Initial design

doc/design/compression

@@ -0,0 +1,110 @@
+
+			Name Compression
+
+Overview.
+
+	BIND 4.x and BIND 8.x only had one methods of compression to deal
+	with 14 bit compression.  BIND 9 has 3 methods of compression
+	to deal with 14 bit, 16 bit and local compression.
+
+	In addition to this the allowed compression methods vary across
+	types and across client revisions thanks to EDNS.
+
+	To be able to compress a domain name you need to some or all of
+	the following pieces of information.
+
+	1. where the message starts.
+	2. where the current rdata starts in the message (local compression).
+	3. what the current owner name is (local compression).
+	4. existing global 14 bit compression targets.
+	5. existing global 16 bit compression targets.
+	6. existing local compression targets.
+	7. the current domain name.
+	8. what are allowable compression methods, these are not constant
+	   across a message.
+
+	BIND 4.x and BIND 8.x used a table of existing 14 bit compression
+	targets. 
+
+	The implicit assumption is that we will use compression whenever
+	possible and when ever there are multiple alternatives available
+	we will choose the one that minimises the size of the message.
+
+	We will need functions that determine the allowable compression
+	methods, find the "best" match among the available compression
+	targets, add new compression targets.
+
+	We need to be able to back out any changes made to the compression
+	targets if we are unable to add a complete RR (RRset?). This is
+	only a problem for the global compression targets.
+
+Implementation:
+
+	We will maintain two RBT, one for local compression targets and
+	one for global compression targets.  The data for these RBT will
+	be the offset values.  The local compression RBT only needs to
+	be maintained when local compression is possible. The global
+	compression RBT is maintained regardless.  Unless there is a
+	perfect match (or the name is ".") we will add the name to the
+	compression RBTs provide the offset would not be too large for
+	the valid compression methods of the RBT.  All nodes of the RBT
+	will have an offset excluding the root node.
+
+	The local compression RBT will be initalised with the owner name
+	and the start of the rdata will be recorded.
+
+	We will use deepest partial match to find the potential
+	compression targets.
+
+	We only need to maintain one global RBT as 16 bit compression
+	pointers are either valid or invalid for the whole message.
+
+Functions:
+	unsigned int
+	dns_compress_allowed(dns_rdatatype_t type, int edns,
+				  isc_boolean_t isowner);
+
+	Returns allowed compression methods based on type, edns, and whether
+	we are about to compress a owner name.
+
+	void
+	dns_compress_init(dns_compress_t *cctx, isc_boolean_t global16,
+			  isc_mem_t *mctx);
+
+	Initalises cctx to empty and sets whether 16 bit global
+	compression targets are to be added to the global RBT.
+
+	void
+	dns_compress_localinit(dns_compress_t *cctx, dns_name_t owner,
+			       isc_buffer_t *target);
+
+	Initalise a RBT for local compression, freeing and existing RBT.
+	Record current offset.
+
+	dns_compress_invalidate(dns_compress_t *cctx);
+
+	Free any RBT's and make empty.
+
+	void
+	dns_compress_setmethods(dns_compress_t *cctx, unsigned int allowed);
+
+	unsigned int
+	dns_compress_getmethods(dns_compress_t *cctx);
+
+	dns_result_t
+	dns_name_towire(dns_name_t *name, dns_compress_t *cctx,
+			isc_buffer_t *target);
+
+	'name' contains the current name to be added to the message 'target'.
+	'target' is assumed to only contain the message.
+	'cctx' contains the compression context and has to hold all the
+	information required that cannot be obtained from 'name' or 'target'.
+
+	struct dns_compress {
+		unsigned int allowed;	/* Allowed methods. */
+		unsigned int rdata;	/* Start of local rdata */
+		isc_boolean_t global16;	/* 16 bit offsets allowed */
+		dns_rbt_t *local;	/* Local RBT */
+		dns_rbt_t *global;	/* Global RBT */
+		isc_mem_t *mctx;	/* Required by RBT */
+	};