TCP网络通讯如何解决分包粘包问题

TCP作为常用的网络传输协议，数据流解析是网络应用开发人员永远绕不开的一个问题。
TCP数据传输是以无边界的数据流传输形式，所谓无边界是指数据发送端发送的字节数，在数据接收端接受时并不一定等于发送的字节数，可能会出现粘包情况。

一、TCP粘包情况：

1. 发送端发送了数量比较的数据，接收端读取数据时候数据分批到达，造成一次发送多次读取；通常网络路由的缓存大小有关系，一个数据段大小超过缓存大小，那么就要拆包发送。
2. 发送端发送了几次数据，接收端一次性读取了所有数据，造成多次发送一次读取；通常是网络流量优化，把多个小的数据段集满达到一定的数据量，从而减少网络链路中的传输次数。



TCP粘包的解决方案有很多种方法，最简单的一种就是发送的数据协议定义发送的数据包的结构：
1. 数据头：数据包的大小，固定长度。
2. 数据内容：数据内容，长度为数据头定义的长度大小。
实际操作如下：
a）发送端：先发送数据包的大小，再发送数据内容。
b）接收端：先解析本次数据包的大小N，在读取N个字节，这N个字节就是一个完整的数据内容。
具体流程如下：



实现源码

/**
* read size of len from sock into buf.
*/
bool readPack(int sock, char* buf, size_t len) {
if (NULL == buf || len < 1) {
return false;
}
memset(buf, 0, len); // only reset buffer len.
ssize_t read_len = 0, readsum = 0;
do {
read_len = read(sock, buf + readsum, len - readsum);
if (-1 == read_len) { // ignore error case
return false;
}
printf("receive data: %s\n", buf + readsum);
readsum += read_len;
} while (readsum < len && 0 != read_len);
return true;
}

二、测试用例介绍

本篇提供的demo主要流程如下：
1. 客户端负责模拟发送数据，服务端负责接受数据，处理粘包问题
a）emulate_subpackage
模拟情况1，一个长数据经过多次才到达目的地，
在客户端字符串“This is a test case for client send subpackage data. data is not send complete at once.”每次只发送6个字节长度。服务端要把字符串集满才能处理数据（打印字符串）
b）emulate_adheringpackage
模拟情况2，多个数据在一次性到达目的地
在客户端将字符串“Hello I'm lucky. Nice too me you”切成三个数据段（都包含数据头和数据内容），然后一次性发送，服务端读取数据时对三个数据段逐个处理。

三、源码实现

server.cpp

#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <errno.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <arpa/inet.h>
#include <unistd.h>

void newclient(int sock);
bool readPack(int sock, char* buf, size_t len);
void safe_close(int &sock);

int main(int argc, char *argv[]) {
int sockfd = -1, newsockfd = -1;
socklen_t c = 0;
struct sockaddr_in serv_addr, cli_addr;

// Create socket
sockfd = socket(AF_INET, SOCK_STREAM, 0);
if (-1 == sockfd) {
printf("new socket failed. errno: %d, error: %s\n", errno, strerror(errno));
exit(-1);
}

// Prepare the sockaddr_in structure
serv_addr.sin_family = AF_INET;
serv_addr.sin_addr.s_addr = INADDR_ANY;
serv_addr.sin_port = htons(7890);

// bind
if (bind(sockfd, (struct sockaddr *)&serv_addr, sizeof(serv_addr)) < 0) {
printf("bind failed. errno: %d, error: %s\n", errno, strerror(errno));
exit(-1);
}

// listen
listen(sockfd, 5);

printf("listening...\n");
// accept new connection.
c = sizeof(struct sockaddr_in);
int i = 0;
while (i++ < 3) {
printf("waiting for new socket accept.\n");
newsockfd = accept(sockfd, (struct sockaddr*)&cli_addr, (socklen_t*)&c);
if (newsockfd < 0) {
printf("accept connect failed. errno: %d, error: %s\n", errno, strerror(errno));
safe_close(sockfd);
exit(-1);
}
pid_t pid = fork();
if (0 == pid) {
newclient(newsockfd);
safe_close(sockfd);
break;
} else if (pid > 0) {
safe_close(newsockfd);
}
}
safe_close(sockfd);
return 0;
}

void newclient(int sock) {
printf("newclient sock fd: %d\n", sock);
int datasize = 0;
const int HEAD_SIZE = 9;
char buf[512] = {0};
while (true) {
memset(buf, 0, sizeof(buf));
if (! readPack(sock, buf, HEAD_SIZE)) {
printf("read head buffer failed.\n");
safe_close(sock);
return;
}

datasize = atoi(buf);
printf("data size: %s, value:%d\n", buf, datasize);
memset(buf, 0, sizeof(buf));
if (! readPack(sock, buf, datasize)) {
printf("read data buffer failed\n");
safe_close(sock);
return;
}
printf("data size: %d, text: %s\n", datasize, buf);
if (0 == strcmp(buf, "exit")) {
break;
}
}
memset(buf, 0, sizeof(buf));
snprintf(buf, sizeof(buf), "from server read complete.");
write(sock, buf, strlen(buf) + 1);
printf("newclient sockfd: %d, finish.\n", sock);
safe_close(sock);
}

void safe_close(int &sock) {
if (sock > 0) {
close(sock);
sock = -1;
}
}

/**
* read size of len from sock into buf.
*/
bool readPack(int sock, char* buf, size_t len) {
if (NULL == buf || len < 1) {
return false;
}
memset(buf, 0, len); // only reset buffer len.
ssize_t read_len = 0, readsum = 0;
do {
read_len = read(sock, buf + readsum, len - readsum);
if (-1 == read_len) { // ignore error case
return false;
}
printf("receive data: %s\n", buf + readsum);
readsum += read_len;
} while (readsum < len && 0 != read_len);
return true;
}

client.cpp

#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <time.h>
#include <errno.h>
#include <sys/socket.h>
#include <arpa/inet.h>
#include <unistd.h>

void safe_close(int &sock);
void emulate_subpackage(int sock);
void emulate_adheringpackage(int sock);

int main(int argc, char *argv[]) {
char buf[128] = {0};
int sockfd = -1;
struct sockaddr_in serv_addr;

// Create sock
sockfd = socket(AF_INET, SOCK_STREAM, 0);
if (-1 == sockfd) {
printf("new socket failed. errno: %d, error: %s\n", errno, strerror(errno));
exit(-1);
}

serv_addr.sin_addr.s_addr = inet_addr("127.0.0.1");
serv_addr.sin_family = AF_INET;
serv_addr.sin_port = htons(7890);

// Connect to remote server
if (connect(sockfd, (struct sockaddr *)&serv_addr, sizeof(serv_addr)) < 0) {
printf("connection failed. errno: %d, error: %s\n", errno, strerror(errno));
exit(-1);
}
emulate_subpackage(sockfd);
emulate_adheringpackage(sockfd);

const int HEAD_SIZE = 9;
const char temp[] = "exit";
memset(buf, 0, sizeof(buf));
snprintf(buf, sizeof(buf), "%0.*zu", HEAD_SIZE - 1, sizeof(temp));
write(sockfd, buf, HEAD_SIZE);
write(sockfd, temp, sizeof(temp));

printf("send complete.\n");
memset(buf, 0, sizeof(buf));
read(sockfd, buf, sizeof(buf));
printf("receive data: %s\n", buf);
printf("client finish.\n");

safe_close(sockfd);
return 0;
}

void safe_close(int &sock) {
if (sock > 0) {
close(sock);
sock = -1;
}
}

/**
* emulate socket data write multi part.
*/
void emulate_subpackage(int sock) {
printf("emulate_subpackage...\n");
char text[] = "This is a test case for client send subpackage data. data is not send complete at once.";
const size_t TEXTSIZE = sizeof(text);
ssize_t len = 0;
size_t sendsize = 0, sendsum = 0;

const int HEAD_SIZE = 9;
char buf[64] = {0};
snprintf(buf, HEAD_SIZE, "%08zu", TEXTSIZE);
write(sock, buf, HEAD_SIZE);
printf("send data size: %s\n", buf);

do {
sendsize = 6;
if (sendsum + sendsize > TEXTSIZE) {
sendsize = TEXTSIZE - sendsum;
}
len = write(sock, text + sendsum, sendsize);
if (-1 == len) {
printf("send data failed. errno: %d, error: %s\n", errno, strerror(errno));
return;
}
memset(buf, 0, sizeof(buf));
snprintf(buf, len + 1, text + sendsum);
printf("send data: %s\n", buf);
sendsum += len;
sleep(1);
} while (sendsum < TEXTSIZE && 0 != len);
}

/**
* emualte socket data write adhering.
*/
void emulate_adheringpackage(int sock) {
printf("emulate_adheringpackage...\n");
const int HEAD_SIZE = 9;
char buf[1024] = {0};
char text[128] = {0};
char *pstart = buf;

// append text
memset(text, 0, sizeof(text));
snprintf(text, sizeof(text), "Hello ");
snprintf(pstart, HEAD_SIZE, "%08zu", strlen(text) + 1);
pstart += HEAD_SIZE;
snprintf(pstart, strlen(text) + 1, "%s", text);
pstart += strlen(text) + 1;

// append text
memset(text, 0, sizeof(text));
snprintf(text, sizeof(text), "I'm lucky.");
snprintf(pstart, HEAD_SIZE, "%08zu", strlen(text) + 1);
pstart += HEAD_SIZE;
snprintf(pstart, strlen(text) + 1, "%s", text);
pstart += strlen(text) + 1;

// append text
memset(text, 0, sizeof(text));
snprintf(text, sizeof(text), "Nice too me you");
snprintf(pstart, HEAD_SIZE, "%08zu", strlen(text) + 1);
pstart += HEAD_SIZE;
snprintf(pstart, strlen(text) + 1, "%s", text);
pstart += strlen(text) + 1;
write(sock, buf, pstart - buf);
}

Makefile

CC=g++
CFLAGS=-I

all: server.o client.o

server.o: server.cpp
$(CC) -o server.o server.cpp

client.o: client.cpp
$(CC) -o client.o client.cpp

clean:
rm *.o

四、测试结果

编译及运行
$ make
g++ -o server.o server.cpp
g++ -o client.o client.cpp
客户端模拟发送数据
$ ./client.o
emulate_subpackage...
send data size: 00000088
send data: This i
send data: s a te
send data: st cas
send data: e for
send data: client
send data: send
send data: subpac
send data: kage d
send data: ata. d
send data: ata is
send data: not s
send data: end co
send data: mplete
send data: at on
send data: ce.
emulate_adheringpackage...
send complete.
receive data: from server read complete.
client finish.

服务端模拟接受数据
$ ./server.o
listening...
waiting for new socket accept.
waiting for new socket accept.
newclient sock fd: 4
receive data: 00000088
data size: 00000088, value:88
receive data: This i
receive data: s a te
receive data: st cas
receive data: e for
receive data: client
receive data: send
receive data: subpac
receive data: kage d
receive data: ata. d
receive data: ata is
receive data: not s
receive data: end co
receive data: mplete
receive data: at on
receive data: ce.
data size: 88, text: This is a test case for client send subpackage data. data is not send complete at once.
receive data: 00000007
data size: 00000007, value:7
receive data: Hello
data size: 7, text: Hello
receive data: 00000011
data size: 00000011, value:11
receive data: I'm lucky.
data size: 11, text: I'm lucky.
receive data: 00000016
data size: 00000016, value:16
receive data: Nice too me you
data size: 16, text: Nice too me you
receive data: 00000005
data size: 00000005, value:5
receive data: exit
data size: 5, text: exit
newclient sockfd: 4, finish.