Mongoose源碼剖析：核心處理模塊

引言

前面我們介紹了Mongoose所有的幾個主要的數據結構mg_context、mg_connection、mg_request_info，還有Mongoose的生命主線。有了這些基礎就可以來看看Mongoose的核心處理工作是怎樣的。如果你還沒有閱讀前面的文章，你可以通過下面的隧道直通：

1. Mongoose源碼剖析：外篇之web服務器
2. Mongoose源碼剖析：Introduction and Installation
3. Mongoose源碼剖析：數據結構篇
4. Mongoose源碼剖析：mongoose的工作模型

本文從下面幾個方面去介紹Mongoose的核心處理模塊，連接建立之後的：

1. 請求解析
2. 請求驗證
3. 請求滿足

1、連接的建立

Mongoose的主線程master_thread在接受一個新的client連接請求時，會將client的socket地址放入一個queue（調用put_socket()方法）；而當worker_thread線程處理client的請求時，是通過get_socket()方法從queue取出client的socket地址，然後與它建立連接。

建立連接就用到了數據結構mg_connection，該結構保存了client的連接信息。該結構體中有兩個非常重要的成員：mg_request_info用於保存client的請求信息、mg_context用於保存該client請求的mongoose上下文。建立連接的代碼片段如下：

while (get_socket(ctx, &conn.client) == TRUE)
{
        conn.birth_time = time(NULL);
        conn.ctx = ctx;

        if (conn.client.is_ssl &&
            (conn.ssl = SSL_new(conn.ctx->ssl_ctx)) == NULL) {
            cry(&conn, "%s: SSL_new: %d", __func__, ERRNO);
        } else if (conn.client.is_ssl &&
            SSL_set_fd(conn.ssl, conn.client.sock) != 1) {
            cry(&conn, "%s: SSL_set_fd: %d", __func__, ERRNO);
        } else if (conn.client.is_ssl && SSL_accept(conn.ssl) != 1) {
            cry(&conn, "%s: SSL handshake error", __func__);
        } else {
            process_new_connection(&conn);
        }

        close_connection(&conn);
}

其中以SSL_開頭的函數都是加載自SSL的庫，加載庫調用瞭如下接口：static bool_t set_ssl_option(struct mg_context *ctx, const char *pem)，有興趣的話你可以追蹤下去。

2、請求信息獲取

建立連接之後，在process_new_connection中會去讀取client的請求信息，然後纔去解析請求。讀取client端的請求的信息用到了下面的方法：

/*
 * Keep reading the input (either opened file descriptor fd, or socket sock,
 * or SSL descriptor ssl) into buffer buf, until \r\n\r\n appears in the
 * buffer (which marks the end of HTTP request). Buffer buf may already
 * have some data. The length of the data is stored in nread.
 * Upon every read operation, increase nread by the number of bytes read.
*/
static int
read_request(FILE *fp, SOCKET sock, SSL *ssl, char *buf, int bufsiz, int *nread)
{
    int n, request_len;

    request_len = 0;
    while (*nread < bufsiz && request_len == 0) {
        n = pull(fp, sock, ssl, buf + *nread, bufsiz - *nread);
        if (n <= 0) {
            break;
        } else {
            *nread += n;
            request_len = get_request_len(buf, (size_t) *nread);
        }
    }

    return (request_len);
}

其中pull()方法的代碼如下：

/*
 * Read from IO channel - opened file descriptor, socket, or SSL descriptor.
 * Return number of bytes read.
*/
static int
pull(FILE *fp, SOCKET sock, SSL *ssl, char *buf, int len)
{
    int nread;

    if (ssl != NULL) {
        nread = SSL_read(ssl, buf, len);
    } else if (fp != NULL) {
        nread = fread(buf, 1, (size_t) len, fp);
        if (ferror(fp))
            nread = -1;
    } else {
        nread = recv(sock, buf, (size_t) len, 0);
    }

    return (nread);
}

這樣client發送的HTTP請求消息就被worker_thread讀取到了，並存儲在buf中， 接下來的工作就是解析讀取到的請求信息，明白client到底想幹嘛，說白了就從buf中提取信息並存儲到結構體mg_request_info中去。

3、請求解析

請求解析的工作都封裝在parse_http_request()函數匯中，它的代碼如下：

/*
 * Parse HTTP request, fill in mg_request_info structure.
*/
static bool_t
parse_http_request(char *buf, struct mg_request_info *ri, const struct usa *usa)
{
    char    *http_version;
    int n, success_code = FALSE;

    ri->request_method = skip(&buf, " ");
    ri->uri = skip(&buf, " ");
    http_version = skip(&buf, "\r\n");

    if (is_known_http_method(ri->request_method) &&
        ri->uri[0] == '/' &&
        sscanf(http_version, "HTTP/%d.%d%n",
        &ri->http_version_major, &ri->http_version_minor, &n) == 2 &&
        http_version[n] == '\0') {
        parse_http_headers(&buf, ri);
        ri->remote_port = ntohs(usa->u.sin.sin_port);
        (void) memcpy(&ri->remote_ip, &usa->u.sin.sin_addr.s_addr, 4);
        ri->remote_ip = ntohl(ri->remote_ip);
        success_code = TRUE;
    }

    return (success_code);
}

它的主要工作就是從buf中提取出信息放到ri（一個mg_request_info結構）中去，因爲buf是一個無結構的字符串數組。要將它存儲到ri中去，需要找到對應的子串。

這裏主要用到了skip()、parse_http_headers()方法，其中skip()很關鍵，代碼如下：

/*
 * Skip the characters until one of the delimiters characters found.
 * 0-terminate resulting word. Skip the rest of the delimiters if any.
 * Advance pointer to buffer to the next word. Return found 0-terminated word.
*/
static char *
skip(char **buf, const char *delimiters)
{
    char    *p, *begin_word, *end_word, *end_delimiters;

    begin_word = *buf;
    end_word = begin_word + strcspn(begin_word, delimiters);
    end_delimiters = end_word + strspn(end_word, delimiters);

    for (p = end_word; p < end_delimiters; p++)
        *p = '\0';

    *buf = end_delimiters;

    return (begin_word);
}

我們來分析一下skip的作用及實現。如要從buf中解析出client請求的methods是哪個（PUT、GET、POST等等）？只需要這樣做就可以了：

ri->request_method = skip(&buf, " ");

爲了分析，到底是如何實現這個的，我在porcess_new_connection()中加入下面一行輸出buf信息的代碼：

看當我們想mongoose發送的請求信息，這時我們在瀏覽其中輸入http://ip:8080，終端會輸出buf的信息，如下：

看到第一行就是GET /favicon.ico HTTP/1.1。知道了buf中的字符信息，但在我們分析skip(&buf, " ")是如何提取出GET的之前，還要知道strcspn、strspn的作用，下面是它們的原型：

#include <string.h>
size_t strspn(const char *s, const char *accept);
size_t strcspn(const char *s, const char *reject);

下面解釋它們的作用：

DESCRIPTION
       The strspn() function calculates the length of the initial segment of s
which consists entirely of characters in accept.

       The strcspn() function calculates the length of the initial segment  of s which consists entirely of characters not in reject.

RETURN VALUE
       The  strspn()  function returns the number of characters in the initial segment of s which consist only of characters from accept.

       The strcspn() function returns the number of characters in the  initial segment of s which are not in the string reject.

現在已經萬事俱備了，skip(&buf, " ")的執行情況如下：

其餘的解析工作也是類似地進行的，我就不一一闡述了。

4、請求驗證

請求驗證分佈在從連接請求開始到請求得到迴應的整個過程中。在請求解析之前，比如驗證socket的合法性等。在請求解析之後，從buf中解析出HTTP請求消息的各個字段之後，就做一些簡單的驗證工作，比如說HTTP版本的驗證。如果在解析buf時出錯，說明請求的格式不對。

而且在滿足client請求的時候也要進行一些驗證，諸如是否有瀏覽目錄的權限、請求的文件是否存在等等，我就不在詳述了。

5、請求滿足

在parse_http_request()之後，調用analyze_request()去滿足client的請求。這是Mongoose的核心內容，也是不同web服務器軟件相區別的地方。analyze_request()封裝了一些操作，即調用了一些接口去滿足client的請求，代碼如下：

/*
 * This is the heart of the Mongoose's logic.
 * This function is called when the request is read, parsed and validated,
 * and Mongoose must decide what action to take: serve a file, or
 * a directory, or call embedded function, etcetera.
*/
static void
analyze_request(struct mg_connection *conn)
{
    struct mg_request_info *ri = &conn->request_info;
    char            path[FILENAME_MAX], *uri = ri->uri;
    struct mgstat       st;
    const struct callback   *cb;

    if ((conn->request_info.query_string = strchr(uri, '?')) != NULL)
        * conn->request_info.query_string++ = '\0';

    (void) url_decode(uri, (int) strlen(uri), uri, strlen(uri) + 1, FALSE);
    remove_double_dots_and_double_slashes(uri);
    convert_uri_to_file_name(conn, uri, path, sizeof(path));

    if (!check_authorization(conn, path)) {
        send_authorization_request(conn);
    } else if (check_embedded_authorization(conn) == FALSE) {
        /*
         * Embedded code failed authorization. Do nothing here, since
         * an embedded code must handle this itself by either
         * showing proper error message, or redirecting to some
         * sort of login page, or something else.
         */
    } else if ((cb = find_callback(conn->ctx, FALSE, uri, -1)) != NULL) {
        if ((strcmp(ri->request_method, "POST") != 0 &&
            strcmp(ri->request_method, "PUT") != 0) ||
            handle_request_body(conn, NULL))
            cb->func(conn, &conn->request_info, cb->user_data);
    } else if (strstr(path, PASSWORDS_FILE_NAME)) {
        /* Do not allow to view passwords files */
        send_error(conn, 403, "Forbidden", "Access Forbidden");
    } else if ((!strcmp(ri->request_method, "PUT") ||
        !strcmp(ri->request_method, "DELETE")) &&
        (conn->ctx->options[OPT_AUTH_PUT] == NULL ||
         !is_authorized_for_put(conn))) {
        send_authorization_request(conn);
    } else if (!strcmp(ri->request_method, "PUT")) {
        put_file(conn, path);
    } else if (!strcmp(ri->request_method, "DELETE")) {
        if (mg_remove(path) == 0)
            send_error(conn, 200, "OK", "");
        else
            send_error(conn, 500, http_500_error,
                "remove(%s): %s", path, strerror(ERRNO));
    } else if (mg_stat(path, &st) != 0) {
        send_error(conn, 404, "Not Found", "%s", "File not found");
    } else if (st.is_directory && uri[strlen(uri) - 1] != '/') {
        (void) mg_printf(conn,
            "HTTP/1.1 301 Moved Permanently\r\n"
            "Location: %s/\r\n\r\n", uri);
    } else if (st.is_directory &&
        substitute_index_file(conn, path, sizeof(path), &st) == FALSE) {
        if (is_true(conn->ctx->options[OPT_DIR_LIST])) {
            send_directory(conn, path);
        } else {
            send_error(conn, 403, "Directory Listing Denied",
                "Directory listing denied");
        }
#if !defined(NO_CGI)
    } else if (match_extension(path,
        conn->ctx->options[OPT_CGI_EXTENSIONS])) {
        if (strcmp(ri->request_method, "POST") &&
            strcmp(ri->request_method, "GET")) {
            send_error(conn, 501, "Not Implemented",
                "Method %s is not implemented", ri->request_method);
        } else {
            send_cgi(conn, path);
        }
#endif /* NO_CGI */
#if !defined(NO_SSI)
    } else if (match_extension(path,
        conn->ctx->options[OPT_SSI_EXTENSIONS])) {
        send_ssi(conn, path);
#endif /* NO_SSI */
    } else if (is_not_modified(conn, &st)) {
        send_error(conn, 304, "Not Modified", "");
    } else {
        send_file(conn, path, &st);
    }
}

上面的代碼比較好理解我就不把它嚼爛之後，再展現給你！那樣也沒有意思~\(≧▽≦)/~啦啦啦。感興趣的自己去逐一分析。