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 > 0) {
234 bdev->bd_fsfreeze_count++;
235 sb = get_super(bdev);
236 mutex_unlock(&bdev->bd_fsfreeze_mutex);
239 bdev->bd_fsfreeze_count++;
241 sb = get_super(bdev);
242 if (sb && !(sb->s_flags & MS_RDONLY)) {
243 sb->s_frozen = SB_FREEZE_WRITE;
248 sb->s_frozen = SB_FREEZE_TRANS;
251 sync_blockdev(sb->s_bdev);
253 if (sb->s_op->freeze_fs) {
254 error = sb->s_op->freeze_fs(sb);
257 "VFS:Filesystem freeze failed\n");
258 sb->s_frozen = SB_UNFROZEN;
260 bdev->bd_fsfreeze_count--;
261 mutex_unlock(&bdev->bd_fsfreeze_mutex);
262 return ERR_PTR(error);
268 mutex_unlock(&bdev->bd_fsfreeze_mutex);
270 return sb; /* thaw_bdev releases s->s_umount */
272 EXPORT_SYMBOL(freeze_bdev);
275 * thaw_bdev -- unlock filesystem
276 * @bdev: blockdevice to unlock
277 * @sb: associated superblock
279 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
281 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
285 mutex_lock(&bdev->bd_fsfreeze_mutex);
286 if (!bdev->bd_fsfreeze_count) {
287 mutex_unlock(&bdev->bd_fsfreeze_mutex);
291 bdev->bd_fsfreeze_count--;
292 if (bdev->bd_fsfreeze_count > 0) {
295 mutex_unlock(&bdev->bd_fsfreeze_mutex);
300 BUG_ON(sb->s_bdev != bdev);
301 if (!(sb->s_flags & MS_RDONLY)) {
302 if (sb->s_op->unfreeze_fs) {
303 error = sb->s_op->unfreeze_fs(sb);
306 "VFS:Filesystem thaw failed\n");
307 sb->s_frozen = SB_FREEZE_TRANS;
308 bdev->bd_fsfreeze_count++;
309 mutex_unlock(&bdev->bd_fsfreeze_mutex);
313 sb->s_frozen = SB_UNFROZEN;
315 wake_up(&sb->s_wait_unfrozen);
320 mutex_unlock(&bdev->bd_fsfreeze_mutex);
323 EXPORT_SYMBOL(thaw_bdev);
325 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
327 return block_write_full_page(page, blkdev_get_block, wbc);
330 static int blkdev_readpage(struct file * file, struct page * page)
332 return block_read_full_page(page, blkdev_get_block);
335 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
336 loff_t pos, unsigned len, unsigned flags,
337 struct page **pagep, void **fsdata)
340 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
344 static int blkdev_write_end(struct file *file, struct address_space *mapping,
345 loff_t pos, unsigned len, unsigned copied,
346 struct page *page, void *fsdata)
349 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
352 page_cache_release(page);
359 * for a block special file file->f_path.dentry->d_inode->i_size is zero
360 * so we compute the size by hand (just as in block_read/write above)
362 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
364 struct inode *bd_inode = file->f_mapping->host;
368 mutex_lock(&bd_inode->i_mutex);
369 size = i_size_read(bd_inode);
376 offset += file->f_pos;
379 if (offset >= 0 && offset <= size) {
380 if (offset != file->f_pos) {
381 file->f_pos = offset;
385 mutex_unlock(&bd_inode->i_mutex);
390 * Filp is never NULL; the only case when ->fsync() is called with
391 * NULL first argument is nfsd_sync_dir() and that's not a directory.
394 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
396 return sync_blockdev(I_BDEV(filp->f_mapping->host));
403 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
404 static struct kmem_cache * bdev_cachep __read_mostly;
406 static struct inode *bdev_alloc_inode(struct super_block *sb)
408 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
411 return &ei->vfs_inode;
414 static void bdev_destroy_inode(struct inode *inode)
416 struct bdev_inode *bdi = BDEV_I(inode);
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 INIT_LIST_HEAD(&bdev->bd_inodes);
429 INIT_LIST_HEAD(&bdev->bd_list);
431 INIT_LIST_HEAD(&bdev->bd_holder_list);
433 inode_init_once(&ei->vfs_inode);
434 /* Initialize mutex for freeze. */
435 mutex_init(&bdev->bd_fsfreeze_mutex);
438 static inline void __bd_forget(struct inode *inode)
440 list_del_init(&inode->i_devices);
441 inode->i_bdev = NULL;
442 inode->i_mapping = &inode->i_data;
445 static void bdev_clear_inode(struct inode *inode)
447 struct block_device *bdev = &BDEV_I(inode)->bdev;
449 spin_lock(&bdev_lock);
450 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
451 __bd_forget(list_entry(p, struct inode, i_devices));
453 list_del_init(&bdev->bd_list);
454 spin_unlock(&bdev_lock);
457 static const struct super_operations bdev_sops = {
458 .statfs = simple_statfs,
459 .alloc_inode = bdev_alloc_inode,
460 .destroy_inode = bdev_destroy_inode,
461 .drop_inode = generic_delete_inode,
462 .clear_inode = bdev_clear_inode,
465 static int bd_get_sb(struct file_system_type *fs_type,
466 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
468 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
471 static struct file_system_type bd_type = {
474 .kill_sb = kill_anon_super,
477 struct super_block *blockdev_superblock __read_mostly;
479 void __init bdev_cache_init(void)
482 struct vfsmount *bd_mnt;
484 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
485 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
486 SLAB_MEM_SPREAD|SLAB_PANIC),
488 err = register_filesystem(&bd_type);
490 panic("Cannot register bdev pseudo-fs");
491 bd_mnt = kern_mount(&bd_type);
493 panic("Cannot create bdev pseudo-fs");
495 * This vfsmount structure is only used to obtain the
496 * blockdev_superblock, so tell kmemleak not to report it.
498 kmemleak_not_leak(bd_mnt);
499 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
503 * Most likely _very_ bad one - but then it's hardly critical for small
504 * /dev and can be fixed when somebody will need really large one.
505 * Keep in mind that it will be fed through icache hash function too.
507 static inline unsigned long hash(dev_t dev)
509 return MAJOR(dev)+MINOR(dev);
512 static int bdev_test(struct inode *inode, void *data)
514 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
517 static int bdev_set(struct inode *inode, void *data)
519 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
523 static LIST_HEAD(all_bdevs);
525 struct block_device *bdget(dev_t dev)
527 struct block_device *bdev;
530 inode = iget5_locked(blockdev_superblock, hash(dev),
531 bdev_test, bdev_set, &dev);
536 bdev = &BDEV_I(inode)->bdev;
538 if (inode->i_state & I_NEW) {
539 bdev->bd_contains = NULL;
540 bdev->bd_inode = inode;
541 bdev->bd_block_size = (1 << inode->i_blkbits);
542 bdev->bd_part_count = 0;
543 bdev->bd_invalidated = 0;
544 inode->i_mode = S_IFBLK;
546 inode->i_bdev = bdev;
547 inode->i_data.a_ops = &def_blk_aops;
548 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
549 inode->i_data.backing_dev_info = &default_backing_dev_info;
550 spin_lock(&bdev_lock);
551 list_add(&bdev->bd_list, &all_bdevs);
552 spin_unlock(&bdev_lock);
553 unlock_new_inode(inode);
558 EXPORT_SYMBOL(bdget);
561 * bdgrab -- Grab a reference to an already referenced block device
562 * @bdev: Block device to grab a reference to.
564 struct block_device *bdgrab(struct block_device *bdev)
566 atomic_inc(&bdev->bd_inode->i_count);
570 long nr_blockdev_pages(void)
572 struct block_device *bdev;
574 spin_lock(&bdev_lock);
575 list_for_each_entry(bdev, &all_bdevs, bd_list) {
576 ret += bdev->bd_inode->i_mapping->nrpages;
578 spin_unlock(&bdev_lock);
582 void bdput(struct block_device *bdev)
584 iput(bdev->bd_inode);
587 EXPORT_SYMBOL(bdput);
589 static struct block_device *bd_acquire(struct inode *inode)
591 struct block_device *bdev;
593 spin_lock(&bdev_lock);
594 bdev = inode->i_bdev;
596 atomic_inc(&bdev->bd_inode->i_count);
597 spin_unlock(&bdev_lock);
600 spin_unlock(&bdev_lock);
602 bdev = bdget(inode->i_rdev);
604 spin_lock(&bdev_lock);
605 if (!inode->i_bdev) {
607 * We take an additional bd_inode->i_count for inode,
608 * and it's released in clear_inode() of inode.
609 * So, we can access it via ->i_mapping always
612 atomic_inc(&bdev->bd_inode->i_count);
613 inode->i_bdev = bdev;
614 inode->i_mapping = bdev->bd_inode->i_mapping;
615 list_add(&inode->i_devices, &bdev->bd_inodes);
617 spin_unlock(&bdev_lock);
622 /* Call when you free inode */
624 void bd_forget(struct inode *inode)
626 struct block_device *bdev = NULL;
628 spin_lock(&bdev_lock);
630 if (!sb_is_blkdev_sb(inode->i_sb))
631 bdev = inode->i_bdev;
634 spin_unlock(&bdev_lock);
637 iput(bdev->bd_inode);
640 int bd_claim(struct block_device *bdev, void *holder)
643 spin_lock(&bdev_lock);
645 /* first decide result */
646 if (bdev->bd_holder == holder)
647 res = 0; /* already a holder */
648 else if (bdev->bd_holder != NULL)
649 res = -EBUSY; /* held by someone else */
650 else if (bdev->bd_contains == bdev)
651 res = 0; /* is a whole device which isn't held */
653 else if (bdev->bd_contains->bd_holder == bd_claim)
654 res = 0; /* is a partition of a device that is being partitioned */
655 else if (bdev->bd_contains->bd_holder != NULL)
656 res = -EBUSY; /* is a partition of a held device */
658 res = 0; /* is a partition of an un-held device */
660 /* now impose change */
662 /* note that for a whole device bd_holders
663 * will be incremented twice, and bd_holder will
664 * be set to bd_claim before being set to holder
666 bdev->bd_contains->bd_holders ++;
667 bdev->bd_contains->bd_holder = bd_claim;
669 bdev->bd_holder = holder;
671 spin_unlock(&bdev_lock);
675 EXPORT_SYMBOL(bd_claim);
677 void bd_release(struct block_device *bdev)
679 spin_lock(&bdev_lock);
680 if (!--bdev->bd_contains->bd_holders)
681 bdev->bd_contains->bd_holder = NULL;
682 if (!--bdev->bd_holders)
683 bdev->bd_holder = NULL;
684 spin_unlock(&bdev_lock);
687 EXPORT_SYMBOL(bd_release);
691 * Functions for bd_claim_by_kobject / bd_release_from_kobject
693 * If a kobject is passed to bd_claim_by_kobject()
694 * and the kobject has a parent directory,
695 * following symlinks are created:
696 * o from the kobject to the claimed bdev
697 * o from "holders" directory of the bdev to the parent of the kobject
698 * bd_release_from_kobject() removes these symlinks.
701 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
702 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
703 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
704 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
707 static int add_symlink(struct kobject *from, struct kobject *to)
711 return sysfs_create_link(from, to, kobject_name(to));
714 static void del_symlink(struct kobject *from, struct kobject *to)
718 sysfs_remove_link(from, kobject_name(to));
722 * 'struct bd_holder' contains pointers to kobjects symlinked by
723 * bd_claim_by_kobject.
724 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
727 struct list_head list; /* chain of holders of the bdev */
728 int count; /* references from the holder */
729 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
730 struct kobject *hdev; /* e.g. "/block/dm-0" */
731 struct kobject *hdir; /* e.g. "/block/sda/holders" */
732 struct kobject *sdev; /* e.g. "/block/sda" */
736 * Get references of related kobjects at once.
737 * Returns 1 on success. 0 on failure.
739 * Should call bd_holder_release_dirs() after successful use.
741 static int bd_holder_grab_dirs(struct block_device *bdev,
742 struct bd_holder *bo)
747 bo->sdir = kobject_get(bo->sdir);
751 bo->hdev = kobject_get(bo->sdir->parent);
755 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
759 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
766 kobject_put(bo->sdev);
768 kobject_put(bo->hdev);
770 kobject_put(bo->sdir);
775 /* Put references of related kobjects at once. */
776 static void bd_holder_release_dirs(struct bd_holder *bo)
778 kobject_put(bo->hdir);
779 kobject_put(bo->sdev);
780 kobject_put(bo->hdev);
781 kobject_put(bo->sdir);
784 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
786 struct bd_holder *bo;
788 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
798 static void free_bd_holder(struct bd_holder *bo)
804 * find_bd_holder - find matching struct bd_holder from the block device
806 * @bdev: struct block device to be searched
807 * @bo: target struct bd_holder
809 * Returns matching entry with @bo in @bdev->bd_holder_list.
810 * If found, increment the reference count and return the pointer.
811 * If not found, returns NULL.
813 static struct bd_holder *find_bd_holder(struct block_device *bdev,
814 struct bd_holder *bo)
816 struct bd_holder *tmp;
818 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
819 if (tmp->sdir == bo->sdir) {
828 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
830 * @bdev: block device to be bd_claimed
831 * @bo: preallocated and initialized by alloc_bd_holder()
833 * Add @bo to @bdev->bd_holder_list, create symlinks.
835 * Returns 0 if symlinks are created.
836 * Returns -ve if something fails.
838 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
845 if (!bd_holder_grab_dirs(bdev, bo))
848 err = add_symlink(bo->sdir, bo->sdev);
852 err = add_symlink(bo->hdir, bo->hdev);
854 del_symlink(bo->sdir, bo->sdev);
858 list_add_tail(&bo->list, &bdev->bd_holder_list);
863 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
865 * @bdev: block device to be bd_claimed
866 * @kobj: holder's kobject
868 * If there is matching entry with @kobj in @bdev->bd_holder_list
869 * and no other bd_claim() from the same kobject,
870 * remove the struct bd_holder from the list, delete symlinks for it.
872 * Returns a pointer to the struct bd_holder when it's removed from the list
873 * and ready to be freed.
874 * Returns NULL if matching claim isn't found or there is other bd_claim()
875 * by the same kobject.
877 static struct bd_holder *del_bd_holder(struct block_device *bdev,
878 struct kobject *kobj)
880 struct bd_holder *bo;
882 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
883 if (bo->sdir == kobj) {
885 BUG_ON(bo->count < 0);
888 del_symlink(bo->sdir, bo->sdev);
889 del_symlink(bo->hdir, bo->hdev);
890 bd_holder_release_dirs(bo);
901 * bd_claim_by_kobject - bd_claim() with additional kobject signature
903 * @bdev: block device to be claimed
904 * @holder: holder's signature
905 * @kobj: holder's kobject
907 * Do bd_claim() and if it succeeds, create sysfs symlinks between
908 * the bdev and the holder's kobject.
909 * Use bd_release_from_kobject() when relesing the claimed bdev.
911 * Returns 0 on success. (same as bd_claim())
912 * Returns errno on failure.
914 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
915 struct kobject *kobj)
918 struct bd_holder *bo, *found;
923 bo = alloc_bd_holder(kobj);
927 mutex_lock(&bdev->bd_mutex);
929 err = bd_claim(bdev, holder);
933 found = find_bd_holder(bdev, bo);
937 err = add_bd_holder(bdev, bo);
943 mutex_unlock(&bdev->bd_mutex);
949 * bd_release_from_kobject - bd_release() with additional kobject signature
951 * @bdev: block device to be released
952 * @kobj: holder's kobject
954 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
956 static void bd_release_from_kobject(struct block_device *bdev,
957 struct kobject *kobj)
962 mutex_lock(&bdev->bd_mutex);
964 free_bd_holder(del_bd_holder(bdev, kobj));
965 mutex_unlock(&bdev->bd_mutex);
969 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
971 * @bdev: block device to be claimed
972 * @holder: holder's signature
973 * @disk: holder's gendisk
975 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
977 int bd_claim_by_disk(struct block_device *bdev, void *holder,
978 struct gendisk *disk)
980 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
982 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
985 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
987 * @bdev: block device to be claimed
988 * @disk: holder's gendisk
990 * Call bd_release_from_kobject() and put @disk->slave_dir.
992 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
994 bd_release_from_kobject(bdev, disk->slave_dir);
995 kobject_put(disk->slave_dir);
997 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1001 * Tries to open block device by device number. Use it ONLY if you
1002 * really do not have anything better - i.e. when you are behind a
1003 * truly sucky interface and all you are given is a device number. _Never_
1004 * to be used for internal purposes. If you ever need it - reconsider
1007 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1009 struct block_device *bdev = bdget(dev);
1012 err = blkdev_get(bdev, mode);
1013 return err ? ERR_PTR(err) : bdev;
1016 EXPORT_SYMBOL(open_by_devnum);
1019 * flush_disk - invalidates all buffer-cache entries on a disk
1021 * @bdev: struct block device to be flushed
1023 * Invalidates all buffer-cache entries on a disk. It should be called
1024 * when a disk has been changed -- either by a media change or online
1027 static void flush_disk(struct block_device *bdev)
1029 if (__invalidate_device(bdev)) {
1030 char name[BDEVNAME_SIZE] = "";
1033 disk_name(bdev->bd_disk, 0, name);
1034 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1035 "resized disk %s\n", name);
1040 if (disk_partitionable(bdev->bd_disk))
1041 bdev->bd_invalidated = 1;
1045 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1046 * @disk: struct gendisk to check
1047 * @bdev: struct bdev to adjust.
1049 * This routine checks to see if the bdev size does not match the disk size
1050 * and adjusts it if it differs.
1052 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1054 loff_t disk_size, bdev_size;
1056 disk_size = (loff_t)get_capacity(disk) << 9;
1057 bdev_size = i_size_read(bdev->bd_inode);
1058 if (disk_size != bdev_size) {
1059 char name[BDEVNAME_SIZE];
1061 disk_name(disk, 0, name);
1063 "%s: detected capacity change from %lld to %lld\n",
1064 name, bdev_size, disk_size);
1065 i_size_write(bdev->bd_inode, disk_size);
1069 EXPORT_SYMBOL(check_disk_size_change);
1072 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1073 * @disk: struct gendisk to be revalidated
1075 * This routine is a wrapper for lower-level driver's revalidate_disk
1076 * call-backs. It is used to do common pre and post operations needed
1077 * for all revalidate_disk operations.
1079 int revalidate_disk(struct gendisk *disk)
1081 struct block_device *bdev;
1084 if (disk->fops->revalidate_disk)
1085 ret = disk->fops->revalidate_disk(disk);
1087 bdev = bdget_disk(disk, 0);
1091 mutex_lock(&bdev->bd_mutex);
1092 check_disk_size_change(disk, bdev);
1093 mutex_unlock(&bdev->bd_mutex);
1097 EXPORT_SYMBOL(revalidate_disk);
1100 * This routine checks whether a removable media has been changed,
1101 * and invalidates all buffer-cache-entries in that case. This
1102 * is a relatively slow routine, so we have to try to minimize using
1103 * it. Thus it is called only upon a 'mount' or 'open'. This
1104 * is the best way of combining speed and utility, I think.
1105 * People changing diskettes in the middle of an operation deserve
1108 int check_disk_change(struct block_device *bdev)
1110 struct gendisk *disk = bdev->bd_disk;
1111 const struct block_device_operations *bdops = disk->fops;
1113 if (!bdops->media_changed)
1115 if (!bdops->media_changed(bdev->bd_disk))
1119 if (bdops->revalidate_disk)
1120 bdops->revalidate_disk(bdev->bd_disk);
1124 EXPORT_SYMBOL(check_disk_change);
1126 void bd_set_size(struct block_device *bdev, loff_t size)
1128 unsigned bsize = bdev_logical_block_size(bdev);
1130 bdev->bd_inode->i_size = size;
1131 while (bsize < PAGE_CACHE_SIZE) {
1136 bdev->bd_block_size = bsize;
1137 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1139 EXPORT_SYMBOL(bd_set_size);
1141 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1146 * mutex_lock(part->bd_mutex)
1147 * mutex_lock_nested(whole->bd_mutex, 1)
1150 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1152 struct gendisk *disk;
1157 if (mode & FMODE_READ)
1159 if (mode & FMODE_WRITE)
1162 * hooks: /n/, see "layering violations".
1164 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1174 disk = get_gendisk(bdev->bd_dev, &partno);
1176 goto out_unlock_kernel;
1178 mutex_lock_nested(&bdev->bd_mutex, for_part);
1179 if (!bdev->bd_openers) {
1180 bdev->bd_disk = disk;
1181 bdev->bd_contains = bdev;
1183 struct backing_dev_info *bdi;
1186 bdev->bd_part = disk_get_part(disk, partno);
1190 if (disk->fops->open) {
1191 ret = disk->fops->open(bdev, mode);
1192 if (ret == -ERESTARTSYS) {
1193 /* Lost a race with 'disk' being
1194 * deleted, try again.
1197 disk_put_part(bdev->bd_part);
1198 bdev->bd_part = NULL;
1199 module_put(disk->fops->owner);
1201 bdev->bd_disk = NULL;
1202 mutex_unlock(&bdev->bd_mutex);
1208 if (!bdev->bd_openers) {
1209 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1210 bdi = blk_get_backing_dev_info(bdev);
1212 bdi = &default_backing_dev_info;
1213 bdev->bd_inode->i_data.backing_dev_info = bdi;
1215 if (bdev->bd_invalidated)
1216 rescan_partitions(disk, bdev);
1218 struct block_device *whole;
1219 whole = bdget_disk(disk, 0);
1224 ret = __blkdev_get(whole, mode, 1);
1227 bdev->bd_contains = whole;
1228 bdev->bd_inode->i_data.backing_dev_info =
1229 whole->bd_inode->i_data.backing_dev_info;
1230 bdev->bd_part = disk_get_part(disk, partno);
1231 if (!(disk->flags & GENHD_FL_UP) ||
1232 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1236 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1240 module_put(disk->fops->owner);
1242 if (bdev->bd_contains == bdev) {
1243 if (bdev->bd_disk->fops->open) {
1244 ret = bdev->bd_disk->fops->open(bdev, mode);
1246 goto out_unlock_bdev;
1248 if (bdev->bd_invalidated)
1249 rescan_partitions(bdev->bd_disk, bdev);
1254 bdev->bd_part_count++;
1255 mutex_unlock(&bdev->bd_mutex);
1260 disk_put_part(bdev->bd_part);
1261 bdev->bd_disk = NULL;
1262 bdev->bd_part = NULL;
1263 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1264 if (bdev != bdev->bd_contains)
1265 __blkdev_put(bdev->bd_contains, mode, 1);
1266 bdev->bd_contains = NULL;
1268 mutex_unlock(&bdev->bd_mutex);
1273 module_put(disk->fops->owner);
1280 int blkdev_get(struct block_device *bdev, fmode_t mode)
1282 return __blkdev_get(bdev, mode, 0);
1284 EXPORT_SYMBOL(blkdev_get);
1286 static int blkdev_open(struct inode * inode, struct file * filp)
1288 struct block_device *bdev;
1292 * Preserve backwards compatibility and allow large file access
1293 * even if userspace doesn't ask for it explicitly. Some mkfs
1294 * binary needs it. We might want to drop this workaround
1295 * during an unstable branch.
1297 filp->f_flags |= O_LARGEFILE;
1299 if (filp->f_flags & O_NDELAY)
1300 filp->f_mode |= FMODE_NDELAY;
1301 if (filp->f_flags & O_EXCL)
1302 filp->f_mode |= FMODE_EXCL;
1303 if ((filp->f_flags & O_ACCMODE) == 3)
1304 filp->f_mode |= FMODE_WRITE_IOCTL;
1306 bdev = bd_acquire(inode);
1310 filp->f_mapping = bdev->bd_inode->i_mapping;
1312 res = blkdev_get(bdev, filp->f_mode);
1316 if (filp->f_mode & FMODE_EXCL) {
1317 res = bd_claim(bdev, filp);
1319 goto out_blkdev_put;
1325 blkdev_put(bdev, filp->f_mode);
1329 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1332 struct gendisk *disk = bdev->bd_disk;
1333 struct block_device *victim = NULL;
1335 mutex_lock_nested(&bdev->bd_mutex, for_part);
1338 bdev->bd_part_count--;
1340 if (!--bdev->bd_openers) {
1341 sync_blockdev(bdev);
1344 if (bdev->bd_contains == bdev) {
1345 if (disk->fops->release)
1346 ret = disk->fops->release(disk, mode);
1348 if (!bdev->bd_openers) {
1349 struct module *owner = disk->fops->owner;
1353 disk_put_part(bdev->bd_part);
1354 bdev->bd_part = NULL;
1355 bdev->bd_disk = NULL;
1356 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1357 if (bdev != bdev->bd_contains)
1358 victim = bdev->bd_contains;
1359 bdev->bd_contains = NULL;
1362 mutex_unlock(&bdev->bd_mutex);
1365 __blkdev_put(victim, mode, 1);
1369 int blkdev_put(struct block_device *bdev, fmode_t mode)
1371 return __blkdev_put(bdev, mode, 0);
1373 EXPORT_SYMBOL(blkdev_put);
1375 static int blkdev_close(struct inode * inode, struct file * filp)
1377 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1378 if (bdev->bd_holder == filp)
1380 return blkdev_put(bdev, filp->f_mode);
1383 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1385 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1386 fmode_t mode = file->f_mode;
1389 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1390 * to updated it before every ioctl.
1392 if (file->f_flags & O_NDELAY)
1393 mode |= FMODE_NDELAY;
1395 mode &= ~FMODE_NDELAY;
1397 return blkdev_ioctl(bdev, mode, cmd, arg);
1401 * Write data to the block device. Only intended for the block device itself
1402 * and the raw driver which basically is a fake block device.
1404 * Does not take i_mutex for the write and thus is not for general purpose
1407 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1408 unsigned long nr_segs, loff_t pos)
1410 struct file *file = iocb->ki_filp;
1413 BUG_ON(iocb->ki_pos != pos);
1415 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1416 if (ret > 0 || ret == -EIOCBQUEUED) {
1419 err = generic_write_sync(file, pos, ret);
1420 if (err < 0 && ret > 0)
1425 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1428 * Try to release a page associated with block device when the system
1429 * is under memory pressure.
1431 static int blkdev_releasepage(struct page *page, gfp_t wait)
1433 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1435 if (super && super->s_op->bdev_try_to_free_page)
1436 return super->s_op->bdev_try_to_free_page(super, page, wait);
1438 return try_to_free_buffers(page);
1441 static const struct address_space_operations def_blk_aops = {
1442 .readpage = blkdev_readpage,
1443 .writepage = blkdev_writepage,
1444 .sync_page = block_sync_page,
1445 .write_begin = blkdev_write_begin,
1446 .write_end = blkdev_write_end,
1447 .writepages = generic_writepages,
1448 .releasepage = blkdev_releasepage,
1449 .direct_IO = blkdev_direct_IO,
1452 const struct file_operations def_blk_fops = {
1453 .open = blkdev_open,
1454 .release = blkdev_close,
1455 .llseek = block_llseek,
1456 .read = do_sync_read,
1457 .write = do_sync_write,
1458 .aio_read = generic_file_aio_read,
1459 .aio_write = blkdev_aio_write,
1460 .mmap = generic_file_mmap,
1461 .fsync = block_fsync,
1462 .unlocked_ioctl = block_ioctl,
1463 #ifdef CONFIG_COMPAT
1464 .compat_ioctl = compat_blkdev_ioctl,
1466 .splice_read = generic_file_splice_read,
1467 .splice_write = generic_file_splice_write,
1470 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1473 mm_segment_t old_fs = get_fs();
1475 res = blkdev_ioctl(bdev, 0, cmd, arg);
1480 EXPORT_SYMBOL(ioctl_by_bdev);
1483 * lookup_bdev - lookup a struct block_device by name
1484 * @pathname: special file representing the block device
1486 * Get a reference to the blockdevice at @pathname in the current
1487 * namespace if possible and return it. Return ERR_PTR(error)
1490 struct block_device *lookup_bdev(const char *pathname)
1492 struct block_device *bdev;
1493 struct inode *inode;
1497 if (!pathname || !*pathname)
1498 return ERR_PTR(-EINVAL);
1500 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1502 return ERR_PTR(error);
1504 inode = path.dentry->d_inode;
1506 if (!S_ISBLK(inode->i_mode))
1509 if (path.mnt->mnt_flags & MNT_NODEV)
1512 bdev = bd_acquire(inode);
1519 bdev = ERR_PTR(error);
1522 EXPORT_SYMBOL(lookup_bdev);
1525 * open_bdev_exclusive - open a block device by name and set it up for use
1527 * @path: special file representing the block device
1528 * @mode: FMODE_... combination to pass be used
1529 * @holder: owner for exclusion
1531 * Open the blockdevice described by the special file at @path, claim it
1534 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1536 struct block_device *bdev;
1539 bdev = lookup_bdev(path);
1543 error = blkdev_get(bdev, mode);
1545 return ERR_PTR(error);
1547 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1549 error = bd_claim(bdev, holder);
1556 blkdev_put(bdev, mode);
1557 return ERR_PTR(error);
1560 EXPORT_SYMBOL(open_bdev_exclusive);
1563 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1565 * @bdev: blockdevice to close
1566 * @mode: mode, must match that used to open.
1568 * This is the counterpart to open_bdev_exclusive().
1570 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1573 blkdev_put(bdev, mode);
1576 EXPORT_SYMBOL(close_bdev_exclusive);
1578 int __invalidate_device(struct block_device *bdev)
1580 struct super_block *sb = get_super(bdev);
1585 * no need to lock the super, get_super holds the
1586 * read mutex so the filesystem cannot go away
1587 * under us (->put_super runs with the write lock
1590 shrink_dcache_sb(sb);
1591 res = invalidate_inodes(sb);
1594 invalidate_bdev(bdev);
1597 EXPORT_SYMBOL(__invalidate_device);