Re: [PATCH v2 12/12] nvmet: use bio_chain_and_submit to simplify bio chaining

Stephen Zhang <[email protected]> Mon, 1 Dec 2025 10:41:06 +0800
Newsgroups org.kernel.vger.linux-bcache,dev.linux.lists.gfs2,dev.linux.lists.ntfs3,dev.linux.lists.nvdimm,dev.linux.lists.virtualization,org.infradead.lists.linux-nvme,org.kernel.vger.linux-block,org.kernel.vger.linux-kernel,org.kernel.vger.linux-xfs
Message-ID <CANubcdWAk2Mh5b9stjTh8N84jq+XAgaR3n2-VYRinU9ERtJLUw@mail.gmail.com>
Sagi Grimberg <[email protected]> 于2025年12月1日周一 07:03写道:
>
> Acked-by: Sagi Grimberg <[email protected]>

Hello,

I already dropped this patch in v3:
https://lore.kernel.org/all/[email protected]/
The reason is that the order of operations is critical. In the original code::
----------------
...
bio->bi_end_io = nvmet_bio_done;

for_each_sg(req->sg, sg, req->sg_cnt, i) {
...
          struct bio *prev = bio;
....
          bio_chain(bio, prev);
          submit_bio(prev);
}
----------------

the oldest bio (i.e., prev) retains the real bi_end_io function:

bio -> bio -> ... -> prev
However, using bio_chain_and_submit(prev, bio) would create the reverse chain:

prev -> prev -> ... -> bio

where the newest bio would hold the real bi_end_io function, which does not
match the required behavior in this context.

Thanks,
Shida