1 /* $OpenBSD: slcompress.c,v 1.9 2007/02/14 00:53:48 jsg Exp $ */
2 /* $NetBSD: slcompress.c,v 1.17 1997/05/17 21:12:10 christos Exp $ */
3
4 /*
5 * Copyright (c) 1989, 1993, 1994
6 * The Regents of the University of California. All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)slcompress.c 8.2 (Berkeley) 4/16/94
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 #include <sys/param.h>
44 #include <sys/mbuf.h>
45 #include <sys/systm.h>
46
47 #include <netinet/in.h>
48 #include <netinet/in_systm.h>
49 #include <netinet/ip.h>
50 #include <netinet/tcp.h>
51
52 #include <net/slcompress.h>
53
54 #ifndef SL_NO_STATS
55 #define INCR(counter) ++comp->counter;
56 #else
57 #define INCR(counter)
58 #endif
59
60 #define BCMP(p1, p2, n) bcmp((char *)(p1), (char *)(p2), (int)(n))
61 #define BCOPY(p1, p2, n) bcopy((char *)(p1), (char *)(p2), (int)(n))
62 #ifndef _KERNEL
63 #define ovbcopy bcopy
64 #endif
65
66
67 void
68 sl_compress_init(comp)
69 struct slcompress *comp;
70 {
71 u_int i;
72 struct cstate *tstate = comp->tstate;
73
74 bzero((char *)comp, sizeof(*comp));
75 for (i = MAX_STATES - 1; 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_STATES - 1];
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
88 /*
89 * Like sl_compress_init, but we get to specify the maximum connection
90 * ID to use on transmission.
91 */
92 void
93 sl_compress_setup(comp, max_state)
94 struct slcompress *comp;
95 int max_state;
96 {
97 u_int i;
98 struct cstate *tstate = comp->tstate;
99
100 if (max_state == -1) {
101 max_state = MAX_STATES - 1;
102 bzero((char *)comp, sizeof(*comp));
103 } else {
104 /* Don't reset statistics */
105 bzero((char *)comp->tstate, sizeof(comp->tstate));
106 bzero((char *)comp->rstate, sizeof(comp->rstate));
107 }
108 for (i = max_state; i > 0; --i) {
109 tstate[i].cs_id = i;
110 tstate[i].cs_next = &tstate[i - 1];
111 }
112 tstate[0].cs_next = &tstate[max_state];
113 tstate[0].cs_id = 0;
114 comp->last_cs = &tstate[0];
115 comp->last_recv = 255;
116 comp->last_xmit = 255;
117 comp->flags = SLF_TOSS;
118 }
119
120
121 /* ENCODE encodes a number that is known to be non-zero. ENCODEZ
122 * checks for zero (since zero has to be encoded in the long, 3 byte
123 * form).
124 */
125 #define ENCODE(n) { \
126 if ((u_int16_t)(n) >= 256) { \
127 *cp++ = 0; \
128 cp[1] = (n); \
129 cp[0] = (n) >> 8; \
130 cp += 2; \
131 } else { \
132 *cp++ = (n); \
133 } \
134 }
135 #define ENCODEZ(n) { \
136 if ((u_int16_t)(n) >= 256 || (u_int16_t)(n) == 0) { \
137 *cp++ = 0; \
138 cp[1] = (n); \
139 cp[0] = (n) >> 8; \
140 cp += 2; \
141 } else { \
142 *cp++ = (n); \
143 } \
144 }
145
146 #define DECODEL(f) { \
147 if (*cp == 0) {\
148 (f) = htonl(ntohl(f) + ((cp[1] << 8) | cp[2])); \
149 cp += 3; \
150 } else { \
151 (f) = htonl(ntohl(f) + (u_int32_t)*cp++); \
152 } \
153 }
154
155 #define DECODES(f) { \
156 if (*cp == 0) {\
157 (f) = htons(ntohs(f) + ((cp[1] << 8) | cp[2])); \
158 cp += 3; \
159 } else { \
160 (f) = htons(ntohs(f) + (u_int32_t)*cp++); \
161 } \
162 }
163
164 #define DECODEU(f) { \
165 if (*cp == 0) {\
166 (f) = htons((cp[1] << 8) | cp[2]); \
167 cp += 3; \
168 } else { \
169 (f) = htons((u_int32_t)*cp++); \
170 } \
171 }
172
173 u_int
174 sl_compress_tcp(m, ip, comp, compress_cid)
175 struct mbuf *m;
176 struct ip *ip;
177 struct slcompress *comp;
178 int compress_cid;
179 {
180 struct cstate *cs = comp->last_cs->cs_next;
181 u_int hlen = ip->ip_hl;
182 struct tcphdr *oth;
183 struct tcphdr *th;
184 u_int deltaS, deltaA;
185 u_int changes = 0;
186 u_char new_seq[16];
187 u_char *cp = new_seq;
188
189 /*
190 * Bail if this is an IP fragment or if the TCP packet isn't
191 * `compressible' (i.e., ACK isn't set or some other control bit is
192 * set). (We assume that the caller has already made sure the
193 * packet is IP proto TCP).
194 */
195 if ((ip->ip_off & htons(0x3fff)) || m->m_len < 40)
196 return (TYPE_IP);
197
198 th = (struct tcphdr *)&((int32_t *)ip)[hlen];
199 if ((th->th_flags & (TH_SYN|TH_FIN|TH_RST|TH_ACK)) != TH_ACK)
200 return (TYPE_IP);
201 /*
202 * Packet is compressible -- we're going to send either a
203 * COMPRESSED_TCP or UNCOMPRESSED_TCP packet. Either way we need
204 * to locate (or create) the connection state. Special case the
205 * most recently used connection since it's most likely to be used
206 * again & we don't have to do any reordering if it's used.
207 */
208 INCR(sls_packets)
209 if (ip->ip_src.s_addr != cs->cs_ip.ip_src.s_addr ||
210 ip->ip_dst.s_addr != cs->cs_ip.ip_dst.s_addr ||
211 *(int32_t *)th != ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl]) {
212 /*
213 * Wasn't the first -- search for it.
214 *
215 * States are kept in a circularly linked list with
216 * last_cs pointing to the end of the list. The
217 * list is kept in lru order by moving a state to the
218 * head of the list whenever it is referenced. Since
219 * the list is short and, empirically, the connection
220 * we want is almost always near the front, we locate
221 * states via linear search. If we don't find a state
222 * for the datagram, the oldest state is (re-)used.
223 */
224 struct cstate *lcs;
225 struct cstate *lastcs = comp->last_cs;
226
227 do {
228 lcs = cs; cs = cs->cs_next;
229 INCR(sls_searches)
230 if (ip->ip_src.s_addr == cs->cs_ip.ip_src.s_addr
231 && ip->ip_dst.s_addr == cs->cs_ip.ip_dst.s_addr
232 && *(int32_t *)th ==
233 ((int32_t *)&cs->cs_ip)[cs->cs_ip.ip_hl])
234 goto found;
235 } while (cs != lastcs);
236
237 /*
238 * Didn't find it -- re-use oldest cstate. Send an
239 * uncompressed packet that tells the other side what
240 * connection number we're using for this conversation.
241 * Note that since the state list is circular, the oldest
242 * state points to the newest and we only need to set
243 * last_cs to update the lru linkage.
244 */
245 INCR(sls_misses)
246 comp->last_cs = lcs;
247 hlen += th->th_off;
248 hlen <<= 2;
249 goto uncompressed;
250
251 found:
252 /*
253 * Found it -- move to the front on the connection list.
254 */
255 if (cs == lastcs)
256 comp->last_cs = lcs;
257 else {
258 lcs->cs_next = cs->cs_next;
259 cs->cs_next = lastcs->cs_next;
260 lastcs->cs_next = cs;
261 }
262 }
263
264 /*
265 * Make sure that only what we expect to change changed. The first
266 * line of the `if' checks the IP protocol version, header length &
267 * type of service. The 2nd line checks the "Don't fragment" bit.
268 * The 3rd line checks the time-to-live and protocol (the protocol
269 * check is unnecessary but costless). The 4th line checks the TCP
270 * header length. The 5th line checks IP options, if any. The 6th
271 * line checks TCP options, if any. If any of these things are
272 * different between the previous & current datagram, we send the
273 * current datagram `uncompressed'.
274 */
275 oth = (struct tcphdr *)&((int32_t *)&cs->cs_ip)[hlen];
276 deltaS = hlen;
277 hlen += th->th_off;
278 hlen <<= 2;
279
280 if (((u_int16_t *)ip)[0] != ((u_int16_t *)&cs->cs_ip)[0] ||
281 ((u_int16_t *)ip)[3] != ((u_int16_t *)&cs->cs_ip)[3] ||
282 ((u_int16_t *)ip)[4] != ((u_int16_t *)&cs->cs_ip)[4] ||
283 th->th_off != oth->th_off ||
284 (deltaS > 5 &&
285 BCMP(ip + 1, &cs->cs_ip + 1, (deltaS - 5) << 2)) ||
286 (th->th_off > 5 &&
287 BCMP(th + 1, oth + 1, (th->th_off - 5) << 2)))
288 goto uncompressed;
289
290 /*
291 * Figure out which of the changing fields changed. The
292 * receiver expects changes in the order: urgent, window,
293 * ack, seq (the order minimizes the number of temporaries
294 * needed in this section of code).
295 */
296 if (th->th_flags & TH_URG) {
297 deltaS = ntohs(th->th_urp);
298 ENCODEZ(deltaS);
299 changes |= NEW_U;
300 } else if (th->th_urp != oth->th_urp)
301 /* argh! URG not set but urp changed -- a sensible
302 * implementation should never do this but RFC793
303 * doesn't prohibit the change so we have to deal
304 * with it. */
305 goto uncompressed;
306
307 deltaS = (u_int16_t)(ntohs(th->th_win) - ntohs(oth->th_win));
308 if (deltaS) {
309 ENCODE(deltaS);
310 changes |= NEW_W;
311 }
312
313 deltaA = ntohl(th->th_ack) - ntohl(oth->th_ack);
314 if (deltaA) {
315 if (deltaA > 0xffff)
316 goto uncompressed;
317 ENCODE(deltaA);
318 changes |= NEW_A;
319 }
320
321 deltaS = ntohl(th->th_seq) - ntohl(oth->th_seq);
322 if (deltaS) {
323 if (deltaS > 0xffff)
324 goto uncompressed;
325 ENCODE(deltaS);
326 changes |= NEW_S;
327 }
328
329 switch(changes) {
330
331 case 0:
332 /*
333 * Nothing changed. If this packet contains data and the
334 * last one didn't, this is probably a data packet following
335 * an ack (normal on an interactive connection) and we send
336 * it compressed. Otherwise it's probably a retransmit,
337 * retransmitted ack or window probe. Send it uncompressed
338 * in case the other side missed the compressed version.
339 */
340 if (ip->ip_len != cs->cs_ip.ip_len &&
341 ntohs(cs->cs_ip.ip_len) == hlen)
342 break;
343
344 /* FALLTHROUGH */
345
346 case SPECIAL_I:
347 case SPECIAL_D:
348 /*
349 * actual changes match one of our special case encodings --
350 * send packet uncompressed.
351 */
352 goto uncompressed;
353
354 case NEW_S|NEW_A:
355 if (deltaS == deltaA &&
356 deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
357 /* special case for echoed terminal traffic */
358 changes = SPECIAL_I;
359 cp = new_seq;
360 }
361 break;
362
363 case NEW_S:
364 if (deltaS == ntohs(cs->cs_ip.ip_len) - hlen) {
365 /* special case for data xfer */
366 changes = SPECIAL_D;
367 cp = new_seq;
368 }
369 break;
370 }
371
372 deltaS = ntohs(ip->ip_id) - ntohs(cs->cs_ip.ip_id);
373 if (deltaS != 1) {
374 ENCODEZ(deltaS);
375 changes |= NEW_I;
376 }
377 if (th->th_flags & TH_PUSH)
378 changes |= TCP_PUSH_BIT;
379 /*
380 * Grab the cksum before we overwrite it below. Then update our
381 * state with this packet's header.
382 */
383 deltaA = ntohs(th->th_sum);
384 BCOPY(ip, &cs->cs_ip, hlen);
385
386 /*
387 * We want to use the original packet as our compressed packet.
388 * (cp - new_seq) is the number of bytes we need for compressed
389 * sequence numbers. In addition we need one byte for the change
390 * mask, one for the connection id and two for the tcp checksum.
391 * So, (cp - new_seq) + 4 bytes of header are needed. hlen is how
392 * many bytes of the original packet to toss so subtract the two to
393 * get the new packet size.
394 */
395 deltaS = cp - new_seq;
396 cp = (u_char *)ip;
397 if (compress_cid == 0 || comp->last_xmit != cs->cs_id) {
398 comp->last_xmit = cs->cs_id;
399 hlen -= deltaS + 4;
400 cp += hlen;
401 *cp++ = changes | NEW_C;
402 *cp++ = cs->cs_id;
403 } else {
404 hlen -= deltaS + 3;
405 cp += hlen;
406 *cp++ = changes;
407 }
408 m->m_len -= hlen;
409 m->m_data += hlen;
410 *cp++ = deltaA >> 8;
411 *cp++ = deltaA;
412 BCOPY(new_seq, cp, deltaS);
413 INCR(sls_compressed)
414 return (TYPE_COMPRESSED_TCP);
415
416 /*
417 * Update connection state cs & send uncompressed packet ('uncompressed'
418 * means a regular ip/tcp packet but with the 'conversation id' we hope
419 * to use on future compressed packets in the protocol field).
420 */
421 uncompressed:
422 BCOPY(ip, &cs->cs_ip, hlen);
423 ip->ip_p = cs->cs_id;
424 comp->last_xmit = cs->cs_id;
425 return (TYPE_UNCOMPRESSED_TCP);
426 }
427
428
429 int
430 sl_uncompress_tcp(bufp, len, type, comp)
431 u_char **bufp;
432 int len;
433 u_int type;
434 struct slcompress *comp;
435 {
436 u_char *hdr, *cp;
437 int hlen, vjlen;
438
439 cp = bufp? *bufp: NULL;
440 vjlen = sl_uncompress_tcp_core(cp, len, len, type, comp, &hdr, &hlen);
441 if (vjlen < 0)
442 return (0); /* error */
443 if (vjlen == 0)
444 return (len); /* was uncompressed already */
445
446 cp += vjlen;
447 len -= vjlen;
448
449 /*
450 * At this point, cp points to the first byte of data in the
451 * packet. If we're not aligned on a 4-byte boundary, copy the
452 * data down so the ip & tcp headers will be aligned. Then back up
453 * cp by the tcp/ip header length to make room for the reconstructed
454 * header (we assume the packet we were handed has enough space to
455 * prepend 128 bytes of header).
456 */
457 if ((long)cp & 3) {
458 if (len > 0)
459 (void) ovbcopy(cp, (caddr_t)((long)cp &~ 3), len);
460 cp = (u_char *)((long)cp &~ 3);
461 }
462 cp -= hlen;
463 len += hlen;
464 BCOPY(hdr, cp, hlen);
465
466 *bufp = cp;
467 return (len);
468 }
469
470 /*
471 * Uncompress a packet of total length total_len. The first buflen
472 * bytes are at buf; this must include the entire (compressed or
473 * uncompressed) TCP/IP header. This procedure returns the length
474 * of the VJ header, with a pointer to the uncompressed IP header
475 * in *hdrp and its length in *hlenp.
476 */
477 int
478 sl_uncompress_tcp_core(buf, buflen, total_len, type, comp, hdrp, hlenp)
479 u_char *buf;
480 int buflen, total_len;
481 u_int type;
482 struct slcompress *comp;
483 u_char **hdrp;
484 u_int *hlenp;
485 {
486 u_char *cp;
487 u_int hlen, changes;
488 struct tcphdr *th;
489 struct cstate *cs;
490 struct ip *ip;
491 u_int16_t *bp;
492 u_int vjlen;
493
494 switch (type) {
495
496 case TYPE_UNCOMPRESSED_TCP:
497 ip = (struct ip *) buf;
498 if (ip->ip_p >= MAX_STATES)
499 goto bad;
500 cs = &comp->rstate[comp->last_recv = ip->ip_p];
501 comp->flags &=~ SLF_TOSS;
502 ip->ip_p = IPPROTO_TCP;
503 /*
504 * Calculate the size of the TCP/IP header and make sure that
505 * we don't overflow the space we have available for it.
506 */
507 hlen = ip->ip_hl << 2;
508 if (hlen + sizeof(struct tcphdr) > buflen)
509 goto bad;
510 hlen += ((struct tcphdr *)&((char *)ip)[hlen])->th_off << 2;
511 if (hlen > MAX_HDR || hlen > buflen)
512 goto bad;
513 BCOPY(ip, &cs->cs_ip, hlen);
514 cs->cs_hlen = hlen;
515 INCR(sls_uncompressedin)
516 *hdrp = (u_char *) &cs->cs_ip;
517 *hlenp = hlen;
518 return (0);
519
520 default:
521 goto bad;
522
523 case TYPE_COMPRESSED_TCP:
524 break;
525 }
526 /* We've got a compressed packet. */
527 INCR(sls_compressedin)
528 cp = buf;
529 changes = *cp++;
530 if (changes & NEW_C) {
531 /* Make sure the state index is in range, then grab the state.
532 * If we have a good state index, clear the 'discard' flag. */
533 if (*cp >= MAX_STATES)
534 goto bad;
535
536 comp->flags &=~ SLF_TOSS;
537 comp->last_recv = *cp++;
538 } else {
539 /* this packet has an implicit state index. If we've
540 * had a line error since the last time we got an
541 * explicit state index, we have to toss the packet. */
542 if (comp->flags & SLF_TOSS) {
543 INCR(sls_tossed)
544 return (-1);
545 }
546 }
547 cs = &comp->rstate[comp->last_recv];
548 hlen = cs->cs_ip.ip_hl << 2;
549 th = (struct tcphdr *)&((u_char *)&cs->cs_ip)[hlen];
550 th->th_sum = htons((*cp << 8) | cp[1]);
551 cp += 2;
552 if (changes & TCP_PUSH_BIT)
553 th->th_flags |= TH_PUSH;
554 else
555 th->th_flags &=~ TH_PUSH;
556
557 switch (changes & SPECIALS_MASK) {
558 case SPECIAL_I:
559 {
560 u_int i = ntohs(cs->cs_ip.ip_len) - cs->cs_hlen;
561 th->th_ack = htonl(ntohl(th->th_ack) + i);
562 th->th_seq = htonl(ntohl(th->th_seq) + i);
563 }
564 break;
565
566 case SPECIAL_D:
567 th->th_seq = htonl(ntohl(th->th_seq) + ntohs(cs->cs_ip.ip_len)
568 - cs->cs_hlen);
569 break;
570
571 default:
572 if (changes & NEW_U) {
573 th->th_flags |= TH_URG;
574 DECODEU(th->th_urp)
575 } else
576 th->th_flags &=~ TH_URG;
577 if (changes & NEW_W)
578 DECODES(th->th_win)
579 if (changes & NEW_A)
580 DECODEL(th->th_ack)
581 if (changes & NEW_S)
582 DECODEL(th->th_seq)
583 break;
584 }
585 if (changes & NEW_I) {
586 DECODES(cs->cs_ip.ip_id)
587 } else
588 cs->cs_ip.ip_id = htons(ntohs(cs->cs_ip.ip_id) + 1);
589
590 /*
591 * At this point, cp points to the first byte of data in the
592 * packet. Fill in the IP total length and update the IP
593 * header checksum.
594 */
595 vjlen = cp - buf;
596 buflen -= vjlen;
597 if (buflen < 0)
598 /* we must have dropped some characters (crc should detect
599 * this but the old slip framing won't) */
600 goto bad;
601
602 total_len += cs->cs_hlen - vjlen;
603 cs->cs_ip.ip_len = htons(total_len);
604
605 /* recompute the ip header checksum */
606 bp = (u_int16_t *) &cs->cs_ip;
607 cs->cs_ip.ip_sum = 0;
608 for (changes = 0; hlen > 0; hlen -= 2)
609 changes += *bp++;
610 changes = (changes & 0xffff) + (changes >> 16);
611 changes = (changes & 0xffff) + (changes >> 16);
612 cs->cs_ip.ip_sum = ~ changes;
613
614 *hdrp = (u_char *) &cs->cs_ip;
615 *hlenp = cs->cs_hlen;
616 return vjlen;
617
618 bad:
619 comp->flags |= SLF_TOSS;
620 INCR(sls_errorin)
621 return (-1);
622 }