Re: [PATCH] RDMA/siw: Fix the sendmsg byte count in siw_tcp_sendpages

[Pedro Falcato]

On Wed, Jul 23, 2025 at 02:52:12PM +0000, Bernard Metzler wrote:
> 
> 
> > -----Original Message-----
> > From: Pedro Falcato <pfalcato@xxxxxxx>
> > Sent: Wednesday, 23 July 2025 12:41
> > To: Jason Gunthorpe <jgg@xxxxxxxx>; Bernard Metzler <BMT@xxxxxxxxxxxxxx>;
> > Leon Romanovsky <leon@xxxxxxxxxx>; Vlastimil Babka <vbabka@xxxxxxx>
> > Cc: Jakub Kicinski <kuba@xxxxxxxxxx>; David Howells <dhowells@xxxxxxxxxx>;
> > Tom Talpey <tom@xxxxxxxxxx>; linux-rdma@xxxxxxxxxxxxxxx; linux-
> > kernel@xxxxxxxxxxxxxxx; linux-mm@xxxxxxxxx; Pedro Falcato
> > <pfalcato@xxxxxxx>; stable@xxxxxxxxxxxxxxx; kernel test robot
> > <oliver.sang@xxxxxxxxx>
> [snip]
> > ---
> >  drivers/infiniband/sw/siw/siw_qp_tx.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/infiniband/sw/siw/siw_qp_tx.c
> > b/drivers/infiniband/sw/siw/siw_qp_tx.c
> > index 3a08f57d2211..9576a2b766c4 100644
> > --- a/drivers/infiniband/sw/siw/siw_qp_tx.c
> > +++ b/drivers/infiniband/sw/siw/siw_qp_tx.c
> > @@ -340,11 +340,11 @@ static int siw_tcp_sendpages(struct socket *s, struct
> > page **page, int offset,
> >  		if (!sendpage_ok(page[i]))
> >  			msg.msg_flags &= ~MSG_SPLICE_PAGES;
> >  		bvec_set_page(&bvec, page[i], bytes, offset);
> > -		iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
> > +		iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, bytes);
> > 
> >  try_page_again:
> >  		lock_sock(sk);
> > -		rv = tcp_sendmsg_locked(sk, &msg, size);
> > +		rv = tcp_sendmsg_locked(sk, &msg, bytes);
> >  		release_sock(sk);
> > 
> 
> Pedro, many thanks for catching this! I completely
> missed it during my too sloppy review of that patch.
> It's a serious bug which must be fixed asap.
> BUT, looking closer, I do not see the offset being taken
> into account when retrying a current segment. So,
> resend attempts seem to send old data which are already
> out. Shouldn't the try_page_again: label be above
> bvec_set_page()??

This was raised off-list by Vlastimil - I think it's harmless to bump (but not use)
the offset here, because by reusing the iov_iter we progressively consume the data
(it keeps its own size and offset tracking internally). So the only thing we
need to track is the size we pass to tcp_sendmsg_locked[1].

If desired (and if my logic is correct!) I can send a v2 deleting that bit.


[1] Assuming tcp_sendmsg_locked guarantees it will never consume something out
of the iovec_iter without reporting it as bytes copied, which from a code reading
it seems like it won't...


-- 
Pedro