 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1016 bd_release_from_kobject(bdev, disk->slave_dir);
1017 kobject_put(disk->slave_dir);
1019 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1023 * Tries to open block device by device number. Use it ONLY if you
1024 * really do not have anything better - i.e. when you are behind a
1025 * truly sucky interface and all you are given is a device number. _Never_
1026 * to be used for internal purposes. If you ever need it - reconsider
1029 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1031 struct block_device *bdev = bdget(dev);
1034 err = blkdev_get(bdev, mode);
1035 return err ? ERR_PTR(err) : bdev;
1038 EXPORT_SYMBOL(open_by_devnum);
1041 * flush_disk - invalidates all buffer-cache entries on a disk
1043 * @bdev: struct block device to be flushed
1045 * Invalidates all buffer-cache entries on a disk. It should be called
1046 * when a disk has been changed -- either by a media change or online
1049 static void flush_disk(struct block_device *bdev)
1051 if (__invalidate_device(bdev)) {
1052 char name[BDEVNAME_SIZE] = "";
1055 disk_name(bdev->bd_disk, 0, name);
1056 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1057 "resized disk %s\n", name);
1062 if (disk_partitionable(bdev->bd_disk))
1063 bdev->bd_invalidated = 1;
1067 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1068 * @disk: struct gendisk to check
1069 * @bdev: struct bdev to adjust.
1071 * This routine checks to see if the bdev size does not match the disk size
1072 * and adjusts it if it differs.
1074 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1076 loff_t disk_size, bdev_size;
1078 disk_size = (loff_t)get_capacity(disk) << 9;
1079 bdev_size = i_size_read(bdev->bd_inode);
1080 if (disk_size != bdev_size) {
1081 char name[BDEVNAME_SIZE];
1083 disk_name(disk, 0, name);
1085 "%s: detected capacity change from %lld to %lld\n",
1086 name, bdev_size, disk_size);
1087 i_size_write(bdev->bd_inode, disk_size);
1091 EXPORT_SYMBOL(check_disk_size_change);
1094 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1095 * @disk: struct gendisk to be revalidated
1097 * This routine is a wrapper for lower-level driver's revalidate_disk
1098 * call-backs. It is used to do common pre and post operations needed
1099 * for all revalidate_disk operations.
1101 int revalidate_disk(struct gendisk *disk)
1103 struct block_device *bdev;
1106 if (disk->fops->revalidate_disk)
1107 ret = disk->fops->revalidate_disk(disk);
1109 bdev = bdget_disk(disk, 0);
1113 mutex_lock(&bdev->bd_mutex);
1114 check_disk_size_change(disk, bdev);
1115 mutex_unlock(&bdev->bd_mutex);
1119 EXPORT_SYMBOL(revalidate_disk);
1122 * This routine checks whether a removable media has been changed,
1123 * and invalidates all buffer-cache-entries in that case. This
1124 * is a relatively slow routine, so we have to try to minimize using
1125 * it. Thus it is called only upon a 'mount' or 'open'. This
1126 * is the best way of combining speed and utility, I think.
1127 * People changing diskettes in the middle of an operation deserve
1130 int check_disk_change(struct block_device *bdev)
1132 struct gendisk *disk = bdev->bd_disk;
1133 const struct block_device_operations *bdops = disk->fops;
1135 if (!bdops->media_changed)
1137 if (!bdops->media_changed(bdev->bd_disk))
1141 if (bdops->revalidate_disk)
1142 bdops->revalidate_disk(bdev->bd_disk);
1146 EXPORT_SYMBOL(check_disk_change);
1148 void bd_set_size(struct block_device *bdev, loff_t size)
1150 unsigned bsize = bdev_logical_block_size(bdev);
1152 bdev->bd_inode->i_size = size;
1153 while (bsize < PAGE_CACHE_SIZE) {
1158 bdev->bd_block_size = bsize;
1159 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1161 EXPORT_SYMBOL(bd_set_size);
1163 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1168 * mutex_lock(part->bd_mutex)
1169 * mutex_lock_nested(whole->bd_mutex, 1)
1172 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1174 struct gendisk *disk;
1179 if (mode & FMODE_READ)
1181 if (mode & FMODE_WRITE)
1184 * hooks: /n/, see "layering violations".
1186 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1196 disk = get_gendisk(bdev->bd_dev, &partno);
1198 goto out_unlock_kernel;
1200 mutex_lock_nested(&bdev->bd_mutex, for_part);
1201 if (!bdev->bd_openers) {
1202 bdev->bd_disk = disk;
1203 bdev->bd_contains = bdev;
1205 struct backing_dev_info *bdi;
1208 bdev->bd_part = disk_get_part(disk, partno);
1212 if (disk->fops->open) {
1213 ret = disk->fops->open(bdev, mode);
1214 if (ret == -ERESTARTSYS) {
1215 /* Lost a race with 'disk' being
1216 * deleted, try again.
1219 disk_put_part(bdev->bd_part);
1220 bdev->bd_part = NULL;
1221 module_put(disk->fops->owner);
1223 bdev->bd_disk = NULL;
1224 mutex_unlock(&bdev->bd_mutex);
1230 if (!bdev->bd_openers) {
1231 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1232 bdi = blk_get_backing_dev_info(bdev);
1234 bdi = &default_backing_dev_info;
1235 bdev->bd_inode->i_data.backing_dev_info = bdi;
1237 if (bdev->bd_invalidated)
1238 rescan_partitions(disk, bdev);
1240 struct block_device *whole;
1241 whole = bdget_disk(disk, 0);
1246 ret = __blkdev_get(whole, mode, 1);
1249 bdev->bd_contains = whole;
1250 bdev->bd_inode->i_data.backing_dev_info =
1251 whole->bd_inode->i_data.backing_dev_info;
1252 bdev->bd_part = disk_get_part(disk, partno);
1253 if (!(disk->flags & GENHD_FL_UP) ||
1254 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1258 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1261 module_put(disk->fops->owner);
1264 if (bdev->bd_contains == bdev) {
1265 if (bdev->bd_disk->fops->open) {
1266 ret = bdev->bd_disk->fops->open(bdev, mode);
1268 goto out_unlock_bdev;
1270 if (bdev->bd_invalidated)
1271 rescan_partitions(bdev->bd_disk, bdev);
1276 bdev->bd_part_count++;
1277 mutex_unlock(&bdev->bd_mutex);
1282 disk_put_part(bdev->bd_part);
1283 bdev->bd_disk = NULL;
1284 bdev->bd_part = NULL;
1285 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1286 if (bdev != bdev->bd_contains)
1287 __blkdev_put(bdev->bd_contains, mode, 1);
1288 bdev->bd_contains = NULL;
1290 mutex_unlock(&bdev->bd_mutex);
1295 module_put(disk->fops->owner);
1302 int blkdev_get(struct block_device *bdev, fmode_t mode)
1304 return __blkdev_get(bdev, mode, 0);
1306 EXPORT_SYMBOL(blkdev_get);
1308 static int blkdev_open(struct inode * inode, struct file * filp)
1310 struct block_device *bdev;
1314 * Preserve backwards compatibility and allow large file access
1315 * even if userspace doesn't ask for it explicitly. Some mkfs
1316 * binary needs it. We might want to drop this workaround
1317 * during an unstable branch.
1319 filp->f_flags |= O_LARGEFILE;
1321 if (filp->f_flags & O_NDELAY)
1322 filp->f_mode |= FMODE_NDELAY;
1323 if (filp->f_flags & O_EXCL)
1324 filp->f_mode |= FMODE_EXCL;
1325 if ((filp->f_flags & O_ACCMODE) == 3)
1326 filp->f_mode |= FMODE_WRITE_IOCTL;
1328 bdev = bd_acquire(inode);
1332 filp->f_mapping = bdev->bd_inode->i_mapping;
1334 res = blkdev_get(bdev, filp->f_mode);
1338 if (filp->f_mode & FMODE_EXCL) {
1339 res = bd_claim(bdev, filp);
1341 goto out_blkdev_put;
1347 blkdev_put(bdev, filp->f_mode);
1351 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1354 struct gendisk *disk = bdev->bd_disk;
1355 struct block_device *victim = NULL;
1357 mutex_lock_nested(&bdev->bd_mutex, for_part);
1360 bdev->bd_part_count--;
1362 if (!--bdev->bd_openers) {
1363 sync_blockdev(bdev);
1366 if (bdev->bd_contains == bdev) {
1367 if (disk->fops->release)
1368 ret = disk->fops->release(disk, mode);
1370 if (!bdev->bd_openers) {
1371 struct module *owner = disk->fops->owner;
1375 disk_put_part(bdev->bd_part);
1376 bdev->bd_part = NULL;
1377 bdev->bd_disk = NULL;
1378 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1379 if (bdev != bdev->bd_contains)
1380 victim = bdev->bd_contains;
1381 bdev->bd_contains = NULL;
1384 mutex_unlock(&bdev->bd_mutex);
1387 __blkdev_put(victim, mode, 1);
1391 int blkdev_put(struct block_device *bdev, fmode_t mode)
1393 return __blkdev_put(bdev, mode, 0);
1395 EXPORT_SYMBOL(blkdev_put);
1397 static int blkdev_close(struct inode * inode, struct file * filp)
1399 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1400 if (bdev->bd_holder == filp)
1402 return blkdev_put(bdev, filp->f_mode);
1405 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1407 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1408 fmode_t mode = file->f_mode;
1411 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1412 * to updated it before every ioctl.
1414 if (file->f_flags & O_NDELAY)
1415 mode |= FMODE_NDELAY;
1417 mode &= ~FMODE_NDELAY;
1419 return blkdev_ioctl(bdev, mode, cmd, arg);
1423 * Write data to the block device. Only intended for the block device itself
1424 * and the raw driver which basically is a fake block device.
1426 * Does not take i_mutex for the write and thus is not for general purpose
1429 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1430 unsigned long nr_segs, loff_t pos)
1432 struct file *file = iocb->ki_filp;
1435 BUG_ON(iocb->ki_pos != pos);
1437 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1438 if (ret > 0 || ret == -EIOCBQUEUED) {
1441 err = generic_write_sync(file, pos, ret);
1442 if (err < 0 && ret > 0)
1447 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1450 * Try to release a page associated with block device when the system
1451 * is under memory pressure.
1453 static int blkdev_releasepage(struct page *page, gfp_t wait)
1455 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1457 if (super && super->s_op->bdev_try_to_free_page)
1458 return super->s_op->bdev_try_to_free_page(super, page, wait);
1460 return try_to_free_buffers(page);
1463 static const struct address_space_operations def_blk_aops = {
1464 .readpage = blkdev_readpage,
1465 .writepage = blkdev_writepage,
1466 .sync_page = block_sync_page,
1467 .write_begin = blkdev_write_begin,
1468 .write_end = blkdev_write_end,
1469 .writepages = generic_writepages,
1470 .releasepage = blkdev_releasepage,
1471 .direct_IO = blkdev_direct_IO,
1474 const struct file_operations def_blk_fops = {
1475 .open = blkdev_open,
1476 .release = blkdev_close,
1477 .llseek = block_llseek,
1478 .read = do_sync_read,
1479 .write = do_sync_write,
1480 .aio_read = generic_file_aio_read,
1481 .aio_write = blkdev_aio_write,
1482 .mmap = generic_file_mmap,
1483 .fsync = block_fsync,
1484 .unlocked_ioctl = block_ioctl,
1485 #ifdef CONFIG_COMPAT
1486 .compat_ioctl = compat_blkdev_ioctl,
1488 .splice_read = generic_file_splice_read,
1489 .splice_write = generic_file_splice_write,
1492 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1495 mm_segment_t old_fs = get_fs();
1497 res = blkdev_ioctl(bdev, 0, cmd, arg);
1502 EXPORT_SYMBOL(ioctl_by_bdev);
1505 * lookup_bdev - lookup a struct block_device by name
1506 * @pathname: special file representing the block device
1508 * Get a reference to the blockdevice at @pathname in the current
1509 * namespace if possible and return it. Return ERR_PTR(error)
1512 struct block_device *lookup_bdev(const char *pathname)
1514 struct block_device *bdev;
1515 struct inode *inode;
1519 if (!pathname || !*pathname)
1520 return ERR_PTR(-EINVAL);
1522 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1524 return ERR_PTR(error);
1526 inode = path.dentry->d_inode;
1528 if (!S_ISBLK(inode->i_mode))
1531 if (path.mnt->mnt_flags & MNT_NODEV)
1534 bdev = bd_acquire(inode);
1541 bdev = ERR_PTR(error);
1544 EXPORT_SYMBOL(lookup_bdev);
1547 * open_bdev_exclusive - open a block device by name and set it up for use
1549 * @path: special file representing the block device
1550 * @mode: FMODE_... combination to pass be used
1551 * @holder: owner for exclusion
1553 * Open the blockdevice described by the special file at @path, claim it
1556 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1558 struct block_device *bdev;
1561 bdev = lookup_bdev(path);
1565 error = blkdev_get(bdev, mode);
1567 return ERR_PTR(error);
1569 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1571 error = bd_claim(bdev, holder);
1578 blkdev_put(bdev, mode);
1579 return ERR_PTR(error);
1582 EXPORT_SYMBOL(open_bdev_exclusive);
1585 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1587 * @bdev: blockdevice to close
1588 * @mode: mode, must match that used to open.
1590 * This is the counterpart to open_bdev_exclusive().
1592 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1595 blkdev_put(bdev, mode);
1598 EXPORT_SYMBOL(close_bdev_exclusive);
1600 int __invalidate_device(struct block_device *bdev)
1602 struct super_block *sb = get_super(bdev);
1607 * no need to lock the super, get_super holds the
1608 * read mutex so the filesystem cannot go away
1609 * under us (->put_super runs with the write lock
1612 shrink_dcache_sb(sb);
1613 res = invalidate_inodes(sb);
1616 invalidate_bdev(bdev);
1619 EXPORT_SYMBOL(__invalidate_device);