Warning - this is a first (fast) draft that needs further revision!
Several changes in Apache 2.0 affect the internal request processing mechanics. Module authors need to be aware of these changes so they may take advantage of the optimizations and security enhancements.
The first major change is to the subrequest and redirect
mechanisms. There were a number of different code paths in
Apache 1.3 to attempt to optimize subrequest or redirect
behavior. As patches were introduced to 2.0, these
optimizations (and the server behavior) were quickly broken due
to this duplication of code. All duplicate code has been folded
back into ap_process_internal_request()
to prevent
the code from falling out of sync again.
This means that much of the existing code was 'unoptimized'. It is the Apache HTTP Project's first goal to create a robust and correct implementation of the HTTP server RFC. Additional goals include security, scalability and optimization. New methods were sought to optimize the server (beyond the performance of Apache 1.3) without introducing fragile or insecure code.
All requests pass through
ap_process_request_internal()
in request.c,
including subrequests and redirects. If a module doesn't pass
generated requests through this code, the author is cautioned
that the module may be broken by future changes to request
processing.
To streamline requests, the module author can take advantage of the hooks offered to drop out of the request cycle early, or to bypass core Apache hooks which are irrelevant (and costly in terms of CPU.)
The request's parsed_uri path is unescaped, once and only once, at the beginning of internal request processing.
This step is bypassed if the proxyreq flag is set, or the parsed_uri.path element is unset. The module has no further control of this one-time unescape operation, either failing to unescape or multiply unescaping the URL leads to security reprecussions.
All /../
and /./
elements are
removed by ap_getparents()
. This helps to ensure
the path is (nearly) absolute before the request processing
continues.
This step cannot be bypassed.
Every request is subject to an
ap_location_walk()
call. This ensures that
<Location > sections are consistently enforced for all
requests. If the request is an internal redirect or a
sub-request, it may borrow some or all of the processing from
the previous or parent request's ap_location_walk, so this step
is generally very efficient after processing the main
request.
Modules can determine the file name, or alter the given URI in this step. For example, mod_vhost_alias will translate the URI's path into the configured virtual host, mod_alias will translate the path to an alias path, and if the request falls back on the core, the DocumentRoot is prepended to the request resource.
If all modules DECLINE this phase, an error 500 is returned to the browser, and a "couldn't translate name" error is logged automatically.
After the file or correct URI was determined, the appropriate per-dir configurations are merged together. For example, mod_proxy compares and merges the appropriate <Proxy > sections. If the URI is nothing more than a local (non-proxy) TRACE request, the core handles the request and returns DONE. If no module answers this hook with OK or DONE, the core will run the request filename against the <Directory > and <Files > sections. If the request 'filename' isn't an absolute, legal filename, a note is set for later termination.
Every request is hardened by a second
ap_location_walk()
call. This reassures that a
translated request is still subjected to the configured
<Location > sections. The request again borrows some or
all of the processing from its previous location_walk above,
so this step is almost always very efficient unless the
translated URI mapped to a substantially different path or
Virtual Host.
The main request then parses the client's headers. This prepares the remaining request processing steps to better serve the client's request.
Needs Documentation. Code is;
switch (ap_satisfies(r)) { case SATISFY_ALL: case SATISFY_NOSPEC: if ((access_status = ap_run_access_checker(r)) != 0) { return decl_die(access_status, "check access", r); } if (ap_some_auth_required(r)) { if (((access_status = ap_run_check_user_id(r)) != 0) || !ap_auth_type(r)) { return decl_die(access_status, ap_auth_type(r) ? "check user. No user file?" : "perform authentication. AuthType not set!", r); } if (((access_status = ap_run_auth_checker(r)) != 0) || !ap_auth_type(r)) { return decl_die(access_status, ap_auth_type(r) ? "check access. No groups file?" : "perform authentication. AuthType not set!", r); } } break; case SATISFY_ANY: if (((access_status = ap_run_access_checker(r)) != 0) || !ap_auth_type(r)) { if (!ap_some_auth_required(r)) { return decl_die(access_status, ap_auth_type(r) ? "check access" : "perform authentication. AuthType not set!", r); } if (((access_status = ap_run_check_user_id(r)) != 0) || !ap_auth_type(r)) { return decl_die(access_status, ap_auth_type(r) ? "check user. No user file?" : "perform authentication. AuthType not set!", r); } if (((access_status = ap_run_auth_checker(r)) != 0) || !ap_auth_type(r)) { return decl_die(access_status, ap_auth_type(r) ? "check access. No groups file?" : "perform authentication. AuthType not set!", r); } } break; }
The modules have an opportunity to test the URI or filename against the target resource, and set mime information for the request. Both mod_mime and mod_mime_magic use this phase to compare the file name or contents against the administrator's configuration and set the content type, language, character set and request handler. Some modules may set up their filters or other request handling parameters at this time.
If all modules DECLINE this phase, an error 500 is returned to the browser, and a "couldn't find types" error is logged automatically.
Many modules are 'trounced' by some phase above. The fixups phase is used by modules to 'reassert' their ownership or force the request's fields to their appropriate values. It isn't always the cleanest mechanism, but occasionally it's the only option.
This phase is not part of the
processing in ap_process_request_internal()
. Many
modules prepare one or more subrequests prior to creating any
content at all. After the core, or a module calls
ap_process_request_internal()
it then calls
ap_invoke_handler()
to generate the request.
Modules that transform the content in some way can insert their values and override existing filters, such that if the user configured a more advanced filter out-of-order, then the module can move its order as need be. There is no result code, so actions in this hook better be trusted to always succeed.
The module finally has a chance to serve the request in its handler hook. Note that not every prepared request is sent to the handler hook. Many modules, such as mod_autoindex, will create subrequests for a given URI, and then never serve the subrequest, but simply lists it for the user. Remember not to put required teardown from the hooks above into this module, but register pool cleanups against the request pool to free resources as required.