xref: /freebsd/sys/net/slcompress.c (revision 0b3105a37d7adcadcb720112fed4dc4e8040be99)
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  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)slcompress.c	8.2 (Berkeley) 4/16/94
30  * $FreeBSD$
31  */
32 
33 /*
34  * Routines to compress and uncompess tcp packets (for transmission
35  * over low speed serial lines.
36  *
37  * Van Jacobson (van@helios.ee.lbl.gov), Dec 31, 1989:
38  *	- Initial distribution.
39  *
40  */
41 
42 #include <sys/param.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 
46 #include <netinet/in.h>
47 #include <netinet/in_systm.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((void *)(p1), (void *)(p2), (int)(n))
60 #define BCOPY(p1, p2, n) bcopy((void *)(p1), (void *)(p2), (int)(n))
61 
62 void
63 sl_compress_init(comp, max_state)
64 	struct slcompress *comp;
65 	int max_state;
66 {
67 	register u_int i;
68 	register struct cstate *tstate = comp->tstate;
69 
70 	if (max_state == -1) {
71 		max_state = MAX_STATES - 1;
72 		bzero((char *)comp, sizeof(*comp));
73 	} else {
74 		/* Don't reset statistics */
75 		bzero((char *)comp->tstate, sizeof(comp->tstate));
76 		bzero((char *)comp->rstate, sizeof(comp->rstate));
77 	}
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_int16_t)(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_int16_t)(n) >= 256 || (u_int16_t)(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_int32_t)*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_int32_t)*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_int32_t)*cp++); \
140 	} \
141 }
142 
143 /*
144  * Attempt to compress an outgoing TCP packet and return the type of
145  * the result.  The caller must have already verified that the protocol
146  * is TCP.  The first mbuf must contain the complete IP and TCP headers,
147  * and "ip" must be == mtod(m, struct ip *).  "comp" supplies the
148  * compression state, and "compress_cid" tells us whether it is OK
149  * to leave out the CID field when feasible.
150  *
151  * The caller is responsible for adjusting m->m_pkthdr.len upon return,
152  * if m is an M_PKTHDR mbuf.
153  */
154 u_int
155 sl_compress_tcp(m, ip, comp, compress_cid)
156 	struct mbuf *m;
157 	register struct ip *ip;
158 	struct slcompress *comp;
159 	int compress_cid;
160 {
161 	register struct cstate *cs = comp->last_cs->cs_next;
162 	register u_int hlen = ip->ip_hl;
163 	register struct tcphdr *oth;
164 	register struct tcphdr *th;
165 	register u_int deltaS, deltaA;
166 	register u_int changes = 0;
167 	u_char new_seq[16];
168 	register u_char *cp = new_seq;
169 
170 	/*
171 	 * Bail if this is an IP fragment or if the TCP packet isn't
172 	 * `compressible' (i.e., ACK isn't set or some other control bit is
173 	 * set).  (We assume that the caller has already made sure the
174 	 * packet is IP proto TCP).
175 	 */
176 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
177 		return (TYPE_IP);
178 
179 	th = (struct tcphdr *)&((int32_t *)ip)[hlen];
180 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
181 		return (TYPE_IP);
182 	/*
183 	 * Packet is compressible -- we're going to send either a
184 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
185 	 * to locate (or create) the connection state.  Special case the
186 	 * most recently used connection since it's most likely to be used
187 	 * again & we don't have to do any reordering if it's used.
188 	 */
189 	INCR(sls_packets)
190 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
191 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
192 	    *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
193 		/*
194 		 * Wasn't the first -- search for it.
195 		 *
196 		 * States are kept in a circularly linked list with
197 		 * last_cs pointing to the end of the list.  The
198 		 * list is kept in lru order by moving a state to the
199 		 * head of the list whenever it is referenced.  Since
200 		 * the list is short and, empirically, the connection
201 		 * we want is almost always near the front, we locate
202 		 * states via linear search.  If we don't find a state
203 		 * for the datagram, the oldest state is (re-)used.
204 		 */
205 		register struct cstate *lcs;
206 		register struct cstate *lastcs = comp->last_cs;
207 
208 		do {
209 			lcs = cs; cs = cs->cs_next;
210 			INCR(sls_searches)
211 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
212 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
213 			    && *(int32_t *)th ==
214 			    ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
215 				goto found;
216 		} while (cs != lastcs);
217 
218 		/*
219 		 * Didn't find it -- re-use oldest cstate.  Send an
220 		 * uncompressed packet that tells the other side what
221 		 * connection number we're using for this conversation.
222 		 * Note that since the state list is circular, the oldest
223 		 * state points to the newest and we only need to set
224 		 * last_cs to update the lru linkage.
225 		 */
226 		INCR(sls_misses)
227 		comp->last_cs = lcs;
228 		hlen += th->th_off;
229 		hlen <<= 2;
230 		if (hlen > m->m_len)
231 		    return TYPE_IP;
232 		goto uncompressed;
233 
234 	found:
235 		/*
236 		 * Found it -- move to the front on the connection list.
237 		 */
238 		if (cs == lastcs)
239 			comp->last_cs = lcs;
240 		else {
241 			lcs->cs_next = cs->cs_next;
242 			cs->cs_next = lastcs->cs_next;
243 			lastcs->cs_next = cs;
244 		}
245 	}
246 
247 	/*
248 	 * Make sure that only what we expect to change changed. The first
249 	 * line of the `if' checks the IP protocol version, header length &
250 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
251 	 * The 3rd line checks the time-to-live and protocol (the protocol
252 	 * check is unnecessary but costless).  The 4th line checks the TCP
253 	 * header length.  The 5th line checks IP options, if any.  The 6th
254 	 * line checks TCP options, if any.  If any of these things are
255 	 * different between the previous & current datagram, we send the
256 	 * current datagram `uncompressed'.
257 	 */
258 	oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
259 	deltaS = hlen;
260 	hlen += th->th_off;
261 	hlen <<= 2;
262 	if (hlen > m->m_len)
263 	    return TYPE_IP;
264 
265 	if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
266 	    ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
267 	    ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
268 	    th->th_off != oth->th_off ||
269 	    (deltaS > 5 &&
270 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
271 	    (th->th_off > 5 &&
272 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
273 		goto uncompressed;
274 
275 	/*
276 	 * Figure out which of the changing fields changed.  The
277 	 * receiver expects changes in the order: urgent, window,
278 	 * ack, seq (the order minimizes the number of temporaries
279 	 * needed in this section of code).
280 	 */
281 	if (th->th_flags & TH_URG) {
282 		deltaS = ntohs(th->th_urp);
283 		ENCODEZ(deltaS);
284 		changes |= NEW_U;
285 	} else if (th->th_urp != oth->th_urp)
286 		/* argh! URG not set but urp changed -- a sensible
287 		 * implementation should never do this but RFC793
288 		 * doesn't prohibit the change so we have to deal
289 		 * with it. */
290 		 goto uncompressed;
291 
292 	deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
293 	if (deltaS) {
294 		ENCODE(deltaS);
295 		changes |= NEW_W;
296 	}
297 
298 	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
299 	if (deltaA) {
300 		if (deltaA > 0xffff)
301 			goto uncompressed;
302 		ENCODE(deltaA);
303 		changes |= NEW_A;
304 	}
305 
306 	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
307 	if (deltaS) {
308 		if (deltaS > 0xffff)
309 			goto uncompressed;
310 		ENCODE(deltaS);
311 		changes |= NEW_S;
312 	}
313 
314 	switch(changes) {
315 
316 	case 0:
317 		/*
318 		 * Nothing changed. If this packet contains data and the
319 		 * last one didn't, this is probably a data packet following
320 		 * an ack (normal on an interactive connection) and we send
321 		 * it compressed.  Otherwise it's probably a retransmit,
322 		 * retransmitted ack or window probe.  Send it uncompressed
323 		 * in case the other side missed the compressed version.
324 		 */
325 		if (ip->ip_len != cs->cs_ip.ip_len &&
326 		    ntohs(cs->cs_ip.ip_len) == hlen)
327 			break;
328 
329 		/* FALLTHROUGH */
330 
331 	case SPECIAL_I:
332 	case SPECIAL_D:
333 		/*
334 		 * actual changes match one of our special case encodings --
335 		 * send packet uncompressed.
336 		 */
337 		goto uncompressed;
338 
339 	case NEW_S|NEW_A:
340 		if (deltaS == deltaA &&
341 		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
342 			/* special case for echoed terminal traffic */
343 			changes = SPECIAL_I;
344 			cp = new_seq;
345 		}
346 		break;
347 
348 	case NEW_S:
349 		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
350 			/* special case for data xfer */
351 			changes = SPECIAL_D;
352 			cp = new_seq;
353 		}
354 		break;
355 	}
356 
357 	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
358 	if (deltaS != 1) {
359 		ENCODEZ(deltaS);
360 		changes |= NEW_I;
361 	}
362 	if (th->th_flags & TH_PUSH)
363 		changes |= TCP_PUSH_BIT;
364 	/*
365 	 * Grab the cksum before we overwrite it below.  Then update our
366 	 * state with this packet's header.
367 	 */
368 	deltaA = ntohs(th->th_sum);
369 	BCOPY(ip, &cs->cs_ip, hlen);
370 
371 	/*
372 	 * We want to use the original packet as our compressed packet.
373 	 * (cp - new_seq) is the number of bytes we need for compressed
374 	 * sequence numbers.  In addition we need one byte for the change
375 	 * mask, one for the connection id and two for the tcp checksum.
376 	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
377 	 * many bytes of the original packet to toss so subtract the two to
378 	 * get the new packet size.
379 	 */
380 	deltaS = cp - new_seq;
381 	cp = (u_char *)ip;
382 	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
383 		comp->last_xmit = cs->cs_id;
384 		hlen -= deltaS + 4;
385 		cp += hlen;
386 		*cp++ = changes | NEW_C;
387 		*cp++ = cs->cs_id;
388 	} else {
389 		hlen -= deltaS + 3;
390 		cp += hlen;
391 		*cp++ = changes;
392 	}
393 	m->m_len -= hlen;
394 	m->m_data += hlen;
395 	*cp++ = deltaA >> 8;
396 	*cp++ = deltaA;
397 	BCOPY(new_seq, cp, deltaS);
398 	INCR(sls_compressed)
399 	return (TYPE_COMPRESSED_TCP);
400 
401 	/*
402 	 * Update connection state cs & send uncompressed packet ('uncompressed'
403 	 * means a regular ip/tcp packet but with the 'conversation id' we hope
404 	 * to use on future compressed packets in the protocol field).
405 	 */
406 uncompressed:
407 	BCOPY(ip, &cs->cs_ip, hlen);
408 	ip->ip_p = cs->cs_id;
409 	comp->last_xmit = cs->cs_id;
410 	return (TYPE_UNCOMPRESSED_TCP);
411 }
412 
413 
414 int
415 sl_uncompress_tcp(bufp, len, type, comp)
416 	u_char **bufp;
417 	int len;
418 	u_int type;
419 	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 ((intptr_t)cp & 3) {
443 		if (len > 0)
444 			BCOPY(cp, ((intptr_t)cp &~ 3), len);
445 		cp = (u_char *)((intptr_t)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  */
462 int
463 sl_uncompress_tcp_core(buf, buflen, total_len, type, comp, hdrp, hlenp)
464 	u_char *buf;
465 	int buflen, total_len;
466 	u_int type;
467 	struct slcompress *comp;
468 	u_char **hdrp;
469 	u_int *hlenp;
470 {
471 	register u_char *cp;
472 	register u_int hlen, changes;
473 	register struct tcphdr *th;
474 	register struct cstate *cs;
475 	register struct ip *ip;
476 	register u_int16_t *bp;
477 	register u_int vjlen;
478 
479 	switch (type) {
480 
481 	case TYPE_UNCOMPRESSED_TCP:
482 		ip = (struct ip *) buf;
483 		if (ip->ip_p >= MAX_STATES)
484 			goto bad;
485 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
486 		comp->flags &=~ SLF_TOSS;
487 		ip->ip_p = IPPROTO_TCP;
488 		/*
489 		 * Calculate the size of the TCP/IP header and make sure that
490 		 * we don't overflow the space we have available for it.
491 		 */
492 		hlen = ip->ip_hl << 2;
493 		if (hlen + sizeof(struct tcphdr) > buflen)
494 			goto bad;
495 		hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
496 		if (hlen > MAX_HDR || hlen > buflen)
497 			goto bad;
498 		BCOPY(ip, &cs->cs_ip, hlen);
499 		cs->cs_hlen = hlen;
500 		INCR(sls_uncompressedin)
501 		*hdrp = (u_char *) &cs->cs_ip;
502 		*hlenp = hlen;
503 		return (0);
504 
505 	default:
506 		goto bad;
507 
508 	case TYPE_COMPRESSED_TCP:
509 		break;
510 	}
511 	/* We've got a compressed packet. */
512 	INCR(sls_compressedin)
513 	cp = buf;
514 	changes = *cp++;
515 	if (changes & NEW_C) {
516 		/* Make sure the state index is in range, then grab the state.
517 		 * If we have a good state index, clear the 'discard' flag. */
518 		if (*cp >= MAX_STATES)
519 			goto bad;
520 
521 		comp->flags &=~ SLF_TOSS;
522 		comp->last_recv = *cp++;
523 	} else {
524 		/* this packet has an implicit state index.  If we've
525 		 * had a line error since the last time we got an
526 		 * explicit state index, we have to toss the packet. */
527 		if (comp->flags & SLF_TOSS) {
528 			INCR(sls_tossed)
529 			return (-1);
530 		}
531 	}
532 	cs = &comp->rstate[comp->last_recv];
533 	hlen = cs->cs_ip.ip_hl << 2;
534 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
535 	th->th_sum = htons((*cp << 8) | cp[1]);
536 	cp += 2;
537 	if (changes & TCP_PUSH_BIT)
538 		th->th_flags |= TH_PUSH;
539 	else
540 		th->th_flags &=~ TH_PUSH;
541 
542 	switch (changes & SPECIALS_MASK) {
543 	case SPECIAL_I:
544 		{
545 		register u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
546 		th->th_ack = htonl(ntohl(th->th_ack) + i);
547 		th->th_seq = htonl(ntohl(th->th_seq) + i);
548 		}
549 		break;
550 
551 	case SPECIAL_D:
552 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
553 				   - cs->cs_hlen);
554 		break;
555 
556 	default:
557 		if (changes & NEW_U) {
558 			th->th_flags |= TH_URG;
559 			DECODEU(th->th_urp)
560 		} else
561 			th->th_flags &=~ TH_URG;
562 		if (changes & NEW_W)
563 			DECODES(th->th_win)
564 		if (changes & NEW_A)
565 			DECODEL(th->th_ack)
566 		if (changes & NEW_S)
567 			DECODEL(th->th_seq)
568 		break;
569 	}
570 	if (changes & NEW_I) {
571 		DECODES(cs->cs_ip.ip_id)
572 	} else
573 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
574 
575 	/*
576 	 * At this point, cp points to the first byte of data in the
577 	 * packet.  Fill in the IP total length and update the IP
578 	 * header checksum.
579 	 */
580 	vjlen = cp - buf;
581 	buflen -= vjlen;
582 	if (buflen < 0)
583 		/* we must have dropped some characters (crc should detect
584 		 * this but the old slip framing won't) */
585 		goto bad;
586 
587 	total_len += cs->cs_hlen - vjlen;
588 	cs->cs_ip.ip_len = htons(total_len);
589 
590 	/* recompute the ip header checksum */
591 	bp = (u_int16_t *) &cs->cs_ip;
592 	cs->cs_ip.ip_sum = 0;
593 		for (changes = 0; hlen > 0; hlen -= 2)
594 			changes += *bp++;
595 		changes = (changes & 0xffff) + (changes >> 16);
596 		changes = (changes & 0xffff) + (changes >> 16);
597 	cs->cs_ip.ip_sum = ~ changes;
598 
599 	*hdrp = (u_char *) &cs->cs_ip;
600 	*hlenp = cs->cs_hlen;
601 	return vjlen;
602 
603 bad:
604 	comp->flags |= SLF_TOSS;
605 	INCR(sls_errorin)
606 	return (-1);
607 }
608