Re: [PATCH v2] RDMA/siw: Fix the sendmsg byte count in siw_tcp_sendpages

[Pedro Falcato]

On Tue, Jul 29, 2025 at 08:53:02PM +0200, Bernard Metzler wrote:
> On 29.07.2025 14:03, Pedro Falcato wrote:
> > Ever since commit c2ff29e99a76 ("siw: Inline do_tcp_sendpages()"),
> > we have been doing this:
> > 
> > static int siw_tcp_sendpages(struct socket *s, struct page **page, int offset,
> >                               size_t size)
> > [...]
> >          /* Calculate the number of bytes we need to push, for this page
> >           * specifically */
> >          size_t bytes = min_t(size_t, PAGE_SIZE - offset, size);
> >          /* If we can't splice it, then copy it in, as normal */
> >          if (!sendpage_ok(page[i]))
> >                  msg.msg_flags &= ~MSG_SPLICE_PAGES;
> >          /* Set the bvec pointing to the page, with len $bytes */
> >          bvec_set_page(&bvec, page[i], bytes, offset);
> >          /* Set the iter to $size, aka the size of the whole sendpages (!!!) */
> >          iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, size);
> > try_page_again:
> >          lock_sock(sk);
> >          /* Sendmsg with $size size (!!!) */
> >          rv = tcp_sendmsg_locked(sk, &msg, size);
> > 
> > This means we've been sending oversized iov_iters and tcp_sendmsg calls
> > for a while. This has a been a benign bug because sendpage_ok() always
> > returned true. With the recent slab allocator changes being slowly
> > introduced into next (that disallow sendpage on large kmalloc
> > allocations), we have recently hit out-of-bounds crashes, due to slight
> > differences in iov_iter behavior between the MSG_SPLICE_PAGES and
> > "regular" copy paths:
> > 
> > (MSG_SPLICE_PAGES)
> > skb_splice_from_iter
> >    iov_iter_extract_pages
> >      iov_iter_extract_bvec_pages
> >        uses i->nr_segs to correctly stop in its tracks before OoB'ing everywhere
> >    skb_splice_from_iter gets a "short" read
> > 
> > (!MSG_SPLICE_PAGES)
> > skb_copy_to_page_nocache copy=iov_iter_count
> >   [...]
> >     copy_from_iter
> >          /* this doesn't help */
> >          if (unlikely(iter->count < len))
> >                  len = iter->count;
> >            iterate_bvec
> >              ... and we run off the bvecs
> > 
> > Fix this by properly setting the iov_iter's byte count, plus sending the
> > correct byte count to tcp_sendmsg_locked.
> > 
> > Cc: stable@xxxxxxxxxxxxxxx
> > Fixes: c2ff29e99a76 ("siw: Inline do_tcp_sendpages()")
> > Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
> > Closes: https://lore.kernel.org/oe-lkp/202507220801.50a7210-lkp@xxxxxxxxx
> > Reviewed-by: David Howells <dhowells@xxxxxxxxxx>
> > Signed-off-by: Pedro Falcato <pfalcato@xxxxxxx>
> > ---
> > 
> > v2:
> >   - Add David Howells's Rb on the original patch
> >   - Remove the offset increment, since it's dead code
> > 
> >   drivers/infiniband/sw/siw/siw_qp_tx.c | 5 ++---
> >   1 file changed, 2 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/infiniband/sw/siw/siw_qp_tx.c b/drivers/infiniband/sw/siw/siw_qp_tx.c
> > index 3a08f57d2211..f7dd32c6e5ba 100644
> > --- a/drivers/infiniband/sw/siw/siw_qp_tx.c
> > +++ b/drivers/infiniband/sw/siw/siw_qp_tx.c
> > @@ -340,18 +340,17 @@ 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);
> >   		if (rv > 0) {
> >   			size -= rv;
> >   			sent += rv;
> >   			if (rv != bytes) {
> > -				offset += rv;
> >   				bytes -= rv;
> >   				goto try_page_again;
> >   			}
> 
> Acked-by: Bernard Metzler <bernard.metzler@xxxxxxxxx>


Thanks!

Do you want to take the fix through your tree? Otherwise I suspect Vlastimil
could simply take it (and possibly resubmit the SLAB PR, which hasn't been
merged yet).

-- 
Pedro