xref: /freebsd/sys/net/slcompress.c (revision 1da7f3f6f72b2245e458fc7195733268ae4a1136)
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 
32 /*
33  * Routines to compress and uncompess tcp packets (for transmission
34  * over low speed serial lines.
35  *
36  * Van Jacobson (van@helios.ee.lbl.gov), Dec 31, 1989:
37  *	- Initial distribution.
38  *
39  */
40 
41 #include <sys/param.h>
42 #include <sys/mbuf.h>
43 #include <sys/systm.h>
44 
45 #include <netinet/in.h>
46 #include <netinet/in_systm.h>
47 #include <netinet/ip.h>
48 #include <netinet/tcp.h>
49 
50 #include <net/slcompress.h>
51 
52 #ifndef SL_NO_STATS
53 #define INCR(counter) ++comp->counter;
54 #else
55 #define INCR(counter)
56 #endif
57 
58 #define BCMP(p1, p2, n) bcmp((void *)(p1), (void *)(p2), (int)(n))
59 #define BCOPY(p1, p2, n) bcopy((void *)(p1), (void *)(p2), (int)(n))
60 
61 void
62 sl_compress_init(struct slcompress *comp, int max_state)
63 {
64 	u_int i;
65 	struct cstate *tstate = comp->tstate;
66 
67 	if (max_state == -1) {
68 		max_state = MAX_STATES - 1;
69 		bzero((char *)comp, sizeof(*comp));
70 	} else {
71 		/* Don't reset statistics */
72 		bzero((char *)comp->tstate, sizeof(comp->tstate));
73 		bzero((char *)comp->rstate, sizeof(comp->rstate));
74 	}
75   	for (i = max_state; i > 0; --i) {
76 		tstate[i].cs_id = i;
77 		tstate[i].cs_next = &tstate[i - 1];
78 	}
79 	tstate[0].cs_next = &tstate[max_state];
80 	tstate[0].cs_id = 0;
81 	comp->last_cs = &tstate[0];
82 	comp->last_recv = 255;
83 	comp->last_xmit = 255;
84 	comp->flags = SLF_TOSS;
85 }
86 
87 /* ENCODE encodes a number that is known to be non-zero.  ENCODEZ
88  * checks for zero (since zero has to be encoded in the long, 3 byte
89  * form).
90  */
91 #define ENCODE(n) { \
92 	if ((u_int16_t)(n) >= 256) { \
93 		*cp++ = 0; \
94 		cp[1] = (n); \
95 		cp[0] = (n) >> 8; \
96 		cp += 2; \
97 	} else { \
98 		*cp++ = (n); \
99 	} \
100 }
101 #define ENCODEZ(n) { \
102 	if ((u_int16_t)(n) >= 256 || (u_int16_t)(n) == 0) { \
103 		*cp++ = 0; \
104 		cp[1] = (n); \
105 		cp[0] = (n) >> 8; \
106 		cp += 2; \
107 	} else { \
108 		*cp++ = (n); \
109 	} \
110 }
111 
112 #define DECODEL(f) { \
113 	if (*cp == 0) {\
114 		(f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
115 		cp += 3; \
116 	} else { \
117 		(f) = htonl(ntohl(f) + (u_int32_t)*cp++); \
118 	} \
119 }
120 
121 #define DECODES(f) { \
122 	if (*cp == 0) {\
123 		(f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
124 		cp += 3; \
125 	} else { \
126 		(f) = htons(ntohs(f) + (u_int32_t)*cp++); \
127 	} \
128 }
129 
130 #define DECODEU(f) { \
131 	if (*cp == 0) {\
132 		(f) = htons((cp[1] << 8) | cp[2]); \
133 		cp += 3; \
134 	} else { \
135 		(f) = htons((u_int32_t)*cp++); \
136 	} \
137 }
138 
139 /*
140  * Attempt to compress an outgoing TCP packet and return the type of
141  * the result.  The caller must have already verified that the protocol
142  * is TCP.  The first mbuf must contain the complete IP and TCP headers,
143  * and "ip" must be == mtod(m, struct ip *).  "comp" supplies the
144  * compression state, and "compress_cid" tells us whether it is OK
145  * to leave out the CID field when feasible.
146  *
147  * The caller is responsible for adjusting m->m_pkthdr.len upon return,
148  * if m is an M_PKTHDR mbuf.
149  */
150 u_int
151 sl_compress_tcp(struct mbuf *m, struct ip *ip, struct slcompress *comp,
152     int compress_cid)
153 {
154 	struct cstate *cs = comp->last_cs->cs_next;
155 	u_int hlen = ip->ip_hl;
156 	struct tcphdr *oth;
157 	struct tcphdr *th;
158 	u_int deltaS, deltaA;
159 	u_int changes = 0;
160 	u_char new_seq[16];
161 	u_char *cp = new_seq;
162 
163 	/*
164 	 * Bail if this is an IP fragment or if the TCP packet isn't
165 	 * `compressible' (i.e., ACK isn't set or some other control bit is
166 	 * set).  (We assume that the caller has already made sure the
167 	 * packet is IP proto TCP).
168 	 */
169 	if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
170 		return (TYPE_IP);
171 
172 	th = (struct tcphdr *)&((int32_t *)ip)[hlen];
173 	if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
174 		return (TYPE_IP);
175 	/*
176 	 * Packet is compressible -- we're going to send either a
177 	 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet.  Either way we need
178 	 * to locate (or create) the connection state.  Special case the
179 	 * most recently used connection since it's most likely to be used
180 	 * again & we don't have to do any reordering if it's used.
181 	 */
182 	INCR(sls_packets)
183 	if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
184 	    ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
185 	    *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
186 		/*
187 		 * Wasn't the first -- search for it.
188 		 *
189 		 * States are kept in a circularly linked list with
190 		 * last_cs pointing to the end of the list.  The
191 		 * list is kept in lru order by moving a state to the
192 		 * head of the list whenever it is referenced.  Since
193 		 * the list is short and, empirically, the connection
194 		 * we want is almost always near the front, we locate
195 		 * states via linear search.  If we don't find a state
196 		 * for the datagram, the oldest state is (re-)used.
197 		 */
198 		struct cstate *lcs;
199 		struct cstate *lastcs = comp->last_cs;
200 
201 		do {
202 			lcs = cs; cs = cs->cs_next;
203 			INCR(sls_searches)
204 			if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
205 			    && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
206 			    && *(int32_t *)th ==
207 			    ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
208 				goto found;
209 		} while (cs != lastcs);
210 
211 		/*
212 		 * Didn't find it -- re-use oldest cstate.  Send an
213 		 * uncompressed packet that tells the other side what
214 		 * connection number we're using for this conversation.
215 		 * Note that since the state list is circular, the oldest
216 		 * state points to the newest and we only need to set
217 		 * last_cs to update the lru linkage.
218 		 */
219 		INCR(sls_misses)
220 		comp->last_cs = lcs;
221 		hlen += th->th_off;
222 		hlen <<= 2;
223 		if (hlen > m->m_len)
224 		    return TYPE_IP;
225 		goto uncompressed;
226 
227 	found:
228 		/*
229 		 * Found it -- move to the front on the connection list.
230 		 */
231 		if (cs == lastcs)
232 			comp->last_cs = lcs;
233 		else {
234 			lcs->cs_next = cs->cs_next;
235 			cs->cs_next = lastcs->cs_next;
236 			lastcs->cs_next = cs;
237 		}
238 	}
239 
240 	/*
241 	 * Make sure that only what we expect to change changed. The first
242 	 * line of the `if' checks the IP protocol version, header length &
243 	 * type of service.  The 2nd line checks the "Don't fragment" bit.
244 	 * The 3rd line checks the time-to-live and protocol (the protocol
245 	 * check is unnecessary but costless).  The 4th line checks the TCP
246 	 * header length.  The 5th line checks IP options, if any.  The 6th
247 	 * line checks TCP options, if any.  If any of these things are
248 	 * different between the previous & current datagram, we send the
249 	 * current datagram `uncompressed'.
250 	 */
251 	oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
252 	deltaS = hlen;
253 	hlen += th->th_off;
254 	hlen <<= 2;
255 	if (hlen > m->m_len)
256 	    return TYPE_IP;
257 
258 	if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
259 	    ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
260 	    ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
261 	    th->th_off != oth->th_off ||
262 	    (deltaS > 5 &&
263 	     BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
264 	    (th->th_off > 5 &&
265 	     BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
266 		goto uncompressed;
267 
268 	/*
269 	 * Figure out which of the changing fields changed.  The
270 	 * receiver expects changes in the order: urgent, window,
271 	 * ack, seq (the order minimizes the number of temporaries
272 	 * needed in this section of code).
273 	 */
274 	if (th->th_flags & TH_URG) {
275 		deltaS = ntohs(th->th_urp);
276 		ENCODEZ(deltaS);
277 		changes |= NEW_U;
278 	} else if (th->th_urp != oth->th_urp)
279 		/* argh! URG not set but urp changed -- a sensible
280 		 * implementation should never do this but RFC793
281 		 * doesn't prohibit the change so we have to deal
282 		 * with it. */
283 		 goto uncompressed;
284 
285 	deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
286 	if (deltaS) {
287 		ENCODE(deltaS);
288 		changes |= NEW_W;
289 	}
290 
291 	deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
292 	if (deltaA) {
293 		if (deltaA > 0xffff)
294 			goto uncompressed;
295 		ENCODE(deltaA);
296 		changes |= NEW_A;
297 	}
298 
299 	deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
300 	if (deltaS) {
301 		if (deltaS > 0xffff)
302 			goto uncompressed;
303 		ENCODE(deltaS);
304 		changes |= NEW_S;
305 	}
306 
307 	switch(changes) {
308 	case 0:
309 		/*
310 		 * Nothing changed. If this packet contains data and the
311 		 * last one didn't, this is probably a data packet following
312 		 * an ack (normal on an interactive connection) and we send
313 		 * it compressed.  Otherwise it's probably a retransmit,
314 		 * retransmitted ack or window probe.  Send it uncompressed
315 		 * in case the other side missed the compressed version.
316 		 */
317 		if (ip->ip_len != cs->cs_ip.ip_len &&
318 		    ntohs(cs->cs_ip.ip_len) == hlen)
319 			break;
320 
321 		/* FALLTHROUGH */
322 
323 	case SPECIAL_I:
324 	case SPECIAL_D:
325 		/*
326 		 * actual changes match one of our special case encodings --
327 		 * send packet uncompressed.
328 		 */
329 		goto uncompressed;
330 
331 	case NEW_S|NEW_A:
332 		if (deltaS == deltaA &&
333 		    deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
334 			/* special case for echoed terminal traffic */
335 			changes = SPECIAL_I;
336 			cp = new_seq;
337 		}
338 		break;
339 
340 	case NEW_S:
341 		if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
342 			/* special case for data xfer */
343 			changes = SPECIAL_D;
344 			cp = new_seq;
345 		}
346 		break;
347 	}
348 
349 	deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
350 	if (deltaS != 1) {
351 		ENCODEZ(deltaS);
352 		changes |= NEW_I;
353 	}
354 	if (th->th_flags & TH_PUSH)
355 		changes |= TCP_PUSH_BIT;
356 	/*
357 	 * Grab the cksum before we overwrite it below.  Then update our
358 	 * state with this packet's header.
359 	 */
360 	deltaA = ntohs(th->th_sum);
361 	BCOPY(ip, &cs->cs_ip, hlen);
362 
363 	/*
364 	 * We want to use the original packet as our compressed packet.
365 	 * (cp - new_seq) is the number of bytes we need for compressed
366 	 * sequence numbers.  In addition we need one byte for the change
367 	 * mask, one for the connection id and two for the tcp checksum.
368 	 * So, (cp - new_seq) + 4 bytes of header are needed.  hlen is how
369 	 * many bytes of the original packet to toss so subtract the two to
370 	 * get the new packet size.
371 	 */
372 	deltaS = cp - new_seq;
373 	cp = (u_char *)ip;
374 	if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
375 		comp->last_xmit = cs->cs_id;
376 		hlen -= deltaS + 4;
377 		cp += hlen;
378 		*cp++ = changes | NEW_C;
379 		*cp++ = cs->cs_id;
380 	} else {
381 		hlen -= deltaS + 3;
382 		cp += hlen;
383 		*cp++ = changes;
384 	}
385 	m->m_len -= hlen;
386 	m->m_data += hlen;
387 	*cp++ = deltaA >> 8;
388 	*cp++ = deltaA;
389 	BCOPY(new_seq, cp, deltaS);
390 	INCR(sls_compressed)
391 	return (TYPE_COMPRESSED_TCP);
392 
393 	/*
394 	 * Update connection state cs & send uncompressed packet ('uncompressed'
395 	 * means a regular ip/tcp packet but with the 'conversation id' we hope
396 	 * to use on future compressed packets in the protocol field).
397 	 */
398 uncompressed:
399 	BCOPY(ip, &cs->cs_ip, hlen);
400 	ip->ip_p = cs->cs_id;
401 	comp->last_xmit = cs->cs_id;
402 	return (TYPE_UNCOMPRESSED_TCP);
403 }
404 
405 int
406 sl_uncompress_tcp(u_char **bufp, int len, u_int type, struct slcompress *comp)
407 {
408 	u_char *hdr, *cp;
409 	int hlen, vjlen;
410 
411 	cp = bufp? *bufp: NULL;
412 	vjlen = sl_uncompress_tcp_core(cp, len, len, type, comp, &hdr, &hlen);
413 	if (vjlen < 0)
414 		return (0);	/* error */
415 	if (vjlen == 0)
416 		return (len);	/* was uncompressed already */
417 
418 	cp += vjlen;
419 	len -= vjlen;
420 
421 	/*
422 	 * At this point, cp points to the first byte of data in the
423 	 * packet.  If we're not aligned on a 4-byte boundary, copy the
424 	 * data down so the ip & tcp headers will be aligned.  Then back up
425 	 * cp by the tcp/ip header length to make room for the reconstructed
426 	 * header (we assume the packet we were handed has enough space to
427 	 * prepend 128 bytes of header).
428 	 */
429 	if ((intptr_t)cp & 3) {
430 		if (len > 0)
431 			BCOPY(cp, ((intptr_t)cp &~ 3), len);
432 		cp = (u_char *)((intptr_t)cp &~ 3);
433 	}
434 	cp -= hlen;
435 	len += hlen;
436 	BCOPY(hdr, cp, hlen);
437 
438 	*bufp = cp;
439 	return (len);
440 }
441 
442 /*
443  * Uncompress a packet of total length total_len.  The first buflen
444  * bytes are at buf; this must include the entire (compressed or
445  * uncompressed) TCP/IP header.  This procedure returns the length
446  * of the VJ header, with a pointer to the uncompressed IP header
447  * in *hdrp and its length in *hlenp.
448  */
449 int
450 sl_uncompress_tcp_core(u_char *buf, int buflen, int total_len, u_int type,
451     struct slcompress *comp, u_char **hdrp, u_int *hlenp)
452 {
453 	u_char *cp;
454 	u_int hlen, changes;
455 	struct tcphdr *th;
456 	struct cstate *cs;
457 	struct ip *ip;
458 	u_int16_t *bp;
459 	u_int vjlen;
460 
461 	switch (type) {
462 	case TYPE_UNCOMPRESSED_TCP:
463 		ip = (struct ip *) buf;
464 		if (ip->ip_p >= MAX_STATES)
465 			goto bad;
466 		cs = &comp->rstate[comp->last_recv = ip->ip_p];
467 		comp->flags &=~ SLF_TOSS;
468 		ip->ip_p = IPPROTO_TCP;
469 		/*
470 		 * Calculate the size of the TCP/IP header and make sure that
471 		 * we don't overflow the space we have available for it.
472 		 */
473 		hlen = ip->ip_hl << 2;
474 		if (hlen + sizeof(struct tcphdr) > buflen)
475 			goto bad;
476 		hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
477 		if (hlen > MAX_HDR || hlen > buflen)
478 			goto bad;
479 		BCOPY(ip, &cs->cs_ip, hlen);
480 		cs->cs_hlen = hlen;
481 		INCR(sls_uncompressedin)
482 		*hdrp = (u_char *) &cs->cs_ip;
483 		*hlenp = hlen;
484 		return (0);
485 
486 	default:
487 		goto bad;
488 
489 	case TYPE_COMPRESSED_TCP:
490 		break;
491 	}
492 	/* We've got a compressed packet. */
493 	INCR(sls_compressedin)
494 	cp = buf;
495 	changes = *cp++;
496 	if (changes & NEW_C) {
497 		/* Make sure the state index is in range, then grab the state.
498 		 * If we have a good state index, clear the 'discard' flag. */
499 		if (*cp >= MAX_STATES)
500 			goto bad;
501 
502 		comp->flags &=~ SLF_TOSS;
503 		comp->last_recv = *cp++;
504 	} else {
505 		/* this packet has an implicit state index.  If we've
506 		 * had a line error since the last time we got an
507 		 * explicit state index, we have to toss the packet. */
508 		if (comp->flags & SLF_TOSS) {
509 			INCR(sls_tossed)
510 			return (-1);
511 		}
512 	}
513 	cs = &comp->rstate[comp->last_recv];
514 	hlen = cs->cs_ip.ip_hl << 2;
515 	th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
516 	th->th_sum = htons((*cp << 8) | cp[1]);
517 	cp += 2;
518 	if (changes & TCP_PUSH_BIT)
519 		th->th_flags |= TH_PUSH;
520 	else
521 		th->th_flags &=~ TH_PUSH;
522 
523 	switch (changes & SPECIALS_MASK) {
524 	case SPECIAL_I:
525 		{
526 		u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
527 		th->th_ack = htonl(ntohl(th->th_ack) + i);
528 		th->th_seq = htonl(ntohl(th->th_seq) + i);
529 		}
530 		break;
531 
532 	case SPECIAL_D:
533 		th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
534 				   - cs->cs_hlen);
535 		break;
536 
537 	default:
538 		if (changes & NEW_U) {
539 			th->th_flags |= TH_URG;
540 			DECODEU(th->th_urp)
541 		} else
542 			th->th_flags &=~ TH_URG;
543 		if (changes & NEW_W)
544 			DECODES(th->th_win)
545 		if (changes & NEW_A)
546 			DECODEL(th->th_ack)
547 		if (changes & NEW_S)
548 			DECODEL(th->th_seq)
549 		break;
550 	}
551 	if (changes & NEW_I) {
552 		DECODES(cs->cs_ip.ip_id)
553 	} else
554 		cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
555 
556 	/*
557 	 * At this point, cp points to the first byte of data in the
558 	 * packet.  Fill in the IP total length and update the IP
559 	 * header checksum.
560 	 */
561 	vjlen = cp - buf;
562 	buflen -= vjlen;
563 	if (buflen < 0)
564 		/* we must have dropped some characters (crc should detect
565 		 * this but the old slip framing won't) */
566 		goto bad;
567 
568 	total_len += cs->cs_hlen - vjlen;
569 	cs->cs_ip.ip_len = htons(total_len);
570 
571 	/* recompute the ip header checksum */
572 	bp = (u_int16_t *) &cs->cs_ip;
573 	cs->cs_ip.ip_sum = 0;
574 		for (changes = 0; hlen > 0; hlen -= 2)
575 			changes += *bp++;
576 		changes = (changes & 0xffff) + (changes >> 16);
577 		changes = (changes & 0xffff) + (changes >> 16);
578 	cs->cs_ip.ip_sum = ~ changes;
579 
580 	*hdrp = (u_char *) &cs->cs_ip;
581 	*hlenp = cs->cs_hlen;
582 	return vjlen;
583 
584 bad:
585 	comp->flags |= SLF_TOSS;
586 	INCR(sls_errorin)
587 	return (-1);
588 }
589