mod_negotiation.c

/* Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

/*
 * mod_negotiation.c: keeps track of MIME types the client is willing to
 * accept, and contains code to handle type arbitration.
 *
 * rst
 */

#include "apr.h"
#include "apr_strings.h"
#include "apr_file_io.h"
#include "apr_lib.h"

#define APR_WANT_STRFUNC
#include "apr_want.h"

#include "ap_config.h"
#include "httpd.h"
#include "http_config.h"
#include "http_request.h"
#include "http_protocol.h"
#include "http_core.h"
#include "http_log.h"
#include "util_script.h"


#define MAP_FILE_MAGIC_TYPE "application/x-type-map"

/* Commands --- configuring document caching on a per (virtual?)
 * server basis...
 */

typedef struct {
    int forcelangpriority;
    apr_array_header_t *language_priority;
} neg_dir_config;

/* forcelangpriority flags
 */
#define FLP_UNDEF    0    /* Same as FLP_DEFAULT, but base overrides */
#define FLP_NONE     1    /* Return 406, HTTP_NOT_ACCEPTABLE */
#define FLP_PREFER   2    /* Use language_priority rather than MC */
#define FLP_FALLBACK 4    /* Use language_priority rather than NA */

#define FLP_DEFAULT  FLP_PREFER

/* env evaluation
 */
#define DISCARD_ALL_ENCODINGS 1  /* no-gzip */
#define DISCARD_ALL_BUT_HTML  2  /* gzip-only-text/html */

module AP_MODULE_DECLARE_DATA negotiation_module;

static void *create_neg_dir_config(apr_pool_t *p, char *dummy)
{
    neg_dir_config *new = (neg_dir_config *) apr_palloc(p,
                                                        sizeof(neg_dir_config));

    new->forcelangpriority = FLP_UNDEF;
    new->language_priority = NULL;
    return new;
}

static void *merge_neg_dir_configs(apr_pool_t *p, void *basev, void *addv)
{
    neg_dir_config *base = (neg_dir_config *) basev;
    neg_dir_config *add = (neg_dir_config *) addv;
    neg_dir_config *new = (neg_dir_config *) apr_palloc(p,
                                                        sizeof(neg_dir_config));

    /* give priority to the config in the subdirectory */
    new->forcelangpriority = (add->forcelangpriority != FLP_UNDEF)
                                ? add->forcelangpriority
                                : base->forcelangpriority;
    new->language_priority = add->language_priority
                                ? add->language_priority
                                : base->language_priority;
    return new;
}

static const char *set_language_priority(cmd_parms *cmd, void *n_,
                                         const char *lang)
{
    neg_dir_config *n = n_;
    const char **langp;

    if (!n->language_priority)
        n->language_priority = apr_array_make(cmd->pool, 4, sizeof(char *));

    langp = (const char **) apr_array_push(n->language_priority);
    *langp = lang;
    return NULL;
}

static const char *set_force_priority(cmd_parms *cmd, void *n_, const char *w)
{
    neg_dir_config *n = n_;

    if (!strcasecmp(w, "None")) {
        if (n->forcelangpriority & ~FLP_NONE) {
            return "Cannot combine ForceLanguagePriority options with None";
        }
        n->forcelangpriority = FLP_NONE;
    }
    else if (!strcasecmp(w, "Prefer")) {
        if (n->forcelangpriority & FLP_NONE) {
            return "Cannot combine ForceLanguagePriority options None and "
                   "Prefer";
        }
        n->forcelangpriority |= FLP_PREFER;
    }
    else if (!strcasecmp(w, "Fallback")) {
        if (n->forcelangpriority & FLP_NONE) {
            return "Cannot combine ForceLanguagePriority options None and "
                   "Fallback";
        }
        n->forcelangpriority |= FLP_FALLBACK;
    }
    else {
        return apr_pstrcat(cmd->pool, "Invalid ForceLanguagePriority option ",
                           w, NULL);
    }

    return NULL;
}

static const char *cache_negotiated_docs(cmd_parms *cmd, void *dummy,
                                         int arg)
{
    ap_set_module_config(cmd->server->module_config, &negotiation_module,
                         (arg ? "Cache" : NULL));
    return NULL;
}

static int do_cache_negotiated_docs(server_rec *s)
{
    return (ap_get_module_config(s->module_config,
                                 &negotiation_module) != NULL);
}

static const command_rec negotiation_cmds[] =
{
    AP_INIT_FLAG("CacheNegotiatedDocs", cache_negotiated_docs, NULL, RSRC_CONF,
                 "Either 'on' or 'off' (default)"),
    AP_INIT_ITERATE("LanguagePriority", set_language_priority, NULL,
                    OR_FILEINFO,
                    "space-delimited list of MIME language abbreviations"),
    AP_INIT_ITERATE("ForceLanguagePriority", set_force_priority, NULL,
                    OR_FILEINFO,
                    "Force LanguagePriority elections, either None, or "
                    "Fallback and/or Prefer"),
    {NULL}
};

/*
 * Record of available info on a media type specified by the client
 * (we also use 'em for encodings and languages)
 */

typedef struct accept_rec {
    char *name;                 /* MUST be lowercase */
    float quality;
    float level;
    char *charset;              /* for content-type only */
} accept_rec;

/*
 * Record of available info on a particular variant
 *
 * Note that a few of these fields are updated by the actual negotiation
 * code.  These are:
 *
 * level_matched --- initialized to zero.  Set to the value of level
 *             if the client actually accepts this media type at that
 *             level (and *not* if it got in on a wildcard).  See level_cmp
 *             below.
 * mime_stars -- initialized to zero.  Set to the number of stars
 *               present in the best matching Accept header element.
 *               1 for star/star, 2 for type/star and 3 for
 *               type/subtype.
 *
 * definite -- initialized to 1.  Set to 0 if there is a match which
 *             makes the variant non-definite according to the rules
 *             in rfc2296.
 */

typedef struct var_rec {
    request_rec *sub_req;       /* May be NULL (is, for map files) */
    const char *mime_type;      /* MUST be lowercase */
    const char *file_name;      /* Set to 'this' (for map file body content) */
    apr_off_t body;             /* Only for map file body content */
    const char *content_encoding;
    apr_array_header_t *content_languages; /* list of lang. for this variant */
    const char *content_charset;
    const char *description;

    /* The next five items give the quality values for the dimensions
     * of negotiation for this variant. They are obtained from the
     * appropriate header lines, except for source_quality, which
     * is obtained from the variant itself (the 'qs' parameter value
     * from the variant's mime-type). Apart from source_quality,
     * these values are set when we find the quality for each variant
     * (see best_match()). source_quality is set from the 'qs' parameter
     * of the variant description or mime type: see set_mime_fields().
     */
    float lang_quality;         /* quality of this variant's language */
    float encoding_quality;     /* ditto encoding */
    float charset_quality;      /* ditto charset */
    float mime_type_quality;    /* ditto media type */
    float source_quality;       /* source quality for this variant */

    /* Now some special values */
    float level;                /* Auxiliary to content-type... */
    apr_off_t bytes;            /* content length, if known */
    int lang_index;             /* Index into LanguagePriority list */
    int is_pseudo_html;         /* text/html, *or* the INCLUDES_MAGIC_TYPEs */

    /* Above are all written-once properties of the variant.  The
     * three fields below are changed during negotiation:
     */

    float level_matched;
    int mime_stars;
    int definite;
} var_rec;

/* Something to carry around the state of negotiation (and to keep
 * all of this thread-safe)...
 */

typedef struct {
    apr_pool_t *pool;
    request_rec *r;
    neg_dir_config *conf;
    char *dir_name;
    int accept_q;               /* 1 if an Accept item has a q= param */
    float default_lang_quality; /* fiddle lang q for variants with no lang */

    /* the array pointers below are NULL if the corresponding accept
     * headers are not present
     */
    apr_array_header_t *accepts;            /* accept_recs */
    apr_array_header_t *accept_encodings;   /* accept_recs */
    apr_array_header_t *accept_charsets;    /* accept_recs */
    apr_array_header_t *accept_langs;       /* accept_recs */

    apr_array_header_t *avail_vars;         /* available variants */

    int count_multiviews_variants;    /* number of variants found on disk */

    int is_transparent;       /* 1 if this resource is trans. negotiable */

    int dont_fiddle_headers;  /* 1 if we may not fiddle with accept hdrs */
    int ua_supports_trans;    /* 1 if ua supports trans negotiation */
    int send_alternates;      /* 1 if we want to send an Alternates header */
    int may_choose;           /* 1 if we may choose a variant for the client */
    int use_rvsa;             /* 1 if we must use RVSA/1.0 negotiation algo */
} negotiation_state;

/* A few functions to manipulate var_recs.
 * Cleaning out the fields...
 */

static void clean_var_rec(var_rec *mime_info)
{
    mime_info->sub_req = NULL;
    mime_info->mime_type = "";
    mime_info->file_name = "";
    mime_info->body = 0;
    mime_info->content_encoding = NULL;
    mime_info->content_languages = NULL;
    mime_info->content_charset = "";
    mime_info->description = "";

    mime_info->is_pseudo_html = 0;
    mime_info->level = 0.0f;
    mime_info->level_matched = 0.0f;
    mime_info->bytes = -1;
    mime_info->lang_index = -1;
    mime_info->mime_stars = 0;
    mime_info->definite = 1;

    mime_info->charset_quality = 1.0f;
    mime_info->encoding_quality = 1.0f;
    mime_info->lang_quality = 1.0f;
    mime_info->mime_type_quality = 1.0f;
    mime_info->source_quality = 0.0f;
}

/* Initializing the relevant fields of a variant record from the
 * accept_info read out of its content-type, one way or another.
 */

static void set_mime_fields(var_rec *var, accept_rec *mime_info)
{
    var->mime_type = mime_info->name;
    var->source_quality = mime_info->quality;
    var->level = mime_info->level;
    var->content_charset = mime_info->charset;

    var->is_pseudo_html = (!strcmp(var->mime_type, "text/html")
                           || !strcmp(var->mime_type, INCLUDES_MAGIC_TYPE)
                           || !strcmp(var->mime_type, INCLUDES_MAGIC_TYPE3));
}

/* Create a variant list validator in r using info from vlistr. */

static void set_vlist_validator(request_rec *r, request_rec *vlistr)
{
    /* Calculating the variant list validator is similar to
     * calculating an etag for the source of the variant list
     * information, so we use ap_make_etag().  Note that this
     * validator can be 'weak' in extreme case.
     */
    ap_update_mtime(vlistr, vlistr->finfo.mtime);
    r->vlist_validator = ap_make_etag(vlistr, 0);

    /* ap_set_etag will later take r->vlist_validator into account
     * when creating the etag header
     */
}


/*****************************************************************
 *
 * Parsing (lists of) media types and their parameters, as seen in
 * HTTPD header lines and elsewhere.
 */

/*
 * parse quality value. atof(3) is not well-usable here, because it
 * depends on the locale (argh).
 *
 * However, RFC 2616 states:
 * 3.9 Quality Values
 *
 * [...] HTTP/1.1 applications MUST NOT generate more than three digits
 * after the decimal point. User configuration of these values SHOULD also
 * be limited in this fashion.
 *
 *     qvalue         = ( "0" [ "." 0*3DIGIT ] )
 *                    | ( "1" [ "." 0*3("0") ] )
 *
 * This is quite easy. If the supplied string doesn't match the above
 * definition (loosely), we simply return 1 (same as if there's no qvalue)
 */

static float atoq(const char *string)
{
    if (!string || !*string) {
        return  1.0f;
    }

    while (apr_isspace(*string)) {
        ++string;
    }

    /* be tolerant and accept qvalues without leading zero
     * (also for backwards compat, where atof() was in use)
     */
    if (*string != '.' && *string++ != '0') {
        return 1.0f;
    }

    if (*string == '.') {
        /* better only one division later, than dealing with fscking
         * IEEE format 0.1 factors ...
         */
        int i = 0;

        if (*++string >= '0' && *string <= '9') {
            i += (*string - '0') * 100;

            if (*++string >= '0' && *string <= '9') {
                i += (*string - '0') * 10;

                if (*++string > '0' && *string <= '9') {
                    i += (*string - '0');
                }
            }
        }

        return (float)i / 1000.0f;
    }

    return 0.0f;
}

/*
 * Get a single mime type entry --- one media type and parameters;
 * enter the values we recognize into the argument accept_rec
 */

static const char *get_entry(apr_pool_t *p, accept_rec *result,
                             const char *accept_line)
{
    result->quality = 1.0f;
    result->level = 0.0f;
    result->charset = "";

    /*
     * Note that this handles what I gather is the "old format",
     *
     *    Accept: text/html text/plain moo/zot
     *
     * without any compatibility kludges --- if the token after the
     * MIME type begins with a semicolon, we know we're looking at parms,
     * otherwise, we know we aren't.  (So why all the pissing and moaning
     * in the CERN server code?  I must be missing something).
     */

    result->name = ap_get_token(p, &accept_line, 0);
    ap_str_tolower(result->name);     /* You want case insensitive,
                                       * you'll *get* case insensitive.
                                       */

    /* KLUDGE!!! Default HTML to level 2.0 unless the browser
     * *explicitly* says something else.
     */

    if (!strcmp(result->name, "text/html") && (result->level == 0.0)) {
        result->level = 2.0f;
    }
    else if (!strcmp(result->name, INCLUDES_MAGIC_TYPE)) {
        result->level = 2.0f;
    }
    else if (!strcmp(result->name, INCLUDES_MAGIC_TYPE3)) {
        result->level = 3.0f;
    }

    while (*accept_line == ';') {
        /* Parameters ... */

        char *parm;
        char *cp;
        char *end;

        ++accept_line;
        parm = ap_get_token(p, &accept_line, 1);

        /* Look for 'var = value' --- and make sure the var is in lcase. */

        for (cp = parm; (*cp && !apr_isspace(*cp) && *cp != '='); ++cp) {
            *cp = apr_tolower(*cp);
        }

        if (!*cp) {
            continue;           /* No '='; just ignore it. */
        }

        *cp++ = '\0';           /* Delimit var */
        while (apr_isspace(*cp) || *cp == '=') {
            ++cp;
        }

        if (*cp == '"') {
            ++cp;
            for (end = cp;
                 (*end && *end != '\n' && *end != '\r' && *end != '\"');
                 end++);
        }
        else {
            for (end = cp; (*end && !apr_isspace(*end)); end++);
        }
        if (*end) {
            *end = '\0';        /* strip ending quote or return */
        }
        ap_str_tolower(cp);

        if (parm[0] == 'q'
            && (parm[1] == '\0' || (parm[1] == 's' && parm[2] == '\0'))) {
            result->quality = atoq(cp);
        }
        else if (parm[0] == 'l' && !strcmp(&parm[1], "evel")) {
            result->level = (float)atoi(cp);
        }
        else if (!strcmp(parm, "charset")) {
            result->charset = cp;
        }
    }

    if (*accept_line == ',') {
        ++accept_line;
    }

    return accept_line;
}

/*****************************************************************
 *
 * Dealing with header lines ...
 *
 * Accept, Accept-Charset, Accept-Language and Accept-Encoding
 * are handled by do_header_line() - they all have the same
 * basic structure of a list of items of the format
 *    name; q=N; charset=TEXT
 *
 * where charset is only valid in Accept.
 */

static apr_array_header_t *do_header_line(apr_pool_t *p,
                                          const char *accept_line)
{
    apr_array_header_t *accept_recs;

    if (!accept_line) {
        return NULL;
    }

    accept_recs = apr_array_make(p, 40, sizeof(accept_rec));

    while (*accept_line) {
        accept_rec *new = (accept_rec *) apr_array_push(accept_recs);
        accept_line = get_entry(p, new, accept_line);
    }

    return accept_recs;
}

/* Given the text of the Content-Languages: line from the var map file,
 * return an array containing the languages of this variant
 */

static apr_array_header_t *do_languages_line(apr_pool_t *p,
                                             const char **lang_line)
{
    apr_array_header_t *lang_recs = apr_array_make(p, 2, sizeof(char *));

    if (!lang_line) {
        return lang_recs;
    }

    while (**lang_line) {
        char **new = (char **) apr_array_push(lang_recs);
        *new = ap_get_token(p, lang_line, 0);
        ap_str_tolower(*new);
        if (**lang_line == ',' || **lang_line == ';') {
            ++(*lang_line);
        }
    }

    return lang_recs;
}

/*****************************************************************
 *
 * Handling header lines from clients...
 */

static negotiation_state *parse_accept_headers(request_rec *r)
{
    negotiation_state *new =
        (negotiation_state *) apr_pcalloc(r->pool, sizeof(negotiation_state));
    accept_rec *elts;
    apr_table_t *hdrs = r->headers_in;
    int i;

    new->pool = r->pool;
    new->r = r;
    new->conf = (neg_dir_config *)ap_get_module_config(r->per_dir_config,
                                                       &negotiation_module);

    new->dir_name = ap_make_dirstr_parent(r->pool, r->filename);

    new->accepts = do_header_line(r->pool, apr_table_get(hdrs, "Accept"));

    /* calculate new->accept_q value */
    if (new->accepts) {
        elts = (accept_rec *) new->accepts->elts;

        for (i = 0; i < new->accepts->nelts; ++i) {
            if (elts[i].quality < 1.0) {
                new->accept_q = 1;
            }
        }
    }

    new->accept_encodings =
        do_header_line(r->pool, apr_table_get(hdrs, "Accept-Encoding"));
    new->accept_langs =
        do_header_line(r->pool, apr_table_get(hdrs, "Accept-Language"));
    new->accept_charsets =
        do_header_line(r->pool, apr_table_get(hdrs, "Accept-Charset"));

    /* This is possibly overkill for some servers, heck, we have
     * only 33 index.html variants in docs/docroot (today).
     * Make this configurable?
     */
    new->avail_vars = apr_array_make(r->pool, 40, sizeof(var_rec));

    return new;
}


static void parse_negotiate_header(request_rec *r, negotiation_state *neg)
{
    const char *negotiate = apr_table_get(r->headers_in, "Negotiate");
    char *tok;

    /* First, default to no TCN, no Alternates, and the original Apache
     * negotiation algorithm with fiddles for broken browser configs.
     *
     * To save network bandwidth, we do not configure to send an
     * Alternates header to the user agent by default.  User
     * agents that want an Alternates header for agent-driven
     * negotiation will have to request it by sending an
     * appropriate Negotiate header.
     */
    neg->ua_supports_trans   = 0;
    neg->send_alternates     = 0;
    neg->may_choose          = 1;
    neg->use_rvsa            = 0;
    neg->dont_fiddle_headers = 0;

    if (!negotiate)
        return;

    if (strcmp(negotiate, "trans") == 0) {
        /* Lynx 2.7 and 2.8 send 'negotiate: trans' even though they
         * do not support transparent content negotiation, so for Lynx we
         * ignore the negotiate header when its contents are exactly "trans".
         * If future versions of Lynx ever need to say 'negotiate: trans',
         * they can send the equivalent 'negotiate: trans, trans' instead
         * to avoid triggering the workaround below.
         */
        const char *ua = apr_table_get(r->headers_in, "User-Agent");

        if (ua && (strncmp(ua, "Lynx", 4) == 0))
            return;
    }

    neg->may_choose = 0;  /* An empty Negotiate would require 300 response */

    while ((tok = ap_get_list_item(neg->pool, &negotiate)) != NULL) {

        if (strcmp(tok, "trans") == 0 ||
            strcmp(tok, "vlist") == 0 ||
            strcmp(tok, "guess-small") == 0 ||
            apr_isdigit(tok[0]) ||
            strcmp(tok, "*") == 0) {

            /* The user agent supports transparent negotiation */
            neg->ua_supports_trans = 1;

            /* Send-alternates could be configurable, but note
             * that it must be 1 if we have 'vlist' in the
             * negotiate header.
             */
            neg->send_alternates = 1;

            if (strcmp(tok, "1.0") == 0) {
                /* we may use the RVSA/1.0 algorithm, configure for it */
                neg->may_choose = 1;
                neg->use_rvsa = 1;
                neg->dont_fiddle_headers = 1;
            }
            else if (tok[0] == '*') {
                /* we may use any variant selection algorithm, configure
                 * to use the Apache algorithm
                 */
                neg->may_choose = 1;

                /* We disable header fiddles on the assumption that a
                 * client sending Negotiate knows how to send correct
                 * headers which don't need fiddling.
                 */
                neg->dont_fiddle_headers = 1;
            }
        }
    }

#ifdef NEG_DEBUG
    ap_log_error(APLOG_MARK, APLOG_STARTUP, 0, NULL, APLOGNO(00680)
            "dont_fiddle_headers=%d use_rvsa=%d ua_supports_trans=%d "
            "send_alternates=%d, may_choose=%d",
            neg->dont_fiddle_headers, neg->use_rvsa,
            neg->ua_supports_trans, neg->send_alternates, neg->may_choose);
#endif

}

/* Sometimes clients will give us no Accept info at all; this routine sets
 * up the standard default for that case, and also arranges for us to be
 * willing to run a CGI script if we find one.  (In fact, we set up to
 * dramatically prefer CGI scripts in cases where that's appropriate,
 * e.g., POST or when URI includes query args or extra path info).
 */
static void maybe_add_default_accepts(negotiation_state *neg,
                                      int prefer_scripts)
{
    accept_rec *new_accept;

    if (!neg->accepts) {
        neg->accepts = apr_array_make(neg->pool, 4, sizeof(accept_rec));

        new_accept = (accept_rec *) apr_array_push(neg->accepts);

        new_accept->name = "*/*";
        new_accept->quality = 1.0f;
        new_accept->level = 0.0f;
    }

    new_accept = (accept_rec *) apr_array_push(neg->accepts);

    new_accept->name = CGI_MAGIC_TYPE;
    if (neg->use_rvsa) {
        new_accept->quality = 0;
    }
    else {
        new_accept->quality = prefer_scripts ? 2.0f : 0.001f;
    }
    new_accept->level = 0.0f;
}

/*****************************************************************
 *
 * Parsing type-map files, in Roy's meta/http format augmented with
 * #-comments.
 */

/* Reading RFC822-style header lines, ignoring #-comments and
 * handling continuations.
 */

enum header_state {
    header_eof, header_seen, header_sep
};

static enum header_state get_header_line(char *buffer, int len, apr_file_t *map)
{
    char *buf_end = buffer + len;
    char *cp;
    char c;

    /* Get a noncommented line */

    do {
        if (apr_file_gets(buffer, MAX_STRING_LEN, map) != APR_SUCCESS) {
            return header_eof;
        }
    } while (buffer[0] == '#');

    /* If blank, just return it --- this ends information on this variant */

    for (cp = buffer; apr_isspace(*cp); ++cp) {
        continue;
    }

    if (*cp == '\0') {
        return header_sep;
    }

    /* If non-blank, go looking for header lines, but note that we still
     * have to treat comments specially...
     */

    cp += strlen(cp);

    /* We need to shortcut the rest of this block following the Body:
     * tag - we will not look for continutation after this line.
     */
    if (!strncasecmp(buffer, "Body:", 5))
        return header_seen;

    while (apr_file_getc(&c, map) != APR_EOF) {
        if (c == '#') {
            /* Comment line */
            while (apr_file_getc(&c, map) != APR_EOF && c != '\n') {
                continue;
            }
        }
        else if (apr_isspace(c)) {
            /* Leading whitespace.  POSSIBLE continuation line
             * Also, possibly blank --- if so, we ungetc() the final newline
             * so that we will pick up the blank line the next time 'round.
             */

            while (c != '\n' && apr_isspace(c)) {
                if (apr_file_getc(&c, map) != APR_SUCCESS) {
                    break;
                }
            }

            apr_file_ungetc(c, map);

            if (c == '\n') {
                return header_seen;     /* Blank line */
            }

            /* Continuation */

            while (   cp < buf_end - 2
                   && (apr_file_getc(&c, map)) != APR_EOF
                   && c != '\n') {
                *cp++ = c;
            }

            *cp++ = '\n';
            *cp = '\0';
        }
        else {

            /* Line beginning with something other than whitespace */

            apr_file_ungetc(c, map);
            return header_seen;
        }
    }

    return header_seen;
}

static apr_off_t get_body(char *buffer, apr_size_t *len, const char *tag,
                          apr_file_t *map)
{
    char *endbody;
    apr_size_t bodylen;
    apr_off_t pos;


    /* We are at the first character following a body:tag\n entry
     * Suck in the body, then backspace to the first char after the
     * closing tag entry.  If we fail to read, find the tag or back
     * up then we have a hosed file, so give up already
     */
    --*len; /* Reserve space for '\0' */
    if (apr_file_read(map, buffer, len) != APR_SUCCESS) {
        return -1;
    }
    buffer[*len] = '\0';

    endbody = ap_strstr(buffer, tag);
    if (!endbody) {
        return -1;
    }
    bodylen = endbody - buffer;
    endbody += strlen(tag);
    /* Skip all the trailing cruft after the end tag to the next line */
    while (*endbody) {
        if (*endbody == '\n') {
            ++endbody;
            break;
        }
        ++endbody;
    }

    pos = -(apr_off_t)(*len - (endbody - buffer));
    if (apr_file_seek(map, APR_CUR, &pos) != APR_SUCCESS) {
        return -1;
    }

    /* Give the caller back the actual body's file offset and length */
    *len = bodylen;
    return pos - (endbody - buffer);
}


/* Stripping out RFC822 comments */

static void strip_paren_comments(char *hdr)
{
    /* Hmmm... is this correct?  In Roy's latest draft, (comments) can nest! */
    /* Nope, it isn't correct.  Fails to handle backslash escape as well.    */

    while (*hdr) {
        if (*hdr == '"') {
            hdr = strchr(hdr, '"');
            if (hdr == NULL) {
                return;
            }
            ++hdr;
        }
        else if (*hdr == '(') {
            while (*hdr && *hdr != ')') {
                *hdr++ = ' ';
            }

            if (*hdr) {
                *hdr++ = ' ';
            }
        }
        else {
            ++hdr;
        }
    }
}

/* Getting to a header body from the header */

static char *lcase_header_name_return_body(char *header, request_rec *r)
{
    char *cp = header;

    for ( ; *cp && *cp != ':' ; ++cp) {
        *cp = apr_tolower(*cp);
    }

    if (!*cp) {
        ap_log_rerror(APLOG_MARK, APLOG_ERR, 0, r, APLOGNO(00681)
                      "Syntax error in type map, no ':' in %s for header %s",
                      r->filename, header);
        return NULL;
    }

    do {
        ++cp;
    } while (apr_isspace(*cp));

    if (!*cp) {
        ap_log_rerror(APLOG_MARK, APLOG_ERR, 0, r, APLOGNO(00682)
                      "Syntax error in type map --- no header body: %s for %s",
                      r->filename, header);
        return NULL;
    }

    return cp;
}

static int read_type_map(apr_file_t **map, negotiation_state *neg,
                         request_rec *rr)
{
    request_rec *r = neg->r;
    apr_file_t *map_ = NULL;
    apr_status_t status;
    char buffer[MAX_STRING_LEN];
    enum header_state hstate;
    struct var_rec mime_info;
    int has_content;

    if (!map)
        map = &map_;

    /* We are not using multiviews */
    neg->count_multiviews_variants = 0;

    if ((status = apr_file_open(map, rr->filename, APR_READ | APR_BUFFERED,
                APR_OS_DEFAULT, neg->pool)) != APR_SUCCESS) {
        ap_log_rerror(APLOG_MARK, APLOG_ERR, status, r, APLOGNO(00683)
                      "cannot access type map file: %s", rr->filename);
        if (APR_STATUS_IS_ENOTDIR(status) || APR_STATUS_IS_ENOENT(status)) {
            return HTTP_NOT_FOUND;
        }
        else {
            return HTTP_FORBIDDEN;
        }
    }

    clean_var_rec(&mime_info);
    has_content = 0;

    do {
        hstate = get_header_line(buffer, MAX_STRING_LEN, *map);

        if (hstate == header_seen) {
            char *body1 = lcase_header_name_return_body(buffer, neg->r);
            const char *body;

            if (body1 == NULL) {
                return HTTP_INTERNAL_SERVER_ERROR;
            }

            strip_paren_comments(body1);
            body = body1;

            if (!strncmp(buffer, "uri:", 4)) {
                mime_info.file_name = ap_get_token(neg->pool, &body, 0);
            }
            else if (!strncmp(buffer, "content-type:", 13)) {
                struct accept_rec accept_info;

                get_entry(neg->pool, &accept_info, body);
                set_mime_fields(&mime_info, &accept_info);
                has_content = 1;
            }
            else if (!strncmp(buffer, "content-length:", 15)) {
                char *errp;
                apr_off_t number;

                body1 = ap_get_token(neg->pool, &body, 0);
                if (apr_strtoff(&number, body1, &errp, 10) != APR_SUCCESS
                    || *errp || number < 0) {
                    ap_log_rerror(APLOG_MARK, APLOG_ERR, 0, r, APLOGNO(00684)
                                  "Parse error in type map, Content-Length: "
                                  "'%s' in %s is invalid.",
                                  body1, r->filename);