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 sb->s_frozen = SB_FREEZE_TRANS;
250 up_write(&sb->s_umount);
251 mutex_unlock(&bdev->bd_fsfreeze_mutex);
255 sb->s_frozen = SB_FREEZE_WRITE;
260 sb->s_frozen = SB_FREEZE_TRANS;
263 sync_blockdev(sb->s_bdev);
265 if (sb->s_op->freeze_fs) {
266 error = sb->s_op->freeze_fs(sb);
269 "VFS:Filesystem freeze failed\n");
270 sb->s_frozen = SB_UNFROZEN;
271 deactivate_locked_super(sb);
272 bdev->bd_fsfreeze_count--;
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
274 return ERR_PTR(error);
277 up_write(&sb->s_umount);
281 mutex_unlock(&bdev->bd_fsfreeze_mutex);
282 return sb; /* thaw_bdev releases s->s_umount */
284 EXPORT_SYMBOL(freeze_bdev);
287 * thaw_bdev -- unlock filesystem
288 * @bdev: blockdevice to unlock
289 * @sb: associated superblock
291 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
293 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
297 mutex_lock(&bdev->bd_fsfreeze_mutex);
298 if (!bdev->bd_fsfreeze_count)
302 if (--bdev->bd_fsfreeze_count > 0)
308 BUG_ON(sb->s_bdev != bdev);
309 down_write(&sb->s_umount);
310 if (sb->s_flags & MS_RDONLY)
313 if (sb->s_op->unfreeze_fs) {
314 error = sb->s_op->unfreeze_fs(sb);
317 "VFS:Filesystem thaw failed\n");
318 sb->s_frozen = SB_FREEZE_TRANS;
319 bdev->bd_fsfreeze_count++;
320 mutex_unlock(&bdev->bd_fsfreeze_mutex);
326 sb->s_frozen = SB_UNFROZEN;
328 wake_up(&sb->s_wait_unfrozen);
331 deactivate_locked_super(sb);
333 mutex_unlock(&bdev->bd_fsfreeze_mutex);
336 EXPORT_SYMBOL(thaw_bdev);
338 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
340 return block_write_full_page(page, blkdev_get_block, wbc);
343 static int blkdev_readpage(struct file * file, struct page * page)
345 return block_read_full_page(page, blkdev_get_block);
348 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
353 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
357 static int blkdev_write_end(struct file *file, struct address_space *mapping,
358 loff_t pos, unsigned len, unsigned copied,
359 struct page *page, void *fsdata)
362 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
365 page_cache_release(page);
372 * for a block special file file->f_path.dentry->d_inode->i_size is zero
373 * so we compute the size by hand (just as in block_read/write above)
375 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
377 struct inode *bd_inode = file->f_mapping->host;
381 mutex_lock(&bd_inode->i_mutex);
382 size = i_size_read(bd_inode);
389 offset += file->f_pos;
392 if (offset >= 0 && offset <= size) {
393 if (offset != file->f_pos) {
394 file->f_pos = offset;
398 mutex_unlock(&bd_inode->i_mutex);
403 * Filp is never NULL; the only case when ->fsync() is called with
404 * NULL first argument is nfsd_sync_dir() and that's not a directory.
407 int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
409 return sync_blockdev(I_BDEV(filp->f_mapping->host));
416 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
417 static struct kmem_cache * bdev_cachep __read_mostly;
419 static struct inode *bdev_alloc_inode(struct super_block *sb)
421 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
424 return &ei->vfs_inode;
426 EXPORT_SYMBOL(block_fsync);
428 static void bdev_destroy_inode(struct inode *inode)
430 struct bdev_inode *bdi = BDEV_I(inode);
432 kmem_cache_free(bdev_cachep, bdi);
435 static void init_once(void *foo)
437 struct bdev_inode *ei = (struct bdev_inode *) foo;
438 struct block_device *bdev = &ei->bdev;
440 memset(bdev, 0, sizeof(*bdev));
441 mutex_init(&bdev->bd_mutex);
442 INIT_LIST_HEAD(&bdev->bd_inodes);
443 INIT_LIST_HEAD(&bdev->bd_list);
445 INIT_LIST_HEAD(&bdev->bd_holder_list);
447 inode_init_once(&ei->vfs_inode);
448 /* Initialize mutex for freeze. */
449 mutex_init(&bdev->bd_fsfreeze_mutex);
452 static inline void __bd_forget(struct inode *inode)
454 list_del_init(&inode->i_devices);
455 inode->i_bdev = NULL;
456 inode->i_mapping = &inode->i_data;
459 static void bdev_clear_inode(struct inode *inode)
461 struct block_device *bdev = &BDEV_I(inode)->bdev;
463 spin_lock(&bdev_lock);
464 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
465 __bd_forget(list_entry(p, struct inode, i_devices));
467 list_del_init(&bdev->bd_list);
468 spin_unlock(&bdev_lock);
471 static const struct super_operations bdev_sops = {
472 .statfs = simple_statfs,
473 .alloc_inode = bdev_alloc_inode,
474 .destroy_inode = bdev_destroy_inode,
475 .drop_inode = generic_delete_inode,
476 .clear_inode = bdev_clear_inode,
479 static int bd_get_sb(struct file_system_type *fs_type,
480 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
482 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
485 static struct file_system_type bd_type = {
488 .kill_sb = kill_anon_super,
491 struct super_block *blockdev_superblock __read_mostly;
493 void __init bdev_cache_init(void)
496 struct vfsmount *bd_mnt;
498 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
499 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
500 SLAB_MEM_SPREAD|SLAB_PANIC),
502 err = register_filesystem(&bd_type);
504 panic("Cannot register bdev pseudo-fs");
505 bd_mnt = kern_mount(&bd_type);
507 panic("Cannot create bdev pseudo-fs");
509 * This vfsmount structure is only used to obtain the
510 * blockdev_superblock, so tell kmemleak not to report it.
512 kmemleak_not_leak(bd_mnt);
513 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
517 * Most likely _very_ bad one - but then it's hardly critical for small
518 * /dev and can be fixed when somebody will need really large one.
519 * Keep in mind that it will be fed through icache hash function too.
521 static inline unsigned long hash(dev_t dev)
523 return MAJOR(dev)+MINOR(dev);
526 static int bdev_test(struct inode *inode, void *data)
528 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
531 static int bdev_set(struct inode *inode, void *data)
533 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
537 static LIST_HEAD(all_bdevs);
539 struct block_device *bdget(dev_t dev)
541 struct block_device *bdev;
544 inode = iget5_locked(blockdev_superblock, hash(dev),
545 bdev_test, bdev_set, &dev);
550 bdev = &BDEV_I(inode)->bdev;
552 if (inode->i_state & I_NEW) {
553 bdev->bd_contains = NULL;
554 bdev->bd_inode = inode;
555 bdev->bd_block_size = (1 << inode->i_blkbits);
556 bdev->bd_part_count = 0;
557 bdev->bd_invalidated = 0;
558 inode->i_mode = S_IFBLK;
560 inode->i_bdev = bdev;
561 inode->i_data.a_ops = &def_blk_aops;
562 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
563 inode->i_data.backing_dev_info = &default_backing_dev_info;
564 spin_lock(&bdev_lock);
565 list_add(&bdev->bd_list, &all_bdevs);
566 spin_unlock(&bdev_lock);
567 unlock_new_inode(inode);
572 EXPORT_SYMBOL(bdget);
575 * bdgrab -- Grab a reference to an already referenced block device
576 * @bdev: Block device to grab a reference to.
578 struct block_device *bdgrab(struct block_device *bdev)
580 atomic_inc(&bdev->bd_inode->i_count);
584 long nr_blockdev_pages(void)
586 struct block_device *bdev;
588 spin_lock(&bdev_lock);
589 list_for_each_entry(bdev, &all_bdevs, bd_list) {
590 ret += bdev->bd_inode->i_mapping->nrpages;
592 spin_unlock(&bdev_lock);
596 void bdput(struct block_device *bdev)
598 iput(bdev->bd_inode);
601 EXPORT_SYMBOL(bdput);
603 static struct block_device *bd_acquire(struct inode *inode)
605 struct block_device *bdev;
607 spin_lock(&bdev_lock);
608 bdev = inode->i_bdev;
610 atomic_inc(&bdev->bd_inode->i_count);
611 spin_unlock(&bdev_lock);
614 spin_unlock(&bdev_lock);
616 bdev = bdget(inode->i_rdev);
618 spin_lock(&bdev_lock);
619 if (!inode->i_bdev) {
621 * We take an additional bd_inode->i_count for inode,
622 * and it's released in clear_inode() of inode.
623 * So, we can access it via ->i_mapping always
626 atomic_inc(&bdev->bd_inode->i_count);
627 inode->i_bdev = bdev;
628 inode->i_mapping = bdev->bd_inode->i_mapping;
629 list_add(&inode->i_devices, &bdev->bd_inodes);
631 spin_unlock(&bdev_lock);
636 /* Call when you free inode */
638 void bd_forget(struct inode *inode)
640 struct block_device *bdev = NULL;
642 spin_lock(&bdev_lock);
644 if (!sb_is_blkdev_sb(inode->i_sb))
645 bdev = inode->i_bdev;
648 spin_unlock(&bdev_lock);
651 iput(bdev->bd_inode);
654 int bd_claim(struct block_device *bdev, void *holder)
657 spin_lock(&bdev_lock);
659 /* first decide result */
660 if (bdev->bd_holder == holder)
661 res = 0; /* already a holder */
662 else if (bdev->bd_holder != NULL)
663 res = -EBUSY; /* held by someone else */
664 else if (bdev->bd_contains == bdev)
665 res = 0; /* is a whole device which isn't held */
667 else if (bdev->bd_contains->bd_holder == bd_claim)
668 res = 0; /* is a partition of a device that is being partitioned */
669 else if (bdev->bd_contains->bd_holder != NULL)
670 res = -EBUSY; /* is a partition of a held device */
672 res = 0; /* is a partition of an un-held device */
674 /* now impose change */
676 /* note that for a whole device bd_holders
677 * will be incremented twice, and bd_holder will
678 * be set to bd_claim before being set to holder
680 bdev->bd_contains->bd_holders ++;
681 bdev->bd_contains->bd_holder = bd_claim;
683 bdev->bd_holder = holder;
685 spin_unlock(&bdev_lock);
689 EXPORT_SYMBOL(bd_claim);
691 void bd_release(struct block_device *bdev)
693 spin_lock(&bdev_lock);
694 if (!--bdev->bd_contains->bd_holders)
695 bdev->bd_contains->bd_holder = NULL;
696 if (!--bdev->bd_holders)
697 bdev->bd_holder = NULL;
698 spin_unlock(&bdev_lock);
701 EXPORT_SYMBOL(bd_release);
705 * Functions for bd_claim_by_kobject / bd_release_from_kobject
707 * If a kobject is passed to bd_claim_by_kobject()
708 * and the kobject has a parent directory,
709 * following symlinks are created:
710 * o from the kobject to the claimed bdev
711 * o from "holders" directory of the bdev to the parent of the kobject
712 * bd_release_from_kobject() removes these symlinks.
715 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
716 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
717 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
718 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
721 static int add_symlink(struct kobject *from, struct kobject *to)
725 return sysfs_create_link(from, to, kobject_name(to));
728 static void del_symlink(struct kobject *from, struct kobject *to)
732 sysfs_remove_link(from, kobject_name(to));
736 * 'struct bd_holder' contains pointers to kobjects symlinked by
737 * bd_claim_by_kobject.
738 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
741 struct list_head list; /* chain of holders of the bdev */
742 int count; /* references from the holder */
743 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
744 struct kobject *hdev; /* e.g. "/block/dm-0" */
745 struct kobject *hdir; /* e.g. "/block/sda/holders" */
746 struct kobject *sdev; /* e.g. "/block/sda" */
750 * Get references of related kobjects at once.
751 * Returns 1 on success. 0 on failure.
753 * Should call bd_holder_release_dirs() after successful use.
755 static int bd_holder_grab_dirs(struct block_device *bdev,
756 struct bd_holder *bo)
761 bo->sdir = kobject_get(bo->sdir);
765 bo->hdev = kobject_get(bo->sdir->parent);
769 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
773 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
780 kobject_put(bo->sdev);
782 kobject_put(bo->hdev);
784 kobject_put(bo->sdir);
789 /* Put references of related kobjects at once. */
790 static void bd_holder_release_dirs(struct bd_holder *bo)
792 kobject_put(bo->hdir);
793 kobject_put(bo->sdev);
794 kobject_put(bo->hdev);
795 kobject_put(bo->sdir);
798 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
800 struct bd_holder *bo;
802 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
812 static void free_bd_holder(struct bd_holder *bo)
818 * find_bd_holder - find matching struct bd_holder from the block device
820 * @bdev: struct block device to be searched
821 * @bo: target struct bd_holder
823 * Returns matching entry with @bo in @bdev->bd_holder_list.
824 * If found, increment the reference count and return the pointer.
825 * If not found, returns NULL.
827 static struct bd_holder *find_bd_holder(struct block_device *bdev,
828 struct bd_holder *bo)
830 struct bd_holder *tmp;
832 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
833 if (tmp->sdir == bo->sdir) {
842 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
844 * @bdev: block device to be bd_claimed
845 * @bo: preallocated and initialized by alloc_bd_holder()
847 * Add @bo to @bdev->bd_holder_list, create symlinks.
849 * Returns 0 if symlinks are created.
850 * Returns -ve if something fails.
852 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
859 if (!bd_holder_grab_dirs(bdev, bo))
862 err = add_symlink(bo->sdir, bo->sdev);
866 err = add_symlink(bo->hdir, bo->hdev);
868 del_symlink(bo->sdir, bo->sdev);
872 list_add_tail(&bo->list, &bdev->bd_holder_list);
877 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
879 * @bdev: block device to be bd_claimed
880 * @kobj: holder's kobject
882 * If there is matching entry with @kobj in @bdev->bd_holder_list
883 * and no other bd_claim() from the same kobject,
884 * remove the struct bd_holder from the list, delete symlinks for it.
886 * Returns a pointer to the struct bd_holder when it's removed from the list
887 * and ready to be freed.
888 * Returns NULL if matching claim isn't found or there is other bd_claim()
889 * by the same kobject.
891 static struct bd_holder *del_bd_holder(struct block_device *bdev,
892 struct kobject *kobj)
894 struct bd_holder *bo;
896 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
897 if (bo->sdir == kobj) {
899 BUG_ON(bo->count < 0);
902 del_symlink(bo->sdir, bo->sdev);
903 del_symlink(bo->hdir, bo->hdev);
904 bd_holder_release_dirs(bo);
915 * bd_claim_by_kobject - bd_claim() with additional kobject signature
917 * @bdev: block device to be claimed
918 * @holder: holder's signature
919 * @kobj: holder's kobject
921 * Do bd_claim() and if it succeeds, create sysfs symlinks between
922 * the bdev and the holder's kobject.
923 * Use bd_release_from_kobject() when relesing the claimed bdev.
925 * Returns 0 on success. (same as bd_claim())
926 * Returns errno on failure.
928 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
929 struct kobject *kobj)
932 struct bd_holder *bo, *found;
937 bo = alloc_bd_holder(kobj);
941 mutex_lock(&bdev->bd_mutex);
943 err = bd_claim(bdev, holder);
947 found = find_bd_holder(bdev, bo);
951 err = add_bd_holder(bdev, bo);
957 mutex_unlock(&bdev->bd_mutex);
963 * bd_release_from_kobject - bd_release() with additional kobject signature
965 * @bdev: block device to be released
966 * @kobj: holder's kobject
968 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
970 static void bd_release_from_kobject(struct block_device *bdev,
971 struct kobject *kobj)
976 mutex_lock(&bdev->bd_mutex);
978 free_bd_holder(del_bd_holder(bdev, kobj));
979 mutex_unlock(&bdev->bd_mutex);
983 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
985 * @bdev: block device to be claimed
986 * @holder: holder's signature
987 * @disk: holder's gendisk
989 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
991 int bd_claim_by_disk(struct block_device *bdev, void *holder,
992 struct gendisk *disk)
994 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
996 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
999 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1001 * @bdev: block device to be claimed
1002 * @disk: holder's gendisk
1004 * Call bd_release_from_kobject() and put @disk->slave_dir.
1006 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1008 bd_release_from_kobject(bdev, disk->slave_dir);
1009 kobject_put(disk->slave_dir);
1011 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1015 * Tries to open block device by device number. Use it ONLY if you
1016 * really do not have anything better - i.e. when you are behind a
1017 * truly sucky interface and all you are given is a device number. _Never_
1018 * to be used for internal purposes. If you ever need it - reconsider
1021 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
1023 struct block_device *bdev = bdget(dev);
1026 err = blkdev_get(bdev, mode);
1027 return err ? ERR_PTR(err) : bdev;
1030 EXPORT_SYMBOL(open_by_devnum);
1033 * flush_disk - invalidates all buffer-cache entries on a disk
1035 * @bdev: struct block device to be flushed
1037 * Invalidates all buffer-cache entries on a disk. It should be called
1038 * when a disk has been changed -- either by a media change or online
1041 static void flush_disk(struct block_device *bdev)
1043 if (__invalidate_device(bdev)) {
1044 char name[BDEVNAME_SIZE] = "";
1047 disk_name(bdev->bd_disk, 0, name);
1048 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1049 "resized disk %s\n", name);
1054 if (disk_partitionable(bdev->bd_disk))
1055 bdev->bd_invalidated = 1;
1059 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1060 * @disk: struct gendisk to check
1061 * @bdev: struct bdev to adjust.
1063 * This routine checks to see if the bdev size does not match the disk size
1064 * and adjusts it if it differs.
1066 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1068 loff_t disk_size, bdev_size;
1070 disk_size = (loff_t)get_capacity(disk) << 9;
1071 bdev_size = i_size_read(bdev->bd_inode);
1072 if (disk_size != bdev_size) {
1073 char name[BDEVNAME_SIZE];
1075 disk_name(disk, 0, name);
1077 "%s: detected capacity change from %lld to %lld\n",
1078 name, bdev_size, disk_size);
1079 i_size_write(bdev->bd_inode, disk_size);
1083 EXPORT_SYMBOL(check_disk_size_change);
1086 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1087 * @disk: struct gendisk to be revalidated
1089 * This routine is a wrapper for lower-level driver's revalidate_disk
1090 * call-backs. It is used to do common pre and post operations needed
1091 * for all revalidate_disk operations.
1093 int revalidate_disk(struct gendisk *disk)
1095 struct block_device *bdev;
1098 if (disk->fops->revalidate_disk)
1099 ret = disk->fops->revalidate_disk(disk);
1101 bdev = bdget_disk(disk, 0);
1105 mutex_lock(&bdev->bd_mutex);
1106 check_disk_size_change(disk, bdev);
1107 mutex_unlock(&bdev->bd_mutex);
1111 EXPORT_SYMBOL(revalidate_disk);
1114 * This routine checks whether a removable media has been changed,
1115 * and invalidates all buffer-cache-entries in that case. This
1116 * is a relatively slow routine, so we have to try to minimize using
1117 * it. Thus it is called only upon a 'mount' or 'open'. This
1118 * is the best way of combining speed and utility, I think.
1119 * People changing diskettes in the middle of an operation deserve
1122 int check_disk_change(struct block_device *bdev)
1124 struct gendisk *disk = bdev->bd_disk;
1125 const struct block_device_operations *bdops = disk->fops;
1127 if (!bdops->media_changed)
1129 if (!bdops->media_changed(bdev->bd_disk))
1133 if (bdops->revalidate_disk)
1134 bdops->revalidate_disk(bdev->bd_disk);
1138 EXPORT_SYMBOL(check_disk_change);
1140 void bd_set_size(struct block_device *bdev, loff_t size)
1142 unsigned bsize = bdev_logical_block_size(bdev);
1144 bdev->bd_inode->i_size = size;
1145 while (bsize < PAGE_CACHE_SIZE) {
1150 bdev->bd_block_size = bsize;
1151 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1153 EXPORT_SYMBOL(bd_set_size);
1155 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1160 * mutex_lock(part->bd_mutex)
1161 * mutex_lock_nested(whole->bd_mutex, 1)
1164 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1166 struct gendisk *disk;
1171 if (mode & FMODE_READ)
1173 if (mode & FMODE_WRITE)
1176 * hooks: /n/, see "layering violations".
1179 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1190 disk = get_gendisk(bdev->bd_dev, &partno);
1192 goto out_unlock_kernel;
1194 mutex_lock_nested(&bdev->bd_mutex, for_part);
1195 if (!bdev->bd_openers) {
1196 bdev->bd_disk = disk;
1197 bdev->bd_contains = bdev;
1199 struct backing_dev_info *bdi;
1202 bdev->bd_part = disk_get_part(disk, partno);
1206 if (disk->fops->open) {
1207 ret = disk->fops->open(bdev, mode);
1208 if (ret == -ERESTARTSYS) {
1209 /* Lost a race with 'disk' being
1210 * deleted, try again.
1213 disk_put_part(bdev->bd_part);
1214 bdev->bd_part = NULL;
1215 module_put(disk->fops->owner);
1217 bdev->bd_disk = NULL;
1218 mutex_unlock(&bdev->bd_mutex);
1224 if (!bdev->bd_openers) {
1225 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1226 bdi = blk_get_backing_dev_info(bdev);
1228 bdi = &default_backing_dev_info;
1229 bdev->bd_inode->i_data.backing_dev_info = bdi;
1231 if (bdev->bd_invalidated)
1232 rescan_partitions(disk, bdev);
1234 struct block_device *whole;
1235 whole = bdget_disk(disk, 0);
1240 ret = __blkdev_get(whole, mode, 1);
1243 bdev->bd_contains = whole;
1244 bdev->bd_inode->i_data.backing_dev_info =
1245 whole->bd_inode->i_data.backing_dev_info;
1246 bdev->bd_part = disk_get_part(disk, partno);
1247 if (!(disk->flags & GENHD_FL_UP) ||
1248 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1252 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1255 module_put(disk->fops->owner);
1258 if (bdev->bd_contains == bdev) {
1259 if (bdev->bd_disk->fops->open) {
1260 ret = bdev->bd_disk->fops->open(bdev, mode);
1262 goto out_unlock_bdev;
1264 if (bdev->bd_invalidated)
1265 rescan_partitions(bdev->bd_disk, bdev);
1270 bdev->bd_part_count++;
1271 mutex_unlock(&bdev->bd_mutex);
1276 disk_put_part(bdev->bd_part);
1277 bdev->bd_disk = NULL;
1278 bdev->bd_part = NULL;
1279 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1280 if (bdev != bdev->bd_contains)
1281 __blkdev_put(bdev->bd_contains, mode, 1);
1282 bdev->bd_contains = NULL;
1284 mutex_unlock(&bdev->bd_mutex);
1289 module_put(disk->fops->owner);
1296 int blkdev_get(struct block_device *bdev, fmode_t mode)
1298 return __blkdev_get(bdev, mode, 0);
1300 EXPORT_SYMBOL(blkdev_get);
1302 static int blkdev_open(struct inode * inode, struct file * filp)
1304 struct block_device *bdev;
1308 * Preserve backwards compatibility and allow large file access
1309 * even if userspace doesn't ask for it explicitly. Some mkfs
1310 * binary needs it. We might want to drop this workaround
1311 * during an unstable branch.
1313 filp->f_flags |= O_LARGEFILE;
1315 if (filp->f_flags & O_NDELAY)
1316 filp->f_mode |= FMODE_NDELAY;
1317 if (filp->f_flags & O_EXCL)
1318 filp->f_mode |= FMODE_EXCL;
1319 if ((filp->f_flags & O_ACCMODE) == 3)
1320 filp->f_mode |= FMODE_WRITE_IOCTL;
1322 bdev = bd_acquire(inode);
1326 filp->f_mapping = bdev->bd_inode->i_mapping;
1328 res = blkdev_get(bdev, filp->f_mode);
1332 if (filp->f_mode & FMODE_EXCL) {
1333 res = bd_claim(bdev, filp);
1335 goto out_blkdev_put;
1341 blkdev_put(bdev, filp->f_mode);
1345 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1348 struct gendisk *disk = bdev->bd_disk;
1349 struct block_device *victim = NULL;
1351 mutex_lock_nested(&bdev->bd_mutex, for_part);
1354 bdev->bd_part_count--;
1356 if (!--bdev->bd_openers) {
1357 sync_blockdev(bdev);
1360 if (bdev->bd_contains == bdev) {
1361 if (disk->fops->release)
1362 ret = disk->fops->release(disk, mode);
1364 if (!bdev->bd_openers) {
1365 struct module *owner = disk->fops->owner;
1369 disk_put_part(bdev->bd_part);
1370 bdev->bd_part = NULL;
1371 bdev->bd_disk = NULL;
1372 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1373 if (bdev != bdev->bd_contains)
1374 victim = bdev->bd_contains;
1375 bdev->bd_contains = NULL;
1378 mutex_unlock(&bdev->bd_mutex);
1381 __blkdev_put(victim, mode, 1);
1385 int blkdev_put(struct block_device *bdev, fmode_t mode)
1387 return __blkdev_put(bdev, mode, 0);
1389 EXPORT_SYMBOL(blkdev_put);
1391 static int blkdev_close(struct inode * inode, struct file * filp)
1393 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1394 if (bdev->bd_holder == filp)
1396 return blkdev_put(bdev, filp->f_mode);
1399 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1401 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1402 fmode_t mode = file->f_mode;
1405 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1406 * to updated it before every ioctl.
1408 if (file->f_flags & O_NDELAY)
1409 mode |= FMODE_NDELAY;
1411 mode &= ~FMODE_NDELAY;
1413 return blkdev_ioctl(bdev, mode, cmd, arg);
1417 * Write data to the block device. Only intended for the block device itself
1418 * and the raw driver which basically is a fake block device.
1420 * Does not take i_mutex for the write and thus is not for general purpose
1423 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1424 unsigned long nr_segs, loff_t pos)
1426 struct file *file = iocb->ki_filp;
1429 BUG_ON(iocb->ki_pos != pos);
1431 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1432 if (ret > 0 || ret == -EIOCBQUEUED) {
1435 err = generic_write_sync(file, pos, ret);
1436 if (err < 0 && ret > 0)
1441 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1444 * Try to release a page associated with block device when the system
1445 * is under memory pressure.
1447 static int blkdev_releasepage(struct page *page, gfp_t wait)
1449 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1451 if (super && super->s_op->bdev_try_to_free_page)
1452 return super->s_op->bdev_try_to_free_page(super, page, wait);
1454 return try_to_free_buffers(page);
1457 static const struct address_space_operations def_blk_aops = {
1458 .readpage = blkdev_readpage,
1459 .writepage = blkdev_writepage,
1460 .sync_page = block_sync_page,
1461 .write_begin = blkdev_write_begin,
1462 .write_end = blkdev_write_end,
1463 .writepages = generic_writepages,
1464 .releasepage = blkdev_releasepage,
1465 .direct_IO = blkdev_direct_IO,
1468 const struct file_operations def_blk_fops = {
1469 .open = blkdev_open,
1470 .release = blkdev_close,
1471 .llseek = block_llseek,
1472 .read = do_sync_read,
1473 .write = do_sync_write,
1474 .aio_read = generic_file_aio_read,
1475 .aio_write = blkdev_aio_write,
1476 .mmap = generic_file_mmap,
1477 .fsync = block_fsync,
1478 .unlocked_ioctl = block_ioctl,
1479 #ifdef CONFIG_COMPAT
1480 .compat_ioctl = compat_blkdev_ioctl,
1482 .splice_read = generic_file_splice_read,
1483 .splice_write = generic_file_splice_write,
1486 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1489 mm_segment_t old_fs = get_fs();
1491 res = blkdev_ioctl(bdev, 0, cmd, arg);
1496 EXPORT_SYMBOL(ioctl_by_bdev);
1499 * lookup_bdev - lookup a struct block_device by name
1500 * @pathname: special file representing the block device
1502 * Get a reference to the blockdevice at @pathname in the current
1503 * namespace if possible and return it. Return ERR_PTR(error)
1506 struct block_device *lookup_bdev(const char *pathname)
1508 struct block_device *bdev;
1509 struct inode *inode;
1513 if (!pathname || !*pathname)
1514 return ERR_PTR(-EINVAL);
1516 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1518 return ERR_PTR(error);
1520 inode = path.dentry->d_inode;
1522 if (!S_ISBLK(inode->i_mode))
1525 if (path.mnt->mnt_flags & MNT_NODEV)
1528 bdev = bd_acquire(inode);
1535 bdev = ERR_PTR(error);
1538 EXPORT_SYMBOL(lookup_bdev);
1541 * open_bdev_exclusive - open a block device by name and set it up for use
1543 * @path: special file representing the block device
1544 * @mode: FMODE_... combination to pass be used
1545 * @holder: owner for exclusion
1547 * Open the blockdevice described by the special file at @path, claim it
1550 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1552 struct block_device *bdev;
1555 bdev = lookup_bdev(path);
1559 error = blkdev_get(bdev, mode);
1561 return ERR_PTR(error);
1563 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1565 error = bd_claim(bdev, holder);
1572 blkdev_put(bdev, mode);
1573 return ERR_PTR(error);
1576 EXPORT_SYMBOL(open_bdev_exclusive);
1579 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1581 * @bdev: blockdevice to close
1582 * @mode: mode, must match that used to open.
1584 * This is the counterpart to open_bdev_exclusive().
1586 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1589 blkdev_put(bdev, mode);
1592 EXPORT_SYMBOL(close_bdev_exclusive);
1594 int __invalidate_device(struct block_device *bdev)
1596 struct super_block *sb = get_super(bdev);
1601 * no need to lock the super, get_super holds the
1602 * read mutex so the filesystem cannot go away
1603 * under us (->put_super runs with the write lock
1606 shrink_dcache_sb(sb);
1607 res = invalidate_inodes(sb);
1610 invalidate_bdev(bdev);
1613 EXPORT_SYMBOL(__invalidate_device);