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/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/buffer_head.h>
20 #include <linux/writeback.h>
21 #include <linux/mpage.h>
22 #include <linux/mount.h>
23 #include <linux/uio.h>
24 #include <linux/namei.h>
25 #include <linux/log2.h>
26 #include <asm/uaccess.h>
30 struct block_device bdev;
31 struct inode vfs_inode;
34 static inline struct bdev_inode *BDEV_I(struct inode *inode)
36 return container_of(inode, struct bdev_inode, vfs_inode);
39 inline struct block_device *I_BDEV(struct inode *inode)
41 return &BDEV_I(inode)->bdev;
44 EXPORT_SYMBOL(I_BDEV);
46 static sector_t max_block(struct block_device *bdev)
48 sector_t retval = ~((sector_t)0);
49 loff_t sz = i_size_read(bdev->bd_inode);
52 unsigned int size = block_size(bdev);
53 unsigned int sizebits = blksize_bits(size);
54 retval = (sz >> sizebits);
59 /* Kill _all_ buffers and pagecache , dirty or not.. */
60 static void kill_bdev(struct block_device *bdev)
62 if (bdev->bd_inode->i_mapping->nrpages == 0)
65 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
68 int set_blocksize(struct block_device *bdev, int size)
70 /* Size must be a power of two, and between 512 and PAGE_SIZE */
71 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
74 /* Size cannot be smaller than the size supported by the device */
75 if (size < bdev_hardsect_size(bdev))
78 /* Don't change the size if it is same as current */
79 if (bdev->bd_block_size != size) {
81 bdev->bd_block_size = size;
82 bdev->bd_inode->i_blkbits = blksize_bits(size);
88 EXPORT_SYMBOL(set_blocksize);
90 int sb_set_blocksize(struct super_block *sb, int size)
92 if (set_blocksize(sb->s_bdev, size))
94 /* If we get here, we know size is power of two
95 * and it's value is between 512 and PAGE_SIZE */
96 sb->s_blocksize = size;
97 sb->s_blocksize_bits = blksize_bits(size);
98 return sb->s_blocksize;
101 EXPORT_SYMBOL(sb_set_blocksize);
103 int sb_min_blocksize(struct super_block *sb, int size)
105 int minsize = bdev_hardsect_size(sb->s_bdev);
108 return sb_set_blocksize(sb, size);
111 EXPORT_SYMBOL(sb_min_blocksize);
114 blkdev_get_block(struct inode *inode, sector_t iblock,
115 struct buffer_head *bh, int create)
117 if (iblock >= max_block(I_BDEV(inode))) {
122 * for reads, we're just trying to fill a partial page.
123 * return a hole, they will have to call get_block again
124 * before they can fill it, and they will get -EIO at that
129 bh->b_bdev = I_BDEV(inode);
130 bh->b_blocknr = iblock;
131 set_buffer_mapped(bh);
136 blkdev_get_blocks(struct inode *inode, sector_t iblock,
137 struct buffer_head *bh, int create)
139 sector_t end_block = max_block(I_BDEV(inode));
140 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
142 if ((iblock + max_blocks) > end_block) {
143 max_blocks = end_block - iblock;
144 if ((long)max_blocks <= 0) {
146 return -EIO; /* write fully beyond EOF */
148 * It is a read which is fully beyond EOF. We return
149 * a !buffer_mapped buffer
155 bh->b_bdev = I_BDEV(inode);
156 bh->b_blocknr = iblock;
157 bh->b_size = max_blocks << inode->i_blkbits;
159 set_buffer_mapped(bh);
164 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
165 loff_t offset, unsigned long nr_segs)
167 struct file *file = iocb->ki_filp;
168 struct inode *inode = file->f_mapping->host;
170 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
171 iov, offset, nr_segs, blkdev_get_blocks, NULL);
175 static int blk_end_aio(struct bio *bio, unsigned int bytes_done, int error)
177 struct kiocb *iocb = bio->bi_private;
178 atomic_t *bio_count = &iocb->ki_bio_count;
180 if (bio_data_dir(bio) == READ)
181 bio_check_pages_dirty(bio);
183 bio_release_pages(bio);
187 /* iocb->ki_nbytes stores error code from LLDD */
189 iocb->ki_nbytes = -EIO;
191 if (atomic_dec_and_test(bio_count)) {
192 if ((long)iocb->ki_nbytes < 0)
193 aio_complete(iocb, iocb->ki_nbytes, 0);
195 aio_complete(iocb, iocb->ki_left, 0);
205 struct page *page[VEC_SIZE];
208 #define PAGES_SPANNED(addr, len) \
209 (DIV_ROUND_UP((addr) + (len), PAGE_SIZE) - (addr) / PAGE_SIZE);
212 * get page pointer for user addr, we internally cache struct page array for
213 * (addr, count) range in pvec to avoid frequent call to get_user_pages. If
214 * internal page list is exhausted, a batch count of up to VEC_SIZE is used
215 * to get next set of page struct.
217 static struct page *blk_get_page(unsigned long addr, size_t count, int rw,
221 if (pvec->idx == pvec->nr) {
222 nr_pages = PAGES_SPANNED(addr, count);
223 nr_pages = min(nr_pages, VEC_SIZE);
224 down_read(¤t->mm->mmap_sem);
225 ret = get_user_pages(current, current->mm, addr, nr_pages,
226 rw == READ, 0, pvec->page, NULL);
227 up_read(¤t->mm->mmap_sem);
233 return pvec->page[pvec->idx++];
236 /* return a page back to pvec array */
237 static void blk_unget_page(struct page *page, struct pvec *pvec)
239 pvec->page[--pvec->idx] = page;
243 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
244 loff_t pos, unsigned long nr_segs)
246 struct inode *inode = iocb->ki_filp->f_mapping->host;
247 unsigned blkbits = blksize_bits(bdev_hardsect_size(I_BDEV(inode)));
248 unsigned blocksize_mask = (1 << blkbits) - 1;
249 unsigned long seg = 0; /* iov segment iterator */
250 unsigned long nvec; /* number of bio vec needed */
251 unsigned long cur_off; /* offset into current page */
252 unsigned long cur_len; /* I/O len of current page, up to PAGE_SIZE */
254 unsigned long addr; /* user iovec address */
255 size_t count; /* user iovec len */
256 size_t nbytes = iocb->ki_nbytes = iocb->ki_left; /* total xfer size */
257 loff_t size; /* size of block device */
259 atomic_t *bio_count = &iocb->ki_bio_count;
266 if (pos & blocksize_mask)
269 size = i_size_read(inode);
270 if (pos + nbytes > size) {
272 iocb->ki_left = nbytes;
276 * check first non-zero iov alignment, the remaining
277 * iov alignment is checked inside bio loop below.
280 addr = (unsigned long) iov[seg].iov_base;
281 count = min(iov[seg].iov_len, nbytes);
282 if (addr & blocksize_mask || count & blocksize_mask)
284 } while (!count && ++seg < nr_segs);
285 atomic_set(bio_count, 1);
288 /* roughly estimate number of bio vec needed */
289 nvec = (nbytes + PAGE_SIZE - 1) / PAGE_SIZE;
290 nvec = max(nvec, nr_segs - seg);
291 nvec = min(nvec, (unsigned long) BIO_MAX_PAGES);
293 /* bio_alloc should not fail with GFP_KERNEL flag */
294 bio = bio_alloc(GFP_KERNEL, nvec);
295 bio->bi_bdev = I_BDEV(inode);
296 bio->bi_end_io = blk_end_aio;
297 bio->bi_private = iocb;
298 bio->bi_sector = pos >> blkbits;
300 cur_off = addr & ~PAGE_MASK;
301 cur_len = PAGE_SIZE - cur_off;
305 page = blk_get_page(addr, count, rw, &pvec);
306 if (unlikely(IS_ERR(page)))
309 if (bio_add_page(bio, page, cur_len, cur_off)) {
317 while (++seg < nr_segs) {
318 addr = (unsigned long) iov[seg].iov_base;
319 count = iov[seg].iov_len;
322 if (unlikely(addr & blocksize_mask ||
323 count & blocksize_mask)) {
324 page = ERR_PTR(-EINVAL);
327 count = min(count, nbytes);
331 blk_unget_page(page, &pvec);
334 /* bio is ready, submit it */
336 bio_set_pages_dirty(bio);
337 atomic_inc(bio_count);
342 iocb->ki_left -= nbytes;
343 nbytes = iocb->ki_left;
344 iocb->ki_pos += nbytes;
346 blk_run_address_space(inode->i_mapping);
347 if (atomic_dec_and_test(bio_count))
348 aio_complete(iocb, nbytes, 0);
354 * back out nbytes count constructed so far for this bio,
355 * we will throw away current bio.
357 nbytes += bio->bi_size;
358 bio_release_pages(bio);
362 * if no bio was submmitted, return the error code.
363 * otherwise, proceed with pending I/O completion.
365 if (atomic_read(bio_count) == 1)
366 return PTR_ERR(page);
371 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
373 return block_write_full_page(page, blkdev_get_block, wbc);
376 static int blkdev_readpage(struct file * file, struct page * page)
378 return block_read_full_page(page, blkdev_get_block);
381 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
383 return block_prepare_write(page, from, to, blkdev_get_block);
386 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
388 return block_commit_write(page, from, to);
393 * for a block special file file->f_path.dentry->d_inode->i_size is zero
394 * so we compute the size by hand (just as in block_read/write above)
396 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
398 struct inode *bd_inode = file->f_mapping->host;
402 mutex_lock(&bd_inode->i_mutex);
403 size = i_size_read(bd_inode);
410 offset += file->f_pos;
413 if (offset >= 0 && offset <= size) {
414 if (offset != file->f_pos) {
415 file->f_pos = offset;
419 mutex_unlock(&bd_inode->i_mutex);
424 * Filp is never NULL; the only case when ->fsync() is called with
425 * NULL first argument is nfsd_sync_dir() and that's not a directory.
428 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
430 return sync_blockdev(I_BDEV(filp->f_mapping->host));
437 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
438 static struct kmem_cache * bdev_cachep __read_mostly;
440 static struct inode *bdev_alloc_inode(struct super_block *sb)
442 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
445 return &ei->vfs_inode;
448 static void bdev_destroy_inode(struct inode *inode)
450 struct bdev_inode *bdi = BDEV_I(inode);
452 bdi->bdev.bd_inode_backing_dev_info = NULL;
453 kmem_cache_free(bdev_cachep, bdi);
456 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
458 struct bdev_inode *ei = (struct bdev_inode *) foo;
459 struct block_device *bdev = &ei->bdev;
461 memset(bdev, 0, sizeof(*bdev));
462 mutex_init(&bdev->bd_mutex);
463 sema_init(&bdev->bd_mount_sem, 1);
464 INIT_LIST_HEAD(&bdev->bd_inodes);
465 INIT_LIST_HEAD(&bdev->bd_list);
467 INIT_LIST_HEAD(&bdev->bd_holder_list);
469 inode_init_once(&ei->vfs_inode);
472 static inline void __bd_forget(struct inode *inode)
474 list_del_init(&inode->i_devices);
475 inode->i_bdev = NULL;
476 inode->i_mapping = &inode->i_data;
479 static void bdev_clear_inode(struct inode *inode)
481 struct block_device *bdev = &BDEV_I(inode)->bdev;
483 spin_lock(&bdev_lock);
484 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
485 __bd_forget(list_entry(p, struct inode, i_devices));
487 list_del_init(&bdev->bd_list);
488 spin_unlock(&bdev_lock);
491 static const struct super_operations bdev_sops = {
492 .statfs = simple_statfs,
493 .alloc_inode = bdev_alloc_inode,
494 .destroy_inode = bdev_destroy_inode,
495 .drop_inode = generic_delete_inode,
496 .clear_inode = bdev_clear_inode,
499 static int bd_get_sb(struct file_system_type *fs_type,
500 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
502 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
505 static struct file_system_type bd_type = {
508 .kill_sb = kill_anon_super,
511 static struct vfsmount *bd_mnt __read_mostly;
512 struct super_block *blockdev_superblock;
514 void __init bdev_cache_init(void)
517 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
518 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
519 SLAB_MEM_SPREAD|SLAB_PANIC),
521 err = register_filesystem(&bd_type);
523 panic("Cannot register bdev pseudo-fs");
524 bd_mnt = kern_mount(&bd_type);
525 err = PTR_ERR(bd_mnt);
527 panic("Cannot create bdev pseudo-fs");
528 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
532 * Most likely _very_ bad one - but then it's hardly critical for small
533 * /dev and can be fixed when somebody will need really large one.
534 * Keep in mind that it will be fed through icache hash function too.
536 static inline unsigned long hash(dev_t dev)
538 return MAJOR(dev)+MINOR(dev);
541 static int bdev_test(struct inode *inode, void *data)
543 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
546 static int bdev_set(struct inode *inode, void *data)
548 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
552 static LIST_HEAD(all_bdevs);
554 struct block_device *bdget(dev_t dev)
556 struct block_device *bdev;
559 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
560 bdev_test, bdev_set, &dev);
565 bdev = &BDEV_I(inode)->bdev;
567 if (inode->i_state & I_NEW) {
568 bdev->bd_contains = NULL;
569 bdev->bd_inode = inode;
570 bdev->bd_block_size = (1 << inode->i_blkbits);
571 bdev->bd_part_count = 0;
572 bdev->bd_invalidated = 0;
573 inode->i_mode = S_IFBLK;
575 inode->i_bdev = bdev;
576 inode->i_data.a_ops = &def_blk_aops;
577 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
578 inode->i_data.backing_dev_info = &default_backing_dev_info;
579 spin_lock(&bdev_lock);
580 list_add(&bdev->bd_list, &all_bdevs);
581 spin_unlock(&bdev_lock);
582 unlock_new_inode(inode);
587 EXPORT_SYMBOL(bdget);
589 long nr_blockdev_pages(void)
593 spin_lock(&bdev_lock);
594 list_for_each(p, &all_bdevs) {
595 struct block_device *bdev;
596 bdev = list_entry(p, struct block_device, bd_list);
597 ret += bdev->bd_inode->i_mapping->nrpages;
599 spin_unlock(&bdev_lock);
603 void bdput(struct block_device *bdev)
605 iput(bdev->bd_inode);
608 EXPORT_SYMBOL(bdput);
610 static struct block_device *bd_acquire(struct inode *inode)
612 struct block_device *bdev;
614 spin_lock(&bdev_lock);
615 bdev = inode->i_bdev;
617 atomic_inc(&bdev->bd_inode->i_count);
618 spin_unlock(&bdev_lock);
621 spin_unlock(&bdev_lock);
623 bdev = bdget(inode->i_rdev);
625 spin_lock(&bdev_lock);
626 if (!inode->i_bdev) {
628 * We take an additional bd_inode->i_count for inode,
629 * and it's released in clear_inode() of inode.
630 * So, we can access it via ->i_mapping always
633 atomic_inc(&bdev->bd_inode->i_count);
634 inode->i_bdev = bdev;
635 inode->i_mapping = bdev->bd_inode->i_mapping;
636 list_add(&inode->i_devices, &bdev->bd_inodes);
638 spin_unlock(&bdev_lock);
643 /* Call when you free inode */
645 void bd_forget(struct inode *inode)
647 struct block_device *bdev = NULL;
649 spin_lock(&bdev_lock);
651 if (inode->i_sb != blockdev_superblock)
652 bdev = inode->i_bdev;
655 spin_unlock(&bdev_lock);
658 iput(bdev->bd_inode);
661 int bd_claim(struct block_device *bdev, void *holder)
664 spin_lock(&bdev_lock);
666 /* first decide result */
667 if (bdev->bd_holder == holder)
668 res = 0; /* already a holder */
669 else if (bdev->bd_holder != NULL)
670 res = -EBUSY; /* held by someone else */
671 else if (bdev->bd_contains == bdev)
672 res = 0; /* is a whole device which isn't held */
674 else if (bdev->bd_contains->bd_holder == bd_claim)
675 res = 0; /* is a partition of a device that is being partitioned */
676 else if (bdev->bd_contains->bd_holder != NULL)
677 res = -EBUSY; /* is a partition of a held device */
679 res = 0; /* is a partition of an un-held device */
681 /* now impose change */
683 /* note that for a whole device bd_holders
684 * will be incremented twice, and bd_holder will
685 * be set to bd_claim before being set to holder
687 bdev->bd_contains->bd_holders ++;
688 bdev->bd_contains->bd_holder = bd_claim;
690 bdev->bd_holder = holder;
692 spin_unlock(&bdev_lock);
696 EXPORT_SYMBOL(bd_claim);
698 void bd_release(struct block_device *bdev)
700 spin_lock(&bdev_lock);
701 if (!--bdev->bd_contains->bd_holders)
702 bdev->bd_contains->bd_holder = NULL;
703 if (!--bdev->bd_holders)
704 bdev->bd_holder = NULL;
705 spin_unlock(&bdev_lock);
708 EXPORT_SYMBOL(bd_release);
712 * Functions for bd_claim_by_kobject / bd_release_from_kobject
714 * If a kobject is passed to bd_claim_by_kobject()
715 * and the kobject has a parent directory,
716 * following symlinks are created:
717 * o from the kobject to the claimed bdev
718 * o from "holders" directory of the bdev to the parent of the kobject
719 * bd_release_from_kobject() removes these symlinks.
722 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
723 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
724 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
725 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
728 static struct kobject *bdev_get_kobj(struct block_device *bdev)
730 if (bdev->bd_contains != bdev)
731 return kobject_get(&bdev->bd_part->kobj);
733 return kobject_get(&bdev->bd_disk->kobj);
736 static struct kobject *bdev_get_holder(struct block_device *bdev)
738 if (bdev->bd_contains != bdev)
739 return kobject_get(bdev->bd_part->holder_dir);
741 return kobject_get(bdev->bd_disk->holder_dir);
744 static int add_symlink(struct kobject *from, struct kobject *to)
748 return sysfs_create_link(from, to, kobject_name(to));
751 static void del_symlink(struct kobject *from, struct kobject *to)
755 sysfs_remove_link(from, kobject_name(to));
759 * 'struct bd_holder' contains pointers to kobjects symlinked by
760 * bd_claim_by_kobject.
761 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
764 struct list_head list; /* chain of holders of the bdev */
765 int count; /* references from the holder */
766 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
767 struct kobject *hdev; /* e.g. "/block/dm-0" */
768 struct kobject *hdir; /* e.g. "/block/sda/holders" */
769 struct kobject *sdev; /* e.g. "/block/sda" */
773 * Get references of related kobjects at once.
774 * Returns 1 on success. 0 on failure.
776 * Should call bd_holder_release_dirs() after successful use.
778 static int bd_holder_grab_dirs(struct block_device *bdev,
779 struct bd_holder *bo)
784 bo->sdir = kobject_get(bo->sdir);
788 bo->hdev = kobject_get(bo->sdir->parent);
792 bo->sdev = bdev_get_kobj(bdev);
796 bo->hdir = bdev_get_holder(bdev);
803 kobject_put(bo->sdev);
805 kobject_put(bo->hdev);
807 kobject_put(bo->sdir);
812 /* Put references of related kobjects at once. */
813 static void bd_holder_release_dirs(struct bd_holder *bo)
815 kobject_put(bo->hdir);
816 kobject_put(bo->sdev);
817 kobject_put(bo->hdev);
818 kobject_put(bo->sdir);
821 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
823 struct bd_holder *bo;
825 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
835 static void free_bd_holder(struct bd_holder *bo)
841 * find_bd_holder - find matching struct bd_holder from the block device
843 * @bdev: struct block device to be searched
844 * @bo: target struct bd_holder
846 * Returns matching entry with @bo in @bdev->bd_holder_list.
847 * If found, increment the reference count and return the pointer.
848 * If not found, returns NULL.
850 static struct bd_holder *find_bd_holder(struct block_device *bdev,
851 struct bd_holder *bo)
853 struct bd_holder *tmp;
855 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
856 if (tmp->sdir == bo->sdir) {
865 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
867 * @bdev: block device to be bd_claimed
868 * @bo: preallocated and initialized by alloc_bd_holder()
870 * Add @bo to @bdev->bd_holder_list, create symlinks.
872 * Returns 0 if symlinks are created.
873 * Returns -ve if something fails.
875 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
882 if (!bd_holder_grab_dirs(bdev, bo))
885 ret = add_symlink(bo->sdir, bo->sdev);
887 ret = add_symlink(bo->hdir, bo->hdev);
889 del_symlink(bo->sdir, bo->sdev);
892 list_add_tail(&bo->list, &bdev->bd_holder_list);
897 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
899 * @bdev: block device to be bd_claimed
900 * @kobj: holder's kobject
902 * If there is matching entry with @kobj in @bdev->bd_holder_list
903 * and no other bd_claim() from the same kobject,
904 * remove the struct bd_holder from the list, delete symlinks for it.
906 * Returns a pointer to the struct bd_holder when it's removed from the list
907 * and ready to be freed.
908 * Returns NULL if matching claim isn't found or there is other bd_claim()
909 * by the same kobject.
911 static struct bd_holder *del_bd_holder(struct block_device *bdev,
912 struct kobject *kobj)
914 struct bd_holder *bo;
916 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
917 if (bo->sdir == kobj) {
919 BUG_ON(bo->count < 0);
922 del_symlink(bo->sdir, bo->sdev);
923 del_symlink(bo->hdir, bo->hdev);
924 bd_holder_release_dirs(bo);
935 * bd_claim_by_kobject - bd_claim() with additional kobject signature
937 * @bdev: block device to be claimed
938 * @holder: holder's signature
939 * @kobj: holder's kobject
941 * Do bd_claim() and if it succeeds, create sysfs symlinks between
942 * the bdev and the holder's kobject.
943 * Use bd_release_from_kobject() when relesing the claimed bdev.
945 * Returns 0 on success. (same as bd_claim())
946 * Returns errno on failure.
948 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
949 struct kobject *kobj)
952 struct bd_holder *bo, *found;
957 bo = alloc_bd_holder(kobj);
961 mutex_lock(&bdev->bd_mutex);
962 res = bd_claim(bdev, holder);
964 found = find_bd_holder(bdev, bo);
966 res = add_bd_holder(bdev, bo);
974 mutex_unlock(&bdev->bd_mutex);
980 * bd_release_from_kobject - bd_release() with additional kobject signature
982 * @bdev: block device to be released
983 * @kobj: holder's kobject
985 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
987 static void bd_release_from_kobject(struct block_device *bdev,
988 struct kobject *kobj)
990 struct bd_holder *bo;
995 mutex_lock(&bdev->bd_mutex);
997 if ((bo = del_bd_holder(bdev, kobj)))
999 mutex_unlock(&bdev->bd_mutex);
1003 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1005 * @bdev: block device to be claimed
1006 * @holder: holder's signature
1007 * @disk: holder's gendisk
1009 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1011 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1012 struct gendisk *disk)
1014 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1016 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1019 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1021 * @bdev: block device to be claimed
1022 * @disk: holder's gendisk
1024 * Call bd_release_from_kobject() and put @disk->slave_dir.
1026 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1028 bd_release_from_kobject(bdev, disk->slave_dir);
1029 kobject_put(disk->slave_dir);
1031 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1035 * Tries to open block device by device number. Use it ONLY if you
1036 * really do not have anything better - i.e. when you are behind a
1037 * truly sucky interface and all you are given is a device number. _Never_
1038 * to be used for internal purposes. If you ever need it - reconsider
1041 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
1043 struct block_device *bdev = bdget(dev);
1045 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
1047 err = blkdev_get(bdev, mode, flags);
1048 return err ? ERR_PTR(err) : bdev;
1051 EXPORT_SYMBOL(open_by_devnum);
1054 * This routine checks whether a removable media has been changed,
1055 * and invalidates all buffer-cache-entries in that case. This
1056 * is a relatively slow routine, so we have to try to minimize using
1057 * it. Thus it is called only upon a 'mount' or 'open'. This
1058 * is the best way of combining speed and utility, I think.
1059 * People changing diskettes in the middle of an operation deserve
1062 int check_disk_change(struct block_device *bdev)
1064 struct gendisk *disk = bdev->bd_disk;
1065 struct block_device_operations * bdops = disk->fops;
1067 if (!bdops->media_changed)
1069 if (!bdops->media_changed(bdev->bd_disk))
1072 if (__invalidate_device(bdev))
1073 printk("VFS: busy inodes on changed media.\n");
1075 if (bdops->revalidate_disk)
1076 bdops->revalidate_disk(bdev->bd_disk);
1077 if (bdev->bd_disk->minors > 1)
1078 bdev->bd_invalidated = 1;
1082 EXPORT_SYMBOL(check_disk_change);
1084 void bd_set_size(struct block_device *bdev, loff_t size)
1086 unsigned bsize = bdev_hardsect_size(bdev);
1088 bdev->bd_inode->i_size = size;
1089 while (bsize < PAGE_CACHE_SIZE) {
1094 bdev->bd_block_size = bsize;
1095 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1097 EXPORT_SYMBOL(bd_set_size);
1099 static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1101 static int __blkdev_put(struct block_device *bdev, int for_part);
1106 * mutex_lock(part->bd_mutex)
1107 * mutex_lock_nested(whole->bd_mutex, 1)
1110 static int do_open(struct block_device *bdev, struct file *file, int for_part)
1112 struct module *owner = NULL;
1113 struct gendisk *disk;
1117 file->f_mapping = bdev->bd_inode->i_mapping;
1119 disk = get_gendisk(bdev->bd_dev, &part);
1125 owner = disk->fops->owner;
1127 mutex_lock_nested(&bdev->bd_mutex, for_part);
1128 if (!bdev->bd_openers) {
1129 bdev->bd_disk = disk;
1130 bdev->bd_contains = bdev;
1132 struct backing_dev_info *bdi;
1133 if (disk->fops->open) {
1134 ret = disk->fops->open(bdev->bd_inode, file);
1138 if (!bdev->bd_openers) {
1139 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1140 bdi = blk_get_backing_dev_info(bdev);
1142 bdi = &default_backing_dev_info;
1143 bdev->bd_inode->i_data.backing_dev_info = bdi;
1145 if (bdev->bd_invalidated)
1146 rescan_partitions(disk, bdev);
1148 struct hd_struct *p;
1149 struct block_device *whole;
1150 whole = bdget_disk(disk, 0);
1155 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
1158 bdev->bd_contains = whole;
1159 p = disk->part[part - 1];
1160 bdev->bd_inode->i_data.backing_dev_info =
1161 whole->bd_inode->i_data.backing_dev_info;
1162 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
1166 kobject_get(&p->kobj);
1168 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
1173 if (bdev->bd_contains == bdev) {
1174 if (bdev->bd_disk->fops->open) {
1175 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1179 if (bdev->bd_invalidated)
1180 rescan_partitions(bdev->bd_disk, bdev);
1185 bdev->bd_part_count++;
1186 mutex_unlock(&bdev->bd_mutex);
1191 bdev->bd_disk = NULL;
1192 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1193 if (bdev != bdev->bd_contains)
1194 __blkdev_put(bdev->bd_contains, 1);
1195 bdev->bd_contains = NULL;
1199 mutex_unlock(&bdev->bd_mutex);
1206 static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1210 * This crockload is due to bad choice of ->open() type.
1212 * For now, block device ->open() routine must _not_
1213 * examine anything in 'inode' argument except ->i_rdev.
1215 struct file fake_file = {};
1216 struct dentry fake_dentry = {};
1217 fake_file.f_mode = mode;
1218 fake_file.f_flags = flags;
1219 fake_file.f_path.dentry = &fake_dentry;
1220 fake_dentry.d_inode = bdev->bd_inode;
1222 return do_open(bdev, &fake_file, for_part);
1225 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1227 return __blkdev_get(bdev, mode, flags, 0);
1229 EXPORT_SYMBOL(blkdev_get);
1231 static int blkdev_open(struct inode * inode, struct file * filp)
1233 struct block_device *bdev;
1237 * Preserve backwards compatibility and allow large file access
1238 * even if userspace doesn't ask for it explicitly. Some mkfs
1239 * binary needs it. We might want to drop this workaround
1240 * during an unstable branch.
1242 filp->f_flags |= O_LARGEFILE;
1244 bdev = bd_acquire(inode);
1248 res = do_open(bdev, filp, 0);
1252 if (!(filp->f_flags & O_EXCL) )
1255 if (!(res = bd_claim(bdev, filp)))
1262 static int __blkdev_put(struct block_device *bdev, int for_part)
1265 struct inode *bd_inode = bdev->bd_inode;
1266 struct gendisk *disk = bdev->bd_disk;
1267 struct block_device *victim = NULL;
1269 mutex_lock_nested(&bdev->bd_mutex, for_part);
1272 bdev->bd_part_count--;
1274 if (!--bdev->bd_openers) {
1275 sync_blockdev(bdev);
1278 if (bdev->bd_contains == bdev) {
1279 if (disk->fops->release)
1280 ret = disk->fops->release(bd_inode, NULL);
1282 if (!bdev->bd_openers) {
1283 struct module *owner = disk->fops->owner;
1288 if (bdev->bd_contains != bdev) {
1289 kobject_put(&bdev->bd_part->kobj);
1290 bdev->bd_part = NULL;
1292 bdev->bd_disk = NULL;
1293 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1294 if (bdev != bdev->bd_contains)
1295 victim = bdev->bd_contains;
1296 bdev->bd_contains = NULL;
1299 mutex_unlock(&bdev->bd_mutex);
1302 __blkdev_put(victim, 1);
1306 int blkdev_put(struct block_device *bdev)
1308 return __blkdev_put(bdev, 0);
1310 EXPORT_SYMBOL(blkdev_put);
1312 static int blkdev_close(struct inode * inode, struct file * filp)
1314 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1315 if (bdev->bd_holder == filp)
1317 return blkdev_put(bdev);
1320 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1322 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1325 const struct address_space_operations def_blk_aops = {
1326 .readpage = blkdev_readpage,
1327 .writepage = blkdev_writepage,
1328 .sync_page = block_sync_page,
1329 .prepare_write = blkdev_prepare_write,
1330 .commit_write = blkdev_commit_write,
1331 .writepages = generic_writepages,
1332 .direct_IO = blkdev_direct_IO,
1335 const struct file_operations def_blk_fops = {
1336 .open = blkdev_open,
1337 .release = blkdev_close,
1338 .llseek = block_llseek,
1339 .read = do_sync_read,
1340 .write = do_sync_write,
1341 .aio_read = generic_file_aio_read,
1342 .aio_write = generic_file_aio_write_nolock,
1343 .mmap = generic_file_mmap,
1344 .fsync = block_fsync,
1345 .unlocked_ioctl = block_ioctl,
1346 #ifdef CONFIG_COMPAT
1347 .compat_ioctl = compat_blkdev_ioctl,
1349 .sendfile = generic_file_sendfile,
1350 .splice_read = generic_file_splice_read,
1351 .splice_write = generic_file_splice_write,
1354 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1357 mm_segment_t old_fs = get_fs();
1359 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1364 EXPORT_SYMBOL(ioctl_by_bdev);
1367 * lookup_bdev - lookup a struct block_device by name
1369 * @path: special file representing the block device
1371 * Get a reference to the blockdevice at @path in the current
1372 * namespace if possible and return it. Return ERR_PTR(error)
1375 struct block_device *lookup_bdev(const char *path)
1377 struct block_device *bdev;
1378 struct inode *inode;
1379 struct nameidata nd;
1382 if (!path || !*path)
1383 return ERR_PTR(-EINVAL);
1385 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1387 return ERR_PTR(error);
1389 inode = nd.dentry->d_inode;
1391 if (!S_ISBLK(inode->i_mode))
1394 if (nd.mnt->mnt_flags & MNT_NODEV)
1397 bdev = bd_acquire(inode);
1404 bdev = ERR_PTR(error);
1409 * open_bdev_excl - open a block device by name and set it up for use
1411 * @path: special file representing the block device
1412 * @flags: %MS_RDONLY for opening read-only
1413 * @holder: owner for exclusion
1415 * Open the blockdevice described by the special file at @path, claim it
1418 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1420 struct block_device *bdev;
1421 mode_t mode = FMODE_READ;
1424 bdev = lookup_bdev(path);
1428 if (!(flags & MS_RDONLY))
1429 mode |= FMODE_WRITE;
1430 error = blkdev_get(bdev, mode, 0);
1432 return ERR_PTR(error);
1434 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1436 error = bd_claim(bdev, holder);
1444 return ERR_PTR(error);
1447 EXPORT_SYMBOL(open_bdev_excl);
1450 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1452 * @bdev: blockdevice to close
1454 * This is the counterpart to open_bdev_excl().
1456 void close_bdev_excl(struct block_device *bdev)
1462 EXPORT_SYMBOL(close_bdev_excl);
1464 int __invalidate_device(struct block_device *bdev)
1466 struct super_block *sb = get_super(bdev);
1471 * no need to lock the super, get_super holds the
1472 * read mutex so the filesystem cannot go away
1473 * under us (->put_super runs with the write lock
1476 shrink_dcache_sb(sb);
1477 res = invalidate_inodes(sb);
1480 invalidate_bdev(bdev);
1483 EXPORT_SYMBOL(__invalidate_device);