xref: /netbsd-src/sys/net/slcompress.c (revision 6ea46cb5e46c49111a6ecf3bcbe3c7e2730fe9f6)
1 /*	$NetBSD: slcompress.c,v 1.9 1994/06/29 06:36:48 cgd Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)slcompress.c	8.2 (Berkeley) 4/16/94
36  */
37 
38 /*
39  * Routines to compress and uncompess tcp packets (for transmission
40  * over low speed serial lines.
41  *
42  * Van Jacobson (van@helios.ee.lbl.gov), Dec 31, 1989:
43  *	- Initial distribution.
44  */
45 
46 #include <sys/param.h>
47 #include <sys/mbuf.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/in_systm.h>
51 #include <netinet/ip.h>
52 #include <netinet/tcp.h>
53 
54 #include <net/slcompress.h>
55 
56 #ifndef SL_NO_STATS
57 #define INCR(counter) ++comp->counter;
58 #else
59 #define INCR(counter)
60 #endif
61 
62 #define BCMP(p1, p2, n) bcmp((char *)(p1), (char *)(p2), (int)(n))
63 #define BCOPY(p1, p2, n) bcopy((char *)(p1), (char *)(p2), (int)(n))
64 #ifndef KERNEL
65 #define ovbcopy bcopy
66 #endif
67 
68 void
69 sl_compress_init(comp, max_state)
70 	struct slcompress *comp;
71 	int max_state;
72 {
73 	register u_int i;
74 	register struct cstate *tstate = comp->tstate;
75 
76 	if (max_state == -1)
77 		max_state = MAX_STATES - 1;
78 	bzero((char *)comp, sizeof(*comp));
79 	for (i = max_state; i > 0; --i) {
80 		tstate[i].cs_id = i;
81 		tstate[i].cs_next = &tstate[i - 1];
82 	}
83 	tstate[0].cs_next = &tstate[max_state];
84 	tstate[0].cs_id = 0;
85 	comp->last_cs = &tstate[0];
86 	comp->last_recv = 255;
87 	comp->last_xmit = 255;
88 	comp->flags = SLF_TOSS;
89 }
90 
91 
92 /* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
93  * checks for zero (since zero has to be encoded in the long, 3 byte
94  * form).
95  */
96 #define ENCODE(n) { \
97 	if ((u_short)(n) >= 256) { \
98 		*cp++ = 0; \
99 		cp[1] = (n); \
100 		cp[0] = (n) >> 8; \
101 		cp += 2; \
102 	} else { \
103 		*cp++ = (n); \
104 	} \
105 }
106 #define ENCODEZ(n) { \
107 	if ((u_short)(n) >= 256 || (u_short)(n) == 0) { \
108 		*cp++ = 0; \
109 		cp[1] = (n); \
110 		cp[0] = (n) >> 8; \
111 		cp += 2; \
112 	} else { \
113 		*cp++ = (n); \
114 	} \
115 }
116 
117 #define DECODEL(f) { \
118 	if (*cp == 0) {\
119 		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
120 		cp += 3; \
121 	} else { \
122 		(f) = htonl(ntohl(f) + (u_long)*cp++); \
123 	} \
124 }
125 
126 #define DECODES(f) { \
127 	if (*cp == 0) {\
128 		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
129 		cp += 3; \
130 	} else { \
131 		(f) = htons(ntohs(f) + (u_long)*cp++); \
132 	} \
133 }
134 
135 #define DECODEU(f) { \
136 	if (*cp == 0) {\
137 		(f) = htons((cp[1] << 8) | cp[2]); \
138 		cp += 3; \
139 	} else { \
140 		(f) = htons((u_long)*cp++); \
141 	} \
142 }
143 
144 u_int
145 sl_compress_tcp(m, ip, comp, compress_cid)
146 	struct mbuf *m;
147 	register struct ip *ip;
148 	struct slcompress *comp;
149 	int compress_cid;
150 {
151 	register struct cstate *cs = comp->last_cs->cs_next;
152 	register u_int hlen = ip->ip_hl;
153 	register struct tcphdr *oth;
154 	register struct tcphdr *th;
155 	register u_int deltaS, deltaA;
156 	register u_int changes = 0;
157 	u_char new_seq[16];
158 	register u_char *cp = new_seq;
159 
160 	/*
161 	 * Bail if this is an IP fragment or if the TCP packet isn't
162 	 * `compressible' (i.e., ACK isn't set or some other control bit is
163 	 * set).  (We assume that the caller has already made sure the
164 	 * packet is IP proto TCP).
165 	 */
166 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
167 		return (TYPE_IP);
168 
169 	th = (struct tcphdr *)&((int *)ip)[hlen];
170 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
171 		return (TYPE_IP);
172 	/*
173 	 * Packet is compressible -- we're going to send either a
174 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
175 	 * to locate (or create) the connection state.  Special case the
176 	 * most recently used connection since it's most likely to be used
177 	 * again & we don't have to do any reordering if it's used.
178 	 */
179 	INCR(sls_packets)
180 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
181 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
182 	    *(int *)th != ((int *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
183 		/*
184 		 * Wasn't the first -- search for it.
185 		 *
186 		 * States are kept in a circularly linked list with
187 		 * last_cs pointing to the end of the list.  The
188 		 * list is kept in lru order by moving a state to the
189 		 * head of the list whenever it is referenced.  Since
190 		 * the list is short and, empirically, the connection
191 		 * we want is almost always near the front, we locate
192 		 * states via linear search.  If we don't find a state
193 		 * for the datagram, the oldest state is (re-)used.
194 		 */
195 		register struct cstate *lcs;
196 		register struct cstate *lastcs = comp->last_cs;
197 
198 		do {
199 			lcs = cs; cs = cs->cs_next;
200 			INCR(sls_searches)
201 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
202 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
203 			    && *(int *)th == ((int *)&cs->cs_ip)[cs->cs_ip.ip_hl])
204 				goto found;
205 		} while (cs != lastcs);
206 
207 		/*
208 		 * Didn't find it -- re-use oldest cstate.  Send an
209 		 * uncompressed packet that tells the other side what
210 		 * connection number we're using for this conversation.
211 		 * Note that since the state list is circular, the oldest
212 		 * state points to the newest and we only need to set
213 		 * last_cs to update the lru linkage.
214 		 */
215 		INCR(sls_misses)
216 		comp->last_cs = lcs;
217 		hlen += th->th_off;
218 		hlen <<= 2;
219 		goto uncompressed;
220 
221 	found:
222 		/*
223 		 * Found it -- move to the front on the connection list.
224 		 */
225 		if (cs == lastcs)
226 			comp->last_cs = lcs;
227 		else {
228 			lcs->cs_next = cs->cs_next;
229 			cs->cs_next = lastcs->cs_next;
230 			lastcs->cs_next = cs;
231 		}
232 	}
233 
234 	/*
235 	 * Make sure that only what we expect to change changed. The first
236 	 * line of the `if' checks the IP protocol version, header length &
237 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
238 	 * The 3rd line checks the time-to-live and protocol (the protocol
239 	 * check is unnecessary but costless).  The 4th line checks the TCP
240 	 * header length.  The 5th line checks IP options, if any.  The 6th
241 	 * line checks TCP options, if any.  If any of these things are
242 	 * different between the previous & current datagram, we send the
243 	 * current datagram `uncompressed'.
244 	 */
245 	oth = (struct tcphdr *)&((int *)&cs->cs_ip)[hlen];
246 	deltaS = hlen;
247 	hlen += th->th_off;
248 	hlen <<= 2;
249 
250 	if (((u_short *)ip)[0] != ((u_short *)&cs->cs_ip)[0] ||
251 	    ((u_short *)ip)[3] != ((u_short *)&cs->cs_ip)[3] ||
252 	    ((u_short *)ip)[4] != ((u_short *)&cs->cs_ip)[4] ||
253 	    th->th_off != oth->th_off ||
254 	    (deltaS > 5 &&
255 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
256 	    (th->th_off > 5 &&
257 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
258 		goto uncompressed;
259 
260 	/*
261 	 * Figure out which of the changing fields changed.  The
262 	 * receiver expects changes in the order: urgent, window,
263 	 * ack, seq (the order minimizes the number of temporaries
264 	 * needed in this section of code).
265 	 */
266 	if (th->th_flags & TH_URG) {
267 		deltaS = ntohs(th->th_urp);
268 		ENCODEZ(deltaS);
269 		changes |= NEW_U;
270 	} else if (th->th_urp != oth->th_urp)
271 		/* argh! URG not set but urp changed -- a sensible
272 		 * implementation should never do this but RFC793
273 		 * doesn't prohibit the change so we have to deal
274 		 * with it. */
275 		 goto uncompressed;
276 
277 	if (deltaS = (u_short)(ntohs(th->th_win) - ntohs(oth->th_win))) {
278 		ENCODE(deltaS);
279 		changes |= NEW_W;
280 	}
281 
282 	if (deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack)) {
283 		if (deltaA > 0xffff)
284 			goto uncompressed;
285 		ENCODE(deltaA);
286 		changes |= NEW_A;
287 	}
288 
289 	if (deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq)) {
290 		if (deltaS > 0xffff)
291 			goto uncompressed;
292 		ENCODE(deltaS);
293 		changes |= NEW_S;
294 	}
295 
296 	switch(changes) {
297 
298 	case 0:
299 		/*
300 		 * Nothing changed. If this packet contains data and the
301 		 * last one didn't, this is probably a data packet following
302 		 * an ack (normal on an interactive connection) and we send
303 		 * it compressed.  Otherwise it's probably a retransmit,
304 		 * retransmitted ack or window probe.  Send it uncompressed
305 		 * in case the other side missed the compressed version.
306 		 */
307 		if (ip->ip_len != cs->cs_ip.ip_len &&
308 		    ntohs(cs->cs_ip.ip_len) == hlen)
309 			break;
310 
311 		/* (fall through) */
312 
313 	case SPECIAL_I:
314 	case SPECIAL_D:
315 		/*
316 		 * actual changes match one of our special case encodings --
317 		 * send packet uncompressed.
318 		 */
319 		goto uncompressed;
320 
321 	case NEW_S|NEW_A:
322 		if (deltaS == deltaA &&
323 		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
324 			/* special case for echoed terminal traffic */
325 			changes = SPECIAL_I;
326 			cp = new_seq;
327 		}
328 		break;
329 
330 	case NEW_S:
331 		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
332 			/* special case for data xfer */
333 			changes = SPECIAL_D;
334 			cp = new_seq;
335 		}
336 		break;
337 	}
338 
339 	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
340 	if (deltaS != 1) {
341 		ENCODEZ(deltaS);
342 		changes |= NEW_I;
343 	}
344 	if (th->th_flags & TH_PUSH)
345 		changes |= TCP_PUSH_BIT;
346 	/*
347 	 * Grab the cksum before we overwrite it below.  Then update our
348 	 * state with this packet's header.
349 	 */
350 	deltaA = ntohs(th->th_sum);
351 	BCOPY(ip, &cs->cs_ip, hlen);
352 
353 	/*
354 	 * We want to use the original packet as our compressed packet.
355 	 * (cp - new_seq) is the number of bytes we need for compressed
356 	 * sequence numbers.  In addition we need one byte for the change
357 	 * mask, one for the connection id and two for the tcp checksum.
358 	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
359 	 * many bytes of the original packet to toss so subtract the two to
360 	 * get the new packet size.
361 	 */
362 	deltaS = cp - new_seq;
363 	cp = (u_char *)ip;
364 	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
365 		comp->last_xmit = cs->cs_id;
366 		hlen -= deltaS + 4;
367 		cp += hlen;
368 		*cp++ = changes | NEW_C;
369 		*cp++ = cs->cs_id;
370 	} else {
371 		hlen -= deltaS + 3;
372 		cp += hlen;
373 		*cp++ = changes;
374 	}
375 	m->m_len -= hlen;
376 	m->m_data += hlen;
377 	*cp++ = deltaA >> 8;
378 	*cp++ = deltaA;
379 	BCOPY(new_seq, cp, deltaS);
380 	INCR(sls_compressed)
381 	return (TYPE_COMPRESSED_TCP);
382 
383 	/*
384 	 * Update connection state cs & send uncompressed packet ('uncompressed'
385 	 * means a regular ip/tcp packet but with the 'conversation id' we hope
386 	 * to use on future compressed packets in the protocol field).
387 	 */
388 uncompressed:
389 	BCOPY(ip, &cs->cs_ip, hlen);
390 	ip->ip_p = cs->cs_id;
391 	comp->last_xmit = cs->cs_id;
392 	return (TYPE_UNCOMPRESSED_TCP);
393 }
394 
395 
396 int
397 sl_uncompress_tcp(bufp, len, type, comp)
398 	u_char **bufp;
399 	int len;
400 	u_int type;
401 	struct slcompress *comp;
402 {
403 
404 	return sl_uncompress_tcp_part(bufp, len, len, type, comp);
405 }
406 
407 /*
408  * Uncompress a packet of total length total_len.  The first buflen
409  * bytes are at *bufp; this must include the entire (compressed or
410  * uncompressed) TCP/IP header.  In addition, there must be enough
411  * clear space before *bufp to build a full-length TCP/IP header.
412  */
413 int
414 sl_uncompress_tcp_part(bufp, buflen, total_len, type, comp)
415 	u_char **bufp;
416 	int buflen, total_len;
417 	u_int type;
418 	struct slcompress *comp;
419 {
420 	register u_char *cp;
421 	register u_int hlen, changes;
422 	register struct tcphdr *th;
423 	register struct cstate *cs;
424 	register struct ip *ip;
425 
426 	switch (type) {
427 
428 	case TYPE_UNCOMPRESSED_TCP:
429 		ip = (struct ip *) *bufp;
430 		if (ip->ip_p >= MAX_STATES)
431 			goto bad;
432 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
433 		comp->flags &=~ SLF_TOSS;
434 		ip->ip_p = IPPROTO_TCP;
435 		hlen = ip->ip_hl;
436 		hlen += ((struct tcphdr *)&((int *)ip)[hlen])->th_off;
437 		hlen <<= 2;
438 		BCOPY(ip, &cs->cs_ip, hlen);
439 		cs->cs_ip.ip_sum = 0;
440 		cs->cs_hlen = hlen;
441 		INCR(sls_uncompressedin)
442 		return (total_len);
443 
444 	default:
445 		goto bad;
446 
447 	case TYPE_COMPRESSED_TCP:
448 		break;
449 	}
450 	/* We've got a compressed packet. */
451 	INCR(sls_compressedin)
452 	cp = *bufp;
453 	changes = *cp++;
454 	if (changes & NEW_C) {
455 		/* Make sure the state index is in range, then grab the state.
456 		 * If we have a good state index, clear the 'discard' flag. */
457 		if (*cp >= MAX_STATES)
458 			goto bad;
459 
460 		comp->flags &=~ SLF_TOSS;
461 		comp->last_recv = *cp++;
462 	} else {
463 		/* this packet has an implicit state index.  If we've
464 		 * had a line error since the last time we got an
465 		 * explicit state index, we have to toss the packet. */
466 		if (comp->flags & SLF_TOSS) {
467 			INCR(sls_tossed)
468 			return (0);
469 		}
470 	}
471 	cs = &comp->rstate[comp->last_recv];
472 	hlen = cs->cs_ip.ip_hl << 2;
473 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
474 	th->th_sum = htons((*cp << 8) | cp[1]);
475 	cp += 2;
476 	if (changes & TCP_PUSH_BIT)
477 		th->th_flags |= TH_PUSH;
478 	else
479 		th->th_flags &=~ TH_PUSH;
480 
481 	switch (changes & SPECIALS_MASK) {
482 	case SPECIAL_I:
483 		{
484 		register u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
485 		th->th_ack = htonl(ntohl(th->th_ack) + i);
486 		th->th_seq = htonl(ntohl(th->th_seq) + i);
487 		}
488 		break;
489 
490 	case SPECIAL_D:
491 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
492 				   - cs->cs_hlen);
493 		break;
494 
495 	default:
496 		if (changes & NEW_U) {
497 			th->th_flags |= TH_URG;
498 			DECODEU(th->th_urp)
499 		} else
500 			th->th_flags &=~ TH_URG;
501 		if (changes & NEW_W)
502 			DECODES(th->th_win)
503 		if (changes & NEW_A)
504 			DECODEL(th->th_ack)
505 		if (changes & NEW_S)
506 			DECODEL(th->th_seq)
507 		break;
508 	}
509 	if (changes & NEW_I) {
510 		DECODES(cs->cs_ip.ip_id)
511 	} else
512 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
513 
514 	/*
515 	 * At this point, cp points to the first byte of data in the
516 	 * packet.  If we're not aligned on a 4-byte boundary, copy the
517 	 * data down so the ip & tcp headers will be aligned.  Then back up
518 	 * cp by the tcp/ip header length to make room for the reconstructed
519 	 * header (we assume the packet we were handed has enough space to
520 	 * prepend 128 bytes of header).  Adjust the length to account for
521 	 * the new header & fill in the IP total length.
522 	 */
523 	buflen -= (cp - *bufp);
524 	total_len -= (cp - *bufp);
525 	if (buflen < 0)
526 		/* we must have dropped some characters (crc should detect
527 		 * this but the old slip framing won't) */
528 		goto bad;
529 
530 	if ((int)cp & 3) {
531 		if (buflen > 0)
532 			(void) ovbcopy(cp, (caddr_t)((int)cp &~ 3), buflen);
533 		cp = (u_char *)((int)cp &~ 3);
534 	}
535 	cp -= cs->cs_hlen;
536 	total_len += cs->cs_hlen;
537 	cs->cs_ip.ip_len = htons(total_len);
538 	BCOPY(&cs->cs_ip, cp, cs->cs_hlen);
539 	*bufp = cp;
540 
541 	/* recompute the ip header checksum */
542 	{
543 		register u_short *bp = (u_short *)cp;
544 		for (changes = 0; hlen > 0; hlen -= 2)
545 			changes += *bp++;
546 		changes = (changes & 0xffff) + (changes >> 16);
547 		changes = (changes & 0xffff) + (changes >> 16);
548 		((struct ip *)cp)->ip_sum = ~ changes;
549 	}
550 	return (total_len);
551 bad:
552 	comp->flags |= SLF_TOSS;
553 	INCR(sls_errorin)
554 	return (0);
555 }
556