On 2023-03-23 11:35, Christoph Hellwig wrote:

On Wed, Mar 22, 2023 at 02:50:09PM +0100, Pankaj Raghav wrote:

quoted

-	if (!parent)
-		bio->bi_end_io = zram_page_end_io;
-	else
+	if (parent)

I don't think a non-chained bio without and end_io handler can work.

Hmm. Is it because in the case of non-chained bio, zram driver owns the bio,
and it is the responsibility of the driver to call bio_put in the end_io handler?

This !parent case seems to come from writeback_store, and as far as
I can tell is broken already in the current code as it just fires
off an async read without ever waiting for it, using an on-stack bio
just to make things complicated.

The bvec reading code in zram is a mess, but I have an idea how
to clean it up with a little series that should also help with
this issue.

Sounds good.

As a part of this series, should I just have an end_io which has a
call to bio_put then?

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index b7bb52f8dfbd..faa78fce327e 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c

@@ -608,10 +608,6 @@ static void free_block_bdev(struct zram *zram, unsigned long blk_idx)

 static void zram_page_end_io(struct bio *bio)
 {
-       struct page *page = bio_first_page_all(bio);
-
-       page_endio(page, op_is_write(bio_op(bio)),
-                       blk_status_to_errno(bio->bi_status));
        bio_put(bio);
 }