[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Tue, 9 Sep 2014 14:44:56 -0700
From: Martin Kelly <martin@...tingkelly.com>
To: netdev@...r.kernel.org
Cc: "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>,
Stephen Hemminger <stephen@...workplumber.org>
Subject: Question about synchronize_net() in AF_PACKET close()
Hi,
I have been reading the code in net/packet/af_packet.c in order to
optimize the runtime for raw socket close(). In certain cases, close()
can take a long time to return because of the synchronize_net()
overhead. In pursuit of speeding up close(), I have been testing a patch
that defers the socket buffer memory release via call_rcu() in order to
make close() return faster. However, I hit a tricky RCU question for
which I don't have the answer and wanted to know if any RCU/networking
experts could provide some guidance.
In net/packet/af_packet.c, I noticed the following few lines within
packet_release:
synchronize_net();
/*
* Now the socket is dead. No more input will appear.
*/
sock_orphan(sk);
sock->sk = NULL;
>From testing and code analysis, I have found that it appears to be safe
to move sock_orphan and sock->sk = NULL before the synchronize_net()
call like so:
/*
* Now the socket is dead. No more input will appear.
*/
sock_orphan(sk);
sock->sk = NULL;
synchronize_net();
Could some RCU and/or networking experts chime in about whether this a
safe operation? For all I know, there is some deep, fundamental reason
why those lines are in the order they are. On the other hand, perhaps
there is not.
Thanks,
Martin
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists