xref: /netbsd-src/sys/net/slcompress.c (revision ae1bfcddc410612bc8c58b807e1830becb69a24c)
1 /*
2  * Copyright (c) 1989, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	from: @(#)slcompress.c	8.2 (Berkeley) 4/16/94
34  *	$Id: slcompress.c,v 1.8 1994/05/13 06:03:25 mycroft Exp $
35  */
36 
37 /*
38  * Routines to compress and uncompess tcp packets (for transmission
39  * over low speed serial lines.
40  *
41  * Van Jacobson (van@helios.ee.lbl.gov), Dec 31, 1989:
42  *	- Initial distribution.
43  */
44 
45 #include <sys/param.h>
46 #include <sys/mbuf.h>
47 
48 #include <netinet/in.h>
49 #include <netinet/in_systm.h>
50 #include <netinet/ip.h>
51 #include <netinet/tcp.h>
52 
53 #include <net/slcompress.h>
54 
55 #ifndef SL_NO_STATS
56 #define INCR(counter) ++comp->counter;
57 #else
58 #define INCR(counter)
59 #endif
60 
61 #define BCMP(p1, p2, n) bcmp((char *)(p1), (char *)(p2), (int)(n))
62 #define BCOPY(p1, p2, n) bcopy((char *)(p1), (char *)(p2), (int)(n))
63 #ifndef KERNEL
64 #define ovbcopy bcopy
65 #endif
66 
67 void
68 sl_compress_init(comp, max_state)
69 	struct slcompress *comp;
70 	int max_state;
71 {
72 	register u_int i;
73 	register struct cstate *tstate = comp->tstate;
74 
75 	if (max_state == -1)
76 		max_state = MAX_STATES - 1;
77 	bzero((char *)comp, sizeof(*comp));
78 	for (i = max_state; i > 0; --i) {
79 		tstate[i].cs_id = i;
80 		tstate[i].cs_next = &tstate[i - 1];
81 	}
82 	tstate[0].cs_next = &tstate[max_state];
83 	tstate[0].cs_id = 0;
84 	comp->last_cs = &tstate[0];
85 	comp->last_recv = 255;
86 	comp->last_xmit = 255;
87 	comp->flags = SLF_TOSS;
88 }
89 
90 
91 /* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
92  * checks for zero (since zero has to be encoded in the long, 3 byte
93  * form).
94  */
95 #define ENCODE(n) { \
96 	if ((u_short)(n) >= 256) { \
97 		*cp++ = 0; \
98 		cp[1] = (n); \
99 		cp[0] = (n) >> 8; \
100 		cp += 2; \
101 	} else { \
102 		*cp++ = (n); \
103 	} \
104 }
105 #define ENCODEZ(n) { \
106 	if ((u_short)(n) >= 256 || (u_short)(n) == 0) { \
107 		*cp++ = 0; \
108 		cp[1] = (n); \
109 		cp[0] = (n) >> 8; \
110 		cp += 2; \
111 	} else { \
112 		*cp++ = (n); \
113 	} \
114 }
115 
116 #define DECODEL(f) { \
117 	if (*cp == 0) {\
118 		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
119 		cp += 3; \
120 	} else { \
121 		(f) = htonl(ntohl(f) + (u_long)*cp++); \
122 	} \
123 }
124 
125 #define DECODES(f) { \
126 	if (*cp == 0) {\
127 		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
128 		cp += 3; \
129 	} else { \
130 		(f) = htons(ntohs(f) + (u_long)*cp++); \
131 	} \
132 }
133 
134 #define DECODEU(f) { \
135 	if (*cp == 0) {\
136 		(f) = htons((cp[1] << 8) | cp[2]); \
137 		cp += 3; \
138 	} else { \
139 		(f) = htons((u_long)*cp++); \
140 	} \
141 }
142 
143 u_int
144 sl_compress_tcp(m, ip, comp, compress_cid)
145 	struct mbuf *m;
146 	register struct ip *ip;
147 	struct slcompress *comp;
148 	int compress_cid;
149 {
150 	register struct cstate *cs = comp->last_cs->cs_next;
151 	register u_int hlen = ip->ip_hl;
152 	register struct tcphdr *oth;
153 	register struct tcphdr *th;
154 	register u_int deltaS, deltaA;
155 	register u_int changes = 0;
156 	u_char new_seq[16];
157 	register u_char *cp = new_seq;
158 
159 	/*
160 	 * Bail if this is an IP fragment or if the TCP packet isn't
161 	 * `compressible' (i.e., ACK isn't set or some other control bit is
162 	 * set).  (We assume that the caller has already made sure the
163 	 * packet is IP proto TCP).
164 	 */
165 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
166 		return (TYPE_IP);
167 
168 	th = (struct tcphdr *)&((int *)ip)[hlen];
169 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
170 		return (TYPE_IP);
171 	/*
172 	 * Packet is compressible -- we're going to send either a
173 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
174 	 * to locate (or create) the connection state.  Special case the
175 	 * most recently used connection since it's most likely to be used
176 	 * again & we don't have to do any reordering if it's used.
177 	 */
178 	INCR(sls_packets)
179 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
180 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
181 	    *(int *)th != ((int *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
182 		/*
183 		 * Wasn't the first -- search for it.
184 		 *
185 		 * States are kept in a circularly linked list with
186 		 * last_cs pointing to the end of the list.  The
187 		 * list is kept in lru order by moving a state to the
188 		 * head of the list whenever it is referenced.  Since
189 		 * the list is short and, empirically, the connection
190 		 * we want is almost always near the front, we locate
191 		 * states via linear search.  If we don't find a state
192 		 * for the datagram, the oldest state is (re-)used.
193 		 */
194 		register struct cstate *lcs;
195 		register struct cstate *lastcs = comp->last_cs;
196 
197 		do {
198 			lcs = cs; cs = cs->cs_next;
199 			INCR(sls_searches)
200 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
201 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
202 			    && *(int *)th == ((int *)&cs->cs_ip)[cs->cs_ip.ip_hl])
203 				goto found;
204 		} while (cs != lastcs);
205 
206 		/*
207 		 * Didn't find it -- re-use oldest cstate.  Send an
208 		 * uncompressed packet that tells the other side what
209 		 * connection number we're using for this conversation.
210 		 * Note that since the state list is circular, the oldest
211 		 * state points to the newest and we only need to set
212 		 * last_cs to update the lru linkage.
213 		 */
214 		INCR(sls_misses)
215 		comp->last_cs = lcs;
216 		hlen += th->th_off;
217 		hlen <<= 2;
218 		goto uncompressed;
219 
220 	found:
221 		/*
222 		 * Found it -- move to the front on the connection list.
223 		 */
224 		if (cs == lastcs)
225 			comp->last_cs = lcs;
226 		else {
227 			lcs->cs_next = cs->cs_next;
228 			cs->cs_next = lastcs->cs_next;
229 			lastcs->cs_next = cs;
230 		}
231 	}
232 
233 	/*
234 	 * Make sure that only what we expect to change changed. The first
235 	 * line of the `if' checks the IP protocol version, header length &
236 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
237 	 * The 3rd line checks the time-to-live and protocol (the protocol
238 	 * check is unnecessary but costless).  The 4th line checks the TCP
239 	 * header length.  The 5th line checks IP options, if any.  The 6th
240 	 * line checks TCP options, if any.  If any of these things are
241 	 * different between the previous & current datagram, we send the
242 	 * current datagram `uncompressed'.
243 	 */
244 	oth = (struct tcphdr *)&((int *)&cs->cs_ip)[hlen];
245 	deltaS = hlen;
246 	hlen += th->th_off;
247 	hlen <<= 2;
248 
249 	if (((u_short *)ip)[0] != ((u_short *)&cs->cs_ip)[0] ||
250 	    ((u_short *)ip)[3] != ((u_short *)&cs->cs_ip)[3] ||
251 	    ((u_short *)ip)[4] != ((u_short *)&cs->cs_ip)[4] ||
252 	    th->th_off != oth->th_off ||
253 	    (deltaS > 5 &&
254 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
255 	    (th->th_off > 5 &&
256 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
257 		goto uncompressed;
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 	if (deltaS = (u_short)(ntohs(th->th_win) - ntohs(oth->th_win))) {
277 		ENCODE(deltaS);
278 		changes |= NEW_W;
279 	}
280 
281 	if (deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack)) {
282 		if (deltaA > 0xffff)
283 			goto uncompressed;
284 		ENCODE(deltaA);
285 		changes |= NEW_A;
286 	}
287 
288 	if (deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq)) {
289 		if (deltaS > 0xffff)
290 			goto uncompressed;
291 		ENCODE(deltaS);
292 		changes |= NEW_S;
293 	}
294 
295 	switch(changes) {
296 
297 	case 0:
298 		/*
299 		 * Nothing changed. If this packet contains data and the
300 		 * last one didn't, this is probably a data packet following
301 		 * an ack (normal on an interactive connection) and we send
302 		 * it compressed.  Otherwise it's probably a retransmit,
303 		 * retransmitted ack or window probe.  Send it uncompressed
304 		 * in case the other side missed the compressed version.
305 		 */
306 		if (ip->ip_len != cs->cs_ip.ip_len &&
307 		    ntohs(cs->cs_ip.ip_len) == hlen)
308 			break;
309 
310 		/* (fall through) */
311 
312 	case SPECIAL_I:
313 	case SPECIAL_D:
314 		/*
315 		 * actual changes match one of our special case encodings --
316 		 * send packet uncompressed.
317 		 */
318 		goto uncompressed;
319 
320 	case NEW_S|NEW_A:
321 		if (deltaS == deltaA &&
322 		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
323 			/* special case for echoed terminal traffic */
324 			changes = SPECIAL_I;
325 			cp = new_seq;
326 		}
327 		break;
328 
329 	case NEW_S:
330 		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
331 			/* special case for data xfer */
332 			changes = SPECIAL_D;
333 			cp = new_seq;
334 		}
335 		break;
336 	}
337 
338 	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
339 	if (deltaS != 1) {
340 		ENCODEZ(deltaS);
341 		changes |= NEW_I;
342 	}
343 	if (th->th_flags & TH_PUSH)
344 		changes |= TCP_PUSH_BIT;
345 	/*
346 	 * Grab the cksum before we overwrite it below.  Then update our
347 	 * state with this packet's header.
348 	 */
349 	deltaA = ntohs(th->th_sum);
350 	BCOPY(ip, &cs->cs_ip, hlen);
351 
352 	/*
353 	 * We want to use the original packet as our compressed packet.
354 	 * (cp - new_seq) is the number of bytes we need for compressed
355 	 * sequence numbers.  In addition we need one byte for the change
356 	 * mask, one for the connection id and two for the tcp checksum.
357 	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
358 	 * many bytes of the original packet to toss so subtract the two to
359 	 * get the new packet size.
360 	 */
361 	deltaS = cp - new_seq;
362 	cp = (u_char *)ip;
363 	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
364 		comp->last_xmit = cs->cs_id;
365 		hlen -= deltaS + 4;
366 		cp += hlen;
367 		*cp++ = changes | NEW_C;
368 		*cp++ = cs->cs_id;
369 	} else {
370 		hlen -= deltaS + 3;
371 		cp += hlen;
372 		*cp++ = changes;
373 	}
374 	m->m_len -= hlen;
375 	m->m_data += hlen;
376 	*cp++ = deltaA >> 8;
377 	*cp++ = deltaA;
378 	BCOPY(new_seq, cp, deltaS);
379 	INCR(sls_compressed)
380 	return (TYPE_COMPRESSED_TCP);
381 
382 	/*
383 	 * Update connection state cs & send uncompressed packet ('uncompressed'
384 	 * means a regular ip/tcp packet but with the 'conversation id' we hope
385 	 * to use on future compressed packets in the protocol field).
386 	 */
387 uncompressed:
388 	BCOPY(ip, &cs->cs_ip, hlen);
389 	ip->ip_p = cs->cs_id;
390 	comp->last_xmit = cs->cs_id;
391 	return (TYPE_UNCOMPRESSED_TCP);
392 }
393 
394 
395 int
396 sl_uncompress_tcp(bufp, len, type, comp)
397 	u_char **bufp;
398 	int len;
399 	u_int type;
400 	struct slcompress *comp;
401 {
402 
403 	return sl_uncompress_tcp_part(bufp, len, len, type, comp);
404 }
405 
406 /*
407  * Uncompress a packet of total length total_len.  The first buflen
408  * bytes are at *bufp; this must include the entire (compressed or
409  * uncompressed) TCP/IP header.  In addition, there must be enough
410  * clear space before *bufp to build a full-length TCP/IP header.
411  */
412 int
413 sl_uncompress_tcp_part(bufp, buflen, total_len, type, comp)
414 	u_char **bufp;
415 	int buflen, total_len;
416 	u_int type;
417 	struct slcompress *comp;
418 {
419 	register u_char *cp;
420 	register u_int hlen, changes;
421 	register struct tcphdr *th;
422 	register struct cstate *cs;
423 	register struct ip *ip;
424 
425 	switch (type) {
426 
427 	case TYPE_UNCOMPRESSED_TCP:
428 		ip = (struct ip *) *bufp;
429 		if (ip->ip_p >= MAX_STATES)
430 			goto bad;
431 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
432 		comp->flags &=~ SLF_TOSS;
433 		ip->ip_p = IPPROTO_TCP;
434 		hlen = ip->ip_hl;
435 		hlen += ((struct tcphdr *)&((int *)ip)[hlen])->th_off;
436 		hlen <<= 2;
437 		BCOPY(ip, &cs->cs_ip, hlen);
438 		cs->cs_ip.ip_sum = 0;
439 		cs->cs_hlen = hlen;
440 		INCR(sls_uncompressedin)
441 		return (total_len);
442 
443 	default:
444 		goto bad;
445 
446 	case TYPE_COMPRESSED_TCP:
447 		break;
448 	}
449 	/* We've got a compressed packet. */
450 	INCR(sls_compressedin)
451 	cp = *bufp;
452 	changes = *cp++;
453 	if (changes & NEW_C) {
454 		/* Make sure the state index is in range, then grab the state.
455 		 * If we have a good state index, clear the 'discard' flag. */
456 		if (*cp >= MAX_STATES)
457 			goto bad;
458 
459 		comp->flags &=~ SLF_TOSS;
460 		comp->last_recv = *cp++;
461 	} else {
462 		/* this packet has an implicit state index.  If we've
463 		 * had a line error since the last time we got an
464 		 * explicit state index, we have to toss the packet. */
465 		if (comp->flags & SLF_TOSS) {
466 			INCR(sls_tossed)
467 			return (0);
468 		}
469 	}
470 	cs = &comp->rstate[comp->last_recv];
471 	hlen = cs->cs_ip.ip_hl << 2;
472 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
473 	th->th_sum = htons((*cp << 8) | cp[1]);
474 	cp += 2;
475 	if (changes & TCP_PUSH_BIT)
476 		th->th_flags |= TH_PUSH;
477 	else
478 		th->th_flags &=~ TH_PUSH;
479 
480 	switch (changes & SPECIALS_MASK) {
481 	case SPECIAL_I:
482 		{
483 		register u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
484 		th->th_ack = htonl(ntohl(th->th_ack) + i);
485 		th->th_seq = htonl(ntohl(th->th_seq) + i);
486 		}
487 		break;
488 
489 	case SPECIAL_D:
490 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
491 				   - cs->cs_hlen);
492 		break;
493 
494 	default:
495 		if (changes & NEW_U) {
496 			th->th_flags |= TH_URG;
497 			DECODEU(th->th_urp)
498 		} else
499 			th->th_flags &=~ TH_URG;
500 		if (changes & NEW_W)
501 			DECODES(th->th_win)
502 		if (changes & NEW_A)
503 			DECODEL(th->th_ack)
504 		if (changes & NEW_S)
505 			DECODEL(th->th_seq)
506 		break;
507 	}
508 	if (changes & NEW_I) {
509 		DECODES(cs->cs_ip.ip_id)
510 	} else
511 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
512 
513 	/*
514 	 * At this point, cp points to the first byte of data in the
515 	 * packet.  If we're not aligned on a 4-byte boundary, copy the
516 	 * data down so the ip & tcp headers will be aligned.  Then back up
517 	 * cp by the tcp/ip header length to make room for the reconstructed
518 	 * header (we assume the packet we were handed has enough space to
519 	 * prepend 128 bytes of header).  Adjust the length to account for
520 	 * the new header & fill in the IP total length.
521 	 */
522 	buflen -= (cp - *bufp);
523 	total_len -= (cp - *bufp);
524 	if (buflen < 0)
525 		/* we must have dropped some characters (crc should detect
526 		 * this but the old slip framing won't) */
527 		goto bad;
528 
529 	if ((int)cp & 3) {
530 		if (buflen > 0)
531 			(void) ovbcopy(cp, (caddr_t)((int)cp &~ 3), buflen);
532 		cp = (u_char *)((int)cp &~ 3);
533 	}
534 	cp -= cs->cs_hlen;
535 	total_len += cs->cs_hlen;
536 	cs->cs_ip.ip_len = htons(total_len);
537 	BCOPY(&cs->cs_ip, cp, cs->cs_hlen);
538 	*bufp = cp;
539 
540 	/* recompute the ip header checksum */
541 	{
542 		register u_short *bp = (u_short *)cp;
543 		for (changes = 0; hlen > 0; hlen -= 2)
544 			changes += *bp++;
545 		changes = (changes & 0xffff) + (changes >> 16);
546 		changes = (changes & 0xffff) + (changes >> 16);
547 		((struct ip *)cp)->ip_sum = ~ changes;
548 	}
549 	return (total_len);
550 bad:
551 	comp->flags |= SLF_TOSS;
552 	INCR(sls_errorin)
553 	return (0);
554 }
555