知之者不如好之者，好之者不如乐之者
实验目的
通过实验简单了解 packetdrill 时间模型中的其他时间。
时间模型
由于许多协议对时间非常敏感，所以在「packetdrill」 脚本中增加了对时间灵活性的支持。每个语句都有一个时间戳，由 「packetdrill」 强制执行：如果事件没有在指定的时间发生，则 「packetdrill」 会标记一个错误并报告实际时间。
「packetdrill」时间模型
|
Model |
Syntax |
Description |
|
Absolute |
0.750 |
Specifies the specific time at which an event should occur. |
|
Relative |
+0.2 |
Specifies the interval after the last event at which an event should occur. |
|
Wildcard |
* |
Allows an event to occur at any time. |
|
Range |
0.750~0.9 |
Requires the given event to occur with in the timer ange. |
|
Loose |
--tolerance_usecs=800 |
Allows all events to happen with in a range(from the command line). |
|
Blocking |
0.750...0.9 |
Specifies a blocking system call that starts/returns at the given times. |
避免伪造失败
使用 --tolerance_usecs 参数值 4ms，使得事件只要在期望时间的 4ms 范围内发生，即认为测试是成功的。
基础脚本
# cat tcp_3hs_000.pkt// TCP 基础之三次握手0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>+0 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4
其他时间
以 TCP 三次握手为基础，测试时间模型中的其他时间，包括 Wildcard、Range、Loose 和 Blocking。
实验测试 Wildcard
初步修改的脚本，TCP 三次握手阶段时间模型全部改为 * 模式。
# cat tcp_3hs_time_008.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0* < S 0:0(0) win 10000 <mss 1460>* > S. 0:0(0) ack 1 <...>* < . 1:1(0) ack 1 win 10000* accept(3, ..., ...) = 4
测试结果，实际测试执行失败，提示了相关错误，意思是只能用于出站方向的数据包，也就是 > 。
# packetdrill tcp_3hs_time_008.pkttcp_3hs_time_008.pkt:6: semantic error: event time <star> can only be used with outbound packets#
继续修改脚本，执行成功。
# cat tcp_3hs_time_008.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>* > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+ accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_008.pkt#
抓包结果，一切正常。
# tcpdump -i any -nn port 8080tcpdump: data link type LINUX_SLL2tcpdump: verbose output suppressed, use -v[v]... for full protocol decodelistening on any, link-type LINUX_SLL2 (Linux cooked v2), snapshot length 262144 bytes21:00:46.828389 ? In IP 192.0.2.1.37099 > 192.168.40.249.8080: Flags [S], seq 0, win 10000, options [mss 1460], length 021:00:46.828402 ? Out IP 192.168.40.249.8080 > 192.0.2.1.37099: Flags [S.], seq 1935225306, ack 1, win 65535, options [mss 1460], length 021:00:46.838495 ? In IP 192.0.2.1.37099 > 192.168.40.249.8080: Flags [.], ack 1, win 10000, length 021:00:46.838545 ? Out IP 192.168.40.249.8080 > 192.0.2.1.37099: Flags [F.], seq 1, ack 1, win 65535, length 021:00:46.838553 ? In IP 192.0.2.1.37099 > 192.168.40.249.8080: Flags [R.], seq 1, ack 1, win 10000, length 0
实验测试 Range
初步修改脚本，同样只能用于出站方向的数据包，也就是 > ，时间范围为 0.044~0.054 。
# cat tcp_3hs_time_009.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>0.044~0.054 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4
测试结果，实际测试执行失败，提示了相关错误，事件发生的时间 0.000059 sec 不在时间范围 0.044000~0.054000 内。
# packetdrill tcp_3hs_time_009.pkttcp_3hs_time_009.pkt:7: error handling packet: timing error: expected outbound packet in time range 0.044000~0.054000 sec but happened at 0.000059 secscript packet: 0.044000 S. 0:0(0) ack 1actual packet: 0.000059 S. 0:0(0) ack 1 win 65535 <mss 1460>#
继续修改脚本，执行成功。
# cat tcp_3hs_time_009.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>0.004~0.005 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_009.pkt#
抓包结果，一切正常。
# tcpdump -i any -nn port 8080tcpdump: data link type LINUX_SLL2tcpdump: verbose output suppressed, use -v[v]... for full protocol decodelistening on any, link-type LINUX_SLL2 (Linux cooked v2), snapshot length 262144 bytes22:14:42.288390 ? In IP 192.0.2.1.43605 > 192.168.66.212.8080: Flags [S], seq 0, win 10000, options [mss 1460], length 022:14:42.288406 ? Out IP 192.168.66.212.8080 > 192.0.2.1.43605: Flags [S.], seq 386595060, ack 1, win 65535, options [mss 1460], length 022:14:42.298508 ? In IP 192.0.2.1.43605 > 192.168.66.212.8080: Flags [.], ack 1, win 10000, length 022:14:42.298567 ? Out IP 192.168.66.212.8080 > 192.0.2.1.43605: Flags [F.], seq 1, ack 1, win 65535, length 022:14:42.298576 ? In IP 192.0.2.1.43605 > 192.168.66.212.8080: Flags [R.], seq 1, ack 1, win 10000, length 0
当然也可以以绝对时间来编写测试，执行同样成功。
# cat tcp_3hs_time_010.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 30.01 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 00.02 bind(3, ..., ...) = 00.03 listen(3, 1) = 00.04 < S 0:0(0) win 10000 <mss 1460>0.044~0.054 > S. 0:0(0) ack 1 <...>0.05 < . 1:1(0) ack 1 win 100000.06 accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_010.pkt#
实验测试 Loose
Loose 模式实际是调整的 --tolerance_usecs 参数值，默认值是 4ms。
在《理解时间模型之相对时间》文章中，曾有一个测试脚本测试未成功，是由于 script packet: 0.004059 秒，还有一个自身执行的时间偏离，如果是 0.004059 秒，加上偏移就是在[0.000059,0.008059]内，真实数据包的 0.000054 秒不在其中，因此测试失败 。之后通过修改 0.004 为 0.003 后测试成功，因为这样 script packet 的执行时间加上偏移，肯定是在 [0.000000,0.007000]内，脚本会执行成功。
# cat tcp_3hs_time_006.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>+0.004 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_006.pkttcp_3hs_time_006.pkt:7: error handling packet: timing error: expected outbound packet at 0.004059 sec but happened at 0.000054 sec; tolerance 0.004000 secscript packet: 0.004059 S. 0:0(0) ack 1actual packet: 0.000054 S. 0:0(0) ack 1 win 65535 <mss 1460>#
那么通过修改 --tolerance_usecs 参数值为 5ms，也可以成功 ，因为这样 script packet 的执行时间加上偏移，肯定是在[0.000000,0.009000]内，脚本会执行成功。
# cat tcp_3hs_time_011.pkt--tolerance_usecs = 50000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>+0.004 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_011.pkt#
抓包结果，一切正常。
# tcpdump -i any -nn port 8080tcpdump: data link type LINUX_SLL2tcpdump: verbose output suppressed, use -v[v]... for full protocol decodelistening on any, link-type LINUX_SLL2 (Linux cooked v2), snapshot length 262144 bytes22:35:22.220397 ? In IP 192.0.2.1.40423 > 192.168.75.11.8080: Flags [S], seq 0, win 10000, options [mss 1460], length 022:35:22.220418 ? Out IP 192.168.75.11.8080 > 192.0.2.1.40423: Flags [S.], seq 3758488528, ack 1, win 65535, options [mss 1460], length 022:35:22.230509 ? In IP 192.0.2.1.40423 > 192.168.75.11.8080: Flags [.], ack 1, win 10000, length 022:35:22.230562 ? Out IP 192.168.75.11.8080 > 192.0.2.1.40423: Flags [F.], seq 1, ack 1, win 65535, length 022:35:22.230570 ? In IP 192.0.2.1.40423 > 192.168.75.11.8080: Flags [R.], seq 1, ack 1, win 10000, length 0
实验测试 Blocking
Blocking 模式也比较特殊，定义的是指定阻塞系统调用在给定的时间开始/返回。
基础脚本，通过strace输出系统调用时间，各系统调用时间间隔基本都在几十到百微秒之内。
# strace -Ttt -e trace=network packetdrill tcp_3hs_000.pkt...11:37:03.808371 socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 8 <0.000011>11:37:03.808504 setsockopt(8, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0 <0.000004>11:37:03.808584 bind(8, {sa_family=AF_INET, sin_port=htons(8080), sin_addr=inet_addr("192.168.103.108")}, 16) = 0 <0.000006>11:37:03.808656 listen(8, 1) = 0 <0.000006>...
修改脚本，预期0.10...0.105，测试执行失败，提示系统调用返回实际发生在 0.100521 sec，而期望的是在 0.105000 sec，也就是[0.101000,0.109000]内，因此失败。
# cat tcp_3hs_time_012.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 30.10...0.105 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>+0 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_012.pkttcp_3hs_time_012.pkt:2: timing error: expected system call return at 0.105000 sec but happened at 0.100521 sec; tolerance 0.004000 sec#
继续修改脚本，0.10...0.1043，也就是[0.100300,0.108300]内，执行成功。
# cat tcp_3hs_time_013.pkt0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 30.10...0.1043 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0+0 bind(3, ..., ...) = 0+0 listen(3, 1) = 0+0 < S 0:0(0) win 10000 <mss 1460>+0 > S. 0:0(0) ack 1 <...>+0.01 < . 1:1(0) ack 1 win 10000+0 accept(3, ..., ...) = 4## packetdrill tcp_3hs_time_013.pkt#
可以通过 -f 参数得到输出，可见 setsockopt(）相较于 socket() 开始时间确实间隔了约 10ms，符合 0.10 的设定。
...strace: Process 75716 attached[] 11:46:37.300375 socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 8 <0.000015>[] 11:46:37.400964 setsockopt(8, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0 <0.000008>[] 11:46:37.401169 bind(8, {sa_family=AF_INET, sin_port=htons(8080), sin_addr=inet_addr("192.168.77.51")}, 16) = 0 <0.000008>[] 11:46:37.401250 listen(8, 1) = 0 <0.000005>...
抓包结果，因为并没有修改三次握手的数据包，所以没有什么变化。
# tcpdump -i any -nn port 8080tcpdump: data link type LINUX_SLL2tcpdump: verbose output suppressed, use -v[v]... for full protocol decodelistening on any, link-type LINUX_SLL2 (Linux cooked v2), snapshot length 262144 bytes22:28:30.210863 ? In IP 192.0.2.1.59281 > 192.168.8.128.8080: Flags [S], seq 0, win 10000, options [mss 1460], length 022:28:30.210891 ? Out IP 192.168.8.128.8080 > 192.0.2.1.59281: Flags [S.], seq 1173665299, ack 1, win 65535, options [mss 1460], length 022:28:30.222750 ? In IP 192.0.2.1.59281 > 192.168.8.128.8080: Flags [.], ack 1, win 10000, length 022:28:30.223030 ? Out IP 192.168.8.128.8080 > 192.0.2.1.59281: Flags [F.], seq 1, ack 1, win 65535, length 022:28:30.223045 ? In IP 192.0.2.1.59281 > 192.168.8.128.8080: Flags [R.], seq 1, ack 1, win 10000, length 0
往期推荐
原文始发于微信公众号（Echo Reply）：Wireshark & Packetdrill | 理解时间模型之其他时间
- 左青龙
- 微信扫一扫
-
- 右白虎
- 微信扫一扫
-
评论