Version reporting & module ABI change.

[exim.git] / src / src / lookups / README

$Cambridge: exim/src/src/lookups/README,v 1.2 2005/08/01 13:20:28 ph10 Exp $

LOOKUPS
-------

Each lookup type is implemented by 6 functions, xxx_open(), xxx_check(),
xxx_find(), xxx_close(), xxx_tidy(), and xxx_quote(), where xxx is the name of
the lookup type (e.g. lsearch, dbm, or whatever). In addition, there is
a version reporting function used to trace compile-vs-runtime conflicts and
to help administrators ensure that the modules from the correct build are
in use by the main binary.

The xxx_check(), xxx_close(), xxx_tidy(), and xxx_quote() functions need not
exist. There is a table in drtables.c which links the lookup names to the
various sets of functions, with NULL entries for any that don't exist. When
the code for a lookup type is omitted from the binary, all its entries are
NULL.

One of the fields in the table contains flags describing the kind of lookup.
These are

  lookup_querystyle

This is a "query style" lookup without a file name, as opposed to the "single
key" style, where the key is associated with a "file name".

  lookup_absfile

For single key lookups, this means that the file name must be an absolute path.
It is set for lsearch and dbm, but not for NIS, for example.

  lookup_absfilequery

This is a query-style lookup that must start with an absolute file name. For
example, the sqlite lookup is of this type.

When a single-key or absfilequery lookup file is opened, the handle returned by
the xxx_open() function is saved, along with the file name and lookup type, in
a tree. The xxx_close() function is not called when the first lookup is
completed. If there are subsequent lookups of the same type that quote the same
file name, xxx_open() isn't called; instead the cached handle is re-used.

Exim calls the function search_tidyup() at strategic points in its processing
(e.g. after all routing and directing has been done) and this function walks
the tree and calls the xxx_close() functions for all the cached handles.

Query-style lookups don't have the concept of an open file that can be cached
this way. Of course, the local code for the lookup can manage its own caching
information in any way it pleases. This means that the xxx_close()
function, even it it exists, is never called. However, if an xxx_tidy()
function exists, it is called once whenever Exim calls search_tidyup().

A single-key lookup type may also have an xxx_tidy() function, which is called
by search_tidyup() after all cached handles have been closed via the
xxx_close() function.

The lookup functions are wrapped into a special store pool (POOL_SEARCH). You
can safely use store_get to allocate store for your handle caching. The store
will be reset after all xxx_tidy() functions are called.

The function interfaces are as follows:


xxx_open()
----------

This function is called to initiate the lookup. For things that involve files
it should do a real open; for other kinds of lookup it may do nothing at all.
The arguments are:

  uschar *filename    the name of the "file" to open, for non-query-style
                        lookups; NULL for query-style lookups
  uschar **errmsg     where to put an error message if there is a problem

The yield of xxx_open() is a void * value representing the open file or
database. For real files is is normally the FILE or DBM value. For other
kinds of lookup, if there is no natural value to use, (-1) is recommended.
The value should not be NULL (or 0) as that is taken to indicate failure of
the xxx_open() function. For single-key lookups, the handle is cached along
with the filename and type, and may be used for several lookups.


xxx_check()
-----------

If this function exists, it is called after a successful open to check on the
ownership and mode of the file(s). The arguments are:

  void *handle        the handle passed back from xxx_open()
  uschar *filename    the filename passed to xxx_open()
  int modemask        mode bits that must not be set
  int *owners         permitted owners of the file(s)
  int *owngroups      permitted group owners of the file(s)
  uschar **errmsg     where to put an error message if there is a problem

In the owners and owngroups vectors, the first element is the count of the
remaining elements. There is a common function that can be called to check
a file:

int search_check_file(int fd, char *filename, int modemask, int *owners,
  int *owngroups, char *type, char **errmsg);

If fd is >= 0, it is checked using fstat(), and filename is used only in
error messages. If fd is < 0 then filename is checked using stat(). The yield
is zero if all is well, +1 if the mode or uid or gid is wrong, or -1 if the
stat() fails.

The yield of xxx_check() is TRUE if all is well, FALSE otherwise. The
function should not close the file(s) on failure. That is done from outside
by calling the xxx_close() function.


xxx_find()
----------

This is called to search an open file/database. The result is OK, FAIL, or
DEFER. The arguments are:

  void *handle        the handle passed back from xxx_open()
  uschar *filename    the filename passed to xxx_open() (NULL for querystyle)
  uschar *keyquery    the key to look up, or query to process, zero-terminated
  int  length         the length of the key
  uschar **result     point to the yield, in dynamic store, on success
  uschar **errmsg     where to put an error message on failure;
                      this is initially set to "", and should be left
                      as that for a standard "entry not found" error
  BOOL *do_cache      the lookup should set this to FALSE when it changes data.
                      This is TRUE by default. When set to FALSE the cache tree
                      of the current search handle will be cleaned and the
                      current result will NOT be cached. Currently the mysql
                      and pgsql lookups use this when UPDATE/INSERT queries are
                      executed.

Even though the key is zero-terminated, the length is passed because in the
common case it has been computed already and is often needed.


xxx_close()
-----------

This is called for single-key lookups when a file is finished with. There is no
yield, and the only argument is the handle that was passed back from
xxx_open(). It is not called for query style lookups.


xxx_tidy()
----------

This function is called once at the end of search_tidyup() for every lookup
type for which it exists.


xxx_quote()
-----------

This is called by the string expansion code for expansions of the form
${quote_xxx:<string>}, if it exists. If not, the expansion code makes no change
to the string. The function must apply any quoting rules that are specific to
the lookup, and return a pointer to the revised string. If quoting is not
needed, it can return its single argument, which is a uschar *. This function
does NOT use the POOL_SEARCH store, because it's usually never called from any
lookup code.

xxx_report_version()
--------------------

This is called to report diagnostic information to a file stream.
Typically it would report both compile-time and run-time version information.
The arguments are:

  FILE *stream    where to fprintf() the data to


****