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>
55 * PPP "BSD compress" compression
56 * The differences between this compression and the classic BSD LZW
57 * source are obvious from the requirement that the classic code worked
58 * with files while this handles arbitrarily long streams that
59 * are broken into packets. They are:
61 * When the code size expands, a block of junk is not emitted by
62 * the compressor and not expected by the decompressor.
64 * New codes are not necessarily assigned every time an old
65 * code is output by the compressor. This is because a packet
66 * end forces a code to be emitted, but does not imply that a
67 * new sequence has been seen.
69 * The compression ratio is checked at the first end of a packet
70 * after the appropriate gap. Besides simplifying and speeding
71 * things up, this makes it more likely that the transmitter
72 * and receiver will agree when the dictionary is cleared when
73 * compression is not going well.
77 * A dictionary for doing BSD compress.
80 int totlen
; /* length of this structure */
81 u_int hsize
; /* size of the hash table */
82 u_char hshift
; /* used in hash function */
83 u_char n_bits
; /* current bits/code */
87 u_short seqno
; /* sequence number of next packet */
88 u_int hdrlen
; /* header length to preallocate */
90 u_int maxmaxcode
; /* largest valid code */
91 u_int max_ent
; /* largest code in use */
92 u_int in_count
; /* uncompressed bytes, aged */
93 u_int bytes_out
; /* compressed bytes, aged */
94 u_int ratio
; /* recent compression ratio */
95 u_int checkpoint
; /* when to next check the ratio */
96 u_int clear_count
; /* times dictionary cleared */
97 u_int incomp_count
; /* incompressible packets */
98 u_int incomp_bytes
; /* incompressible bytes */
99 u_int uncomp_count
; /* uncompressed packets */
100 u_int uncomp_bytes
; /* uncompressed bytes */
101 u_int comp_count
; /* compressed packets */
102 u_int comp_bytes
; /* compressed bytes */
103 u_short
*lens
; /* array of lengths of codes */
105 union { /* hash value */
108 #ifdef BSD_LITTLE_ENDIAN
109 u_short prefix
; /* preceding code */
110 u_char suffix
; /* last character of new code */
114 u_char suffix
; /* last character of new code */
115 u_short prefix
; /* preceding code */
119 u_short codem1
; /* output of hash table -1 */
120 u_short cptr
; /* map code to hash table entry */
124 #define BSD_OVHD 2 /* BSD compress overhead/packet */
125 #define BSD_INIT_BITS BSD_MIN_BITS
127 static void *bsd_decomp_alloc
__P((u_char
*options
, int opt_len
));
128 static void bsd_free
__P((void *state
));
129 static int bsd_decomp_init
__P((void *state
, u_char
*options
, int opt_len
,
130 int unit
, int hdrlen
, int mru
, int debug
));
131 static void bsd_incomp
__P((void *state
, u_char
*dmsg
, int len
));
132 static int bsd_decompress
__P((void *state
, u_char
*cmp
, int inlen
,
133 u_char
*dmp
, int *outlen
));
134 static void bsd_reset
__P((void *state
));
135 static void bsd_comp_stats
__P((void *state
, struct compstat
*stats
));
138 * Exported procedures.
140 struct compressor ppp_bsd_compress
= {
141 CI_BSD_COMPRESS
, /* compress_proto */
142 bsd_decomp_alloc
, /* decomp_alloc */
143 bsd_free
, /* decomp_free */
144 bsd_decomp_init
, /* decomp_init */
145 bsd_reset
, /* decomp_reset */
146 bsd_decompress
, /* decompress */
147 bsd_incomp
, /* incomp */
148 bsd_comp_stats
, /* decomp_stat */
152 * the next two codes should not be changed lightly, as they must not
153 * lie within the contiguous general code space.
155 #define CLEAR 256 /* table clear output code */
156 #define FIRST 257 /* first free entry */
159 #define MAXCODE(b) ((1 << (b)) - 1)
160 #define BADCODEM1 MAXCODE(BSD_MAX_BITS)
162 #define BSD_HASH(prefix,suffix,hshift) ((((u_int32_t)(suffix)) << (hshift)) \
163 ^ (u_int32_t)(prefix))
164 #define BSD_KEY(prefix,suffix) ((((u_int32_t)(suffix)) << 16) \
165 + (u_int32_t)(prefix))
167 #define CHECK_GAP 10000 /* Ratio check interval */
169 #define RATIO_SCALE_LOG 8
170 #define RATIO_SCALE (1<<RATIO_SCALE_LOG)
171 #define RATIO_MAX (0x7fffffff>>RATIO_SCALE_LOG)
174 * clear the dictionary
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 */
208 if (db
->in_count
>= db
->checkpoint
) {
209 /* age the ratio by limiting the size of the counts */
210 if (db
->in_count
>= RATIO_MAX
211 || db
->bytes_out
>= RATIO_MAX
) {
212 db
->in_count
-= db
->in_count
/4;
213 db
->bytes_out
-= db
->bytes_out
/4;
216 db
->checkpoint
= db
->in_count
+ CHECK_GAP
;
218 if (db
->max_ent
>= db
->maxmaxcode
) {
219 /* Reset the dictionary only if the ratio is worse,
220 * or if it looks as if it has been poisoned
221 * by incompressible data.
223 * This does not overflow, because
224 * db->in_count <= RATIO_MAX.
226 new_ratio
= db
->in_count
<< RATIO_SCALE_LOG
;
227 if (db
->bytes_out
!= 0)
228 new_ratio
/= db
->bytes_out
;
230 if (new_ratio
< db
->ratio
|| new_ratio
< 1 * RATIO_SCALE
) {
234 db
->ratio
= new_ratio
;
244 bsd_comp_stats(state
, stats
)
246 struct compstat
*stats
;
248 struct bsd_db
*db
= (struct bsd_db
*) state
;
251 stats
->unc_bytes
= db
->uncomp_bytes
;
252 stats
->unc_packets
= db
->uncomp_count
;
253 stats
->comp_bytes
= db
->comp_bytes
;
254 stats
->comp_packets
= db
->comp_count
;
255 stats
->inc_bytes
= db
->incomp_bytes
;
256 stats
->inc_packets
= db
->incomp_count
;
257 stats
->ratio
= db
->in_count
;
259 if (stats
->ratio
<= 0x7fffff)
268 * Reset state, as on a CCP ResetReq.
274 struct bsd_db
*db
= (struct bsd_db
*) state
;
282 * Allocate space for a (de) compressor.
285 bsd_alloc(options
, opt_len
, decomp
)
290 u_int newlen
, hsize
, hshift
, maxmaxcode
;
293 if (opt_len
!= 3 || options
[0] != CI_BSD_COMPRESS
|| options
[1] != 3
294 || BSD_VERSION(options
[2]) != BSD_CURRENT_VERSION
)
297 bits
= BSD_NBITS(options
[2]);
299 case 9: /* needs 82152 for both directions */
300 case 10: /* needs 84144 */
301 case 11: /* needs 88240 */
302 case 12: /* needs 96432 */
306 case 13: /* needs 176784 */
310 case 14: /* needs 353744 */
314 case 15: /* needs 691440 */
318 case 16: /* needs 1366160--far too much, */
319 /* hsize = 69001; */ /* and 69001 is too big for cptr */
320 /* hshift = 8; */ /* in struct bsd_db */
326 maxmaxcode
= MAXCODE(bits
);
327 newlen
= sizeof(*db
) + (hsize
-1) * (sizeof(db
->dict
[0]));
328 db
= (struct bsd_db
*) malloc(newlen
);
331 memset(db
, 0, sizeof(*db
) - sizeof(db
->dict
));
336 db
->lens
= (u_short
*) malloc((maxmaxcode
+1) * sizeof(db
->lens
[0]));
346 db
->maxmaxcode
= maxmaxcode
;
356 struct bsd_db
*db
= (struct bsd_db
*) state
;
364 bsd_decomp_alloc(options
, opt_len
)
368 return bsd_alloc(options
, opt_len
, 1);
372 * Initialize the database.
375 bsd_init(db
, options
, opt_len
, unit
, hdrlen
, mru
, debug
, decomp
)
378 int opt_len
, unit
, hdrlen
, mru
, debug
, decomp
;
382 if (opt_len
< CILEN_BSD_COMPRESS
383 || options
[0] != CI_BSD_COMPRESS
|| options
[1] != CILEN_BSD_COMPRESS
384 || BSD_VERSION(options
[2]) != BSD_CURRENT_VERSION
385 || BSD_NBITS(options
[2]) != db
->maxbits
386 || decomp
&& db
->lens
== NULL
)
396 db
->dict
[--i
].codem1
= BADCODEM1
;
397 db
->dict
[i
].cptr
= 0;
412 bsd_decomp_init(state
, options
, opt_len
, unit
, hdrlen
, mru
, debug
)
415 int opt_len
, unit
, hdrlen
, mru
, debug
;
417 return bsd_init((struct bsd_db
*) state
, options
, opt_len
,
418 unit
, hdrlen
, mru
, debug
, 1);
423 * Update the "BSD Compress" dictionary on the receiver for
424 * incompressible data by pretending to compress the incoming data.
427 bsd_incomp(state
, dmsg
, mlen
)
432 struct bsd_db
*db
= (struct bsd_db
*) state
;
433 u_int hshift
= db
->hshift
;
434 u_int max_ent
= db
->max_ent
;
435 u_int n_bits
= db
->n_bits
;
436 struct bsd_dict
*dictp
;
446 ent
= rptr
[0]; /* get the protocol */
452 if ((ent
& 1) == 0 || ent
< 0x21 || ent
> 0xf9)
456 ilen
= 1; /* count the protocol as 1 byte */
458 slen
= dmsg
+ mlen
- rptr
;
460 for (; slen
> 0; --slen
) {
462 fcode
= BSD_KEY(ent
, c
);
463 hval
= BSD_HASH(ent
, c
, hshift
);
464 dictp
= &db
->dict
[hval
];
466 /* validate and then check the entry */
467 if (dictp
->codem1
>= max_ent
)
469 if (dictp
->f
.fcode
== fcode
) {
470 ent
= dictp
->codem1
+1;
471 continue; /* found (prefix,suffix) */
474 /* continue probing until a match or invalid entry */
475 disp
= (hval
== 0) ? 1 : hval
;
478 if (hval
>= db
->hsize
)
480 dictp
= &db
->dict
[hval
];
481 if (dictp
->codem1
>= max_ent
)
483 } while (dictp
->f
.fcode
!= fcode
);
484 ent
= dictp
->codem1
+1;
485 continue; /* finally found (prefix,suffix) */
487 nomatch
: /* output (count) the prefix */
490 /* code -> hashtable */
491 if (max_ent
< db
->maxmaxcode
) {
492 struct bsd_dict
*dictp2
;
493 /* expand code size if needed */
494 if (max_ent
>= MAXCODE(n_bits
))
495 db
->n_bits
= ++n_bits
;
497 /* Invalidate previous hash table entry
498 * assigned this code, and then take it over.
500 dictp2
= &db
->dict
[max_ent
+1];
501 if (db
->dict
[dictp2
->cptr
].codem1
== max_ent
)
502 db
->dict
[dictp2
->cptr
].codem1
= BADCODEM1
;
504 dictp
->codem1
= max_ent
;
505 dictp
->f
.fcode
= fcode
;
507 db
->max_ent
= ++max_ent
;
508 db
->lens
[max_ent
] = db
->lens
[ent
]+1;
512 bitno
+= n_bits
; /* output (count) the last code */
513 db
->bytes_out
+= bitno
/8;
514 db
->in_count
+= ilen
;
518 db
->incomp_bytes
+= ilen
;
520 db
->uncomp_bytes
+= ilen
;
522 /* Increase code size if we would have without the packet
523 * boundary and as the decompressor will.
525 if (max_ent
>= MAXCODE(n_bits
) && max_ent
< db
->maxmaxcode
)
531 * Decompress "BSD Compress"
533 * Because of patent problems, we return DECOMP_ERROR for errors
534 * found by inspecting the input data and for system problems, but
535 * DECOMP_FATALERROR for any errors which could possibly be said to
536 * be being detected "after" decompression. For DECOMP_ERROR,
537 * we can issue a CCP reset-request; for DECOMP_FATALERROR, we may be
538 * infringing a patent of Motorola's if we do, so we take CCP down
541 * Given that the frame has the correct sequence number and a good FCS,
542 * errors such as invalid codes in the input most likely indicate a
543 * bug, so we return DECOMP_FATALERROR for them in order to turn off
544 * compression, even though they are detected by inspecting the input.
547 bsd_decompress(state
, cmsg
, inlen
, dmp
, outlenp
)
552 struct bsd_db
*db
= (struct bsd_db
*) state
;
553 u_int max_ent
= db
->max_ent
;
555 u_int bitno
= 32; /* 1st valid bit in accm */
556 u_int n_bits
= db
->n_bits
;
557 u_int tgtbitno
= 32-n_bits
; /* bitno when we have a code */
558 struct bsd_dict
*dictp
;
559 int explen
, i
, seq
, len
;
560 u_int incode
, oldcode
, finchar
;
561 u_char
*p
, *rptr
, *wptr
;
563 int dlen
, space
, codelen
, extra
;
568 ++rptr
; /* skip protocol (assumed 0xfd) */
569 seq
= (rptr
[0] << 8) + rptr
[1];
571 ilen
= len
= cmsg
+ inlen
- rptr
;
574 * Check the sequence number and give up if it is not what we expect.
576 if (seq
!= db
->seqno
++) {
578 printf("bsd_decomp%d: bad sequence # %d, expected %d\n",
579 db
->unit
, seq
, db
->seqno
- 1);
583 wptr
= dmp
+ db
->hdrlen
;
589 * Accumulate bytes until we have a complete code.
590 * Then get the next code, relying on the 32-bit,
591 * unsigned accm to mask the result.
594 accm
|= *rptr
++ << bitno
;
596 if (tgtbitno
< bitno
)
598 incode
= accm
>> tgtbitno
;
602 if (incode
== CLEAR
) {
604 * The dictionary must only be cleared at
605 * the end of a packet. But there could be an
606 * empty message block at the end.
610 printf("bsd_decomp%d: bad CLEAR\n", db
->unit
);
611 return DECOMP_FATALERROR
;
618 if (incode
> max_ent
+ 2 || incode
> db
->maxmaxcode
619 || incode
> max_ent
&& oldcode
== CLEAR
) {
621 printf("bsd_decomp%d: bad code 0x%x oldcode=0x%x ",
622 db
->unit
, incode
, oldcode
);
623 printf("max_ent=0x%x dlen=%d seqno=%d\n",
624 max_ent
, dlen
, db
->seqno
);
626 return DECOMP_FATALERROR
; /* probably a bug */
629 /* Special case for KwKwK string. */
630 if (incode
> max_ent
) {
638 codelen
= db
->lens
[finchar
];
639 explen
+= codelen
+ extra
;
640 if (explen
> db
->mru
+ 1) {
642 printf("bsd_decomp%d: ran out of mru\n", db
->unit
);
643 return DECOMP_FATALERROR
;
647 * Decode this code and install it in the decompressed buffer.
649 p
= (wptr
+= codelen
);
650 while (finchar
> LAST
) {
651 dictp
= &db
->dict
[db
->dict
[finchar
].cptr
];
655 printf("bsd_decomp%d: fell off end of chain ", db
->unit
);
656 printf("0x%x at 0x%x by 0x%x, max_ent=0x%x\n",
657 incode
, finchar
, db
->dict
[finchar
].cptr
, max_ent
);
658 return DECOMP_FATALERROR
;
660 if (dictp
->codem1
!= finchar
-1) {
661 printf("bsd_decomp%d: bad code chain 0x%x finchar=0x%x ",
662 db
->unit
, incode
, finchar
);
663 printf("oldcode=0x%x cptr=0x%x codem1=0x%x\n", oldcode
,
664 db
->dict
[finchar
].cptr
, dictp
->codem1
);
665 return DECOMP_FATALERROR
;
668 *--p
= dictp
->f
.hs
.suffix
;
669 finchar
= dictp
->f
.hs
.prefix
;
675 printf("bsd_decomp%d: short by %d after code 0x%x, max_ent=0x%x\n",
676 db
->unit
, codelen
, incode
, max_ent
);
679 if (extra
) /* the KwKwK case again */
683 * If not first code in a packet, and
684 * if not out of code space, then allocate a new code.
686 * Keep the hash table correct so it can be used
687 * with uncompressed packets.
689 if (oldcode
!= CLEAR
&& max_ent
< db
->maxmaxcode
) {
690 struct bsd_dict
*dictp2
;
694 fcode
= BSD_KEY(oldcode
,finchar
);
695 hval
= BSD_HASH(oldcode
,finchar
,db
->hshift
);
696 dictp
= &db
->dict
[hval
];
698 /* look for a free hash table entry */
699 if (dictp
->codem1
< max_ent
) {
700 disp
= (hval
== 0) ? 1 : hval
;
703 if (hval
>= db
->hsize
)
705 dictp
= &db
->dict
[hval
];
706 } while (dictp
->codem1
< max_ent
);
710 * Invalidate previous hash table entry
711 * assigned this code, and then take it over
713 dictp2
= &db
->dict
[max_ent
+1];
714 if (db
->dict
[dictp2
->cptr
].codem1
== max_ent
) {
715 db
->dict
[dictp2
->cptr
].codem1
= BADCODEM1
;
718 dictp
->codem1
= max_ent
;
719 dictp
->f
.fcode
= fcode
;
721 db
->max_ent
= ++max_ent
;
722 db
->lens
[max_ent
] = db
->lens
[oldcode
]+1;
724 /* Expand code size if needed. */
725 if (max_ent
>= MAXCODE(n_bits
) && max_ent
< db
->maxmaxcode
) {
726 db
->n_bits
= ++n_bits
;
727 tgtbitno
= 32-n_bits
;
732 *outlenp
= wptr
- (dmp
+ db
->hdrlen
);
735 * Keep the checkpoint right so that incompressible packets
736 * clear the dictionary at the right times.
738 db
->bytes_out
+= ilen
;
739 db
->in_count
+= explen
;
740 if (bsd_check(db
) && db
->debug
) {
741 printf("bsd_decomp%d: peer should have cleared dictionary\n",
746 db
->comp_bytes
+= ilen
+ BSD_OVHD
;
748 db
->uncomp_bytes
+= explen
;
752 #endif /* DO_BSD_COMPRESS */