xref: /freebsd/sys/net/slcompress.c (revision a0409676120c1e558d0ade943019934e0f15118d)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)slcompress.c	8.2 (Berkeley) 4/16/94
32  * $FreeBSD$
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 
44 #include <sys/param.h>
45 #include <sys/mbuf.h>
46 #include <sys/systm.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((void *)(p1), (void *)(p2), (int)(n))
62 #define BCOPY(p1, p2, n) bcopy((void *)(p1), (void *)(p2), (int)(n))
63 
64 void
65 sl_compress_init(struct slcompress *comp, int max_state)
66 {
67 	u_int i;
68 	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 /* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
91  * checks for zero (since zero has to be encoded in the long, 3 byte
92  * form).
93  */
94 #define ENCODE(n) { \
95 	if ((u_int16_t)(n) >= 256) { \
96 		*cp++ = 0; \
97 		cp[1] = (n); \
98 		cp[0] = (n) >> 8; \
99 		cp += 2; \
100 	} else { \
101 		*cp++ = (n); \
102 	} \
103 }
104 #define ENCODEZ(n) { \
105 	if ((u_int16_t)(n) >= 256 || (u_int16_t)(n) == 0) { \
106 		*cp++ = 0; \
107 		cp[1] = (n); \
108 		cp[0] = (n) >> 8; \
109 		cp += 2; \
110 	} else { \
111 		*cp++ = (n); \
112 	} \
113 }
114 
115 #define DECODEL(f) { \
116 	if (*cp == 0) {\
117 		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
118 		cp += 3; \
119 	} else { \
120 		(f) = htonl(ntohl(f) + (u_int32_t)*cp++); \
121 	} \
122 }
123 
124 #define DECODES(f) { \
125 	if (*cp == 0) {\
126 		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
127 		cp += 3; \
128 	} else { \
129 		(f) = htons(ntohs(f) + (u_int32_t)*cp++); \
130 	} \
131 }
132 
133 #define DECODEU(f) { \
134 	if (*cp == 0) {\
135 		(f) = htons((cp[1] << 8) | cp[2]); \
136 		cp += 3; \
137 	} else { \
138 		(f) = htons((u_int32_t)*cp++); \
139 	} \
140 }
141 
142 /*
143  * Attempt to compress an outgoing TCP packet and return the type of
144  * the result.  The caller must have already verified that the protocol
145  * is TCP.  The first mbuf must contain the complete IP and TCP headers,
146  * and "ip" must be == mtod(m, struct ip *).  "comp" supplies the
147  * compression state, and "compress_cid" tells us whether it is OK
148  * to leave out the CID field when feasible.
149  *
150  * The caller is responsible for adjusting m->m_pkthdr.len upon return,
151  * if m is an M_PKTHDR mbuf.
152  */
153 u_int
154 sl_compress_tcp(struct mbuf *m, struct ip *ip, struct slcompress *comp,
155     int compress_cid)
156 {
157 	struct cstate *cs = comp->last_cs->cs_next;
158 	u_int hlen = ip->ip_hl;
159 	struct tcphdr *oth;
160 	struct tcphdr *th;
161 	u_int deltaS, deltaA;
162 	u_int changes = 0;
163 	u_char new_seq[16];
164 	u_char *cp = new_seq;
165 
166 	/*
167 	 * Bail if this is an IP fragment or if the TCP packet isn't
168 	 * `compressible' (i.e., ACK isn't set or some other control bit is
169 	 * set).  (We assume that the caller has already made sure the
170 	 * packet is IP proto TCP).
171 	 */
172 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
173 		return (TYPE_IP);
174 
175 	th = (struct tcphdr *)&((int32_t *)ip)[hlen];
176 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
177 		return (TYPE_IP);
178 	/*
179 	 * Packet is compressible -- we're going to send either a
180 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
181 	 * to locate (or create) the connection state.  Special case the
182 	 * most recently used connection since it's most likely to be used
183 	 * again & we don't have to do any reordering if it's used.
184 	 */
185 	INCR(sls_packets)
186 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
187 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
188 	    *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
189 		/*
190 		 * Wasn't the first -- search for it.
191 		 *
192 		 * States are kept in a circularly linked list with
193 		 * last_cs pointing to the end of the list.  The
194 		 * list is kept in lru order by moving a state to the
195 		 * head of the list whenever it is referenced.  Since
196 		 * the list is short and, empirically, the connection
197 		 * we want is almost always near the front, we locate
198 		 * states via linear search.  If we don't find a state
199 		 * for the datagram, the oldest state is (re-)used.
200 		 */
201 		struct cstate *lcs;
202 		struct cstate *lastcs = comp->last_cs;
203 
204 		do {
205 			lcs = cs; cs = cs->cs_next;
206 			INCR(sls_searches)
207 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
208 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
209 			    && *(int32_t *)th ==
210 			    ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
211 				goto found;
212 		} while (cs != lastcs);
213 
214 		/*
215 		 * Didn't find it -- re-use oldest cstate.  Send an
216 		 * uncompressed packet that tells the other side what
217 		 * connection number we're using for this conversation.
218 		 * Note that since the state list is circular, the oldest
219 		 * state points to the newest and we only need to set
220 		 * last_cs to update the lru linkage.
221 		 */
222 		INCR(sls_misses)
223 		comp->last_cs = lcs;
224 		hlen += th->th_off;
225 		hlen <<= 2;
226 		if (hlen > m->m_len)
227 		    return TYPE_IP;
228 		goto uncompressed;
229 
230 	found:
231 		/*
232 		 * Found it -- move to the front on the connection list.
233 		 */
234 		if (cs == lastcs)
235 			comp->last_cs = lcs;
236 		else {
237 			lcs->cs_next = cs->cs_next;
238 			cs->cs_next = lastcs->cs_next;
239 			lastcs->cs_next = cs;
240 		}
241 	}
242 
243 	/*
244 	 * Make sure that only what we expect to change changed. The first
245 	 * line of the `if' checks the IP protocol version, header length &
246 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
247 	 * The 3rd line checks the time-to-live and protocol (the protocol
248 	 * check is unnecessary but costless).  The 4th line checks the TCP
249 	 * header length.  The 5th line checks IP options, if any.  The 6th
250 	 * line checks TCP options, if any.  If any of these things are
251 	 * different between the previous & current datagram, we send the
252 	 * current datagram `uncompressed'.
253 	 */
254 	oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
255 	deltaS = hlen;
256 	hlen += th->th_off;
257 	hlen <<= 2;
258 	if (hlen > m->m_len)
259 	    return TYPE_IP;
260 
261 	if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
262 	    ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
263 	    ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
264 	    th->th_off != oth->th_off ||
265 	    (deltaS > 5 &&
266 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
267 	    (th->th_off > 5 &&
268 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
269 		goto uncompressed;
270 
271 	/*
272 	 * Figure out which of the changing fields changed.  The
273 	 * receiver expects changes in the order: urgent, window,
274 	 * ack, seq (the order minimizes the number of temporaries
275 	 * needed in this section of code).
276 	 */
277 	if (th->th_flags & TH_URG) {
278 		deltaS = ntohs(th->th_urp);
279 		ENCODEZ(deltaS);
280 		changes |= NEW_U;
281 	} else if (th->th_urp != oth->th_urp)
282 		/* argh! URG not set but urp changed -- a sensible
283 		 * implementation should never do this but RFC793
284 		 * doesn't prohibit the change so we have to deal
285 		 * with it. */
286 		 goto uncompressed;
287 
288 	deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
289 	if (deltaS) {
290 		ENCODE(deltaS);
291 		changes |= NEW_W;
292 	}
293 
294 	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
295 	if (deltaA) {
296 		if (deltaA > 0xffff)
297 			goto uncompressed;
298 		ENCODE(deltaA);
299 		changes |= NEW_A;
300 	}
301 
302 	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
303 	if (deltaS) {
304 		if (deltaS > 0xffff)
305 			goto uncompressed;
306 		ENCODE(deltaS);
307 		changes |= NEW_S;
308 	}
309 
310 	switch(changes) {
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 int
409 sl_uncompress_tcp(u_char **bufp, int len, u_int type, struct slcompress *comp)
410 {
411 	u_char *hdr, *cp;
412 	int hlen, vjlen;
413 
414 	cp = bufp? *bufp: NULL;
415 	vjlen = sl_uncompress_tcp_core(cp, len, len, type, comp, &hdr, &hlen);
416 	if (vjlen < 0)
417 		return (0);	/* error */
418 	if (vjlen == 0)
419 		return (len);	/* was uncompressed already */
420 
421 	cp += vjlen;
422 	len -= vjlen;
423 
424 	/*
425 	 * At this point, cp points to the first byte of data in the
426 	 * packet.  If we're not aligned on a 4-byte boundary, copy the
427 	 * data down so the ip & tcp headers will be aligned.  Then back up
428 	 * cp by the tcp/ip header length to make room for the reconstructed
429 	 * header (we assume the packet we were handed has enough space to
430 	 * prepend 128 bytes of header).
431 	 */
432 	if ((intptr_t)cp & 3) {
433 		if (len > 0)
434 			BCOPY(cp, ((intptr_t)cp &~ 3), len);
435 		cp = (u_char *)((intptr_t)cp &~ 3);
436 	}
437 	cp -= hlen;
438 	len += hlen;
439 	BCOPY(hdr, cp, hlen);
440 
441 	*bufp = cp;
442 	return (len);
443 }
444 
445 /*
446  * Uncompress a packet of total length total_len.  The first buflen
447  * bytes are at buf; this must include the entire (compressed or
448  * uncompressed) TCP/IP header.  This procedure returns the length
449  * of the VJ header, with a pointer to the uncompressed IP header
450  * in *hdrp and its length in *hlenp.
451  */
452 int
453 sl_uncompress_tcp_core(u_char *buf, int buflen, int total_len, u_int type,
454     struct slcompress *comp, u_char **hdrp, u_int *hlenp)
455 {
456 	u_char *cp;
457 	u_int hlen, changes;
458 	struct tcphdr *th;
459 	struct cstate *cs;
460 	struct ip *ip;
461 	u_int16_t *bp;
462 	u_int vjlen;
463 
464 	switch (type) {
465 	case TYPE_UNCOMPRESSED_TCP:
466 		ip = (struct ip *) buf;
467 		if (ip->ip_p >= MAX_STATES)
468 			goto bad;
469 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
470 		comp->flags &=~ SLF_TOSS;
471 		ip->ip_p = IPPROTO_TCP;
472 		/*
473 		 * Calculate the size of the TCP/IP header and make sure that
474 		 * we don't overflow the space we have available for it.
475 		 */
476 		hlen = ip->ip_hl << 2;
477 		if (hlen + sizeof(struct tcphdr) > buflen)
478 			goto bad;
479 		hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
480 		if (hlen > MAX_HDR || hlen > buflen)
481 			goto bad;
482 		BCOPY(ip, &cs->cs_ip, hlen);
483 		cs->cs_hlen = hlen;
484 		INCR(sls_uncompressedin)
485 		*hdrp = (u_char *) &cs->cs_ip;
486 		*hlenp = hlen;
487 		return (0);
488 
489 	default:
490 		goto bad;
491 
492 	case TYPE_COMPRESSED_TCP:
493 		break;
494 	}
495 	/* We've got a compressed packet. */
496 	INCR(sls_compressedin)
497 	cp = buf;
498 	changes = *cp++;
499 	if (changes & NEW_C) {
500 		/* Make sure the state index is in range, then grab the state.
501 		 * If we have a good state index, clear the 'discard' flag. */
502 		if (*cp >= MAX_STATES)
503 			goto bad;
504 
505 		comp->flags &=~ SLF_TOSS;
506 		comp->last_recv = *cp++;
507 	} else {
508 		/* this packet has an implicit state index.  If we've
509 		 * had a line error since the last time we got an
510 		 * explicit state index, we have to toss the packet. */
511 		if (comp->flags & SLF_TOSS) {
512 			INCR(sls_tossed)
513 			return (-1);
514 		}
515 	}
516 	cs = &comp->rstate[comp->last_recv];
517 	hlen = cs->cs_ip.ip_hl << 2;
518 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
519 	th->th_sum = htons((*cp << 8) | cp[1]);
520 	cp += 2;
521 	if (changes & TCP_PUSH_BIT)
522 		th->th_flags |= TH_PUSH;
523 	else
524 		th->th_flags &=~ TH_PUSH;
525 
526 	switch (changes & SPECIALS_MASK) {
527 	case SPECIAL_I:
528 		{
529 		u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
530 		th->th_ack = htonl(ntohl(th->th_ack) + i);
531 		th->th_seq = htonl(ntohl(th->th_seq) + i);
532 		}
533 		break;
534 
535 	case SPECIAL_D:
536 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
537 				   - cs->cs_hlen);
538 		break;
539 
540 	default:
541 		if (changes & NEW_U) {
542 			th->th_flags |= TH_URG;
543 			DECODEU(th->th_urp)
544 		} else
545 			th->th_flags &=~ TH_URG;
546 		if (changes & NEW_W)
547 			DECODES(th->th_win)
548 		if (changes & NEW_A)
549 			DECODEL(th->th_ack)
550 		if (changes & NEW_S)
551 			DECODEL(th->th_seq)
552 		break;
553 	}
554 	if (changes & NEW_I) {
555 		DECODES(cs->cs_ip.ip_id)
556 	} else
557 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
558 
559 	/*
560 	 * At this point, cp points to the first byte of data in the
561 	 * packet.  Fill in the IP total length and update the IP
562 	 * header checksum.
563 	 */
564 	vjlen = cp - buf;
565 	buflen -= vjlen;
566 	if (buflen < 0)
567 		/* we must have dropped some characters (crc should detect
568 		 * this but the old slip framing won't) */
569 		goto bad;
570 
571 	total_len += cs->cs_hlen - vjlen;
572 	cs->cs_ip.ip_len = htons(total_len);
573 
574 	/* recompute the ip header checksum */
575 	bp = (u_int16_t *) &cs->cs_ip;
576 	cs->cs_ip.ip_sum = 0;
577 		for (changes = 0; hlen > 0; hlen -= 2)
578 			changes += *bp++;
579 		changes = (changes & 0xffff) + (changes >> 16);
580 		changes = (changes & 0xffff) + (changes >> 16);
581 	cs->cs_ip.ip_sum = ~ changes;
582 
583 	*hdrp = (u_char *) &cs->cs_ip;
584 	*hlenp = cs->cs_hlen;
585 	return vjlen;
586 
587 bad:
588 	comp->flags |= SLF_TOSS;
589 	INCR(sls_errorin)
590 	return (-1);
591 }
592