slcompress.c revision 15190
1/*
2 * Routines to compress and uncompess tcp packets (for transmission
3 * over low speed serial lines.
4 *
5 * Copyright (c) 1989 Regents of the University of California.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms are permitted
9 * provided that the above copyright notice and this paragraph are
10 * duplicated in all such forms and that any documentation,
11 * advertising materials, and other materials related to such
12 * distribution and use acknowledge that the software was developed
13 * by the University of California, Berkeley.  The name of the
14 * University may not be used to endorse or promote products derived
15 * from this software without specific prior written permission.
16 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
18 * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
19 *
20 * $Id: slcompress.c,v 1.4 1996/01/11 17:48:58 phk Exp $
21 *
22 *	Van Jacobson (van@helios.ee.lbl.gov), Dec 31, 1989:
23 *	- Initial distribution.
24 */
25#ifndef lint
26static char const rcsid[] = "$Id: slcompress.c,v 1.4 1996/01/11 17:48:58 phk Exp $";
27#endif
28
29#include "defs.h"
30#include <netinet/in_systm.h>
31#include <netinet/in.h>
32#include <netinet/tcp.h>
33#include <netinet/ip.h>
34#include "slcompress.h"
35
36struct slstat slstat;
37
38#define INCR(counter)	slstat.counter++;
39
40#define BCMP(p1, p2, n) bcmp((char *)(p1), (char *)(p2), (int)(n))
41#define BCOPY(p1, p2, n) bcopy((char *)(p1), (char *)(p2), (int)(n))
42#ifndef KERNEL
43#define ovbcopy bcopy
44#endif
45
46void
47sl_compress_init(comp)
48	struct slcompress *comp;
49{
50	register u_int i;
51	register struct cstate *tstate = comp->tstate;
52
53	bzero((char *)comp, sizeof(*comp));
54	for (i = MAX_STATES - 1; i > 0; --i) {
55		tstate[i].cs_id = i;
56		tstate[i].cs_next = &tstate[i - 1];
57	}
58	tstate[0].cs_next = &tstate[MAX_STATES - 1];
59	tstate[0].cs_id = 0;
60	comp->last_cs = &tstate[0];
61	comp->last_recv = 255;
62	comp->last_xmit = 255;
63	comp->flags = SLF_TOSS;
64}
65
66
67/* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
68 * checks for zero (since zero has to be encoded in the long, 3 byte
69 * form).
70 */
71#define ENCODE(n) { \
72	if ((u_short)(n) >= 256) { \
73		*cp++ = 0; \
74		cp[1] = (n); \
75		cp[0] = (n) >> 8; \
76		cp += 2; \
77	} else { \
78		*cp++ = (n); \
79	} \
80}
81#define ENCODEZ(n) { \
82	if ((u_short)(n) >= 256 || (u_short)(n) == 0) { \
83		*cp++ = 0; \
84		cp[1] = (n); \
85		cp[0] = (n) >> 8; \
86		cp += 2; \
87	} else { \
88		*cp++ = (n); \
89	} \
90}
91
92#define DECODEL(f) { \
93	if (*cp == 0) {\
94		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
95		cp += 3; \
96	} else { \
97		(f) = htonl(ntohl(f) + (u_long)*cp++); \
98	} \
99}
100
101#define DECODES(f) { \
102	if (*cp == 0) {\
103		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
104		cp += 3; \
105	} else { \
106		(f) = htons(ntohs(f) + (u_long)*cp++); \
107	} \
108}
109
110#define DECODEU(f) { \
111	if (*cp == 0) {\
112		(f) = htons((cp[1] << 8) | cp[2]); \
113		cp += 3; \
114	} else { \
115		(f) = htons((u_long)*cp++); \
116	} \
117}
118
119
120u_char
121sl_compress_tcp(m, ip, comp, compress_cid)
122	struct mbuf *m;
123	register struct ip *ip;
124	struct slcompress *comp;
125	int compress_cid;
126{
127	register struct cstate *cs = comp->last_cs->cs_next;
128	register u_int hlen = ip->ip_hl;
129	register struct tcphdr *oth;
130	register struct tcphdr *th;
131	register u_int deltaS, deltaA;
132	register u_int changes = 0;
133	u_char new_seq[16];
134	register u_char *cp = new_seq;
135
136	/*
137	 * Bail if this is an IP fragment or if the TCP packet isn't
138	 * `compressible' (i.e., ACK isn't set or some other control bit is
139	 * set).  (We assume that the caller has already made sure the
140	 * packet is IP proto TCP).
141	 */
142#ifdef DEBUG
143	if ((ip->ip_off & htons(0x3fff)) || m->cnt < 40) {
144		logprintf("??? 1 ip_off = %x, cnt = %d\n", ip->ip_off, m->cnt);
145		DumpBp(m);
146		return (TYPE_IP);
147	}
148#else
149	if ((ip->ip_off & htons(0x3fff)) || m->cnt < 40)
150		return (TYPE_IP);
151#endif
152
153	th = (struct tcphdr *)&((int *)ip)[hlen];
154#ifdef DEBUG
155	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK) {
156		logprintf("??? 2 th_flags = %x\n", th->th_flags);
157		DumpBp(m);
158		return (TYPE_IP);
159	}
160#else
161	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
162		return (TYPE_IP);
163#endif
164
165	/*
166	 * Packet is compressible -- we're going to send either a
167	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
168	 * to locate (or create) the connection state.  Special case the
169	 * most recently used connection since it's most likely to be used
170	 * again & we don't have to do any reordering if it's used.
171	 */
172	INCR(sls_packets)
173	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
174	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
175	    *(int *)th != ((int *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
176		/*
177		 * Wasn't the first -- search for it.
178		 *
179		 * States are kept in a circularly linked list with
180		 * last_cs pointing to the end of the list.  The
181		 * list is kept in lru order by moving a state to the
182		 * head of the list whenever it is referenced.  Since
183		 * the list is short and, empirically, the connection
184		 * we want is almost always near the front, we locate
185		 * states via linear search.  If we don't find a state
186		 * for the datagram, the oldest state is (re-)used.
187		 */
188		register struct cstate *lcs;
189		register struct cstate *lastcs = comp->last_cs;
190
191		do {
192			lcs = cs; cs = cs->cs_next;
193			INCR(sls_searches)
194			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
195			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
196			    && *(int *)th == ((int *)&cs->cs_ip)[cs->cs_ip.ip_hl])
197				goto found;
198		} while (cs != lastcs);
199
200		/*
201		 * Didn't find it -- re-use oldest cstate.  Send an
202		 * uncompressed packet that tells the other side what
203		 * connection number we're using for this conversation.
204		 * Note that since the state list is circular, the oldest
205		 * state points to the newest and we only need to set
206		 * last_cs to update the lru linkage.
207		 */
208		INCR(sls_misses)
209		comp->last_cs = lcs;
210#define	THOFFSET(th)	(th->th_off)
211		hlen += th->th_off;
212		hlen <<= 2;
213		if (hlen > m->cnt)
214			return(TYPE_IP);
215		goto uncompressed;
216
217	found:
218		/*
219		 * Found it -- move to the front on the connection list.
220		 */
221		if (cs == lastcs)
222			comp->last_cs = lcs;
223		else {
224			lcs->cs_next = cs->cs_next;
225			cs->cs_next = lastcs->cs_next;
226			lastcs->cs_next = cs;
227		}
228	}
229
230	/*
231	 * Make sure that only what we expect to change changed. The first
232	 * line of the `if' checks the IP protocol version, header length &
233	 * type of service.  The 2nd line checks the "Don't fragment" bit.
234	 * The 3rd line checks the time-to-live and protocol (the protocol
235	 * check is unnecessary but costless).  The 4th line checks the TCP
236	 * header length.  The 5th line checks IP options, if any.  The 6th
237	 * line checks TCP options, if any.  If any of these things are
238	 * different between the previous & current datagram, we send the
239	 * current datagram `uncompressed'.
240	 */
241	oth = (struct tcphdr *)&((int *)&cs->cs_ip)[hlen];
242	deltaS = hlen;
243	hlen += th->th_off;
244	hlen <<= 2;
245	if (hlen > m->cnt)
246		return(TYPE_IP);
247
248	if (((u_short *)ip)[0] != ((u_short *)&cs->cs_ip)[0] ||
249	    ((u_short *)ip)[3] != ((u_short *)&cs->cs_ip)[3] ||
250	    ((u_short *)ip)[4] != ((u_short *)&cs->cs_ip)[4] ||
251	    THOFFSET(th) != THOFFSET(oth) ||
252	    (deltaS > 5 &&
253	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
254	    (THOFFSET(th) > 5 &&
255	     BCMP(th + 1, oth + 1, (THOFFSET(th) - 5) << 2))) {
256		goto uncompressed;
257	}
258
259	/*
260	 * Figure out which of the changing fields changed.  The
261	 * receiver expects changes in the order: urgent, window,
262	 * ack, seq (the order minimizes the number of temporaries
263	 * needed in this section of code).
264	 */
265	if (th->th_flags & TH_URG) {
266		deltaS = ntohs(th->th_urp);
267		ENCODEZ(deltaS);
268		changes |= NEW_U;
269	} else if (th->th_urp != oth->th_urp) {
270		/* argh! URG not set but urp changed -- a sensible
271		 * implementation should never do this but RFC793
272		 * doesn't prohibit the change so we have to deal
273		 * with it. */
274		 goto uncompressed;
275	}
276
277	deltaS = (u_short)(ntohs(th->th_win) - ntohs(oth->th_win));
278	if (deltaS) {
279		ENCODE(deltaS);
280		changes |= NEW_W;
281	}
282
283	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
284	if (deltaA) {
285		if (deltaA > 0xffff) {
286			goto uncompressed;
287		}
288		ENCODE(deltaA);
289		changes |= NEW_A;
290	}
291
292	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
293	if (deltaS) {
294		if (deltaS > 0xffff) {
295			goto uncompressed;
296		}
297		ENCODE(deltaS);
298		changes |= NEW_S;
299	}
300
301	switch(changes) {
302
303	case 0:
304		/*
305		 * Nothing changed. If this packet contains data and the
306		 * last one didn't, this is probably a data packet following
307		 * an ack (normal on an interactive connection) and we send
308		 * it compressed.  Otherwise it's probably a retransmit,
309		 * retransmitted ack or window probe.  Send it uncompressed
310		 * in case the other side missed the compressed version.
311		 */
312		if (ip->ip_len != cs->cs_ip.ip_len &&
313		    ntohs(cs->cs_ip.ip_len) == hlen)
314			break;
315
316		/* (fall through) */
317
318	case SPECIAL_I:
319	case SPECIAL_D:
320		/*
321		 * actual changes match one of our special case encodings --
322		 * send packet uncompressed.
323		 */
324		goto uncompressed;
325
326	case NEW_S|NEW_A:
327		if (deltaS == deltaA &&
328		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
329			/* special case for echoed terminal traffic */
330			changes = SPECIAL_I;
331			cp = new_seq;
332		}
333		break;
334
335	case NEW_S:
336		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
337			/* special case for data xfer */
338			changes = SPECIAL_D;
339			cp = new_seq;
340		}
341		break;
342	}
343
344	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
345	if (deltaS != 1) {
346		ENCODEZ(deltaS);
347		changes |= NEW_I;
348	}
349	if (th->th_flags & TH_PUSH)
350		changes |= TCP_PUSH_BIT;
351	/*
352	 * Grab the cksum before we overwrite it below.  Then update our
353	 * state with this packet's header.
354	 */
355	deltaA = ntohs(th->th_sum);
356	BCOPY(ip, &cs->cs_ip, hlen);
357
358	/*
359	 * We want to use the original packet as our compressed packet.
360	 * (cp - new_seq) is the number of bytes we need for compressed
361	 * sequence numbers.  In addition we need one byte for the change
362	 * mask, one for the connection id and two for the tcp checksum.
363	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
364	 * many bytes of the original packet to toss so subtract the two to
365	 * get the new packet size.
366	 */
367	deltaS = cp - new_seq;
368	cp = (u_char *)ip;
369
370        /*
371         * Since fastq traffic can jump ahead of the background traffic,
372         * we don't know what order packets will go on the line.  In this
373         * case, we always send a "new" connection id so the receiver state
374         * stays synchronized.
375         */
376#ifdef SL_NOFASTQ
377        if (comp->last_xmit == cs->cs_id) {
378                hlen -= deltaS + 3;
379                cp += hlen;
380                *cp++ = changes;
381        } else
382#endif
383	{
384		comp->last_xmit = cs->cs_id;
385		hlen -= deltaS + 4;
386		cp += hlen;
387		*cp++ = changes | NEW_C;
388		*cp++ = cs->cs_id;
389	}
390	m->cnt -= hlen;
391	m->offset += hlen;
392	*cp++ = deltaA >> 8;
393	*cp++ = deltaA;
394	BCOPY(new_seq, cp, deltaS);
395	INCR(sls_compressed)
396	return (TYPE_COMPRESSED_TCP);
397
398	/*
399	 * Update connection state cs & send uncompressed packet ('uncompressed'
400	 * means a regular ip/tcp packet but with the 'conversation id' we hope
401	 * to use on future compressed packets in the protocol field).
402	 */
403uncompressed:
404	BCOPY(ip, &cs->cs_ip, hlen);
405	ip->ip_p = cs->cs_id;
406	comp->last_xmit = cs->cs_id;
407	return (TYPE_UNCOMPRESSED_TCP);
408}
409
410
411int
412sl_uncompress_tcp(bufp, len, type, comp)
413	u_char **bufp;
414	int len;
415	u_int type;
416	struct slcompress *comp;
417{
418	register u_char *cp;
419	register u_int hlen, changes;
420	register struct tcphdr *th;
421	register struct cstate *cs;
422	register struct ip *ip;
423
424	switch (type) {
425
426	case TYPE_UNCOMPRESSED_TCP:
427		ip = (struct ip *) *bufp;
428		if (ip->ip_p >= MAX_STATES)
429			goto bad;
430		cs = &comp->rstate[comp->last_recv = ip->ip_p];
431		comp->flags &=~ SLF_TOSS;
432		ip->ip_p = IPPROTO_TCP;
433		/*
434		 * Calculate the size of the TCP/IP header and make sure that
435		 * we don't overflow the space we have available for it.
436		*/
437		hlen = ip->ip_hl << 2;
438		if (hlen + sizeof(struct tcphdr) > len)
439			goto bad;
440		th = (struct tcphdr *)&((int *)ip)[hlen];
441		hlen += THOFFSET(th) << 2;
442		if (hlen > MAX_HDR)
443			goto bad;
444		BCOPY(ip, &cs->cs_ip, hlen);
445		cs->cs_ip.ip_sum = 0;
446		cs->cs_hlen = hlen;
447		INCR(sls_uncompressedin)
448		return (len);
449
450	default:
451		goto bad;
452
453	case TYPE_COMPRESSED_TCP:
454		break;
455	}
456	/* We've got a compressed packet. */
457	INCR(sls_compressedin)
458	cp = *bufp;
459	changes = *cp++;
460#ifdef DEBUG
461	logprintf("compressed: changes = %02x\n", changes);
462#endif
463	if (changes & NEW_C) {
464		/* Make sure the state index is in range, then grab the state.
465		 * If we have a good state index, clear the 'discard' flag. */
466		if (*cp >= MAX_STATES || comp->last_recv == 255)
467			goto bad;
468
469		comp->flags &=~ SLF_TOSS;
470		comp->last_recv = *cp++;
471	} else {
472		/* this packet has an implicit state index.  If we've
473		 * had a line error since the last time we got an
474		 * explicit state index, we have to toss the packet. */
475		if (comp->flags & SLF_TOSS) {
476			INCR(sls_tossed)
477			return (0);
478		}
479	}
480	cs = &comp->rstate[comp->last_recv];
481	hlen = cs->cs_ip.ip_hl << 2;
482	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
483	th->th_sum = htons((*cp << 8) | cp[1]);
484	cp += 2;
485	if (changes & TCP_PUSH_BIT)
486		th->th_flags |= TH_PUSH;
487	else
488		th->th_flags &=~ TH_PUSH;
489
490	switch (changes & SPECIALS_MASK) {
491	case SPECIAL_I:
492		{
493		register u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
494		th->th_ack = htonl(ntohl(th->th_ack) + i);
495		th->th_seq = htonl(ntohl(th->th_seq) + i);
496		}
497		break;
498
499	case SPECIAL_D:
500		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
501				   - cs->cs_hlen);
502		break;
503
504	default:
505		if (changes & NEW_U) {
506			th->th_flags |= TH_URG;
507			DECODEU(th->th_urp)
508		} else
509			th->th_flags &=~ TH_URG;
510		if (changes & NEW_W)
511			DECODES(th->th_win)
512		if (changes & NEW_A)
513			DECODEL(th->th_ack)
514		if (changes & NEW_S) {
515#ifdef DEBUG
516		  logprintf("NEW_S: %02x, %02x, %02x\r\n", *cp, cp[1], cp[2]);
517#endif
518			DECODEL(th->th_seq)
519		}
520		break;
521	}
522	if (changes & NEW_I) {
523		DECODES(cs->cs_ip.ip_id)
524	} else
525		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
526#ifdef DEBUG
527	logprintf("id = %04x, seq = %08x\r\n", cs->cs_ip.ip_id, ntohl(th->th_seq));
528#endif
529
530	/*
531	 * At this point, cp points to the first byte of data in the
532	 * packet.  If we're not aligned on a 4-byte boundary, copy the
533	 * data down so the ip & tcp headers will be aligned.  Then back up
534	 * cp by the tcp/ip header length to make room for the reconstructed
535	 * header (we assume the packet we were handed has enough space to
536	 * prepend 128 bytes of header).  Adjust the length to account for
537	 * the new header & fill in the IP total length.
538	 */
539	len -= (cp - *bufp);
540	if (len < 0)
541		/* we must have dropped some characters (crc should detect
542		 * this but the old slip framing won't) */
543		goto bad;
544
545#ifdef notdef
546	if ((int)cp & 3) {
547		if (len > 0)
548			(void) ovbcopy(cp, (caddr_t)((int)cp &~ 3), len);
549		cp = (u_char *)((int)cp &~ 3);
550	}
551#endif
552
553	cp -= cs->cs_hlen;
554	len += cs->cs_hlen;
555	cs->cs_ip.ip_len = htons(len);
556	BCOPY(&cs->cs_ip, cp, cs->cs_hlen);
557	*bufp = cp;
558
559	/* recompute the ip header checksum */
560	{
561		register u_short *bp = (u_short *)cp;
562		for (changes = 0; hlen > 0; hlen -= 2)
563			changes += *bp++;
564		changes = (changes & 0xffff) + (changes >> 16);
565		changes = (changes & 0xffff) + (changes >> 16);
566		((struct ip *)cp)->ip_sum = ~ changes;
567	}
568	return (len);
569bad:
570	comp->flags |= SLF_TOSS;
571	INCR(sls_errorin)
572	return (0);
573}
574
575int
576ReportCompress()
577{
578  printf("Out:  %d (compress) / %d (total)",
579	slstat.sls_compressed, slstat.sls_packets);
580  printf("  %d (miss) / %d (search)\n",
581	slstat.sls_misses, slstat.sls_searches);
582  printf("In:  %d (compress), %d (uncompress)",
583	slstat.sls_compressedin, slstat.sls_uncompressedin);
584  printf("  %d (error),  %d (tossed)\n",
585	slstat.sls_errorin, slstat.sls_tossed);
586  return(1);
587}
588