4 * Copyright (C) 1991, 1992 Linus Torvalds
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 int set_blocksize(struct block_device *bdev, int size)
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
113 /* Size cannot be smaller than the size supported by the device */
114 if (size < bdev_logical_block_size(bdev))
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
127 EXPORT_SYMBOL(set_blocksize);
129 int sb_set_blocksize(struct super_block *sb, int size)
131 if (set_blocksize(sb->s_bdev, size))
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
136 sb->s_blocksize_bits = blksize_bits(size);
137 return sb->s_blocksize;
140 EXPORT_SYMBOL(sb_set_blocksize);
142 int sb_min_blocksize(struct super_block *sb, int size)
144 int minsize = bdev_logical_block_size(sb->s_bdev);
147 return sb_set_blocksize(sb, size);
150 EXPORT_SYMBOL(sb_min_blocksize);
153 blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
162 static struct inode *bdev_file_inode(struct file *file)
164 return file->f_mapping->host;
167 static unsigned int dio_bio_write_op(struct kiocb *iocb)
169 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
171 /* avoid the need for a I/O completion work item */
172 if (iocb->ki_flags & IOCB_DSYNC)
177 #define DIO_INLINE_BIO_VECS 4
179 static void blkdev_bio_end_io_simple(struct bio *bio)
181 struct task_struct *waiter = bio->bi_private;
183 WRITE_ONCE(bio->bi_private, NULL);
184 blk_wake_io_task(waiter);
188 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
191 struct file *file = iocb->ki_filp;
192 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
193 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
194 loff_t pos = iocb->ki_pos;
195 bool should_dirty = false;
201 if ((pos | iov_iter_alignment(iter)) &
202 (bdev_logical_block_size(bdev) - 1))
205 if (nr_pages <= DIO_INLINE_BIO_VECS)
208 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
214 bio_init(&bio, vecs, nr_pages);
215 bio_set_dev(&bio, bdev);
216 bio.bi_iter.bi_sector = pos >> 9;
217 bio.bi_write_hint = iocb->ki_hint;
218 bio.bi_private = current;
219 bio.bi_end_io = blkdev_bio_end_io_simple;
220 bio.bi_ioprio = iocb->ki_ioprio;
222 ret = bio_iov_iter_get_pages(&bio, iter);
225 ret = bio.bi_iter.bi_size;
227 if (iov_iter_rw(iter) == READ) {
228 bio.bi_opf = REQ_OP_READ;
229 if (iter_is_iovec(iter))
232 bio.bi_opf = dio_bio_write_op(iocb);
233 task_io_account_write(ret);
235 if (iocb->ki_flags & IOCB_HIPRI)
236 bio.bi_opf |= REQ_HIPRI;
238 qc = submit_bio(&bio);
240 set_current_state(TASK_UNINTERRUPTIBLE);
241 if (!READ_ONCE(bio.bi_private))
243 if (!(iocb->ki_flags & IOCB_HIPRI) ||
244 !blk_poll(bdev_get_queue(bdev), qc, true))
247 __set_current_state(TASK_RUNNING);
249 bio_for_each_segment_all(bvec, &bio, i) {
250 if (should_dirty && !PageCompound(bvec->bv_page))
251 set_page_dirty_lock(bvec->bv_page);
252 put_page(bvec->bv_page);
255 if (unlikely(bio.bi_status))
256 ret = blk_status_to_errno(bio.bi_status);
259 if (vecs != inline_vecs)
270 struct task_struct *waiter;
275 bool should_dirty : 1;
280 static struct bio_set blkdev_dio_pool;
282 static void blkdev_bio_end_io(struct bio *bio)
284 struct blkdev_dio *dio = bio->bi_private;
285 bool should_dirty = dio->should_dirty;
287 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
288 if (bio->bi_status && !dio->bio.bi_status)
289 dio->bio.bi_status = bio->bi_status;
292 struct kiocb *iocb = dio->iocb;
295 if (likely(!dio->bio.bi_status)) {
299 ret = blk_status_to_errno(dio->bio.bi_status);
302 dio->iocb->ki_complete(iocb, ret, 0);
306 struct task_struct *waiter = dio->waiter;
308 WRITE_ONCE(dio->waiter, NULL);
309 blk_wake_io_task(waiter);
314 bio_check_pages_dirty(bio);
316 struct bio_vec *bvec;
319 bio_for_each_segment_all(bvec, bio, i)
320 put_page(bvec->bv_page);
326 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
328 struct file *file = iocb->ki_filp;
329 struct inode *inode = bdev_file_inode(file);
330 struct block_device *bdev = I_BDEV(inode);
331 struct blk_plug plug;
332 struct blkdev_dio *dio;
334 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
335 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
336 loff_t pos = iocb->ki_pos;
337 blk_qc_t qc = BLK_QC_T_NONE;
340 if ((pos | iov_iter_alignment(iter)) &
341 (bdev_logical_block_size(bdev) - 1))
344 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
346 dio = container_of(bio, struct blkdev_dio, bio);
347 dio->is_sync = is_sync = is_sync_kiocb(iocb);
349 dio->waiter = current;
356 dio->multi_bio = false;
357 dio->should_dirty = is_read && iter_is_iovec(iter);
360 * Don't plug for HIPRI/polled IO, as those should go straight
364 blk_start_plug(&plug);
367 bio_set_dev(bio, bdev);
368 bio->bi_iter.bi_sector = pos >> 9;
369 bio->bi_write_hint = iocb->ki_hint;
370 bio->bi_private = dio;
371 bio->bi_end_io = blkdev_bio_end_io;
372 bio->bi_ioprio = iocb->ki_ioprio;
374 ret = bio_iov_iter_get_pages(bio, iter);
376 bio->bi_status = BLK_STS_IOERR;
382 bio->bi_opf = REQ_OP_READ;
383 if (dio->should_dirty)
384 bio_set_pages_dirty(bio);
386 bio->bi_opf = dio_bio_write_op(iocb);
387 task_io_account_write(bio->bi_iter.bi_size);
390 dio->size += bio->bi_iter.bi_size;
391 pos += bio->bi_iter.bi_size;
393 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
395 if (iocb->ki_flags & IOCB_HIPRI)
396 bio->bi_opf |= REQ_HIPRI;
398 qc = submit_bio(bio);
402 if (!dio->multi_bio) {
404 * AIO needs an extra reference to ensure the dio
405 * structure which is embedded into the first bio
410 dio->multi_bio = true;
411 atomic_set(&dio->ref, 2);
413 atomic_inc(&dio->ref);
417 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_poll(bdev_get_queue(bdev), qc, true))
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 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
464 module_init(blkdev_init);
466 int __sync_blockdev(struct block_device *bdev, int wait)
471 return filemap_flush(bdev->bd_inode->i_mapping);
472 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
476 * Write out and wait upon all the dirty data associated with a block
477 * device via its mapping. Does not take the superblock lock.
479 int sync_blockdev(struct block_device *bdev)
481 return __sync_blockdev(bdev, 1);
483 EXPORT_SYMBOL(sync_blockdev);
486 * Write out and wait upon all dirty data associated with this
487 * device. Filesystem data as well as the underlying block
488 * device. Takes the superblock lock.
490 int fsync_bdev(struct block_device *bdev)
492 struct super_block *sb = get_super(bdev);
494 int res = sync_filesystem(sb);
498 return sync_blockdev(bdev);
500 EXPORT_SYMBOL(fsync_bdev);
503 * freeze_bdev -- lock a filesystem and force it into a consistent state
504 * @bdev: blockdevice to lock
506 * If a superblock is found on this device, we take the s_umount semaphore
507 * on it to make sure nobody unmounts until the snapshot creation is done.
508 * The reference counter (bd_fsfreeze_count) guarantees that only the last
509 * unfreeze process can unfreeze the frozen filesystem actually when multiple
510 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
511 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
514 struct super_block *freeze_bdev(struct block_device *bdev)
516 struct super_block *sb;
519 mutex_lock(&bdev->bd_fsfreeze_mutex);
520 if (++bdev->bd_fsfreeze_count > 1) {
522 * We don't even need to grab a reference - the first call
523 * to freeze_bdev grab an active reference and only the last
524 * thaw_bdev drops it.
526 sb = get_super(bdev);
529 mutex_unlock(&bdev->bd_fsfreeze_mutex);
533 sb = get_active_super(bdev);
536 if (sb->s_op->freeze_super)
537 error = sb->s_op->freeze_super(sb);
539 error = freeze_super(sb);
541 deactivate_super(sb);
542 bdev->bd_fsfreeze_count--;
543 mutex_unlock(&bdev->bd_fsfreeze_mutex);
544 return ERR_PTR(error);
546 deactivate_super(sb);
549 mutex_unlock(&bdev->bd_fsfreeze_mutex);
550 return sb; /* thaw_bdev releases s->s_umount */
552 EXPORT_SYMBOL(freeze_bdev);
555 * thaw_bdev -- unlock filesystem
556 * @bdev: blockdevice to unlock
557 * @sb: associated superblock
559 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
561 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
565 mutex_lock(&bdev->bd_fsfreeze_mutex);
566 if (!bdev->bd_fsfreeze_count)
570 if (--bdev->bd_fsfreeze_count > 0)
576 if (sb->s_op->thaw_super)
577 error = sb->s_op->thaw_super(sb);
579 error = thaw_super(sb);
581 bdev->bd_fsfreeze_count++;
583 mutex_unlock(&bdev->bd_fsfreeze_mutex);
586 EXPORT_SYMBOL(thaw_bdev);
588 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
590 return block_write_full_page(page, blkdev_get_block, wbc);
593 static int blkdev_readpage(struct file * file, struct page * page)
595 return block_read_full_page(page, blkdev_get_block);
598 static int blkdev_readpages(struct file *file, struct address_space *mapping,
599 struct list_head *pages, unsigned nr_pages)
601 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
604 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
605 loff_t pos, unsigned len, unsigned flags,
606 struct page **pagep, void **fsdata)
608 return block_write_begin(mapping, pos, len, flags, pagep,
612 static int blkdev_write_end(struct file *file, struct address_space *mapping,
613 loff_t pos, unsigned len, unsigned copied,
614 struct page *page, void *fsdata)
617 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
627 * for a block special file file_inode(file)->i_size is zero
628 * so we compute the size by hand (just as in block_read/write above)
630 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
632 struct inode *bd_inode = bdev_file_inode(file);
635 inode_lock(bd_inode);
636 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
637 inode_unlock(bd_inode);
641 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
643 struct inode *bd_inode = bdev_file_inode(filp);
644 struct block_device *bdev = I_BDEV(bd_inode);
647 error = file_write_and_wait_range(filp, start, end);
652 * There is no need to serialise calls to blkdev_issue_flush with
653 * i_mutex and doing so causes performance issues with concurrent
654 * O_SYNC writers to a block device.
656 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
657 if (error == -EOPNOTSUPP)
662 EXPORT_SYMBOL(blkdev_fsync);
665 * bdev_read_page() - Start reading a page from a block device
666 * @bdev: The device to read the page from
667 * @sector: The offset on the device to read the page to (need not be aligned)
668 * @page: The page to read
670 * On entry, the page should be locked. It will be unlocked when the page
671 * has been read. If the block driver implements rw_page synchronously,
672 * that will be true on exit from this function, but it need not be.
674 * Errors returned by this function are usually "soft", eg out of memory, or
675 * queue full; callers should try a different route to read this page rather
676 * than propagate an error back up the stack.
678 * Return: negative errno if an error occurs, 0 if submission was successful.
680 int bdev_read_page(struct block_device *bdev, sector_t sector,
683 const struct block_device_operations *ops = bdev->bd_disk->fops;
684 int result = -EOPNOTSUPP;
686 if (!ops->rw_page || bdev_get_integrity(bdev))
689 result = blk_queue_enter(bdev->bd_queue, 0);
692 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
694 blk_queue_exit(bdev->bd_queue);
697 EXPORT_SYMBOL_GPL(bdev_read_page);
700 * bdev_write_page() - Start writing a page to a block device
701 * @bdev: The device to write the page to
702 * @sector: The offset on the device to write the page to (need not be aligned)
703 * @page: The page to write
704 * @wbc: The writeback_control for the write
706 * On entry, the page should be locked and not currently under writeback.
707 * On exit, if the write started successfully, the page will be unlocked and
708 * under writeback. If the write failed already (eg the driver failed to
709 * queue the page to the device), the page will still be locked. If the
710 * caller is a ->writepage implementation, it will need to unlock the page.
712 * Errors returned by this function are usually "soft", eg out of memory, or
713 * queue full; callers should try a different route to write this page rather
714 * than propagate an error back up the stack.
716 * Return: negative errno if an error occurs, 0 if submission was successful.
718 int bdev_write_page(struct block_device *bdev, sector_t sector,
719 struct page *page, struct writeback_control *wbc)
722 const struct block_device_operations *ops = bdev->bd_disk->fops;
724 if (!ops->rw_page || bdev_get_integrity(bdev))
726 result = blk_queue_enter(bdev->bd_queue, 0);
730 set_page_writeback(page);
731 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
734 end_page_writeback(page);
736 clean_page_buffers(page);
739 blk_queue_exit(bdev->bd_queue);
742 EXPORT_SYMBOL_GPL(bdev_write_page);
748 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
749 static struct kmem_cache * bdev_cachep __read_mostly;
751 static struct inode *bdev_alloc_inode(struct super_block *sb)
753 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
756 return &ei->vfs_inode;
759 static void bdev_i_callback(struct rcu_head *head)
761 struct inode *inode = container_of(head, struct inode, i_rcu);
762 struct bdev_inode *bdi = BDEV_I(inode);
764 kmem_cache_free(bdev_cachep, bdi);
767 static void bdev_destroy_inode(struct inode *inode)
769 call_rcu(&inode->i_rcu, bdev_i_callback);
772 static void init_once(void *foo)
774 struct bdev_inode *ei = (struct bdev_inode *) foo;
775 struct block_device *bdev = &ei->bdev;
777 memset(bdev, 0, sizeof(*bdev));
778 mutex_init(&bdev->bd_mutex);
779 INIT_LIST_HEAD(&bdev->bd_list);
781 INIT_LIST_HEAD(&bdev->bd_holder_disks);
783 bdev->bd_bdi = &noop_backing_dev_info;
784 inode_init_once(&ei->vfs_inode);
785 /* Initialize mutex for freeze. */
786 mutex_init(&bdev->bd_fsfreeze_mutex);
789 static void bdev_evict_inode(struct inode *inode)
791 struct block_device *bdev = &BDEV_I(inode)->bdev;
792 truncate_inode_pages_final(&inode->i_data);
793 invalidate_inode_buffers(inode); /* is it needed here? */
795 spin_lock(&bdev_lock);
796 list_del_init(&bdev->bd_list);
797 spin_unlock(&bdev_lock);
798 /* Detach inode from wb early as bdi_put() may free bdi->wb */
799 inode_detach_wb(inode);
800 if (bdev->bd_bdi != &noop_backing_dev_info) {
801 bdi_put(bdev->bd_bdi);
802 bdev->bd_bdi = &noop_backing_dev_info;
806 static const struct super_operations bdev_sops = {
807 .statfs = simple_statfs,
808 .alloc_inode = bdev_alloc_inode,
809 .destroy_inode = bdev_destroy_inode,
810 .drop_inode = generic_delete_inode,
811 .evict_inode = bdev_evict_inode,
814 static struct dentry *bd_mount(struct file_system_type *fs_type,
815 int flags, const char *dev_name, void *data)
818 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
820 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
824 static struct file_system_type bd_type = {
827 .kill_sb = kill_anon_super,
830 struct super_block *blockdev_superblock __read_mostly;
831 EXPORT_SYMBOL_GPL(blockdev_superblock);
833 void __init bdev_cache_init(void)
836 static struct vfsmount *bd_mnt;
838 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
839 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
840 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
842 err = register_filesystem(&bd_type);
844 panic("Cannot register bdev pseudo-fs");
845 bd_mnt = kern_mount(&bd_type);
847 panic("Cannot create bdev pseudo-fs");
848 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
852 * Most likely _very_ bad one - but then it's hardly critical for small
853 * /dev and can be fixed when somebody will need really large one.
854 * Keep in mind that it will be fed through icache hash function too.
856 static inline unsigned long hash(dev_t dev)
858 return MAJOR(dev)+MINOR(dev);
861 static int bdev_test(struct inode *inode, void *data)
863 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
866 static int bdev_set(struct inode *inode, void *data)
868 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
872 static LIST_HEAD(all_bdevs);
875 * If there is a bdev inode for this device, unhash it so that it gets evicted
876 * as soon as last inode reference is dropped.
878 void bdev_unhash_inode(dev_t dev)
882 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
884 remove_inode_hash(inode);
889 struct block_device *bdget(dev_t dev)
891 struct block_device *bdev;
894 inode = iget5_locked(blockdev_superblock, hash(dev),
895 bdev_test, bdev_set, &dev);
900 bdev = &BDEV_I(inode)->bdev;
902 if (inode->i_state & I_NEW) {
903 bdev->bd_contains = NULL;
904 bdev->bd_super = NULL;
905 bdev->bd_inode = inode;
906 bdev->bd_block_size = i_blocksize(inode);
907 bdev->bd_part_count = 0;
908 bdev->bd_invalidated = 0;
909 inode->i_mode = S_IFBLK;
911 inode->i_bdev = bdev;
912 inode->i_data.a_ops = &def_blk_aops;
913 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
914 spin_lock(&bdev_lock);
915 list_add(&bdev->bd_list, &all_bdevs);
916 spin_unlock(&bdev_lock);
917 unlock_new_inode(inode);
922 EXPORT_SYMBOL(bdget);
925 * bdgrab -- Grab a reference to an already referenced block device
926 * @bdev: Block device to grab a reference to.
928 struct block_device *bdgrab(struct block_device *bdev)
930 ihold(bdev->bd_inode);
933 EXPORT_SYMBOL(bdgrab);
935 long nr_blockdev_pages(void)
937 struct block_device *bdev;
939 spin_lock(&bdev_lock);
940 list_for_each_entry(bdev, &all_bdevs, bd_list) {
941 ret += bdev->bd_inode->i_mapping->nrpages;
943 spin_unlock(&bdev_lock);
947 void bdput(struct block_device *bdev)
949 iput(bdev->bd_inode);
952 EXPORT_SYMBOL(bdput);
954 static struct block_device *bd_acquire(struct inode *inode)
956 struct block_device *bdev;
958 spin_lock(&bdev_lock);
959 bdev = inode->i_bdev;
960 if (bdev && !inode_unhashed(bdev->bd_inode)) {
962 spin_unlock(&bdev_lock);
965 spin_unlock(&bdev_lock);
968 * i_bdev references block device inode that was already shut down
969 * (corresponding device got removed). Remove the reference and look
970 * up block device inode again just in case new device got
971 * reestablished under the same device number.
976 bdev = bdget(inode->i_rdev);
978 spin_lock(&bdev_lock);
979 if (!inode->i_bdev) {
981 * We take an additional reference to bd_inode,
982 * and it's released in clear_inode() of inode.
983 * So, we can access it via ->i_mapping always
987 inode->i_bdev = bdev;
988 inode->i_mapping = bdev->bd_inode->i_mapping;
990 spin_unlock(&bdev_lock);
995 /* Call when you free inode */
997 void bd_forget(struct inode *inode)
999 struct block_device *bdev = NULL;
1001 spin_lock(&bdev_lock);
1002 if (!sb_is_blkdev_sb(inode->i_sb))
1003 bdev = inode->i_bdev;
1004 inode->i_bdev = NULL;
1005 inode->i_mapping = &inode->i_data;
1006 spin_unlock(&bdev_lock);
1013 * bd_may_claim - test whether a block device can be claimed
1014 * @bdev: block device of interest
1015 * @whole: whole block device containing @bdev, may equal @bdev
1016 * @holder: holder trying to claim @bdev
1018 * Test whether @bdev can be claimed by @holder.
1021 * spin_lock(&bdev_lock).
1024 * %true if @bdev can be claimed, %false otherwise.
1026 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1029 if (bdev->bd_holder == holder)
1030 return true; /* already a holder */
1031 else if (bdev->bd_holder != NULL)
1032 return false; /* held by someone else */
1033 else if (whole == bdev)
1034 return true; /* is a whole device which isn't held */
1036 else if (whole->bd_holder == bd_may_claim)
1037 return true; /* is a partition of a device that is being partitioned */
1038 else if (whole->bd_holder != NULL)
1039 return false; /* is a partition of a held device */
1041 return true; /* is a partition of an un-held device */
1045 * bd_prepare_to_claim - prepare to claim a block device
1046 * @bdev: block device of interest
1047 * @whole: the whole device containing @bdev, may equal @bdev
1048 * @holder: holder trying to claim @bdev
1050 * Prepare to claim @bdev. This function fails if @bdev is already
1051 * claimed by another holder and waits if another claiming is in
1052 * progress. This function doesn't actually claim. On successful
1053 * return, the caller has ownership of bd_claiming and bd_holder[s].
1056 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1057 * it multiple times.
1060 * 0 if @bdev can be claimed, -EBUSY otherwise.
1062 static int bd_prepare_to_claim(struct block_device *bdev,
1063 struct block_device *whole, void *holder)
1066 /* if someone else claimed, fail */
1067 if (!bd_may_claim(bdev, whole, holder))
1070 /* if claiming is already in progress, wait for it to finish */
1071 if (whole->bd_claiming) {
1072 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1075 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1076 spin_unlock(&bdev_lock);
1078 finish_wait(wq, &wait);
1079 spin_lock(&bdev_lock);
1087 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1089 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1094 * Now that we hold gendisk reference we make sure bdev we looked up is
1095 * not stale. If it is, it means device got removed and created before
1096 * we looked up gendisk and we fail open in such case. Associating
1097 * unhashed bdev with newly created gendisk could lead to two bdevs
1098 * (and thus two independent caches) being associated with one device
1101 if (inode_unhashed(bdev->bd_inode)) {
1102 put_disk_and_module(disk);
1109 * bd_start_claiming - start claiming a block device
1110 * @bdev: block device of interest
1111 * @holder: holder trying to claim @bdev
1113 * @bdev is about to be opened exclusively. Check @bdev can be opened
1114 * exclusively and mark that an exclusive open is in progress. Each
1115 * successful call to this function must be matched with a call to
1116 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1119 * This function is used to gain exclusive access to the block device
1120 * without actually causing other exclusive open attempts to fail. It
1121 * should be used when the open sequence itself requires exclusive
1122 * access but may subsequently fail.
1128 * Pointer to the block device containing @bdev on success, ERR_PTR()
1131 static struct block_device *bd_start_claiming(struct block_device *bdev,
1134 struct gendisk *disk;
1135 struct block_device *whole;
1141 * @bdev might not have been initialized properly yet, look up
1142 * and grab the outer block device the hard way.
1144 disk = bdev_get_gendisk(bdev, &partno);
1146 return ERR_PTR(-ENXIO);
1149 * Normally, @bdev should equal what's returned from bdget_disk()
1150 * if partno is 0; however, some drivers (floppy) use multiple
1151 * bdev's for the same physical device and @bdev may be one of the
1152 * aliases. Keep @bdev if partno is 0. This means claimer
1153 * tracking is broken for those devices but it has always been that
1157 whole = bdget_disk(disk, 0);
1159 whole = bdgrab(bdev);
1161 put_disk_and_module(disk);
1163 return ERR_PTR(-ENOMEM);
1165 /* prepare to claim, if successful, mark claiming in progress */
1166 spin_lock(&bdev_lock);
1168 err = bd_prepare_to_claim(bdev, whole, holder);
1170 whole->bd_claiming = holder;
1171 spin_unlock(&bdev_lock);
1174 spin_unlock(&bdev_lock);
1176 return ERR_PTR(err);
1181 struct bd_holder_disk {
1182 struct list_head list;
1183 struct gendisk *disk;
1187 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1188 struct gendisk *disk)
1190 struct bd_holder_disk *holder;
1192 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1193 if (holder->disk == disk)
1198 static int add_symlink(struct kobject *from, struct kobject *to)
1200 return sysfs_create_link(from, to, kobject_name(to));
1203 static void del_symlink(struct kobject *from, struct kobject *to)
1205 sysfs_remove_link(from, kobject_name(to));
1209 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1210 * @bdev: the claimed slave bdev
1211 * @disk: the holding disk
1213 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1215 * This functions creates the following sysfs symlinks.
1217 * - from "slaves" directory of the holder @disk to the claimed @bdev
1218 * - from "holders" directory of the @bdev to the holder @disk
1220 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1221 * passed to bd_link_disk_holder(), then:
1223 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1224 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1226 * The caller must have claimed @bdev before calling this function and
1227 * ensure that both @bdev and @disk are valid during the creation and
1228 * lifetime of these symlinks.
1234 * 0 on success, -errno on failure.
1236 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1238 struct bd_holder_disk *holder;
1241 mutex_lock(&bdev->bd_mutex);
1243 WARN_ON_ONCE(!bdev->bd_holder);
1245 /* FIXME: remove the following once add_disk() handles errors */
1246 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1249 holder = bd_find_holder_disk(bdev, disk);
1255 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1261 INIT_LIST_HEAD(&holder->list);
1262 holder->disk = disk;
1265 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1269 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1273 * bdev could be deleted beneath us which would implicitly destroy
1274 * the holder directory. Hold on to it.
1276 kobject_get(bdev->bd_part->holder_dir);
1278 list_add(&holder->list, &bdev->bd_holder_disks);
1282 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1286 mutex_unlock(&bdev->bd_mutex);
1289 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1292 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1293 * @bdev: the calimed slave bdev
1294 * @disk: the holding disk
1296 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1301 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1303 struct bd_holder_disk *holder;
1305 mutex_lock(&bdev->bd_mutex);
1307 holder = bd_find_holder_disk(bdev, disk);
1309 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1310 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1311 del_symlink(bdev->bd_part->holder_dir,
1312 &disk_to_dev(disk)->kobj);
1313 kobject_put(bdev->bd_part->holder_dir);
1314 list_del_init(&holder->list);
1318 mutex_unlock(&bdev->bd_mutex);
1320 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1324 * flush_disk - invalidates all buffer-cache entries on a disk
1326 * @bdev: struct block device to be flushed
1327 * @kill_dirty: flag to guide handling of dirty inodes
1329 * Invalidates all buffer-cache entries on a disk. It should be called
1330 * when a disk has been changed -- either by a media change or online
1333 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1335 if (__invalidate_device(bdev, kill_dirty)) {
1336 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1337 "resized disk %s\n",
1338 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1343 if (disk_part_scan_enabled(bdev->bd_disk))
1344 bdev->bd_invalidated = 1;
1348 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1349 * @disk: struct gendisk to check
1350 * @bdev: struct bdev to adjust.
1351 * @verbose: if %true log a message about a size change if there is any
1353 * This routine checks to see if the bdev size does not match the disk size
1354 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1357 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1360 loff_t disk_size, bdev_size;
1362 disk_size = (loff_t)get_capacity(disk) << 9;
1363 bdev_size = i_size_read(bdev->bd_inode);
1364 if (disk_size != bdev_size) {
1367 "%s: detected capacity change from %lld to %lld\n",
1368 disk->disk_name, bdev_size, disk_size);
1370 i_size_write(bdev->bd_inode, disk_size);
1371 if (bdev_size > disk_size)
1372 flush_disk(bdev, false);
1377 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1378 * @disk: struct gendisk to be revalidated
1380 * This routine is a wrapper for lower-level driver's revalidate_disk
1381 * call-backs. It is used to do common pre and post operations needed
1382 * for all revalidate_disk operations.
1384 int revalidate_disk(struct gendisk *disk)
1386 struct block_device *bdev;
1389 if (disk->fops->revalidate_disk)
1390 ret = disk->fops->revalidate_disk(disk);
1391 bdev = bdget_disk(disk, 0);
1395 mutex_lock(&bdev->bd_mutex);
1396 check_disk_size_change(disk, bdev, ret == 0);
1397 bdev->bd_invalidated = 0;
1398 mutex_unlock(&bdev->bd_mutex);
1402 EXPORT_SYMBOL(revalidate_disk);
1405 * This routine checks whether a removable media has been changed,
1406 * and invalidates all buffer-cache-entries in that case. This
1407 * is a relatively slow routine, so we have to try to minimize using
1408 * it. Thus it is called only upon a 'mount' or 'open'. This
1409 * is the best way of combining speed and utility, I think.
1410 * People changing diskettes in the middle of an operation deserve
1413 int check_disk_change(struct block_device *bdev)
1415 struct gendisk *disk = bdev->bd_disk;
1416 const struct block_device_operations *bdops = disk->fops;
1417 unsigned int events;
1419 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1420 DISK_EVENT_EJECT_REQUEST);
1421 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1424 flush_disk(bdev, true);
1425 if (bdops->revalidate_disk)
1426 bdops->revalidate_disk(bdev->bd_disk);
1430 EXPORT_SYMBOL(check_disk_change);
1432 void bd_set_size(struct block_device *bdev, loff_t size)
1434 unsigned bsize = bdev_logical_block_size(bdev);
1436 inode_lock(bdev->bd_inode);
1437 i_size_write(bdev->bd_inode, size);
1438 inode_unlock(bdev->bd_inode);
1439 while (bsize < PAGE_SIZE) {
1444 bdev->bd_block_size = bsize;
1445 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1447 EXPORT_SYMBOL(bd_set_size);
1449 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1454 * mutex_lock(part->bd_mutex)
1455 * mutex_lock_nested(whole->bd_mutex, 1)
1458 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1460 struct gendisk *disk;
1464 bool first_open = false;
1466 if (mode & FMODE_READ)
1468 if (mode & FMODE_WRITE)
1471 * hooks: /n/, see "layering violations".
1474 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1484 disk = bdev_get_gendisk(bdev, &partno);
1488 disk_block_events(disk);
1489 mutex_lock_nested(&bdev->bd_mutex, for_part);
1490 if (!bdev->bd_openers) {
1492 bdev->bd_disk = disk;
1493 bdev->bd_queue = disk->queue;
1494 bdev->bd_contains = bdev;
1495 bdev->bd_partno = partno;
1499 bdev->bd_part = disk_get_part(disk, partno);
1504 if (disk->fops->open) {
1505 ret = disk->fops->open(bdev, mode);
1506 if (ret == -ERESTARTSYS) {
1507 /* Lost a race with 'disk' being
1508 * deleted, try again.
1511 disk_put_part(bdev->bd_part);
1512 bdev->bd_part = NULL;
1513 bdev->bd_disk = NULL;
1514 bdev->bd_queue = NULL;
1515 mutex_unlock(&bdev->bd_mutex);
1516 disk_unblock_events(disk);
1517 put_disk_and_module(disk);
1523 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1526 * If the device is invalidated, rescan partition
1527 * if open succeeded or failed with -ENOMEDIUM.
1528 * The latter is necessary to prevent ghost
1529 * partitions on a removed medium.
1531 if (bdev->bd_invalidated) {
1533 rescan_partitions(disk, bdev);
1534 else if (ret == -ENOMEDIUM)
1535 invalidate_partitions(disk, bdev);
1541 struct block_device *whole;
1542 whole = bdget_disk(disk, 0);
1547 ret = __blkdev_get(whole, mode, 1);
1550 bdev->bd_contains = whole;
1551 bdev->bd_part = disk_get_part(disk, partno);
1552 if (!(disk->flags & GENHD_FL_UP) ||
1553 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1557 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1560 if (bdev->bd_bdi == &noop_backing_dev_info)
1561 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1563 if (bdev->bd_contains == bdev) {
1565 if (bdev->bd_disk->fops->open)
1566 ret = bdev->bd_disk->fops->open(bdev, mode);
1567 /* the same as first opener case, read comment there */
1568 if (bdev->bd_invalidated) {
1570 rescan_partitions(bdev->bd_disk, bdev);
1571 else if (ret == -ENOMEDIUM)
1572 invalidate_partitions(bdev->bd_disk, bdev);
1575 goto out_unlock_bdev;
1580 bdev->bd_part_count++;
1581 mutex_unlock(&bdev->bd_mutex);
1582 disk_unblock_events(disk);
1583 /* only one opener holds refs to the module and disk */
1585 put_disk_and_module(disk);
1589 disk_put_part(bdev->bd_part);
1590 bdev->bd_disk = NULL;
1591 bdev->bd_part = NULL;
1592 bdev->bd_queue = NULL;
1593 if (bdev != bdev->bd_contains)
1594 __blkdev_put(bdev->bd_contains, mode, 1);
1595 bdev->bd_contains = NULL;
1597 mutex_unlock(&bdev->bd_mutex);
1598 disk_unblock_events(disk);
1599 put_disk_and_module(disk);
1607 * blkdev_get - open a block device
1608 * @bdev: block_device to open
1609 * @mode: FMODE_* mask
1610 * @holder: exclusive holder identifier
1612 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1613 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1614 * @holder is invalid. Exclusive opens may nest for the same @holder.
1616 * On success, the reference count of @bdev is unchanged. On failure,
1623 * 0 on success, -errno on failure.
1625 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1627 struct block_device *whole = NULL;
1630 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1632 if ((mode & FMODE_EXCL) && holder) {
1633 whole = bd_start_claiming(bdev, holder);
1634 if (IS_ERR(whole)) {
1636 return PTR_ERR(whole);
1640 res = __blkdev_get(bdev, mode, 0);
1643 struct gendisk *disk = whole->bd_disk;
1645 /* finish claiming */
1646 mutex_lock(&bdev->bd_mutex);
1647 spin_lock(&bdev_lock);
1650 BUG_ON(!bd_may_claim(bdev, whole, holder));
1652 * Note that for a whole device bd_holders
1653 * will be incremented twice, and bd_holder
1654 * will be set to bd_may_claim before being
1657 whole->bd_holders++;
1658 whole->bd_holder = bd_may_claim;
1660 bdev->bd_holder = holder;
1663 /* tell others that we're done */
1664 BUG_ON(whole->bd_claiming != holder);
1665 whole->bd_claiming = NULL;
1666 wake_up_bit(&whole->bd_claiming, 0);
1668 spin_unlock(&bdev_lock);
1671 * Block event polling for write claims if requested. Any
1672 * write holder makes the write_holder state stick until
1673 * all are released. This is good enough and tracking
1674 * individual writeable reference is too fragile given the
1675 * way @mode is used in blkdev_get/put().
1677 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1678 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1679 bdev->bd_write_holder = true;
1680 disk_block_events(disk);
1683 mutex_unlock(&bdev->bd_mutex);
1689 EXPORT_SYMBOL(blkdev_get);
1692 * blkdev_get_by_path - open a block device by name
1693 * @path: path to the block device to open
1694 * @mode: FMODE_* mask
1695 * @holder: exclusive holder identifier
1697 * Open the blockdevice described by the device file at @path. @mode
1698 * and @holder are identical to blkdev_get().
1700 * On success, the returned block_device has reference count of one.
1706 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1708 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1711 struct block_device *bdev;
1714 bdev = lookup_bdev(path);
1718 err = blkdev_get(bdev, mode, holder);
1720 return ERR_PTR(err);
1722 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1723 blkdev_put(bdev, mode);
1724 return ERR_PTR(-EACCES);
1729 EXPORT_SYMBOL(blkdev_get_by_path);
1732 * blkdev_get_by_dev - open a block device by device number
1733 * @dev: device number of block device to open
1734 * @mode: FMODE_* mask
1735 * @holder: exclusive holder identifier
1737 * Open the blockdevice described by device number @dev. @mode and
1738 * @holder are identical to blkdev_get().
1740 * Use it ONLY if you really do not have anything better - i.e. when
1741 * you are behind a truly sucky interface and all you are given is a
1742 * device number. _Never_ to be used for internal purposes. If you
1743 * ever need it - reconsider your API.
1745 * On success, the returned block_device has reference count of one.
1751 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1753 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1755 struct block_device *bdev;
1760 return ERR_PTR(-ENOMEM);
1762 err = blkdev_get(bdev, mode, holder);
1764 return ERR_PTR(err);
1768 EXPORT_SYMBOL(blkdev_get_by_dev);
1770 static int blkdev_open(struct inode * inode, struct file * filp)
1772 struct block_device *bdev;
1775 * Preserve backwards compatibility and allow large file access
1776 * even if userspace doesn't ask for it explicitly. Some mkfs
1777 * binary needs it. We might want to drop this workaround
1778 * during an unstable branch.
1780 filp->f_flags |= O_LARGEFILE;
1782 filp->f_mode |= FMODE_NOWAIT;
1784 if (filp->f_flags & O_NDELAY)
1785 filp->f_mode |= FMODE_NDELAY;
1786 if (filp->f_flags & O_EXCL)
1787 filp->f_mode |= FMODE_EXCL;
1788 if ((filp->f_flags & O_ACCMODE) == 3)
1789 filp->f_mode |= FMODE_WRITE_IOCTL;
1791 bdev = bd_acquire(inode);
1795 filp->f_mapping = bdev->bd_inode->i_mapping;
1796 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1798 return blkdev_get(bdev, filp->f_mode, filp);
1801 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1803 struct gendisk *disk = bdev->bd_disk;
1804 struct block_device *victim = NULL;
1806 mutex_lock_nested(&bdev->bd_mutex, for_part);
1808 bdev->bd_part_count--;
1810 if (!--bdev->bd_openers) {
1811 WARN_ON_ONCE(bdev->bd_holders);
1812 sync_blockdev(bdev);
1815 bdev_write_inode(bdev);
1817 if (bdev->bd_contains == bdev) {
1818 if (disk->fops->release)
1819 disk->fops->release(disk, mode);
1821 if (!bdev->bd_openers) {
1822 disk_put_part(bdev->bd_part);
1823 bdev->bd_part = NULL;
1824 bdev->bd_disk = NULL;
1825 if (bdev != bdev->bd_contains)
1826 victim = bdev->bd_contains;
1827 bdev->bd_contains = NULL;
1829 put_disk_and_module(disk);
1831 mutex_unlock(&bdev->bd_mutex);
1834 __blkdev_put(victim, mode, 1);
1837 void blkdev_put(struct block_device *bdev, fmode_t mode)
1839 mutex_lock(&bdev->bd_mutex);
1841 if (mode & FMODE_EXCL) {
1845 * Release a claim on the device. The holder fields
1846 * are protected with bdev_lock. bd_mutex is to
1847 * synchronize disk_holder unlinking.
1849 spin_lock(&bdev_lock);
1851 WARN_ON_ONCE(--bdev->bd_holders < 0);
1852 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1854 /* bd_contains might point to self, check in a separate step */
1855 if ((bdev_free = !bdev->bd_holders))
1856 bdev->bd_holder = NULL;
1857 if (!bdev->bd_contains->bd_holders)
1858 bdev->bd_contains->bd_holder = NULL;
1860 spin_unlock(&bdev_lock);
1863 * If this was the last claim, remove holder link and
1864 * unblock evpoll if it was a write holder.
1866 if (bdev_free && bdev->bd_write_holder) {
1867 disk_unblock_events(bdev->bd_disk);
1868 bdev->bd_write_holder = false;
1873 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1874 * event. This is to ensure detection of media removal commanded
1875 * from userland - e.g. eject(1).
1877 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1879 mutex_unlock(&bdev->bd_mutex);
1881 __blkdev_put(bdev, mode, 0);
1883 EXPORT_SYMBOL(blkdev_put);
1885 static int blkdev_close(struct inode * inode, struct file * filp)
1887 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1888 blkdev_put(bdev, filp->f_mode);
1892 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1894 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1895 fmode_t mode = file->f_mode;
1898 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1899 * to updated it before every ioctl.
1901 if (file->f_flags & O_NDELAY)
1902 mode |= FMODE_NDELAY;
1904 mode &= ~FMODE_NDELAY;
1906 return blkdev_ioctl(bdev, mode, cmd, arg);
1910 * Write data to the block device. Only intended for the block device itself
1911 * and the raw driver which basically is a fake block device.
1913 * Does not take i_mutex for the write and thus is not for general purpose
1916 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1918 struct file *file = iocb->ki_filp;
1919 struct inode *bd_inode = bdev_file_inode(file);
1920 loff_t size = i_size_read(bd_inode);
1921 struct blk_plug plug;
1924 if (bdev_read_only(I_BDEV(bd_inode)))
1927 if (!iov_iter_count(from))
1930 if (iocb->ki_pos >= size)
1933 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1936 iov_iter_truncate(from, size - iocb->ki_pos);
1938 blk_start_plug(&plug);
1939 ret = __generic_file_write_iter(iocb, from);
1941 ret = generic_write_sync(iocb, ret);
1942 blk_finish_plug(&plug);
1945 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1947 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1949 struct file *file = iocb->ki_filp;
1950 struct inode *bd_inode = bdev_file_inode(file);
1951 loff_t size = i_size_read(bd_inode);
1952 loff_t pos = iocb->ki_pos;
1958 iov_iter_truncate(to, size);
1959 return generic_file_read_iter(iocb, to);
1961 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1964 * Try to release a page associated with block device when the system
1965 * is under memory pressure.
1967 static int blkdev_releasepage(struct page *page, gfp_t wait)
1969 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1971 if (super && super->s_op->bdev_try_to_free_page)
1972 return super->s_op->bdev_try_to_free_page(super, page, wait);
1974 return try_to_free_buffers(page);
1977 static int blkdev_writepages(struct address_space *mapping,
1978 struct writeback_control *wbc)
1980 return generic_writepages(mapping, wbc);
1983 static const struct address_space_operations def_blk_aops = {
1984 .readpage = blkdev_readpage,
1985 .readpages = blkdev_readpages,
1986 .writepage = blkdev_writepage,
1987 .write_begin = blkdev_write_begin,
1988 .write_end = blkdev_write_end,
1989 .writepages = blkdev_writepages,
1990 .releasepage = blkdev_releasepage,
1991 .direct_IO = blkdev_direct_IO,
1992 .migratepage = buffer_migrate_page_norefs,
1993 .is_dirty_writeback = buffer_check_dirty_writeback,
1996 #define BLKDEV_FALLOC_FL_SUPPORTED \
1997 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1998 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2000 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2003 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2004 struct address_space *mapping;
2005 loff_t end = start + len - 1;
2009 /* Fail if we don't recognize the flags. */
2010 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2013 /* Don't go off the end of the device. */
2014 isize = i_size_read(bdev->bd_inode);
2018 if (mode & FALLOC_FL_KEEP_SIZE) {
2019 len = isize - start;
2020 end = start + len - 1;
2026 * Don't allow IO that isn't aligned to logical block size.
2028 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2031 /* Invalidate the page cache, including dirty pages. */
2032 mapping = bdev->bd_inode->i_mapping;
2033 truncate_inode_pages_range(mapping, start, end);
2036 case FALLOC_FL_ZERO_RANGE:
2037 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2038 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2039 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2041 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2042 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2043 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2045 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2046 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2056 * Invalidate again; if someone wandered in and dirtied a page,
2057 * the caller will be given -EBUSY. The third argument is
2058 * inclusive, so the rounding here is safe.
2060 return invalidate_inode_pages2_range(mapping,
2061 start >> PAGE_SHIFT,
2065 const struct file_operations def_blk_fops = {
2066 .open = blkdev_open,
2067 .release = blkdev_close,
2068 .llseek = block_llseek,
2069 .read_iter = blkdev_read_iter,
2070 .write_iter = blkdev_write_iter,
2071 .mmap = generic_file_mmap,
2072 .fsync = blkdev_fsync,
2073 .unlocked_ioctl = block_ioctl,
2074 #ifdef CONFIG_COMPAT
2075 .compat_ioctl = compat_blkdev_ioctl,
2077 .splice_read = generic_file_splice_read,
2078 .splice_write = iter_file_splice_write,
2079 .fallocate = blkdev_fallocate,
2082 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2085 mm_segment_t old_fs = get_fs();
2087 res = blkdev_ioctl(bdev, 0, cmd, arg);
2092 EXPORT_SYMBOL(ioctl_by_bdev);
2095 * lookup_bdev - lookup a struct block_device by name
2096 * @pathname: special file representing the block device
2098 * Get a reference to the blockdevice at @pathname in the current
2099 * namespace if possible and return it. Return ERR_PTR(error)
2102 struct block_device *lookup_bdev(const char *pathname)
2104 struct block_device *bdev;
2105 struct inode *inode;
2109 if (!pathname || !*pathname)
2110 return ERR_PTR(-EINVAL);
2112 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2114 return ERR_PTR(error);
2116 inode = d_backing_inode(path.dentry);
2118 if (!S_ISBLK(inode->i_mode))
2121 if (!may_open_dev(&path))
2124 bdev = bd_acquire(inode);
2131 bdev = ERR_PTR(error);
2134 EXPORT_SYMBOL(lookup_bdev);
2136 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2138 struct super_block *sb = get_super(bdev);
2143 * no need to lock the super, get_super holds the
2144 * read mutex so the filesystem cannot go away
2145 * under us (->put_super runs with the write lock
2148 shrink_dcache_sb(sb);
2149 res = invalidate_inodes(sb, kill_dirty);
2152 invalidate_bdev(bdev);
2155 EXPORT_SYMBOL(__invalidate_device);
2157 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2159 struct inode *inode, *old_inode = NULL;
2161 spin_lock(&blockdev_superblock->s_inode_list_lock);
2162 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2163 struct address_space *mapping = inode->i_mapping;
2164 struct block_device *bdev;
2166 spin_lock(&inode->i_lock);
2167 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2168 mapping->nrpages == 0) {
2169 spin_unlock(&inode->i_lock);
2173 spin_unlock(&inode->i_lock);
2174 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2176 * We hold a reference to 'inode' so it couldn't have been
2177 * removed from s_inodes list while we dropped the
2178 * s_inode_list_lock We cannot iput the inode now as we can
2179 * be holding the last reference and we cannot iput it under
2180 * s_inode_list_lock. So we keep the reference and iput it
2185 bdev = I_BDEV(inode);
2187 mutex_lock(&bdev->bd_mutex);
2188 if (bdev->bd_openers)
2190 mutex_unlock(&bdev->bd_mutex);
2192 spin_lock(&blockdev_superblock->s_inode_list_lock);
2194 spin_unlock(&blockdev_superblock->s_inode_list_lock);