 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18


 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|          Source Port          |       Destination Port        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                        Sequence Number                        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                    Acknowledgement Number                      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|  Data |           |U|A|P|R|S|F|                               |
| Offset| Reserved  |R|C|S|S|Y|I|            Window             |
|       |           |G|K|H|T|N|N|                               |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|           Checksum            |         Urgent Pointer        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                    Options                    |    Padding    |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                             Data                              |

size: 20bytes ~ 60bytes(40 bytes options)

port: source or destination; 2^16;
number:sequence and acknowledge number；
flags
1. SYN
2. ACK
3. FIN
4. PST: push
5. RST:reset the connection
6. URG: urgent data (priority data)

client create connection port range:

1

net.ipv4.ip_local_port_range

61000 - 32768 = 28232

tcp status

server status: listen: passive open

client status: sync-sent: ative open

tcp tuple

tcp tuple: tcp use (srcip,srcport, desip, des) as key to identifyu a socket

server bind same port for tcp and udp: yes,不同协议，不会冲突

为什么重启服务，报错Address already in use: 重启时，服务器发起close,需要等待 2MSL 时间才能完全关闭connection, 持续占用ip+port

客户端time_wait耗尽端口: 当客户端对同一个server 发起请求后又迅速关闭conmnection

大量关闭请求在time_wait, port被占用

tcp vs udp

tcp vs udp:

more reliable: tcp
more fast: udp

reliable:

connection-oriented
1. 握手建立连接
2. 消息确认机制
error- check
1. checksum
2. 超时重传
流量调节
1. 流量控制
2. 拥塞控制

three-way handshake

process:

sync: syncX
ack+sync: ackX+1, syncY
ack: ackY+1

random sequeuece number: decrease attack 有效的 ack = syncNum+ 1

可以从服务端发起握手吗: 技术上可行, 需要公网ip + port, 客户端一般没有公网ip

handshake server

tcp server:

listen state: passive open
three handshake:
1. receive syn: create req socket
2. receive ack: move socket to accept queue
accept: dequeue socket from accept queue
read: read from accepted socket

code

 1
 2
 3
 4
 5
 6
 7
 8
 9
10


address := localhost+pport 
listFD = new socket();
bind(listenFD, address)
listen(listenFD); 
while(1){
   // get ready conne
   acceptFd =  accept(listenFD);  //
   recv(accetpFd, To: appBuffer);
   close(acceptFd)
}

syn flood attac: syncQueue.length > maxLenght, 导致其他请求无法连接 why: client故意不完成握手(发起syn 但不回复ack)

four-way terminate

process:

active: FinX
passive: ACKX+1
passive: FinY
active: ACK Y+1, 等待2MSL

2MSL: max segement lifetime, 报文传输时间的最大值

1

 sysctl net.ipv4.tcp_fin_timeout

等待2msl 后才关闭连接 , 在此期间

socket无法被复用
delay packet将被di

why 2MSL:

确保能重传ack 从而关闭被动方
确保排干本次连接延迟的包

为什么挥手需要四次:

完整的握手四次
握手阶段合并ack 和 syn
挥手的时候无法合并

ack and retransmission

ack

how:

sender: seq x, datasize=n
receiver: ack x+n

Retransmission

what: 当包丢失的时候，触发重传机制

retransmit timeout

RTT: round-trip time, 往返时间 smooth RTT: 平滑 RTT，加权平均RTT deviatedf RTT: 加权平均偏差 RTT偏差 smaple RTT: 采样RTT, 上一次 RTT

RTTs(somthed RTT) = (7/8) * old_RTTs + (1/8)RTTSample; RTTd = (3/4) * old_RTTd * 1/4(RTTSample-RTTs) RTO= RTTs + 4RTTd

feature:

能根据网络状况动态调整，网络越差，超时时间越长

fast retransmit

how: sender: 重复ack 最后一个有序的包 receiver: 连续收到三个重复的ack 会重传丢失的包

problem: 不知道完整的丢失包 for example :
sender: seq 1,2,3,4 receive:

receiv: 1, 4;
ack: 2,2,2, sender只知道seq2包已丢失，sender只发送seq包
then ack: 3,3,3, sender 再发送 seq3包

selective Ack

fast transmit + sack: 通过 sack告知已经收到的包

flow control

what: 发送方根据接收方的接受能力调整发送量；最大发送量= size of slide window，超过窗口大小，停止发送

window change:

sender:
1. window = unack + toBeSended
2. buffer = window + spare
receiver:
1. window = rec.Next + windowSize
2. buffer = unread + window +spare if unack incrase, sender window decreae if receiver window decrease, sender window decreae

receiver window decrease: unread data increase, window decrease

example:

window problem

zero windw and lost ack:

1
2
3


when windowsize==0
	startProbe:
		getWindowSize 

sillly winow: small packet problem, 持续的发送小包，没有充分利用带宽

solve:

sender：

1

if windowSize <= mss || buffer/2, windowSize =0

receiver: 累计足够的size 再发送

1
2


if  toBeSendDta >= mss, sent
if  unackDta==0, sent 

congestion control

what: 发送方根据网络情况动态调整发送量；

how: maxData<= congestion window

types:

慢启动: 窗口指数增加
拥塞避免: 窗口线性增加
拥塞发生
快速恢复

tcp keep alive

what: 通过发送探测包确保tcp连接alive的机制

tcp非正常关闭:

对方节点crash
被proxy 关闭, delete from connection table, and drop packet

configs:

tcp_keepalive_time: 7200
tcp_keepalive_probes: 9
tcp_keepalive_intvl: 75

how:

If the connection is silent for tcp_keepalive_time seconds, send a single empty ACK packet
Did the server respond with a corresponding ACK of its own?
- No
  1. Wait tcp_keepalive_intvl seconds, then send another ACK
  2. Repeat until the number of ACK probes that have been sent equals tcp_keepalive_probes.
  3. If no response has been received at this point, send a RST and terminate the connection.
- Yes: Return to step 1

数据包拆和黏包

divide;

1. how

data must divide into N block; the data is limited by phyical trasmit size;

tcp layer:

Get MSS(max segement size) By handshake; Dymanic

ip layer:

Get MTU(Maximum transmission unit) By Mtu discovery; Dymanic

tcp

userData Nsize: 1. tcp segement: {segement1…segement2| segement.size=mss, } 2. Ip datagram:{datagram1…datagramN | datagram.size <= mtu}

黏包(sticky)

what ?

get image1,image2,image3, 复用同一个请求；

数据流: 111112222233333;

一个连接里存在多个请求的数据，用户端如何知道这个数据长度(边界)从而正确读取数据；
how to resolve
1. 元数据指定大小；
  
  content-length: NK;
  
  Read(fd, Lenght: N);
2. 终结符(delimiter):
  
  1111\02222\0 ;

Contents

tcp protocol

tcp header && status

header