socket网络编程快速上手（二）——细节问题（2）

2.TCP数据包接收问题

  对初学者来说，很多都会认为：客户端与服务器最终的打印数据接收或者发送条数都该是一致的，1000条发送打印，1000条接收打印，长度都为1000。但是，事实上并不是这样，发送打印基本不会有什么问题（只是一般情况，如果发生调度或者其他情况，有可能导致差别，因此也要注意封装），接收打印却不是固定的，下面是测试代码：

测试客户端程序：

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netdb.h>

#define  PORT        1234
#define  MAXDATASIZE 1000

int main(int argc, char *argv[])
{
    int  sockfd, num;
    char  buf[MAXDATASIZE + 1] = {0};
    struct sockaddr_in server;
    int   iCount = 0;
    
    if (argc != 2) 
    {
        printf("Usage:%s <IP Address>
", argv[0]);
        exit(1);
    }
    
    if ((sockfd=socket(AF_INET, SOCK_STREAM, 0)) == -1)
    {
        printf("socket()error
");
        exit(1);
    }
    bzero(&server, sizeof(server));
    server.sin_family = AF_INET;
    server.sin_port = htons(PORT);
    server.sin_addr.s_addr = inet_addr(argv[1]);
    if (connect(sockfd, (struct sockaddr *)&server, sizeof(server)) == -1)
    {
        printf("connect()error
");
        exit(1);
    }

    while (1)
    {
        memset(buf, 0, sizeof(buf));
        if ((num = recv(sockfd, buf, MAXDATASIZE,0)) == -1)
        {
            printf("recv() error
");
            exit(1);
        }
        buf[num - 1]='';
        printf("%dth Recv Length: %d
", iCount++, num);
    }
    
    close(sockfd);
    
    return 0;
}

测试服务器程序：

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <signal.h>

#define  PORT         1234
#define  BACKLOG      5
#define  MAXDATASIZE  1000

int main()
{
    int  listenfd, connectfd;
    struct  sockaddr_in server;
    struct  sockaddr_in client;
    socklen_t  addrlen;
    char    szbuf[MAXDATASIZE] = {0};
    int     iCount = 0;
    int     iLength = 0;
    
    if ((listenfd = socket(AF_INET, SOCK_STREAM, 0)) == -1)
    {
        perror("Creating  socket failed.");
        exit(1);
    }
    
    int opt = SO_REUSEADDR;
    setsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt));
    
    bzero(&server, sizeof(server));
    server.sin_family = AF_INET;
    server.sin_port = htons(PORT);
    server.sin_addr.s_addr = htonl(INADDR_ANY);
    if (bind(listenfd, (struct sockaddr *)&server, sizeof(server)) == -1) 
    {
        perror("Bind()error.");
        exit(1);
    }   
    if (listen(listenfd, BACKLOG) == -1)
    {
        perror("listen()error
");
        exit(1);
    }
    
    addrlen = sizeof(client);
    if ((connectfd = accept(listenfd, (struct sockaddr*)&client, &addrlen)) == -1) 
    {
        perror("accept()error
");
        exit(1);
    }
    printf("You got a connection from cient's ip is %s, prot is %d
", inet_ntoa(client.sin_addr), htons(client.sin_port));

    memset(szbuf, 'a', sizeof(szbuf));
    while (iCount < 1000)
    {
        iLength = send(connectfd, szbuf, sizeof(szbuf), 0);
        printf("%dth Server Send Length %d
", iCount++, iLength);
    }
    
    printf("send over!
");
    sleep(10);
    
    close(connectfd);
    close(listenfd);
    
    return 0;
}

客户端接收打印片段如下：

936th Recv Length: 1000
937th Recv Length: 1000
938th Recv Length: 1000
939th Recv Length: 1000
940th Recv Length: 1000
941th Recv Length: 1000
942th Recv Length: 384
943th Recv Length: 616
944th Recv Length: 1000
945th Recv Length: 1000
946th Recv Length: 1000
947th Recv Length: 1000
948th Recv Length: 1000
949th Recv Length: 1000
950th Recv Length: 1000
951th Recv Length: 1000
952th Recv Length: 1000
953th Recv Length: 1000
954th Recv Length: 1000
955th Recv Length: 1000
956th Recv Length: 1000
957th Recv Length: 1000
958th Recv Length: 1000
959th Recv Length: 1000
960th Recv Length: 1000
961th Recv Length: 1000
962th Recv Length: 384
963th Recv Length: 616
964th Recv Length: 1000
965th Recv Length: 1000
966th Recv Length: 1000
967th Recv Length: 1000
968th Recv Length: 1000
969th Recv Length: 1000
970th Recv Length: 1000
971th Recv Length: 1000
972th Recv Length: 1000
973th Recv Length: 1000
974th Recv Length: 1000
975th Recv Length: 1000
976th Recv Length: 1000
977th Recv Length: 1000
978th Recv Length: 1000
979th Recv Length: 1000
980th Recv Length: 1000
981th Recv Length: 1000
982th Recv Length: 384
983th Recv Length: 616
984th Recv Length: 1000
985th Recv Length: 1000
986th Recv Length: 1000
987th Recv Length: 1000
988th Recv Length: 1000
989th Recv Length: 1000
990th Recv Length: 1000
991th Recv Length: 1000
992th Recv Length: 1000
993th Recv Length: 1000
994th Recv Length: 1000
995th Recv Length: 1000
996th Recv Length: 1000
997th Recv Length: 1000
998th Recv Length: 1000
999th Recv Length: 1000
1000th Recv Length: 1000
1001th Recv Length: 1000
1002th Recv Length: 384
1003th Recv Length: 616
1004th Recv Length: 1000
1005th Recv Length: 1000
1006th Recv Length: 1000
1007th Recv Length: 1000
1008th Recv Length: 1000
1009th Recv Length: 1000
1010th Recv Length: 1000
1011th Recv Length: 1000
1012th Recv Length: 1000
1013th Recv Length: 1000
1014th Recv Length: 1000
1015th Recv Length: 1000
1016th Recv Length: 1000
1017th Recv Length: 1000
1018th Recv Length: 1000
1019th Recv Length: 1000
1020th Recv Length: 1000
1021th Recv Length: 1000
1022th Recv Length: 384
1023th Recv Length: 616
1024th Recv Length: 1000
1025th Recv Length: 1000
1026th Recv Length: 1000
1027th Recv Length: 1000
1028th Recv Length: 1000
1029th Recv Length: 1000
1030th Recv Length: 1000
1031th Recv Length: 1000
1032th Recv Length: 1000
1033th Recv Length: 1000
1034th Recv Length: 1000
1035th Recv Length: 1000
1036th Recv Length: 1000
1037th Recv Length: 1000
1038th Recv Length: 1000
1039th Recv Length: 1000
1040th Recv Length: 1000
1041th Recv Length: 1000
1042th Recv Length: 384
1043th Recv Length: 616
1044th Recv Length: 1000
1045th Recv Length: 1000
1046th Recv Length: 1000
1047th Recv Length: 1000
1048th Recv Length: 1000
1049th Recv Length: 1000
1050th Recv Length: 1000

服务器发送打印片段整理时发现丢失了，大家可以自己试试，没有问题。

 不难发现，服务器发送正常，客户端在接收时却和我们想的很不一样，但发送和接收的总数据量是一致的，就是说数据没有丢失。如果编程者认为TCP情况下发送和接收的数据长度都一致的，那就极有可能在代码中体现出这一思想，最终出现问题。

  其实，这就是所谓的“粘包”现象，Stevens很明确地已经指出了这一点，他说，“UDP是长度固定的、无连接的不可靠报文传输；TCP是有序、可靠、双向的面向连接字节流”。他没说TCP是长度固定的，有没有？当然我更倾向于这样的理解，UDP是面向报文的，报文在传输时是不能被分割的（只是从应用层来看）；TCP是面向字节流的，接收多少数据完全取决于发送和接收的速度了，有多少数据recv就返回多少，数据长度并不和send保持一致，也没这个必要。

  那么这个问题怎么解决呢？其实，我们只要将recv封装一层就可以了，那就是我们熟悉的readn函数（该函数不是系统调用），代码如下：

int readn(int connfd, void *vptr, int n)
{
    int    nleft;
    int    nread;
    char *ptr;
    struct timeval     select_timeout;
    fd_set rset;

    ptr = vptr;
    nleft = n;

    while (nleft > 0)
    {
        FD_ZERO(&rset);
        FD_SET(connfd, &rset);
        select_timeout.tv_sec = 5;
        select_timeout.tv_usec = 0;
        if (select(connfd+1, &rset, NULL, NULL, &select_timeout) <= 0)
        {
            return -1;
        }
        if ((nread = recv(connfd, ptr, nleft, 0)) < 0)
        {
            if(errno == EINTR)
            {
                nread = 0;
            }
            else
            {
                return -1;
            }
        }
        else if (nread == 0)
        {
            break;
        }
        nleft -= nread;
        ptr   += nread;
    }
    return(n - nleft);
}

相应的也有writen函数

int writen(int connfd, void *vptr, size_t n)
{
    int nleft, nwritten;
     char    *ptr;

    ptr = vptr;
    nleft = n;

    while(nleft>0)
    {
        if((nwritten = send(connfd, ptr, nleft, 0)) == ERROR)
        {
            if(errnoGet() == EINTR)
            {
                //PRT_ERR(("EINTR
"));
                nwritten = 0;
            }
            else 
            {
                //PRT_ERR(("Send() error, 0x%x
", errnoGet()));
                return ERROR;
            }
        }
        nleft -= nwritten;
        ptr   += nwritten;
    }

    return(n);
}

函数中为什么对EINTR进行处理后面再说，也是必不可少的。

  在处理TCP发送和接收部分时，可以说必须要使用上述封装，否则等到造成数据不完整或者不一致后再去找问题，可能就麻烦了。这个是必不可少滴。