Re: [PATCH vfs.all 22/26] block: stash a bdev_file to read/write raw blcok_device

[Yu Kuai <yukuai1@xxxxxxxxxxxxxxx>]

Hi,

在 2024/04/09 18:23, Christian Brauner 写道:
> > +static int __stash_bdev_file(struct block_device *bdev)
>
> I've said that on the previous version. I think that this is really
> error prone and seems overall like an unpleasant solution. I would
> really like to avoid going down that route.

Yes, I see your point, and it's indeed reasonable.

> I think a chunk of this series is good though specicially simple
> conversions of individual filesystems where file_inode() or f_mapping
> makes sense. There's a few exceptions where we might be better of
> replacing the current apis with something else (I think Al touched on
> that somewhere further down the thread.).
>
> I'd suggest the straightforward bd_inode removals into a separate series
> that I can take.
>
> Thanks for working on all of this. It's certainly a contentious area.

How about following simple patch to expose bdev_mapping() for
fs/buffer.c for now?

Thanks,
Kuai

diff --git a/block/blk.h b/block/blk.h
index a34bb590cce6..f8bcb43a12c6 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -428,7 +428,6 @@ static inline int blkdev_zone_mgmt_ioctl(struct 
block_device *bdev,
 #endif /* CONFIG_BLK_DEV_ZONED */

 struct inode *bdev_inode(struct block_device *bdev);
-struct address_space *bdev_mapping(struct block_device *bdev);
 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
 void bdev_add(struct block_device *bdev, dev_t dev);

diff --git a/fs/buffer.c b/fs/buffer.c
index 4f73d23c2c46..e2bd19e3fe48 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -189,8 +189,8 @@ EXPORT_SYMBOL(end_buffer_write_sync);
 static struct buffer_head *
 __find_get_block_slow(struct block_device *bdev, sector_t block)
 {
-       struct inode *bd_inode = bdev->bd_inode;
-       struct address_space *bd_mapping = bd_inode->i_mapping;
+       struct address_space *bd_mapping = bdev_mapping(bdev);
+       struct inode *bd_inode = bd_mapping->host;
        struct buffer_head *ret = NULL;
        pgoff_t index;
        struct buffer_head *bh;
@@ -1034,12 +1034,12 @@ static sector_t folio_init_buffers(struct folio 
*folio,
 static bool grow_dev_folio(struct block_device *bdev, sector_t block,
                pgoff_t index, unsigned size, gfp_t gfp)
 {
-       struct inode *inode = bdev->bd_inode;
+       struct address_space *bd_mapping = bdev_mapping(bdev);
        struct folio *folio;
        struct buffer_head *bh;
        sector_t end_block = 0;

-       folio = __filemap_get_folio(inode->i_mapping, index,
+       folio = __filemap_get_folio(bd_mapping, index,
                        FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp);
        if (IS_ERR(folio))
                return false;
@@ -1073,10 +1073,10 @@ static bool grow_dev_folio(struct block_device 
*bdev, sector_t block,
         * lock to be atomic wrt __find_get_block(), which does not
         * run under the folio lock.
         */
-       spin_lock(&inode->i_mapping->i_private_lock);
+       spin_lock(&bd_mapping->i_private_lock);
        link_dev_buffers(folio, bh);
        end_block = folio_init_buffers(folio, bdev, size);
-       spin_unlock(&inode->i_mapping->i_private_lock);
+       spin_unlock(&bd_mapping->i_private_lock);
 unlock:
        folio_unlock(folio);
        folio_put(folio);
@@ -1463,7 +1463,7 @@ __bread_gfp(struct block_device *bdev, sector_t block,
 {
        struct buffer_head *bh;

-       gfp |= mapping_gfp_constraint(bdev->bd_inode->i_mapping, ~__GFP_FS);
+       gfp |= mapping_gfp_constraint(bdev_mapping(bdev), ~__GFP_FS);

        /*
         * Prefer looping in the allocator rather than here, at least that
@@ -1696,8 +1696,8 @@ EXPORT_SYMBOL(create_empty_buffers);
  */
 void clean_bdev_aliases(struct block_device *bdev, sector_t block, 
sector_t len)
 {
-       struct inode *bd_inode = bdev->bd_inode;
-       struct address_space *bd_mapping = bd_inode->i_mapping;
+       struct address_space *bd_mapping = bdev_mapping(bdev);
+       struct inode *bd_inode = bd_mapping->host;
        struct folio_batch fbatch;
        pgoff_t index = ((loff_t)block << bd_inode->i_blkbits) / PAGE_SIZE;
        pgoff_t end;
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index bc840e0fb6e5..bbae55535d53 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1527,6 +1527,7 @@ void blkdev_put_no_open(struct block_device *bdev);

 struct block_device *I_BDEV(struct inode *inode);
 struct block_device *file_bdev(struct file *bdev_file);
+struct address_space *bdev_mapping(struct block_device *bdev);
 bool disk_live(struct gendisk *disk);
 unsigned int block_size(struct block_device *bdev);

> .