xref: /freebsd/sys/net/slcompress.c (revision 5c831a5bd61576cacb48b39f8eeb47b92707a355)
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. 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(struct slcompress *comp, int max_state)
64 {
65 	u_int i;
66 	struct cstate *tstate = comp->tstate;
67 
68 	if (max_state == -1) {
69 		max_state = MAX_STATES - 1;
70 		bzero((char *)comp, sizeof(*comp));
71 	} else {
72 		/* Don't reset statistics */
73 		bzero((char *)comp->tstate, sizeof(comp->tstate));
74 		bzero((char *)comp->rstate, sizeof(comp->rstate));
75 	}
76   	for (i = max_state; i > 0; --i) {
77 		tstate[i].cs_id = i;
78 		tstate[i].cs_next = &tstate[i - 1];
79 	}
80 	tstate[0].cs_next = &tstate[max_state];
81 	tstate[0].cs_id = 0;
82 	comp->last_cs = &tstate[0];
83 	comp->last_recv = 255;
84 	comp->last_xmit = 255;
85 	comp->flags = SLF_TOSS;
86 }
87 
88 
89 /* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
90  * checks for zero (since zero has to be encoded in the long, 3 byte
91  * form).
92  */
93 #define ENCODE(n) { \
94 	if ((u_int16_t)(n) >= 256) { \
95 		*cp++ = 0; \
96 		cp[1] = (n); \
97 		cp[0] = (n) >> 8; \
98 		cp += 2; \
99 	} else { \
100 		*cp++ = (n); \
101 	} \
102 }
103 #define ENCODEZ(n) { \
104 	if ((u_int16_t)(n) >= 256 || (u_int16_t)(n) == 0) { \
105 		*cp++ = 0; \
106 		cp[1] = (n); \
107 		cp[0] = (n) >> 8; \
108 		cp += 2; \
109 	} else { \
110 		*cp++ = (n); \
111 	} \
112 }
113 
114 #define DECODEL(f) { \
115 	if (*cp == 0) {\
116 		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
117 		cp += 3; \
118 	} else { \
119 		(f) = htonl(ntohl(f) + (u_int32_t)*cp++); \
120 	} \
121 }
122 
123 #define DECODES(f) { \
124 	if (*cp == 0) {\
125 		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
126 		cp += 3; \
127 	} else { \
128 		(f) = htons(ntohs(f) + (u_int32_t)*cp++); \
129 	} \
130 }
131 
132 #define DECODEU(f) { \
133 	if (*cp == 0) {\
134 		(f) = htons((cp[1] << 8) | cp[2]); \
135 		cp += 3; \
136 	} else { \
137 		(f) = htons((u_int32_t)*cp++); \
138 	} \
139 }
140 
141 /*
142  * Attempt to compress an outgoing TCP packet and return the type of
143  * the result.  The caller must have already verified that the protocol
144  * is TCP.  The first mbuf must contain the complete IP and TCP headers,
145  * and "ip" must be == mtod(m, struct ip *).  "comp" supplies the
146  * compression state, and "compress_cid" tells us whether it is OK
147  * to leave out the CID field when feasible.
148  *
149  * The caller is responsible for adjusting m->m_pkthdr.len upon return,
150  * if m is an M_PKTHDR mbuf.
151  */
152 u_int
153 sl_compress_tcp(struct mbuf *m, struct ip *ip, struct slcompress *comp,
154     int compress_cid)
155 {
156 	struct cstate *cs = comp->last_cs->cs_next;
157 	u_int hlen = ip->ip_hl;
158 	struct tcphdr *oth;
159 	struct tcphdr *th;
160 	u_int deltaS, deltaA;
161 	u_int changes = 0;
162 	u_char new_seq[16];
163 	u_char *cp = new_seq;
164 
165 	/*
166 	 * Bail if this is an IP fragment or if the TCP packet isn't
167 	 * `compressible' (i.e., ACK isn't set or some other control bit is
168 	 * set).  (We assume that the caller has already made sure the
169 	 * packet is IP proto TCP).
170 	 */
171 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
172 		return (TYPE_IP);
173 
174 	th = (struct tcphdr *)&((int32_t *)ip)[hlen];
175 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
176 		return (TYPE_IP);
177 	/*
178 	 * Packet is compressible -- we're going to send either a
179 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
180 	 * to locate (or create) the connection state.  Special case the
181 	 * most recently used connection since it's most likely to be used
182 	 * again & we don't have to do any reordering if it's used.
183 	 */
184 	INCR(sls_packets)
185 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
186 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
187 	    *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
188 		/*
189 		 * Wasn't the first -- search for it.
190 		 *
191 		 * States are kept in a circularly linked list with
192 		 * last_cs pointing to the end of the list.  The
193 		 * list is kept in lru order by moving a state to the
194 		 * head of the list whenever it is referenced.  Since
195 		 * the list is short and, empirically, the connection
196 		 * we want is almost always near the front, we locate
197 		 * states via linear search.  If we don't find a state
198 		 * for the datagram, the oldest state is (re-)used.
199 		 */
200 		struct cstate *lcs;
201 		struct cstate *lastcs = comp->last_cs;
202 
203 		do {
204 			lcs = cs; cs = cs->cs_next;
205 			INCR(sls_searches)
206 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
207 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
208 			    && *(int32_t *)th ==
209 			    ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
210 				goto found;
211 		} while (cs != lastcs);
212 
213 		/*
214 		 * Didn't find it -- re-use oldest cstate.  Send an
215 		 * uncompressed packet that tells the other side what
216 		 * connection number we're using for this conversation.
217 		 * Note that since the state list is circular, the oldest
218 		 * state points to the newest and we only need to set
219 		 * last_cs to update the lru linkage.
220 		 */
221 		INCR(sls_misses)
222 		comp->last_cs = lcs;
223 		hlen += th->th_off;
224 		hlen <<= 2;
225 		if (hlen > m->m_len)
226 		    return TYPE_IP;
227 		goto uncompressed;
228 
229 	found:
230 		/*
231 		 * Found it -- move to the front on the connection list.
232 		 */
233 		if (cs == lastcs)
234 			comp->last_cs = lcs;
235 		else {
236 			lcs->cs_next = cs->cs_next;
237 			cs->cs_next = lastcs->cs_next;
238 			lastcs->cs_next = cs;
239 		}
240 	}
241 
242 	/*
243 	 * Make sure that only what we expect to change changed. The first
244 	 * line of the `if' checks the IP protocol version, header length &
245 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
246 	 * The 3rd line checks the time-to-live and protocol (the protocol
247 	 * check is unnecessary but costless).  The 4th line checks the TCP
248 	 * header length.  The 5th line checks IP options, if any.  The 6th
249 	 * line checks TCP options, if any.  If any of these things are
250 	 * different between the previous & current datagram, we send the
251 	 * current datagram `uncompressed'.
252 	 */
253 	oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
254 	deltaS = hlen;
255 	hlen += th->th_off;
256 	hlen <<= 2;
257 	if (hlen > m->m_len)
258 	    return TYPE_IP;
259 
260 	if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
261 	    ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
262 	    ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
263 	    th->th_off != oth->th_off ||
264 	    (deltaS > 5 &&
265 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
266 	    (th->th_off > 5 &&
267 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
268 		goto uncompressed;
269 
270 	/*
271 	 * Figure out which of the changing fields changed.  The
272 	 * receiver expects changes in the order: urgent, window,
273 	 * ack, seq (the order minimizes the number of temporaries
274 	 * needed in this section of code).
275 	 */
276 	if (th->th_flags & TH_URG) {
277 		deltaS = ntohs(th->th_urp);
278 		ENCODEZ(deltaS);
279 		changes |= NEW_U;
280 	} else if (th->th_urp != oth->th_urp)
281 		/* argh! URG not set but urp changed -- a sensible
282 		 * implementation should never do this but RFC793
283 		 * doesn't prohibit the change so we have to deal
284 		 * with it. */
285 		 goto uncompressed;
286 
287 	deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
288 	if (deltaS) {
289 		ENCODE(deltaS);
290 		changes |= NEW_W;
291 	}
292 
293 	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
294 	if (deltaA) {
295 		if (deltaA > 0xffff)
296 			goto uncompressed;
297 		ENCODE(deltaA);
298 		changes |= NEW_A;
299 	}
300 
301 	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
302 	if (deltaS) {
303 		if (deltaS > 0xffff)
304 			goto uncompressed;
305 		ENCODE(deltaS);
306 		changes |= NEW_S;
307 	}
308 
309 	switch(changes) {
310 
311 	case 0:
312 		/*
313 		 * Nothing changed. If this packet contains data and the
314 		 * last one didn't, this is probably a data packet following
315 		 * an ack (normal on an interactive connection) and we send
316 		 * it compressed.  Otherwise it's probably a retransmit,
317 		 * retransmitted ack or window probe.  Send it uncompressed
318 		 * in case the other side missed the compressed version.
319 		 */
320 		if (ip->ip_len != cs->cs_ip.ip_len &&
321 		    ntohs(cs->cs_ip.ip_len) == hlen)
322 			break;
323 
324 		/* FALLTHROUGH */
325 
326 	case SPECIAL_I:
327 	case SPECIAL_D:
328 		/*
329 		 * actual changes match one of our special case encodings --
330 		 * send packet uncompressed.
331 		 */
332 		goto uncompressed;
333 
334 	case NEW_S|NEW_A:
335 		if (deltaS == deltaA &&
336 		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
337 			/* special case for echoed terminal traffic */
338 			changes = SPECIAL_I;
339 			cp = new_seq;
340 		}
341 		break;
342 
343 	case NEW_S:
344 		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
345 			/* special case for data xfer */
346 			changes = SPECIAL_D;
347 			cp = new_seq;
348 		}
349 		break;
350 	}
351 
352 	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
353 	if (deltaS != 1) {
354 		ENCODEZ(deltaS);
355 		changes |= NEW_I;
356 	}
357 	if (th->th_flags & TH_PUSH)
358 		changes |= TCP_PUSH_BIT;
359 	/*
360 	 * Grab the cksum before we overwrite it below.  Then update our
361 	 * state with this packet's header.
362 	 */
363 	deltaA = ntohs(th->th_sum);
364 	BCOPY(ip, &cs->cs_ip, hlen);
365 
366 	/*
367 	 * We want to use the original packet as our compressed packet.
368 	 * (cp - new_seq) is the number of bytes we need for compressed
369 	 * sequence numbers.  In addition we need one byte for the change
370 	 * mask, one for the connection id and two for the tcp checksum.
371 	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
372 	 * many bytes of the original packet to toss so subtract the two to
373 	 * get the new packet size.
374 	 */
375 	deltaS = cp - new_seq;
376 	cp = (u_char *)ip;
377 	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
378 		comp->last_xmit = cs->cs_id;
379 		hlen -= deltaS + 4;
380 		cp += hlen;
381 		*cp++ = changes | NEW_C;
382 		*cp++ = cs->cs_id;
383 	} else {
384 		hlen -= deltaS + 3;
385 		cp += hlen;
386 		*cp++ = changes;
387 	}
388 	m->m_len -= hlen;
389 	m->m_data += hlen;
390 	*cp++ = deltaA >> 8;
391 	*cp++ = deltaA;
392 	BCOPY(new_seq, cp, deltaS);
393 	INCR(sls_compressed)
394 	return (TYPE_COMPRESSED_TCP);
395 
396 	/*
397 	 * Update connection state cs & send uncompressed packet ('uncompressed'
398 	 * means a regular ip/tcp packet but with the 'conversation id' we hope
399 	 * to use on future compressed packets in the protocol field).
400 	 */
401 uncompressed:
402 	BCOPY(ip, &cs->cs_ip, hlen);
403 	ip->ip_p = cs->cs_id;
404 	comp->last_xmit = cs->cs_id;
405 	return (TYPE_UNCOMPRESSED_TCP);
406 }
407 
408 
409 int
410 sl_uncompress_tcp(u_char **bufp, int len, u_int type, struct slcompress *comp)
411 {
412 	u_char *hdr, *cp;
413 	int hlen, vjlen;
414 
415 	cp = bufp? *bufp: NULL;
416 	vjlen = sl_uncompress_tcp_core(cp, len, len, type, comp, &hdr, &hlen);
417 	if (vjlen < 0)
418 		return (0);	/* error */
419 	if (vjlen == 0)
420 		return (len);	/* was uncompressed already */
421 
422 	cp += vjlen;
423 	len -= vjlen;
424 
425 	/*
426 	 * At this point, cp points to the first byte of data in the
427 	 * packet.  If we're not aligned on a 4-byte boundary, copy the
428 	 * data down so the ip & tcp headers will be aligned.  Then back up
429 	 * cp by the tcp/ip header length to make room for the reconstructed
430 	 * header (we assume the packet we were handed has enough space to
431 	 * prepend 128 bytes of header).
432 	 */
433 	if ((intptr_t)cp & 3) {
434 		if (len > 0)
435 			BCOPY(cp, ((intptr_t)cp &~ 3), len);
436 		cp = (u_char *)((intptr_t)cp &~ 3);
437 	}
438 	cp -= hlen;
439 	len += hlen;
440 	BCOPY(hdr, cp, hlen);
441 
442 	*bufp = cp;
443 	return (len);
444 }
445 
446 /*
447  * Uncompress a packet of total length total_len.  The first buflen
448  * bytes are at buf; this must include the entire (compressed or
449  * uncompressed) TCP/IP header.  This procedure returns the length
450  * of the VJ header, with a pointer to the uncompressed IP header
451  * in *hdrp and its length in *hlenp.
452  */
453 int
454 sl_uncompress_tcp_core(u_char *buf, int buflen, int total_len, u_int type,
455     struct slcompress *comp, u_char **hdrp, u_int *hlenp)
456 {
457 	u_char *cp;
458 	u_int hlen, changes;
459 	struct tcphdr *th;
460 	struct cstate *cs;
461 	struct ip *ip;
462 	u_int16_t *bp;
463 	u_int vjlen;
464 
465 	switch (type) {
466 
467 	case TYPE_UNCOMPRESSED_TCP:
468 		ip = (struct ip *) buf;
469 		if (ip->ip_p >= MAX_STATES)
470 			goto bad;
471 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
472 		comp->flags &=~ SLF_TOSS;
473 		ip->ip_p = IPPROTO_TCP;
474 		/*
475 		 * Calculate the size of the TCP/IP header and make sure that
476 		 * we don't overflow the space we have available for it.
477 		 */
478 		hlen = ip->ip_hl << 2;
479 		if (hlen + sizeof(struct tcphdr) > buflen)
480 			goto bad;
481 		hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
482 		if (hlen > MAX_HDR || hlen > buflen)
483 			goto bad;
484 		BCOPY(ip, &cs->cs_ip, hlen);
485 		cs->cs_hlen = hlen;
486 		INCR(sls_uncompressedin)
487 		*hdrp = (u_char *) &cs->cs_ip;
488 		*hlenp = hlen;
489 		return (0);
490 
491 	default:
492 		goto bad;
493 
494 	case TYPE_COMPRESSED_TCP:
495 		break;
496 	}
497 	/* We've got a compressed packet. */
498 	INCR(sls_compressedin)
499 	cp = buf;
500 	changes = *cp++;
501 	if (changes & NEW_C) {
502 		/* Make sure the state index is in range, then grab the state.
503 		 * If we have a good state index, clear the 'discard' flag. */
504 		if (*cp >= MAX_STATES)
505 			goto bad;
506 
507 		comp->flags &=~ SLF_TOSS;
508 		comp->last_recv = *cp++;
509 	} else {
510 		/* this packet has an implicit state index.  If we've
511 		 * had a line error since the last time we got an
512 		 * explicit state index, we have to toss the packet. */
513 		if (comp->flags & SLF_TOSS) {
514 			INCR(sls_tossed)
515 			return (-1);
516 		}
517 	}
518 	cs = &comp->rstate[comp->last_recv];
519 	hlen = cs->cs_ip.ip_hl << 2;
520 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
521 	th->th_sum = htons((*cp << 8) | cp[1]);
522 	cp += 2;
523 	if (changes & TCP_PUSH_BIT)
524 		th->th_flags |= TH_PUSH;
525 	else
526 		th->th_flags &=~ TH_PUSH;
527 
528 	switch (changes & SPECIALS_MASK) {
529 	case SPECIAL_I:
530 		{
531 		u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
532 		th->th_ack = htonl(ntohl(th->th_ack) + i);
533 		th->th_seq = htonl(ntohl(th->th_seq) + i);
534 		}
535 		break;
536 
537 	case SPECIAL_D:
538 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
539 				   - cs->cs_hlen);
540 		break;
541 
542 	default:
543 		if (changes & NEW_U) {
544 			th->th_flags |= TH_URG;
545 			DECODEU(th->th_urp)
546 		} else
547 			th->th_flags &=~ TH_URG;
548 		if (changes & NEW_W)
549 			DECODES(th->th_win)
550 		if (changes & NEW_A)
551 			DECODEL(th->th_ack)
552 		if (changes & NEW_S)
553 			DECODEL(th->th_seq)
554 		break;
555 	}
556 	if (changes & NEW_I) {
557 		DECODES(cs->cs_ip.ip_id)
558 	} else
559 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
560 
561 	/*
562 	 * At this point, cp points to the first byte of data in the
563 	 * packet.  Fill in the IP total length and update the IP
564 	 * header checksum.
565 	 */
566 	vjlen = cp - buf;
567 	buflen -= vjlen;
568 	if (buflen < 0)
569 		/* we must have dropped some characters (crc should detect
570 		 * this but the old slip framing won't) */
571 		goto bad;
572 
573 	total_len += cs->cs_hlen - vjlen;
574 	cs->cs_ip.ip_len = htons(total_len);
575 
576 	/* recompute the ip header checksum */
577 	bp = (u_int16_t *) &cs->cs_ip;
578 	cs->cs_ip.ip_sum = 0;
579 		for (changes = 0; hlen > 0; hlen -= 2)
580 			changes += *bp++;
581 		changes = (changes & 0xffff) + (changes >> 16);
582 		changes = (changes & 0xffff) + (changes >> 16);
583 	cs->cs_ip.ip_sum = ~ changes;
584 
585 	*hdrp = (u_char *) &cs->cs_ip;
586 	*hlenp = cs->cs_hlen;
587 	return vjlen;
588 
589 bad:
590 	comp->flags |= SLF_TOSS;
591 	INCR(sls_errorin)
592 	return (-1);
593 }
594