Re: [PATCH] sctp: avoid irq lock inversion while call sk->sk_data_ready()

[Vlad Yasevich <vladislav.yasevich@hp.com>]

Wei Yongjun wrote:
> 
> Vlad Yasevich wrote:
>> Wei Yongjun wrote:
>>   
>>> sk->sk_data_ready() of sctp socket can be called from both BH and non-BH
>>> contexts, but the default sk->sk_data_ready(), sock_def_readable(), can
>>> not be used in this case. Therefore, we have to make a new function
>>> sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.
>>>
>>>     
>> Wouldn't the same inversion happen in TCP as well?  TCP can call that
>> function in _bh and user contexts as well.
>>   
> 
> Not sure, but TCP does not call that function in user context at all.
> 

Wei

Can you trigger this problem with this patch applied?

diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index 3a95fcb..dabdc50 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -3717,9 +3717,9 @@ SCTP_STATIC int sctp_init_sock(struct sock *sk)
 	sp->hmac = NULL;

 	SCTP_DBG_OBJCNT_INC(sock);
-	percpu_counter_inc(&sctp_sockets_allocated);

 	local_bh_disable();
+	percpu_counter_inc(&sctp_sockets_allocated);
 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 	local_bh_enable();

-vlad

>> -vlad
>>
>>   
>>> ============================>>> [ INFO: possible irq lock inversion dependency detected ]
>>> 2.6.33-rc6 #129
>>> ---------------------------------------------------------
>>> sctp_darn/1517 just changed the state of lock:
>>>  (clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
>>> but this lock took another, SOFTIRQ-unsafe lock in the past:
>>>  (slock-AF_INET){+.-...}
>>>
>>> and interrupts could create inverse lock ordering between them.
>>>
>>> other info that might help us debug this:
>>> 1 lock held by sctp_darn/1517:
>>>  #0:  (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]
>>>
>>> Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
>>> ---
>>>  include/net/sctp/sctp.h |    1 +
>>>  net/sctp/endpointola.c  |    1 +
>>>  net/sctp/socket.c       |   10 ++++++++++
>>>  3 files changed, 12 insertions(+), 0 deletions(-)
>>>
>>> diff --git a/include/net/sctp/sctp.h b/include/net/sctp/sctp.h
>>> index 78740ec..fa6cde5 100644
>>> --- a/include/net/sctp/sctp.h
>>> +++ b/include/net/sctp/sctp.h
>>> @@ -128,6 +128,7 @@ extern int sctp_register_pf(struct sctp_pf *, sa_family_t);
>>>  int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb);
>>>  int sctp_inet_listen(struct socket *sock, int backlog);
>>>  void sctp_write_space(struct sock *sk);
>>> +void sctp_data_ready(struct sock *sk, int len);
>>>  unsigned int sctp_poll(struct file *file, struct socket *sock,
>>>  		poll_table *wait);
>>>  void sctp_sock_rfree(struct sk_buff *skb);
>>> diff --git a/net/sctp/endpointola.c b/net/sctp/endpointola.c
>>> index 905fda5..7ec09ba 100644
>>> --- a/net/sctp/endpointola.c
>>> +++ b/net/sctp/endpointola.c
>>> @@ -144,6 +144,7 @@ static struct sctp_endpoint *sctp_endpoint_init(struct sctp_endpoint *ep,
>>>  	/* Use SCTP specific send buffer space queues.  */
>>>  	ep->sndbuf_policy = sctp_sndbuf_policy;
>>>  
>>> +	sk->sk_data_ready = sctp_data_ready;
>>>  	sk->sk_write_space = sctp_write_space;
>>>  	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
>>>  
>>> diff --git a/net/sctp/socket.c b/net/sctp/socket.c
>>> index 67fdac9..b437e2a 100644
>>> --- a/net/sctp/socket.c
>>> +++ b/net/sctp/socket.c
>>> @@ -6185,6 +6185,16 @@ do_nonblock:
>>>  	goto out;
>>>  }
>>>  
>>> +void sctp_data_ready(struct sock *sk, int len)
>>> +{
>>> +	read_lock_bh(&sk->sk_callback_lock);
>>> +	if (sk_has_sleeper(sk))
>>> +		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
>>> +						POLLRDNORM | POLLRDBAND);
>>> +	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
>>> +	read_unlock_bh(&sk->sk_callback_lock);
>>> +}
>>> +
>>>  /* If socket sndbuf has changed, wake up all per association waiters.  */
>>>  void sctp_write_space(struct sock *sk)
>>>  {
>>>     
>>
>>   
>