Re: [PATCH] tipc: fix a lockdep warning

[Ying Xue <ying.xue@windriver.com>]

Hi Weidong,

Please ignore my last mail because you are to fix the issue which is
different with my mentioned one, and please see my below comments about
your patch;

On 11/22/2013 04:18 PM, wangweidong wrote:
> PC1:tipc-config -netid=1234 -a=1.1.2 -be=eth:eth0/1.1.0
> PC2:tipc-config -netid=1234 -a=1.1.3 -be=eth:eth0/1.1.0
> 
> I used a server code Like this:
> ----------------
>     sk=socket(AF_TIPC,SOCK_RDM,0);
>     bind(sk, &addr, len);
>     while(1) {
>         recvfrom(sk,...);
>         ...
>         sendto(sk,...);
>     }
> ----------------
> 
> when I did ./server in PC1, I got a lockdep as bellow:
> 
> ======================================================
> [ INFO: possible circular locking dependency detected ]
> 3.12.0-lockdep+ #4 Not tainted
> -------------------------------------------------------
> server/3772 is trying to acquire lock:
> (tipc_net_lock){++.-..}, at: [<ffffffffa02e324f>] tipc_link_send+0x2f/0xc0 [tipc]
> 
> but task is already holding lock:
> (tipc_nametbl_lock){++--..}, at: [<ffffffffa02e83e6>] tipc_nametbl_publish+0x46/0xc0 [tipc]
>  which lock already depends on the new lock.
> 
> the existing dependency chain (in reverse order) is:
>  -> #1 (tipc_nametbl_lock){++--..}:
>         [<ffffffff810a2547>] validate_chain+0x6a7/0x7d0
>         [<ffffffff810a29d1>] __lock_acquire+0x361/0x610
>         [<ffffffff810a2d62>] lock_acquire+0xe2/0x110
>         [<ffffffff8151e061>] _raw_write_lock_bh+0x31/0x40
>         [<ffffffffa02e5d40>] tipc_named_reinit+0x10/0x70 [tipc]
>         [<ffffffffa02e8512>] tipc_net_start+0x22/0x80 [tipc]
>         [<ffffffffa02dff0e>] tipc_core_start_net+0xe/0x40 [tipc]
>         [<ffffffffa02df625>] cfg_set_own_addr+0x75/0xc0 [tipc]
>         [<ffffffffa02df8f5>] tipc_cfg_do_cmd+0x135/0x550 [tipc]
>         [<ffffffffa02e87f9>] handle_cmd+0x49/0xe0 [tipc]
>         [<ffffffff814764fd>] genl_family_rcv_msg+0x22d/0x3c0
>         [<ffffffff81476700>] genl_rcv_msg+0x70/0xd0
>         [<ffffffff81474dc9>] netlink_rcv_skb+0x89/0xb0
>         [<ffffffff81475f87>] genl_rcv+0x27/0x40
>         [<ffffffff81474b1e>] netlink_unicast+0x14e/0x1a0
>         [<ffffffff81475735>] netlink_sendmsg+0x245/0x420
>         [<ffffffff814294f6>] __sock_sendmsg+0x66/0x80
>         [<ffffffff814295c2>] sock_aio_write+0xb2/0xc0
>         [<ffffffff811968f0>] do_sync_write+0x60/0x90
>         [<ffffffff81198891>] vfs_write+0x1d1/0x1e0
>         [<ffffffff811989bd>] SyS_write+0x5d/0xa0
>         [<ffffffff81527522>] system_call_fastpath+0x16/0x1b
> 
>  -> #0 (tipc_net_lock){++.-..}:
>         [<ffffffff810a1e2e>] check_prev_add+0x41e/0x490
>         [<ffffffff810a2547>] validate_chain+0x6a7/0x7d0
>         [<ffffffff810a29d1>] __lock_acquire+0x361/0x610
>         [<ffffffff810a2d62>] lock_acquire+0xe2/0x110
>         [<ffffffff8151e2f4>] _raw_read_lock_bh+0x34/0x50
>         [<ffffffffa02e324f>] tipc_link_send+0x2f/0xc0 [tipc]
>         [<ffffffffa02e617b>] named_cluster_distribute+0x6b/0x80 [tipc]
>         [<ffffffffa02e62ab>] tipc_named_publish+0x7b/0x90 [tipc]
>         [<ffffffffa02e841b>] tipc_nametbl_publish+0x7b/0xc0 [tipc]
>         [<ffffffffa02e9958>] tipc_publish+0x98/0xf0 [tipc]
>         [<ffffffffa02ebf58>] bind+0x78/0xb0 [tipc]
>         [<ffffffff81428dc0>] SyS_bind+0xb0/0xd0
>         [<ffffffff81527522>] system_call_fastpath+0x16/0x1b
> 
>  other info that might help us debug this:
> 
>   Possible unsafe locking scenario:
> 
>         CPU0                    CPU1
>         ----                    ----
>    lock(tipc_nametbl_lock);
>                                 lock(tipc_net_lock);
>                                 lock(tipc_nametbl_lock);
>    lock(tipc_net_lock);
> 
>   *** DEADLOCK ***
> ----------------------------------------------------------
> 
> problem is that tipc_nametbl_publish which will hold tipc_nametbl_lock
> and acquire tipc_net_lock, while the tipc_net_start which hold
> tipc_net_lock and acquir tipc_nametbl_lock, so the dead lock occurs.
> 
> tipc_link_send protected by tipc_net_lock, we can unlock the
> tipc_nametbl_lock, and it no need the tipc_nametbl_lock to protect it.
> so I just unlock the tbl_lock before it, and lock the tbl_lock after it.
> 
> Signed-off-by: Wang Weidong <wangweidong1@huawei.com>
> ---
>  net/tipc/name_distr.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/net/tipc/name_distr.c b/net/tipc/name_distr.c
> index e0d0805..ab8f96c 100644
> --- a/net/tipc/name_distr.c
> +++ b/net/tipc/name_distr.c
> @@ -138,7 +138,9 @@ static void named_cluster_distribute(struct sk_buff *buf)
>  			if (!buf_copy)
>  				break;
>  			msg_set_destnode(buf_msg(buf_copy), n_ptr->addr);
> +			write_unlock_bh(&tipc_nametbl_lock);
>  			tipc_link_send(buf_copy, n_ptr->addr, n_ptr->addr);
> +			write_lock_bh(&tipc_nametbl_lock);

We cannot temporarily release/hold tipc_nametbl_lock here, please see
below call path:

tipc_nametbl_withdraw()
  tipc_named_withdraw()
    named_cluster_distribute()
      tipc_link_send()

Especially in tipc_nametbl_withdraw(), we must hold tipc_nametbl_lock to
protect name table before tipc_named_withdraw() is called. If we
temporarily release tipc_nametbl_lock in named_cluster_distribute(), I
am afraid that name table might be changed by another thread at the
moment, having name table inconsistent possibly.

Actually we are implementing another patchset purging the tipc_net_lock
from TIPC stack. If the patchset is involved, I guess the issue would
disappear.

If you have an interesting to see how to purge to tipc_net_lock, please
monitor tipc-discussion mail list.

Regards,
Ying


>  		}
>  	}
>  
> 


------------------------------------------------------------------------------
Shape the Mobile Experience: Free Subscription
Software experts and developers: Be at the forefront of tech innovation.
Intel(R) Software Adrenaline delivers strategic insight and game-changing 
conversations that shape the rapidly evolving mobile landscape. Sign up now. 
http://pubads.g.doubleclick.net/gampad/clk?id=63431311&iu=/4140/ostg.clktrk