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/device_cgroup.h>
15 #include <linux/highmem.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/dax.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
88 EXPORT_SYMBOL(kill_bdev);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device *bdev)
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
95 if (mapping->nrpages) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping);
105 EXPORT_SYMBOL(invalidate_bdev);
107 static void set_init_blocksize(struct block_device *bdev)
109 unsigned bsize = bdev_logical_block_size(bdev);
110 loff_t size = i_size_read(bdev->bd_inode);
112 while (bsize < PAGE_SIZE) {
117 bdev->bd_block_size = bsize;
118 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
121 int set_blocksize(struct block_device *bdev, int size)
123 /* Size must be a power of two, and between 512 and PAGE_SIZE */
124 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
127 /* Size cannot be smaller than the size supported by the device */
128 if (size < bdev_logical_block_size(bdev))
131 /* Don't change the size if it is same as current */
132 if (bdev->bd_block_size != size) {
134 bdev->bd_block_size = size;
135 bdev->bd_inode->i_blkbits = blksize_bits(size);
141 EXPORT_SYMBOL(set_blocksize);
143 int sb_set_blocksize(struct super_block *sb, int size)
145 if (set_blocksize(sb->s_bdev, size))
147 /* If we get here, we know size is power of two
148 * and it's value is between 512 and PAGE_SIZE */
149 sb->s_blocksize = size;
150 sb->s_blocksize_bits = blksize_bits(size);
151 return sb->s_blocksize;
154 EXPORT_SYMBOL(sb_set_blocksize);
156 int sb_min_blocksize(struct super_block *sb, int size)
158 int minsize = bdev_logical_block_size(sb->s_bdev);
161 return sb_set_blocksize(sb, size);
164 EXPORT_SYMBOL(sb_min_blocksize);
167 blkdev_get_block(struct inode *inode, sector_t iblock,
168 struct buffer_head *bh, int create)
170 bh->b_bdev = I_BDEV(inode);
171 bh->b_blocknr = iblock;
172 set_buffer_mapped(bh);
176 static struct inode *bdev_file_inode(struct file *file)
178 return file->f_mapping->host;
181 static unsigned int dio_bio_write_op(struct kiocb *iocb)
183 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
185 /* avoid the need for a I/O completion work item */
186 if (iocb->ki_flags & IOCB_DSYNC)
191 #define DIO_INLINE_BIO_VECS 4
193 static void blkdev_bio_end_io_simple(struct bio *bio)
195 struct task_struct *waiter = bio->bi_private;
197 WRITE_ONCE(bio->bi_private, NULL);
198 wake_up_process(waiter);
202 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
205 struct file *file = iocb->ki_filp;
206 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
207 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
208 loff_t pos = iocb->ki_pos;
209 bool should_dirty = false;
215 if ((pos | iov_iter_alignment(iter)) &
216 (bdev_logical_block_size(bdev) - 1))
219 if (nr_pages <= DIO_INLINE_BIO_VECS)
222 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
228 bio_init(&bio, vecs, nr_pages);
229 bio_set_dev(&bio, bdev);
230 bio.bi_iter.bi_sector = pos >> 9;
231 bio.bi_write_hint = iocb->ki_hint;
232 bio.bi_private = current;
233 bio.bi_end_io = blkdev_bio_end_io_simple;
234 bio.bi_ioprio = iocb->ki_ioprio;
236 ret = bio_iov_iter_get_pages(&bio, iter);
239 ret = bio.bi_iter.bi_size;
241 if (iov_iter_rw(iter) == READ) {
242 bio.bi_opf = REQ_OP_READ;
243 if (iter_is_iovec(iter))
246 bio.bi_opf = dio_bio_write_op(iocb);
247 task_io_account_write(ret);
250 qc = submit_bio(&bio);
252 set_current_state(TASK_UNINTERRUPTIBLE);
253 if (!READ_ONCE(bio.bi_private))
255 if (!(iocb->ki_flags & IOCB_HIPRI) ||
256 !blk_poll(bdev_get_queue(bdev), qc))
259 __set_current_state(TASK_RUNNING);
261 bio_for_each_segment_all(bvec, &bio, i) {
262 if (should_dirty && !PageCompound(bvec->bv_page))
263 set_page_dirty_lock(bvec->bv_page);
264 put_page(bvec->bv_page);
267 if (unlikely(bio.bi_status))
268 ret = blk_status_to_errno(bio.bi_status);
271 if (vecs != inline_vecs)
282 struct task_struct *waiter;
287 bool should_dirty : 1;
292 static struct bio_set blkdev_dio_pool;
294 static void blkdev_bio_end_io(struct bio *bio)
296 struct blkdev_dio *dio = bio->bi_private;
297 bool should_dirty = dio->should_dirty;
299 if (bio->bi_status && !dio->bio.bi_status)
300 dio->bio.bi_status = bio->bi_status;
302 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
304 struct kiocb *iocb = dio->iocb;
307 if (likely(!dio->bio.bi_status)) {
311 ret = blk_status_to_errno(dio->bio.bi_status);
314 dio->iocb->ki_complete(iocb, ret, 0);
317 struct task_struct *waiter = dio->waiter;
319 WRITE_ONCE(dio->waiter, NULL);
320 wake_up_process(waiter);
325 bio_check_pages_dirty(bio);
327 struct bio_vec *bvec;
330 bio_for_each_segment_all(bvec, bio, i)
331 put_page(bvec->bv_page);
337 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
339 struct file *file = iocb->ki_filp;
340 struct inode *inode = bdev_file_inode(file);
341 struct block_device *bdev = I_BDEV(inode);
342 struct blk_plug plug;
343 struct blkdev_dio *dio;
345 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
346 loff_t pos = iocb->ki_pos;
347 blk_qc_t qc = BLK_QC_T_NONE;
350 if ((pos | iov_iter_alignment(iter)) &
351 (bdev_logical_block_size(bdev) - 1))
354 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
355 bio_get(bio); /* extra ref for the completion handler */
357 dio = container_of(bio, struct blkdev_dio, bio);
358 dio->is_sync = is_sync = is_sync_kiocb(iocb);
360 dio->waiter = current;
365 dio->multi_bio = false;
366 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
368 blk_start_plug(&plug);
370 bio_set_dev(bio, bdev);
371 bio->bi_iter.bi_sector = pos >> 9;
372 bio->bi_write_hint = iocb->ki_hint;
373 bio->bi_private = dio;
374 bio->bi_end_io = blkdev_bio_end_io;
375 bio->bi_ioprio = iocb->ki_ioprio;
377 ret = bio_iov_iter_get_pages(bio, iter);
379 bio->bi_status = BLK_STS_IOERR;
385 bio->bi_opf = REQ_OP_READ;
386 if (dio->should_dirty)
387 bio_set_pages_dirty(bio);
389 bio->bi_opf = dio_bio_write_op(iocb);
390 task_io_account_write(bio->bi_iter.bi_size);
393 dio->size += bio->bi_iter.bi_size;
394 pos += bio->bi_iter.bi_size;
396 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
398 qc = submit_bio(bio);
402 if (!dio->multi_bio) {
403 dio->multi_bio = true;
404 atomic_set(&dio->ref, 2);
406 atomic_inc(&dio->ref);
410 bio = bio_alloc(GFP_KERNEL, nr_pages);
412 blk_finish_plug(&plug);
418 set_current_state(TASK_UNINTERRUPTIBLE);
419 if (!READ_ONCE(dio->waiter))
422 if (!(iocb->ki_flags & IOCB_HIPRI) ||
423 !blk_poll(bdev_get_queue(bdev), qc))
426 __set_current_state(TASK_RUNNING);
429 ret = blk_status_to_errno(dio->bio.bi_status);
438 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
442 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
445 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
446 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
448 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
451 static __init int blkdev_init(void)
453 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
455 module_init(blkdev_init);
457 int __sync_blockdev(struct block_device *bdev, int wait)
462 return filemap_flush(bdev->bd_inode->i_mapping);
463 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
467 * Write out and wait upon all the dirty data associated with a block
468 * device via its mapping. Does not take the superblock lock.
470 int sync_blockdev(struct block_device *bdev)
472 return __sync_blockdev(bdev, 1);
474 EXPORT_SYMBOL(sync_blockdev);
477 * Write out and wait upon all dirty data associated with this
478 * device. Filesystem data as well as the underlying block
479 * device. Takes the superblock lock.
481 int fsync_bdev(struct block_device *bdev)
483 struct super_block *sb = get_super(bdev);
485 int res = sync_filesystem(sb);
489 return sync_blockdev(bdev);
491 EXPORT_SYMBOL(fsync_bdev);
494 * freeze_bdev -- lock a filesystem and force it into a consistent state
495 * @bdev: blockdevice to lock
497 * If a superblock is found on this device, we take the s_umount semaphore
498 * on it to make sure nobody unmounts until the snapshot creation is done.
499 * The reference counter (bd_fsfreeze_count) guarantees that only the last
500 * unfreeze process can unfreeze the frozen filesystem actually when multiple
501 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
502 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
505 struct super_block *freeze_bdev(struct block_device *bdev)
507 struct super_block *sb;
510 mutex_lock(&bdev->bd_fsfreeze_mutex);
511 if (++bdev->bd_fsfreeze_count > 1) {
513 * We don't even need to grab a reference - the first call
514 * to freeze_bdev grab an active reference and only the last
515 * thaw_bdev drops it.
517 sb = get_super(bdev);
520 mutex_unlock(&bdev->bd_fsfreeze_mutex);
524 sb = get_active_super(bdev);
527 if (sb->s_op->freeze_super)
528 error = sb->s_op->freeze_super(sb);
530 error = freeze_super(sb);
532 deactivate_super(sb);
533 bdev->bd_fsfreeze_count--;
534 mutex_unlock(&bdev->bd_fsfreeze_mutex);
535 return ERR_PTR(error);
537 deactivate_super(sb);
540 mutex_unlock(&bdev->bd_fsfreeze_mutex);
541 return sb; /* thaw_bdev releases s->s_umount */
543 EXPORT_SYMBOL(freeze_bdev);
546 * thaw_bdev -- unlock filesystem
547 * @bdev: blockdevice to unlock
548 * @sb: associated superblock
550 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
552 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
556 mutex_lock(&bdev->bd_fsfreeze_mutex);
557 if (!bdev->bd_fsfreeze_count)
561 if (--bdev->bd_fsfreeze_count > 0)
567 if (sb->s_op->thaw_super)
568 error = sb->s_op->thaw_super(sb);
570 error = thaw_super(sb);
572 bdev->bd_fsfreeze_count++;
574 mutex_unlock(&bdev->bd_fsfreeze_mutex);
577 EXPORT_SYMBOL(thaw_bdev);
579 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
581 return block_write_full_page(page, blkdev_get_block, wbc);
584 static int blkdev_readpage(struct file * file, struct page * page)
586 return block_read_full_page(page, blkdev_get_block);
589 static int blkdev_readpages(struct file *file, struct address_space *mapping,
590 struct list_head *pages, unsigned nr_pages)
592 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
595 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
596 loff_t pos, unsigned len, unsigned flags,
597 struct page **pagep, void **fsdata)
599 return block_write_begin(mapping, pos, len, flags, pagep,
603 static int blkdev_write_end(struct file *file, struct address_space *mapping,
604 loff_t pos, unsigned len, unsigned copied,
605 struct page *page, void *fsdata)
608 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
618 * for a block special file file_inode(file)->i_size is zero
619 * so we compute the size by hand (just as in block_read/write above)
621 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
623 struct inode *bd_inode = bdev_file_inode(file);
626 inode_lock(bd_inode);
627 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
628 inode_unlock(bd_inode);
632 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
634 struct inode *bd_inode = bdev_file_inode(filp);
635 struct block_device *bdev = I_BDEV(bd_inode);
638 error = file_write_and_wait_range(filp, start, end);
643 * There is no need to serialise calls to blkdev_issue_flush with
644 * i_mutex and doing so causes performance issues with concurrent
645 * O_SYNC writers to a block device.
647 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
648 if (error == -EOPNOTSUPP)
653 EXPORT_SYMBOL(blkdev_fsync);
656 * bdev_read_page() - Start reading a page from a block device
657 * @bdev: The device to read the page from
658 * @sector: The offset on the device to read the page to (need not be aligned)
659 * @page: The page to read
661 * On entry, the page should be locked. It will be unlocked when the page
662 * has been read. If the block driver implements rw_page synchronously,
663 * that will be true on exit from this function, but it need not be.
665 * Errors returned by this function are usually "soft", eg out of memory, or
666 * queue full; callers should try a different route to read this page rather
667 * than propagate an error back up the stack.
669 * Return: negative errno if an error occurs, 0 if submission was successful.
671 int bdev_read_page(struct block_device *bdev, sector_t sector,
674 const struct block_device_operations *ops = bdev->bd_disk->fops;
675 int result = -EOPNOTSUPP;
677 if (!ops->rw_page || bdev_get_integrity(bdev))
680 result = blk_queue_enter(bdev->bd_queue, 0);
683 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
685 blk_queue_exit(bdev->bd_queue);
688 EXPORT_SYMBOL_GPL(bdev_read_page);
691 * bdev_write_page() - Start writing a page to a block device
692 * @bdev: The device to write the page to
693 * @sector: The offset on the device to write the page to (need not be aligned)
694 * @page: The page to write
695 * @wbc: The writeback_control for the write
697 * On entry, the page should be locked and not currently under writeback.
698 * On exit, if the write started successfully, the page will be unlocked and
699 * under writeback. If the write failed already (eg the driver failed to
700 * queue the page to the device), the page will still be locked. If the
701 * caller is a ->writepage implementation, it will need to unlock the page.
703 * Errors returned by this function are usually "soft", eg out of memory, or
704 * queue full; callers should try a different route to write this page rather
705 * than propagate an error back up the stack.
707 * Return: negative errno if an error occurs, 0 if submission was successful.
709 int bdev_write_page(struct block_device *bdev, sector_t sector,
710 struct page *page, struct writeback_control *wbc)
713 const struct block_device_operations *ops = bdev->bd_disk->fops;
715 if (!ops->rw_page || bdev_get_integrity(bdev))
717 result = blk_queue_enter(bdev->bd_queue, 0);
721 set_page_writeback(page);
722 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
725 end_page_writeback(page);
727 clean_page_buffers(page);
730 blk_queue_exit(bdev->bd_queue);
733 EXPORT_SYMBOL_GPL(bdev_write_page);
739 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
740 static struct kmem_cache * bdev_cachep __read_mostly;
742 static struct inode *bdev_alloc_inode(struct super_block *sb)
744 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
747 return &ei->vfs_inode;
750 static void bdev_i_callback(struct rcu_head *head)
752 struct inode *inode = container_of(head, struct inode, i_rcu);
753 struct bdev_inode *bdi = BDEV_I(inode);
755 kmem_cache_free(bdev_cachep, bdi);
758 static void bdev_destroy_inode(struct inode *inode)
760 call_rcu(&inode->i_rcu, bdev_i_callback);
763 static void init_once(void *foo)
765 struct bdev_inode *ei = (struct bdev_inode *) foo;
766 struct block_device *bdev = &ei->bdev;
768 memset(bdev, 0, sizeof(*bdev));
769 mutex_init(&bdev->bd_mutex);
770 INIT_LIST_HEAD(&bdev->bd_list);
772 INIT_LIST_HEAD(&bdev->bd_holder_disks);
774 bdev->bd_bdi = &noop_backing_dev_info;
775 inode_init_once(&ei->vfs_inode);
776 /* Initialize mutex for freeze. */
777 mutex_init(&bdev->bd_fsfreeze_mutex);
780 static void bdev_evict_inode(struct inode *inode)
782 struct block_device *bdev = &BDEV_I(inode)->bdev;
783 truncate_inode_pages_final(&inode->i_data);
784 invalidate_inode_buffers(inode); /* is it needed here? */
786 spin_lock(&bdev_lock);
787 list_del_init(&bdev->bd_list);
788 spin_unlock(&bdev_lock);
789 /* Detach inode from wb early as bdi_put() may free bdi->wb */
790 inode_detach_wb(inode);
791 if (bdev->bd_bdi != &noop_backing_dev_info) {
792 bdi_put(bdev->bd_bdi);
793 bdev->bd_bdi = &noop_backing_dev_info;
797 static const struct super_operations bdev_sops = {
798 .statfs = simple_statfs,
799 .alloc_inode = bdev_alloc_inode,
800 .destroy_inode = bdev_destroy_inode,
801 .drop_inode = generic_delete_inode,
802 .evict_inode = bdev_evict_inode,
805 static struct dentry *bd_mount(struct file_system_type *fs_type,
806 int flags, const char *dev_name, void *data)
809 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
811 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
815 static struct file_system_type bd_type = {
818 .kill_sb = kill_anon_super,
821 struct super_block *blockdev_superblock __read_mostly;
822 EXPORT_SYMBOL_GPL(blockdev_superblock);
824 void __init bdev_cache_init(void)
827 static struct vfsmount *bd_mnt;
829 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
830 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
831 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
833 err = register_filesystem(&bd_type);
835 panic("Cannot register bdev pseudo-fs");
836 bd_mnt = kern_mount(&bd_type);
838 panic("Cannot create bdev pseudo-fs");
839 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
843 * Most likely _very_ bad one - but then it's hardly critical for small
844 * /dev and can be fixed when somebody will need really large one.
845 * Keep in mind that it will be fed through icache hash function too.
847 static inline unsigned long hash(dev_t dev)
849 return MAJOR(dev)+MINOR(dev);
852 static int bdev_test(struct inode *inode, void *data)
854 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
857 static int bdev_set(struct inode *inode, void *data)
859 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
863 static LIST_HEAD(all_bdevs);
866 * If there is a bdev inode for this device, unhash it so that it gets evicted
867 * as soon as last inode reference is dropped.
869 void bdev_unhash_inode(dev_t dev)
873 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
875 remove_inode_hash(inode);
880 struct block_device *bdget(dev_t dev)
882 struct block_device *bdev;
885 inode = iget5_locked(blockdev_superblock, hash(dev),
886 bdev_test, bdev_set, &dev);
891 bdev = &BDEV_I(inode)->bdev;
893 if (inode->i_state & I_NEW) {
894 bdev->bd_contains = NULL;
895 bdev->bd_super = NULL;
896 bdev->bd_inode = inode;
897 bdev->bd_block_size = i_blocksize(inode);
898 bdev->bd_part_count = 0;
899 bdev->bd_invalidated = 0;
900 inode->i_mode = S_IFBLK;
902 inode->i_bdev = bdev;
903 inode->i_data.a_ops = &def_blk_aops;
904 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
905 spin_lock(&bdev_lock);
906 list_add(&bdev->bd_list, &all_bdevs);
907 spin_unlock(&bdev_lock);
908 unlock_new_inode(inode);
913 EXPORT_SYMBOL(bdget);
916 * bdgrab -- Grab a reference to an already referenced block device
917 * @bdev: Block device to grab a reference to.
919 struct block_device *bdgrab(struct block_device *bdev)
921 ihold(bdev->bd_inode);
924 EXPORT_SYMBOL(bdgrab);
926 long nr_blockdev_pages(void)
928 struct block_device *bdev;
930 spin_lock(&bdev_lock);
931 list_for_each_entry(bdev, &all_bdevs, bd_list) {
932 ret += bdev->bd_inode->i_mapping->nrpages;
934 spin_unlock(&bdev_lock);
938 void bdput(struct block_device *bdev)
940 iput(bdev->bd_inode);
943 EXPORT_SYMBOL(bdput);
945 static struct block_device *bd_acquire(struct inode *inode)
947 struct block_device *bdev;
949 spin_lock(&bdev_lock);
950 bdev = inode->i_bdev;
951 if (bdev && !inode_unhashed(bdev->bd_inode)) {
953 spin_unlock(&bdev_lock);
956 spin_unlock(&bdev_lock);
959 * i_bdev references block device inode that was already shut down
960 * (corresponding device got removed). Remove the reference and look
961 * up block device inode again just in case new device got
962 * reestablished under the same device number.
967 bdev = bdget(inode->i_rdev);
969 spin_lock(&bdev_lock);
970 if (!inode->i_bdev) {
972 * We take an additional reference to bd_inode,
973 * and it's released in clear_inode() of inode.
974 * So, we can access it via ->i_mapping always
978 inode->i_bdev = bdev;
979 inode->i_mapping = bdev->bd_inode->i_mapping;
981 spin_unlock(&bdev_lock);
986 /* Call when you free inode */
988 void bd_forget(struct inode *inode)
990 struct block_device *bdev = NULL;
992 spin_lock(&bdev_lock);
993 if (!sb_is_blkdev_sb(inode->i_sb))
994 bdev = inode->i_bdev;
995 inode->i_bdev = NULL;
996 inode->i_mapping = &inode->i_data;
997 spin_unlock(&bdev_lock);
1004 * bd_may_claim - test whether a block device can be claimed
1005 * @bdev: block device of interest
1006 * @whole: whole block device containing @bdev, may equal @bdev
1007 * @holder: holder trying to claim @bdev
1009 * Test whether @bdev can be claimed by @holder.
1012 * spin_lock(&bdev_lock).
1015 * %true if @bdev can be claimed, %false otherwise.
1017 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1020 if (bdev->bd_holder == holder)
1021 return true; /* already a holder */
1022 else if (bdev->bd_holder != NULL)
1023 return false; /* held by someone else */
1024 else if (whole == bdev)
1025 return true; /* is a whole device which isn't held */
1027 else if (whole->bd_holder == bd_may_claim)
1028 return true; /* is a partition of a device that is being partitioned */
1029 else if (whole->bd_holder != NULL)
1030 return false; /* is a partition of a held device */
1032 return true; /* is a partition of an un-held device */
1036 * bd_prepare_to_claim - prepare to claim a block device
1037 * @bdev: block device of interest
1038 * @whole: the whole device containing @bdev, may equal @bdev
1039 * @holder: holder trying to claim @bdev
1041 * Prepare to claim @bdev. This function fails if @bdev is already
1042 * claimed by another holder and waits if another claiming is in
1043 * progress. This function doesn't actually claim. On successful
1044 * return, the caller has ownership of bd_claiming and bd_holder[s].
1047 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1048 * it multiple times.
1051 * 0 if @bdev can be claimed, -EBUSY otherwise.
1053 static int bd_prepare_to_claim(struct block_device *bdev,
1054 struct block_device *whole, void *holder)
1057 /* if someone else claimed, fail */
1058 if (!bd_may_claim(bdev, whole, holder))
1061 /* if claiming is already in progress, wait for it to finish */
1062 if (whole->bd_claiming) {
1063 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1066 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1067 spin_unlock(&bdev_lock);
1069 finish_wait(wq, &wait);
1070 spin_lock(&bdev_lock);
1078 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1080 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1085 * Now that we hold gendisk reference we make sure bdev we looked up is
1086 * not stale. If it is, it means device got removed and created before
1087 * we looked up gendisk and we fail open in such case. Associating
1088 * unhashed bdev with newly created gendisk could lead to two bdevs
1089 * (and thus two independent caches) being associated with one device
1092 if (inode_unhashed(bdev->bd_inode)) {
1093 put_disk_and_module(disk);
1100 * bd_start_claiming - start claiming a block device
1101 * @bdev: block device of interest
1102 * @holder: holder trying to claim @bdev
1104 * @bdev is about to be opened exclusively. Check @bdev can be opened
1105 * exclusively and mark that an exclusive open is in progress. Each
1106 * successful call to this function must be matched with a call to
1107 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1110 * This function is used to gain exclusive access to the block device
1111 * without actually causing other exclusive open attempts to fail. It
1112 * should be used when the open sequence itself requires exclusive
1113 * access but may subsequently fail.
1119 * Pointer to the block device containing @bdev on success, ERR_PTR()
1122 static struct block_device *bd_start_claiming(struct block_device *bdev,
1125 struct gendisk *disk;
1126 struct block_device *whole;
1132 * @bdev might not have been initialized properly yet, look up
1133 * and grab the outer block device the hard way.
1135 disk = bdev_get_gendisk(bdev, &partno);
1137 return ERR_PTR(-ENXIO);
1140 * Normally, @bdev should equal what's returned from bdget_disk()
1141 * if partno is 0; however, some drivers (floppy) use multiple
1142 * bdev's for the same physical device and @bdev may be one of the
1143 * aliases. Keep @bdev if partno is 0. This means claimer
1144 * tracking is broken for those devices but it has always been that
1148 whole = bdget_disk(disk, 0);
1150 whole = bdgrab(bdev);
1152 put_disk_and_module(disk);
1154 return ERR_PTR(-ENOMEM);
1156 /* prepare to claim, if successful, mark claiming in progress */
1157 spin_lock(&bdev_lock);
1159 err = bd_prepare_to_claim(bdev, whole, holder);
1161 whole->bd_claiming = holder;
1162 spin_unlock(&bdev_lock);
1165 spin_unlock(&bdev_lock);
1167 return ERR_PTR(err);
1172 struct bd_holder_disk {
1173 struct list_head list;
1174 struct gendisk *disk;
1178 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1179 struct gendisk *disk)
1181 struct bd_holder_disk *holder;
1183 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1184 if (holder->disk == disk)
1189 static int add_symlink(struct kobject *from, struct kobject *to)
1191 return sysfs_create_link(from, to, kobject_name(to));
1194 static void del_symlink(struct kobject *from, struct kobject *to)
1196 sysfs_remove_link(from, kobject_name(to));
1200 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1201 * @bdev: the claimed slave bdev
1202 * @disk: the holding disk
1204 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1206 * This functions creates the following sysfs symlinks.
1208 * - from "slaves" directory of the holder @disk to the claimed @bdev
1209 * - from "holders" directory of the @bdev to the holder @disk
1211 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1212 * passed to bd_link_disk_holder(), then:
1214 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1215 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1217 * The caller must have claimed @bdev before calling this function and
1218 * ensure that both @bdev and @disk are valid during the creation and
1219 * lifetime of these symlinks.
1225 * 0 on success, -errno on failure.
1227 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1229 struct bd_holder_disk *holder;
1232 mutex_lock(&bdev->bd_mutex);
1234 WARN_ON_ONCE(!bdev->bd_holder);
1236 /* FIXME: remove the following once add_disk() handles errors */
1237 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1240 holder = bd_find_holder_disk(bdev, disk);
1246 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1252 INIT_LIST_HEAD(&holder->list);
1253 holder->disk = disk;
1256 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1260 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1264 * bdev could be deleted beneath us which would implicitly destroy
1265 * the holder directory. Hold on to it.
1267 kobject_get(bdev->bd_part->holder_dir);
1269 list_add(&holder->list, &bdev->bd_holder_disks);
1273 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1277 mutex_unlock(&bdev->bd_mutex);
1280 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1283 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1284 * @bdev: the calimed slave bdev
1285 * @disk: the holding disk
1287 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1292 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1294 struct bd_holder_disk *holder;
1296 mutex_lock(&bdev->bd_mutex);
1298 holder = bd_find_holder_disk(bdev, disk);
1300 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1301 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1302 del_symlink(bdev->bd_part->holder_dir,
1303 &disk_to_dev(disk)->kobj);
1304 kobject_put(bdev->bd_part->holder_dir);
1305 list_del_init(&holder->list);
1309 mutex_unlock(&bdev->bd_mutex);
1311 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1315 * flush_disk - invalidates all buffer-cache entries on a disk
1317 * @bdev: struct block device to be flushed
1318 * @kill_dirty: flag to guide handling of dirty inodes
1320 * Invalidates all buffer-cache entries on a disk. It should be called
1321 * when a disk has been changed -- either by a media change or online
1324 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1326 if (__invalidate_device(bdev, kill_dirty)) {
1327 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1328 "resized disk %s\n",
1329 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1331 bdev->bd_invalidated = 1;
1335 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1336 * @disk: struct gendisk to check
1337 * @bdev: struct bdev to adjust.
1338 * @verbose: if %true log a message about a size change if there is any
1340 * This routine checks to see if the bdev size does not match the disk size
1341 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1344 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1347 loff_t disk_size, bdev_size;
1349 disk_size = (loff_t)get_capacity(disk) << 9;
1350 bdev_size = i_size_read(bdev->bd_inode);
1351 if (disk_size != bdev_size) {
1354 "%s: detected capacity change from %lld to %lld\n",
1355 disk->disk_name, bdev_size, disk_size);
1357 i_size_write(bdev->bd_inode, disk_size);
1358 if (bdev_size > disk_size)
1359 flush_disk(bdev, false);
1364 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1365 * @disk: struct gendisk to be revalidated
1367 * This routine is a wrapper for lower-level driver's revalidate_disk
1368 * call-backs. It is used to do common pre and post operations needed
1369 * for all revalidate_disk operations.
1371 int revalidate_disk(struct gendisk *disk)
1373 struct block_device *bdev;
1376 if (disk->fops->revalidate_disk)
1377 ret = disk->fops->revalidate_disk(disk);
1378 bdev = bdget_disk(disk, 0);
1382 mutex_lock(&bdev->bd_mutex);
1383 check_disk_size_change(disk, bdev, ret == 0);
1384 bdev->bd_invalidated = 0;
1385 mutex_unlock(&bdev->bd_mutex);
1389 EXPORT_SYMBOL(revalidate_disk);
1392 * This routine checks whether a removable media has been changed,
1393 * and invalidates all buffer-cache-entries in that case. This
1394 * is a relatively slow routine, so we have to try to minimize using
1395 * it. Thus it is called only upon a 'mount' or 'open'. This
1396 * is the best way of combining speed and utility, I think.
1397 * People changing diskettes in the middle of an operation deserve
1400 int check_disk_change(struct block_device *bdev)
1402 struct gendisk *disk = bdev->bd_disk;
1403 const struct block_device_operations *bdops = disk->fops;
1404 unsigned int events;
1406 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1407 DISK_EVENT_EJECT_REQUEST);
1408 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1411 flush_disk(bdev, true);
1412 if (bdops->revalidate_disk)
1413 bdops->revalidate_disk(bdev->bd_disk);
1417 EXPORT_SYMBOL(check_disk_change);
1419 void bd_set_size(struct block_device *bdev, loff_t size)
1421 inode_lock(bdev->bd_inode);
1422 i_size_write(bdev->bd_inode, size);
1423 inode_unlock(bdev->bd_inode);
1425 EXPORT_SYMBOL(bd_set_size);
1427 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1429 static void bdev_disk_changed(struct block_device *bdev, bool invalidate)
1431 if (disk_part_scan_enabled(bdev->bd_disk)) {
1433 invalidate_partitions(bdev->bd_disk, bdev);
1435 rescan_partitions(bdev->bd_disk, bdev);
1437 check_disk_size_change(bdev->bd_disk, bdev, !invalidate);
1438 bdev->bd_invalidated = 0;
1445 * mutex_lock(part->bd_mutex)
1446 * mutex_lock_nested(whole->bd_mutex, 1)
1449 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1451 struct gendisk *disk;
1455 bool first_open = false;
1457 if (mode & FMODE_READ)
1459 if (mode & FMODE_WRITE)
1462 * hooks: /n/, see "layering violations".
1465 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1475 disk = bdev_get_gendisk(bdev, &partno);
1479 disk_block_events(disk);
1480 mutex_lock_nested(&bdev->bd_mutex, for_part);
1481 if (!bdev->bd_openers) {
1483 bdev->bd_disk = disk;
1484 bdev->bd_queue = disk->queue;
1485 bdev->bd_contains = bdev;
1486 bdev->bd_partno = partno;
1490 bdev->bd_part = disk_get_part(disk, partno);
1495 if (disk->fops->open) {
1496 ret = disk->fops->open(bdev, mode);
1497 if (ret == -ERESTARTSYS) {
1498 /* Lost a race with 'disk' being
1499 * deleted, try again.
1502 disk_put_part(bdev->bd_part);
1503 bdev->bd_part = NULL;
1504 bdev->bd_disk = NULL;
1505 bdev->bd_queue = NULL;
1506 mutex_unlock(&bdev->bd_mutex);
1507 disk_unblock_events(disk);
1508 put_disk_and_module(disk);
1514 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1515 set_init_blocksize(bdev);
1519 * If the device is invalidated, rescan partition
1520 * if open succeeded or failed with -ENOMEDIUM.
1521 * The latter is necessary to prevent ghost
1522 * partitions on a removed medium.
1524 if (bdev->bd_invalidated &&
1525 (!ret || ret == -ENOMEDIUM))
1526 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1531 struct block_device *whole;
1532 whole = bdget_disk(disk, 0);
1537 ret = __blkdev_get(whole, mode, 1);
1540 bdev->bd_contains = whole;
1541 bdev->bd_part = disk_get_part(disk, partno);
1542 if (!(disk->flags & GENHD_FL_UP) ||
1543 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1547 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1548 set_init_blocksize(bdev);
1551 if (bdev->bd_bdi == &noop_backing_dev_info)
1552 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1554 if (bdev->bd_contains == bdev) {
1556 if (bdev->bd_disk->fops->open)
1557 ret = bdev->bd_disk->fops->open(bdev, mode);
1558 /* the same as first opener case, read comment there */
1559 if (bdev->bd_invalidated &&
1560 (!ret || ret == -ENOMEDIUM))
1561 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1563 goto out_unlock_bdev;
1568 bdev->bd_part_count++;
1569 mutex_unlock(&bdev->bd_mutex);
1570 disk_unblock_events(disk);
1571 /* only one opener holds refs to the module and disk */
1573 put_disk_and_module(disk);
1577 disk_put_part(bdev->bd_part);
1578 bdev->bd_disk = NULL;
1579 bdev->bd_part = NULL;
1580 bdev->bd_queue = NULL;
1581 if (bdev != bdev->bd_contains)
1582 __blkdev_put(bdev->bd_contains, mode, 1);
1583 bdev->bd_contains = NULL;
1585 mutex_unlock(&bdev->bd_mutex);
1586 disk_unblock_events(disk);
1587 put_disk_and_module(disk);
1595 * blkdev_get - open a block device
1596 * @bdev: block_device to open
1597 * @mode: FMODE_* mask
1598 * @holder: exclusive holder identifier
1600 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1601 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1602 * @holder is invalid. Exclusive opens may nest for the same @holder.
1604 * On success, the reference count of @bdev is unchanged. On failure,
1611 * 0 on success, -errno on failure.
1613 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1615 struct block_device *whole = NULL;
1618 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1620 if ((mode & FMODE_EXCL) && holder) {
1621 whole = bd_start_claiming(bdev, holder);
1622 if (IS_ERR(whole)) {
1624 return PTR_ERR(whole);
1628 res = __blkdev_get(bdev, mode, 0);
1631 struct gendisk *disk = whole->bd_disk;
1633 /* finish claiming */
1634 mutex_lock(&bdev->bd_mutex);
1635 spin_lock(&bdev_lock);
1638 BUG_ON(!bd_may_claim(bdev, whole, holder));
1640 * Note that for a whole device bd_holders
1641 * will be incremented twice, and bd_holder
1642 * will be set to bd_may_claim before being
1645 whole->bd_holders++;
1646 whole->bd_holder = bd_may_claim;
1648 bdev->bd_holder = holder;
1651 /* tell others that we're done */
1652 BUG_ON(whole->bd_claiming != holder);
1653 whole->bd_claiming = NULL;
1654 wake_up_bit(&whole->bd_claiming, 0);
1656 spin_unlock(&bdev_lock);
1659 * Block event polling for write claims if requested. Any
1660 * write holder makes the write_holder state stick until
1661 * all are released. This is good enough and tracking
1662 * individual writeable reference is too fragile given the
1663 * way @mode is used in blkdev_get/put().
1665 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1666 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1667 bdev->bd_write_holder = true;
1668 disk_block_events(disk);
1671 mutex_unlock(&bdev->bd_mutex);
1677 EXPORT_SYMBOL(blkdev_get);
1680 * blkdev_get_by_path - open a block device by name
1681 * @path: path to the block device to open
1682 * @mode: FMODE_* mask
1683 * @holder: exclusive holder identifier
1685 * Open the blockdevice described by the device file at @path. @mode
1686 * and @holder are identical to blkdev_get().
1688 * On success, the returned block_device has reference count of one.
1694 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1696 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1699 struct block_device *bdev;
1702 bdev = lookup_bdev(path);
1706 err = blkdev_get(bdev, mode, holder);
1708 return ERR_PTR(err);
1710 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1711 blkdev_put(bdev, mode);
1712 return ERR_PTR(-EACCES);
1717 EXPORT_SYMBOL(blkdev_get_by_path);
1720 * blkdev_get_by_dev - open a block device by device number
1721 * @dev: device number of block device to open
1722 * @mode: FMODE_* mask
1723 * @holder: exclusive holder identifier
1725 * Open the blockdevice described by device number @dev. @mode and
1726 * @holder are identical to blkdev_get().
1728 * Use it ONLY if you really do not have anything better - i.e. when
1729 * you are behind a truly sucky interface and all you are given is a
1730 * device number. _Never_ to be used for internal purposes. If you
1731 * ever need it - reconsider your API.
1733 * On success, the returned block_device has reference count of one.
1739 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1741 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1743 struct block_device *bdev;
1748 return ERR_PTR(-ENOMEM);
1750 err = blkdev_get(bdev, mode, holder);
1752 return ERR_PTR(err);
1756 EXPORT_SYMBOL(blkdev_get_by_dev);
1758 static int blkdev_open(struct inode * inode, struct file * filp)
1760 struct block_device *bdev;
1763 * Preserve backwards compatibility and allow large file access
1764 * even if userspace doesn't ask for it explicitly. Some mkfs
1765 * binary needs it. We might want to drop this workaround
1766 * during an unstable branch.
1768 filp->f_flags |= O_LARGEFILE;
1770 filp->f_mode |= FMODE_NOWAIT;
1772 if (filp->f_flags & O_NDELAY)
1773 filp->f_mode |= FMODE_NDELAY;
1774 if (filp->f_flags & O_EXCL)
1775 filp->f_mode |= FMODE_EXCL;
1776 if ((filp->f_flags & O_ACCMODE) == 3)
1777 filp->f_mode |= FMODE_WRITE_IOCTL;
1779 bdev = bd_acquire(inode);
1783 filp->f_mapping = bdev->bd_inode->i_mapping;
1784 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1786 return blkdev_get(bdev, filp->f_mode, filp);
1789 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1791 struct gendisk *disk = bdev->bd_disk;
1792 struct block_device *victim = NULL;
1794 mutex_lock_nested(&bdev->bd_mutex, for_part);
1796 bdev->bd_part_count--;
1798 if (!--bdev->bd_openers) {
1799 WARN_ON_ONCE(bdev->bd_holders);
1800 sync_blockdev(bdev);
1803 bdev_write_inode(bdev);
1805 if (bdev->bd_contains == bdev) {
1806 if (disk->fops->release)
1807 disk->fops->release(disk, mode);
1809 if (!bdev->bd_openers) {
1810 disk_put_part(bdev->bd_part);
1811 bdev->bd_part = NULL;
1812 bdev->bd_disk = NULL;
1813 if (bdev != bdev->bd_contains)
1814 victim = bdev->bd_contains;
1815 bdev->bd_contains = NULL;
1817 put_disk_and_module(disk);
1819 mutex_unlock(&bdev->bd_mutex);
1822 __blkdev_put(victim, mode, 1);
1825 void blkdev_put(struct block_device *bdev, fmode_t mode)
1827 mutex_lock(&bdev->bd_mutex);
1829 if (mode & FMODE_EXCL) {
1833 * Release a claim on the device. The holder fields
1834 * are protected with bdev_lock. bd_mutex is to
1835 * synchronize disk_holder unlinking.
1837 spin_lock(&bdev_lock);
1839 WARN_ON_ONCE(--bdev->bd_holders < 0);
1840 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1842 /* bd_contains might point to self, check in a separate step */
1843 if ((bdev_free = !bdev->bd_holders))
1844 bdev->bd_holder = NULL;
1845 if (!bdev->bd_contains->bd_holders)
1846 bdev->bd_contains->bd_holder = NULL;
1848 spin_unlock(&bdev_lock);
1851 * If this was the last claim, remove holder link and
1852 * unblock evpoll if it was a write holder.
1854 if (bdev_free && bdev->bd_write_holder) {
1855 disk_unblock_events(bdev->bd_disk);
1856 bdev->bd_write_holder = false;
1861 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1862 * event. This is to ensure detection of media removal commanded
1863 * from userland - e.g. eject(1).
1865 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1867 mutex_unlock(&bdev->bd_mutex);
1869 __blkdev_put(bdev, mode, 0);
1871 EXPORT_SYMBOL(blkdev_put);
1873 static int blkdev_close(struct inode * inode, struct file * filp)
1875 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1876 blkdev_put(bdev, filp->f_mode);
1880 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1882 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1883 fmode_t mode = file->f_mode;
1886 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1887 * to updated it before every ioctl.
1889 if (file->f_flags & O_NDELAY)
1890 mode |= FMODE_NDELAY;
1892 mode &= ~FMODE_NDELAY;
1894 return blkdev_ioctl(bdev, mode, cmd, arg);
1898 * Write data to the block device. Only intended for the block device itself
1899 * and the raw driver which basically is a fake block device.
1901 * Does not take i_mutex for the write and thus is not for general purpose
1904 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1906 struct file *file = iocb->ki_filp;
1907 struct inode *bd_inode = bdev_file_inode(file);
1908 loff_t size = i_size_read(bd_inode);
1909 struct blk_plug plug;
1912 if (bdev_read_only(I_BDEV(bd_inode)))
1915 if (!iov_iter_count(from))
1918 if (iocb->ki_pos >= size)
1921 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1924 iov_iter_truncate(from, size - iocb->ki_pos);
1926 blk_start_plug(&plug);
1927 ret = __generic_file_write_iter(iocb, from);
1929 ret = generic_write_sync(iocb, ret);
1930 blk_finish_plug(&plug);
1933 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1935 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1937 struct file *file = iocb->ki_filp;
1938 struct inode *bd_inode = bdev_file_inode(file);
1939 loff_t size = i_size_read(bd_inode);
1940 loff_t pos = iocb->ki_pos;
1946 iov_iter_truncate(to, size);
1947 return generic_file_read_iter(iocb, to);
1949 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1952 * Try to release a page associated with block device when the system
1953 * is under memory pressure.
1955 static int blkdev_releasepage(struct page *page, gfp_t wait)
1957 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1959 if (super && super->s_op->bdev_try_to_free_page)
1960 return super->s_op->bdev_try_to_free_page(super, page, wait);
1962 return try_to_free_buffers(page);
1965 static int blkdev_writepages(struct address_space *mapping,
1966 struct writeback_control *wbc)
1968 return generic_writepages(mapping, wbc);
1971 static const struct address_space_operations def_blk_aops = {
1972 .readpage = blkdev_readpage,
1973 .readpages = blkdev_readpages,
1974 .writepage = blkdev_writepage,
1975 .write_begin = blkdev_write_begin,
1976 .write_end = blkdev_write_end,
1977 .writepages = blkdev_writepages,
1978 .releasepage = blkdev_releasepage,
1979 .direct_IO = blkdev_direct_IO,
1980 .is_dirty_writeback = buffer_check_dirty_writeback,
1983 #define BLKDEV_FALLOC_FL_SUPPORTED \
1984 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1985 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1987 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1990 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1991 struct address_space *mapping;
1992 loff_t end = start + len - 1;
1996 /* Fail if we don't recognize the flags. */
1997 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2000 /* Don't go off the end of the device. */
2001 isize = i_size_read(bdev->bd_inode);
2005 if (mode & FALLOC_FL_KEEP_SIZE) {
2006 len = isize - start;
2007 end = start + len - 1;
2013 * Don't allow IO that isn't aligned to logical block size.
2015 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2018 /* Invalidate the page cache, including dirty pages. */
2019 mapping = bdev->bd_inode->i_mapping;
2020 truncate_inode_pages_range(mapping, start, end);
2023 case FALLOC_FL_ZERO_RANGE:
2024 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2025 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2026 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2028 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2029 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2030 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2032 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2033 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2043 * Invalidate again; if someone wandered in and dirtied a page,
2044 * the caller will be given -EBUSY. The third argument is
2045 * inclusive, so the rounding here is safe.
2047 return invalidate_inode_pages2_range(mapping,
2048 start >> PAGE_SHIFT,
2052 const struct file_operations def_blk_fops = {
2053 .open = blkdev_open,
2054 .release = blkdev_close,
2055 .llseek = block_llseek,
2056 .read_iter = blkdev_read_iter,
2057 .write_iter = blkdev_write_iter,
2058 .mmap = generic_file_mmap,
2059 .fsync = blkdev_fsync,
2060 .unlocked_ioctl = block_ioctl,
2061 #ifdef CONFIG_COMPAT
2062 .compat_ioctl = compat_blkdev_ioctl,
2064 .splice_read = generic_file_splice_read,
2065 .splice_write = iter_file_splice_write,
2066 .fallocate = blkdev_fallocate,
2069 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2072 mm_segment_t old_fs = get_fs();
2074 res = blkdev_ioctl(bdev, 0, cmd, arg);
2079 EXPORT_SYMBOL(ioctl_by_bdev);
2082 * lookup_bdev - lookup a struct block_device by name
2083 * @pathname: special file representing the block device
2085 * Get a reference to the blockdevice at @pathname in the current
2086 * namespace if possible and return it. Return ERR_PTR(error)
2089 struct block_device *lookup_bdev(const char *pathname)
2091 struct block_device *bdev;
2092 struct inode *inode;
2096 if (!pathname || !*pathname)
2097 return ERR_PTR(-EINVAL);
2099 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2101 return ERR_PTR(error);
2103 inode = d_backing_inode(path.dentry);
2105 if (!S_ISBLK(inode->i_mode))
2108 if (!may_open_dev(&path))
2111 bdev = bd_acquire(inode);
2118 bdev = ERR_PTR(error);
2121 EXPORT_SYMBOL(lookup_bdev);
2123 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2125 struct super_block *sb = get_super(bdev);
2130 * no need to lock the super, get_super holds the
2131 * read mutex so the filesystem cannot go away
2132 * under us (->put_super runs with the write lock
2135 shrink_dcache_sb(sb);
2136 res = invalidate_inodes(sb, kill_dirty);
2139 invalidate_bdev(bdev);
2142 EXPORT_SYMBOL(__invalidate_device);
2144 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2146 struct inode *inode, *old_inode = NULL;
2148 spin_lock(&blockdev_superblock->s_inode_list_lock);
2149 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2150 struct address_space *mapping = inode->i_mapping;
2151 struct block_device *bdev;
2153 spin_lock(&inode->i_lock);
2154 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2155 mapping->nrpages == 0) {
2156 spin_unlock(&inode->i_lock);
2160 spin_unlock(&inode->i_lock);
2161 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2163 * We hold a reference to 'inode' so it couldn't have been
2164 * removed from s_inodes list while we dropped the
2165 * s_inode_list_lock We cannot iput the inode now as we can
2166 * be holding the last reference and we cannot iput it under
2167 * s_inode_list_lock. So we keep the reference and iput it
2172 bdev = I_BDEV(inode);
2174 mutex_lock(&bdev->bd_mutex);
2175 if (bdev->bd_openers)
2177 mutex_unlock(&bdev->bd_mutex);
2179 spin_lock(&blockdev_superblock->s_inode_list_lock);
2181 spin_unlock(&blockdev_superblock->s_inode_list_lock);