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 * This takes the block device bd_mount_sem to make sure no new mounts
220 * happen on bdev until thaw_bdev() is called.
221 * If a superblock is found on this device, we take the s_umount semaphore
222 * on it to make sure nobody unmounts until the snapshot creation is done.
223 * The reference counter (bd_fsfreeze_count) guarantees that only the last
224 * unfreeze process can unfreeze the frozen filesystem actually when multiple
225 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
226 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
229 struct super_block *freeze_bdev(struct block_device *bdev)
231 struct super_block *sb;
234 mutex_lock(&bdev->bd_fsfreeze_mutex);
235 if (bdev->bd_fsfreeze_count > 0) {
236 bdev->bd_fsfreeze_count++;
237 sb = get_super(bdev);
238 mutex_unlock(&bdev->bd_fsfreeze_mutex);
241 bdev->bd_fsfreeze_count++;
243 down(&bdev->bd_mount_sem);
244 sb = get_super(bdev);
245 if (sb && !(sb->s_flags & MS_RDONLY)) {
246 sb->s_frozen = SB_FREEZE_WRITE;
251 sb->s_frozen = SB_FREEZE_TRANS;
254 sync_blockdev(sb->s_bdev);
256 if (sb->s_op->freeze_fs) {
257 error = sb->s_op->freeze_fs(sb);
260 "VFS:Filesystem freeze failed\n");
261 sb->s_frozen = SB_UNFROZEN;
263 up(&bdev->bd_mount_sem);
264 bdev->bd_fsfreeze_count--;
265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
266 return ERR_PTR(error);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
274 return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */
276 EXPORT_SYMBOL(freeze_bdev);
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
285 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
289 mutex_lock(&bdev->bd_fsfreeze_mutex);
290 if (!bdev->bd_fsfreeze_count) {
291 mutex_unlock(&bdev->bd_fsfreeze_mutex);
295 bdev->bd_fsfreeze_count--;
296 if (bdev->bd_fsfreeze_count > 0) {
299 mutex_unlock(&bdev->bd_fsfreeze_mutex);
304 BUG_ON(sb->s_bdev != bdev);
305 if (!(sb->s_flags & MS_RDONLY)) {
306 if (sb->s_op->unfreeze_fs) {
307 error = sb->s_op->unfreeze_fs(sb);
310 "VFS:Filesystem thaw failed\n");
311 sb->s_frozen = SB_FREEZE_TRANS;
312 bdev->bd_fsfreeze_count++;
313 mutex_unlock(&bdev->bd_fsfreeze_mutex);
317 sb->s_frozen = SB_UNFROZEN;
319 wake_up(&sb->s_wait_unfrozen);
324 up(&bdev->bd_mount_sem);
325 mutex_unlock(&bdev->bd_fsfreeze_mutex);
328 EXPORT_SYMBOL(thaw_bdev);
330 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
332 return block_write_full_page(page, blkdev_get_block, wbc);
335 static int blkdev_readpage(struct file * file, struct page * page)
337 return block_read_full_page(page, blkdev_get_block);
340 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
341 loff_t pos, unsigned len, unsigned flags,
342 struct page **pagep, void **fsdata)
345 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
349 static int blkdev_write_end(struct file *file, struct address_space *mapping,
350 loff_t pos, unsigned len, unsigned copied,
351 struct page *page, void *fsdata)
354 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
357 page_cache_release(page);
364 * for a block special file file->f_path.dentry->d_inode->i_size is zero
365 * so we compute the size by hand (just as in block_read/write above)
367 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
369 struct inode *bd_inode = file->f_mapping->host;
373 mutex_lock(&bd_inode->i_mutex);
374 size = i_size_read(bd_inode);
381 offset += file->f_pos;
384 if (offset >= 0 && offset <= size) {
385 if (offset != file->f_pos) {
386 file->f_pos = offset;
390 mutex_unlock(&bd_inode->i_mutex);
395 * Filp is never NULL; the only case when ->fsync() is called with
396 * NULL first argument is nfsd_sync_dir() and that's not a directory.
399 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
401 return sync_blockdev(I_BDEV(filp->f_mapping->host));
408 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
409 static struct kmem_cache * bdev_cachep __read_mostly;
411 static struct inode *bdev_alloc_inode(struct super_block *sb)
413 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
416 return &ei->vfs_inode;
419 static void bdev_destroy_inode(struct inode *inode)
421 struct bdev_inode *bdi = BDEV_I(inode);
423 bdi->bdev.bd_inode_backing_dev_info = NULL;
424 kmem_cache_free(bdev_cachep, bdi);
427 static void init_once(void *foo)
429 struct bdev_inode *ei = (struct bdev_inode *) foo;
430 struct block_device *bdev = &ei->bdev;
432 memset(bdev, 0, sizeof(*bdev));
433 mutex_init(&bdev->bd_mutex);
434 sema_init(&bdev->bd_mount_sem, 1);
435 INIT_LIST_HEAD(&bdev->bd_inodes);
436 INIT_LIST_HEAD(&bdev->bd_list);
438 INIT_LIST_HEAD(&bdev->bd_holder_list);
440 inode_init_once(&ei->vfs_inode);
441 /* Initialize mutex for freeze. */
442 mutex_init(&bdev->bd_fsfreeze_mutex);
445 static inline void __bd_forget(struct inode *inode)
447 list_del_init(&inode->i_devices);
448 inode->i_bdev = NULL;
449 inode->i_mapping = &inode->i_data;
452 static void bdev_clear_inode(struct inode *inode)
454 struct block_device *bdev = &BDEV_I(inode)->bdev;
456 spin_lock(&bdev_lock);
457 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
458 __bd_forget(list_entry(p, struct inode, i_devices));
460 list_del_init(&bdev->bd_list);
461 spin_unlock(&bdev_lock);
464 static const struct super_operations bdev_sops = {
465 .statfs = simple_statfs,
466 .alloc_inode = bdev_alloc_inode,
467 .destroy_inode = bdev_destroy_inode,
468 .drop_inode = generic_delete_inode,
469 .clear_inode = bdev_clear_inode,
472 static int bd_get_sb(struct file_system_type *fs_type,
473 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
475 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
478 static struct file_system_type bd_type = {
481 .kill_sb = kill_anon_super,
484 struct super_block *blockdev_superblock __read_mostly;
486 void __init bdev_cache_init(void)
489 struct vfsmount *bd_mnt;
491 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
492 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
493 SLAB_MEM_SPREAD|SLAB_PANIC),
495 err = register_filesystem(&bd_type);
497 panic("Cannot register bdev pseudo-fs");
498 bd_mnt = kern_mount(&bd_type);
500 panic("Cannot create bdev pseudo-fs");
502 * This vfsmount structure is only used to obtain the
503 * blockdev_superblock, so tell kmemleak not to report it.
505 kmemleak_not_leak(bd_mnt);
506 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
510 * Most likely _very_ bad one - but then it's hardly critical for small
511 * /dev and can be fixed when somebody will need really large one.
512 * Keep in mind that it will be fed through icache hash function too.
514 static inline unsigned long hash(dev_t dev)
516 return MAJOR(dev)+MINOR(dev);
519 static int bdev_test(struct inode *inode, void *data)
521 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
524 static int bdev_set(struct inode *inode, void *data)
526 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
530 static LIST_HEAD(all_bdevs);
532 struct block_device *bdget(dev_t dev)
534 struct block_device *bdev;
537 inode = iget5_locked(blockdev_superblock, hash(dev),
538 bdev_test, bdev_set, &dev);
543 bdev = &BDEV_I(inode)->bdev;
545 if (inode->i_state & I_NEW) {
546 bdev->bd_contains = NULL;
547 bdev->bd_inode = inode;
548 bdev->bd_block_size = (1 << inode->i_blkbits);
549 bdev->bd_part_count = 0;
550 bdev->bd_invalidated = 0;
551 inode->i_mode = S_IFBLK;
553 inode->i_bdev = bdev;
554 inode->i_data.a_ops = &def_blk_aops;
555 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
556 inode->i_data.backing_dev_info = &default_backing_dev_info;
557 spin_lock(&bdev_lock);
558 list_add(&bdev->bd_list, &all_bdevs);
559 spin_unlock(&bdev_lock);
560 unlock_new_inode(inode);
565 EXPORT_SYMBOL(bdget);
567 long nr_blockdev_pages(void)
569 struct block_device *bdev;
571 spin_lock(&bdev_lock);
572 list_for_each_entry(bdev, &all_bdevs, bd_list) {
573 ret += bdev->bd_inode->i_mapping->nrpages;
575 spin_unlock(&bdev_lock);
579 void bdput(struct block_device *bdev)
581 iput(bdev->bd_inode);
584 EXPORT_SYMBOL(bdput);
586 static struct block_device *bd_acquire(struct inode *inode)
588 struct block_device *bdev;
590 spin_lock(&bdev_lock);
591 bdev = inode->i_bdev;
593 atomic_inc(&bdev->bd_inode->i_count);
594 spin_unlock(&bdev_lock);
597 spin_unlock(&bdev_lock);
599 bdev = bdget(inode->i_rdev);
601 spin_lock(&bdev_lock);
602 if (!inode->i_bdev) {
604 * We take an additional bd_inode->i_count for inode,
605 * and it's released in clear_inode() of inode.
606 * So, we can access it via ->i_mapping always
609 atomic_inc(&bdev->bd_inode->i_count);
610 inode->i_bdev = bdev;
611 inode->i_mapping = bdev->bd_inode->i_mapping;
612 list_add(&inode->i_devices, &bdev->bd_inodes);
614 spin_unlock(&bdev_lock);
619 /* Call when you free inode */
621 void bd_forget(struct inode *inode)
623 struct block_device *bdev = NULL;
625 spin_lock(&bdev_lock);
627 if (!sb_is_blkdev_sb(inode->i_sb))
628 bdev = inode->i_bdev;
631 spin_unlock(&bdev_lock);
634 iput(bdev->bd_inode);
637 int bd_claim(struct block_device *bdev, void *holder)
640 spin_lock(&bdev_lock);
642 /* first decide result */
643 if (bdev->bd_holder == holder)
644 res = 0; /* already a holder */
645 else if (bdev->bd_holder != NULL)
646 res = -EBUSY; /* held by someone else */
647 else if (bdev->bd_contains == bdev)
648 res = 0; /* is a whole device which isn't held */
650 else if (bdev->bd_contains->bd_holder == bd_claim)
651 res = 0; /* is a partition of a device that is being partitioned */
652 else if (bdev->bd_contains->bd_holder != NULL)
653 res = -EBUSY; /* is a partition of a held device */
655 res = 0; /* is a partition of an un-held device */
657 /* now impose change */
659 /* note that for a whole device bd_holders
660 * will be incremented twice, and bd_holder will
661 * be set to bd_claim before being set to holder
663 bdev->bd_contains->bd_holders ++;
664 bdev->bd_contains->bd_holder = bd_claim;
666 bdev->bd_holder = holder;
668 spin_unlock(&bdev_lock);
672 EXPORT_SYMBOL(bd_claim);
674 void bd_release(struct block_device *bdev)
676 spin_lock(&bdev_lock);
677 if (!--bdev->bd_contains->bd_holders)
678 bdev->bd_contains->bd_holder = NULL;
679 if (!--bdev->bd_holders)
680 bdev->bd_holder = NULL;
681 spin_unlock(&bdev_lock);
684 EXPORT_SYMBOL(bd_release);
688 * Functions for bd_claim_by_kobject / bd_release_from_kobject
690 * If a kobject is passed to bd_claim_by_kobject()
691 * and the kobject has a parent directory,
692 * following symlinks are created:
693 * o from the kobject to the claimed bdev
694 * o from "holders" directory of the bdev to the parent of the kobject
695 * bd_release_from_kobject() removes these symlinks.
698 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
699 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
700 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
701 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
704 static int add_symlink(struct kobject *from, struct kobject *to)
708 return sysfs_create_link(from, to, kobject_name(to));
711 static void del_symlink(struct kobject *from, struct kobject *to)
715 sysfs_remove_link(from, kobject_name(to));
719 * 'struct bd_holder' contains pointers to kobjects symlinked by
720 * bd_claim_by_kobject.
721 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
724 struct list_head list; /* chain of holders of the bdev */
725 int count; /* references from the holder */
726 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
727 struct kobject *hdev; /* e.g. "/block/dm-0" */
728 struct kobject *hdir; /* e.g. "/block/sda/holders" */
729 struct kobject *sdev; /* e.g. "/block/sda" */
733 * Get references of related kobjects at once.
734 * Returns 1 on success. 0 on failure.
736 * Should call bd_holder_release_dirs() after successful use.
738 static int bd_holder_grab_dirs(struct block_device *bdev,
739 struct bd_holder *bo)
744 bo->sdir = kobject_get(bo->sdir);
748 bo->hdev = kobject_get(bo->sdir->parent);
752 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
756 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
763 kobject_put(bo->sdev);
765 kobject_put(bo->hdev);
767 kobject_put(bo->sdir);
772 /* Put references of related kobjects at once. */
773 static void bd_holder_release_dirs(struct bd_holder *bo)
775 kobject_put(bo->hdir);
776 kobject_put(bo->sdev);
777 kobject_put(bo->hdev);
778 kobject_put(bo->sdir);
781 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
783 struct bd_holder *bo;
785 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
795 static void free_bd_holder(struct bd_holder *bo)
801 * find_bd_holder - find matching struct bd_holder from the block device
803 * @bdev: struct block device to be searched
804 * @bo: target struct bd_holder
806 * Returns matching entry with @bo in @bdev->bd_holder_list.
807 * If found, increment the reference count and return the pointer.
808 * If not found, returns NULL.
810 static struct bd_holder *find_bd_holder(struct block_device *bdev,
811 struct bd_holder *bo)
813 struct bd_holder *tmp;
815 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
816 if (tmp->sdir == bo->sdir) {
825 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
827 * @bdev: block device to be bd_claimed
828 * @bo: preallocated and initialized by alloc_bd_holder()
830 * Add @bo to @bdev->bd_holder_list, create symlinks.
832 * Returns 0 if symlinks are created.
833 * Returns -ve if something fails.
835 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
842 if (!bd_holder_grab_dirs(bdev, bo))
845 err = add_symlink(bo->sdir, bo->sdev);
849 err = add_symlink(bo->hdir, bo->hdev);
851 del_symlink(bo->sdir, bo->sdev);
855 list_add_tail(&bo->list, &bdev->bd_holder_list);
860 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
862 * @bdev: block device to be bd_claimed
863 * @kobj: holder's kobject
865 * If there is matching entry with @kobj in @bdev->bd_holder_list
866 * and no other bd_claim() from the same kobject,
867 * remove the struct bd_holder from the list, delete symlinks for it.
869 * Returns a pointer to the struct bd_holder when it's removed from the list
870 * and ready to be freed.
871 * Returns NULL if matching claim isn't found or there is other bd_claim()
872 * by the same kobject.
874 static struct bd_holder *del_bd_holder(struct block_device *bdev,
875 struct kobject *kobj)
877 struct bd_holder *bo;
879 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
880 if (bo->sdir == kobj) {
882 BUG_ON(bo->count < 0);
885 del_symlink(bo->sdir, bo->sdev);
886 del_symlink(bo->hdir, bo->hdev);
887 bd_holder_release_dirs(bo);
898 * bd_claim_by_kobject - bd_claim() with additional kobject signature
900 * @bdev: block device to be claimed
901 * @holder: holder's signature
902 * @kobj: holder's kobject
904 * Do bd_claim() and if it succeeds, create sysfs symlinks between
905 * the bdev and the holder's kobject.
906 * Use bd_release_from_kobject() when relesing the claimed bdev.
908 * Returns 0 on success. (same as bd_claim())
909 * Returns errno on failure.
911 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
912 struct kobject *kobj)
915 struct bd_holder *bo, *found;
920 bo = alloc_bd_holder(kobj);
924 mutex_lock(&bdev->bd_mutex);
926 err = bd_claim(bdev, holder);
930 found = find_bd_holder(bdev, bo);
934 err = add_bd_holder(bdev, bo);
940 mutex_unlock(&bdev->bd_mutex);
946 * bd_release_from_kobject - bd_release() with additional kobject signature
948 * @bdev: block device to be released
949 * @kobj: holder's kobject
951 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
953 static void bd_release_from_kobject(struct block_device *bdev,
954 struct kobject *kobj)
959 mutex_lock(&bdev->bd_mutex);
961 free_bd_holder(del_bd_holder(bdev, kobj));
962 mutex_unlock(&bdev->bd_mutex);
966 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
968 * @bdev: block device to be claimed
969 * @holder: holder's signature
970 * @disk: holder's gendisk
972 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
974 int bd_claim_by_disk(struct block_device *bdev, void *holder,
975 struct gendisk *disk)
977 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
979 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
982 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
984 * @bdev: block device to be claimed
985 * @disk: holder's gendisk
987 * Call bd_release_from_kobject() and put @disk->slave_dir.
989 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
991 bd_release_from_kobject(bdev, disk->slave_dir);
992 kobject_put(disk->slave_dir);
994 EXPORT_SYMBOL_GPL(bd_release_from_disk);
998 * Tries to open block device by device number. Use it ONLY if you
999 * really do not have anything better - i.e. when you are behind a
1000 * truly sucky interface and all you are given is a device number. _Never_
1001 * to be used for internal purposes. If you ever need it - reconsider
1004 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1006 struct block_device *bdev = bdget(dev);
1009 err = blkdev_get(bdev, mode);
1010 return err ? ERR_PTR(err) : bdev;
1013 EXPORT_SYMBOL(open_by_devnum);
1016 * flush_disk - invalidates all buffer-cache entries on a disk
1018 * @bdev: struct block device to be flushed
1020 * Invalidates all buffer-cache entries on a disk. It should be called
1021 * when a disk has been changed -- either by a media change or online
1024 static void flush_disk(struct block_device *bdev)
1026 if (__invalidate_device(bdev)) {
1027 char name[BDEVNAME_SIZE] = "";
1030 disk_name(bdev->bd_disk, 0, name);
1031 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1032 "resized disk %s\n", name);
1037 if (disk_partitionable(bdev->bd_disk))
1038 bdev->bd_invalidated = 1;
1042 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1043 * @disk: struct gendisk to check
1044 * @bdev: struct bdev to adjust.
1046 * This routine checks to see if the bdev size does not match the disk size
1047 * and adjusts it if it differs.
1049 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1051 loff_t disk_size, bdev_size;
1053 disk_size = (loff_t)get_capacity(disk) << 9;
1054 bdev_size = i_size_read(bdev->bd_inode);
1055 if (disk_size != bdev_size) {
1056 char name[BDEVNAME_SIZE];
1058 disk_name(disk, 0, name);
1060 "%s: detected capacity change from %lld to %lld\n",
1061 name, bdev_size, disk_size);
1062 i_size_write(bdev->bd_inode, disk_size);
1066 EXPORT_SYMBOL(check_disk_size_change);
1069 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1070 * @disk: struct gendisk to be revalidated
1072 * This routine is a wrapper for lower-level driver's revalidate_disk
1073 * call-backs. It is used to do common pre and post operations needed
1074 * for all revalidate_disk operations.
1076 int revalidate_disk(struct gendisk *disk)
1078 struct block_device *bdev;
1081 if (disk->fops->revalidate_disk)
1082 ret = disk->fops->revalidate_disk(disk);
1084 bdev = bdget_disk(disk, 0);
1088 mutex_lock(&bdev->bd_mutex);
1089 check_disk_size_change(disk, bdev);
1090 mutex_unlock(&bdev->bd_mutex);
1094 EXPORT_SYMBOL(revalidate_disk);
1097 * This routine checks whether a removable media has been changed,
1098 * and invalidates all buffer-cache-entries in that case. This
1099 * is a relatively slow routine, so we have to try to minimize using
1100 * it. Thus it is called only upon a 'mount' or 'open'. This
1101 * is the best way of combining speed and utility, I think.
1102 * People changing diskettes in the middle of an operation deserve
1105 int check_disk_change(struct block_device *bdev)
1107 struct gendisk *disk = bdev->bd_disk;
1108 struct block_device_operations * bdops = disk->fops;
1110 if (!bdops->media_changed)
1112 if (!bdops->media_changed(bdev->bd_disk))
1116 if (bdops->revalidate_disk)
1117 bdops->revalidate_disk(bdev->bd_disk);
1121 EXPORT_SYMBOL(check_disk_change);
1123 void bd_set_size(struct block_device *bdev, loff_t size)
1125 unsigned bsize = bdev_logical_block_size(bdev);
1127 bdev->bd_inode->i_size = size;
1128 while (bsize < PAGE_CACHE_SIZE) {
1133 bdev->bd_block_size = bsize;
1134 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1136 EXPORT_SYMBOL(bd_set_size);
1138 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1143 * mutex_lock(part->bd_mutex)
1144 * mutex_lock_nested(whole->bd_mutex, 1)
1147 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1149 struct gendisk *disk;
1154 if (mode & FMODE_READ)
1156 if (mode & FMODE_WRITE)
1159 * hooks: /n/, see "layering violations".
1161 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1171 disk = get_gendisk(bdev->bd_dev, &partno);
1173 goto out_unlock_kernel;
1175 mutex_lock_nested(&bdev->bd_mutex, for_part);
1176 if (!bdev->bd_openers) {
1177 bdev->bd_disk = disk;
1178 bdev->bd_contains = bdev;
1180 struct backing_dev_info *bdi;
1183 bdev->bd_part = disk_get_part(disk, partno);
1187 if (disk->fops->open) {
1188 ret = disk->fops->open(bdev, mode);
1189 if (ret == -ERESTARTSYS) {
1190 /* Lost a race with 'disk' being
1191 * deleted, try again.
1194 disk_put_part(bdev->bd_part);
1195 bdev->bd_part = NULL;
1196 module_put(disk->fops->owner);
1198 bdev->bd_disk = NULL;
1199 mutex_unlock(&bdev->bd_mutex);
1205 if (!bdev->bd_openers) {
1206 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1207 bdi = blk_get_backing_dev_info(bdev);
1209 bdi = &default_backing_dev_info;
1210 bdev->bd_inode->i_data.backing_dev_info = bdi;
1212 if (bdev->bd_invalidated)
1213 rescan_partitions(disk, bdev);
1215 struct block_device *whole;
1216 whole = bdget_disk(disk, 0);
1221 ret = __blkdev_get(whole, mode, 1);
1224 bdev->bd_contains = whole;
1225 bdev->bd_inode->i_data.backing_dev_info =
1226 whole->bd_inode->i_data.backing_dev_info;
1227 bdev->bd_part = disk_get_part(disk, partno);
1228 if (!(disk->flags & GENHD_FL_UP) ||
1229 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1233 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1237 module_put(disk->fops->owner);
1239 if (bdev->bd_contains == bdev) {
1240 if (bdev->bd_disk->fops->open) {
1241 ret = bdev->bd_disk->fops->open(bdev, mode);
1243 goto out_unlock_bdev;
1245 if (bdev->bd_invalidated)
1246 rescan_partitions(bdev->bd_disk, bdev);
1251 bdev->bd_part_count++;
1252 mutex_unlock(&bdev->bd_mutex);
1257 disk_put_part(bdev->bd_part);
1258 bdev->bd_disk = NULL;
1259 bdev->bd_part = NULL;
1260 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1261 if (bdev != bdev->bd_contains)
1262 __blkdev_put(bdev->bd_contains, mode, 1);
1263 bdev->bd_contains = NULL;
1265 mutex_unlock(&bdev->bd_mutex);
1270 module_put(disk->fops->owner);
1277 int blkdev_get(struct block_device *bdev, fmode_t mode)
1279 return __blkdev_get(bdev, mode, 0);
1281 EXPORT_SYMBOL(blkdev_get);
1283 static int blkdev_open(struct inode * inode, struct file * filp)
1285 struct block_device *bdev;
1289 * Preserve backwards compatibility and allow large file access
1290 * even if userspace doesn't ask for it explicitly. Some mkfs
1291 * binary needs it. We might want to drop this workaround
1292 * during an unstable branch.
1294 filp->f_flags |= O_LARGEFILE;
1296 if (filp->f_flags & O_NDELAY)
1297 filp->f_mode |= FMODE_NDELAY;
1298 if (filp->f_flags & O_EXCL)
1299 filp->f_mode |= FMODE_EXCL;
1300 if ((filp->f_flags & O_ACCMODE) == 3)
1301 filp->f_mode |= FMODE_WRITE_IOCTL;
1303 bdev = bd_acquire(inode);
1307 filp->f_mapping = bdev->bd_inode->i_mapping;
1309 res = blkdev_get(bdev, filp->f_mode);
1313 if (filp->f_mode & FMODE_EXCL) {
1314 res = bd_claim(bdev, filp);
1316 goto out_blkdev_put;
1322 blkdev_put(bdev, filp->f_mode);
1326 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1329 struct gendisk *disk = bdev->bd_disk;
1330 struct block_device *victim = NULL;
1332 mutex_lock_nested(&bdev->bd_mutex, for_part);
1335 bdev->bd_part_count--;
1337 if (!--bdev->bd_openers) {
1338 sync_blockdev(bdev);
1341 if (bdev->bd_contains == bdev) {
1342 if (disk->fops->release)
1343 ret = disk->fops->release(disk, mode);
1345 if (!bdev->bd_openers) {
1346 struct module *owner = disk->fops->owner;
1350 disk_put_part(bdev->bd_part);
1351 bdev->bd_part = NULL;
1352 bdev->bd_disk = NULL;
1353 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1354 if (bdev != bdev->bd_contains)
1355 victim = bdev->bd_contains;
1356 bdev->bd_contains = NULL;
1359 mutex_unlock(&bdev->bd_mutex);
1362 __blkdev_put(victim, mode, 1);
1366 int blkdev_put(struct block_device *bdev, fmode_t mode)
1368 return __blkdev_put(bdev, mode, 0);
1370 EXPORT_SYMBOL(blkdev_put);
1372 static int blkdev_close(struct inode * inode, struct file * filp)
1374 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1375 if (bdev->bd_holder == filp)
1377 return blkdev_put(bdev, filp->f_mode);
1380 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1382 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1383 fmode_t mode = file->f_mode;
1386 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1387 * to updated it before every ioctl.
1389 if (file->f_flags & O_NDELAY)
1390 mode |= FMODE_NDELAY;
1392 mode &= ~FMODE_NDELAY;
1394 return blkdev_ioctl(bdev, mode, cmd, arg);
1398 * Try to release a page associated with block device when the system
1399 * is under memory pressure.
1401 static int blkdev_releasepage(struct page *page, gfp_t wait)
1403 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1405 if (super && super->s_op->bdev_try_to_free_page)
1406 return super->s_op->bdev_try_to_free_page(super, page, wait);
1408 return try_to_free_buffers(page);
1411 static const struct address_space_operations def_blk_aops = {
1412 .readpage = blkdev_readpage,
1413 .writepage = blkdev_writepage,
1414 .sync_page = block_sync_page,
1415 .write_begin = blkdev_write_begin,
1416 .write_end = blkdev_write_end,
1417 .writepages = generic_writepages,
1418 .releasepage = blkdev_releasepage,
1419 .direct_IO = blkdev_direct_IO,
1422 const struct file_operations def_blk_fops = {
1423 .open = blkdev_open,
1424 .release = blkdev_close,
1425 .llseek = block_llseek,
1426 .read = do_sync_read,
1427 .write = do_sync_write,
1428 .aio_read = generic_file_aio_read,
1429 .aio_write = generic_file_aio_write_nolock,
1430 .mmap = generic_file_mmap,
1431 .fsync = block_fsync,
1432 .unlocked_ioctl = block_ioctl,
1433 #ifdef CONFIG_COMPAT
1434 .compat_ioctl = compat_blkdev_ioctl,
1436 .splice_read = generic_file_splice_read,
1437 .splice_write = generic_file_splice_write,
1440 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1443 mm_segment_t old_fs = get_fs();
1445 res = blkdev_ioctl(bdev, 0, cmd, arg);
1450 EXPORT_SYMBOL(ioctl_by_bdev);
1453 * lookup_bdev - lookup a struct block_device by name
1454 * @pathname: special file representing the block device
1456 * Get a reference to the blockdevice at @pathname in the current
1457 * namespace if possible and return it. Return ERR_PTR(error)
1460 struct block_device *lookup_bdev(const char *pathname)
1462 struct block_device *bdev;
1463 struct inode *inode;
1467 if (!pathname || !*pathname)
1468 return ERR_PTR(-EINVAL);
1470 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1472 return ERR_PTR(error);
1474 inode = path.dentry->d_inode;
1476 if (!S_ISBLK(inode->i_mode))
1479 if (path.mnt->mnt_flags & MNT_NODEV)
1482 bdev = bd_acquire(inode);
1489 bdev = ERR_PTR(error);
1492 EXPORT_SYMBOL(lookup_bdev);
1495 * open_bdev_exclusive - open a block device by name and set it up for use
1497 * @path: special file representing the block device
1498 * @mode: FMODE_... combination to pass be used
1499 * @holder: owner for exclusion
1501 * Open the blockdevice described by the special file at @path, claim it
1504 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1506 struct block_device *bdev;
1509 bdev = lookup_bdev(path);
1513 error = blkdev_get(bdev, mode);
1515 return ERR_PTR(error);
1517 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1519 error = bd_claim(bdev, holder);
1526 blkdev_put(bdev, mode);
1527 return ERR_PTR(error);
1530 EXPORT_SYMBOL(open_bdev_exclusive);
1533 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1535 * @bdev: blockdevice to close
1536 * @mode: mode, must match that used to open.
1538 * This is the counterpart to open_bdev_exclusive().
1540 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1543 blkdev_put(bdev, mode);
1546 EXPORT_SYMBOL(close_bdev_exclusive);
1548 int __invalidate_device(struct block_device *bdev)
1550 struct super_block *sb = get_super(bdev);
1555 * no need to lock the super, get_super holds the
1556 * read mutex so the filesystem cannot go away
1557 * under us (->put_super runs with the write lock
1560 shrink_dcache_sb(sb);
1561 res = invalidate_inodes(sb);
1564 invalidate_bdev(bdev);
1567 EXPORT_SYMBOL(__invalidate_device);