Re: Fragment flooding in 2.4.x/2.5.x

[kuznet@ms2.inr.ac.ru]

Hello!

> > Did you not solve this problem using right write_space?
> 
> Sure, I can add specific checks for (atomic_read(&sk->wmem_alloc) < 
> sk->sndbuf) in the RPC layer,

But it is there now.

static void
udp_write_space(struct sock *sk)
{
	struct rpc_xprt *xprt;

	if (!(xprt = xprt_from_sock(sk)))
		return;
	if (xprt->shutdown)
		return;

	/* Wait until we have enough socket memory. */
	if (sock_writeable(sk))
		return;

So, I do not understand what you speak about.



>		    	 Sending partial messages isn't a feature 
> that sounds like it would be particularly useful for any other applications 
> either.

The thing, which is really useless, is that your patch preparing skbs
and dropping them in the next line. With the same success you could
trigger BUG() there. :-) Right application just should not reach
this condition.

Anyway, I have to repeat:

>>Better way exists. Just use forced sock_wmalloc instead of
>>sock_alloc_send_skb on non-blocking send of all the fragments
>>but the first.


> However what if the actual call to alloc_skb() fails?

The same as if it would be lost by network.

Alexey