1 /* Because this code is derived from the 4.3BSD compress source:
4 * Copyright (c) 1985, 1986 The Regents of the University of California.
7 * This code is derived from software contributed to Berkeley by
8 * James A. Woods, derived from original work by Spencer Thomas
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * $Id: bsd-comp.c,v 1.1 1999/03/23 03:21:57 paulus Exp $
44 #include <sys/types.h>
52 * PPP "BSD compress" compression
53 * The differences between this compression and the classic BSD LZW
54 * source are obvious from the requirement that the classic code worked
55 * with files while this handles arbitrarily long streams that
56 * are broken into packets. They are:
58 * When the code size expands, a block of junk is not emitted by
59 * the compressor and not expected by the decompressor.
61 * New codes are not necessarily assigned every time an old
62 * code is output by the compressor. This is because a packet
63 * end forces a code to be emitted, but does not imply that a
64 * new sequence has been seen.
66 * The compression ratio is checked at the first end of a packet
67 * after the appropriate gap. Besides simplifying and speeding
68 * things up, this makes it more likely that the transmitter
69 * and receiver will agree when the dictionary is cleared when
70 * compression is not going well.
74 * A dictionary for doing BSD compress.
77 int totlen; /* length of this structure */
78 u_int hsize; /* size of the hash table */
79 u_char hshift; /* used in hash function */
80 u_char n_bits; /* current bits/code */
84 u_short seqno; /* sequence number of next packet */
85 u_int hdrlen; /* header length to preallocate */
87 u_int maxmaxcode; /* largest valid code */
88 u_int max_ent; /* largest code in use */
89 u_int in_count; /* uncompressed bytes, aged */
90 u_int bytes_out; /* compressed bytes, aged */
91 u_int ratio; /* recent compression ratio */
92 u_int checkpoint; /* when to next check the ratio */
93 u_int clear_count; /* times dictionary cleared */
94 u_int incomp_count; /* incompressible packets */
95 u_int incomp_bytes; /* incompressible bytes */
96 u_int uncomp_count; /* uncompressed packets */
97 u_int uncomp_bytes; /* uncompressed bytes */
98 u_int comp_count; /* compressed packets */
99 u_int comp_bytes; /* compressed bytes */
100 u_short *lens; /* array of lengths of codes */
102 union { /* hash value */
105 #ifdef BSD_LITTLE_ENDIAN
106 u_short prefix; /* preceding code */
107 u_char suffix; /* last character of new code */
111 u_char suffix; /* last character of new code */
112 u_short prefix; /* preceding code */
116 u_short codem1; /* output of hash table -1 */
117 u_short cptr; /* map code to hash table entry */
121 #define BSD_OVHD 2 /* BSD compress overhead/packet */
122 #define BSD_INIT_BITS BSD_MIN_BITS
124 static void *bsd_decomp_alloc __P((u_char *options, int opt_len));
125 static void bsd_free __P((void *state));
126 static int bsd_decomp_init __P((void *state, u_char *options, int opt_len,
127 int unit, int hdrlen, int mru, int debug));
128 static void bsd_incomp __P((void *state, u_char *dmsg, int len));
129 static int bsd_decompress __P((void *state, u_char *cmp, int inlen,
130 u_char *dmp, int *outlen));
131 static void bsd_reset __P((void *state));
132 static void bsd_comp_stats __P((void *state, struct compstat *stats));
135 * Exported procedures.
137 struct compressor ppp_bsd_compress = {
138 CI_BSD_COMPRESS, /* compress_proto */
139 bsd_decomp_alloc, /* decomp_alloc */
140 bsd_free, /* decomp_free */
141 bsd_decomp_init, /* decomp_init */
142 bsd_reset, /* decomp_reset */
143 bsd_decompress, /* decompress */
144 bsd_incomp, /* incomp */
145 bsd_comp_stats, /* decomp_stat */
149 * the next two codes should not be changed lightly, as they must not
150 * lie within the contiguous general code space.
152 #define CLEAR 256 /* table clear output code */
153 #define FIRST 257 /* first free entry */
156 #define MAXCODE(b) ((1 << (b)) - 1)
157 #define BADCODEM1 MAXCODE(BSD_MAX_BITS)
159 #define BSD_HASH(prefix,suffix,hshift) ((((u_int32_t)(suffix)) << (hshift)) \
160 ^ (u_int32_t)(prefix))
161 #define BSD_KEY(prefix,suffix) ((((u_int32_t)(suffix)) << 16) \
162 + (u_int32_t)(prefix))
164 #define CHECK_GAP 10000 /* Ratio check interval */
166 #define RATIO_SCALE_LOG 8
167 #define RATIO_SCALE (1<<RATIO_SCALE_LOG)
168 #define RATIO_MAX (0x7fffffff>>RATIO_SCALE_LOG)
171 * clear the dictionary
178 db->max_ent = FIRST-1;
179 db->n_bits = BSD_INIT_BITS;
183 db->checkpoint = CHECK_GAP;
187 * If the dictionary is full, then see if it is time to reset it.
189 * Compute the compression ratio using fixed-point arithmetic
190 * with 8 fractional bits.
192 * Since we have an infinite stream instead of a single file,
193 * watch only the local compression ratio.
195 * Since both peers must reset the dictionary at the same time even in
196 * the absence of CLEAR codes (while packets are incompressible), they
197 * must compute the same ratio.
199 static int /* 1=output CLEAR */
205 if (db->in_count >= db->checkpoint) {
206 /* age the ratio by limiting the size of the counts */
207 if (db->in_count >= RATIO_MAX
208 || db->bytes_out >= RATIO_MAX) {
209 db->in_count -= db->in_count/4;
210 db->bytes_out -= db->bytes_out/4;
213 db->checkpoint = db->in_count + CHECK_GAP;
215 if (db->max_ent >= db->maxmaxcode) {
216 /* Reset the dictionary only if the ratio is worse,
217 * or if it looks as if it has been poisoned
218 * by incompressible data.
220 * This does not overflow, because
221 * db->in_count <= RATIO_MAX.
223 new_ratio = db->in_count << RATIO_SCALE_LOG;
224 if (db->bytes_out != 0)
225 new_ratio /= db->bytes_out;
227 if (new_ratio < db->ratio || new_ratio < 1 * RATIO_SCALE) {
231 db->ratio = new_ratio;
241 bsd_comp_stats(state, stats)
243 struct compstat *stats;
245 struct bsd_db *db = (struct bsd_db *) state;
248 stats->unc_bytes = db->uncomp_bytes;
249 stats->unc_packets = db->uncomp_count;
250 stats->comp_bytes = db->comp_bytes;
251 stats->comp_packets = db->comp_count;
252 stats->inc_bytes = db->incomp_bytes;
253 stats->inc_packets = db->incomp_count;
254 stats->ratio = db->in_count;
256 if (stats->ratio <= 0x7fffff)
265 * Reset state, as on a CCP ResetReq.
271 struct bsd_db *db = (struct bsd_db *) state;
279 * Allocate space for a (de) compressor.
282 bsd_alloc(options, opt_len, decomp)
287 u_int newlen, hsize, hshift, maxmaxcode;
290 if (opt_len != 3 || options[0] != CI_BSD_COMPRESS || options[1] != 3
291 || BSD_VERSION(options[2]) != BSD_CURRENT_VERSION)
294 bits = BSD_NBITS(options[2]);
296 case 9: /* needs 82152 for both directions */
297 case 10: /* needs 84144 */
298 case 11: /* needs 88240 */
299 case 12: /* needs 96432 */
303 case 13: /* needs 176784 */
307 case 14: /* needs 353744 */
311 case 15: /* needs 691440 */
315 case 16: /* needs 1366160--far too much, */
316 /* hsize = 69001; */ /* and 69001 is too big for cptr */
317 /* hshift = 8; */ /* in struct bsd_db */
323 maxmaxcode = MAXCODE(bits);
324 newlen = sizeof(*db) + (hsize-1) * (sizeof(db->dict[0]));
325 db = (struct bsd_db *) malloc(newlen);
328 bzero(db, sizeof(*db) - sizeof(db->dict));
333 db->lens = (u_short *) malloc((maxmaxcode+1) * sizeof(db->lens[0]));
343 db->maxmaxcode = maxmaxcode;
353 struct bsd_db *db = (struct bsd_db *) state;
361 bsd_decomp_alloc(options, opt_len)
365 return bsd_alloc(options, opt_len, 1);
369 * Initialize the database.
372 bsd_init(db, options, opt_len, unit, hdrlen, mru, debug, decomp)
375 int opt_len, unit, hdrlen, mru, debug, decomp;
379 if (opt_len < CILEN_BSD_COMPRESS
380 || options[0] != CI_BSD_COMPRESS || options[1] != CILEN_BSD_COMPRESS
381 || BSD_VERSION(options[2]) != BSD_CURRENT_VERSION
382 || BSD_NBITS(options[2]) != db->maxbits
383 || decomp && db->lens == NULL)
393 db->dict[--i].codem1 = BADCODEM1;
394 db->dict[i].cptr = 0;
409 bsd_decomp_init(state, options, opt_len, unit, hdrlen, mru, debug)
412 int opt_len, unit, hdrlen, mru, debug;
414 return bsd_init((struct bsd_db *) state, options, opt_len,
415 unit, hdrlen, mru, debug, 1);
420 * Update the "BSD Compress" dictionary on the receiver for
421 * incompressible data by pretending to compress the incoming data.
424 bsd_incomp(state, dmsg, mlen)
429 struct bsd_db *db = (struct bsd_db *) state;
430 u_int hshift = db->hshift;
431 u_int max_ent = db->max_ent;
432 u_int n_bits = db->n_bits;
433 struct bsd_dict *dictp;
443 ent = rptr[0]; /* get the protocol */
449 if ((ent & 1) == 0 || ent < 0x21 || ent > 0xf9)
453 ilen = 1; /* count the protocol as 1 byte */
455 slen = dmsg + mlen - rptr;
457 for (; slen > 0; --slen) {
459 fcode = BSD_KEY(ent, c);
460 hval = BSD_HASH(ent, c, hshift);
461 dictp = &db->dict[hval];
463 /* validate and then check the entry */
464 if (dictp->codem1 >= max_ent)
466 if (dictp->f.fcode == fcode) {
467 ent = dictp->codem1+1;
468 continue; /* found (prefix,suffix) */
471 /* continue probing until a match or invalid entry */
472 disp = (hval == 0) ? 1 : hval;
475 if (hval >= db->hsize)
477 dictp = &db->dict[hval];
478 if (dictp->codem1 >= max_ent)
480 } while (dictp->f.fcode != fcode);
481 ent = dictp->codem1+1;
482 continue; /* finally found (prefix,suffix) */
484 nomatch: /* output (count) the prefix */
487 /* code -> hashtable */
488 if (max_ent < db->maxmaxcode) {
489 struct bsd_dict *dictp2;
490 /* expand code size if needed */
491 if (max_ent >= MAXCODE(n_bits))
492 db->n_bits = ++n_bits;
494 /* Invalidate previous hash table entry
495 * assigned this code, and then take it over.
497 dictp2 = &db->dict[max_ent+1];
498 if (db->dict[dictp2->cptr].codem1 == max_ent)
499 db->dict[dictp2->cptr].codem1 = BADCODEM1;
501 dictp->codem1 = max_ent;
502 dictp->f.fcode = fcode;
504 db->max_ent = ++max_ent;
505 db->lens[max_ent] = db->lens[ent]+1;
509 bitno += n_bits; /* output (count) the last code */
510 db->bytes_out += bitno/8;
511 db->in_count += ilen;
515 db->incomp_bytes += ilen;
517 db->uncomp_bytes += ilen;
519 /* Increase code size if we would have without the packet
520 * boundary and as the decompressor will.
522 if (max_ent >= MAXCODE(n_bits) && max_ent < db->maxmaxcode)
528 * Decompress "BSD Compress"
530 * Because of patent problems, we return DECOMP_ERROR for errors
531 * found by inspecting the input data and for system problems, but
532 * DECOMP_FATALERROR for any errors which could possibly be said to
533 * be being detected "after" decompression. For DECOMP_ERROR,
534 * we can issue a CCP reset-request; for DECOMP_FATALERROR, we may be
535 * infringing a patent of Motorola's if we do, so we take CCP down
538 * Given that the frame has the correct sequence number and a good FCS,
539 * errors such as invalid codes in the input most likely indicate a
540 * bug, so we return DECOMP_FATALERROR for them in order to turn off
541 * compression, even though they are detected by inspecting the input.
544 bsd_decompress(state, cmsg, inlen, dmp, outlenp)
549 struct bsd_db *db = (struct bsd_db *) state;
550 u_int max_ent = db->max_ent;
552 u_int bitno = 32; /* 1st valid bit in accm */
553 u_int n_bits = db->n_bits;
554 u_int tgtbitno = 32-n_bits; /* bitno when we have a code */
555 struct bsd_dict *dictp;
556 int explen, i, seq, len;
557 u_int incode, oldcode, finchar;
558 u_char *p, *rptr, *wptr;
560 int dlen, space, codelen, extra;
565 ++rptr; /* skip protocol (assumed 0xfd) */
566 seq = (rptr[0] << 8) + rptr[1];
568 ilen = len = cmsg + inlen - rptr;
571 * Check the sequence number and give up if it is not what we expect.
573 if (seq != db->seqno++) {
575 printf("bsd_decomp%d: bad sequence # %d, expected %d\n",
576 db->unit, seq, db->seqno - 1);
580 wptr = dmp + db->hdrlen;
586 * Accumulate bytes until we have a complete code.
587 * Then get the next code, relying on the 32-bit,
588 * unsigned accm to mask the result.
591 accm |= *rptr++ << bitno;
593 if (tgtbitno < bitno)
595 incode = accm >> tgtbitno;
599 if (incode == CLEAR) {
601 * The dictionary must only be cleared at
602 * the end of a packet. But there could be an
603 * empty message block at the end.
607 printf("bsd_decomp%d: bad CLEAR\n", db->unit);
608 return DECOMP_FATALERROR;
615 if (incode > max_ent + 2 || incode > db->maxmaxcode
616 || incode > max_ent && oldcode == CLEAR) {
618 printf("bsd_decomp%d: bad code 0x%x oldcode=0x%x ",
619 db->unit, incode, oldcode);
620 printf("max_ent=0x%x dlen=%d seqno=%d\n",
621 max_ent, dlen, db->seqno);
623 return DECOMP_FATALERROR; /* probably a bug */
626 /* Special case for KwKwK string. */
627 if (incode > max_ent) {
635 codelen = db->lens[finchar];
636 explen += codelen + extra;
637 if (explen > db->mru + 1) {
639 printf("bsd_decomp%d: ran out of mru\n", db->unit);
640 return DECOMP_FATALERROR;
644 * Decode this code and install it in the decompressed buffer.
646 p = (wptr += codelen);
647 while (finchar > LAST) {
648 dictp = &db->dict[db->dict[finchar].cptr];
652 printf("bsd_decomp%d: fell off end of chain ", db->unit);
653 printf("0x%x at 0x%x by 0x%x, max_ent=0x%x\n",
654 incode, finchar, db->dict[finchar].cptr, max_ent);
655 return DECOMP_FATALERROR;
657 if (dictp->codem1 != finchar-1) {
658 printf("bsd_decomp%d: bad code chain 0x%x finchar=0x%x ",
659 db->unit, incode, finchar);
660 printf("oldcode=0x%x cptr=0x%x codem1=0x%x\n", oldcode,
661 db->dict[finchar].cptr, dictp->codem1);
662 return DECOMP_FATALERROR;
665 *--p = dictp->f.hs.suffix;
666 finchar = dictp->f.hs.prefix;
672 printf("bsd_decomp%d: short by %d after code 0x%x, max_ent=0x%x\n",
673 db->unit, codelen, incode, max_ent);
676 if (extra) /* the KwKwK case again */
680 * If not first code in a packet, and
681 * if not out of code space, then allocate a new code.
683 * Keep the hash table correct so it can be used
684 * with uncompressed packets.
686 if (oldcode != CLEAR && max_ent < db->maxmaxcode) {
687 struct bsd_dict *dictp2;
691 fcode = BSD_KEY(oldcode,finchar);
692 hval = BSD_HASH(oldcode,finchar,db->hshift);
693 dictp = &db->dict[hval];
695 /* look for a free hash table entry */
696 if (dictp->codem1 < max_ent) {
697 disp = (hval == 0) ? 1 : hval;
700 if (hval >= db->hsize)
702 dictp = &db->dict[hval];
703 } while (dictp->codem1 < max_ent);
707 * Invalidate previous hash table entry
708 * assigned this code, and then take it over
710 dictp2 = &db->dict[max_ent+1];
711 if (db->dict[dictp2->cptr].codem1 == max_ent) {
712 db->dict[dictp2->cptr].codem1 = BADCODEM1;
715 dictp->codem1 = max_ent;
716 dictp->f.fcode = fcode;
718 db->max_ent = ++max_ent;
719 db->lens[max_ent] = db->lens[oldcode]+1;
721 /* Expand code size if needed. */
722 if (max_ent >= MAXCODE(n_bits) && max_ent < db->maxmaxcode) {
723 db->n_bits = ++n_bits;
724 tgtbitno = 32-n_bits;
729 *outlenp = wptr - (dmp + db->hdrlen);
732 * Keep the checkpoint right so that incompressible packets
733 * clear the dictionary at the right times.
735 db->bytes_out += ilen;
736 db->in_count += explen;
737 if (bsd_check(db) && db->debug) {
738 printf("bsd_decomp%d: peer should have cleared dictionary\n",
743 db->comp_bytes += ilen + BSD_OVHD;
745 db->uncomp_bytes += explen;
749 #endif /* DO_BSD_COMPRESS */