Message-ID: <9e3bb558-ecb1-a6aa-35e4-a2771136b3fe@gmail.com>
Date:   Tue, 11 Jan 2022 20:39:26 +0000
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.4.1
Subject: Re: [PATCH 09/14] ipv6: hand dst refs to cork setup
Content-Language: en-US
To:     Paolo Abeni <pabeni@redhat.com>, netdev@vger.kernel.org,
        "David S . Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>
Cc:     Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        David Ahern <dsahern@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        Willem de Bruijn <willemdebruijn.kernel@gmail.com>,
        linux-kernel@vger.kernel.org
References: <cover.1641863490.git.asml.silence@gmail.com>
 <07031c43d3e5c005fbfc76b60a58e30c66d7c620.1641863490.git.asml.silence@gmail.com>
 <48293134f179d643e9ec7bcbd7bca895df7611ac.camel@redhat.com>
From:   Pavel Begunkov <asml.silence@gmail.com>
In-Reply-To: <48293134f179d643e9ec7bcbd7bca895df7611ac.camel@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 1/11/22 17:11, Paolo Abeni wrote:
> On Tue, 2022-01-11 at 01:21 +0000, Pavel Begunkov wrote:
>> During cork->dst setup, ip6_make_skb() gets an additional reference to
>> a passed in dst. However, udpv6_sendmsg() doesn't need dst after calling
>> ip6_make_skb(), and so we can save two additional atomics by passing
>> dst references to ip6_make_skb(). udpv6_sendmsg() is the only caller, so
>> it's enough to make sure it doesn't use dst afterwards.
> 
> What about the corked path in udp6_sendmsg()? I mean:

It doesn't change it for callers, so the ref stays with udp6_sendmsg() when
corking. To compensate for ip6_setup_cork() there is an explicit dst_hold()
in ip6_append_data, should be fine.

@@ -1784,6 +1784,7 @@ int ip6_append_data(struct sock *sk,
  		/*
  		 * setup for corking
  		 */
+		dst_hold(&rt->dst);
  		err = ip6_setup_cork(sk, &inet->cork, &np->cork,
  				     ipc6, rt);


I don't care much about corking perf, but might be better to implement
this "handing away" for ip6_append_data() as well to be more consistent
with ip6_make_skb().


> udp6_sendmsg(MSG_MORE) -> ip6_append_data() -> ip6_setup_cork()
> 
> what if ip6_setup_cork() errors out in that path?

-- 
Pavel Begunkov