sk->socket is invalid in tcp stack

sk->socket is invalid in tcp stack

[Philippe Veillette (LMC)]

Hi everybody

I've found what could be a problem in the tcp stack with linux-2.4.17 &
2.4.18.  When i run lmbench-2.0-patch2 and that i add the following line of
code in tcp_v4_rcv, it<s get added between the if (!ipsec_sk_policy(sk,skb))
... and if (sk->state == TCP_TIME_WAIT)

if (sk->socket) {	
	if (sk->socket->inode) {
		printk("Boum\n");
	}
}

I get a crash, i can give the dump later but for now, I am just wondering if
the sk->socket could be invalid when we are receiving a tcp packet.  Since
from the search i've done it seems to be initialized only when the sock
struct is initialized in sock_init_data that get called by inet_create.

But what is more frightening, is that it's alright for sometime and then
Boum, crash....

Bye

Philippe Veillette

Re: sk->socket is invalid in tcp stack

[Andi Kleen]

"Philippe Veillette (LMC)" <Philippe.Veillette@ericsson.ca> writes:

> I've found what could be a problem in the tcp stack with linux-2.4.17 &
> 2.4.18.  When i run lmbench-2.0-patch2 and that i add the following line of
> code in tcp_v4_rcv, it<s get added between the if (!ipsec_sk_policy(sk,skb))
> ... and if (sk->state == TCP_TIME_WAIT)
> 
> if (sk->socket) {	
> 	if (sk->socket->inode) {
> 		printk("Boum\n");
> 	}
> }
> 
> I get a crash, i can give the dump later but for now, I am just wondering if
> the sk->socket could be invalid when we are receiving a tcp packet.  Since

It likely did receive to a time-wait socket. time-wait buckets are 
"inherited" by hand from struct sock and live in similar hash tables, 
but only some fields at the beginning are valid. Yes, it's rather ugly, but ...

-Andi

RE: sk->socket is invalid in tcp stack

[Philippe Veillette (LMC)]

> It likely did receive to a time-wait socket. time-wait buckets are 
> "inherited" by hand from struct sock and live in similar hash tables, 
> but only some fields at the beginning are valid. Yes, it's 
> rather ugly, but ...
> 
> -Andi
> 
Thanks, 

That's what I have found after, the sock is in a time wait state, 
guess i will have to add a check...

Phil