mod_negotiation.c revision 1cde33c7e2019830f8fb3224e01649305583916e
/* ====================================================================
* The Apache Software License, Version 1.1
*
* Copyright (c) 2000-2002 The Apache Software Foundation. All rights
* reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* distribution.
*
* 3. The end-user documentation included with the redistribution,
* if any, must include the following acknowledgment:
* "This product includes software developed by the
* Apache Software Foundation (http://www.apache.org/)."
* Alternately, this acknowledgment may appear in the software itself,
* if and wherever such third-party acknowledgments normally appear.
*
* 4. The names "Apache" and "Apache Software Foundation" must
* not be used to endorse or promote products derived from this
* software without prior written permission. For written
* permission, please contact apache@apache.org.
*
* 5. Products derived from this software may not be called "Apache",
* nor may "Apache" appear in their name, without prior written
* permission of the Apache Software Foundation.
*
* THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESSED OR IMPLIED
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE APACHE SOFTWARE FOUNDATION OR
* ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
* USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
* ====================================================================
*
* This software consists of voluntary contributions made by many
* individuals on behalf of the Apache Software Foundation. For more
* information on the Apache Software Foundation, please see
*
* Portions of this software are based upon public domain software
* originally written at the National Center for Supercomputing Applications,
* University of Illinois, Urbana-Champaign.
*/
/*
* 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;
/* forcelangpriority flags
*/
#define FLP_UNDEF 0 /* Same as FLP_DEFAULT, but base overrides */
#define FLP_DEFAULT FLP_PREFER
{
return new;
}
{
/* give priority to the config in the subdirectory */
return new;
}
const char *lang)
{
neg_dir_config *n = n_;
const char **langp;
if (!n->language_priority)
return NULL;
}
{
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 NULL;
}
int arg)
{
return NULL;
}
static int do_cache_negotiated_docs(server_rec *s)
{
}
static const command_rec negotiation_cmds[] =
{
"Either 'on' or 'off' (default)"),
"space-delimited list of MIME language abbreviations"),
{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.
*
* 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 {
const char *mime_type; /* MUST be lowercase */
const char *file_name; /* Set to 'this' (for map file body content) */
const char *content_encoding;
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... */
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 {
request_rec *r;
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
*/
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 */
/* A few functions to manipulate var_recs.
* Cleaning out the fields...
*/
{
mime_info->is_pseudo_html = 0;
mime_info->mime_stars = 0;
}
/* Initializing the relevant fields of a variant record from the
* accept_info read out of its content-type, one way or another.
*/
{
}
/* Create a variant list validator in r using info from 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_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.
*/
/*
* Get a single mime type entry --- one media type and parameters;
* enter the values we recognize into the argument accept_rec
*/
const char *accept_line)
{
/*
* Note that this handles what I gather is the "old format",
*
*
* 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).
*/
* you'll *get* case insensitive.
*/
/* KLUDGE!!! Default HTML to level 2.0 unless the browser
* *explicitly* says something else.
*/
}
}
}
while (*accept_line == ';') {
/* Parameters ... */
char *parm;
char *cp;
char *end;
++accept_line;
/* Look for 'var = value' --- and make sure the var is in lcase. */
}
if (!*cp) {
continue; /* No '='; just ignore it. */
}
++cp;
}
if (*cp == '"') {
++cp;
end++);
}
else {
}
if (*end) {
}
if (parm[0] == 'q'
}
}
}
}
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.
*/
{
if (!accept_line) {
return NULL;
}
while (*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
*/
{
if (!lang_line) {
return lang_recs;
}
while (**lang_line) {
++(*lang_line);
}
}
return lang_recs;
}
/*****************************************************************
*
* Handling header lines from clients...
*/
{
int i;
new->r = r;
/* calculate new->accept_q value */
}
}
}
new->accept_langs =
/* This is possibly overkill for some servers, heck, we have
* only 33 index.html variants in docs/docroot (today).
* Make this configurable?
*/
return new;
}
{
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->dont_fiddle_headers = 0;
if (!negotiate)
return;
/* 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.
*/
return;
}
apr_isdigit(tok[0]) ||
/* The user agent supports transparent negotiation */
/* Send-alternates could be configurable, but note
* that it must be 1 if we have 'vlist' in the
* negotiate header.
*/
/* we may use the RVSA/1.0 algorithm, configure for it */
}
else if (tok[0] == '*') {
/* we may use any variant selection algorithm, configure
* to use the Apache algorithm
*/
/* We disable header fiddles on the assumption that a
* client sending Negotiate knows how to send correct
* headers which don't need fiddling.
*/
}
}
}
#ifdef NEG_DEBUG
"dont_fiddle_headers=%d use_rvsa=%d ua_supports_trans=%d "
"send_alternates=%d, may_choose=%d",
#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).
*/
int prefer_scripts)
{
}
new_accept->quality = 0;
}
else {
}
}
/*****************************************************************
*
* #-comments.
*/
/* Reading RFC822-style header lines, ignoring #-comments and
* handling continuations.
*/
enum header_state {
};
{
char *cp;
char c;
/* Get a noncommented line */
do {
return header_eof;
}
} while (buffer[0] == '#');
/* If blank, just return it --- this ends information on this variant */
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...
*/
/* We need to shortcut the rest of this block following the Body:
* tag - we will not look for continutation after this line.
*/
return header_seen;
if (c == '#') {
/* Comment line */
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)) {
break;
}
apr_file_ungetc(c, map);
if (c == '\n') {
return header_seen; /* Blank line */
}
/* Continuation */
*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;
}
{
char *endbody;
int bodylen;
int taglen;
/* 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
*/
return -1;
}
return -1;
}
/* Skip all the trailing cruft after the end tag to the next line */
while (*endbody) {
if (*endbody == '\n') {
++endbody;
break;
}
++endbody;
}
return -1;
}
/* Give the caller back the actual body's file offset and length */
}
/* 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 == '"') {
return;
}
++hdr;
}
else if (*hdr == '(') {
*hdr++ = ' ';
}
if (*hdr) {
*hdr++ = ' ';
}
}
else {
++hdr;
}
}
}
/* Getting to a header body from the header */
{
}
if (!*cp) {
"Syntax error in type map, no ':' in %s for header %s",
return NULL;
}
do {
++cp;
if (!*cp) {
"Syntax error in type map --- no header body: %s for %s",
return NULL;
}
return cp;
}
{
request_rec *r = neg->r;
char buffer[MAX_STRING_LEN];
enum header_state hstate;
int has_content;
if (!map)
/* We are not using multiviews */
neg->count_multiviews_variants = 0;
return HTTP_FORBIDDEN;
}
has_content = 0;
do {
if (hstate == header_seen) {
const char *body;
return HTTP_INTERNAL_SERVER_ERROR;
}
}
struct accept_rec accept_info;
has_content = 1;
}
has_content = 1;
}
&body);
has_content = 1;
}
has_content = 1;
}
char *cp;
}
}
*eol = '\0';
"Syntax error in type map, no end tag '%s'"
"found in %s for Body: content.",
break;
}
}
}
else {
}
has_content = 0;
}
} while (hstate != header_eof);
if (map_)
set_vlist_validator(r, rr);
return OK;
}
/* Sort function used by read_types_multi. */
/* First key is the source quality, sort in descending order. */
/* XXX: note that we currently implement no method of setting the
* source quality for multiviews variants, so we are always comparing
* 1.0 to 1.0 for now
*/
if (a->source_quality < b->source_quality)
return 1;
if (a->source_quality > b->source_quality)
return -1;
/* Second key is the variant name */
}
/*****************************************************************
*
* Same as read_type_map, except we use a filtered directory listing
* as the map...
*/
{
request_rec *r = neg->r;
char *filp;
int prefix_len;
struct accept_rec accept_info;
void *new_var;
int anymatch = 0;
return DECLINED; /* Weird... */
}
/* XXX this should be more general, and quit using 'specials' */
return DECLINED;
}
++filp;
return HTTP_FORBIDDEN;
}
/* Do we have a match? */
#ifdef CASE_BLIND_FILESYSTEM
#else
#endif
continue;
}
continue;
}
/* Don't negotiate directories and other unusual files
* and we aught to discover if the LNK was interesting.
*
* Of course, this only helps platforms that capture the
* the filetype in apr_dir_read(), which most can once
* they are optimized with some magic [it's known to the
* dirent, not associated to the inode, on most FS's.]
*/
continue;
/* Ok, something's here. Maybe nothing useful. Remember that
* we tried, if we completely fail, so we can reject the request!
*/
anymatch = 1;
/* See if it's something which we have access to, and which
* has a known type and encoding (as opposed to something
* which we'll be slapping default_type on later).
*/
/* Double check, we still don't multi-resolve non-ordinary files
*/
continue;
/* If it has a handler, we'll pretend it's a CGI script,
* since that's a good indication of the sort of thing it
* might be doing.
*/
}
/*
* mod_mime will _always_ provide us the base name in the
* ap-mime-exception-list, if it processed anything. If
* this list is empty, give up immediately, there was
* nothing interesting. For example, looking at the files
* readme.txt and readme.foo, we will throw away .foo if
* it's an insignificant file (e.g. did not identify a
* language, charset, encoding, content type or handler,)
*/
"ap-mime-exceptions-list");
if (!exception_list) {
continue;
}
/* Each unregonized bit better match our base name, in sequence.
* A test of index.html.foo will match index.foo or index.html.foo,
* but it will never transpose the segments and allow index.foo.html
* because that would introduce too much CPU consumption. Better that
* we don't attempt a many-to-many match here.
*/
{
*segend = '\0';
#ifdef CASE_BLIND_FILESYSTEM
#else
#endif
--nexcept;
++cur_except;
}
if (!saveend)
break;
}
if (nexcept) {
/* Something you don't know is, something you don't know...
*/
continue;
}
}
/*
* ###: be warned, the _default_ content type is already
* picked up here! If we failed the subrequest, or don't
* know what we are serving, then continue.
*/
continue;
}
/* If it's a map file, we use that instead of the map
* we're building...
*/
if (((sub_req->content_type) &&
}
}
/* Have reasonable variant --- gather notes. */
if (sub_req->content_encoding) {
}
if (sub_req->content_languages) {
}
}
/* We found some file names that matched. None could be served.
* Rather than fall out to autoindex or some other mapper, this
* request must die.
*/
"Negotiation: discovered file(s) matching request: %s"
" (None could be negotiated).",
r->filename);
return HTTP_NOT_FOUND;
}
set_vlist_validator(r, r);
/* Sort the variants into a canonical order. The negotiation
* result sometimes depends on the order of the variants. By
* sorting the variants into a canonical order, rather than using
* the order in which readdir() happens to return them, we ensure
* that the negotiation result will be consistent over filesystem
*/
sizeof(var_rec), (int (*)(const void *, const void *)) variantsortf);
return OK;
}
/*****************************************************************
* And now for the code you've been waiting for... actually
* finding a match to the client's requirements.
*/
* try mentioning those three characters in a C comment). Using strcmp()
* is legit, because everything has already been smashed to lowercase.
*
* Note also that if we get an exact match on the media type, we update
* level_matched for use in level_cmp below...
*
* We also give a value for mime_stars, which is used later. It should
*/
{
}
return 1;
}
}
return 1;
}
return 1;
}
}
return OK;
}
/* This code implements a piece of the tie-breaking algorithm between
* variants of equal quality. This piece is the treatment of variants
* of the same base media type, but different levels. What we want to
* return is the variant at the highest level that the client explicitly
* claimed to accept.
*
* If all the variants available are at a higher level than that, or if
* the client didn't say anything specific about this media type at all
* and these variants just got in on a wildcard, we prefer the lowest
* level, on grounds that that's the one that the client is least likely
* to choke on.
*
* (This is all motivated by treatment of levels in HTML --- we only
* want to give level 3 to browsers that explicitly ask for it; browsers
* that don't, including HTTP/0.9 browsers that only get the implicit
* "Accept: * / *" [space added to avoid confusing cpp --- no, that
* syntax doesn't really work] should get HTML2 if available).
*
* (Note that this code only comes into play when we are choosing among
* variants of equal quality, where the draft standard gives us a fair
* bit of leeway about what to do. It ain't specified by the standard;
* rather, it is a choice made by this server about what to do in cases
* where the standard does not specify a unique course of action).
*/
{
/* Levels are only comparable between matching media types */
return 0;
}
return 0;
}
/* The result of the above if statements is that, if we get to
* here, both variants have the same mime_type or both are
* pseudo-html.
*/
/* Take highest level that matched, if either did match. */
return 1;
}
return -1;
}
/* Neither matched. Take lowest level, if there's a difference. */
return 1;
}
return -1;
}
/* Tied */
return 0;
}
/* Finding languages. The main entry point is set_language_quality()
* which is called for each variant. It sets two elements in the
* variant record:
* language_quality - the 'q' value of the 'best' matching language
* from Accept-Language: header (HTTP/1.1)
* lang_index - Non-negotiated language priority, using
* position of language on the Accept-Language:
* header, if present, else LanguagePriority
* directive order.
*
* When we do the variant checking for best variant, we use language
* quality first, and if a tie, language_index next (this only applies
* when _not_ using the RVSA/1.0 algorithm). If using the RVSA/1.0
* algorithm, lang_index is never used.
*
* set_language_quality() calls find_lang_index() and find_default_index()
* to set lang_index.
*/
{
const char **alang;
int i;
if (!lang || !accept_langs) {
return -1;
}
for (i = 0; i < accept_langs->nelts; ++i) {
return i;
}
}
return -1;
}
/* set_default_lang_quality() sets the quality we apply to variants
* which have no language assigned to them. If none of the variants
* have a language, we are not negotiating on language, so all are
* acceptable, and we set the default q value to 1.0. However if
* some of the variants have languages, we set this default to 0.0001.
* The value of this default will be applied to all variants with
* no explicit language -- which will have the effect of making them
* acceptable, but only if no variants with an explicit language
* are acceptable. The default q value set here is assigned to variants
* with no language type in set_language_quality().
*
* Note that if using the RVSA/1.0 algorithm, we don't use this
* fiddle.
*/
{
int j;
if (!neg->dont_fiddle_headers) {
if (variant->content_languages &&
return;
}
}
}
}
/* Set the language_quality value in the variant record. Also
* assigns lang_index for ForceLanguagePriority.
*
* To find the language_quality value, we look for the 'q' value
* of the 'best' matching language on the Accept-Language
* header. The 'best' match is the language on Accept-Language
* header which matches the language of this variant either fully,
* or as far as the prefix marker (-). If two or more languages
* match, use the longest string from the Accept-Language header
* (see HTTP/1.1 [14.4])
*
* When a variant has multiple languages, we find the 'best'
* match for each variant language tag as above, then select the
* one with the highest q value. Because both the accept-header
* and variant can have multiple languages, we now have a hairy
* loop-within-a-loop here.
*
* If the variant has no language and we have no Accept-Language
* items, leave the quality at 1.0 and return.
*
* If the variant has no language, we use the default as set by
* set_default_lang_quality() (1.0 if we are not negotiating on
* language, 0.001 if we are).
*
* Following the setting of the language quality, we drop through to
* set the old 'lang_index'. This is set based on either the order
* of the languages on the Accept-Language header, or the
* order on the LanguagePriority directive. This is only used
* in the negotiation if the language qualities tie.
*/
{
if (forcepriority == FLP_UNDEF) {
}
/* This variant has no content-language, so use the default
* quality factor for variants with no content-language
* (previously set by set_default_lang_quality()).
* Leave the factor alone (it remains at 1.0) when we may not fiddle
* with the headers.
*/
if (!neg->dont_fiddle_headers) {
}
if (!neg->accept_langs) {
return; /* no accept-language header */
}
return;
}
else {
/* Variant has one (or more) languages. Look for the best
* match. We do this by going through each language on the
* variant description looking for a match on the
* Accept-Language header. The best match is the longest
* matching language on the header. The final result is the
* best q value from all the languages on the variant
* description.
*/
if (!neg->accept_langs) {
/* no accept-language header makes the variant indefinite */
}
else { /* There is an accept-language with 0 or more items */
char *lang, *p;
float fiddle_q = 0.0f;
int any_match_on_star = 0;
int i, j;
p = NULL;
bestthistag = NULL;
alen = 0;
/* lang is the variant's language-tag, which is the one
* we are allowed to use the prefix of in HTTP/1.1
*/
/* now find the best (i.e. longest) matching
* Accept-Language header language. We put the best match
* for this tag in bestthistag. We cannot update the
* overall best (based on q value) because the best match
* for this tag is the longest language item on the accept
* header, not necessarily the highest q.
*/
if (!star) {
}
continue;
}
/* Find language. We match if either the variant
* language tag exactly matches the language range
* from the accept header, or a prefix of the variant
* language tag up to a '-' character matches the
* whole of the language range in the Accept-Language
* header. Note that HTTP/1.x allows any number of
* '-' characters in a tag or range, currently only
* tags with zero or one '-' characters are defined
* for general use (see rfc1766).
*
* We only use language range in the Accept-Language
* header the best match for the variant language tag
* if it is longer than the previous best match.
*/
if (alen > longest_lang_range_len) {
bestthistag = &accs[i];
}
}
/* The next bit is a fiddle. Some browsers might
* be configured to send more specific language
* ranges than desirable. For example, an
* Accept-Language of en-US should never match
* variants with languages en or en-GB. But US
* English speakers might pick en-US as their
* language choice. So this fiddle checks if the
* language range has a prefix, and if so, it
* matches variants which match that prefix with a
* priority of 0.001. So a request for en-US would
* match variants of types en and en-GB, but at
* much lower priority than matches of en-US
* directly, or of any other language listed on
* the Accept-Language header. Note that this
* fiddle does not handle multi-level prefixes.
*/
fiddle_q = 0.001f;
}
}
}
}
/* Finished looking at Accept-Language headers, the best
* (longest) match is in bestthistag, or NULL if no match
*/
if (!best ||
best = bestthistag;
}
/* See if the tag matches on a * in the Accept-Language
* header. If so, record this fact for later use
*/
if (!bestthistag && star) {
any_match_on_star = 1;
}
}
/* If one of the language tags of the variant matched on *, we
* need to see if its q is better than that of any non-* match
* on any other tag of the variant. If so the * match takes
* precedence and the overall match is not definite.
*/
if ( any_match_on_star &&
(!best)) ) {
}
}
}
/* Handle the ForceDefaultLanguage overrides, based on the best match
* to LanguagePriority order. The best match is the lowest index of
* any LanguagePriority match.
*/
if (((forcepriority & FLP_PREFER)
&& (variant->lang_index < 0))
|| ((forcepriority & FLP_FALLBACK)
&& !variant->lang_quality))
{
int bestidx = -1;
int j;
{
/* lang is the variant's language-tag, which is the one
* we are allowed to use the prefix of in HTTP/1.1
*/
int idx = -1;
/* If we wish to fallback or
* we use our own LanguagePriority index.
*/
}
}
if (bestidx >= 0) {
if (variant->lang_quality) {
if (forcepriority & FLP_PREFER) {
}
}
else {
if (forcepriority & FLP_FALLBACK) {
}
}
}
}
return;
}
/* Determining the content length --- if the map didn't tell us,
* we have to do a stat() and remember for next time.
*/
{
}
else {
}
}
}
}
/* For a given variant, find the best matching Accept: header
* and assign the Accept: header's quality value to the
* mime_type_quality field of the variant, for later use in
* determining the best matching variant.
*/
{
int i;
float q = 0.0f;
int q_definite = 1;
/* if no Accept: header, leave quality alone (will
* remain at the default value of 1)
*
* XXX: This if is currently never true because of the effect of
* maybe_add_default_accepts().
*/
return;
}
/*
* Go through each of the ranges on the Accept: header,
* looking for the 'best' match with this variant's
* content-type. We use the best match's quality
* value (from the Accept: header) for this variant's
* mime_type_quality field.
*
* The best match is determined like this:
*/
int prev_mime_stars;
continue; /* didn't match the content type at all */
}
else {
/* did match - see if there were less or more stars than
* in previous match
*/
continue; /* more stars => not as good a match */
}
}
/* If we are allowed to mess with the q-values
* and have no explicit q= parameters in the accept header,
* make wildcards very low, so we have a low chance
* of ending up with them if there's something better.
*/
q = 0.01f;
}
q = 0.02f;
}
else {
}
}
variant->mime_type_quality = q;
}
/* For a given variant, find the 'q' value of the charset given
* on the Accept-Charset line. If no charsets are listed,
* assume value of '1'.
*/
{
int i;
/* if no Accept-Charset: header, leave quality alone (will
* remain at the default value of 1)
*/
if (!neg->accept_charsets) {
return;
}
/* Charset of variant not known */
/* if not a text / * type, leave quality alone */
))
return;
/* Don't go guessing if we are in strict header mode,
* e.g. when running the rvsa, as any guess won't be reflected
* in the variant list or content-location headers.
*/
if (neg->dont_fiddle_headers)
return;
}
/*
* Go through each of the items on the Accept-Charset header,
* looking for a match with this variant's charset. If none
* match, charset is unacceptable, so set quality to 0.
*/
return;
}
}
}
/* No explicit match */
if (star) {
return;
}
/* If this variant is in charset iso-8859-1, the default is 1.0 */
}
else {
}
}
/* is_identity_encoding is included for back-compat, but does anyone
* use 7bit, 8bin or binary in their var files??
*/
static int is_identity_encoding(const char *enc)
{
}
/*
* set_encoding_quality determines whether the encoding for a particular
* variant is acceptable for the user-agent.
*
* The rules for encoding are that if the user-agent does not supply
* any Accept-Encoding header, then all encodings are allowed but a
* variant with no encoding should be preferred.
* If there is an empty Accept-Encoding header, then no encodings are
* acceptable. If there is a non-empty Accept-Encoding header, then
* any of the listed encodings are acceptable, as well as no encoding
* unless the "identity" encoding is specifically excluded.
*/
{
float value_if_not_found = 0.0f;
int i;
if (!neg->accept_encodings) {
/* We had no Accept-Encoding header, assume that all
* encodings are acceptable with a low quality,
* but we prefer no encoding if available.
*/
else
return;
}
enc = "identity";
value_if_not_found = 0.0001f;
}
/* Go through each of the encodings on the Accept-Encoding: header,
* looking for a match with our encoding. x- prefixes are ignored.
*/
enc += 2;
}
name += 2;
}
return;
}
star = &accept_recs[i];
}
}
/* No explicit match */
if (star) {
return;
}
/* Encoding not found on Accept-Encoding: header, so it is
* _not_ acceptable unless it is the identity (no encoding)
*/
}
/*************************************************************
* Possible results of the variant selection algorithm
*/
enum algorithm_results {
alg_list /* list variants */
};
/* Below is the 'best_match' function. It returns an int, which has
* one of the two values alg_choice or alg_list, which give the result
* of the variant selection algorithm. alg_list means that no best
* variant was found by the algorithm, alg_choice means that a best
* terminology comes from TCN (rfc2295), but is used in a more generic
* way here. The best variant is returned in *pbest. best_match has
* two possible algorithms for determining the best variant: the
* RVSA/1.0 algorithm (from RFC2296), and the standard Apache
* algorithm. These are split out into separate functions
* (is_variant_better_rvsa() and is_variant_better()). Selection of
* one is through the neg->use_rvsa flag.
*
* The call to best_match also creates full information, including
* language, charset, etc quality for _every_ variant. This is needed
* for generating a correct Vary header, and can be used for the
* Alternates header, the human-readable list responses and 406 errors.
*/
/* Firstly, the RVSA/1.0 (HTTP Remote Variant Selection Algorithm
* v1.0) from rfc2296. This is the algorithm that goes together with
* transparent content negotiation (TCN).
*/
{
/* TCN does not cover negotiation on content-encoding. For now,
* we ignore the encoding unless it was explicitly excluded.
*/
return 0;
q = variant->mime_type_quality *
/* RFC 2296 calls for the result to be rounded to 5 decimal places,
* but we don't do that because it serves no useful purpose other
* than to ensure that a remote algorithm operates on the same
* precision as ours. That is silly, since what we obviously want
* is for the algorithm to operate on the best available precision
* regardless of who runs it. Since the above calculation may
* result in significant variance at 1e-12, rounding would be bogus.
*/
#ifdef NEG_DEBUG
"Variant: file=%s type=%s lang=%s sourceq=%1.3f "
"mimeq=%1.3f langq=%1.3f charq=%1.3f encq=%1.3f "
"q=%1.5f definite=%d",
: ""),
q,
#endif
if (q <= 0.0f) {
return 0;
}
if (q > bestq) {
*p_bestq = q;
return 1;
}
if (q == bestq) {
/* If the best variant's encoding is of lesser quality than
* this variant, then we prefer this variant
*/
*p_bestq = q;
return 1;
}
}
return 0;
}
/* Negotiation algorithm as used by previous versions of Apache
* (just about).
*/
{
int levcmp;
/* For non-transparent negotiation, server can choose how
* to handle the negotiation. We'll use the following in
* order: content-type, language, content-type level, charset,
* content encoding, content length.
*
* For each check, we have three possible outcomes:
* This variant is worse than current best: return 0
* This variant is better than the current best:
* assign this variant's q to *p_bestq, and return 1
* This variant is just as desirable as the current best:
* drop through to the next test.
*
* This code is written in this long-winded way to allow future
* customisation, either by the addition of additional
* checks, or to allow the order of the checks to be determined
* by configuration options (e.g. we might prefer to check
* language quality _before_ content type).
*/
/* First though, eliminate this variant if it is not
* acceptable by type, charset, encoding or language.
*/
#ifdef NEG_DEBUG
"Variant: file=%s type=%s lang=%s sourceq=%1.3f "
"mimeq=%1.3f langq=%1.3f langidx=%d charq=%1.3f encq=%1.3f ",
: ""),
#endif
return 0; /* don't consider unacceptables */
}
if (q == 0.0 || q < bestq) {
return 0;
}
*p_bestq = q;
return 1;
}
/* language */
return 0;
}
*p_bestq = q;
return 1;
}
/* if language qualities were equal, try the LanguagePriority stuff */
return 0;
}
*p_bestq = q;
return 1;
}
* support it on other types too)
*/
if (levcmp == -1) {
return 0;
}
if (levcmp == 1) {
*p_bestq = q;
return 1;
}
/* charset */
return 0;
}
/* If the best variant's charset is ISO-8859-1 and this variant has
* the same charset quality, then we prefer this variant
*/
*p_bestq = q;
return 1;
}
/* Prefer the highest value for encoding_quality.
*/
return 0;
}
*p_bestq = q;
return 1;
}
/* content length if all else equal */
return 0;
}
/* ok, to get here means every thing turned out equal, except
* we have a shorter content length, so use this variant
*/
*p_bestq = q;
return 1;
}
{
int j;
float bestq = 0.0f;
/*
* Find the 'best' variant
*/
/* Find all the relevant 'quality' values from the
* Accept... headers, and store in the variant. This also
* prepares for sending an Alternates header etc so we need to
* do it even if we do not actually plan to find a best
* variant.
*/
/* Only do variant selection if we may actually choose a
* variant for the client
*/
if (neg->may_choose) {
/* Now find out if this variant is better than the current
* best, either using the RVSA/1.0 algorithm, or Apache's
* internal server-driven algorithm. Presumably other
* server-driven algorithms are possible, and could be
* implemented here.
*/
}
}
else {
}
}
}
}
/* We now either have a best variant, or no best variant */
/* calculate result for RVSA/1.0 algorithm:
* only a choice response if the best variant has q>0
* and is definite
*/
}
else {
/* calculate result for Apache negotiation algorithm */
}
/* Returning a choice response with a non-neighboring variant is a
* protocol security error in TCN (see rfc2295). We do *not*
* verify here that the variant and URI are neighbors, even though
* we may return alg_choice. We depend on the environment (the
* caller) to only declare the resource transparently negotiable if
* all variants are neighbors.
*/
return algorithm_result;
}
/* Sets response headers for a negotiated response.
* neg->is_transparent determines whether a transparently negotiated
* response or a plain `server driven negotiation' response is
* created. Applicable headers are Alternates, Vary, and TCN.
*
* The Vary header we create is sometimes longer than is required for
* the correct caching of negotiated results by HTTP/1.1 caches. For
* example if we have 3 variants x.html, x.ps.en and x.ps.nl, and if
* the Accept: header assigns a 0 quality to .ps, then the results of
* the two server-side negotiation algorithms we currently implement
* will never depend on Accept-Language so we could return `Vary:
* negotiate, accept' instead of the longer 'Vary: negotiate, accept,
* accept-language' which the code below will return. A routine for
* computing the exact minimal Vary header would be a huge pain to code
* and maintain though, especially because we need to take all possible
* twiddles in the server-side negotiation algorithms into account.
*/
int alg_result)
{
const char *sample_type = NULL;
const char *sample_language = NULL;
const char *sample_encoding = NULL;
const char *sample_charset = NULL;
char *lang;
char *qstr;
char *lenstr;
int first_variant = 1;
int vary_by_type = 0;
int vary_by_language = 0;
int vary_by_charset = 0;
int vary_by_encoding = 0;
int j;
/* In order to avoid O(n^2) memory copies in building Alternates,
* we preallocate a apr_table_t with the maximum substrings possible,
* fill it with the variant list, and then concatenate the entire array.
* Note that if you change the number of substrings pushed, you also
* need to change the calculation of max_vlist_array above.
*/
else
/* Put headers into err_headers_out, since send_http_header()
* outputs both headers_out and err_headers_out.
*/
hdrs = r->err_headers_out;
}
else {
}
/* Calculate Vary by looking for any difference between variants */
if (first_variant) {
}
else {
if (!vary_by_type &&
vary_by_type = 1;
}
if (!vary_by_charset &&
vary_by_charset = 1;
}
if (!vary_by_language &&
vary_by_language = 1;
}
if (!vary_by_encoding &&
vary_by_encoding = 1;
}
}
first_variant = 0;
if (!neg->send_alternates)
continue;
/* Generate the string components for this Alternates entry */
/* Strip trailing zeros (saves those valuable network bytes) */
}
}
}
}
}
if (lang) {
}
/* Strictly speaking, this is non-standard, but so is TCN */
}
/* Note that the Alternates specification (in rfc2295) does
* not require that we include {length x}, so we could omit it
* if determining the length is too expensive. We currently
* always include it though. 22 bytes is enough for 2^64.
*
* If the variant is a CGI script, find_content_length would
* return the length of the script, not the output it
* produces, so we check for the presence of a handler and if
* there is one we don't add a length.
*
* XXX: TODO: This check does not detect a CGI script if we
* get the variant from a type map. This needs to be fixed
* (without breaking things if the type map specifies a
* content-length, which currently leads to the correct result).
*/
}
}
}
}
}
}
/**********************************************************************
*
* Return an HTML list of variants. This is output as part of the
* choice response or 406 status body.
*/
{
int i;
/* In order to avoid O(n^2) memory copies in building the list,
* we preallocate a apr_table_t with the maximum substrings possible,
* fill it with the variant list, and then concatenate the entire array.
*/
/* The format isn't very neat, and it would be nice to make
* the tags human readable (eg replace 'language en' with 'English').
* Note that if you change the number of substrings pushed, you also
* need to change the calculation of max_vlist_array above.
*/
}
languages, ',');
}
}
if (variant->content_encoding) {
}
}
}
{
}
else {
}
}
/* Called if we got a "Choice" response from the variant selection algorithm.
* It checks the result of the chosen variant to see if it
* is itself negotiated (if so, return error HTTP_VARIANT_ALSO_VARIES).
* Otherwise, add the appropriate headers to the current response.
*/
{
const char *sub_vary;
int status;
return status;
}
}
else {
}
/* The variant selection algorithm told us to return a "Choice"
* response. This is the normal variant response, with
* some extra headers. First, ensure that the chosen
* variant did or will not itself engage in transparent negotiation.
* If not, set the appropriate headers, and fall through to
* the normal variant handling
*/
/* This catches the error that a transparent type map selects a
* transparent multiviews resource as the best variant.
*
* XXX: We do not signal an error if a transparent type map
* selects a _non_transparent multiviews resource as the best
* variant, because we can generate a legal negotiation response
* in this case. In this case, the vlist_validator of the
* nontransparent subrequest will be lost however. This could
* lead to cases in which a change in the set of variants or the
* negotiation algorithm of the nontransparent resource is never
* propagated up to a HTTP/1.1 cache which interprets Vary. To be
* completely on the safe side we should return HTTP_VARIANT_ALSO_VARIES
* for this type of recursive negotiation too.
*/
if (neg->is_transparent &&
return HTTP_VARIANT_ALSO_VARIES;
}
/* This catches the error that a transparent type map recursively
* selects, as the best variant, another type map which itself
* causes transparent negotiation to be done.
*
* XXX: Actually, we catch this error by catching all cases of
* type map recursion. There are some borderline recursive type
* map arrangements which would not produce transparent
* negotiation protocol errors or lack of cache propagation
* problems, but such arrangements are very hard to detect at this
* point in the control flow, so we do not bother to single them
* out.
*
* Recursive type maps imply a recursive arrangement of negotiated
* resources which is visible to outside clients, and this is not
* supported by the transparent negotiation caching protocols, so
* if we are to have generic support for recursive type maps, we
* have to create some configuration setting which makes all type
* maps non-transparent when recursion is enabled. Also, if we
* want recursive type map support which ensures propagation of
* type map changes into HTTP/1.1 caches that handle Vary, we
* would have to extend the current mechanism for generating
* variant list validators.
*/
return HTTP_VARIANT_ALSO_VARIES;
}
/* This adds an appropriate Variant-Vary header if the subrequest
* is a multiviews resource.
*
* XXX: TODO: Note that this does _not_ handle any Vary header
* returned by a CGI if sub_req is a CGI script, because we don't
* see that Vary header yet at this point in the control flow.
* This won't cause any cache consistency problems _unless_ the
* CGI script also returns a Cache-Control header marking the
* response as cachable. This needs to be fixed, also there are
* problems if a CGI returns an Etag header which also need to be
* fixed.
*/
/* Move the subreq Vary header into the main request to
* prevent having two Vary headers in the response, which
* would be legal but strange.
*/
}
/* Still to do by caller: add Expires */
return 0;
}
/****************************************************************
*
* Executive...
*/
{
int alg_result; /* result of variant selection algorithm */
int res;
int j;
/* Decide if resource is transparently negotiable */
/* GET or HEAD? (HEAD has same method number as GET) */
if (r->method_number == M_GET) {
/* maybe this should be configurable, see also the comment
* about recursive type maps in setup_choice_response()
*/
/* We can't be transparent if we are a map file in the middle
* of the request URI.
*/
neg->is_transparent = 0;
/* We can't be transparent, because of internal
* assumptions in best_match(), if there is a
* non-neighboring variant. We can have a non-neighboring
* variant when processing a type map.
*/
neg->is_transparent = 0;
}
}
if (neg->is_transparent) {
}
else { /* configure negotiation on non-transparent resource */
}
/* alg_result is one of
* alg_choice: a best variant is chosen
* alg_list: no best variant is chosen
*/
if (alg_result == alg_list) {
/* send a list response or HTTP_NOT_ACCEPTABLE error response */
store_variant_list(r, neg);
/* XXX todo: expires? cachability? */
/* Some HTTP/1.0 clients are known to choke when they get
* a 300 (multiple choices) response without a Location
* header. However the 300 code response we are are about
* to generate will only reach 1.0 clients which support
* transparent negotiation, and they should be OK. The
* response should never reach older 1.0 clients, even if
* we have CacheNegotiatedDocs enabled, because no 1.0
* proxy cache (we know of) will cache and return 300
* responses (they certainly won't if they conform to the
* HTTP/1.0 specification).
*/
return HTTP_MULTIPLE_CHOICES;
}
if (!*bestp) {
"no acceptable variant: %s", r->filename);
return HTTP_NOT_ACCEPTABLE;
}
}
/* Variant selection chose a variant */
/* XXX todo: merge the two cases in the if statement below */
if (neg->is_transparent) {
return res; /* return if error */
}
}
else {
}
/* Make sure caching works - Vary should handle HTTP/1.1, but for
* HTTP/1.0, we can't allow caching at all.
*/
/* XXX: Note that we only set r->no_cache to 1, which causes
* Expires: <now> to be added, when responding to a HTTP/1.0
* client. If we return the response to a 1.1 client, we do not
* add Expires <now>, because doing so would degrade 1.1 cache
* performance by preventing re-use of the response without prior
* revalidation. On the other hand, if the 1.1 client is a proxy
* which was itself contacted by a 1.0 client, or a proxy cache
* which can be contacted later by 1.0 clients, then we currently
* rely on this 1.1 proxy to add the Expires: <now> when it
* forwards the response.
*
* XXX: TODO: Find out if the 1.1 spec requires proxies and
* tunnels to add Expires: <now> when forwarding the response to
* 1.0 clients. I (kh) recall it is rather vague on this point.
* Testing actual 1.1 proxy implementations would also be nice. If
* Expires: <now> is not added by proxies then we need to always
* include Expires: <now> ourselves to ensure correct caching, but
* this would degrade HTTP/1.1 cache efficiency unless we also add
* Cache-Control: max-age=N, which we currently don't.
*
* Roy: No, we are not going to screw over HTTP future just to
* ensure that people who can't be bothered to upgrade their
* clients will always receive perfect server-side negotiation.
* Hell, those clients are sending bogus accept headers anyway.
*
* Manual setting of cache-control/expires always overrides this
* automated kluge, on purpose.
*/
if ((!do_cache_negotiated_docs(r->server)
r->no_cache = 1;
}
return OK;
}
static int handle_map_file(request_rec *r)
{
int res;
char *udir;
return DECLINED;
neg = parse_accept_headers(r);
return res;
}
{
conn_rec *c = r->connection;
apr_bucket *e;
return res;
}
/*if (r->method_number == M_OPTIONS) {
* return ap_send_http_options(r);
*}
*/
return HTTP_METHOD_NOT_ALLOWED;
}
/* ### These may be implemented by adding some 'extra' info
* of the file offset onto the etag
* ap_update_mtime(r, r->finfo.mtime);
* ap_set_last_modified(r);
* ap_set_etag(r);
*/
return res;
}
c->bucket_alloc);
e = apr_bucket_eos_create(c->bucket_alloc);
}
/* remove any path_info from the end of the uri before trying
* to change the filename. r->path_info from the original
* request is passed along on the redirect.
*/
}
return OK;
}
static int handle_multi(request_rec *r)
{
int res;
int j;
|| !(ap_allow_options(r) & OPT_MULTI)) {
return DECLINED;
}
neg = parse_accept_headers(r);
/* free all allocated memory from subrequests */
}
}
return res;
}
return DECLINED;
}
if (res != 0)
goto return_from_multi;
/* We got this out of a map file, so we don't actually have
* a sub_req structure yet. Get one now.
*/
goto return_from_multi;
}
}
/* now do a "fast redirect" ... promotes the sub_req into the main req */
/* give no advise for time on this subrequest. Perhaps we
* should tally the last mtime amoung all variants, and date
* the most recent, but that could confuse the proxies.
*/
r->mtime = 0;
/* clean up all but our favorite variant, since that sub_req
* is now merged into the main request!
*/
}
}
return OK;
}
/**********************************************************************
* There is a problem with content-encoding, as some clients send and
* expect an x- token (e.g. x-gzip) while others expect the plain token
* (i.e. gzip). To try and deal with this as best as possible we do
* the following: if the client sent an Accept-Encoding header and it
* contains a plain token corresponding to the content encoding of the
* response, then set content encoding using the plain token. Else if
* the A-E header contains the x- token use the x- token in the C-E
* header. Else don't do anything.
*
* Note that if no A-E header was sent, or it does not contain a token
* compatible with the final content encoding, then the token in the
* C-E header will be whatever was specified in the AddEncoding
* directive.
*/
static int fix_encoding(request_rec *r)
{
const char *enc = r->content_encoding;
int i;
return DECLINED;
}
enc += 2;
}
return DECLINED;
}
for (i = 0; i < accept_encodings->nelts; ++i) {
r->content_encoding = name;
return OK;
}
}
}
if (x_enc) {
r->content_encoding = x_enc;
return OK;
}
return DECLINED;
}
static void register_hooks(apr_pool_t *p)
{
}
{
create_neg_dir_config, /* dir config creator */
merge_neg_dir_configs, /* dir merger --- default is to override */
NULL, /* server config */
NULL, /* merge server config */
negotiation_cmds, /* command apr_table_t */
register_hooks /* register hooks */
};