1/*	$OpenBSD: slcompress.c,v 1.12 2015/12/03 14:34:48 blambert Exp $	*/
2/*	$NetBSD: slcompress.c,v 1.17 1997/05/17 21:12:10 christos Exp $	*/
3
4/*
5 * Copyright (c) 1989, 1993, 1994
6 *	The Regents of the University of California.  All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	@(#)slcompress.c	8.2 (Berkeley) 4/16/94
33 */
34
35/*
36 * Routines to compress and uncompess tcp packets (for transmission
37 * over low speed serial lines.
38 *
39 * Van Jacobson (van@helios.ee.lbl.gov), Dec 31, 1989:
40 *	- Initial distribution.
41 */
42
43#include <sys/param.h>
44#include <sys/mbuf.h>
45#include <sys/systm.h>
46
47#include <netinet/in.h>
48#include <netinet/ip.h>
49#include <netinet/tcp.h>
50
51#include <net/slcompress.h>
52
53#ifndef SL_NO_STATS
54#define INCR(counter) ++comp->counter;
55#else
56#define INCR(counter)
57#endif
58
59#define BCMP(p1, p2, n) bcmp((char *)(p1), (char *)(p2), (int)(n))
60#define BCOPY(p1, p2, n) bcopy((char *)(p1), (char *)(p2), (int)(n))
61
62void
63sl_compress_init(struct slcompress *comp)
64{
65	u_int i;
66	struct cstate *tstate = comp->tstate;
67
68	bzero((char *)comp, sizeof(*comp));
69	for (i = MAX_STATES - 1; i > 0; --i) {
70		tstate[i].cs_id = i;
71		tstate[i].cs_next = &tstate[i - 1];
72	}
73	tstate[0].cs_next = &tstate[MAX_STATES - 1];
74	tstate[0].cs_id = 0;
75	comp->last_cs = &tstate[0];
76	comp->last_recv = 255;
77	comp->last_xmit = 255;
78	comp->flags = SLF_TOSS;
79}
80
81
82/*
83 * Like sl_compress_init, but we get to specify the maximum connection
84 * ID to use on transmission.
85 */
86void
87sl_compress_setup(struct slcompress *comp, int max_state)
88{
89	u_int i;
90	struct cstate *tstate = comp->tstate;
91
92	if (max_state == -1) {
93		max_state = MAX_STATES - 1;
94		bzero((char *)comp, sizeof(*comp));
95	} else {
96		/* Don't reset statistics */
97		bzero((char *)comp->tstate, sizeof(comp->tstate));
98		bzero((char *)comp->rstate, sizeof(comp->rstate));
99	}
100	for (i = max_state; i > 0; --i) {
101		tstate[i].cs_id = i;
102		tstate[i].cs_next = &tstate[i - 1];
103	}
104	tstate[0].cs_next = &tstate[max_state];
105	tstate[0].cs_id = 0;
106	comp->last_cs = &tstate[0];
107	comp->last_recv = 255;
108	comp->last_xmit = 255;
109	comp->flags = SLF_TOSS;
110}
111
112
113/* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
114 * checks for zero (since zero has to be encoded in the long, 3 byte
115 * form).
116 */
117#define ENCODE(n) { \
118	if ((u_int16_t)(n) >= 256) { \
119		*cp++ = 0; \
120		cp[1] = (n); \
121		cp[0] = (n) >> 8; \
122		cp += 2; \
123	} else { \
124		*cp++ = (n); \
125	} \
126}
127#define ENCODEZ(n) { \
128	if ((u_int16_t)(n) >= 256 || (u_int16_t)(n) == 0) { \
129		*cp++ = 0; \
130		cp[1] = (n); \
131		cp[0] = (n) >> 8; \
132		cp += 2; \
133	} else { \
134		*cp++ = (n); \
135	} \
136}
137
138#define DECODEL(f) { \
139	if (*cp == 0) {\
140		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
141		cp += 3; \
142	} else { \
143		(f) = htonl(ntohl(f) + (u_int32_t)*cp++); \
144	} \
145}
146
147#define DECODES(f) { \
148	if (*cp == 0) {\
149		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
150		cp += 3; \
151	} else { \
152		(f) = htons(ntohs(f) + (u_int32_t)*cp++); \
153	} \
154}
155
156#define DECODEU(f) { \
157	if (*cp == 0) {\
158		(f) = htons((cp[1] << 8) | cp[2]); \
159		cp += 3; \
160	} else { \
161		(f) = htons((u_int32_t)*cp++); \
162	} \
163}
164
165u_int
166sl_compress_tcp(struct mbuf *m, struct ip *ip, struct slcompress *comp,
167    int compress_cid)
168{
169	struct cstate *cs = comp->last_cs->cs_next;
170	u_int hlen = ip->ip_hl;
171	struct tcphdr *oth;
172	struct tcphdr *th;
173	u_int deltaS, deltaA;
174	u_int changes = 0;
175	u_char new_seq[16];
176	u_char *cp = new_seq;
177
178	/*
179	 * Bail if this is an IP fragment or if the TCP packet isn't
180	 * `compressible' (i.e., ACK isn't set or some other control bit is
181	 * set).  (We assume that the caller has already made sure the
182	 * packet is IP proto TCP).
183	 */
184	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
185		return (TYPE_IP);
186
187	th = (struct tcphdr *)&((int32_t *)ip)[hlen];
188	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
189		return (TYPE_IP);
190	/*
191	 * Packet is compressible -- we're going to send either a
192	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
193	 * to locate (or create) the connection state.  Special case the
194	 * most recently used connection since it's most likely to be used
195	 * again & we don't have to do any reordering if it's used.
196	 */
197	INCR(sls_packets)
198	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
199	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
200	    *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
201		/*
202		 * Wasn't the first -- search for it.
203		 *
204		 * States are kept in a circularly linked list with
205		 * last_cs pointing to the end of the list.  The
206		 * list is kept in lru order by moving a state to the
207		 * head of the list whenever it is referenced.  Since
208		 * the list is short and, empirically, the connection
209		 * we want is almost always near the front, we locate
210		 * states via linear search.  If we don't find a state
211		 * for the datagram, the oldest state is (re-)used.
212		 */
213		struct cstate *lcs;
214		struct cstate *lastcs = comp->last_cs;
215
216		do {
217			lcs = cs; cs = cs->cs_next;
218			INCR(sls_searches)
219			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
220			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
221			    && *(int32_t *)th ==
222			    ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
223				goto found;
224		} while (cs != lastcs);
225
226		/*
227		 * Didn't find it -- re-use oldest cstate.  Send an
228		 * uncompressed packet that tells the other side what
229		 * connection number we're using for this conversation.
230		 * Note that since the state list is circular, the oldest
231		 * state points to the newest and we only need to set
232		 * last_cs to update the lru linkage.
233		 */
234		INCR(sls_misses)
235		comp->last_cs = lcs;
236		hlen += th->th_off;
237		hlen <<= 2;
238		goto uncompressed;
239
240	found:
241		/*
242		 * Found it -- move to the front on the connection list.
243		 */
244		if (cs == lastcs)
245			comp->last_cs = lcs;
246		else {
247			lcs->cs_next = cs->cs_next;
248			cs->cs_next = lastcs->cs_next;
249			lastcs->cs_next = cs;
250		}
251	}
252
253	/*
254	 * Make sure that only what we expect to change changed. The first
255	 * line of the `if' checks the IP protocol version, header length &
256	 * type of service.  The 2nd line checks the "Don't fragment" bit.
257	 * The 3rd line checks the time-to-live and protocol (the protocol
258	 * check is unnecessary but costless).  The 4th line checks the TCP
259	 * header length.  The 5th line checks IP options, if any.  The 6th
260	 * line checks TCP options, if any.  If any of these things are
261	 * different between the previous & current datagram, we send the
262	 * current datagram `uncompressed'.
263	 */
264	oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
265	deltaS = hlen;
266	hlen += th->th_off;
267	hlen <<= 2;
268
269	if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
270	    ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
271	    ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
272	    th->th_off != oth->th_off ||
273	    (deltaS > 5 &&
274	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
275	    (th->th_off > 5 &&
276	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
277		goto uncompressed;
278
279	/*
280	 * Figure out which of the changing fields changed.  The
281	 * receiver expects changes in the order: urgent, window,
282	 * ack, seq (the order minimizes the number of temporaries
283	 * needed in this section of code).
284	 */
285	if (th->th_flags & TH_URG) {
286		deltaS = ntohs(th->th_urp);
287		ENCODEZ(deltaS);
288		changes |= NEW_U;
289	} else if (th->th_urp != oth->th_urp)
290		/* argh! URG not set but urp changed -- a sensible
291		 * implementation should never do this but RFC793
292		 * doesn't prohibit the change so we have to deal
293		 * with it. */
294		 goto uncompressed;
295
296	deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
297	if (deltaS) {
298		ENCODE(deltaS);
299		changes |= NEW_W;
300	}
301
302	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
303	if (deltaA) {
304		if (deltaA > 0xffff)
305			goto uncompressed;
306		ENCODE(deltaA);
307		changes |= NEW_A;
308	}
309
310	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
311	if (deltaS) {
312		if (deltaS > 0xffff)
313			goto uncompressed;
314		ENCODE(deltaS);
315		changes |= NEW_S;
316	}
317
318	switch(changes) {
319
320	case 0:
321		/*
322		 * Nothing changed. If this packet contains data and the
323		 * last one didn't, this is probably a data packet following
324		 * an ack (normal on an interactive connection) and we send
325		 * it compressed.  Otherwise it's probably a retransmit,
326		 * retransmitted ack or window probe.  Send it uncompressed
327		 * in case the other side missed the compressed version.
328		 */
329		if (ip->ip_len != cs->cs_ip.ip_len &&
330		    ntohs(cs->cs_ip.ip_len) == hlen)
331			break;
332
333		/* FALLTHROUGH */
334
335	case SPECIAL_I:
336	case SPECIAL_D:
337		/*
338		 * actual changes match one of our special case encodings --
339		 * send packet uncompressed.
340		 */
341		goto uncompressed;
342
343	case NEW_S|NEW_A:
344		if (deltaS == deltaA &&
345		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
346			/* special case for echoed terminal traffic */
347			changes = SPECIAL_I;
348			cp = new_seq;
349		}
350		break;
351
352	case NEW_S:
353		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
354			/* special case for data xfer */
355			changes = SPECIAL_D;
356			cp = new_seq;
357		}
358		break;
359	}
360
361	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
362	if (deltaS != 1) {
363		ENCODEZ(deltaS);
364		changes |= NEW_I;
365	}
366	if (th->th_flags & TH_PUSH)
367		changes |= TCP_PUSH_BIT;
368	/*
369	 * Grab the cksum before we overwrite it below.  Then update our
370	 * state with this packet's header.
371	 */
372	deltaA = ntohs(th->th_sum);
373	BCOPY(ip, &cs->cs_ip, hlen);
374
375	/*
376	 * We want to use the original packet as our compressed packet.
377	 * (cp - new_seq) is the number of bytes we need for compressed
378	 * sequence numbers.  In addition we need one byte for the change
379	 * mask, one for the connection id and two for the tcp checksum.
380	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
381	 * many bytes of the original packet to toss so subtract the two to
382	 * get the new packet size.
383	 */
384	deltaS = cp - new_seq;
385	cp = (u_char *)ip;
386	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
387		comp->last_xmit = cs->cs_id;
388		hlen -= deltaS + 4;
389		cp += hlen;
390		*cp++ = changes | NEW_C;
391		*cp++ = cs->cs_id;
392	} else {
393		hlen -= deltaS + 3;
394		cp += hlen;
395		*cp++ = changes;
396	}
397	m->m_len -= hlen;
398	m->m_data += hlen;
399	*cp++ = deltaA >> 8;
400	*cp++ = deltaA;
401	BCOPY(new_seq, cp, deltaS);
402	INCR(sls_compressed)
403	return (TYPE_COMPRESSED_TCP);
404
405	/*
406	 * Update connection state cs & send uncompressed packet ('uncompressed'
407	 * means a regular ip/tcp packet but with the 'conversation id' we hope
408	 * to use on future compressed packets in the protocol field).
409	 */
410uncompressed:
411	BCOPY(ip, &cs->cs_ip, hlen);
412	ip->ip_p = cs->cs_id;
413	comp->last_xmit = cs->cs_id;
414	return (TYPE_UNCOMPRESSED_TCP);
415}
416
417
418int
419sl_uncompress_tcp(u_char **bufp, int len, u_int type, struct slcompress *comp)
420{
421	u_char *hdr, *cp;
422	int hlen, vjlen;
423
424	cp = bufp? *bufp: NULL;
425	vjlen = sl_uncompress_tcp_core(cp, len, len, type, comp, &hdr, &hlen);
426	if (vjlen < 0)
427		return (0);	/* error */
428	if (vjlen == 0)
429		return (len);	/* was uncompressed already */
430
431	cp += vjlen;
432	len -= vjlen;
433
434	/*
435	 * At this point, cp points to the first byte of data in the
436	 * packet.  If we're not aligned on a 4-byte boundary, copy the
437	 * data down so the ip & tcp headers will be aligned.  Then back up
438	 * cp by the tcp/ip header length to make room for the reconstructed
439	 * header (we assume the packet we were handed has enough space to
440	 * prepend 128 bytes of header).
441	 */
442	if ((long)cp & 3) {
443		if (len > 0)
444			(void) memmove((caddr_t)((long)cp &~ 3), cp, len);
445		cp = (u_char *)((long)cp &~ 3);
446	}
447	cp -= hlen;
448	len += hlen;
449	BCOPY(hdr, cp, hlen);
450
451	*bufp = cp;
452	return (len);
453}
454
455/*
456 * Uncompress a packet of total length total_len.  The first buflen
457 * bytes are at buf; this must include the entire (compressed or
458 * uncompressed) TCP/IP header.  This procedure returns the length
459 * of the VJ header, with a pointer to the uncompressed IP header
460 * in *hdrp and its length in *hlenp.
461 */
462int
463sl_uncompress_tcp_core(u_char *buf, int buflen, int total_len, u_int type,
464    struct slcompress *comp, u_char **hdrp, u_int *hlenp)
465{
466	u_char *cp;
467	u_int hlen, changes;
468	struct tcphdr *th;
469	struct cstate *cs;
470	struct ip *ip;
471	u_int16_t *bp;
472	u_int vjlen;
473
474	switch (type) {
475
476	case TYPE_UNCOMPRESSED_TCP:
477		ip = (struct ip *) buf;
478		if (ip->ip_p >= MAX_STATES)
479			goto bad;
480		cs = &comp->rstate[comp->last_recv = ip->ip_p];
481		comp->flags &=~ SLF_TOSS;
482		ip->ip_p = IPPROTO_TCP;
483		/*
484		 * Calculate the size of the TCP/IP header and make sure that
485		 * we don't overflow the space we have available for it.
486		 */
487		hlen = ip->ip_hl << 2;
488		if (hlen + sizeof(struct tcphdr) > buflen)
489			goto bad;
490		hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
491		if (hlen > MAX_HDR || hlen > buflen)
492			goto bad;
493		BCOPY(ip, &cs->cs_ip, hlen);
494		cs->cs_hlen = hlen;
495		INCR(sls_uncompressedin)
496		*hdrp = (u_char *) &cs->cs_ip;
497		*hlenp = hlen;
498		return (0);
499
500	default:
501		goto bad;
502
503	case TYPE_COMPRESSED_TCP:
504		break;
505	}
506	/* We've got a compressed packet. */
507	INCR(sls_compressedin)
508	cp = buf;
509	changes = *cp++;
510	if (changes & NEW_C) {
511		/* Make sure the state index is in range, then grab the state.
512		 * If we have a good state index, clear the 'discard' flag. */
513		if (*cp >= MAX_STATES)
514			goto bad;
515
516		comp->flags &=~ SLF_TOSS;
517		comp->last_recv = *cp++;
518	} else {
519		/* this packet has an implicit state index.  If we've
520		 * had a line error since the last time we got an
521		 * explicit state index, we have to toss the packet. */
522		if (comp->flags & SLF_TOSS) {
523			INCR(sls_tossed)
524			return (-1);
525		}
526	}
527	cs = &comp->rstate[comp->last_recv];
528	hlen = cs->cs_ip.ip_hl << 2;
529	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
530	th->th_sum = htons((*cp << 8) | cp[1]);
531	cp += 2;
532	if (changes & TCP_PUSH_BIT)
533		th->th_flags |= TH_PUSH;
534	else
535		th->th_flags &=~ TH_PUSH;
536
537	switch (changes & SPECIALS_MASK) {
538	case SPECIAL_I:
539		{
540		u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
541		th->th_ack = htonl(ntohl(th->th_ack) + i);
542		th->th_seq = htonl(ntohl(th->th_seq) + i);
543		}
544		break;
545
546	case SPECIAL_D:
547		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
548				   - cs->cs_hlen);
549		break;
550
551	default:
552		if (changes & NEW_U) {
553			th->th_flags |= TH_URG;
554			DECODEU(th->th_urp)
555		} else
556			th->th_flags &=~ TH_URG;
557		if (changes & NEW_W)
558			DECODES(th->th_win)
559		if (changes & NEW_A)
560			DECODEL(th->th_ack)
561		if (changes & NEW_S)
562			DECODEL(th->th_seq)
563		break;
564	}
565	if (changes & NEW_I) {
566		DECODES(cs->cs_ip.ip_id)
567	} else
568		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
569
570	/*
571	 * At this point, cp points to the first byte of data in the
572	 * packet.  Fill in the IP total length and update the IP
573	 * header checksum.
574	 */
575	vjlen = cp - buf;
576	buflen -= vjlen;
577	if (buflen < 0)
578		/* we must have dropped some characters (crc should detect
579		 * this but the old slip framing won't) */
580		goto bad;
581
582	total_len += cs->cs_hlen - vjlen;
583	cs->cs_ip.ip_len = htons(total_len);
584
585	/* recompute the ip header checksum */
586	bp = (u_int16_t *) &cs->cs_ip;
587	cs->cs_ip.ip_sum = 0;
588	for (changes = 0; hlen > 0; hlen -= 2)
589		changes += *bp++;
590	changes = (changes & 0xffff) + (changes >> 16);
591	changes = (changes & 0xffff) + (changes >> 16);
592	cs->cs_ip.ip_sum = ~ changes;
593
594	*hdrp = (u_char *) &cs->cs_ip;
595	*hlenp = cs->cs_hlen;
596	return vjlen;
597
598bad:
599	comp->flags |= SLF_TOSS;
600	INCR(sls_errorin)
601	return (-1);
602}
603