On Tue, May 14, 2019 at 04:31:02PM +0200, Christoph Hellwig wrote:
> On Tue, May 14, 2019 at 10:27:17PM +0800, Ming Lei wrote:
> > I am wondering if it can be done easily, given mkfs is userspace
> > which only calls write syscall on block device. Or could you share
> > something about how to fix the stupid things?
>
> mkfs.ext4 at least uses buffered I/O on the block device. And the
> block device uses the really old buffer head based address_space ops,
> which will submit one bio per buffer_head, that is per logic block.
> mkfs probably writes much larger sizes than that..
As a first step we could try something like that patch below. Although
the mpage ops still aren't exactly optimal:
diff --git a/fs/block_dev.c b/fs/block_dev.c
index bded2ee3788d..b2ee74f1c669 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -608,12 +608,12 @@ EXPORT_SYMBOL(thaw_bdev);
static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
{
- return block_write_full_page(page, blkdev_get_block, wbc);
+ return mpage_writepage(page, blkdev_get_block, wbc);
}
static int blkdev_readpage(struct file * file, struct page * page)
{
- return block_read_full_page(page, blkdev_get_block);
+ return mpage_readpage(page, blkdev_get_block);
}
static int blkdev_readpages(struct file *file, struct address_space *mapping,
@@ -1984,7 +1984,7 @@ static int blkdev_releasepage(struct page *page, gfp_t
wait)
static int blkdev_writepages(struct address_space *mapping,
struct writeback_control *wbc)
{
- return generic_writepages(mapping, wbc);
+ return mpage_writepages(mapping, wbc, blkdev_get_block);
}
static const struct address_space_operations def_blk_aops = {
