现在的位置: 首页 > 自动控制 > 工业·编程 > 正文

TinyHttp:最简单的Web Server浅析

2019-06-06 17:41 工业·编程 ⁄ 共 12589字 ⁄ 字号 暂无评论

Overview

main() –> startup() –> accept_request() –> execute_cgi()

Main senario:

1

Establish a web server step

  1. 建立连接– accept connection from client
  2. Accept request – 读取一条HTTP报文
  3. Handle request – 解释报文,采取行动
  4. 访问资源 – 访问报文中指定的资源
  5. 构建报文 – 创建带有正确HTTP首部的报文
  6. 发送响应 – give the response to client
  7. 记录事务处理过程 – record the step into log <– No such info in TinyHttp

Create connection

如果client已经打开了持久链接,那么可以复用,否则客户端可以打开一条新连接。

Web服务器会将新连接添加到现存的Web服务器列表中,做好监视链接上传数据的准备。

Web server可以随意拒绝or立即关闭任意一条链接<==?? How, Need to check Nginx/APV/..??

客户端主机名识别–会降低性能。So many server close this function or permit it in some special cases.

Accept request message

解析请求行,查找请求方法、指定的资源标识符以及版本号,各项之间以一个空格分隔,并且以一个CRLF \r作为行的结束

读取以 CRLF 结尾的报文首部

检测到以CRLF结尾的、标识首部结束的空行(如果有的话)

长度由content-length决定

eg:

TinyHttp采用多线程模型,没有使用复用IO模型,收到一个请求时,即listenning port get accept request,generate a thread to run accept_request function

    int main(void):

            /*建立socket,监听端口*/ 

             ++server_sock = startup(&port);

             ++while (1):

              |__client_sock = accept(server_sock,

                (struct sockaddr *)&client_name,

                &client_name_len);/*recieve request message*/

              |__if (pthread_create(&newthread , NULL, accept_request, (void*)client_sock) != 0):

                    perror("pthread_create");/* accept_request(client_sock),child thread to handle request*/

            //close    

            ++close(server_sock);

            ++return(0);

注意:pthread_create 第二个参数回调函数也要求是void* func(void*)三个参数是(void*) ,后面

            void* accept_request(void* thread_arg) client_sock 作为参数传进来,用的时候要:

            int client = (int)client_sock

    /*对参数的读取放在get_line()里很关键一点要理解get_line的意思。

    * 我们要知道当在浏览器中输入url后enter之后,它发给服务器是文本型的字符串,

    * 遵循http请求格式,类似下面的:

    * GET / HTTP/1.1

    * HOST:www.abc.com

    * Content-type:text/html

    *...

    * get_line干的事就是读取一行,并且不管原来是以\n还是\r\n结束,均转化为以\n再加\0字符结束。其实现如下:*/

int get_line(int sock, char* buf, int size)

        {

           int i = 0;

           int rcv_num = 0;

           char c = '\0'; /*recieve one character each time*/

           /*consider \r\n*/

           while((i<size-1) && (c != '\n')) {

               rcv_num = recv(sock, &c, 1, 0);

               if (rcv_num > 0) {

                   if (c=='\r') {/*收到 \r 则继续接收下个字节,因为换行符可能是 \r\n*/

                       /*use MSG_PEEK to keep window unmoved, next recv read the same character*/

                       rcv_num =  recv(sock, &c, 1, MSG_PEEK);

                       if ((rcv_num > 0) && (c == '\n'))

                           rcv_num =  recv(sock, &c, 1, 0);/*this time,use '\n' to replace '\r\n' to put in 'c'*/

                       else

                           c = '\n'

                   }

                   buf[i] = c; /*assign recved value to buf */

                   i++;

               } else {

                   c = '\n'

               }

               buf[i] = '\0';  

           }

           return i;

        }

Handle request

一旦收到请求,就可以根据方法、资源、首部和可选的主体部分来对请求进行处理,比如POST,要求请求报文中必须带有实体主体部分

主要处理逻辑在accept_request()

对资源的映射以及访问

通常,web server文件系统会有一个专门的文件夹来存放Web内容,这个文件夹通常被称为 root根目录。web 服务器从请求中获取uri,并将其附加到文档根目录后面。

动态内容资源映射–映射到按需动态生成内容的程序上去,Web server要可以分辨出资源什么时候是动态的,动态内容生成程序位于何处

参见accept_request 程序:

eg:

浏览器里输入127.0.0.1:37776–numchars = get_line(client, buf,sizeof(buf));/*get each line*/

gdb 中看 $2 = “GET / HTTP/1.1\n”, ‘\000’

–> url is /

–> path is htdocs/

–> path is htdocs/index.html,如果不输入路径的话,那么就从htdocs找,这是默认的根目录

–> if ((st.st_mode & S_IFMT) == S_IFDIR) /*dir in request*/

–> strcat(path, "/index.html"); 目录的话,再找index.html

–> 找 是否需要走cgi 处理 CGI:通用网关接口(Common Gateway Interface)

代码

void* accept_request(void* thread_arg)

        {

            char buf[1024];

            int numchars;

            char method[255];/*GET or POST*/

            char url[255]; /*path, eg:GET /color.cgi?color=red ==>/color.cgi*/

            char path[512];

            size_t i,j;

            struct stat st;

            int cgi = 0; /* becomes true if server decides this is a CGI */

            char *query_string = NULL; /*para info of GET method eg: GET /color.cgi?color=red ==> color=red */

            int client = (int)thread_arg;

            numchars = get_line(client, buf,sizeof(buf));/*get each line*/

            if (numchars == 0) {

                printf("No header exsited\n");

            /*  error_die("get_line");*/

            }

            i = j = 0;

            /*Get method string*/

            while(!IS_SPACE(buf[j]) && (i < sizeof(method)-1)) {

                method[i++] = buf[j++];

            }

             method[i] = '\0';

            /*The method can not work*/

            if (strcasecmp(method, "GET") && strcasecmp(method, "POST")) {

                unimplemented(client);

                return NULL;

            }

            if (strcasecmp(method, "POST") == 0)

                cgi = 1;

            i = 0;

            /*Get url*/

            while(IS_SPACE(buf[j]) && (j<sizeof(buf))) {

                j++;

            }

              while(!IS_SPACE(buf[j]) && (j<sizeof(buf)) && (i<sizeof(url)-1)) {

                  url[i++] = buf[j++];

            }

             url[i] = '\0';

          if (strcasecmp(method, "GET") == 0) {

                query_string = url;

                /*if there is '?', it's indicate that para after '?' ,it's dynamic request*/

                while((*query_string != '?') && (*query_string != '\0'))

                  query_string++;

              if (*query_string == '?') {  GET /color.cgi?color=red

                    *query_string = '\0'; /* discard the content after '?' */

                     query_string++; /*query string is pointing to the para of dynamic request*/

                     cgi = 1; <==query_string 指向 color=red

                }

            }

         /*get the path of request file*/

        sprintf(path, "htdocs%s", url);

        int last = strlen(path)-1;

        if (path[last] == '/')

            strcat(path, "index.html");

        /*no such file, return not frond*/

        if (stat(path, &st) == -1) {

            // 读header

            while (numchars>0 && strcmp(buf,"\n"))

                numchars = get_line(client, buf, sizeof(buf));

            not_found(client);

        } else {

            /*find it*/

            if ((st.st_mode & S_IFMT) == S_IFDIR) /*dir in request*/

                strcat(path, "/index.html");

            /*check the file permissions, if can be executed, then go to cgi*/

            if ((st.st_mode & S_IXOTH) || (st.st_mode & S_IXGRP) || (st.st_mode & S_IXUSR))

                cgi = 1;

            if (!cgi) /*static page request*/

                serve_file(client, path); -- 执行 htdocs/index.html

            else /*dynamic*/

                execute_cgi(client, path, method, query_string);

        }

        close(client);}

构建响应:

一旦识别出了资源,就执行请求方法中描述的动作,并返回响应报文。响应报文中包括响应码,响应首部,还可能有响应主体

以tinyhttp为例,如果文件路径合法,对于无参数的 GET 请求,直接输出服务器文件到浏览器,即用 HTTP 格式写到套接字上,跳到(10)。其他情况(带参数 GET,POST 方式,url 为可执行文件),则调用 excute_cgi 函数执行 cgi 脚本。

cgi 介绍, 服务器和客户端之间的通信,是客户端的浏览器和服务器端的http服务器之间的HTTP通信,我们只需要知道浏览器请求执行服务器上哪个CGI程序就可以了,其他不必深究细节,因为这些过程不需要程序员去操作。一般情况下,服务器和CGI程序之间是通过标准输入输出来进行数据传递的,而这个过程需要环境变量的协作方可实现。

服务器将URL指向一个应用程序

服务器为应用程序执行做准备

应用程序执行,读取标准输入和有关环境变量

应用程序进行标准输出

参考: http://www.cnblogs.com/lele/articles/3564327.html

static 响应:

返回index.html页面 – 读取header并丢弃

    1: buf = "Host: 127.0.0.1:37776\n", '\000' <repeats 402 times>...

            (gdb) n

            253

            1: buf = "User-Agent: Mozilla/5.0 (X11; Ubuntu; Linux i686; rv:41.0) Gecko/20100101 Firefox/41.0\n", '\000' <repeats 337 times>...

            1: buf = "User-Agent: Mozilla/5.0 (X11; Ubuntu; Linux i686; rv:41.0) Gecko/20100101 Firefox/41.0\n", '\000' <repeats 337 times>...

            1: buf = "Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8\n\000 Firefox/41.0\n", '\000' <repeats 337 times>...

            1: buf = "Accept-Language: en-US,en;q=0.5\n\000ml+xml,application/xml;q=0.9,*/*;q=0.8\n\000 Firefox/41.0\n", '\000' <repeats 337 times>...

            buf = "Connection: keep-alive\n\000eflate\n\000\000ml+xml,application/xml;q=0.9,*/*;q=0.8\n\000 Firefox/41.0\n", '\000' <repeats 337 times>...

2. 发送headers:

    strcpy(buf, "HTTP/1.0 200 OK\r\n");

                 send(client, buf, strlen(buf), 0);

            strcpy(buf, SERVER_STRING);

            send(client, buf, strlen(buf), 0);

            strcpy(buf, "Content-Type:text/html\r\n");

                send(client, buf, strlen(buf), 0);

                strcpy(buf, "\r\n");

                send(client, buf, strlen(buf), 0);

3. 发送index.html

            void cat (int client, FILE* resource)

            {

                char buf[1024];

                fgets(buf, sizeof(buf), resource); /*fgets read one line once time*/

                while(!feof(resource)) {

                send(client, buf, strlen(buf), 0);

                fgets(buf, sizeof(buf), resource);

                }

            }

4. 走到主页,html格式

    <FORM ACTION="color.cgi" METHOD="POST">

    Enter a color: <INPUT TYPE="text" NAME="color">

    <INPUT TYPE="submit">

5. 再输入color会走到动态调用,通过index.html 调用color.cgi, 然后把execute cgi

dynamic 响应:

(gdb) p method

$10 = “POST”, ‘\000’ url is /color.cgi

回一个200:OK buf = "HTTP/1.0 200 OK \r\n\000cation/x-www-form-urlencoded\n\000on/xml;q=0.9,*/*;q=0.8\n\000 Firefox/41.0\n", '\000' <repeats 665 times>...

create pipe, dup.父进程读取请求(点击submit后,实际上时color=red),写出到cgi_input[0](cgi_input[0]和STDIN共用一个描述符),作为cgi程序的输入.

子进程设置好环境变量REQUEST_METHOD=POST. CONTENT_LENGTH = 11 – STDIO中的有效信息长度, then

execl(path, path, NULL);

如下图: cgi_input[1] 负责POST时的写入,如果浏览器直接127.0.0.1:37776/color.cgi,那么不会有这一步,只是cgi的程序输出到STDOUT,和cgi_output[1]共享 – > cgi_output[0] –> recv() –> 浏览器

code:

void execute_cgi(int client, const char *path, const char *method, const char *query_string)

        {

            char buf[1024];

            int cgi_output[2];

            int cgi_input[2];

            pid_t pid;

            int status;/*for waitpid*/

            int i;

            char c;/*for recieve single character*/

            int numchars = 1;

            int content_length = -1;

            buf[0] = 'A';

            buf[1] = '\0'; /*why*/

            /*read and discard headers for "GET"*/

            if (strcasecmp(method, "GET") == 0) {

                while((numchars > 0) && strcmp("\n", buf))

                    numchars = get_line(client, buf, sizeof(buf));

            } else {

                /*POST, find content_length*/

                numchars = get_line(client, buf, sizeof(buf));

                while ((numchars > 0) && strcmp("\n", buf)) {

                    /*use '\0' to split, the buf like this : Content-Length*/

                    buf[15] = '\0';

                    /*each line*/

                    if (strcasecmp(buf, "Content-Length:") == 0)

                        content_length = atoi(&buf[16]);/*eg: Content-Length:25, 25 is pointting to*/

                    numchars = get_line(client, buf, sizeof(buf));

                }

                if (content_length == -1) {/*no change*/

                    bad_request(client);

                    return;

                }

            }

            sprintf(buf, "HTTP/1.0 200 OK \r\n");

            send(client, buf, strlen(buf),0);

            // create pipe

            if (pipe(cgi_output)<0) {

                // 500 internal server error

                cannot_execute(client);

                return;

            }

            if (pipe(cgi_input)<0) {

                // 500 internal server error

                cannot_execute(client);

                return;

            }

            if ((pid = fork())<0) {

                cannot_execute(client);

                return;

            }

            if (pid == 0) {

                /*child process, child: CGI script*/

                char meth_env[255];

                char query_env[255];

                char length_env[255];

                /*pipe read and write*/

                dup2(cgi_output[1], STDOUT_FILENO);

                dup2(cgi_input[0], STDIN_FILENO);  

                close(cgi_output[0]);

                close(cgi_input[1]);

                sprintf(meth_env,"REQUEST_METHOD=%s",method);

                putenv(meth_env);

                if (strcasecmp(method, "GET") == 0) {

                    sprintf(query_env, "QUERY_STRING=%s",query_string);

                    putenv(query_env);

                } else {

                    /*POST*/

                    sprintf(query_env, "CONTENT_LENGTH=%d",content_length);

                    putenv(query_env);

                }  

                /*run cgi*/

                execl(path, path, NULL)

                exit(0);

            } else {

            /*parent process*/

                close(cgi_output[1]);

                close(cgi_input[0]);

                if (strcasecmp(method, "POST") == 0) {

                    for (i=0; i<content_length; i++) {

                        recv(client, &c, 1, 0);

                        /*data of POST write to cgi_input[1] --> STDIN*/

                        if (1 != write(cgi_input[1], &c, 1)) {

                            printf("%c",c);

                            error_die("write");

                        }

                    }

                }

                /*send data to browser*/

                while(read(cgi_output[0], &c, 1)) {

                    send(client, &c, 1, 0);

                }

                close(cgi_output[0]);

                close(cgi_input[1]);

                waitpid(pid, &status, 0);

            }

        }

遗留问题:

tinyhttpd 一样使用线程模型的吗?有考虑或者测试过某个线程异常退出然后会发生什么吗?

—因为在linux环境下一个线程异常退出,整个进程都会退出,所以 nginx 和 Apache 才都使用进程模型(windows 下不存在这个问题,所以windows下 Apache 使用多进程+多线程)

如果 client 端大量连接server,只连接不通信,你的 server 会不停地创建新的线程响应新的连接吗?

—第二个问题是同步多进程/线程模型服务器的一个问题,消耗资源大, 所以最好考虑设定最大并发数

http/0.9、http/1.0、http/1.1 请求是否都能得到正确的返回?http/1.1 中的 expect 请求呢?chunck?gzip?是否都可以得到正确的返回?是否支持 https?

如果 CGI 执行时间过长是否会导致线程阻塞?

没有更改当前工作目录,如果不是在当前目录启动,则服务器将找不到对应的资源目录,当然,代码中使用相对路径是相同弊病 – How

没有将运行、通信情况记入 log,这样对于以往的服务器工作状况、通信状况,我们将无从得知

没有对拒绝服务攻击做出处理,即便不是拒绝服务攻击,仅仅是连接数目过多都会使服务器线程开辟过多而占用过多资源,这一点如果换成 IO 复用模型可以得到一定的改善

给我留言

留言无头像?