Re: [PATCH V2] net: add accounting for socket backlog

[Eric Dumazet <eric.dumazet@gmail.com>]

Le vendredi 26 février 2010 à 17:27 +0800, Zhu Yi a écrit :
> We got system OOM while running some UDP netperf testing on the loopback
> device. The case is multiple senders sent stream UDP packets to a single
> receiver via loopback on local host. Of course, the receiver is not able
> to handle all the packets in time. But we surprisingly found that these
> packets were not discarded due to the receiver's sk->sk_rcvbuf limit.
> Instead, they are kept queuing to sk->sk_backlog and finally ate up all
> the memory. We believe this is a secure hole that a none privileged user
> can crash the system.
> 
> The root cause for this problem is, when the receiver is doing
> __release_sock() (i.e. after userspace recv, kernel udp_recvmsg ->
> skb_free_datagram_locked -> release_sock), it moves skbs from backlog to
> sk_receive_queue with the softirq enabled. In the above case, multiple
> busy senders will almost make it an endless loop. The skbs in the
> backlog end up eat all the system memory.
> 
> The patch fixed this problem by adding accounting for the socket
> backlog. So that the backlog size can be restricted by protocol's choice
> (i.e. UDP).
> 
> Reported-by: Alex Shi <alex.shi@intel.com>
> Cc: David Miller <davem@davemloft.net>
> Cc: Eric Dumazet <eric.dumazet@gmail.com>
> Signed-off-by: Zhu Yi <yi.zhu@intel.com>
> ---
> V2: remove atomic operation for sk_backlog.len
>     limit UDP backlog size to 2*sk->sk_rcvbuf
> 

> +
>  static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
>  {
> +	sk->sk_backlog.len -= skb->truesize;
>  	return sk->sk_backlog_rcv(sk, skb);
>  }
>  

I am afraid sk_backlog_rcv() is not always called with lock held, and
not always called to process backlog (see TCP ucopy.prequeue)

If you take a look at __release_sock() for example, we make the backlog
private to the process before handling it (outside of lock_sock())

Therefore, I suggest doing the 'substraction' outside of
sk_backlog_rcv().

diff --git a/net/core/sock.c b/net/core/sock.c
index e1f6f22..57271cb 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -1520,6 +1520,7 @@ static void __release_sock(struct sock *sk)
 
        do {
                sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
+               sk->sk_backlog.len = 0;
                bh_unlock_sock(sk);
 
                do {


Ah, I see __release_sock() is already doing a preemption check, please
ignore my previous comment, when I said "__release_sock() could run
forever with no preemption, even with a limit on backlog"

Thanks