Core Hooks

The httpd's core modules offer a predefinined list of hooks used during the standard request processing phase. Creating a new hook will expose a function that implements it (see sections below) but it is essential to undestand that you will not extend the httpd's core hooks. Their presence and order in the request processing is in fact a consequence of how they are called in server/request.c (check this section for an overview). The core hooks are listed in the doxygen documentation.

Reading guide for developing modules and request processing before proceeding is highly recomended.

Creating a hook function

In order to create a new hook, four things need to be done:

Declare the hook function

Use the AP_DECLARE_HOOK macro, which needs to be given the return type of the hook function, the name of the hook, and the arguments. For example, if the hook returns an int and takes a request_rec * and an int and is called do_something, then declare it like this:

AP_DECLARE_HOOK(int, do_something, (request_rec *r, int n))

This should go in a header which modules will include if they want to use the hook.

Create the hook structure

Each source file that exports a hook has a private structure which is used to record the module functions that use the hook. This is declared as follows:

APR_HOOK_STRUCT(
  APR_HOOK_LINK(do_something)
  ...
)

Implement the hook caller

The source file that exports the hook has to implement a function that will call the hook. There are currently three possible ways to do this. In all cases, the calling function is called ap_run_hookname().

Void hooks

If the return value of a hook is void, then all the hooks are called, and the caller is implemented like this:

AP_IMPLEMENT_HOOK_VOID(do_something, (request_rec *r, int n), (r, n))

The second and third arguments are the dummy argument declaration and the dummy arguments as they will be used when calling the hook. In other words, this macro expands to something like this:

void ap_run_do_something(request_rec *r, int n)
{
    ...
    do_something(r, n);
}

Hooks that return a value

If the hook returns a value, then it can either be run until the first hook that does something interesting, like so:

AP_IMPLEMENT_HOOK_RUN_FIRST(int, do_something, (request_rec *r, int n), (r, n), DECLINED)

The first hook that does not return DECLINED stops the loop and its return value is returned from the hook caller. Note that DECLINED is the traditional hook return value meaning "I didn't do anything", but it can be whatever suits you.

Alternatively, all hooks can be run until an error occurs. This boils down to permitting two return values, one of which means "I did something, and it was OK" and the other meaning "I did nothing". The first function that returns a value other than one of those two stops the loop, and its return is the return value. Declare these like so:

AP_IMPLEMENT_HOOK_RUN_ALL(int, do_something, (request_rec *r, int n), (r, n), OK, DECLINED)

Again, OK and DECLINED are the traditional values. You can use what you want.

Call the hook callers

At appropriate moments in the code, call the hook caller, like so:

int n, ret;
request_rec *r;

ret=ap_run_do_something(r, n);

Hooking the hook

A module that wants a hook to be called needs to do two things.

Implement the hook function

Include the appropriate header, and define a static function of the correct type:

static int my_something_doer(request_rec *r, int n)
{
    ...
    return OK;
}

Add a hook registering function

During initialisation, the server will call each modules hook registering function, which is included in the module structure:

static void my_register_hooks()
{
    ap_hook_do_something(my_something_doer, NULL, NULL, APR_HOOK_MIDDLE);
}

mode MODULE_VAR_EXPORT my_module =
{
    ...
    my_register_hooks       /* register hooks */
};

Controlling hook calling order

In the example above, we didn't use the three arguments in the hook registration function that control calling order of all the functions registered within the hook. There are two mechanisms for doing this. The first, rather crude, method, allows us to specify roughly where the hook is run relative to other modules. The final argument control this. There are three possible values: APR_HOOK_FIRST, APR_HOOK_MIDDLE and APR_HOOK_LAST.

All modules using any particular value may be run in any order relative to each other, but, of course, all modules using APR_HOOK_FIRST will be run before APR_HOOK_MIDDLE which are before APR_HOOK_LAST. Modules that don't care when they are run should use APR_HOOK_MIDDLE. These values are spaced out, so that positions like APR_HOOK_FIRST-2 are possible to hook slightly earlier than other functions.

Note that there are two more values, APR_HOOK_REALLY_FIRST and APR_HOOK_REALLY_LAST. These should only be used by the hook exporter.

The other method allows finer control. When a module knows that it must be run before (or after) some other modules, it can specify them by name. The second (third) argument is a NULL-terminated array of strings consisting of the names of modules that must be run before (after) the current module. For example, suppose we want "mod_xyz.c" and "mod_abc.c" to run before we do, then we'd hook as follows:

static void register_hooks()
{
    static const char * const aszPre[] = { "mod_xyz.c", "mod_abc.c", NULL };

    ap_hook_do_something(my_something_doer, aszPre, NULL, APR_HOOK_MIDDLE);
}

Note that the sort used to achieve this is stable, so ordering set by APR_HOOK_ORDER is preserved, as far as is possible.