Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <3bffa44a-b98c-85d9-9d5f-1b52943a7f2f@ispras.ru>
Date:   Mon, 29 May 2023 17:13:05 +0300
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.10.0
Subject: Re: [PATCH] udp6: Fix race condition in udp6_sendmsg & connect
Content-Language: ru
To:     Paolo Abeni <pabeni@redhat.com>,
        Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Cc:     "David S. Miller" <davem@davemloft.net>,
        David Ahern <dsahern@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, lvc-project@linuxtesting.org
References: <20230526150806.1457828-1-VEfanov@ispras.ru>
 <27614af23cd7ae4433b909194062c553a6ae16ac.camel@redhat.com>
 <027d28a0-b31b-ab42-9eb6-2826c04c9364@ispras.ru>
 <815ce4d97f6d673799ee7a94d90eeda58b1e51e4.camel@redhat.com>
From:   Vlad Efanov <vefanov@ispras.ru>
In-Reply-To: <815ce4d97f6d673799ee7a94d90eeda58b1e51e4.camel@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Thank you for detail information.

The patch is reworked and being tested now.


Vlad.

On 26.05.2023 21:13, Paolo Abeni wrote:
> On Fri, 2023-05-26 at 18:58 +0300, Ефанов Владислав Александрович
> wrote:
>> I don't think that we can just move sk_dst_set() call.
>>
>> I think we can destroy dst of sendmsg task in this case.
> AFAICS ip6_sk_dst_lookup_flow tries to acquire a reference to the
> cached dst. If the connect() clears the cache, decreasing the refcnt,
> the counter of the dst in use by sendmsg() must still be non zero.
>
> IMHO the problem you see is that sk_setup_caps() keeps using the dst
> after transferring the ownership to the dst cache, which is illegal.
> The suggested patch addressed that.
>
> If I'm wrong your syzkaller repro will keep splatting. Please have just
> have a spin, thanks.
>
> Paolo
>