On Fri, Apr 19, 2019 at 7:02 PM Eric Dumazet <eduma...@google.com> wrote:
>
> tcp sendmsg() and sendpage() normally advance skb->data_len
> and skb->truesize by the payload added to an skb.
>
> But sendmsg(fd, ..., MSG_ZEROCOPY) has to account for whole pages,
> even if a single byte of payload is used in the page.
>
> This means that we can not assume skb->truesize can be adjusted
> by skb->data_len. We must instead overwrite its value.
>
> Otherwise skb->truesize is too big and can hit socket sndbuf limit,
> especially if the skb is recycled multiple times :/
>
> Fixes: 472c2e07eef0 ("tcp: add one skb cache for tx")
> Signed-off-by: Eric Dumazet <eduma...@google.com>
> Cc: Soheil Hassas Yeganeh <soh...@google.com>
> Cc: Willem de Bruijn <will...@google.com>
Acked-by: Soheil Hassas Yeganeh <soh...@google.com>
Nice catch! Thank you for the fix.
> ---
> net/ipv4/tcp.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> index
> 603e770d59b3db96adca9602319d2f6970a56285..f7567a3698eb4d6c24dbe6a2f110413ce14d24a7
> 100644
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -868,7 +868,7 @@ struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int
> size, gfp_t gfp,
> if (likely(!size)) {
> skb = sk->sk_tx_skb_cache;
> if (skb && !skb_cloned(skb)) {
> - skb->truesize -= skb->data_len;
> + skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
> sk->sk_tx_skb_cache = NULL;
> pskb_trim(skb, 0);
> INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
> --
> 2.21.0.593.g511ec345e18-goog
>
