xref: /freebsd/sys/net/slcompress.c (revision 076b94438c7d42c1b4661ed1e12e3b12ca69361a)
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 
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(struct mbuf *m, struct ip *ip, struct slcompress *comp,
156     int compress_cid)
157 {
158 	struct cstate *cs = comp->last_cs->cs_next;
159 	u_int hlen = ip->ip_hl;
160 	struct tcphdr *oth;
161 	struct tcphdr *th;
162 	u_int deltaS, deltaA;
163 	u_int changes = 0;
164 	u_char new_seq[16];
165 	u_char *cp = new_seq;
166 
167 	/*
168 	 * Bail if this is an IP fragment or if the TCP packet isn't
169 	 * `compressible' (i.e., ACK isn't set or some other control bit is
170 	 * set).  (We assume that the caller has already made sure the
171 	 * packet is IP proto TCP).
172 	 */
173 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
174 		return (TYPE_IP);
175 
176 	th = (struct tcphdr *)&((int32_t *)ip)[hlen];
177 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
178 		return (TYPE_IP);
179 	/*
180 	 * Packet is compressible -- we're going to send either a
181 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
182 	 * to locate (or create) the connection state.  Special case the
183 	 * most recently used connection since it's most likely to be used
184 	 * again & we don't have to do any reordering if it's used.
185 	 */
186 	INCR(sls_packets)
187 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
188 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
189 	    *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
190 		/*
191 		 * Wasn't the first -- search for it.
192 		 *
193 		 * States are kept in a circularly linked list with
194 		 * last_cs pointing to the end of the list.  The
195 		 * list is kept in lru order by moving a state to the
196 		 * head of the list whenever it is referenced.  Since
197 		 * the list is short and, empirically, the connection
198 		 * we want is almost always near the front, we locate
199 		 * states via linear search.  If we don't find a state
200 		 * for the datagram, the oldest state is (re-)used.
201 		 */
202 		struct cstate *lcs;
203 		struct cstate *lastcs = comp->last_cs;
204 
205 		do {
206 			lcs = cs; cs = cs->cs_next;
207 			INCR(sls_searches)
208 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
209 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
210 			    && *(int32_t *)th ==
211 			    ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
212 				goto found;
213 		} while (cs != lastcs);
214 
215 		/*
216 		 * Didn't find it -- re-use oldest cstate.  Send an
217 		 * uncompressed packet that tells the other side what
218 		 * connection number we're using for this conversation.
219 		 * Note that since the state list is circular, the oldest
220 		 * state points to the newest and we only need to set
221 		 * last_cs to update the lru linkage.
222 		 */
223 		INCR(sls_misses)
224 		comp->last_cs = lcs;
225 		hlen += th->th_off;
226 		hlen <<= 2;
227 		if (hlen > m->m_len)
228 		    return TYPE_IP;
229 		goto uncompressed;
230 
231 	found:
232 		/*
233 		 * Found it -- move to the front on the connection list.
234 		 */
235 		if (cs == lastcs)
236 			comp->last_cs = lcs;
237 		else {
238 			lcs->cs_next = cs->cs_next;
239 			cs->cs_next = lastcs->cs_next;
240 			lastcs->cs_next = cs;
241 		}
242 	}
243 
244 	/*
245 	 * Make sure that only what we expect to change changed. The first
246 	 * line of the `if' checks the IP protocol version, header length &
247 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
248 	 * The 3rd line checks the time-to-live and protocol (the protocol
249 	 * check is unnecessary but costless).  The 4th line checks the TCP
250 	 * header length.  The 5th line checks IP options, if any.  The 6th
251 	 * line checks TCP options, if any.  If any of these things are
252 	 * different between the previous & current datagram, we send the
253 	 * current datagram `uncompressed'.
254 	 */
255 	oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
256 	deltaS = hlen;
257 	hlen += th->th_off;
258 	hlen <<= 2;
259 	if (hlen > m->m_len)
260 	    return TYPE_IP;
261 
262 	if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
263 	    ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
264 	    ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
265 	    th->th_off != oth->th_off ||
266 	    (deltaS > 5 &&
267 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
268 	    (th->th_off > 5 &&
269 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
270 		goto uncompressed;
271 
272 	/*
273 	 * Figure out which of the changing fields changed.  The
274 	 * receiver expects changes in the order: urgent, window,
275 	 * ack, seq (the order minimizes the number of temporaries
276 	 * needed in this section of code).
277 	 */
278 	if (th->th_flags & TH_URG) {
279 		deltaS = ntohs(th->th_urp);
280 		ENCODEZ(deltaS);
281 		changes |= NEW_U;
282 	} else if (th->th_urp != oth->th_urp)
283 		/* argh! URG not set but urp changed -- a sensible
284 		 * implementation should never do this but RFC793
285 		 * doesn't prohibit the change so we have to deal
286 		 * with it. */
287 		 goto uncompressed;
288 
289 	deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
290 	if (deltaS) {
291 		ENCODE(deltaS);
292 		changes |= NEW_W;
293 	}
294 
295 	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
296 	if (deltaA) {
297 		if (deltaA > 0xffff)
298 			goto uncompressed;
299 		ENCODE(deltaA);
300 		changes |= NEW_A;
301 	}
302 
303 	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
304 	if (deltaS) {
305 		if (deltaS > 0xffff)
306 			goto uncompressed;
307 		ENCODE(deltaS);
308 		changes |= NEW_S;
309 	}
310 
311 	switch(changes) {
312 
313 	case 0:
314 		/*
315 		 * Nothing changed. If this packet contains data and the
316 		 * last one didn't, this is probably a data packet following
317 		 * an ack (normal on an interactive connection) and we send
318 		 * it compressed.  Otherwise it's probably a retransmit,
319 		 * retransmitted ack or window probe.  Send it uncompressed
320 		 * in case the other side missed the compressed version.
321 		 */
322 		if (ip->ip_len != cs->cs_ip.ip_len &&
323 		    ntohs(cs->cs_ip.ip_len) == hlen)
324 			break;
325 
326 		/* FALLTHROUGH */
327 
328 	case SPECIAL_I:
329 	case SPECIAL_D:
330 		/*
331 		 * actual changes match one of our special case encodings --
332 		 * send packet uncompressed.
333 		 */
334 		goto uncompressed;
335 
336 	case NEW_S|NEW_A:
337 		if (deltaS == deltaA &&
338 		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
339 			/* special case for echoed terminal traffic */
340 			changes = SPECIAL_I;
341 			cp = new_seq;
342 		}
343 		break;
344 
345 	case NEW_S:
346 		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
347 			/* special case for data xfer */
348 			changes = SPECIAL_D;
349 			cp = new_seq;
350 		}
351 		break;
352 	}
353 
354 	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
355 	if (deltaS != 1) {
356 		ENCODEZ(deltaS);
357 		changes |= NEW_I;
358 	}
359 	if (th->th_flags & TH_PUSH)
360 		changes |= TCP_PUSH_BIT;
361 	/*
362 	 * Grab the cksum before we overwrite it below.  Then update our
363 	 * state with this packet's header.
364 	 */
365 	deltaA = ntohs(th->th_sum);
366 	BCOPY(ip, &cs->cs_ip, hlen);
367 
368 	/*
369 	 * We want to use the original packet as our compressed packet.
370 	 * (cp - new_seq) is the number of bytes we need for compressed
371 	 * sequence numbers.  In addition we need one byte for the change
372 	 * mask, one for the connection id and two for the tcp checksum.
373 	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
374 	 * many bytes of the original packet to toss so subtract the two to
375 	 * get the new packet size.
376 	 */
377 	deltaS = cp - new_seq;
378 	cp = (u_char *)ip;
379 	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
380 		comp->last_xmit = cs->cs_id;
381 		hlen -= deltaS + 4;
382 		cp += hlen;
383 		*cp++ = changes | NEW_C;
384 		*cp++ = cs->cs_id;
385 	} else {
386 		hlen -= deltaS + 3;
387 		cp += hlen;
388 		*cp++ = changes;
389 	}
390 	m->m_len -= hlen;
391 	m->m_data += hlen;
392 	*cp++ = deltaA >> 8;
393 	*cp++ = deltaA;
394 	BCOPY(new_seq, cp, deltaS);
395 	INCR(sls_compressed)
396 	return (TYPE_COMPRESSED_TCP);
397 
398 	/*
399 	 * Update connection state cs & send uncompressed packet ('uncompressed'
400 	 * means a regular ip/tcp packet but with the 'conversation id' we hope
401 	 * to use on future compressed packets in the protocol field).
402 	 */
403 uncompressed:
404 	BCOPY(ip, &cs->cs_ip, hlen);
405 	ip->ip_p = cs->cs_id;
406 	comp->last_xmit = cs->cs_id;
407 	return (TYPE_UNCOMPRESSED_TCP);
408 }
409 
410 
411 int
412 sl_uncompress_tcp(u_char **bufp, int len, u_int type, struct slcompress *comp)
413 {
414 	u_char *hdr, *cp;
415 	int hlen, vjlen;
416 
417 	cp = bufp? *bufp: NULL;
418 	vjlen = sl_uncompress_tcp_core(cp, len, len, type, comp, &hdr, &hlen);
419 	if (vjlen < 0)
420 		return (0);	/* error */
421 	if (vjlen == 0)
422 		return (len);	/* was uncompressed already */
423 
424 	cp += vjlen;
425 	len -= vjlen;
426 
427 	/*
428 	 * At this point, cp points to the first byte of data in the
429 	 * packet.  If we're not aligned on a 4-byte boundary, copy the
430 	 * data down so the ip & tcp headers will be aligned.  Then back up
431 	 * cp by the tcp/ip header length to make room for the reconstructed
432 	 * header (we assume the packet we were handed has enough space to
433 	 * prepend 128 bytes of header).
434 	 */
435 	if ((intptr_t)cp & 3) {
436 		if (len > 0)
437 			BCOPY(cp, ((intptr_t)cp &~ 3), len);
438 		cp = (u_char *)((intptr_t)cp &~ 3);
439 	}
440 	cp -= hlen;
441 	len += hlen;
442 	BCOPY(hdr, cp, hlen);
443 
444 	*bufp = cp;
445 	return (len);
446 }
447 
448 /*
449  * Uncompress a packet of total length total_len.  The first buflen
450  * bytes are at buf; this must include the entire (compressed or
451  * uncompressed) TCP/IP header.  This procedure returns the length
452  * of the VJ header, with a pointer to the uncompressed IP header
453  * in *hdrp and its length in *hlenp.
454  */
455 int
456 sl_uncompress_tcp_core(u_char *buf, int buflen, int total_len, u_int type,
457     struct slcompress *comp, u_char **hdrp, u_int *hlenp)
458 {
459 	u_char *cp;
460 	u_int hlen, changes;
461 	struct tcphdr *th;
462 	struct cstate *cs;
463 	struct ip *ip;
464 	u_int16_t *bp;
465 	u_int vjlen;
466 
467 	switch (type) {
468 
469 	case TYPE_UNCOMPRESSED_TCP:
470 		ip = (struct ip *) buf;
471 		if (ip->ip_p >= MAX_STATES)
472 			goto bad;
473 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
474 		comp->flags &=~ SLF_TOSS;
475 		ip->ip_p = IPPROTO_TCP;
476 		/*
477 		 * Calculate the size of the TCP/IP header and make sure that
478 		 * we don't overflow the space we have available for it.
479 		 */
480 		hlen = ip->ip_hl << 2;
481 		if (hlen + sizeof(struct tcphdr) > buflen)
482 			goto bad;
483 		hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
484 		if (hlen > MAX_HDR || hlen > buflen)
485 			goto bad;
486 		BCOPY(ip, &cs->cs_ip, hlen);
487 		cs->cs_hlen = hlen;
488 		INCR(sls_uncompressedin)
489 		*hdrp = (u_char *) &cs->cs_ip;
490 		*hlenp = hlen;
491 		return (0);
492 
493 	default:
494 		goto bad;
495 
496 	case TYPE_COMPRESSED_TCP:
497 		break;
498 	}
499 	/* We've got a compressed packet. */
500 	INCR(sls_compressedin)
501 	cp = buf;
502 	changes = *cp++;
503 	if (changes & NEW_C) {
504 		/* Make sure the state index is in range, then grab the state.
505 		 * If we have a good state index, clear the 'discard' flag. */
506 		if (*cp >= MAX_STATES)
507 			goto bad;
508 
509 		comp->flags &=~ SLF_TOSS;
510 		comp->last_recv = *cp++;
511 	} else {
512 		/* this packet has an implicit state index.  If we've
513 		 * had a line error since the last time we got an
514 		 * explicit state index, we have to toss the packet. */
515 		if (comp->flags & SLF_TOSS) {
516 			INCR(sls_tossed)
517 			return (-1);
518 		}
519 	}
520 	cs = &comp->rstate[comp->last_recv];
521 	hlen = cs->cs_ip.ip_hl << 2;
522 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
523 	th->th_sum = htons((*cp << 8) | cp[1]);
524 	cp += 2;
525 	if (changes & TCP_PUSH_BIT)
526 		th->th_flags |= TH_PUSH;
527 	else
528 		th->th_flags &=~ TH_PUSH;
529 
530 	switch (changes & SPECIALS_MASK) {
531 	case SPECIAL_I:
532 		{
533 		u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
534 		th->th_ack = htonl(ntohl(th->th_ack) + i);
535 		th->th_seq = htonl(ntohl(th->th_seq) + i);
536 		}
537 		break;
538 
539 	case SPECIAL_D:
540 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
541 				   - cs->cs_hlen);
542 		break;
543 
544 	default:
545 		if (changes & NEW_U) {
546 			th->th_flags |= TH_URG;
547 			DECODEU(th->th_urp)
548 		} else
549 			th->th_flags &=~ TH_URG;
550 		if (changes & NEW_W)
551 			DECODES(th->th_win)
552 		if (changes & NEW_A)
553 			DECODEL(th->th_ack)
554 		if (changes & NEW_S)
555 			DECODEL(th->th_seq)
556 		break;
557 	}
558 	if (changes & NEW_I) {
559 		DECODES(cs->cs_ip.ip_id)
560 	} else
561 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
562 
563 	/*
564 	 * At this point, cp points to the first byte of data in the
565 	 * packet.  Fill in the IP total length and update the IP
566 	 * header checksum.
567 	 */
568 	vjlen = cp - buf;
569 	buflen -= vjlen;
570 	if (buflen < 0)
571 		/* we must have dropped some characters (crc should detect
572 		 * this but the old slip framing won't) */
573 		goto bad;
574 
575 	total_len += cs->cs_hlen - vjlen;
576 	cs->cs_ip.ip_len = htons(total_len);
577 
578 	/* recompute the ip header checksum */
579 	bp = (u_int16_t *) &cs->cs_ip;
580 	cs->cs_ip.ip_sum = 0;
581 		for (changes = 0; hlen > 0; hlen -= 2)
582 			changes += *bp++;
583 		changes = (changes & 0xffff) + (changes >> 16);
584 		changes = (changes & 0xffff) + (changes >> 16);
585 	cs->cs_ip.ip_sum = ~ changes;
586 
587 	*hdrp = (u_char *) &cs->cs_ip;
588 	*hlenp = cs->cs_hlen;
589 	return vjlen;
590 
591 bad:
592 	comp->flags |= SLF_TOSS;
593 	INCR(sls_errorin)
594 	return (-1);
595 }
596