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 1.4 2004/01/17 05:47:55 carlsonj Exp $
44 #include <sys/types.h>
49 #include <net/ppp_defs.h>
56 * PPP "BSD compress" compression
57 * The differences between this compression and the classic BSD LZW
58 * source are obvious from the requirement that the classic code worked
59 * with files while this handles arbitrarily long streams that
60 * are broken into packets. They are:
62 * When the code size expands, a block of junk is not emitted by
63 * the compressor and not expected by the decompressor.
65 * New codes are not necessarily assigned every time an old
66 * code is output by the compressor. This is because a packet
67 * end forces a code to be emitted, but does not imply that a
68 * new sequence has been seen.
70 * The compression ratio is checked at the first end of a packet
71 * after the appropriate gap. Besides simplifying and speeding
72 * things up, this makes it more likely that the transmitter
73 * and receiver will agree when the dictionary is cleared when
74 * compression is not going well.
78 * A dictionary for doing BSD compress.
81 int totlen; /* length of this structure */
82 u_int hsize; /* size of the hash table */
83 u_char hshift; /* used in hash function */
84 u_char n_bits; /* current bits/code */
88 u_short seqno; /* sequence number of next packet */
89 u_int hdrlen; /* header length to preallocate */
91 u_int maxmaxcode; /* largest valid code */
92 u_int max_ent; /* largest code in use */
93 u_int in_count; /* uncompressed bytes, aged */
94 u_int bytes_out; /* compressed bytes, aged */
95 u_int ratio; /* recent compression ratio */
96 u_int checkpoint; /* when to next check the ratio */
97 u_int clear_count; /* times dictionary cleared */
98 u_int incomp_count; /* incompressible packets */
99 u_int incomp_bytes; /* incompressible bytes */
100 u_int uncomp_count; /* uncompressed packets */
101 u_int uncomp_bytes; /* uncompressed bytes */
102 u_int comp_count; /* compressed packets */
103 u_int comp_bytes; /* compressed bytes */
104 u_short *lens; /* array of lengths of codes */
106 union { /* hash value */
109 #ifdef BSD_LITTLE_ENDIAN
110 u_short prefix; /* preceding code */
111 u_char suffix; /* last character of new code */
115 u_char suffix; /* last character of new code */
116 u_short prefix; /* preceding code */
120 u_short codem1; /* output of hash table -1 */
121 u_short cptr; /* map code to hash table entry */
125 #define BSD_OVHD 2 /* BSD compress overhead/packet */
126 #define BSD_INIT_BITS BSD_MIN_BITS
128 static void *bsd_decomp_alloc(u_char *options, int opt_len);
129 static void bsd_free(void *state);
130 static int bsd_decomp_init(void *state, u_char *options, int opt_len,
131 int unit, int hdrlen, int mru, int debug);
132 static void bsd_incomp(void *state, u_char *dmsg, int len);
133 static int bsd_decompress(void *state, u_char *cmp, int inlen,
134 u_char *dmp, int *outlen);
135 static void bsd_reset(void *state);
136 static void bsd_comp_stats(void *state, struct compstat *stats);
139 * Exported procedures.
141 struct compressor ppp_bsd_compress = {
142 CI_BSD_COMPRESS, /* compress_proto */
143 bsd_decomp_alloc, /* decomp_alloc */
144 bsd_free, /* decomp_free */
145 bsd_decomp_init, /* decomp_init */
146 bsd_reset, /* decomp_reset */
147 bsd_decompress, /* decompress */
148 bsd_incomp, /* incomp */
149 bsd_comp_stats, /* decomp_stat */
153 * the next two codes should not be changed lightly, as they must not
154 * lie within the contiguous general code space.
156 #define CLEAR 256 /* table clear output code */
157 #define FIRST 257 /* first free entry */
160 #define MAXCODE(b) ((1 << (b)) - 1)
161 #define BADCODEM1 MAXCODE(BSD_MAX_BITS)
163 #define BSD_HASH(prefix,suffix,hshift) ((((u_int32_t)(suffix)) << (hshift)) \
164 ^ (u_int32_t)(prefix))
165 #define BSD_KEY(prefix,suffix) ((((u_int32_t)(suffix)) << 16) \
166 + (u_int32_t)(prefix))
168 #define CHECK_GAP 10000 /* Ratio check interval */
170 #define RATIO_SCALE_LOG 8
171 #define RATIO_SCALE (1<<RATIO_SCALE_LOG)
172 #define RATIO_MAX (0x7fffffff>>RATIO_SCALE_LOG)
175 * clear the dictionary
178 bsd_clear(struct bsd_db *db)
181 db->max_ent = FIRST-1;
182 db->n_bits = BSD_INIT_BITS;
186 db->checkpoint = CHECK_GAP;
190 * If the dictionary is full, then see if it is time to reset it.
192 * Compute the compression ratio using fixed-point arithmetic
193 * with 8 fractional bits.
195 * Since we have an infinite stream instead of a single file,
196 * watch only the local compression ratio.
198 * Since both peers must reset the dictionary at the same time even in
199 * the absence of CLEAR codes (while packets are incompressible), they
200 * must compute the same ratio.
202 static int /* 1=output CLEAR */
203 bsd_check(struct bsd_db *db)
207 if (db->in_count >= db->checkpoint) {
208 /* age the ratio by limiting the size of the counts */
209 if (db->in_count >= RATIO_MAX
210 || db->bytes_out >= RATIO_MAX) {
211 db->in_count -= db->in_count/4;
212 db->bytes_out -= db->bytes_out/4;
215 db->checkpoint = db->in_count + CHECK_GAP;
217 if (db->max_ent >= db->maxmaxcode) {
218 /* Reset the dictionary only if the ratio is worse,
219 * or if it looks as if it has been poisoned
220 * by incompressible data.
222 * This does not overflow, because
223 * db->in_count <= RATIO_MAX.
225 new_ratio = db->in_count << RATIO_SCALE_LOG;
226 if (db->bytes_out != 0)
227 new_ratio /= db->bytes_out;
229 if (new_ratio < db->ratio || new_ratio < 1 * RATIO_SCALE) {
233 db->ratio = new_ratio;
243 bsd_comp_stats(void *state, 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;
255 u_int ratio = db->in_count;
257 if (ratio <= 0x7fffff)
262 stats->ratio = ratio / out;
266 * Reset state, as on a CCP ResetReq.
269 bsd_reset(void *state)
271 struct bsd_db *db = (struct bsd_db *) state;
279 * Allocate space for a (de) compressor.
282 bsd_alloc(u_char *options, int opt_len, int decomp)
285 u_int newlen, hsize, hshift, maxmaxcode;
288 if (opt_len != 3 || options[0] != CI_BSD_COMPRESS || options[1] != 3
289 || BSD_VERSION(options[2]) != BSD_CURRENT_VERSION)
292 bits = BSD_NBITS(options[2]);
294 case 9: /* needs 82152 for both directions */
295 case 10: /* needs 84144 */
296 case 11: /* needs 88240 */
297 case 12: /* needs 96432 */
301 case 13: /* needs 176784 */
305 case 14: /* needs 353744 */
309 case 15: /* needs 691440 */
313 case 16: /* needs 1366160--far too much, */
314 /* hsize = 69001; */ /* and 69001 is too big for cptr */
315 /* hshift = 8; */ /* in struct bsd_db */
321 maxmaxcode = MAXCODE(bits);
322 newlen = sizeof(*db) + (hsize-1) * (sizeof(db->dict[0]));
323 db = (struct bsd_db *) malloc(newlen);
326 memset(db, 0, sizeof(*db) - sizeof(db->dict));
331 db->lens = (u_short *) malloc((maxmaxcode+1) * sizeof(db->lens[0]));
341 db->maxmaxcode = maxmaxcode;
348 bsd_free(void *state)
350 struct bsd_db *db = (struct bsd_db *) state;
358 bsd_decomp_alloc(u_char *options, int opt_len)
360 return bsd_alloc(options, opt_len, 1);
364 * Initialize the database.
367 bsd_init(struct bsd_db *db, u_char *options, int opt_len, int unit,
368 int hdrlen, int mru, int debug, int decomp)
372 if (opt_len < CILEN_BSD_COMPRESS
373 || options[0] != CI_BSD_COMPRESS || options[1] != CILEN_BSD_COMPRESS
374 || BSD_VERSION(options[2]) != BSD_CURRENT_VERSION
375 || BSD_NBITS(options[2]) != db->maxbits
376 || (decomp && db->lens == NULL))
386 db->dict[--i].codem1 = BADCODEM1;
387 db->dict[i].cptr = 0;
402 bsd_decomp_init(void *state, u_char *options, int opt_len,
403 int unit, int hdrlen, int mru, int debug)
405 return bsd_init((struct bsd_db *) state, options, opt_len,
406 unit, hdrlen, mru, debug, 1);
411 * Update the "BSD Compress" dictionary on the receiver for
412 * incompressible data by pretending to compress the incoming data.
415 bsd_incomp(void *state, u_char *dmsg, int mlen)
417 struct bsd_db *db = (struct bsd_db *) state;
418 u_int hshift = db->hshift;
419 u_int max_ent = db->max_ent;
420 u_int n_bits = db->n_bits;
421 struct bsd_dict *dictp;
431 ent = rptr[0]; /* get the protocol */
437 if ((ent & 1) == 0 || ent < 0x21 || ent > 0xf9)
441 ilen = 1; /* count the protocol as 1 byte */
443 slen = dmsg + mlen - rptr;
445 for (; slen > 0; --slen) {
447 fcode = BSD_KEY(ent, c);
448 hval = BSD_HASH(ent, c, hshift);
449 dictp = &db->dict[hval];
451 /* validate and then check the entry */
452 if (dictp->codem1 >= max_ent)
454 if (dictp->f.fcode == fcode) {
455 ent = dictp->codem1+1;
456 continue; /* found (prefix,suffix) */
459 /* continue probing until a match or invalid entry */
460 disp = (hval == 0) ? 1 : hval;
463 if (hval >= db->hsize)
465 dictp = &db->dict[hval];
466 if (dictp->codem1 >= max_ent)
468 } while (dictp->f.fcode != fcode);
469 ent = dictp->codem1+1;
470 continue; /* finally found (prefix,suffix) */
472 nomatch: /* output (count) the prefix */
475 /* code -> hashtable */
476 if (max_ent < db->maxmaxcode) {
477 struct bsd_dict *dictp2;
478 /* expand code size if needed */
479 if (max_ent >= MAXCODE(n_bits))
480 db->n_bits = ++n_bits;
482 /* Invalidate previous hash table entry
483 * assigned this code, and then take it over.
485 dictp2 = &db->dict[max_ent+1];
486 if (db->dict[dictp2->cptr].codem1 == max_ent)
487 db->dict[dictp2->cptr].codem1 = BADCODEM1;
489 dictp->codem1 = max_ent;
490 dictp->f.fcode = fcode;
492 db->max_ent = ++max_ent;
493 db->lens[max_ent] = db->lens[ent]+1;
497 bitno += n_bits; /* output (count) the last code */
498 db->bytes_out += bitno/8;
499 db->in_count += ilen;
503 db->incomp_bytes += ilen;
505 db->uncomp_bytes += ilen;
507 /* Increase code size if we would have without the packet
508 * boundary and as the decompressor will.
510 if (max_ent >= MAXCODE(n_bits) && max_ent < db->maxmaxcode)
516 * Decompress "BSD Compress"
518 * Because of patent problems, we return DECOMP_ERROR for errors
519 * found by inspecting the input data and for system problems, but
520 * DECOMP_FATALERROR for any errors which could possibly be said to
521 * be being detected "after" decompression. For DECOMP_ERROR,
522 * we can issue a CCP reset-request; for DECOMP_FATALERROR, we may be
523 * infringing a patent of Motorola's if we do, so we take CCP down
526 * Given that the frame has the correct sequence number and a good FCS,
527 * errors such as invalid codes in the input most likely indicate a
528 * bug, so we return DECOMP_FATALERROR for them in order to turn off
529 * compression, even though they are detected by inspecting the input.
532 bsd_decompress(void *state, u_char *cmsg, int inlen, u_char *dmp, int *outlenp)
534 struct bsd_db *db = (struct bsd_db *) state;
535 u_int max_ent = db->max_ent;
537 u_int bitno = 32; /* 1st valid bit in accm */
538 u_int n_bits = db->n_bits;
539 u_int tgtbitno = 32-n_bits; /* bitno when we have a code */
540 struct bsd_dict *dictp;
541 int explen, seq, len;
542 u_int incode, oldcode, finchar;
543 u_char *p, *rptr, *wptr;
550 ++rptr; /* skip protocol (assumed 0xfd) */
551 seq = (rptr[0] << 8) + rptr[1];
553 ilen = len = cmsg + inlen - rptr;
556 * Check the sequence number and give up if it is not what we expect.
558 if (seq != db->seqno++) {
560 printf("bsd_decomp%d: bad sequence # %d, expected %d\n",
561 db->unit, seq, db->seqno - 1);
565 wptr = dmp + db->hdrlen;
571 * Accumulate bytes until we have a complete code.
572 * Then get the next code, relying on the 32-bit,
573 * unsigned accm to mask the result.
576 accm |= *rptr++ << bitno;
578 if (tgtbitno < bitno)
580 incode = accm >> tgtbitno;
584 if (incode == CLEAR) {
586 * The dictionary must only be cleared at
587 * the end of a packet. But there could be an
588 * empty message block at the end.
592 printf("bsd_decomp%d: bad CLEAR\n", db->unit);
593 return DECOMP_FATALERROR;
600 if (incode > max_ent + 2 || incode > db->maxmaxcode
601 || (incode > max_ent && oldcode == CLEAR)) {
603 printf("bsd_decomp%d: bad code 0x%x oldcode=0x%x ",
604 db->unit, incode, oldcode);
605 printf("max_ent=0x%x seqno=%d\n",
608 return DECOMP_FATALERROR; /* probably a bug */
611 /* Special case for KwKwK string. */
612 if (incode > max_ent) {
620 codelen = db->lens[finchar];
621 explen += codelen + extra;
622 if (explen > db->mru + 1) {
624 printf("bsd_decomp%d: ran out of mru\n", db->unit);
625 return DECOMP_FATALERROR;
629 * Decode this code and install it in the decompressed buffer.
631 p = (wptr += codelen);
632 while (finchar > LAST) {
633 dictp = &db->dict[db->dict[finchar].cptr];
637 printf("bsd_decomp%d: fell off end of chain ", db->unit);
638 printf("0x%x at 0x%x by 0x%x, max_ent=0x%x\n",
639 incode, finchar, db->dict[finchar].cptr, max_ent);
640 return DECOMP_FATALERROR;
642 if (dictp->codem1 != finchar-1) {
643 printf("bsd_decomp%d: bad code chain 0x%x finchar=0x%x ",
644 db->unit, incode, finchar);
645 printf("oldcode=0x%x cptr=0x%x codem1=0x%x\n", oldcode,
646 db->dict[finchar].cptr, dictp->codem1);
647 return DECOMP_FATALERROR;
650 *--p = dictp->f.hs.suffix;
651 finchar = dictp->f.hs.prefix;
657 printf("bsd_decomp%d: short by %d after code 0x%x, max_ent=0x%x\n",
658 db->unit, codelen, incode, max_ent);
661 if (extra) /* the KwKwK case again */
665 * If not first code in a packet, and
666 * if not out of code space, then allocate a new code.
668 * Keep the hash table correct so it can be used
669 * with uncompressed packets.
671 if (oldcode != CLEAR && max_ent < db->maxmaxcode) {
672 struct bsd_dict *dictp2;
676 fcode = BSD_KEY(oldcode,finchar);
677 hval = BSD_HASH(oldcode,finchar,db->hshift);
678 dictp = &db->dict[hval];
680 /* look for a free hash table entry */
681 if (dictp->codem1 < max_ent) {
682 disp = (hval == 0) ? 1 : hval;
685 if (hval >= db->hsize)
687 dictp = &db->dict[hval];
688 } while (dictp->codem1 < max_ent);
692 * Invalidate previous hash table entry
693 * assigned this code, and then take it over
695 dictp2 = &db->dict[max_ent+1];
696 if (db->dict[dictp2->cptr].codem1 == max_ent) {
697 db->dict[dictp2->cptr].codem1 = BADCODEM1;
700 dictp->codem1 = max_ent;
701 dictp->f.fcode = fcode;
703 db->max_ent = ++max_ent;
704 db->lens[max_ent] = db->lens[oldcode]+1;
706 /* Expand code size if needed. */
707 if (max_ent >= MAXCODE(n_bits) && max_ent < db->maxmaxcode) {
708 db->n_bits = ++n_bits;
709 tgtbitno = 32-n_bits;
714 *outlenp = wptr - (dmp + db->hdrlen);
717 * Keep the checkpoint right so that incompressible packets
718 * clear the dictionary at the right times.
720 db->bytes_out += ilen;
721 db->in_count += explen;
722 if (bsd_check(db) && db->debug) {
723 printf("bsd_decomp%d: peer should have cleared dictionary\n",
728 db->comp_bytes += ilen + BSD_OVHD;
730 db->uncomp_bytes += explen;
734 #endif /* DO_BSD_COMPRESS */