docs/manual/developer/new_api_2_4.xml

<?xml version='1.0' encoding='UTF-8' ?>
<!DOCTYPE manualpage SYSTEM "../style/manualpage.dtd">
<?xml-stylesheet type="text/xsl" href="../style/manual.en.xsl"?>
<!-- $LastChangedRevision$ -->

<!--
 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.
-->

<manualpage metafile="new_api_2_4.xml.meta">

<title>API Changes in Apache HTTP Server 2.4 since 2.2</title>

<summary>
  <p>This document describes changes to the Apache HTTPD API from
     version 2.2 to 2.4, that may be of interest to module/application
     developers and core hacks.  As of the first GA release of the
     2.4 branch API compatibility is preserved for the life of the
     2.4 branch.  (The
     <a href="http://svn.apache.org/repos/asf/httpd/httpd/branches/2.4.x/VERSIONING">VERSIONING</a>
     description for the 2.4 release provides more information about API
     compatibility.)</p>

  <p>API changes fall into two categories: APIs that are altogether new,
     and existing APIs that are expanded or changed.  The latter are
     further divided into those where all changes are backwards-compatible
     (so existing modules can ignore them), and those that might
     require attention by maintainers.  As with the transition from
     HTTPD 2.0 to 2.2, existing modules and applications will require
     recompiling and may call for some attention, but most should not
     require any substantial updating (although some may be able to
     take advantage of API changes to offer significant improvements).</p>
  <p>For the purpose of this document, the API is split according
     to the public header files.  These headers are themselves the
     reference documentation, and can be used to generate a browsable
     HTML reference with <code>make docs</code>.</p>
</summary>

<section id="api_changes">
  <title>Changed APIs</title>

  <section id="ap_expr">
    <title>ap_expr (NEW!)</title>
    <p>Introduces a new API to parse and evaluate boolean and algebraic
       expressions, including provision for a standard syntax and
       customized variants.</p>
  </section>

  <section id="ap_listen">
    <title>ap_listen (changed; backwards-compatible)</title>
    <p>Introduces a new API to enable httpd child processes to serve
       different purposes.</p>
  </section>

  <section id="ap_mpm">
    <title>ap_mpm (changed)</title>
  <p><code>ap_mpm_run</code> is replaced by a new <code>mpm</code> hook.
  Also <code>ap_graceful_stop_signalled</code> is lost, and
  <code>ap_mpm_register_timed_callback</code> is new.</p>
  </section>

  <section id="ap_regex">
    <title>ap_regex (changed)</title>
  <p>In addition to the existing regexp wrapper, a new higher-level API
  <code>ap_rxplus</code> is now provided.  This provides the capability to
  compile Perl-style expressions like <code>s/regexp/replacement/flags</code>
  and to execute them against arbitrary strings. Support for regexp
  backreferences is also added.</p>
  </section>

  <section id="ap_slotmem">
    <title>ap_slotmem (NEW!)</title>
    <p>Introduces an API for modules to allocate and manage memory slots,
    most commonly for shared memory.</p>
  </section>

  <section id="ap_socache">
    <title>ap_socache (NEW!)</title>
    <p>API to manage a shared object cache.</p>
  </section>

  <section id="heartbeat">
    <title>heartbeat (NEW!)</title>
    <p>common structures for heartbeat modules</p>
  </section>

  <section id="ap_parse_htaccess">
    <title>ap_parse_htaccess (changed)</title>
    <p>The function signature for <code>ap_parse_htaccess</code> has been
    changed. A <code>apr_table_t</code> of individual directives allowed
    for override must now be passed (override remains).</p>
  </section>

  <section id="http_config">
    <title>http_config (changed)</title>
    <ul>
      <li>Introduces per-module, per-directory loglevels, including macro wrappers.</li>
      <li>New <code>AP_DECLARE_MODULE</code> macro to declare all modules.</li>
      <li>New <code>APLOG_USE_MODULE</code> macro necessary for per-module loglevels in
          multi-file modules.</li>
      <li>New API to retain data across module unload/load</li>
      <li>New <code>check_config</code> hook</li>
      <li>New <code>ap_process_fnmatch_configs()</code> function to process wildcards</li>
      <li>Change <code>ap_configfile_t</code>, <code>ap_cfg_getline()</code>,
          <code>ap_cfg_getc()</code> to return error codes, and add
          <code>ap_pcfg_strerror()</code> for retrieving an error description.</li>
      <li>Any config directive permitted in ACCESS_CONF context must now
          correctly handle being called from an .htaccess file via the new
          <directive module="core">AllowOverrideList</directive> directive.
          ap_check_cmd_context() accepts a new flag NOT_IN_HTACCESS to detect
          this case.</li>
    </ul>
  </section>

  <section id="http_core">
    <title>http_core (changed)</title>
    <ul>
      <li>REMOVED <code>ap_default_type</code>, <code>ap_requires</code>, all
          2.2 authnz API</li>
      <li>Introduces Optional Functions for logio and authnz</li>
      <li>New function <code>ap_get_server_name_for_url</code> to support IPv6
          literals.</li>
      <li>New function <code>ap_register_errorlog_handler</code> to register error log
          format string handlers.</li>
      <li>Arguments of <code>error_log</code> hook have changed. Declaration has moved to
          <code>http_core.h</code>.</li>
      <li>New function <code>ap_state_query</code> to determine if the server is in the
          initial configuration preflight phase or not. This is both easier to
          use and more correct than the old method of creating a pool userdata
          entry in the process pool.</li>
      <li>New function <code>ap_get_conn_socket</code> to get the socket descriptor for a
          connection. This should be used instead of accessing the core
          connection config directly.</li>
    </ul>
  </section>

  <section id="httpd">
    <title>httpd (changed)</title>
    <ul>
      <li>Introduce per-directory, per-module loglevel</li>
      <li>New loglevels <code>APLOG_TRACEn</code></li>
      <li>Introduce errorlog ids for requests and connections</li>
      <li>Support for mod_request kept_body</li>
      <li>Support buffering filter data for async requests</li>
      <li>New <code>CONN_STATE</code> values</li>
      <li>Function changes: <code>ap_escape_html</code> updated;
          <code>ap_unescape_all</code>, <code>ap_escape_path_segment_buffer</code></li>
      <li>Modules that load other modules later than the <code>EXEC_ON_READ</code> config
          reading stage need to call <code>ap_reserve_module_slots()</code> or
          <code>ap_reserve_module_slots_directive()</code> in their
          <code>pre_config hook</code>.</li>
      <li>The useragent IP address per request can now be tracked
          independently of the client IP address of the connection, for
          support of deployments with load balancers.</li>
    </ul>
  </section>

  <section id="http_log">
    <title>http_log (changed)</title>
    <ul>
      <li>Introduce per-directory, per-module loglevel</li>
      <li>New loglevels <code>APLOG_TRACEn</code></li>
      <li><code>ap_log_*error</code> become macro wrappers (backwards-compatible if
          <code>APLOG_MARK</code> macro is used, except that is no longer possible to
          use <code>#ifdef</code> inside the argument list)</li>
      <li>piped logging revamped</li>
      <li><code>module_index</code> added to error_log hook</li>
      <li>new function: <code>ap_log_command_line</code></li>
    </ul>
  </section>

  <section id="http_request">
    <title>http_request (changed)</title>
    <ul>
      <li>New auth_internal API and auth_provider API</li>
      <li>New <code>EOR</code> bucket type</li>
      <li>New function <code>ap_process_async_request</code></li>
      <li>New flags <code>AP_AUTH_INTERNAL_PER_CONF</code> and
          <code>AP_AUTH_INTERNAL_PER_URI</code></li>
      <li>New <code>access_checker_ex</code> hook to apply additional access control
          and/or bypass authentication.</li>
      <li>New functions <code>ap_hook_check_access_ex</code>,
          <code>ap_hook_check_access</code>, <code>ap_hook_check_authn</code>,
          <code>ap_hook_check_authz</code> which accept
          <code>AP_AUTH_INTERNAL_PER_*</code> flags</li>
      <li>DEPRECATED direct use of <code>ap_hook_access_checker</code>,
          <code>access_checker_ex</code>, <code>ap_hook_check_user_id</code>,
          <code>ap_hook_auth_checker</code></li>
    </ul>
    <p>When possible, registering all access control hooks (including
       authentication and authorization hooks) using <code>AP_AUTH_INTERNAL_PER_CONF</code>
       is recommended.  If all modules' access control hooks are registered
       with this flag, then whenever the server handles an internal
       sub-request that matches the same set of access control configuration
       directives as the initial request (which is the common case), it can
       avoid invoking the access control hooks another time.</p>
    <p>If your module requires the old behavior and must perform access
       control checks on every sub-request with a different URI from the
       initial request, even if that URI matches the same set of access
       control configuration directives, then use
       <code>AP_AUTH_INTERNAL_PER_URI</code>.</p>
  </section>

  <section id="mod_auth">
    <title>mod_auth (NEW!)</title>
    <p>Introduces the new provider framework for authn and authz</p>
  </section>

  <section id="mod_cache">
    <title>mod_cache (changed)</title>
    <p>Introduces a <code>commit_entity()</code> function to the cache provider
    interface, allowing atomic writes to cache. Add a <code>cache_status()</code>
    hook to report the cache decision. All private structures and functions were
    removed.</p>
  </section>

  <section id="mod_core">
    <title>mod_core (NEW!)</title>
    <p>This introduces low-level APIs to send arbitrary headers,
    and exposes functions to handle HTTP OPTIONS and TRACE.</p>
  </section>

  <section id="mod_cache_disk">
    <title>mod_cache_disk (changed)</title>
    <p>Changes the disk format of the disk cache to support atomic cache
    updates without locking. The device/inode pair of the body file is
    embedded in the header file, allowing confirmation that the header
    and body belong to one another.</p>
  </section>

  <section id="mod_disk_cache">
    <title>mod_disk_cache (renamed)</title>
    <p>The mod_disk_cache module has been renamed to mod_cache_disk in
    order to be consistent with the naming of other modules within the
    server.</p>
  </section>

  <section id="mod_request">
    <title>mod_request (NEW!)</title>
    <p>The API for <module>mod_request</module>, to make input data
    available to multiple application/handler modules where required,
    and to parse HTML form data.</p>
  </section>

  <section id="mpm_common">
    <title>mpm_common (changed)</title>
    <ul>
      <li>REMOVES: <code>accept</code>, <code>lockfile</code>, <code>lock_mech</code>,
          <code>set_scoreboard</code> (locking uses the new ap_mutex API)</li>
      <li>NEW API to drop privileges (delegates this platform-dependent
          function to modules)</li>
      <li>NEW Hooks: <code>mpm_query</code>, <code>timed_callback</code>, and
          <code>get_name</code></li>
      <li>CHANGED interfaces: <code>monitor</code> hook,
          <code>ap_reclaim_child_processes</code>,
          <code>ap_relieve_child_processes</code></li>
    </ul>
  </section>

  <section id="scoreboard">
    <title>scoreboard (changed)</title>
    <p><code>ap_get_scoreboard_worker</code> is made non-backwards-compatible
    as an alternative version is introduced.  Additional proxy_balancer
    support.  Child status stuff revamped.</p>
  </section>

  <section id="util_cookies">
    <title>util_cookies (NEW!)</title>
    <p>Introduces a new API for managing HTTP Cookies.</p>
  </section>

  <section id="util_ldap">
    <title>util_ldap (changed)</title>
    <p><em>no description available</em></p>
  </section>

  <section id="util_mutex">
    <title>util_mutex (NEW!)</title>
    <p>A wrapper for APR proc and global mutexes in httpd, providing
       common configuration for the underlying mechanism and location
       of lock files.</p>
  </section>

  <section id="util_script">
    <title>util_script (changed)</title>
    <p>NEW: <code>ap_args_to_table</code></p>
  </section>

  <section id="util_time">
    <title>util_time (changed)</title>
    <p>NEW: <code>ap_recent_ctime_ex</code></p>
  </section>

</section>

<section id="upgrading">
  <title>Specific information on upgrading modules from 2.2</title>

  <section id="upgrading_logging">
    <title>Logging</title>
    <p>In order to take advantage of per-module loglevel configuration, any
       source file that calls the <code>ap_log_*</code> functions should declare
       which module it belongs to. If the module's module_struct is called
       <code>foo_module</code>, the following code can be used to remain
       backward compatible with HTTPD 2.0 and 2.2:</p>
    <example>
        #include &lt;http_log.h&gt;<br/>
        <br/>
        #ifdef APLOG_USE_MODULE<br/>
        APLOG_USE_MODULE(foo);<br/>
        #endif
    </example>
    <p>Note: This is absolutely required for C++-language modules.  It
    can be skipped for C-language modules, though that breaks
    module-specific log level support for files without it.</p>
    <p>The number of parameters of the <code>ap_log_*</code> functions and the
       definition of <code>APLOG_MARK</code> has changed. Normally, the change
       is completely transparent. However, changes are required if a
       module uses <code>APLOG_MARK</code> as a parameter to its own functions
       or if a module calls <code>ap_log_*</code> without passing
       <code>APLOG_MARK</code>.  A module which uses wrappers
       around <code>ap_log_*</code> typically uses both of these constructs.</p>

    <p>The easiest way to change code which passes <code>APLOG_MARK</code> to
       its own functions is to define and use a different macro that expands to
       the parameters required by those functions, as <code>APLOG_MARK</code>
       should only be used when calling <code>ap_log_*</code>
       directly.  In this way, the code will remain compatible with HTTPD 2.0
       and 2.2.</p>

    <p>Code which calls <code>ap_log_*</code> without passing
       <code>APLOG_MARK</code> will necessarily differ between 2.4 and earlier
       releases, as 2.4 requires a new third argument,
       <code>APLOG_MODULE_INDEX</code>.</p>

    <example>
       /* code for httpd 2.0/2.2 */<br />
       ap_log_perror(file, line, APLOG_ERR, 0, p, "Failed to allocate dynamic lock structure");<br />
       <br />
       /* code for httpd 2.4 */<br />
       ap_log_perror(file, line, APLOG_MODULE_INDEX, APLOG_ERR, 0, p, "Failed to allocate dynamic lock structure");<br />
       <br />
    </example>

    <p><code>ap_log_*error</code> are now implemented as macros. This means
       that it is no longer possible to use <code>#ifdef</code> inside the
       argument list of <code>ap_log_*error</code>, as this would cause
       undefined behavior according to C99.</p>

    <p>A <code>server_rec</code> pointer must be passed to
       <code>ap_log_error()</code> when called after startup.  This
       was always appropriate, but there are even more limitations with
       a <code>NULL</code> <code>server_rec</code> in 2.4 than in
       previous releases.  Beginning with 2.3.12, the global variable
       <code>ap_server_conf</code> can always be used as
       the <code>server_rec</code> parameter, as it will be
       <code>NULL</code> only when it is valid to pass <code>NULL</code>
       to <code>ap_log_error()</code>.  <code>ap_server_conf</code>
       should be used only when a more appropriate <code>server_rec</code>
       is not available.</p>

    <p>Consider the following changes to take advantage of the new
       <code>APLOG_TRACE1..8</code> log levels:</p>
       <ul>
         <li>Check current use of <code>APLOG_DEBUG</code> and
         consider if one of the <code>APLOG_TRACEn</code> levels is
         more appropriate.</li>
         <li>If your module currently has a mechanism for configuring
         the amount of debug logging which is performed, consider
         eliminating that mechanism and relying on the use of
         different <code>APLOG_TRACEn</code> levels.  If expensive
         trace processing needs to be bypassed depending on the
         configured log level, use the <code>APLOGtrace<em>n</em></code>
         and <code>APLOGrtrace<em>n</em></code> macros to first check
         if tracing is enabled.</li>
       </ul>

    <p>Modules sometimes add process id and/or thread id to their log
       messages.  These ids are now logged by default, so it may not
       be necessary for the module to log them explicitly.  (Users may
       remove them from the error log format, but they can be
       instructed to add it back if necessary for problem diagnosis.)</p>
  </section>

  <section id="upgrading_byfunction">
    <title>If your module uses these existing APIs...</title>

    <dl>
      <dt><code>ap_default_type()</code></dt>
      <dd>This is no longer available; Content-Type must be configured
          explicitly or added by the application.</dd>

      <dt><code>ap_get_server_name()</code></dt>
      <dd>If the returned server name is used in a URL,
      use <code>ap_get_server_name_for_url()</code> instead.  This new
      function handles the odd case where the server name is an IPv6
      literal address.</dd>

      <dt><code>ap_get_server_version()</code></dt>
      <dd>For logging purposes, where detailed information is
          appropriate, use <code>ap_get_server_description()</code>.
          When generating output, where the amount of information
          should be configurable by ServerTokens, use
          <code>ap_get_server_banner()</code>.</dd>

      <dt><code>ap_graceful_stop_signalled()</code></dt>
      <dd>Replace with a call
      to <code>ap_mpm_query(AP_MPMQ_MPM_STATE)</code> and checking for
      state <code>AP_MPMQ_STOPPING</code>.</dd>

      <dt><code>ap_max_daemons_limit</code>, <code>ap_my_generation</code>,
          and <code>ap_threads_per_child</code></dt>
      <dd>Use <code>ap_mpm_query()</code> query codes
          <code>AP_MPMQ_MAX_DAEMON_USED</code>, <code>AP_MPMQ_GENERATION</code>,
          and <code>AP_MPMQ_MAX_THREADS</code>, respectively.</dd>

      <dt><code>ap_mpm_query()</code></dt>
      <dd>Ensure that it is not used until after the register-hooks
          hook has completed.  Otherwise, an MPM built as a DSO
          would not have had a chance to enable support for this
          function.</dd>

      <dt><code>ap_requires()</code></dt>
      <dd>The core server now provides better infrastructure for handling
          <directive module="mod_authz_core">Require</directive> configuration.
          Register an auth provider function for each supported entity using
          <code>ap_register_auth_provider()</code>.  The function will be
          called as necessary during <directive>Require</directive>
          processing.  (Consult bundled modules for detailed examples.)</dd>

      <dt><code>ap_server_conf->process->pool</code>
      userdata</dt>
      <dd>
        Optional:
        <ul>
          <li>If your module uses this to determine which pass of the
	  startup hooks is being run,
	  use <code>ap_state_query(AP_SQ_MAIN_STATE)</code>.</li>
          <li>If your module uses this to maintain data across the
          unloading and reloading of your module, use
          <code>ap_retained_data_create()</code> and
          <code>ap_retained_data_get()</code>.</li>
        </ul>
      </dd>

      <dt><code>apr_global_mutex_create()</code>,
          <code>apr_proc_mutex_create()</code></dt>
      <dd>Optional: See <code>ap_mutex_register()</code>,
          <code>ap_global_mutex_create()</code>, and
          <code>ap_proc_mutex_create()</code>; these allow your
          mutexes to be configurable with
          the <directive module="core">Mutex</directive> directive;
          you can also remove any configuration mechanisms in your
          module for such mutexes
      </dd>

      <dt><code>CORE_PRIVATE</code></dt>
      <dd>This is now unnecessary and ignored.</dd>

      <dt><code>dav_new_error()</code>
      and <code>dav_new_error_tag()</code></dt>
      <dd>Previously, these assumed that <code>errno</code> contained
      information describing the failure.  Now,
      an <code>apr_status_t</code> parameter must be provided.  Pass
      0/APR_SUCCESS if there is no such error information, or a valid
      <code>apr_status_t</code> value otherwise.</dd>

      <dt><code>mpm_default.h</code>, <code>DEFAULT_LOCKFILE</code>,
      <code>DEFAULT_THREAD_LIMIT</code>, <code>DEFAULT_PIDLOG</code>,
      etc.</dt>
      <dd>The header file and most of the default configuration
      values set in it are no longer visible to modules.  (Most can
      still be overridden at build time.)  <code>DEFAULT_PIDLOG</code>
      and <code>DEFAULT_REL_RUNTIMEDIR</code> are now universally
      available via <code>ap_config.h</code>.</dd>

      <dt><code>unixd_config</code></dt>
      <dd>This has been renamed to ap_unixd_config.</dd>

      <dt><code>conn_rec->remote_ip</code> and
          <code>conn_rec->remote_addr</code></dt>
      <dd>These fields have been renamed in order to distinguish between
      the client IP address of the connection and the useragent IP address
      of the request (potentially overridden by a load balancer or proxy).
      References to either of these fields must be updated with one of the
      following options, as appropriate for the module:
      <ul>
        <li>When you require the IP address of the user agent, which
        might be connected directly to the server, or might optionally be
        separated from the server by a transparent load balancer or
        proxy, use <code>request_rec->useragent_ip</code> and
        <code>request_rec->useragent_addr</code>.</li>
        <li>When you require the IP address of the client that is
        connected directly to the server, which might be the useragent or
        might be the load balancer or proxy itself, use
        <code>conn_rec->client_ip</code> and
        <code>conn_rec->client_addr</code>.</li>
      </ul>
      </dd>
    </dl>
  </section>

  <section id="upgrading_byfeature">
    <title>If your module interfaces with this feature...</title>
    <dl>
      <dt>suEXEC</dt>
      <dd>Optional: If your module logs an error
          when <code>ap_unixd_config.suexec_enabled</code> is 0,
          also log the value of the new
          field <code>suexec_disabled_reason</code>, which contains an
          explanation of why it is not available.</dd>

      <dt>Extended status data in the scoreboard</dt>
      <dd>In previous releases, <code>ExtendedStatus</code> had to be
          set to <code>On</code>, which in turn required that
          mod_status was loaded.  In 2.4, just
          set <code>ap_extended_status</code> to <code>1</code> in a
          pre-config hook and the extended status data will be
          available.</dd>
    </dl>
  </section>

  <section id="upgrading_newfeatures">
    <title>Does your module...</title>
    <dl>
    <dt>Parse query args</dt>
    <dd>Consider if <code>ap_args_to_table()</code> would be
    helpful.</dd>

    <dt>Parse form data...</dt>
    <dd>Use <code>ap_parse_form_data()</code>.</dd>

    <dt>Check for request header fields <code>Content-Length</code>
    and <code>Transfer-Encoding</code> to see if a body was
    specified</dt>
    <dd>Use <code>ap_request_has_body()</code>.</dd>

    <dt>Implement cleanups which clear pointer variables</dt>
    <dd>Use <code>ap_pool_cleanup_set_null()</code>.</dd>

    <dt>Create run-time files such as shared memory files, pid files,
    etc.</dt>
    <dd>Use <code>ap_runtime_dir_relative()</code> so that the global
    configuration for the location of such files, either by the
    <code>DEFAULT_REL_RUNTIMEDIR</code> compile setting or the
    <directive module="core">DefaultRuntimeDir</directive> directive,
    will be respected.  <em>Apache httpd 2.4.2 and above.</em></dd>

    </dl>
  </section>

</section>

</manualpage>