netdev - Re: [PATCH net] udp: fix l4 hash after reconnect

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <939fa561-841e-442a-be0b-0e71c6843e5c@redhat.com>
Date: Tue, 7 Jan 2025 08:56:31 +0100
From: Paolo Abeni <pabeni@...hat.com>
To: Philo Lu <lulie@...ux.alibaba.com>, Eric Dumazet <edumazet@...gle.com>
Cc: netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>,
 David Ahern <dsahern@...nel.org>, Jakub Kicinski <kuba@...nel.org>,
 Simon Horman <horms@...nel.org>, Fred Chen <fred.cc@...baba-inc.com>,
 Cambda Zhu <cambda@...ux.alibaba.com>, Willem de Bruijn
 <willemb@...gle.com>, Stefano Brivio <sbrivio@...hat.com>
Subject: Re: [PATCH net] udp: fix l4 hash after reconnect

Hi,

I'm sorry for the latency, I was off in the past days.

On 12/31/24 8:55 AM, Philo Lu wrote:
> On 2024/12/10 16:32, Paolo Abeni wrote:
>> On 12/7/24 03:34, Philo Lu wrote:
>>> On 2024/12/7 00:23, Paolo Abeni wrote:
>>>> On 12/6/24 17:01, Eric Dumazet wrote:
>>>>> BTW, it seems that udp_lib_rehash() does the udp_rehash4()
>>>>> only if the hash2 has changed.
>>>>
>>>> Oh, you are right, that requires a separate fix.
>>>>
>>>> @Philo: could you please have a look at that? basically you need to
>>>> check separately for hash2 and hash4 changes.
>>>
>>> This is a good question. IIUC, the only affected case is when trying to
>>> re-connect another remote address with the same local address
>>
>> AFAICS, there is also another case: when re-connection using a different
>> local addresses with the same l2 hash...
>>
>>> (i.e.,
>>> hash2 unchanged). And this will be handled by udp_lib_hash4(). So in
>>> udp_lib_rehash() I put rehash4() inside hash2 checking, which means a
>>> passive rehash4 following rehash2.
>>
>> ... but even the latter case should be covered from the above.
>>
>>> So I think it's more about the convention for rehash. We can choose the
>>> better one.
>>
>> IIRC a related question raised during code review for the udp L4 hash
>> patches. Perhaps refactoring the code slightly to let udp_rehash()
>> really doing the re-hashing and udp_hash really doing only the hashing
>> could be worth.
>>
> 
> I'm trying to unify rehash() for both hash2 and hash4 in 
> __ip4_datagram_connect, when I noticed the inet_rcv_saddr checking 
> before calling rehash():
> 
> ```
> if (!inet->inet_rcv_saddr) {
> 	inet->inet_rcv_saddr = fl4->saddr;
> 	if (sk->sk_prot->rehash)
> 		sk->sk_prot->rehash(sk);
> }
> ```
> This means inet_rcv_saddr is reset at most once no matter how many times 
> connect() is called. 

... if you make consecutive connect(<dst address>) calls.

 __udp_disconnect() clears saddr, so:

connect(<AF_UNSPEC>); connect(<dst address>);

will yield the expected result

> I'm not sure if this is by-design for some reason? 
> Or can I remove this checking? like:
> 
> --- a/net/ipv4/datagram.c
> +++ b/net/ipv4/datagram.c
> @@ -67,11 +67,9 @@ int __ip4_datagram_connect(struct sock *sk, struct 
> sockaddr *uaddr, int addr_len
>          inet->inet_dport = usin->sin_port;
>          if (!inet->inet_saddr)
>                  inet->inet_saddr = fl4->saddr;
> -       if (!inet->inet_rcv_saddr) {
> -               inet->inet_rcv_saddr = fl4->saddr;
> -               if (sk->sk_prot->rehash)
> -                       sk->sk_prot->rehash(sk);
> -       }
> +       inet->inet_rcv_saddr = fl4->saddr;
> +       if (sk->sk_prot->rehash)
> +               sk->sk_prot->rehash(sk);
>          reuseport_has_conns_set(sk);
>          sk->sk_state = TCP_ESTABLISHED;
>          sk_set_txhash(sk);

This sounds like an unexpected behaviour change which may broke existing
applications.

I suggest retaining the current beheviour.

Thanks,

Paolo