webbench源码浅析

cd webbench
sudo make
sudo make install webbenchpath
./webbench -c 40 -t 10 -f --get http://www.baidu.com/index.html

//socket.c
/* $Id: socket.c 1.1 1995/01/01 07:11:14 cthuang Exp $
 *
 * This module has been modified by Radim Kolar for OS/2 emx
 */

/***********************************************************************
  module:       socket.c
  program:      popclient
  SCCS ID:      @(#)socket.c    1.5  4/1/94
  programmer:   Virginia Tech Computing Center
  compiler:     DEC RISC C compiler (Ultrix 4.1)
  environment:  DEC Ultrix 4.3 
  description:  UNIX sockets code.
 ***********************************************************************/

#include <sys/types.h>
#include <sys/socket.h>
#include <fcntl.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netdb.h>
#include <sys/time.h>
#include <string.h>
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>

int Socket(const char *host, int clientPort)
{
    int sock;
    unsigned long inaddr;
    struct sockaddr_in ad;
    struct hostent *hp;

    /*
    初始化地址
    struct sockaddr_in{
        short sin_family;//Address family一般来说AF_INET（地址族）PF_INET（协议族）
        unsigned short sin_port;//Port number(必须要采用网络数据格式,普通数字可以用htons()函数转换成网络数据格式的数字)
        struct in_addr sin_addr;//IP address in network byte order（Internet address）
        unsigned char sin_zero[8];//Same size as struct sockaddr没有实际意义,只是为了　跟SOCKADDR结构在内存中对齐
    };
    struct in_addr {
        in_addr_t s_addr;
    };
    in_addr_t 一般为 32位的unsigned int,其字节顺序为网络顺序(大端字节序)
    */

    memset(&ad, 0, sizeof(ad));

    //AF_INET（又称 PF_INET）是 IPv4 网络协议的套接字类型,两者有着相同的宏定义.
    ad.sin_family = AF_INET;

    //inet_addr() 将字符串形式的IP地址 -> 网络字节顺序  的整型值
    inaddr = inet_addr(host);

    //INADDR_NONE 是个宏定义,代表无效的IP地址
    if (inaddr != INADDR_NONE)
        memcpy(&ad.sin_addr, &inaddr, sizeof(inaddr));
    else
    {
        /*
        gethostbyname()通过名字获得主机的相关信息
        返回值是一个结构体
         struct hostent{
            char *h_name;
            char ** h_aliases;
            short h_addrtype;
            short h_length;
            char ** h_addr_list;
        };
        */
        hp = gethostbyname(host);
        if (hp == NULL)
            return -1;

        //此处用h_addr是因为有这样一个宏定义
        //#define h_addr h_addr_list[0]
        memcpy(&ad.sin_addr, hp->h_addr, hp->h_length);
    }
    /*
    htons主机字节序 >> 网络字节序
    主机:小端字节序
    网络:大端字节序
    htonl()--"Host to Network Long"
    ntohl()--"Network to Host Long"
    htons()--"Host to Network Short"
    ntohs()--"Network to Host Short"
    */
    ad.sin_port = htons(clientPort);

    /*
    socket()函数用于根据指定的地址族、数据类型和协议来分配一个套接口的描述字及其所用的资源。
    SOCK_STREAM 提供有序的、可靠的、双向的和基于连接的字节流，使用带外数据传送机制，为Internet地址族使用TCP。
    SOCK_DGRAM 支持无连接的、不可靠的和使用固定大小（通常很小）缓冲区的数据报服务，为Internet地址族使用UDP。
    */
    sock = socket(AF_INET, SOCK_STREAM, 0);
    if (sock < 0)
        return sock;
    //connect()用于建立与指定socket的连接。
    if (connect(sock, (struct sockaddr *)&ad, sizeof(ad)) < 0)
        return -1;
    return sock;
}

webbench.c
/*
* (C) Radim Kolar 1997-2004
* This is free software, see GNU Public License version 2 for
* details.
*
* Simple forking WWW Server benchmark:
*
* Usage:
*   webbench --help
*
* Return codes:
*    0 - sucess
*    1 - benchmark failed (server is not on-line)
*    2 - bad param
*    3 - internal error, fork failed
* 
*/
#include "socket.c"
#include <unistd.h>
#include <sys/param.h>
#include <rpc/types.h>
#include <getopt.h>
#include <strings.h>
#include <time.h>
#include <signal.h>

/* values */
//volatile 防止编译器对代码进行优化
//如果到达规定的时间,把该变量置1
volatile int timerexpired=0;
//存储成功的http请求数,最后除以时间,就得到速度.
int speed=0;
//存储失败http连接数
int failed=0;
//传输过的字节数
int bytes=0;

/* globals */
int http10=1; /* 0 - http/0.9, 1 - http/1.0, 2 - http/1.1 */
/* Allow: GET, HEAD, OPTIONS, TRACE */
#define METHOD_GET 0
#define METHOD_HEAD 1
#define METHOD_OPTIONS 2
#define METHOD_TRACE 3
#define PROGRAM_VERSION "1.5"
int method=METHOD_GET;
int clients=1;
//如果命令行参数加了-f,就会将force置1,代表忽略返回的数据
int force=0;
//强制加载,如果添加-r参数,会置1,会将http的pragma字段设置为no-cache,即不使用缓存
int force_reload=0;
int proxyport=80;
char *proxyhost=NULL;
int benchtime=30;

//父子进程用来通信的管道
/* internal */
int mypipe[2];
char host[MAXHOSTNAMELEN];
#define REQUEST_SIZE 2048
char request[REQUEST_SIZE];

/*
struct option指明了一个“长参数”（即形如--name的参数）名称和性质

struct option {
    const char *name; 参数名
    int has_arg； 0-无参数 1-后面一定跟个参数 2-可跟可不跟
    int *flag; 用来决定getopt_long()的返回值是什么,flag是null，则函数会返回与该项option匹配的val值
               参数不为空，那么当选中某个长选项的时候，getopt_long将返回0，并且将flag指针参数指向val值
    int val; 和flag联合决定返回值
}
 */
static const struct option long_options[]=
{
    {"force",no_argument,&force,1},
    {"reload",no_argument,&force_reload,1},
    {"time",required_argument,NULL,'t'},
    {"help",no_argument,NULL,'?'},
    {"http09",no_argument,NULL,'9'},
    {"http10",no_argument,NULL,'1'},
    {"http11",no_argument,NULL,'2'},
    {"get",no_argument,&method,METHOD_GET},
    {"head",no_argument,&method,METHOD_HEAD},
    {"options",no_argument,&method,METHOD_OPTIONS},
    {"trace",no_argument,&method,METHOD_TRACE},
    {"version",no_argument,NULL,'V'},
    {"proxy",required_argument,NULL,'p'},
    {"clients",required_argument,NULL,'c'},
    {NULL,0,NULL,0}
};

/* prototypes原型 */
static void benchcore(const char* host,const int port, const char *request);
static int bench(void);
static void build_request(const char *url);

//子进程中的信号处理函数
static void alarm_handler(int signal)
{
    timerexpired=1;
}	

//用法
static void usage(void)
{
    fprintf(stderr,
            "webbench [option]... URL\n"
            "  -f|--force               Don't wait for reply from server.\n"
            "  -r|--reload              Send reload request - Pragma: no-cache.\n"
            "  -t|--time <sec>          Run benchmark for <sec> seconds. Default 30.\n"
            "  -p|--proxy <server:port> Use proxy server for request.\n"
            "  -c|--clients <n>         Run <n> HTTP clients at once. Default one.\n"
            "  -9|--http09              Use HTTP/0.9 style requests.\n"
            "  -1|--http10              Use HTTP/1.0 protocol.\n"
            "  -2|--http11              Use HTTP/1.1 protocol.\n"
            "  --get                    Use GET request method.\n"
            "  --head                   Use HEAD request method.\n"
            "  --options                Use OPTIONS request method.\n"
            "  --trace                  Use TRACE request method.\n"
            "  -?|-h|--help             This information.\n"
            "  -V|--version             Display program version.\n"
           );
}

int main(int argc, char *argv[])
{
    int opt=0;
    int options_index=0;
    char *tmp=NULL;

    if(argc==1)
    {
        usage();
        return 2;
    } 

    /*
    命令行参数可以分为两类，一类是短选项，一类是长选项，短选项在参数前加一杠"-"，长选项在参数前连续加两杠"--"
    getopt函数只能处理短选项，而getopt_long函数两者都可以
    int getopt_long(int argc,char* const argv[],const char *optstring,const struct option *longopts,int *longindex);
    optstring: 表示短选项字符串
    形式如“a:b::cd:“，分别表示程序支持的命令行短选项有-a、-b、-c、-d，冒号含义如下：
    (1)只有一个字符，不带冒号——只表示选项，如-c 
    (2)一个字符，后接一个冒号——表示选项后面带一个参数，如-a 100
    (3)一个字符，后接两个冒号——表示选项后面带一个可选参数，即参数可有可无，如果带参数，则选项与参数直接不能有空格如-b200
     */
    while((opt=getopt_long(argc,argv,"912Vfrt:p:c:?h",long_options,&options_index))!=EOF )
    {
        switch(opt)
        {
            case  0 : break;
            case 'f': force=1;break;
            case 'r': force_reload=1;break; 
            case '9': http10=0;break;
            case '1': http10=1;break;
            case '2': http10=2;break;
            case 'V': printf(PROGRAM_VERSION"\n");exit(0);
            case 't': benchtime=atoi(optarg);break;	     
            case 'p': 
            /* proxy server parsing server:port */
            // strrchr(const char *str, int c)
            // 在参数str所指向的字符串中搜索最后一次出现字符c(一个无符号字符)的位置,如果未找到该值，则函数返回一个空指针
            tmp=strrchr(optarg,':');
            proxyhost=optarg;
            if(tmp==NULL)
            {
                break;
            }
            if(tmp==optarg)
            {
                fprintf(stderr,"Error in option --proxy %s: Missing hostname.\n",optarg);
                return 2;
            }
            if(tmp==optarg+strlen(optarg)-1)
            {
                fprintf(stderr,"Error in option --proxy %s Port number is missing.\n",optarg);
                return 2;
            }
            *tmp='\0';
            //把字符串转换为一个整数
            proxyport=atoi(tmp+1);break;
            case ':':
            case 'h':
            case '?': usage();return 2;//break; 这里的break永远不会执行到呀
            case 'c': clients=atoi(optarg);break;
        }
    }

    //optind：表示的是下一个将被处理到的参数在argv中的下标值。
    if(optind==argc) {
        fprintf(stderr,"webbench: Missing URL!\n");
        usage();
        return 2;
    }

    if(clients==0) clients=1;
    if(benchtime==0) benchtime=30;
 
    /* Copyright */
    fprintf(stderr,"Webbench - Simple Web Benchmark "PROGRAM_VERSION"\n"
            "Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.\n"
            );

    //构造http的请求消息
    build_request(argv[optind]);
 
    // print request info ,do it in function build_request
    /*printf("Benchmarking: ");
 
    switch(method)
    {
        case METHOD_GET:
        default:
        printf("GET");break;
        case METHOD_OPTIONS:
        printf("OPTIONS");break;
        case METHOD_HEAD:
        printf("HEAD");break;
        case METHOD_TRACE:
        printf("TRACE");break;
    }
    
    printf(" %s",argv[optind]);
    
    switch(http10)
    {
        case 0: printf(" (using HTTP/0.9)");break;
        case 2: printf(" (using HTTP/1.1)");break;
    }
 
    printf("\n");
    */

    printf("Runing info: ");

    if(clients==1) 
        printf("1 client");
    else
        printf("%d clients",clients);

    printf(", running %d sec", benchtime);
    
    if(force) printf(", early socket close");
    if(proxyhost!=NULL) printf(", via proxy server %s:%d",proxyhost,proxyport);
    if(force_reload) printf(", forcing reload");
    
    printf(".\n");
    
    return bench();
}

//用来创建http的请求
void build_request(const char *url)
{
    char tmp[10];
    int i;

    //bzero(host,MAXHOSTNAMELEN);
    //bzero(request,REQUEST_SIZE);
    memset(host,0,MAXHOSTNAMELEN);
    memset(request,0,REQUEST_SIZE);

    if(force_reload && proxyhost!=NULL && http10<1) http10=1;
    if(method==METHOD_HEAD && http10<1) http10=1;
    if(method==METHOD_OPTIONS && http10<2) http10=2;
    if(method==METHOD_TRACE && http10<2) http10=2;

    switch(method)
    {
        default:
        case METHOD_GET: strcpy(request,"GET");break;
        case METHOD_HEAD: strcpy(request,"HEAD");break;
        case METHOD_OPTIONS: strcpy(request,"OPTIONS");break;
        case METHOD_TRACE: strcpy(request,"TRACE");break;
    }

    strcat(request," ");

    if(NULL==strstr(url,"://"))
    {
        fprintf(stderr, "\n%s: is not a valid URL.\n",url);
        exit(2);
    }
    if(strlen(url)>1500)
    {
        fprintf(stderr,"URL is too long.\n");
        exit(2);
    }
    if (0!=strncasecmp("http://",url,7)) 
    { 
        fprintf(stderr,"\nOnly HTTP protocol is directly supported, set --proxy for others.\n");
        exit(2);
    }
    
    /* protocol/host delimiter */
    i=strstr(url,"://")-url+3;

    if(strchr(url+i,'/')==NULL) {
        fprintf(stderr,"\nInvalid URL syntax - hostname don't ends with '/'.\n");
        exit(2);
    }
    
    if(proxyhost==NULL)
    {
        /* get port from hostname */
        if(index(url+i,':')!=NULL && index(url+i,':')<index(url+i,'/'))
        {
            strncpy(host,url+i,strchr(url+i,':')-url-i);
            //bzero(tmp,10);
            memset(tmp,0,10);
            strncpy(tmp,index(url+i,':')+1,strchr(url+i,'/')-index(url+i,':')-1);
            /* printf("tmp=%s\n",tmp); */
            proxyport=atoi(tmp);
            if(proxyport==0) proxyport=80;
        } 
        else
        {
            strncpy(host,url+i,strcspn(url+i,"/"));
        }
        // printf("Host=%s\n",host);
        strcat(request+strlen(request),url+i+strcspn(url+i,"/"));
    } 
    else
    {
        // printf("ProxyHost=%s\nProxyPort=%d\n",proxyhost,proxyport);
        strcat(request,url);
    }

    if(http10==1)
        strcat(request," HTTP/1.0");
    else if (http10==2)
        strcat(request," HTTP/1.1");
  
    strcat(request,"\r\n");
  
    if(http10>0)
        strcat(request,"User-Agent: WebBench "PROGRAM_VERSION"\r\n");
    if(proxyhost==NULL && http10>0)
    {
        strcat(request,"Host: ");
        strcat(request,host);
        strcat(request,"\r\n");
    }
 
    if(force_reload && proxyhost!=NULL)
    {
        strcat(request,"Pragma: no-cache\r\n");
    }
  
    if(http10>1)
        strcat(request,"Connection: close\r\n");
    
    /* add empty line at end */
    if(http10>0) strcat(request,"\r\n"); 
    
    printf("\nRequest:\n%s\n",request);
}

//用来创建多个子进程,然后每个子进程会分别调用benchcore函数对服务器进行测试.
/* vraci system rc error kod */
static int bench(void)
{
    int i,j,k;	
    pid_t pid=0;
    FILE *f;

    //先测试一下此网址是否正常.
    /* check avaibility of target server */
    i=Socket(proxyhost==NULL?host:proxyhost,proxyport);
    if(i<0) { 
        fprintf(stderr,"\nConnect to server failed. Aborting benchmark.\n");
        return 1;
    }
    close(i);
    
    /* create pipe */
    if(pipe(mypipe))
    {
        perror("pipe failed.");
        return 3;
    }

    /* not needed, since we have alarm() in childrens */
    /* wait 4 next system clock tick */
    /*
    cas=time(NULL);
    while(time(NULL)==cas)
    sched_yield();
    */

    /* fork childs */
    for(i=0;i<clients;i++)
    {
        pid=fork();
        if(pid <= (pid_t) 0)
        {
            /* child process or error*/
            sleep(1); /* make childs faster */
            break;
        }
    }

    if( pid < (pid_t) 0)
    {
        fprintf(stderr,"problems forking worker no. %d\n",i);
        perror("fork failed.");
        return 3;
    }

    if(pid == (pid_t) 0)
    {
        //子进程
        /* I am a child */
        if(proxyhost==NULL)
            benchcore(host,proxyport,request);
        else
            benchcore(proxyhost,proxyport,request);

        /* write results to pipe */
        f=fdopen(mypipe[1],"w");
        if(f==NULL)
        {
            perror("open pipe for writing failed.");
            return 3;
        }
        //向管道写入小于PIPE_BUF的数据的操作可以保证是原子的
        /* fprintf(stderr,"Child - %d %d\n",speed,failed); */
        fprintf(f,"%d %d %d\n",speed,failed,bytes);
        fclose(f);

        return 0;
    } 
    else
    {
        //父进程
        f=fdopen(mypipe[0],"r");
        if(f==NULL) 
        {
            perror("open pipe for reading failed.");
            return 3;
        }

        /*
        设置流的缓冲区
        int setvbuf(FILE *stream, char *buf, int type, unsigned size);
        type ： 期望缓冲区的类型：
        _IOFBF(满缓冲）：当缓冲区为空时，从流读入数据。或者当缓冲区满时，向流写入数 据。
        _IOLBF(行缓冲）：每次从流中读入一行数据或向流中写入一行数据。
        _IONBF(无缓冲）：直接从流中读入数据或直接向流中写入数据，而没有缓冲区。
        size ： 缓冲区内字节的数量。
        */
        setvbuf(f,NULL,_IONBF,0);
        
        speed=0;
        failed=0;
        bytes=0;

        //父进程不断地对管道进行读取,汇总每个子进程的测试结果.
        while(1)
        {
            //当管道中无数据时,应该会阻塞吧
            pid=fscanf(f,"%d %d %d",&i,&j,&k);
            if(pid<2)
            {
                fprintf(stderr,"Some of our childrens died.\n");
                break;
            }
            
            speed+=i;
            failed+=j;
            bytes+=k;
        
            /* fprintf(stderr,"*Knock* %d %d read=%d\n",speed,failed,pid); */
            if(--clients==0) break;
        }
    
        fclose(f);

        printf("\nSpeed=%d pages/min, %d bytes/sec.\nRequests: %d susceed, %d failed.\n",
            (int)((speed+failed)/(benchtime/60.0f)),
            (int)(bytes/(float)benchtime),
            speed,
            failed);
    }
    
    return i;
}

//测试的核心函数,该函数在子进程中执行,不断地发送http请求,force==0就会接受数据,force!=0就忽略掉返回的数据.
//该函数给自己设置了个定时器,到达时间就会让自己把timeexpired置为1,然后会退出循环.
void benchcore(const char *host,const int port,const char *req)
{
    int rlen;
    char buf[1500];
    int s,i;
    /*
    struct sigaction {
        void (*sa_handler)(int); 信号处理函数
        void (*sa_sigaction)(int, siginfo_t *, void *);
        sigset_t sa_mask; 信号掩码,执行信号处理函数时将屏蔽这些信号
        int sa_flags;
        void (*sa_restorer)(void);
    }
     */
    struct sigaction sa;

    /* setup alarm signal handler */
    sa.sa_handler=alarm_handler;
    sa.sa_flags=0;
    //设置信号处理函数
    if(sigaction(SIGALRM,&sa,NULL))
        exit(3);

    //设置定时器,benchtime之后就发送信号.
    //unsigned int alarm（unsigned int seconds);
    alarm(benchtime); // after benchtime,then exit

    rlen=strlen(req);

    //该循环负责不断地创建socket连接,然后给服务器发送http请求.
    nexttry:while(1)
    {
        if(timerexpired)
        {
            if(failed>0)
            {
                /* fprintf(stderr,"Correcting failed by signal\n"); */
                failed--;
            }
            return;
        }
        
        s=Socket(host,port);
        //printf("socket discriptr: %d,getpid: %d.\n",s,getpid());
        if(s<0) { failed++;continue;} 
        if(rlen!=write(s,req,rlen)) {failed++;close(s);continue;}
        if(http10==0)
        /*
        禁止在一个套接口上进行数据的接收与发送。
        int shutdown(int sockfd,int how);
        s：用于标识一个套接口的描述字。
        how：标志，用于描述禁止哪些操作。
        how的方式有三种分别是
        SHUT_RD（0）：关闭sockfd上的读功能，此选项将不允许sockfd进行读操作。
        SHUT_WR（1）：关闭sockfd的写功能，此选项将不允许sockfd进行写操作。
        SHUT_RDWR（2）：关闭sockfd的读写功能。
         */
        if(shutdown(s,1)) { failed++;close(s);continue;}
        if(force==0) 
        {
            /* read all available data from socket */
            //该循环负责接受服务器返回的数据,如果force为1,则不接受数据
            while(1)
            {
                if(timerexpired) break; 
                i=read(s,buf,1500);
                /* fprintf(stderr,"%d\n",i); */
                if(i<0) 
                { 
                    failed++;
                    close(s);
                    goto nexttry;
                }
                else if(i==0) break;
                else
                    bytes+=i;
            }
        }
        if(close(s)) {failed++;continue;}
        speed++;
    }
}