[PATCH] nvme-tcp: fix incorrect h2cdata pdu offset accounting

[Keith Busch]

On Sun, Oct 03, 2021 at 01:19:28AM +0300, Sagi Grimberg wrote:
> > Unfortunately this was unsuccessful. The same issue is still occuring. Please
> > let me know if you have another patch to try. I'll also keep looking for a
> > solution as well.
> 
> Really? That's unexpected, this patch should ensure that the request
> is not advanced after the last payload send. The req->data_sent and
> req->data_len are recorded before we actually perform the send so
> the request should not be advanced if (e.g. last send):
> 	(req_data_sent + ret == req_data_len)
> 
> So I'm surprised that the same issue is still occurring.

Only thing I thought of so far is if req_data_len is not aligned with
nvme_tcp_req_cur_length(). When this was working, the request would not
advance based on the nvme_tcp_req_cur_length() value; now the criteria
is based on 'req->data_len'.