当流量遵循非对称路径时，TCP连接无法建立
简介
本文档介绍在SD-WAN交换矩阵中使用非对称路径进行流量转发时出现的问题。
问题
无法从host1(hostname - edgelin1)建立到host2(hostname - edgeclien2)的安全外壳(SSH)连接，但同时SSH在相反方向工作正常。
[root@edgeclient2 user]# ssh user@192.168.40.21 user@192.168.40.21's password: Last login: Sun Feb 10 13:26:32 2019 from 192.168.60.20 [user@edgeclient1 ~]$
[root@edgeclient1 user]# ssh user@192.168.60.20 <nothing happens after that>
或
[user@edgeclient1 ~]$ ssh user@192.168.60.20 ssh_exchange_identification: Connection closed by remote host
edgeclient1和edgeclient2 SSH守护程序和客户端都具有已知的良好配置，并且可以从本地LAN网段成功建立连接：
vedge4# request execute vpn 40 ssh user@192.168.60.20 user@192.168.60.20's password: Last login: Sun Feb 10 13:28:23 2019 from 192.168.60.7 [user@edgeclient2 ~]$
所有其他传输控制协议(TCP)应用都有类似的问题。
拓扑图
诊断
在vEdge1和vEdge3的服务端接口上，在相应方向上配置并应用了此访问控制列表(ACL):
policy
access-list SSH_IN
sequence 10
match
source-ip 192.168.40.21/32
destination-ip 192.168.60.20/32
!
action accept
count SSH_IN
!
!
default-action accept
!
access-list SSH_OUT
sequence 10
match
source-ip 192.168.60.20/32
destination-ip 192.168.40.21/32
!
action accept
count SSH_OUT
!
!
default-action accept
!
!
已在vEdge4上应用镜像ACL:
policy
access-list SSH_IN
sequence 10
match
source-ip 192.168.60.20/32
destination-ip 192.168.40.21/32
!
action accept
count SSH_IN
!
!
default-action accept
!
access-list SSH_OUT
sequence 10
match
source-ip 192.168.40.21/32
destination-ip 192.168.60.20/32
!
action accept
count SSH_OUT
!
!
default-action accept
!
!
此外，所有vEdge路由器上都启用了应用可视性，并且在SSH连接建立阶段检查了流：
vedge1# show app cflowd flows | tab ; show policy access-list-counters
TCP TIME EGRESS INGRESS
SRC DEST IP CNTRL ICMP TOTAL TOTAL MIN MAX TO INTF INTF
VPN SRC IP DEST IP PORT PORT DSCP PROTO BITS OPCODE NHOP IP PKTS BYTES LEN LEN START TIME EXPIRE NAME NAME
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
40 192.168.40.21 192.168.60.20 47866 22 0 6 24 0 192.168.109.7 3 227 66 87 Sun Feb 17 14:13:25 2019 34 ge0/0 ge0/1
COUNTER
NAME NAME PACKETS BYTES
----------------------------------
SSH_IN SSH_IN 3 227
SSH_OUT SSH_OUT 2 140
vedge3# show app cflowd flows | tab ; show policy access-list-counters
TCP TIME EGRESS INGRESS
SRC DEST IP CNTRL ICMP TOTAL TOTAL MIN MAX TO INTF INTF
VPN SRC IP DEST IP PORT PORT DSCP PROTO BITS OPCODE NHOP IP PKTS BYTES LEN LEN START TIME EXPIRE NAME NAME
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
40 192.168.60.20 192.168.40.21 22 47866 0 6 18 0 192.168.40.21 8 480 60 60 Sun Feb 17 14:14:08 2019 51 ge0/1 ge0/0
COUNTER
NAME NAME PACKETS BYTES
----------------------------------
SSH_IN SSH_IN 0 0
SSH_OUT SSH_OUT 7 420
vedge4# show app cflowd flows | tab ; show policy access-list-counters
TCP TIME EGRESS INGRESS
SRC DEST IP CNTRL ICMP TOTAL TOTAL MIN MAX TO INTF INTF
VPN SRC IP DEST IP PORT PORT DSCP PROTO BITS OPCODE NHOP IP PKTS BYTES LEN LEN START TIME EXPIRE NAME NAME
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------
40 192.168.40.21 192.168.60.20 47866 22 0 6 2 0 192.168.60.20 4 240 60 60 Sun Feb 17 14:17:44 2019 37 ge0/2 ge0/0
40 192.168.60.20 192.168.40.21 22 47866 0 6 18 0 192.168.110.6 8 592 74 74 Sun Feb 17 14:17:44 2019 49 ge0/0 ge0/2
COUNTER
NAME NAME PACKETS BYTES
----------------------------------
SSH_IN SSH_IN 8 592
SSH_OUT SSH_OUT 4 240
从这些输出中可以看到，入站和出站流是非对称的。edgeclient1(192.168.40.21)正在尝试与edgeclient2(192.168.60.20)建立SSH会话，传入流量通过vEdge1返回，返回流量通过vEdge3返回。从ACL计数器中，您还可以看到该传入和传出的数量vEdge4上的数据包与vEdge1和vEdge3上相应方向的和不匹配。同时，使用ping测试时没有丢包:
[root@edgeclient1 user]# ping -f 192.168.60.20 -c 10000 PING 192.168.60.20 (192.168.60.20) 56(84) bytes of data. --- 192.168.60.20 ping statistics --- 10000 packets transmitted, 10000 received, 0% packet loss, time 3076ms rtt min/avg/max/mdev = 0.128/0.291/6.607/0.623 ms, ipg/ewma 0.307/0.170 ms
[root@edgeclient2 user]# ping -f 192.168.40.21 -c 10000 PING 192.168.40.21 (192.168.40.21) 56(84) bytes of data. --- 192.168.40.21 ping statistics --- 10000 packets transmitted, 10000 received, 0% packet loss, time 3402ms rtt min/avg/max/mdev = 0.212/0.318/2.766/0.136 ms, ipg/ewma 0.340/0.327 ms
另外，我们还回顾了SSH在反向运行良好，并且文件也可以通过scp/sftp复制，而不会出现任何问题。
解决方案
最初怀疑存在某些深度数据包检测(DPI)配置或数据策略，但没有激活这些配置或数据策略：
vedge3# show policy from-vsmart % No entries found. vedge1# show policy from-vsmart % No entries found.
但最终发现TCP优化已启用：
vedge1# show app tcp-opt active-flows
EGRESS INGRESS
SRC DEST INTF INTF TX RX UNOPT PROXY
VPN SRC IP DEST IP PORT PORT START TIME NAME NAME BYTES BYTES TCP STATE REASON IDENTITY
--------------------------------------------------------------------------------------------------------------------------------------------
40 192.168.40.21 192.168.60.20 47868 22 Sun Feb 17 14:18:13 2019 ge0_0 ge0_1 314 0 In-progress - Client-Proxy
vedge1# show app tcp-opt expired-flows
SRC DEST TX RX UNOPT PROXY
TIMESTAMP VPN SRC IP DEST IP PORT PORT START TIME END TIME BYTES BYTES TCP STATE REASON IDENTITY DELETE REASON
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1549819969608 40 192.168.40.21 192.168.60.7 22 56612 Sun Feb 10 18:32:49 2019 Sun Feb 10 18:36:03 2019 5649 4405 Optimized - Server-Proxy CLOSED
1549820055487 40 192.168.40.21 192.168.60.7 22 56613 Sun Feb 10 18:34:15 2019 Sun Feb 10 19:07:46 2019 5719 4669 Optimized - Server-Proxy CLOSED
1550408210511 40 192.168.40.21 192.168.60.20 47862 22 Sun Feb 17 13:56:50 2019 Sun Feb 17 13:56:58 2019 401 0 Optimized - Client-Proxy STATE-TIMEOUT
1550408981634 40 192.168.40.21 192.168.60.20 47864 22 Sun Feb 17 14:09:41 2019 Sun Feb 17 14:09:49 2019 401 0 Optimized - Client-Proxy STATE-TIMEOUT
1550409205399 40 192.168.40.21 192.168.60.20 47866 22 Sun Feb 17 14:13:25 2019 Sun Feb 17 14:13:33 2019 227 0 Optimized - Client-Proxy STATE-TIMEOUT
1550409493042 40 192.168.40.21 192.168.60.20 47868 22 Sun Feb 17 14:18:13 2019 Sun Feb 17 14:18:21 2019 401 0 Optimized - Client-Proxy STATE-TIMEOUT
此外，在debugs ftm tcpopt CONN_TEARDOWN消息中可以看到。
vedge1# show log /var/log/tmplog/vdebug tail "-f" local7.debug: Feb 17 13:56:50 vedge1 FTMD[662]: ftm_tcpopt_flow_add[268]: Created new tcpflow :- vrid-3 192.168.40.21/47862 192.168.60.20/22 local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_send_conn_tear_down[388]: Trying to pack and send the following message to TCPD local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_send_conn_tear_down[408]: Sending following CONN_TD msg local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_send_conn_tear_down[413]: 192.168.40.21:47862->192.168.60.20:22; vpn:40; syn_seq_num:4172167164; identity:0; cport_prime:0 local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_msgq_tx[354]: Transfering size = 66 bytes data local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_send_conn_tear_down[416]: Successfully sent conn_td msg to TCPD local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpopt_propagate_tear_down[1038]: Sent CONN_TEARDOWN msg to tcpd for existing tcpflow :- vrid-3 192.168.40.21/47862 192.168.60.20/22 ; identity:CLIENT_SIDE_PROXY . Send Successful ! local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpopt_append_expired_err_flow_tbl[958]: Appending flow vrid-3 192.168.40.21/47862 192.168.60.20/22 to the expired flow table at Sun Feb 17 13:56:58 2019 local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpopt_append_expired_err_flow_tbl[980]: Appending flow vrid-3 192.168.40.21/47862 192.168.60.20/22 to the error flow table at Sun Feb 17 13:56:58 2019 local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpopt_flow_delete[293]: Removing tcpflow :- vrid-3 192.168.40.21/47862 192.168.60.20/22 local7.debug: Feb 17 13:56:58 vedge1 TCPD[670]: handle_upstream_connect[538]: Error - BP NULL local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_msg_decode[254]: FTM-TCPD: Received FTM_TCPD__PB_FTM_TCPD_MSG__E_MSG_TYPE__CONN_CLOSED msg local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_handle_conn_closed[139]: FTM-TCPD: Received CONN_CLOSED for following C->S local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_handle_conn_closed[150]: 192.168.40.21:47862->192.168.60.20:22; vpn:40; syn_seq_num:4172167164; identity:0; cport_prime:47862; bind_port:0 local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_handle_conn_closed[184]: FTM-TCPD: Could not find entry in FT for following flow local7.debug: Feb 17 13:56:58 vedge1 FTMD[662]: ftm_tcpd_handle_conn_closed[185]: vrid-3 192.168.40.21/47862 192.168.60.20/22
此处您可以看到TCP优化工作正常时的示例（可以看到CONN_EST消息）：
vedge3# show log /var/log/tmplog/vdebug tail "-f -n 0" local7.debug: Feb 17 15:41:13 vedge3 FTMD[657]: ftm_tcpd_msg_decode[254]: FTM-TCPD: Received FTM_TCPD__PB_FTM_TCPD_MSG__E_MSG_TYPE__CONN_CLOSED msg local7.debug: Feb 17 15:41:13 vedge3 FTMD[657]: ftm_tcpd_handle_conn_closed[139]: FTM-TCPD: Received CONN_CLOSED for following C->S local7.debug: Feb 17 15:41:13 vedge3 FTMD[657]: ftm_tcpd_handle_conn_closed[150]: 192.168.40.21:47876->192.168.60.20:22; vpn:40; syn_seq_num:2779178897; identity:0; cport_prime:47876; bind_port:0 local7.debug: Feb 17 15:41:15 vedge3 FTMD[657]: ftm_tcpd_msg_decode[258]: FTM-TCPD: Received FTM_TCPD__PB_FTM_TCPD_MSG__E_MSG_TYPE__CONN_EST msg local7.debug: Feb 17 15:41:15 vedge3 FTMD[657]: ftm_tcpd_handle_conn_est[202]: FTM-TCPD: Received CONN_EST for following C->S local7.debug: Feb 17 15:41:15 vedge3 FTMD[657]: ftm_tcpd_handle_conn_est[213]: 192.168.40.21:47878->192.168.60.20:22; vpn:40; syn_seq_num:2690847868; identity:0; cport_prime:47878; bind_port:0 local7.debug: Feb 17 15:41:15 vedge3 FTMD[657]: ftm_tcpopt_flow_add[268]: Created new tcpflow :- vrid-3 192.168.40.21/47878 192.168.60.20/22
结论
TCP优化要求流对称，因此要解决此问题，必须禁用TCP优化(no vpn 40 tcp优化)或必须创建数据策略，以强制TCP流在两个方向上采用相同路径。有关此信息，请参阅SD-WAN设计指南第23页的Traffic Symmetry for DPI。
反馈