Re: Panic in udp4_lib_lookup2

Hello,

On Wed, 2021-05-12 at 23:51 +0530, kapandey@codeaurora.org wrote:

We observed panic in udp_lib_lookup with below call trace:
[136523.743271]  (7) Call trace:
[136523.743275]  (7)  udp4_lib_lookup2+0x88/0x1d8
[136523.743277]  (7)  __udp4_lib_lookup+0x168/0x194
[136523.743280]  (7)  udp4_lib_lookup+0x28/0x54
[136523.743285]  (7)  nf_sk_lookup_slow_v4+0x2b4/0x384
[136523.743289]  (7)  owner_mt+0xb8/0x248
[136523.743292]  (7)  ipt_do_table+0x28c/0x6a8
[136523.743295]  (7) iptable_filter_hook+0x24/0x30
[136523.743299]  (7)  nf_hook_slow+0xa8/0x148
[136523.743303]  (7)  ip_local_deliver+0xa8/0x14c
[136523.743305]  (7)  ip_rcv+0xe0/0x134

It would be helpful if you could provide a full, decoded, stack trace
and the relevant kernel version.

We suspect this might happen due to below sequence:

Some email formatting made the "graph" very hard to read...

Time                                                   CPU X             
                                                                          
                                    CPU Y
t0                                inet_release -> udp_lib_close -> 
sk_common_release -> udp_lib_unhash                            
inet_diag_handler_cmd -> udp_diag_destroy -> __udp_diag_destroy -> 
udp_lib_rehash

... but it looks like udp_lib_close() is missing a
lock_sock()/release_sock() pair. Something alike:
---

diff --git a/include/net/udp.h b/include/net/udp.h
index 360df454356c..06586b42db3f 100644
--- a/include/net/udp.h
+++ b/include/net/udp.h

@@ -209,7 +209,9 @@ void udp_lib_rehash(struct sock *sk, u16 new_hash);
 
 static inline void udp_lib_close(struct sock *sk, long timeout)
 {
+	lock_sock(sk);
 	sk_common_release(sk);
+	release_sock(sk);
 }
 
 int udp_lib_get_port(struct sock *sk, unsigned short snum,
---

could u please give the above a spin in your testbed?

Thanks!

Paolo