From mboxrd@z Thu Jan 1 00:00:00 1970 From: Russell King - ARM Linux Subject: Re: NFS/TCP/IPv6 acting strangely in 4.2 Date: Fri, 11 Sep 2015 17:49:38 +0100 Message-ID: <20150911164937.GW21084@n2100.arm.linux.org.uk> References: <20150911113839.GO21084@n2100.arm.linux.org.uk> <1441976691.4619.58.camel@edumazet-glaptop2.roam.corp.google.com> <20150911143347.GQ21084@n2100.arm.linux.org.uk> <20150911150613.GR21084@n2100.arm.linux.org.uk> <1441984723.4619.61.camel@edumazet-glaptop2.roam.corp.google.com> <20150911162416.GV21084@n2100.arm.linux.org.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: netdev-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-nfs-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Trond Myklebust , Anna Schumaker , linux-arm-kernel-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org To: Eric Dumazet Return-path: Content-Disposition: inline In-Reply-To: <20150911162416.GV21084-l+eeeJia6m9vn6HldHNs0ANdhmdF6hFW@public.gmane.org> Sender: linux-nfs-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-Id: netdev.vger.kernel.org On Fri, Sep 11, 2015 at 05:24:17PM +0100, Russell King - ARM Linux wrot= e: > On Fri, Sep 11, 2015 at 08:18:43AM -0700, Eric Dumazet wrote: > > On Fri, 2015-09-11 at 16:06 +0100, Russell King - ARM Linux wrote: > > > On Fri, Sep 11, 2015 at 03:33:47PM +0100, Russell King - ARM Linu= x wrote: > > > > It looks like 0c78789e3a030615c6650fde89546cadf40ec2cc might be= relevant > > > > too, but I don't see that solving the multiple _concurrent_ con= nection > > > > attempts with the same port number - presumably it's somehow tr= ying to > > > > make the same socket repeatedly connect despite a previous conn= ection > > > > being in progress, which would have nothing to do with cleaning= up a > > > > previous attempt. > > >=20 > > > As I suspected, applying the above commit in addition does not so= lve the > > > problem, I still see the same behaviour: SYN SYNACK SYN RSTACK, S= YN > > > SYNACK SYN RSTACK, and eventual SYN storms. > > >=20 > > > I do have this captured as well: > > >=20 > > > 2558 0.834316 armada388 -> n2100 TCP [TCP Port numbers reused] = rpasswd=E2=86=92nfs [SYN] Seq=3D1053655487 Win=3D28800 Len=3D0 MSS=3D14= 40 SACK_PERM=3D1 TSval=3D60001 TSecr=3D0 WS=3D128 > > > 2559 0.834572 n2100 -> armada388 TCP nfs=E2=86=92rpasswd [SYN, = ACK] Seq=3D3076611574 Ack=3D1053655488 Win=3D28560 Len=3D0 MSS=3D1440 S= ACK_PERM=3D1 TSval=3D869622246 TSecr=3D60001 WS=3D64 > > > 2560 0.834666 armada388 -> n2100 TCP [TCP Port numbers reused] = rpasswd=E2=86=92nfs [SYN] Seq=3D1054228544 Win=3D28800 Len=3D0 MSS=3D14= 40 SACK_PERM=3D1 TSval=3D60005 TSecr=3D0 WS=3D128 > > > 2561 0.834895 n2100 -> armada388 TCP nfs=E2=86=92rpasswd [ACK] = Seq=3D3076611575 Ack=3D1053655488 Win=3D28560 Len=3D0 TSval=3D869622246= TSecr=3D60001 > > >=20 > > > The packet at 2561 looks wrong to me - this doesn't follow what I= know > > > would be the standard TCP setup of syn, synack, ack, because that= final > > > ack is in the wrong direction. > > >=20 > >=20 > > This 2561 packet is an ACK packet, because n2100 has a SYN_RECV soc= ket > > created by packet 2558. > >=20 > > It receives a SYN packet (2560) that it interprets as a packet slig= htly > > out of sequence (1054228544 being above 1053655487) for this SYN_RE= CV > >=20 > > The wrong packet is 2560, not 2561 >=20 > Ok. >=20 > Looking deeper at the XPRT sunrpc code, I have to wonder about the > sanity of this: >=20 > void xprt_connect(struct rpc_task *task) > { > ... > if (!xprt_connected(xprt)) { > ... > if (test_bit(XPRT_CLOSING, &xprt->state)) > return; > if (xprt_test_and_set_connecting(xprt)) > return; > xprt->stat.connect_start =3D jiffies; > xprt->ops->connect(xprt, task); >=20 > That calls into xs_connect(), which schedules a workqueue to do the > connection. The workqueue will call xs_tcp_setup_socket(). >=20 > xs_tcp_setup_socket() creates a socket if one didn't exist, otherwise > re-using the previously obtained socket (which'll be why its using th= e > same socket) and then goes on to call xs_tcp_finish_connecting(). >=20 > xs_tcp_finish_connecting() calls kernel_connect(), which will return > -EINPROGRESS. We seem to treat EINPROGRESS as if the connection was > successful: >=20 > case 0: > case -EINPROGRESS: > case -EALREADY: > xprt_unlock_connect(xprt, transport); > xprt_clear_connecting(xprt); > return; >=20 > and the xprt_clear_connecting() results in this whole path being > re-opened: the socket is not yet connected, so xprt_connected() will > return false, and despite the socket connection still being mid-way > through being connected, we clear the "connecting" status, causing > xprt_test_and_set_connecting() to return false. >=20 > That allows us to re-call xprt->ops->connect, re-queue the connect > worker, and re-run the call to kernel_connect() for a socket which is > already mid-way through being connected. >=20 > Shouldn't the "connecting" status only be cleared when either the soc= ket > has _finished_ connecting, or when the connection has _failed_ to con= nect, > and not when it's mid-way through connecting? >=20 > I've not been able to prove this: I've set rpc_debug to 129 to log > just xprt and trans RPC facilities, and that's sufficient to change > the timing such that this doesn't happen. =46ollowing that idea, I just tried the patch below, and it seems to wo= rk. I don't know whether it handles all cases after a call to kernel_connec= t(), but it stops the multiple connection attempts: 1 0.000000 armada388 -> n2100 TCP 1009=E2=86=92nfs [SYN] Seq=3D3794= 066539 Win=3D28560 Len=3D0 MSS=3D1440 SACK_PERM=3D1 TSval=3D15712 TSecr= =3D870317691 WS=3D128 2 0.000414 n2100 -> armada388 TCP nfs=E2=86=921009 [SYN, ACK] Seq=3D= 1884476522 Ack=3D3794066540 Win=3D28560 Len=3D0 MSS=3D1440 SACK_PERM=3D= 1 TSval=3D870318939 TSecr=3D15712 WS=3D64 3 0.000787 armada388 -> n2100 TCP 1009=E2=86=92nfs [ACK] Seq=3D3794= 066540 Ack=3D1884476523 Win=3D28672 Len=3D0 TSval=3D15712 TSecr=3D87031= 8939 4 0.001304 armada388 -> n2100 NFS V3 ACCESS Call, FH: 0x905379cc, [= Check: RD LU MD XT DL] 5 0.001566 n2100 -> armada388 TCP nfs=E2=86=921009 [ACK] Seq=3D1884= 476523 Ack=3D3794066660 Win=3D28608 Len=3D0 TSval=3D870318939 TSecr=3D1= 5712 6 0.001640 armada388 -> n2100 NFS V3 ACCESS Call, FH: 0x905379cc, [= Check: RD LU MD XT DL] 7 0.001866 n2100 -> armada388 TCP nfs=E2=86=921009 [ACK] Seq=3D1884= 476523 Ack=3D3794066780 Win=3D28608 Len=3D0 TSval=3D870318939 TSecr=3D1= 5712 8 0.003070 n2100 -> armada388 NFS V3 ACCESS Reply (Call In 4), [All= owed: RD LU MD XT DL] 9 0.003415 armada388 -> n2100 TCP 1009=E2=86=92nfs [ACK] Seq=3D3794= 066780 Ack=3D1884476647 Win=3D28672 Len=3D0 TSval=3D15712 TSecr=3D87031= 8939 10 0.003592 armada388 -> n2100 NFS V3 ACCESS Call, FH: 0xe15fc9c9, [= Check: RD LU MD XT DL] 11 0.004354 n2100 -> armada388 NFS V3 ACCESS Reply (Call In 6), [All= owed: RD LU MD XT DL] 12 0.004682 armada388 -> n2100 NFS V3 ACCESS Call, FH: 0xe15fc9c9, [= Check: RD LU MD XT DL] 13 0.005365 n2100 -> armada388 NFS V3 ACCESS Reply (Call In 10), [Al= lowed: RD LU MD XT DL] 14 0.005701 armada388 -> n2100 NFS V3 GETATTR Call, FH: 0xe15fc9c9 =2E.. net/sunrpc/xprtsock.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c index ff5b6a2e62c3..c456d6e51c56 100644 --- a/net/sunrpc/xprtsock.c +++ b/net/sunrpc/xprtsock.c @@ -1450,6 +1450,7 @@ static void xs_tcp_state_change(struct sock *sk) switch (sk->sk_state) { case TCP_ESTABLISHED: spin_lock(&xprt->transport_lock); + xprt_clear_connecting(xprt); if (!xprt_test_and_set_connected(xprt)) { struct sock_xprt *transport =3D container_of(xprt, struct sock_xprt, xprt); @@ -1474,12 +1475,14 @@ static void xs_tcp_state_change(struct sock *sk= ) smp_mb__before_atomic(); clear_bit(XPRT_CONNECTED, &xprt->state); clear_bit(XPRT_CLOSE_WAIT, &xprt->state); + clear_bit(XPRT_CONNECTING, &xprt->state); =09 smp_mb__after_atomic(); break; case TCP_CLOSE_WAIT: /* The server initiated a shutdown of the socket */ xprt->connect_cookie++; clear_bit(XPRT_CONNECTED, &xprt->state); + clear_bit(XPRT_CONNECTING, &xprt->state); xs_tcp_force_close(xprt); case TCP_CLOSING: /* @@ -1493,6 +1496,7 @@ static void xs_tcp_state_change(struct sock *sk) set_bit(XPRT_CLOSING, &xprt->state); smp_mb__before_atomic(); clear_bit(XPRT_CONNECTED, &xprt->state); + clear_bit(XPRT_CONNECTING, &xprt->state); smp_mb__after_atomic(); break; case TCP_CLOSE: @@ -2237,11 +2241,13 @@ static void xs_tcp_setup_socket(struct work_str= uct *work) xs_tcp_force_close(xprt); break; case 0: - case -EINPROGRESS: case -EALREADY: xprt_unlock_connect(xprt, transport); xprt_clear_connecting(xprt); return; + case -EINPROGRESS: + xprt_unlock_connect(xprt, transport); + return; case -EINVAL: /* Happens, for instance, if the user specified a link * local IPv6 address without a scope-id. --=20 =46TTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up according to speedtest.net. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at http://vger.kernel.org/majordomo-info.html