Copyright (C) 2004  Internet Systems Consortium, Inc. ("ISC")
Copyright (C) 2000, 2001  Internet Software Consortium.
See COPYRIGHT in the source root or http://isc.org/copyright.html for terms.

Using the BIND 9 Simplified Database Interface

This document describes the care and feeding of the BIND 9 Simplified
Database Interface, which allows you to extend BIND 9 with new ways
of obtaining the data that is published as DNS zones.


The Original BIND 9 Database Interface

BIND 9 has a well-defined "back-end database interface" that makes it
possible to replace the component of the name server responsible for
the storage and retrieval of zone data, called the "database", on a
per-zone basis.  The default database is an in-memory, red-black-tree
data structure commonly referred to as "rbtdb", but it is possible to
write drivers to support any number of alternative database
technologies such as in-memory hash tables, application specific
persistent on-disk databases, object databases, or relational
databases.

The original BIND 9 database interface defined in <dns/db.h> is
designed to efficiently support the full set of database functionality
needed by a name server that implements the complete DNS protocols,
including features such as zone transfers, dynamic update, and DNSSEC.
Each of these aspects of name server operations places its own set of
demands on the data store, with the result that the database API is
quite complex and contains operations that are highly specific to the
DNS.  For example, data are stored in a binary format, the name space
is tree structured, and sets of data records are conceptually
associated with DNSSEC signature sets.  For these reasons, writing a
driver using this interface is a highly nontrivial undertaking.


The Simplified Database Interface

Many BIND users wish to provide access to various data sources through
the DNS, but are not necessarily interested in completely replacing
the in-memory "rbt" database or in supporting features like dynamic
update, DNSSEC, or even zone transfers.

Often, all you want is limited, read-only DNS access to an existing
system.  For example, you may have an existing relational database
containing hostname/address mappings and wish to provide forvard and
reverse DNS lookups based on this information.  Or perhaps you want to
set up a simple DNS-based load balancing system where the name server
answers queries about a single DNS name with a dynamically changing
set of A records.

BIND 9.1 introduced a new, simplified database interface, or "sdb",
which greatly simplifies the writing of drivers for these kinds of
applications.


The sdb Driver

An sdb driver is an object module, typically written in C, which is
linked into the name server and registers itself with the sdb
subsystem.  It provides a set of callback functions, which also serve
to advertise its capabilities.  When the name server receives DNS
queries, invokes the callback functions to obtain the data to respond
with.

Unlike the full database interface, the sdb interface represents all
domain names and resource records as ASCII text.


Writing an sdb Driver

When a driver is registered, it specifies its name, a list of callback
functions, and flags.

The flags specify whether the driver wants to use relative domain
names where possible.

The callback functions are as follows.  The only one that must be
defined is lookup().

  - create(zone, argc, argv, driverdata, dbdata)
	  Create a database object for "zone".

  - destroy(zone, driverdata, dbdata)
	  Destroy the database object for "zone".

  - lookup(zone, name, dbdata, lookup)
	  Return all the records at the domain name "name".

  - authority(zone, dbdata, lookup)
	  Return the SOA and NS records at the zone apex.

  - allnodes(zone, dbdata, allnodes)
	  Return all data in the zone, for zone transfers.

For more detail about these functions and their parameters, see
bind9/lib/dns/include/dns/sdb.h.  For example drivers, see
bind9/contrib/sdb.


Rebuilding the Server

The driver module and header file must be copied to (or linked into)
the bind9/bin/named and bind9/bin/named/include directories
respectively, and must be added to the DBDRIVER_OBJS and DBDRIVER_SRCS
lines in bin/named/Makefile.in (e.g. for the timedb sample sdb driver,
add timedb.c to DBDRIVER_SRCS and timedb.@O@ to DBDRIVER_OBJS).  If
the driver needs additional header files or libraries in nonstandard
places, the DBDRIVER_INCLUDES and DBDRIVER_LIBS lines should also be
updated.

Calls to dns_sdb_register() and dns_sdb_unregister() (or wrappers,
e.g. timedb_init() and timedb_clear() for the timedb sample sdb
driver) must be inserted into the server, in bind9/bin/named/main.c.
Registration should be in setup(), before the call to
ns_server_create().  Unregistration should be in cleanup(),
after the call to ns_server_destroy().  A #include should be added
corresponding to the driver header file.

You should try doing this with one or more of the sample drivers
before attempting to write a driver of your own.


Configuring the Server

To make a zone use a new database driver, specify a "database" option
in its "zone" statement in named.conf.  For example, if the driver
registers itself under the name "acmedb", you might say

   zone "foo.com" {
	   database "acmedb";
   };

You can pass arbitrary arguments to the create() function of the
driver by adding any number of whitespace-separated words after the
driver name:

   zone "foo.com" {
	   database "acmedb -mode sql -connect 10.0.0.1";
   };


Hints for Driver Writers

 - If a driver is generating data on the fly, it probably should
   not implement the allnodes() function, since a zone transfer
   will not be meaningful.  The allnodes() function is more relevant
   with data from a database.

 - The authority() function is necessary if and only if the lookup()
   function will not add SOA and NS records at the zone apex.  If
   SOA and NS records are provided by the lookup() function,
   the authority() function should be NULL.

 - When a driver is registered, an opaque object can be provided.  This
   object is passed into the database create() and destroy() functions.

 - When a database is created, an opaque object can be created that
   is associated with that database.  This object is passed into the
   lookup(), authority(), and allnodes() functions, and is
   destroyed by the destroy() function.


Future Directions

A future release may support dynamic loading of sdb drivers.


$Id: sdb,v 1.6 2004/03/05 05:04:54 marka Exp $