今天同学实现一个经典的Unix/Linux网络编程的例子--时间服务器的时候,遇到了一件怪事:“Connection refused” 。服务器端的端口号已经成功打开,客户端也能够ping通服务器端,但是就是不能建立TCP连接获取服务器时间。代码是很经典的代码,贴一下。
服务器端:
/* timeserv.c - a socket-based time of day server
*/
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netdb.h>
#include <time.h>
#include <strings.h>
#define PORTNUM 13000 /* our time service phone number */
#define HOSTLEN 256
#define oops(msg) { perror(msg) ; exit(1) ; }
int main(int ac, char *av[])
{
struct sockaddr_in saddr; /* build our address here */
struct hostent *hp; /* this is part of our */
char hostname[HOSTLEN]; /* address */
int sock_id,sock_fd; /* line id, file desc */
FILE *sock_fp; /* use socket as stream */
char *ctime(); /* convert secs to string */
time_t thetime; /* the time we report */
/*
* Step 1: ask kernel for a socket
*/
sock_id = socket( PF_INET, SOCK_STREAM, 0 ); /* get a socket */
if ( sock_id == -1 )
oops( "socket" );
/*
* Step 2: bind address to socket. Address is host,port
*/
bzero( (void *)&saddr, sizeof(saddr) ); /* clear out struct */
gethostname( hostname, HOSTLEN ); /* where am I ? */
hp = gethostbyname( hostname ); /* get info about host */
/* fill in host part */
bcopy( (void *)hp->h_addr, (void *)&saddr.sin_addr, hp->h_length);
saddr.sin_port = htons(PORTNUM); /* fill in socket port */
saddr.sin_family = AF_INET ; /* fill in addr family */
if ( bind(sock_id, (struct sockaddr *)&saddr, sizeof(saddr)) != 0 )
oops( "bind" );
/*
* Step 3: allow incoming calls with Qsize=1 on socket
*/
if ( listen(sock_id, 1) != 0 )
oops( "listen" );
/*
* main loop: accept(), write(), close()
*/
while ( 1 ){
sock_fd = accept(sock_id, NULL, NULL); /* wait for call */
printf("Wow! got a call!\n");
if ( sock_fd == -1 )
oops( "accept" ); /* error getting calls */
sock_fp = fdopen(sock_fd,"w"); /* we'll write to the */
if ( sock_fp == NULL ) /* socket as a stream */
oops( "fdopen" ); /* unless we can't */
thetime = time(NULL); /* get time */
/* and convert to strng */
fprintf( sock_fp, "The time here is .." );
fprintf( sock_fp, "%s", ctime(&thetime) );
fclose( sock_fp ); /* release connection */
}
}
客户端:
/* timeclnt.c - a client for timeserv.c
* usage: timeclnt hostname portnumber
*/
#include <stdio.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netdb.h>
#define oops(msg) { perror(msg); exit(1); }
main(int ac, char *av[])
{
struct sockaddr_in servadd; /* the number to call */
struct hostent *hp; /* used to get number */
int sock_id, sock_fd; /* the socket and fd */
char message[BUFSIZ]; /* to receive message */
int messlen; /* for message length */
/*
* Step 1: Get a socket
*/
sock_id = socket( AF_INET, SOCK_STREAM, 0 ); /* get a line */
if ( sock_id == -1 )
oops( "socket" ); /* or fail */
/*
* Step 2: connect to server
* need to build address (host,port) of server first
*/
bzero( &servadd, sizeof( servadd ) ); /* zero the address */
hp = gethostbyname( av[1] ); /* lookup host's ip # */
if (hp == NULL)
oops(av[1]); /* or die */
bcopy(hp->h_addr, (struct sockaddr *)&servadd.sin_addr, hp->h_length);
servadd.sin_port = htons(atoi(av[2])); /* fill in port number */
servadd.sin_family = AF_INET ; /* fill in socket type */
/* now dial */
if ( connect(sock_id,(struct sockaddr *)&servadd, sizeof(servadd)) !=0)
oops( "connect" );
/*
* Step 3: transfer data from server, then hangup
*/
messlen = read(sock_id, message, BUFSIZ); /* read stuff */
if ( messlen == - 1 )
oops("read") ;
if ( write( 1, message, messlen ) != messlen ) /* and write to */
oops( "write" ); /* stdout */
close( sock_id );
}
这个代码我至少在《Unix/Linux编程实践》和《Unix环境高级编程》中见过,但是拷到自己机器上运行,却依旧出现了相同的“Connection refused”的错误提示。Connection refused,就说明服务器端的端口没开放或者压根没连到服务器端。这到底是为什么呢,上面的代码找不到什么错误,纠结了一个晚上,终于搞明白了,其实还是在服务器端bind的时候出现差错了。
看上面服务器端的代码,注意获取本地网络地址以及设置端口号的那一块:
gethostname( hostname, HOSTLEN ); /* where am I ? */
hp = gethostbyname( hostname ); /* get info about host */
/* fill in host part */
bcopy( (void *)hp->h_addr, (void *)&saddr.sin_addr, hp->h_length);
saddr.sin_port = htons(PORTNUM); /* fill in socket port */
saddr.sin_family = AF_INET ; /* fill in addr family */
在这里,服务器程序使用了gethostname()函数来获取本机的网络地址,然后用gethostbyname()函数转化为网络字节序地址。问题就出在gethostname()函数身上。该函数的作用是返回主机名,通常就是TCP/IP网络上主机的名字,也就是IP地址。
这个函数在实现上,会先查找/etc/hosts文件的内容,然后查询DNS服务器。如果/etc/hosts文件没有配置,返回的主机名就是localhost也就是127.0.0.1 。所以,使用bind函数绑定的网络地址其实是一个127.0.0.1 。客户端想connect的话,就找不到连接了。
然后又查了本科时候学习网络编程时做的实验报告(很汗颜,要翻回去查本科时的资料)。借用了这样的处理方法,查询本地网络地址很好用。
稍作一下修改,部分代码如下:
/*
* Step 2: bind address to socket. Address is host,port
*/
bzero( (void *)&saddr, sizeof(saddr) ); /* clear out struct */
saddr.sin_addr.s_addr = htonl(INADDR_ANY);
saddr.sin_port = htons(PORTNUM); /* fill in socket port */
saddr.sin_family = AF_INET ; /* fill in addr family */
if ( bind(sock_id, (struct sockaddr *)&saddr, sizeof(saddr)) != 0 )
oops( "bind" );
关键部分就是红色那行,直接设置网络地址为INADDR_ANY 。有的服务器是多宿主机可能有多个网卡,那么运行在这样的服务器上的服务程序在为其Socket绑定IP地址时可以把htonl(INADDR_ANY)置给s_addr,这样做的好处是不论哪个网段上的客户程序都能与该服务程序通信;如果只给运行在多宿主机上的服务程序的Socket绑定一个固定的IP地址,那么就只有与该IP地址处于同一个网段上的客户程序才能与该服务程序通信。
就是如此这般!
转贴:http://hi.baidu.com/sky_space/blog/item/9fe37506d311647703088119.html