On Mon, 2015-06-29 at 15:06 +0200, Jason A. Donenfeld wrote:
> Some sockets set sock->sk->sk_allocation = GFP_ATOMIC. In spite of this,
> functions that call sock_alloc_send_skb will then call
> sock_alloc_send_pskb, which very often results in sleeping. Since the
> intention of callers setting sk_allocation = GFP_ATOMIC might be to be
> able to send from atomic context, we need to honor this and not sleep.
What exact problem have you noticed ? We need details please.
GFP_ATOMIC in these path does not mean 'do not wait' but instead
'allocate from emergency pools'.
We already have many ways to state ' do not wait', maybe you should use
them.
>
> Signed-off-by: Jason A. Donenfeld <ja...@zx2c4.com>
> ---
> net/core/sock.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 1e1fe9a..f00e691 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -1804,34 +1804,37 @@ struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
> unsigned long header_len,
> unsigned long data_len, int noblock,
> int *errcode, int max_page_order)
> {
> struct sk_buff *skb;
> long timeo;
> int err;
>
> timeo = sock_sndtimeo(sk, noblock);
> for (;;) {
> err = sock_error(sk);
> if (err != 0)
> goto failure;
>
> err = -EPIPE;
> if (sk->sk_shutdown & SEND_SHUTDOWN)
> goto failure;
>
> + if (sk->sk_allocation & GFP_ATOMIC)
> + break;
> +
This is the wrong place to put this test, as following one is probably
the one that is hit most of the times (fast path)
> if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
> break;
Anyway, testing for GFP_ATOMIC 'flag' is wrong.
You probably meant to test __GFP_WAIT instead, but you need to give more
details.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
