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 void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
65 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
69 static void bdev_write_inode(struct block_device *bdev)
71 struct inode *inode = bdev->bd_inode;
74 spin_lock(&inode->i_lock);
75 while (inode->i_state & I_DIRTY) {
76 spin_unlock(&inode->i_lock);
77 ret = write_inode_now(inode, true);
79 char name[BDEVNAME_SIZE];
80 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
81 "for block device %s (err=%d).\n",
82 bdevname(bdev, name), ret);
84 spin_lock(&inode->i_lock);
86 spin_unlock(&inode->i_lock);
89 /* Kill _all_ buffers and pagecache , dirty or not.. */
90 void kill_bdev(struct block_device *bdev)
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
94 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
98 truncate_inode_pages(mapping, 0);
100 EXPORT_SYMBOL(kill_bdev);
102 /* Invalidate clean unused buffers and pagecache. */
103 void invalidate_bdev(struct block_device *bdev)
105 struct address_space *mapping = bdev->bd_inode->i_mapping;
107 if (mapping->nrpages) {
108 invalidate_bh_lrus();
109 lru_add_drain_all(); /* make sure all lru add caches are flushed */
110 invalidate_mapping_pages(mapping, 0, -1);
112 /* 99% of the time, we don't need to flush the cleancache on the bdev.
113 * But, for the strange corners, lets be cautious
115 cleancache_invalidate_inode(mapping);
117 EXPORT_SYMBOL(invalidate_bdev);
119 static void set_init_blocksize(struct block_device *bdev)
121 unsigned bsize = bdev_logical_block_size(bdev);
122 loff_t size = i_size_read(bdev->bd_inode);
124 while (bsize < PAGE_SIZE) {
129 bdev->bd_block_size = bsize;
130 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
133 int set_blocksize(struct block_device *bdev, int size)
135 /* Size must be a power of two, and between 512 and PAGE_SIZE */
136 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
139 /* Size cannot be smaller than the size supported by the device */
140 if (size < bdev_logical_block_size(bdev))
143 /* Don't change the size if it is same as current */
144 if (bdev->bd_block_size != size) {
146 bdev->bd_block_size = size;
147 bdev->bd_inode->i_blkbits = blksize_bits(size);
153 EXPORT_SYMBOL(set_blocksize);
155 int sb_set_blocksize(struct super_block *sb, int size)
157 if (set_blocksize(sb->s_bdev, size))
159 /* If we get here, we know size is power of two
160 * and it's value is between 512 and PAGE_SIZE */
161 sb->s_blocksize = size;
162 sb->s_blocksize_bits = blksize_bits(size);
163 return sb->s_blocksize;
166 EXPORT_SYMBOL(sb_set_blocksize);
168 int sb_min_blocksize(struct super_block *sb, int size)
170 int minsize = bdev_logical_block_size(sb->s_bdev);
173 return sb_set_blocksize(sb, size);
176 EXPORT_SYMBOL(sb_min_blocksize);
179 blkdev_get_block(struct inode *inode, sector_t iblock,
180 struct buffer_head *bh, int create)
182 bh->b_bdev = I_BDEV(inode);
183 bh->b_blocknr = iblock;
184 set_buffer_mapped(bh);
188 static struct inode *bdev_file_inode(struct file *file)
190 return file->f_mapping->host;
193 static unsigned int dio_bio_write_op(struct kiocb *iocb)
195 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
197 /* avoid the need for a I/O completion work item */
198 if (iocb->ki_flags & IOCB_DSYNC)
203 #define DIO_INLINE_BIO_VECS 4
205 static void blkdev_bio_end_io_simple(struct bio *bio)
207 struct task_struct *waiter = bio->bi_private;
209 WRITE_ONCE(bio->bi_private, NULL);
210 wake_up_process(waiter);
214 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
217 struct file *file = iocb->ki_filp;
218 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
219 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
220 loff_t pos = iocb->ki_pos;
221 bool should_dirty = false;
227 if ((pos | iov_iter_alignment(iter)) &
228 (bdev_logical_block_size(bdev) - 1))
231 if (nr_pages <= DIO_INLINE_BIO_VECS)
234 vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
239 bio_init(&bio, vecs, nr_pages);
240 bio_set_dev(&bio, bdev);
241 bio.bi_iter.bi_sector = pos >> 9;
242 bio.bi_write_hint = iocb->ki_hint;
243 bio.bi_private = current;
244 bio.bi_end_io = blkdev_bio_end_io_simple;
246 ret = bio_iov_iter_get_pages(&bio, iter);
249 ret = bio.bi_iter.bi_size;
251 if (iov_iter_rw(iter) == READ) {
252 bio.bi_opf = REQ_OP_READ;
253 if (iter_is_iovec(iter))
256 bio.bi_opf = dio_bio_write_op(iocb);
257 task_io_account_write(ret);
260 qc = submit_bio(&bio);
262 set_current_state(TASK_UNINTERRUPTIBLE);
263 if (!READ_ONCE(bio.bi_private))
265 if (!(iocb->ki_flags & IOCB_HIPRI) ||
266 !blk_mq_poll(bdev_get_queue(bdev), qc))
269 __set_current_state(TASK_RUNNING);
271 bio_for_each_segment_all(bvec, &bio, i) {
272 if (should_dirty && !PageCompound(bvec->bv_page))
273 set_page_dirty_lock(bvec->bv_page);
274 put_page(bvec->bv_page);
277 if (unlikely(bio.bi_status))
278 ret = blk_status_to_errno(bio.bi_status);
281 if (vecs != inline_vecs)
292 struct task_struct *waiter;
297 bool should_dirty : 1;
302 static struct bio_set *blkdev_dio_pool __read_mostly;
304 static void blkdev_bio_end_io(struct bio *bio)
306 struct blkdev_dio *dio = bio->bi_private;
307 bool should_dirty = dio->should_dirty;
309 if (bio->bi_status && !dio->bio.bi_status)
310 dio->bio.bi_status = bio->bi_status;
312 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
314 struct kiocb *iocb = dio->iocb;
317 if (likely(!dio->bio.bi_status)) {
321 ret = blk_status_to_errno(dio->bio.bi_status);
324 dio->iocb->ki_complete(iocb, ret, 0);
327 struct task_struct *waiter = dio->waiter;
329 WRITE_ONCE(dio->waiter, NULL);
330 wake_up_process(waiter);
335 bio_check_pages_dirty(bio);
337 struct bio_vec *bvec;
340 bio_for_each_segment_all(bvec, bio, i)
341 put_page(bvec->bv_page);
347 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
349 struct file *file = iocb->ki_filp;
350 struct inode *inode = bdev_file_inode(file);
351 struct block_device *bdev = I_BDEV(inode);
352 struct blk_plug plug;
353 struct blkdev_dio *dio;
355 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
356 loff_t pos = iocb->ki_pos;
357 blk_qc_t qc = BLK_QC_T_NONE;
360 if ((pos | iov_iter_alignment(iter)) &
361 (bdev_logical_block_size(bdev) - 1))
364 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
365 bio_get(bio); /* extra ref for the completion handler */
367 dio = container_of(bio, struct blkdev_dio, bio);
368 dio->is_sync = is_sync = is_sync_kiocb(iocb);
370 dio->waiter = current;
375 dio->multi_bio = false;
376 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
378 blk_start_plug(&plug);
380 bio_set_dev(bio, bdev);
381 bio->bi_iter.bi_sector = pos >> 9;
382 bio->bi_write_hint = iocb->ki_hint;
383 bio->bi_private = dio;
384 bio->bi_end_io = blkdev_bio_end_io;
386 ret = bio_iov_iter_get_pages(bio, iter);
388 bio->bi_status = BLK_STS_IOERR;
394 bio->bi_opf = REQ_OP_READ;
395 if (dio->should_dirty)
396 bio_set_pages_dirty(bio);
398 bio->bi_opf = dio_bio_write_op(iocb);
399 task_io_account_write(bio->bi_iter.bi_size);
402 dio->size += bio->bi_iter.bi_size;
403 pos += bio->bi_iter.bi_size;
405 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
407 qc = submit_bio(bio);
411 if (!dio->multi_bio) {
412 dio->multi_bio = true;
413 atomic_set(&dio->ref, 2);
415 atomic_inc(&dio->ref);
419 bio = bio_alloc(GFP_KERNEL, nr_pages);
421 blk_finish_plug(&plug);
427 set_current_state(TASK_UNINTERRUPTIBLE);
428 if (!READ_ONCE(dio->waiter))
431 if (!(iocb->ki_flags & IOCB_HIPRI) ||
432 !blk_mq_poll(bdev_get_queue(bdev), qc))
435 __set_current_state(TASK_RUNNING);
438 ret = blk_status_to_errno(dio->bio.bi_status);
447 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
451 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
454 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
455 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
457 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
460 static __init int blkdev_init(void)
462 blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
463 if (!blkdev_dio_pool)
467 module_init(blkdev_init);
469 int __sync_blockdev(struct block_device *bdev, int wait)
474 return filemap_flush(bdev->bd_inode->i_mapping);
475 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
479 * Write out and wait upon all the dirty data associated with a block
480 * device via its mapping. Does not take the superblock lock.
482 int sync_blockdev(struct block_device *bdev)
484 return __sync_blockdev(bdev, 1);
486 EXPORT_SYMBOL(sync_blockdev);
489 * Write out and wait upon all dirty data associated with this
490 * device. Filesystem data as well as the underlying block
491 * device. Takes the superblock lock.
493 int fsync_bdev(struct block_device *bdev)
495 struct super_block *sb = get_super(bdev);
497 int res = sync_filesystem(sb);
501 return sync_blockdev(bdev);
503 EXPORT_SYMBOL(fsync_bdev);
506 * freeze_bdev -- lock a filesystem and force it into a consistent state
507 * @bdev: blockdevice to lock
509 * If a superblock is found on this device, we take the s_umount semaphore
510 * on it to make sure nobody unmounts until the snapshot creation is done.
511 * The reference counter (bd_fsfreeze_count) guarantees that only the last
512 * unfreeze process can unfreeze the frozen filesystem actually when multiple
513 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
514 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
517 struct super_block *freeze_bdev(struct block_device *bdev)
519 struct super_block *sb;
522 mutex_lock(&bdev->bd_fsfreeze_mutex);
523 if (++bdev->bd_fsfreeze_count > 1) {
525 * We don't even need to grab a reference - the first call
526 * to freeze_bdev grab an active reference and only the last
527 * thaw_bdev drops it.
529 sb = get_super(bdev);
532 mutex_unlock(&bdev->bd_fsfreeze_mutex);
536 sb = get_active_super(bdev);
539 if (sb->s_op->freeze_super)
540 error = sb->s_op->freeze_super(sb);
542 error = freeze_super(sb);
544 deactivate_super(sb);
545 bdev->bd_fsfreeze_count--;
546 mutex_unlock(&bdev->bd_fsfreeze_mutex);
547 return ERR_PTR(error);
549 deactivate_super(sb);
552 mutex_unlock(&bdev->bd_fsfreeze_mutex);
553 return sb; /* thaw_bdev releases s->s_umount */
555 EXPORT_SYMBOL(freeze_bdev);
558 * thaw_bdev -- unlock filesystem
559 * @bdev: blockdevice to unlock
560 * @sb: associated superblock
562 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
564 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
568 mutex_lock(&bdev->bd_fsfreeze_mutex);
569 if (!bdev->bd_fsfreeze_count)
573 if (--bdev->bd_fsfreeze_count > 0)
579 if (sb->s_op->thaw_super)
580 error = sb->s_op->thaw_super(sb);
582 error = thaw_super(sb);
584 bdev->bd_fsfreeze_count++;
586 mutex_unlock(&bdev->bd_fsfreeze_mutex);
589 EXPORT_SYMBOL(thaw_bdev);
591 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
593 return block_write_full_page(page, blkdev_get_block, wbc);
596 static int blkdev_readpage(struct file * file, struct page * page)
598 return block_read_full_page(page, blkdev_get_block);
601 static int blkdev_readpages(struct file *file, struct address_space *mapping,
602 struct list_head *pages, unsigned nr_pages)
604 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
607 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
608 loff_t pos, unsigned len, unsigned flags,
609 struct page **pagep, void **fsdata)
611 return block_write_begin(mapping, pos, len, flags, pagep,
615 static int blkdev_write_end(struct file *file, struct address_space *mapping,
616 loff_t pos, unsigned len, unsigned copied,
617 struct page *page, void *fsdata)
620 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
630 * for a block special file file_inode(file)->i_size is zero
631 * so we compute the size by hand (just as in block_read/write above)
633 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
635 struct inode *bd_inode = bdev_file_inode(file);
638 inode_lock(bd_inode);
639 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
640 inode_unlock(bd_inode);
644 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
646 struct inode *bd_inode = bdev_file_inode(filp);
647 struct block_device *bdev = I_BDEV(bd_inode);
650 error = file_write_and_wait_range(filp, start, end);
655 * There is no need to serialise calls to blkdev_issue_flush with
656 * i_mutex and doing so causes performance issues with concurrent
657 * O_SYNC writers to a block device.
659 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
660 if (error == -EOPNOTSUPP)
665 EXPORT_SYMBOL(blkdev_fsync);
668 * bdev_read_page() - Start reading a page from a block device
669 * @bdev: The device to read the page from
670 * @sector: The offset on the device to read the page to (need not be aligned)
671 * @page: The page to read
673 * On entry, the page should be locked. It will be unlocked when the page
674 * has been read. If the block driver implements rw_page synchronously,
675 * that will be true on exit from this function, but it need not be.
677 * Errors returned by this function are usually "soft", eg out of memory, or
678 * queue full; callers should try a different route to read this page rather
679 * than propagate an error back up the stack.
681 * Return: negative errno if an error occurs, 0 if submission was successful.
683 int bdev_read_page(struct block_device *bdev, sector_t sector,
686 const struct block_device_operations *ops = bdev->bd_disk->fops;
687 int result = -EOPNOTSUPP;
689 if (!ops->rw_page || bdev_get_integrity(bdev))
692 result = blk_queue_enter(bdev->bd_queue, false);
695 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
696 blk_queue_exit(bdev->bd_queue);
699 EXPORT_SYMBOL_GPL(bdev_read_page);
702 * bdev_write_page() - Start writing a page to a block device
703 * @bdev: The device to write the page to
704 * @sector: The offset on the device to write the page to (need not be aligned)
705 * @page: The page to write
706 * @wbc: The writeback_control for the write
708 * On entry, the page should be locked and not currently under writeback.
709 * On exit, if the write started successfully, the page will be unlocked and
710 * under writeback. If the write failed already (eg the driver failed to
711 * queue the page to the device), the page will still be locked. If the
712 * caller is a ->writepage implementation, it will need to unlock the page.
714 * Errors returned by this function are usually "soft", eg out of memory, or
715 * queue full; callers should try a different route to write this page rather
716 * than propagate an error back up the stack.
718 * Return: negative errno if an error occurs, 0 if submission was successful.
720 int bdev_write_page(struct block_device *bdev, sector_t sector,
721 struct page *page, struct writeback_control *wbc)
724 const struct block_device_operations *ops = bdev->bd_disk->fops;
726 if (!ops->rw_page || bdev_get_integrity(bdev))
728 result = blk_queue_enter(bdev->bd_queue, false);
732 set_page_writeback(page);
733 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
735 end_page_writeback(page);
737 clean_page_buffers(page);
740 blk_queue_exit(bdev->bd_queue);
743 EXPORT_SYMBOL_GPL(bdev_write_page);
749 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
750 static struct kmem_cache * bdev_cachep __read_mostly;
752 static struct inode *bdev_alloc_inode(struct super_block *sb)
754 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
757 return &ei->vfs_inode;
760 static void bdev_i_callback(struct rcu_head *head)
762 struct inode *inode = container_of(head, struct inode, i_rcu);
763 struct bdev_inode *bdi = BDEV_I(inode);
765 kmem_cache_free(bdev_cachep, bdi);
768 static void bdev_destroy_inode(struct inode *inode)
770 call_rcu(&inode->i_rcu, bdev_i_callback);
773 static void init_once(void *foo)
775 struct bdev_inode *ei = (struct bdev_inode *) foo;
776 struct block_device *bdev = &ei->bdev;
778 memset(bdev, 0, sizeof(*bdev));
779 mutex_init(&bdev->bd_mutex);
780 INIT_LIST_HEAD(&bdev->bd_list);
782 INIT_LIST_HEAD(&bdev->bd_holder_disks);
784 bdev->bd_bdi = &noop_backing_dev_info;
785 inode_init_once(&ei->vfs_inode);
786 /* Initialize mutex for freeze. */
787 mutex_init(&bdev->bd_fsfreeze_mutex);
790 static void bdev_evict_inode(struct inode *inode)
792 struct block_device *bdev = &BDEV_I(inode)->bdev;
793 truncate_inode_pages_final(&inode->i_data);
794 invalidate_inode_buffers(inode); /* is it needed here? */
796 spin_lock(&bdev_lock);
797 list_del_init(&bdev->bd_list);
798 spin_unlock(&bdev_lock);
799 /* Detach inode from wb early as bdi_put() may free bdi->wb */
800 inode_detach_wb(inode);
801 if (bdev->bd_bdi != &noop_backing_dev_info) {
802 bdi_put(bdev->bd_bdi);
803 bdev->bd_bdi = &noop_backing_dev_info;
807 static const struct super_operations bdev_sops = {
808 .statfs = simple_statfs,
809 .alloc_inode = bdev_alloc_inode,
810 .destroy_inode = bdev_destroy_inode,
811 .drop_inode = generic_delete_inode,
812 .evict_inode = bdev_evict_inode,
815 static struct dentry *bd_mount(struct file_system_type *fs_type,
816 int flags, const char *dev_name, void *data)
819 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
821 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
825 static struct file_system_type bd_type = {
828 .kill_sb = kill_anon_super,
831 struct super_block *blockdev_superblock __read_mostly;
832 EXPORT_SYMBOL_GPL(blockdev_superblock);
834 void __init bdev_cache_init(void)
837 static struct vfsmount *bd_mnt;
839 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
840 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
841 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
843 err = register_filesystem(&bd_type);
845 panic("Cannot register bdev pseudo-fs");
846 bd_mnt = kern_mount(&bd_type);
848 panic("Cannot create bdev pseudo-fs");
849 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
853 * Most likely _very_ bad one - but then it's hardly critical for small
854 * /dev and can be fixed when somebody will need really large one.
855 * Keep in mind that it will be fed through icache hash function too.
857 static inline unsigned long hash(dev_t dev)
859 return MAJOR(dev)+MINOR(dev);
862 static int bdev_test(struct inode *inode, void *data)
864 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
867 static int bdev_set(struct inode *inode, void *data)
869 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
873 static LIST_HEAD(all_bdevs);
876 * If there is a bdev inode for this device, unhash it so that it gets evicted
877 * as soon as last inode reference is dropped.
879 void bdev_unhash_inode(dev_t dev)
883 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
885 remove_inode_hash(inode);
890 struct block_device *bdget(dev_t dev)
892 struct block_device *bdev;
895 inode = iget5_locked(blockdev_superblock, hash(dev),
896 bdev_test, bdev_set, &dev);
901 bdev = &BDEV_I(inode)->bdev;
903 if (inode->i_state & I_NEW) {
904 bdev->bd_contains = NULL;
905 bdev->bd_super = NULL;
906 bdev->bd_inode = inode;
907 bdev->bd_block_size = i_blocksize(inode);
908 bdev->bd_part_count = 0;
909 bdev->bd_invalidated = 0;
910 inode->i_mode = S_IFBLK;
912 inode->i_bdev = bdev;
913 inode->i_data.a_ops = &def_blk_aops;
914 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
915 spin_lock(&bdev_lock);
916 list_add(&bdev->bd_list, &all_bdevs);
917 spin_unlock(&bdev_lock);
918 unlock_new_inode(inode);
923 EXPORT_SYMBOL(bdget);
926 * bdgrab -- Grab a reference to an already referenced block device
927 * @bdev: Block device to grab a reference to.
929 struct block_device *bdgrab(struct block_device *bdev)
931 ihold(bdev->bd_inode);
934 EXPORT_SYMBOL(bdgrab);
936 long nr_blockdev_pages(void)
938 struct block_device *bdev;
940 spin_lock(&bdev_lock);
941 list_for_each_entry(bdev, &all_bdevs, bd_list) {
942 ret += bdev->bd_inode->i_mapping->nrpages;
944 spin_unlock(&bdev_lock);
948 void bdput(struct block_device *bdev)
950 iput(bdev->bd_inode);
953 EXPORT_SYMBOL(bdput);
955 static struct block_device *bd_acquire(struct inode *inode)
957 struct block_device *bdev;
959 spin_lock(&bdev_lock);
960 bdev = inode->i_bdev;
961 if (bdev && !inode_unhashed(bdev->bd_inode)) {
963 spin_unlock(&bdev_lock);
966 spin_unlock(&bdev_lock);
969 * i_bdev references block device inode that was already shut down
970 * (corresponding device got removed). Remove the reference and look
971 * up block device inode again just in case new device got
972 * reestablished under the same device number.
977 bdev = bdget(inode->i_rdev);
979 spin_lock(&bdev_lock);
980 if (!inode->i_bdev) {
982 * We take an additional reference to bd_inode,
983 * and it's released in clear_inode() of inode.
984 * So, we can access it via ->i_mapping always
988 inode->i_bdev = bdev;
989 inode->i_mapping = bdev->bd_inode->i_mapping;
991 spin_unlock(&bdev_lock);
996 /* Call when you free inode */
998 void bd_forget(struct inode *inode)
1000 struct block_device *bdev = NULL;
1002 spin_lock(&bdev_lock);
1003 if (!sb_is_blkdev_sb(inode->i_sb))
1004 bdev = inode->i_bdev;
1005 inode->i_bdev = NULL;
1006 inode->i_mapping = &inode->i_data;
1007 spin_unlock(&bdev_lock);
1014 * bd_may_claim - test whether a block device can be claimed
1015 * @bdev: block device of interest
1016 * @whole: whole block device containing @bdev, may equal @bdev
1017 * @holder: holder trying to claim @bdev
1019 * Test whether @bdev can be claimed by @holder.
1022 * spin_lock(&bdev_lock).
1025 * %true if @bdev can be claimed, %false otherwise.
1027 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1030 if (bdev->bd_holder == holder)
1031 return true; /* already a holder */
1032 else if (bdev->bd_holder != NULL)
1033 return false; /* held by someone else */
1034 else if (whole == bdev)
1035 return true; /* is a whole device which isn't held */
1037 else if (whole->bd_holder == bd_may_claim)
1038 return true; /* is a partition of a device that is being partitioned */
1039 else if (whole->bd_holder != NULL)
1040 return false; /* is a partition of a held device */
1042 return true; /* is a partition of an un-held device */
1046 * bd_prepare_to_claim - prepare to claim a block device
1047 * @bdev: block device of interest
1048 * @whole: the whole device containing @bdev, may equal @bdev
1049 * @holder: holder trying to claim @bdev
1051 * Prepare to claim @bdev. This function fails if @bdev is already
1052 * claimed by another holder and waits if another claiming is in
1053 * progress. This function doesn't actually claim. On successful
1054 * return, the caller has ownership of bd_claiming and bd_holder[s].
1057 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1058 * it multiple times.
1061 * 0 if @bdev can be claimed, -EBUSY otherwise.
1063 static int bd_prepare_to_claim(struct block_device *bdev,
1064 struct block_device *whole, void *holder)
1067 /* if someone else claimed, fail */
1068 if (!bd_may_claim(bdev, whole, holder))
1071 /* if claiming is already in progress, wait for it to finish */
1072 if (whole->bd_claiming) {
1073 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1076 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1077 spin_unlock(&bdev_lock);
1079 finish_wait(wq, &wait);
1080 spin_lock(&bdev_lock);
1089 * bd_start_claiming - start claiming a block device
1090 * @bdev: block device of interest
1091 * @holder: holder trying to claim @bdev
1093 * @bdev is about to be opened exclusively. Check @bdev can be opened
1094 * exclusively and mark that an exclusive open is in progress. Each
1095 * successful call to this function must be matched with a call to
1096 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1099 * This function is used to gain exclusive access to the block device
1100 * without actually causing other exclusive open attempts to fail. It
1101 * should be used when the open sequence itself requires exclusive
1102 * access but may subsequently fail.
1108 * Pointer to the block device containing @bdev on success, ERR_PTR()
1111 static struct block_device *bd_start_claiming(struct block_device *bdev,
1114 struct gendisk *disk;
1115 struct block_device *whole;
1121 * @bdev might not have been initialized properly yet, look up
1122 * and grab the outer block device the hard way.
1124 disk = get_gendisk(bdev->bd_dev, &partno);
1126 return ERR_PTR(-ENXIO);
1129 * Normally, @bdev should equal what's returned from bdget_disk()
1130 * if partno is 0; however, some drivers (floppy) use multiple
1131 * bdev's for the same physical device and @bdev may be one of the
1132 * aliases. Keep @bdev if partno is 0. This means claimer
1133 * tracking is broken for those devices but it has always been that
1137 whole = bdget_disk(disk, 0);
1139 whole = bdgrab(bdev);
1141 module_put(disk->fops->owner);
1144 return ERR_PTR(-ENOMEM);
1146 /* prepare to claim, if successful, mark claiming in progress */
1147 spin_lock(&bdev_lock);
1149 err = bd_prepare_to_claim(bdev, whole, holder);
1151 whole->bd_claiming = holder;
1152 spin_unlock(&bdev_lock);
1155 spin_unlock(&bdev_lock);
1157 return ERR_PTR(err);
1162 struct bd_holder_disk {
1163 struct list_head list;
1164 struct gendisk *disk;
1168 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1169 struct gendisk *disk)
1171 struct bd_holder_disk *holder;
1173 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1174 if (holder->disk == disk)
1179 static int add_symlink(struct kobject *from, struct kobject *to)
1181 return sysfs_create_link(from, to, kobject_name(to));
1184 static void del_symlink(struct kobject *from, struct kobject *to)
1186 sysfs_remove_link(from, kobject_name(to));
1190 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1191 * @bdev: the claimed slave bdev
1192 * @disk: the holding disk
1194 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1196 * This functions creates the following sysfs symlinks.
1198 * - from "slaves" directory of the holder @disk to the claimed @bdev
1199 * - from "holders" directory of the @bdev to the holder @disk
1201 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1202 * passed to bd_link_disk_holder(), then:
1204 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1205 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1207 * The caller must have claimed @bdev before calling this function and
1208 * ensure that both @bdev and @disk are valid during the creation and
1209 * lifetime of these symlinks.
1215 * 0 on success, -errno on failure.
1217 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1219 struct bd_holder_disk *holder;
1222 mutex_lock(&bdev->bd_mutex);
1224 WARN_ON_ONCE(!bdev->bd_holder);
1226 /* FIXME: remove the following once add_disk() handles errors */
1227 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1230 holder = bd_find_holder_disk(bdev, disk);
1236 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1242 INIT_LIST_HEAD(&holder->list);
1243 holder->disk = disk;
1246 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1250 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1254 * bdev could be deleted beneath us which would implicitly destroy
1255 * the holder directory. Hold on to it.
1257 kobject_get(bdev->bd_part->holder_dir);
1259 list_add(&holder->list, &bdev->bd_holder_disks);
1263 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1267 mutex_unlock(&bdev->bd_mutex);
1270 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1273 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1274 * @bdev: the calimed slave bdev
1275 * @disk: the holding disk
1277 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1282 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1284 struct bd_holder_disk *holder;
1286 mutex_lock(&bdev->bd_mutex);
1288 holder = bd_find_holder_disk(bdev, disk);
1290 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1291 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1292 del_symlink(bdev->bd_part->holder_dir,
1293 &disk_to_dev(disk)->kobj);
1294 kobject_put(bdev->bd_part->holder_dir);
1295 list_del_init(&holder->list);
1299 mutex_unlock(&bdev->bd_mutex);
1301 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1305 * flush_disk - invalidates all buffer-cache entries on a disk
1307 * @bdev: struct block device to be flushed
1308 * @kill_dirty: flag to guide handling of dirty inodes
1310 * Invalidates all buffer-cache entries on a disk. It should be called
1311 * when a disk has been changed -- either by a media change or online
1314 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1316 if (__invalidate_device(bdev, kill_dirty)) {
1317 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1318 "resized disk %s\n",
1319 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1324 if (disk_part_scan_enabled(bdev->bd_disk))
1325 bdev->bd_invalidated = 1;
1329 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1330 * @disk: struct gendisk to check
1331 * @bdev: struct bdev to adjust.
1333 * This routine checks to see if the bdev size does not match the disk size
1334 * and adjusts it if it differs.
1336 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1338 loff_t disk_size, bdev_size;
1340 disk_size = (loff_t)get_capacity(disk) << 9;
1341 bdev_size = i_size_read(bdev->bd_inode);
1342 if (disk_size != bdev_size) {
1344 "%s: detected capacity change from %lld to %lld\n",
1345 disk->disk_name, bdev_size, disk_size);
1346 i_size_write(bdev->bd_inode, disk_size);
1347 flush_disk(bdev, false);
1350 EXPORT_SYMBOL(check_disk_size_change);
1353 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1354 * @disk: struct gendisk to be revalidated
1356 * This routine is a wrapper for lower-level driver's revalidate_disk
1357 * call-backs. It is used to do common pre and post operations needed
1358 * for all revalidate_disk operations.
1360 int revalidate_disk(struct gendisk *disk)
1362 struct block_device *bdev;
1365 if (disk->fops->revalidate_disk)
1366 ret = disk->fops->revalidate_disk(disk);
1367 bdev = bdget_disk(disk, 0);
1371 mutex_lock(&bdev->bd_mutex);
1372 check_disk_size_change(disk, bdev);
1373 bdev->bd_invalidated = 0;
1374 mutex_unlock(&bdev->bd_mutex);
1378 EXPORT_SYMBOL(revalidate_disk);
1381 * This routine checks whether a removable media has been changed,
1382 * and invalidates all buffer-cache-entries in that case. This
1383 * is a relatively slow routine, so we have to try to minimize using
1384 * it. Thus it is called only upon a 'mount' or 'open'. This
1385 * is the best way of combining speed and utility, I think.
1386 * People changing diskettes in the middle of an operation deserve
1389 int check_disk_change(struct block_device *bdev)
1391 struct gendisk *disk = bdev->bd_disk;
1392 const struct block_device_operations *bdops = disk->fops;
1393 unsigned int events;
1395 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1396 DISK_EVENT_EJECT_REQUEST);
1397 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1400 flush_disk(bdev, true);
1401 if (bdops->revalidate_disk)
1402 bdops->revalidate_disk(bdev->bd_disk);
1406 EXPORT_SYMBOL(check_disk_change);
1408 void bd_set_size(struct block_device *bdev, loff_t size)
1410 inode_lock(bdev->bd_inode);
1411 i_size_write(bdev->bd_inode, size);
1412 inode_unlock(bdev->bd_inode);
1414 EXPORT_SYMBOL(bd_set_size);
1416 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1421 * mutex_lock(part->bd_mutex)
1422 * mutex_lock_nested(whole->bd_mutex, 1)
1425 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1427 struct gendisk *disk;
1428 struct module *owner;
1433 if (mode & FMODE_READ)
1435 if (mode & FMODE_WRITE)
1438 * hooks: /n/, see "layering violations".
1441 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1449 disk = get_gendisk(bdev->bd_dev, &partno);
1452 owner = disk->fops->owner;
1454 disk_block_events(disk);
1455 mutex_lock_nested(&bdev->bd_mutex, for_part);
1456 if (!bdev->bd_openers) {
1457 bdev->bd_disk = disk;
1458 bdev->bd_queue = disk->queue;
1459 bdev->bd_contains = bdev;
1460 bdev->bd_partno = partno;
1464 bdev->bd_part = disk_get_part(disk, partno);
1469 if (disk->fops->open) {
1470 ret = disk->fops->open(bdev, mode);
1471 if (ret == -ERESTARTSYS) {
1472 /* Lost a race with 'disk' being
1473 * deleted, try again.
1476 disk_put_part(bdev->bd_part);
1477 bdev->bd_part = NULL;
1478 bdev->bd_disk = NULL;
1479 bdev->bd_queue = NULL;
1480 mutex_unlock(&bdev->bd_mutex);
1481 disk_unblock_events(disk);
1489 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1490 set_init_blocksize(bdev);
1494 * If the device is invalidated, rescan partition
1495 * if open succeeded or failed with -ENOMEDIUM.
1496 * The latter is necessary to prevent ghost
1497 * partitions on a removed medium.
1499 if (bdev->bd_invalidated) {
1501 rescan_partitions(disk, bdev);
1502 else if (ret == -ENOMEDIUM)
1503 invalidate_partitions(disk, bdev);
1509 struct block_device *whole;
1510 whole = bdget_disk(disk, 0);
1515 ret = __blkdev_get(whole, mode, 1);
1520 bdev->bd_contains = whole;
1521 bdev->bd_part = disk_get_part(disk, partno);
1522 if (!(disk->flags & GENHD_FL_UP) ||
1523 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1527 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1528 set_init_blocksize(bdev);
1531 if (bdev->bd_bdi == &noop_backing_dev_info)
1532 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1534 if (bdev->bd_contains == bdev) {
1536 if (bdev->bd_disk->fops->open)
1537 ret = bdev->bd_disk->fops->open(bdev, mode);
1538 /* the same as first opener case, read comment there */
1539 if (bdev->bd_invalidated) {
1541 rescan_partitions(bdev->bd_disk, bdev);
1542 else if (ret == -ENOMEDIUM)
1543 invalidate_partitions(bdev->bd_disk, bdev);
1546 goto out_unlock_bdev;
1548 /* only one opener holds refs to the module and disk */
1554 bdev->bd_part_count++;
1555 mutex_unlock(&bdev->bd_mutex);
1556 disk_unblock_events(disk);
1560 disk_put_part(bdev->bd_part);
1561 bdev->bd_disk = NULL;
1562 bdev->bd_part = NULL;
1563 bdev->bd_queue = NULL;
1564 if (bdev != bdev->bd_contains)
1565 __blkdev_put(bdev->bd_contains, mode, 1);
1566 bdev->bd_contains = NULL;
1568 mutex_unlock(&bdev->bd_mutex);
1569 disk_unblock_events(disk);
1578 * blkdev_get - open a block device
1579 * @bdev: block_device to open
1580 * @mode: FMODE_* mask
1581 * @holder: exclusive holder identifier
1583 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1584 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1585 * @holder is invalid. Exclusive opens may nest for the same @holder.
1587 * On success, the reference count of @bdev is unchanged. On failure,
1594 * 0 on success, -errno on failure.
1596 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1598 struct block_device *whole = NULL;
1601 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1603 if ((mode & FMODE_EXCL) && holder) {
1604 whole = bd_start_claiming(bdev, holder);
1605 if (IS_ERR(whole)) {
1607 return PTR_ERR(whole);
1611 res = __blkdev_get(bdev, mode, 0);
1614 struct gendisk *disk = whole->bd_disk;
1616 /* finish claiming */
1617 mutex_lock(&bdev->bd_mutex);
1618 spin_lock(&bdev_lock);
1621 BUG_ON(!bd_may_claim(bdev, whole, holder));
1623 * Note that for a whole device bd_holders
1624 * will be incremented twice, and bd_holder
1625 * will be set to bd_may_claim before being
1628 whole->bd_holders++;
1629 whole->bd_holder = bd_may_claim;
1631 bdev->bd_holder = holder;
1634 /* tell others that we're done */
1635 BUG_ON(whole->bd_claiming != holder);
1636 whole->bd_claiming = NULL;
1637 wake_up_bit(&whole->bd_claiming, 0);
1639 spin_unlock(&bdev_lock);
1642 * Block event polling for write claims if requested. Any
1643 * write holder makes the write_holder state stick until
1644 * all are released. This is good enough and tracking
1645 * individual writeable reference is too fragile given the
1646 * way @mode is used in blkdev_get/put().
1648 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1649 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1650 bdev->bd_write_holder = true;
1651 disk_block_events(disk);
1654 mutex_unlock(&bdev->bd_mutex);
1663 EXPORT_SYMBOL(blkdev_get);
1666 * blkdev_get_by_path - open a block device by name
1667 * @path: path to the block device to open
1668 * @mode: FMODE_* mask
1669 * @holder: exclusive holder identifier
1671 * Open the blockdevice described by the device file at @path. @mode
1672 * and @holder are identical to blkdev_get().
1674 * On success, the returned block_device has reference count of one.
1680 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1682 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1685 struct block_device *bdev;
1688 bdev = lookup_bdev(path);
1692 err = blkdev_get(bdev, mode, holder);
1694 return ERR_PTR(err);
1696 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1697 blkdev_put(bdev, mode);
1698 return ERR_PTR(-EACCES);
1703 EXPORT_SYMBOL(blkdev_get_by_path);
1706 * blkdev_get_by_dev - open a block device by device number
1707 * @dev: device number of block device to open
1708 * @mode: FMODE_* mask
1709 * @holder: exclusive holder identifier
1711 * Open the blockdevice described by device number @dev. @mode and
1712 * @holder are identical to blkdev_get().
1714 * Use it ONLY if you really do not have anything better - i.e. when
1715 * you are behind a truly sucky interface and all you are given is a
1716 * device number. _Never_ to be used for internal purposes. If you
1717 * ever need it - reconsider your API.
1719 * On success, the returned block_device has reference count of one.
1725 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1727 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1729 struct block_device *bdev;
1734 return ERR_PTR(-ENOMEM);
1736 err = blkdev_get(bdev, mode, holder);
1738 return ERR_PTR(err);
1742 EXPORT_SYMBOL(blkdev_get_by_dev);
1744 static int blkdev_open(struct inode * inode, struct file * filp)
1746 struct block_device *bdev;
1749 * Preserve backwards compatibility and allow large file access
1750 * even if userspace doesn't ask for it explicitly. Some mkfs
1751 * binary needs it. We might want to drop this workaround
1752 * during an unstable branch.
1754 filp->f_flags |= O_LARGEFILE;
1756 filp->f_mode |= FMODE_NOWAIT;
1758 if (filp->f_flags & O_NDELAY)
1759 filp->f_mode |= FMODE_NDELAY;
1760 if (filp->f_flags & O_EXCL)
1761 filp->f_mode |= FMODE_EXCL;
1762 if ((filp->f_flags & O_ACCMODE) == 3)
1763 filp->f_mode |= FMODE_WRITE_IOCTL;
1765 bdev = bd_acquire(inode);
1769 filp->f_mapping = bdev->bd_inode->i_mapping;
1770 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1772 return blkdev_get(bdev, filp->f_mode, filp);
1775 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1777 struct gendisk *disk = bdev->bd_disk;
1778 struct block_device *victim = NULL;
1781 * Sync early if it looks like we're the last one. If someone else
1782 * opens the block device between now and the decrement of bd_openers
1783 * then we did a sync that we didn't need to, but that's not the end
1784 * of the world and we want to avoid long (could be several minute)
1785 * syncs while holding the mutex.
1787 if (bdev->bd_openers == 1)
1788 sync_blockdev(bdev);
1790 mutex_lock_nested(&bdev->bd_mutex, for_part);
1792 bdev->bd_part_count--;
1794 if (!--bdev->bd_openers) {
1795 WARN_ON_ONCE(bdev->bd_holders);
1796 sync_blockdev(bdev);
1799 bdev_write_inode(bdev);
1801 if (bdev->bd_contains == bdev) {
1802 if (disk->fops->release)
1803 disk->fops->release(disk, mode);
1805 if (!bdev->bd_openers) {
1806 struct module *owner = disk->fops->owner;
1808 disk_put_part(bdev->bd_part);
1809 bdev->bd_part = NULL;
1810 bdev->bd_disk = NULL;
1811 if (bdev != bdev->bd_contains)
1812 victim = bdev->bd_contains;
1813 bdev->bd_contains = NULL;
1818 mutex_unlock(&bdev->bd_mutex);
1821 __blkdev_put(victim, mode, 1);
1824 void blkdev_put(struct block_device *bdev, fmode_t mode)
1826 mutex_lock(&bdev->bd_mutex);
1828 if (mode & FMODE_EXCL) {
1832 * Release a claim on the device. The holder fields
1833 * are protected with bdev_lock. bd_mutex is to
1834 * synchronize disk_holder unlinking.
1836 spin_lock(&bdev_lock);
1838 WARN_ON_ONCE(--bdev->bd_holders < 0);
1839 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1841 /* bd_contains might point to self, check in a separate step */
1842 if ((bdev_free = !bdev->bd_holders))
1843 bdev->bd_holder = NULL;
1844 if (!bdev->bd_contains->bd_holders)
1845 bdev->bd_contains->bd_holder = NULL;
1847 spin_unlock(&bdev_lock);
1850 * If this was the last claim, remove holder link and
1851 * unblock evpoll if it was a write holder.
1853 if (bdev_free && bdev->bd_write_holder) {
1854 disk_unblock_events(bdev->bd_disk);
1855 bdev->bd_write_holder = false;
1860 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1861 * event. This is to ensure detection of media removal commanded
1862 * from userland - e.g. eject(1).
1864 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1866 mutex_unlock(&bdev->bd_mutex);
1868 __blkdev_put(bdev, mode, 0);
1870 EXPORT_SYMBOL(blkdev_put);
1872 static int blkdev_close(struct inode * inode, struct file * filp)
1874 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1875 blkdev_put(bdev, filp->f_mode);
1879 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1881 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1882 fmode_t mode = file->f_mode;
1885 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1886 * to updated it before every ioctl.
1888 if (file->f_flags & O_NDELAY)
1889 mode |= FMODE_NDELAY;
1891 mode &= ~FMODE_NDELAY;
1893 return blkdev_ioctl(bdev, mode, cmd, arg);
1897 * Write data to the block device. Only intended for the block device itself
1898 * and the raw driver which basically is a fake block device.
1900 * Does not take i_mutex for the write and thus is not for general purpose
1903 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1905 struct file *file = iocb->ki_filp;
1906 struct inode *bd_inode = bdev_file_inode(file);
1907 loff_t size = i_size_read(bd_inode);
1908 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 size -= iocb->ki_pos;
1925 if (iov_iter_count(from) > size) {
1926 shorted = iov_iter_count(from) - size;
1927 iov_iter_truncate(from, size);
1930 blk_start_plug(&plug);
1931 ret = __generic_file_write_iter(iocb, from);
1933 ret = generic_write_sync(iocb, ret);
1934 iov_iter_reexpand(from, iov_iter_count(from) + shorted);
1935 blk_finish_plug(&plug);
1938 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1940 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1942 struct file *file = iocb->ki_filp;
1943 struct inode *bd_inode = bdev_file_inode(file);
1944 loff_t size = i_size_read(bd_inode);
1945 loff_t pos = iocb->ki_pos;
1953 if (iov_iter_count(to) > size) {
1954 shorted = iov_iter_count(to) - size;
1955 iov_iter_truncate(to, size);
1958 ret = generic_file_read_iter(iocb, to);
1959 iov_iter_reexpand(to, iov_iter_count(to) + shorted);
1962 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1965 * Try to release a page associated with block device when the system
1966 * is under memory pressure.
1968 static int blkdev_releasepage(struct page *page, gfp_t wait)
1970 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1972 if (super && super->s_op->bdev_try_to_free_page)
1973 return super->s_op->bdev_try_to_free_page(super, page, wait);
1975 return try_to_free_buffers(page);
1978 static int blkdev_writepages(struct address_space *mapping,
1979 struct writeback_control *wbc)
1981 if (dax_mapping(mapping)) {
1982 struct block_device *bdev = I_BDEV(mapping->host);
1984 return dax_writeback_mapping_range(mapping, bdev, wbc);
1986 return generic_writepages(mapping, wbc);
1989 static const struct address_space_operations def_blk_aops = {
1990 .readpage = blkdev_readpage,
1991 .readpages = blkdev_readpages,
1992 .writepage = blkdev_writepage,
1993 .write_begin = blkdev_write_begin,
1994 .write_end = blkdev_write_end,
1995 .writepages = blkdev_writepages,
1996 .releasepage = blkdev_releasepage,
1997 .direct_IO = blkdev_direct_IO,
1998 .is_dirty_writeback = buffer_check_dirty_writeback,
2001 #define BLKDEV_FALLOC_FL_SUPPORTED \
2002 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2003 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2005 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2008 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2009 struct address_space *mapping;
2010 loff_t end = start + len - 1;
2014 /* Fail if we don't recognize the flags. */
2015 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2018 /* Don't go off the end of the device. */
2019 isize = i_size_read(bdev->bd_inode);
2023 if (mode & FALLOC_FL_KEEP_SIZE) {
2024 len = isize - start;
2025 end = start + len - 1;
2031 * Don't allow IO that isn't aligned to logical block size.
2033 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2036 /* Invalidate the page cache, including dirty pages. */
2037 mapping = bdev->bd_inode->i_mapping;
2038 truncate_inode_pages_range(mapping, start, end);
2041 case FALLOC_FL_ZERO_RANGE:
2042 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2043 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2044 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2046 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2047 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2048 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2050 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2051 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2061 * Invalidate again; if someone wandered in and dirtied a page,
2062 * the caller will be given -EBUSY. The third argument is
2063 * inclusive, so the rounding here is safe.
2065 return invalidate_inode_pages2_range(mapping,
2066 start >> PAGE_SHIFT,
2070 const struct file_operations def_blk_fops = {
2071 .open = blkdev_open,
2072 .release = blkdev_close,
2073 .llseek = block_llseek,
2074 .read_iter = blkdev_read_iter,
2075 .write_iter = blkdev_write_iter,
2076 .mmap = generic_file_mmap,
2077 .fsync = blkdev_fsync,
2078 .unlocked_ioctl = block_ioctl,
2079 #ifdef CONFIG_COMPAT
2080 .compat_ioctl = compat_blkdev_ioctl,
2082 .splice_read = generic_file_splice_read,
2083 .splice_write = iter_file_splice_write,
2084 .fallocate = blkdev_fallocate,
2087 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2090 mm_segment_t old_fs = get_fs();
2092 res = blkdev_ioctl(bdev, 0, cmd, arg);
2097 EXPORT_SYMBOL(ioctl_by_bdev);
2100 * lookup_bdev - lookup a struct block_device by name
2101 * @pathname: special file representing the block device
2103 * Get a reference to the blockdevice at @pathname in the current
2104 * namespace if possible and return it. Return ERR_PTR(error)
2107 struct block_device *lookup_bdev(const char *pathname)
2109 struct block_device *bdev;
2110 struct inode *inode;
2114 if (!pathname || !*pathname)
2115 return ERR_PTR(-EINVAL);
2117 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2119 return ERR_PTR(error);
2121 inode = d_backing_inode(path.dentry);
2123 if (!S_ISBLK(inode->i_mode))
2126 if (!may_open_dev(&path))
2129 bdev = bd_acquire(inode);
2136 bdev = ERR_PTR(error);
2139 EXPORT_SYMBOL(lookup_bdev);
2141 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2143 struct super_block *sb = get_super(bdev);
2148 * no need to lock the super, get_super holds the
2149 * read mutex so the filesystem cannot go away
2150 * under us (->put_super runs with the write lock
2153 shrink_dcache_sb(sb);
2154 res = invalidate_inodes(sb, kill_dirty);
2157 invalidate_bdev(bdev);
2160 EXPORT_SYMBOL(__invalidate_device);
2162 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2164 struct inode *inode, *old_inode = NULL;
2166 spin_lock(&blockdev_superblock->s_inode_list_lock);
2167 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2168 struct address_space *mapping = inode->i_mapping;
2169 struct block_device *bdev;
2171 spin_lock(&inode->i_lock);
2172 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2173 mapping->nrpages == 0) {
2174 spin_unlock(&inode->i_lock);
2178 spin_unlock(&inode->i_lock);
2179 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2181 * We hold a reference to 'inode' so it couldn't have been
2182 * removed from s_inodes list while we dropped the
2183 * s_inode_list_lock We cannot iput the inode now as we can
2184 * be holding the last reference and we cannot iput it under
2185 * s_inode_list_lock. So we keep the reference and iput it
2190 bdev = I_BDEV(inode);
2192 mutex_lock(&bdev->bd_mutex);
2193 if (bdev->bd_openers)
2195 mutex_unlock(&bdev->bd_mutex);
2197 spin_lock(&blockdev_superblock->s_inode_list_lock);
2199 spin_unlock(&blockdev_superblock->s_inode_list_lock);