mod_substitute.c

Apache官方在今天放出产品系列2.2的最新版本2.2.11的源码包 最流行的HTTP服务器软件之一

        /*
         * Everything to be passed to the next filter goes in
         * here, our pass brigade.
         */
        ctx->passbb = apr_brigade_create(f->r->pool, f->c->bucket_alloc);
        /* Create our temporary pool only once */
        apr_pool_create(&(ctx->tpool), f->r->pool);
        apr_table_unset(f->r->headers_out, "Content-Length");
    }

    /*
     * Shortcircuit processing
     */
    if (APR_BRIGADE_EMPTY(bb))
        return APR_SUCCESS;

    /*
     * Here's the concept:
     *  Read in the data and look for newlines. Once we
     *  find a full "line", add it to our working brigade.
     *  If we've finished reading the brigade and we have
     *  any left over data (not a "full" line), store that
     *  for the next pass.
     *
     * Note: anything stored in ctx->linebb for sure does not have
     * a newline char, so we don't concat that bb with the
     * new bb, since we would spending time searching for the newline
     * in data we know it doesn't exist. So instead, we simply scan
     * our current bb and, if we see a newline, prepend ctx->linebb
     * to the front of it. This makes the code much less straight-
     * forward (otherwise we could APR_BRIGADE_CONCAT(ctx->linebb, bb)
     * and just scan for newlines and not bother with needing to know
     * when ctx->linebb needs to be reset) but also faster. We'll take
     * the speed.
     *
     * Note: apr_brigade_split_line would be nice here, but we
     * really can't use it since we need more control and we want
     * to re-use already read bucket data.
     *
     * See mod_include if still confused :)
     */

    while ((b = APR_BRIGADE_FIRST(bb)) && (b != APR_BRIGADE_SENTINEL(bb))) {
        if (APR_BUCKET_IS_EOS(b)) {
            /*
             * if we see the EOS, then we need to pass along everything we
             * have. But if the ctx->linebb isn't empty, then we need to add
             * that to the end of what we'll be passing.
             */
            if (!APR_BRIGADE_EMPTY(ctx->linebb)) {
                rv = apr_brigade_pflatten(ctx->linebb, &bflat,
                                          &fbytes, ctx->tpool);
                tmp_b = apr_bucket_transient_create(bflat, fbytes,
                                                f->r->connection->bucket_alloc);
                do_pattmatch(f, tmp_b, ctx->pattbb, ctx->tpool);
                APR_BRIGADE_CONCAT(ctx->passbb, ctx->pattbb);
            }
            apr_brigade_cleanup(ctx->linebb);
            APR_BUCKET_REMOVE(b);
            APR_BRIGADE_INSERT_TAIL(ctx->passbb, b);
        }
        /*
         * No need to handle FLUSH buckets separately as we call
         * ap_pass_brigade anyway at the end of the loop.
         */
        else if (APR_BUCKET_IS_METADATA(b)) {
            APR_BUCKET_REMOVE(b);
            APR_BRIGADE_INSERT_TAIL(ctx->passbb, b);
        }
        else {
            /*
             * We have actual "data" so read in as much as we can and start
             * scanning and splitting from our read buffer
             */
            rv = apr_bucket_read(b, &buff, &bytes, APR_BLOCK_READ);
            if (rv != APR_SUCCESS || bytes == 0) {
                APR_BUCKET_REMOVE(b);
            }
            else {
                int num = 0;
                while (bytes > 0) {
                    nl = memchr(buff, APR_ASCII_LF, bytes);
                    if (nl) {
                        len = (apr_size_t) (nl - buff) + 1;
                        /* split *after* the newline */
                        apr_bucket_split(b, len);
                        /*
                         * We've likely read more data, so bypass rereading
                         * bucket data and continue scanning through this
                         * buffer
                         */
                        bytes -= len;
                        buff += len;
                        /*
                         * we need b to be updated for future potential
                         * splitting
                         */
                        tmp_b = APR_BUCKET_NEXT(b);
                        APR_BUCKET_REMOVE(b);
                        /*
                         * Hey, we found a newline! Don't forget the old
                         * stuff that needs to be added to the front. So we
                         * add the split bucket to the end, flatten the whole
                         * bb, morph the whole shebang into a bucket which is
                         * then added to the tail of the newline bb.
                         */
                        if (!APR_BRIGADE_EMPTY(ctx->linebb)) {
                            APR_BRIGADE_INSERT_TAIL(ctx->linebb, b);
                            rv = apr_brigade_pflatten(ctx->linebb, &bflat,
                                                      &fbytes, ctx->tpool);
                            b = apr_bucket_transient_create(bflat, fbytes,
                                            f->r->connection->bucket_alloc);
                            apr_brigade_cleanup(ctx->linebb);
                        }
                        do_pattmatch(f, b, ctx->pattbb, ctx->tpool);
                        /*
                         * Count how many buckets we have in ctx->passbb
                         * so far. Yes, this is correct we count ctx->passbb
                         * and not ctx->pattbb as we do not reset num on every
                         * iteration.
                         */
                        for (b = APR_BRIGADE_FIRST(ctx->pattbb);
                             b != APR_BRIGADE_SENTINEL(ctx->pattbb);
                             b = APR_BUCKET_NEXT(b)) {
                            num++;
                        }
                        APR_BRIGADE_CONCAT(ctx->passbb, ctx->pattbb);
                        /*
                         * If the number of buckets in ctx->passbb reaches an
                         * "insane" level, we consume much memory for all the
                         * buckets as such. So lets flush them down the chain
                         * in this case and thus clear ctx->passbb. This frees
                         * the buckets memory for further processing.
                         * Usually this condition should not become true, but
                         * it is a safety measure for edge cases.
                         */
                        if (num > AP_MAX_BUCKETS) {
                            b = apr_bucket_flush_create(
                                                f->r->connection->bucket_alloc);
                            APR_BRIGADE_INSERT_TAIL(ctx->passbb, b);
                            rv = ap_pass_brigade(f->next, ctx->passbb);
                            apr_brigade_cleanup(ctx->passbb);
                            num = 0;
                            apr_pool_clear(ctx->tpool);
                            if (rv != APR_SUCCESS)
                                return rv;
                        }
                        b = tmp_b;
                    }
                    else {
                        /*
                         * no newline in whatever is left of this buffer so
                         * tuck data away and get next bucket
                         */
                        APR_BUCKET_REMOVE(b);
                        APR_BRIGADE_INSERT_TAIL(ctx->linebb, b);
                        bytes = 0;
                    }
                }
            }
        }
        if (!APR_BRIGADE_EMPTY(ctx->passbb)) {
            rv = ap_pass_brigade(f->next, ctx->passbb);
            apr_brigade_cleanup(ctx->passbb);
            if (rv != APR_SUCCESS) {
                apr_pool_clear(ctx->tpool);
                return rv;
            }
        }
        apr_pool_clear(ctx->tpool);
    }

    /* Anything left we want to save/setaside for the next go-around */
    if (!APR_BRIGADE_EMPTY(ctx->linebb)) {
        /*
         * Provide ap_save_brigade with an existing empty brigade
         * (ctx->linesbb) to avoid creating a new one.
         */
        ap_save_brigade(f, &(ctx->linesbb), &(ctx->linebb), f->r->pool);
        tmp_bb = ctx->linebb;
        ctx->linebb = ctx->linesbb;
        ctx->linesbb = tmp_bb;
    }

    return APR_SUCCESS;
}

static const char *set_pattern(cmd_parms *cmd, void *cfg, const char *line)
{
    char *from = NULL;
    char *to = NULL;
    char *flags = NULL;
    char *ourline;
    char delim;
    subst_pattern_t *nscript;
    int is_pattern = 0;
    int ignore_case = 0;
    int flatten = 1;
    ap_regex_t *r = NULL;

    if (apr_tolower(*line) != 's') {
        return "Bad Substitute format, must be an s/// pattern";
    }
    ourline = apr_pstrdup(cmd->pool, line);
    delim = *++ourline;
    if (delim)
        from = ++ourline;
    if (from) {
        if (*ourline != delim) {
            while (*++ourline && *ourline != delim);
        }
        if (*ourline) {
            *ourline = '\0';
            to = ++ourline;
        }
    }
    if (to) {
        if (*ourline != delim) {
            while (*++ourline && *ourline != delim);
        }
        if (*ourline) {
            *ourline = '\0';
            flags = ++ourline;
        }
    }

    if (!delim || !from || !*from || !to) {
        return "Bad Substitute format, must be a complete s/// pattern";
    }

    if (flags) {
        while (*flags) {
            delim = apr_tolower(*flags);    /* re-use */
            if (delim == 'i')
                ignore_case = 1;
            else if (delim == 'n')
                is_pattern = 1;
            else if (delim == 'f')
                flatten = 1;
            else if (delim == 'q')
                flatten = 0;
            else
                return "Bad Substitute flag, only s///[infq] are supported";
            flags++;
        }
    }

    /* first see if we can compile the regex */
    if (!is_pattern) {
        r = ap_pregcomp(cmd->pool, from, AP_REG_EXTENDED |
                        (ignore_case ? AP_REG_ICASE : 0));
        if (!r)
            return "Substitute could not compile regex";
    }
    nscript = apr_array_push(((subst_dir_conf *) cfg)->patterns);
    /* init the new entries */
    nscript->pattern = NULL;
    nscript->regexp = NULL;
    nscript->replacement = NULL;
    nscript->patlen = 0;

    if (is_pattern) {
        nscript->patlen = strlen(from);
        nscript->pattern = apr_strmatch_precompile(cmd->pool, from, 
                                                   !ignore_case);
    }
    else {
        nscript->regexp = r;
    }

    nscript->replacement = to;
    nscript->replen = strlen(to);
    nscript->flatten = flatten;

    return NULL;
}

#define PROTO_FLAGS AP_FILTER_PROTO_CHANGE|AP_FILTER_PROTO_CHANGE_LENGTH
static void register_hooks(apr_pool_t *pool)
{
    ap_register_output_filter(substitute_filter_name, substitute_filter,
                              NULL, AP_FTYPE_RESOURCE);
}

static const command_rec substitute_cmds[] = {
    AP_INIT_TAKE1("Substitute", set_pattern, NULL, OR_ALL,
                  "Pattern to filter the response content (s/foo/bar/[inf])"),
    {NULL}
};

module AP_MODULE_DECLARE_DATA substitute_module = {
    STANDARD20_MODULE_STUFF,
    create_substitute_dcfg,     /* dir config creater */
    merge_substitute_dcfg,      /* dir merger --- default is to override */
    NULL,                       /* server config */
    NULL,                       /* merge server config */
    substitute_cmds,            /* command table */
    register_hooks              /* register hooks */
};