1 /* inftrees.c -- generate Huffman trees for efficient decoding
2  * Copyright (C) 1995-2013 Mark Adler
3  * For conditions of distribution and use, see copyright notice in zlib.h
4  */
5 
6 #include "zutil.h"
7 #include "inftrees.h"
8 
9 zRCSID("$MirOS: src/kern/z/inftrees.c,v 1.4 2013/08/05 21:27:34 tg Exp $")
10 
11 #define MAXBITS 15
12 
13 #ifndef SMALL
14 const char inflate_copyright[] =
15    " inflate 1.2.8 Copyright 1995-2013 Mark Adler ";
16 #endif
17 /*
18   If you use the zlib library in a product, an acknowledgment is welcome
19   in the documentation of your product. If for some reason you cannot
20   include such an acknowledgment, I would appreciate that you keep this
21   copyright string in the executable of your product.
22  */
23 
24 /*
25    Build a set of tables to decode the provided canonical Huffman code.
26    The code lengths are lens[0..codes-1].  The result starts at *table,
27    whose indices are 0..2^bits-1.  work is a writable array of at least
28    lens shorts, which is used as a work area.  type is the type of code
29    to be generated, CODES, LENS, or DISTS.  On return, zero is success,
30    -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
31    on return points to the next available entry's address.  bits is the
32    requested root table index bits, and on return it is the actual root
33    table index bits.  It will differ if the request is greater than the
34    longest code or if it is less than the shortest code.
35  */
inflate_table(type,lens,codes,table,bits,work)36 int ZLIB_INTERNAL inflate_table(type, lens, codes, table, bits, work)
37 codetype type;
38 unsigned short FAR *lens;
39 unsigned codes;
40 code FAR * FAR *table;
41 unsigned FAR *bits;
42 unsigned short FAR *work;
43 {
44     unsigned len;               /* a code's length in bits */
45     unsigned sym;               /* index of code symbols */
46     unsigned min, max;          /* minimum and maximum code lengths */
47     unsigned root;              /* number of index bits for root table */
48     unsigned curr;              /* number of index bits for current table */
49     unsigned drop;              /* code bits to drop for sub-table */
50     int left;                   /* number of prefix codes available */
51     unsigned used;              /* code entries in table used */
52     unsigned huff;              /* Huffman code */
53     unsigned incr;              /* for incrementing code, index */
54     unsigned fill;              /* index for replicating entries */
55     unsigned low;               /* low bits for current root entry */
56     unsigned mask;              /* mask for low root bits */
57     code here;                  /* table entry for duplication */
58     code FAR *next;             /* next available space in table */
59     const unsigned short FAR *base;     /* base value table to use */
60     const unsigned short FAR *extra;    /* extra bits table to use */
61     int end;                    /* use base and extra for symbol > end */
62     unsigned short count[MAXBITS+1];    /* number of codes of each length */
63     unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
64     static const unsigned short lbase[31] = { /* Length codes 257..285 base */
65         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
66         35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
67     static const unsigned short lext[31] = { /* Length codes 257..285 extra */
68         16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
69         19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 72, 78};
70     static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
71         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
72         257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
73         8193, 12289, 16385, 24577, 0, 0};
74     static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
75         16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
76         23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
77         28, 28, 29, 29, 64, 64};
78 
79     /*
80        Process a set of code lengths to create a canonical Huffman code.  The
81        code lengths are lens[0..codes-1].  Each length corresponds to the
82        symbols 0..codes-1.  The Huffman code is generated by first sorting the
83        symbols by length from short to long, and retaining the symbol order
84        for codes with equal lengths.  Then the code starts with all zero bits
85        for the first code of the shortest length, and the codes are integer
86        increments for the same length, and zeros are appended as the length
87        increases.  For the deflate format, these bits are stored backwards
88        from their more natural integer increment ordering, and so when the
89        decoding tables are built in the large loop below, the integer codes
90        are incremented backwards.
91 
92        This routine assumes, but does not check, that all of the entries in
93        lens[] are in the range 0..MAXBITS.  The caller must assure this.
94        1..MAXBITS is interpreted as that code length.  zero means that that
95        symbol does not occur in this code.
96 
97        The codes are sorted by computing a count of codes for each length,
98        creating from that a table of starting indices for each length in the
99        sorted table, and then entering the symbols in order in the sorted
100        table.  The sorted table is work[], with that space being provided by
101        the caller.
102 
103        The length counts are used for other purposes as well, i.e. finding
104        the minimum and maximum length codes, determining if there are any
105        codes at all, checking for a valid set of lengths, and looking ahead
106        at length counts to determine sub-table sizes when building the
107        decoding tables.
108      */
109 
110     /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
111     for (len = 0; len <= MAXBITS; len++)
112         count[len] = 0;
113     for (sym = 0; sym < codes; sym++)
114         count[lens[sym]]++;
115 
116     /* bound code lengths, force root to be within code lengths */
117     root = *bits;
118     for (max = MAXBITS; max >= 1; max--)
119         if (count[max] != 0) break;
120     if (root > max) root = max;
121     if (max == 0) {                     /* no symbols to code at all */
122         here.op = (unsigned char)64;    /* invalid code marker */
123         here.bits = (unsigned char)1;
124         here.val = (unsigned short)0;
125         *(*table)++ = here;             /* make a table to force an error */
126         *(*table)++ = here;
127         *bits = 1;
128         return 0;     /* no symbols, but wait for decoding to report error */
129     }
130     for (min = 1; min < max; min++)
131         if (count[min] != 0) break;
132     if (root < min) root = min;
133 
134     /* check for an over-subscribed or incomplete set of lengths */
135     left = 1;
136     for (len = 1; len <= MAXBITS; len++) {
137         left <<= 1;
138         left -= count[len];
139         if (left < 0) return -1;        /* over-subscribed */
140     }
141     if (left > 0 && (type == CODES || max != 1))
142         return -1;                      /* incomplete set */
143 
144     /* generate offsets into symbol table for each length for sorting */
145     offs[1] = 0;
146     for (len = 1; len < MAXBITS; len++)
147         offs[len + 1] = offs[len] + count[len];
148 
149     /* sort symbols by length, by symbol order within each length */
150     for (sym = 0; sym < codes; sym++)
151         if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
152 
153     /*
154        Create and fill in decoding tables.  In this loop, the table being
155        filled is at next and has curr index bits.  The code being used is huff
156        with length len.  That code is converted to an index by dropping drop
157        bits off of the bottom.  For codes where len is less than drop + curr,
158        those top drop + curr - len bits are incremented through all values to
159        fill the table with replicated entries.
160 
161        root is the number of index bits for the root table.  When len exceeds
162        root, sub-tables are created pointed to by the root entry with an index
163        of the low root bits of huff.  This is saved in low to check for when a
164        new sub-table should be started.  drop is zero when the root table is
165        being filled, and drop is root when sub-tables are being filled.
166 
167        When a new sub-table is needed, it is necessary to look ahead in the
168        code lengths to determine what size sub-table is needed.  The length
169        counts are used for this, and so count[] is decremented as codes are
170        entered in the tables.
171 
172        used keeps track of how many table entries have been allocated from the
173        provided *table space.  It is checked for LENS and DIST tables against
174        the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
175        the initial root table size constants.  See the comments in inftrees.h
176        for more information.
177 
178        sym increments through all symbols, and the loop terminates when
179        all codes of length max, i.e. all codes, have been processed.  This
180        routine permits incomplete codes, so another loop after this one fills
181        in the rest of the decoding tables with invalid code markers.
182      */
183 
184     /* set up for code type */
185     switch (type) {
186     case CODES:
187         base = extra = work;    /* dummy value--not used */
188         end = 19;
189         break;
190     case LENS:
191         base = lbase;
192         base -= 257;
193         extra = lext;
194         extra -= 257;
195         end = 256;
196         break;
197     default:            /* DISTS */
198         base = dbase;
199         extra = dext;
200         end = -1;
201     }
202 
203     /* initialize state for loop */
204     huff = 0;                   /* starting code */
205     sym = 0;                    /* starting code symbol */
206     len = min;                  /* starting code length */
207     next = *table;              /* current table to fill in */
208     curr = root;                /* current table index bits */
209     drop = 0;                   /* current bits to drop from code for index */
210     low = (unsigned)(-1);       /* trigger new sub-table when len > root */
211     used = 1U << root;          /* use root table entries */
212     mask = used - 1;            /* mask for comparing low */
213 
214     /* check available table space */
215     if ((type == LENS && used > ENOUGH_LENS) ||
216         (type == DISTS && used > ENOUGH_DISTS))
217         return 1;
218 
219     /* process all codes and make table entries */
220     for (;;) {
221         /* create table entry */
222         here.bits = (unsigned char)(len - drop);
223         if ((int)(work[sym]) < end) {
224             here.op = (unsigned char)0;
225             here.val = work[sym];
226         }
227         else if ((int)(work[sym]) > end) {
228             here.op = (unsigned char)(extra[work[sym]]);
229             here.val = base[work[sym]];
230         }
231         else {
232             here.op = (unsigned char)(32 + 64);         /* end of block */
233             here.val = 0;
234         }
235 
236         /* replicate for those indices with low len bits equal to huff */
237         incr = 1U << (len - drop);
238         fill = 1U << curr;
239         min = fill;                 /* save offset to next table */
240         do {
241             fill -= incr;
242             next[(huff >> drop) + fill] = here;
243         } while (fill != 0);
244 
245         /* backwards increment the len-bit code huff */
246         incr = 1U << (len - 1);
247         while (huff & incr)
248             incr >>= 1;
249         if (incr != 0) {
250             huff &= incr - 1;
251             huff += incr;
252         }
253         else
254             huff = 0;
255 
256         /* go to next symbol, update count, len */
257         sym++;
258         if (--(count[len]) == 0) {
259             if (len == max) break;
260             len = lens[work[sym]];
261         }
262 
263         /* create new sub-table if needed */
264         if (len > root && (huff & mask) != low) {
265             /* if first time, transition to sub-tables */
266             if (drop == 0)
267                 drop = root;
268 
269             /* increment past last table */
270             next += min;            /* here min is 1 << curr */
271 
272             /* determine length of next table */
273             curr = len - drop;
274             left = (int)(1 << curr);
275             while (curr + drop < max) {
276                 left -= count[curr + drop];
277                 if (left <= 0) break;
278                 curr++;
279                 left <<= 1;
280             }
281 
282             /* check for enough space */
283             used += 1U << curr;
284             if ((type == LENS && used > ENOUGH_LENS) ||
285                 (type == DISTS && used > ENOUGH_DISTS))
286                 return 1;
287 
288             /* point entry in root table to sub-table */
289             low = huff & mask;
290             (*table)[low].op = (unsigned char)curr;
291             (*table)[low].bits = (unsigned char)root;
292             (*table)[low].val = (unsigned short)(next - *table);
293         }
294     }
295 
296     /* fill in remaining table entry if code is incomplete (guaranteed to have
297        at most one remaining entry, since if the code is incomplete, the
298        maximum code length that was allowed to get this far is one bit) */
299     if (huff != 0) {
300         here.op = (unsigned char)64;            /* invalid code marker */
301         here.bits = (unsigned char)(len - drop);
302         here.val = (unsigned short)0;
303         next[huff] = here;
304     }
305 
306     /* set return parameters */
307     *table += used;
308     *bits = root;
309     return 0;
310 }
311