1/* inftrees.c -- generate Huffman trees for efficient decoding
2 * Copyright (C) 1995-2017 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#define MAXBITS 15
10
11const char inflate_copyright[] =
12   " inflate 1.2.11 Copyright 1995-2017 Mark Adler ";
13/*
14  If you use the zlib library in a product, an acknowledgment is welcome
15  in the documentation of your product. If for some reason you cannot
16  include such an acknowledgment, I would appreciate that you keep this
17  copyright string in the executable of your product.
18 */
19
20/*
21   Build a set of tables to decode the provided canonical Huffman code.
22   The code lengths are lens[0..codes-1].  The result starts at *table,
23   whose indices are 0..2^bits-1.  work is a writable array of at least
24   lens shorts, which is used as a work area.  type is the type of code
25   to be generated, CODES, LENS, or DISTS.  On return, zero is success,
26   -1 is an invalid code, and +1 means that ENOUGH isn't enough.  table
27   on return points to the next available entry's address.  bits is the
28   requested root table index bits, and on return it is the actual root
29   table index bits.  It will differ if the request is greater than the
30   longest code or if it is less than the shortest code.
31 */
32int ZLIB_INTERNAL inflate_table(codetype type, unsigned short FAR *lens,
33    unsigned codes, code FAR * FAR *table, unsigned FAR *bits,
34    unsigned short FAR *work)
35{
36    unsigned len;               /* a code's length in bits */
37    unsigned sym;               /* index of code symbols */
38    unsigned min, max;          /* minimum and maximum code lengths */
39    unsigned root;              /* number of index bits for root table */
40    unsigned curr;              /* number of index bits for current table */
41    unsigned drop;              /* code bits to drop for sub-table */
42    int left;                   /* number of prefix codes available */
43    unsigned used;              /* code entries in table used */
44    unsigned huff;              /* Huffman code */
45    unsigned incr;              /* for incrementing code, index */
46    unsigned fill;              /* index for replicating entries */
47    unsigned low;               /* low bits for current root entry */
48    unsigned mask;              /* mask for low root bits */
49    code here;                  /* table entry for duplication */
50    code FAR *next;             /* next available space in table */
51    const unsigned short FAR *base;     /* base value table to use */
52    const unsigned short FAR *extra;    /* extra bits table to use */
53    unsigned match;             /* use base and extra for symbol >= match */
54    unsigned short count[MAXBITS+1];    /* number of codes of each length */
55    unsigned short offs[MAXBITS+1];     /* offsets in table for each length */
56    static const unsigned short lbase[31] = { /* Length codes 257..285 base */
57        3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
58        35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
59    static const unsigned short lext[31] = { /* Length codes 257..285 extra */
60        16, 16, 16, 16, 16, 16, 16, 16, 17, 17, 17, 17, 18, 18, 18, 18,
61        19, 19, 19, 19, 20, 20, 20, 20, 21, 21, 21, 21, 16, 77, 202};
62    static const unsigned short dbase[32] = { /* Distance codes 0..29 base */
63        1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
64        257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
65        8193, 12289, 16385, 24577, 0, 0};
66    static const unsigned short dext[32] = { /* Distance codes 0..29 extra */
67        16, 16, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, 21, 21, 22, 22,
68        23, 23, 24, 24, 25, 25, 26, 26, 27, 27,
69        28, 28, 29, 29, 64, 64};
70
71    /*
72       Process a set of code lengths to create a canonical Huffman code.  The
73       code lengths are lens[0..codes-1].  Each length corresponds to the
74       symbols 0..codes-1.  The Huffman code is generated by first sorting the
75       symbols by length from short to long, and retaining the symbol order
76       for codes with equal lengths.  Then the code starts with all zero bits
77       for the first code of the shortest length, and the codes are integer
78       increments for the same length, and zeros are appended as the length
79       increases.  For the deflate format, these bits are stored backwards
80       from their more natural integer increment ordering, and so when the
81       decoding tables are built in the large loop below, the integer codes
82       are incremented backwards.
83
84       This routine assumes, but does not check, that all of the entries in
85       lens[] are in the range 0..MAXBITS.  The caller must assure this.
86       1..MAXBITS is interpreted as that code length.  zero means that that
87       symbol does not occur in this code.
88
89       The codes are sorted by computing a count of codes for each length,
90       creating from that a table of starting indices for each length in the
91       sorted table, and then entering the symbols in order in the sorted
92       table.  The sorted table is work[], with that space being provided by
93       the caller.
94
95       The length counts are used for other purposes as well, i.e. finding
96       the minimum and maximum length codes, determining if there are any
97       codes at all, checking for a valid set of lengths, and looking ahead
98       at length counts to determine sub-table sizes when building the
99       decoding tables.
100     */
101
102    /* accumulate lengths for codes (assumes lens[] all in 0..MAXBITS) */
103    for (len = 0; len <= MAXBITS; len++)
104        count[len] = 0;
105    for (sym = 0; sym < codes; sym++)
106        count[lens[sym]]++;
107
108    /* bound code lengths, force root to be within code lengths */
109    root = *bits;
110    for (max = MAXBITS; max >= 1; max--)
111        if (count[max] != 0) break;
112    if (root > max) root = max;
113    if (max == 0) {                     /* no symbols to code at all */
114        here.op = (unsigned char)64;    /* invalid code marker */
115        here.bits = (unsigned char)1;
116        here.val = (unsigned short)0;
117        *(*table)++ = here;             /* make a table to force an error */
118        *(*table)++ = here;
119        *bits = 1;
120        return 0;     /* no symbols, but wait for decoding to report error */
121    }
122    for (min = 1; min < max; min++)
123        if (count[min] != 0) break;
124    if (root < min) root = min;
125
126    /* check for an over-subscribed or incomplete set of lengths */
127    left = 1;
128    for (len = 1; len <= MAXBITS; len++) {
129        left <<= 1;
130        left -= count[len];
131        if (left < 0) return -1;        /* over-subscribed */
132    }
133    if (left > 0 && (type == CODES || max != 1))
134        return -1;                      /* incomplete set */
135
136    /* generate offsets into symbol table for each length for sorting */
137    offs[1] = 0;
138    for (len = 1; len < MAXBITS; len++)
139        offs[len + 1] = offs[len] + count[len];
140
141    /* sort symbols by length, by symbol order within each length */
142    for (sym = 0; sym < codes; sym++)
143        if (lens[sym] != 0) work[offs[lens[sym]]++] = (unsigned short)sym;
144
145    /*
146       Create and fill in decoding tables.  In this loop, the table being
147       filled is at next and has curr index bits.  The code being used is huff
148       with length len.  That code is converted to an index by dropping drop
149       bits off of the bottom.  For codes where len is less than drop + curr,
150       those top drop + curr - len bits are incremented through all values to
151       fill the table with replicated entries.
152
153       root is the number of index bits for the root table.  When len exceeds
154       root, sub-tables are created pointed to by the root entry with an index
155       of the low root bits of huff.  This is saved in low to check for when a
156       new sub-table should be started.  drop is zero when the root table is
157       being filled, and drop is root when sub-tables are being filled.
158
159       When a new sub-table is needed, it is necessary to look ahead in the
160       code lengths to determine what size sub-table is needed.  The length
161       counts are used for this, and so count[] is decremented as codes are
162       entered in the tables.
163
164       used keeps track of how many table entries have been allocated from the
165       provided *table space.  It is checked for LENS and DIST tables against
166       the constants ENOUGH_LENS and ENOUGH_DISTS to guard against changes in
167       the initial root table size constants.  See the comments in inftrees.h
168       for more information.
169
170       sym increments through all symbols, and the loop terminates when
171       all codes of length max, i.e. all codes, have been processed.  This
172       routine permits incomplete codes, so another loop after this one fills
173       in the rest of the decoding tables with invalid code markers.
174     */
175
176    /* set up for code type */
177    switch (type) {
178    case CODES:
179        base = extra = work;    /* dummy value--not used */
180        match = 20;
181        break;
182    case LENS:
183        base = lbase;
184        extra = lext;
185        match = 257;
186        break;
187    default:    /* DISTS */
188        base = dbase;
189        extra = dext;
190        match = 0;
191    }
192
193    /* initialize state for loop */
194    huff = 0;                   /* starting code */
195    sym = 0;                    /* starting code symbol */
196    len = min;                  /* starting code length */
197    next = *table;              /* current table to fill in */
198    curr = root;                /* current table index bits */
199    drop = 0;                   /* current bits to drop from code for index */
200    low = (unsigned)(-1);       /* trigger new sub-table when len > root */
201    used = 1U << root;          /* use root table entries */
202    mask = used - 1;            /* mask for comparing low */
203
204    /* check available table space */
205    if ((type == LENS && used > ENOUGH_LENS) ||
206        (type == DISTS && used > ENOUGH_DISTS))
207        return 1;
208
209    /* process all codes and make table entries */
210    for (;;) {
211        /* create table entry */
212        here.bits = (unsigned char)(len - drop);
213        if (work[sym] + 1U < match) {
214            here.op = (unsigned char)0;
215            here.val = work[sym];
216        }
217        else if (work[sym] >= match) {
218            here.op = (unsigned char)(extra[work[sym] - match]);
219            here.val = base[work[sym] - match];
220        }
221        else {
222            here.op = (unsigned char)(32 + 64);         /* end of block */
223            here.val = 0;
224        }
225
226        /* replicate for those indices with low len bits equal to huff */
227        incr = 1U << (len - drop);
228        fill = 1U << curr;
229        min = fill;                 /* save offset to next table */
230        do {
231            fill -= incr;
232            next[(huff >> drop) + fill] = here;
233        } while (fill != 0);
234
235        /* backwards increment the len-bit code huff */
236        incr = 1U << (len - 1);
237        while (huff & incr)
238            incr >>= 1;
239        if (incr != 0) {
240            huff &= incr - 1;
241            huff += incr;
242        }
243        else
244            huff = 0;
245
246        /* go to next symbol, update count, len */
247        sym++;
248        if (--(count[len]) == 0) {
249            if (len == max) break;
250            len = lens[work[sym]];
251        }
252
253        /* create new sub-table if needed */
254        if (len > root && (huff & mask) != low) {
255            /* if first time, transition to sub-tables */
256            if (drop == 0)
257                drop = root;
258
259            /* increment past last table */
260            next += min;            /* here min is 1 << curr */
261
262            /* determine length of next table */
263            curr = len - drop;
264            left = (int)(1 << curr);
265            while (curr + drop < max) {
266                left -= count[curr + drop];
267                if (left <= 0) break;
268                curr++;
269                left <<= 1;
270            }
271
272            /* check for enough space */
273            used += 1U << curr;
274            if ((type == LENS && used > ENOUGH_LENS) ||
275                (type == DISTS && used > ENOUGH_DISTS))
276                return 1;
277
278            /* point entry in root table to sub-table */
279            low = huff & mask;
280            (*table)[low].op = (unsigned char)curr;
281            (*table)[low].bits = (unsigned char)root;
282            (*table)[low].val = (unsigned short)(next - *table);
283        }
284    }
285
286    /* fill in remaining table entry if code is incomplete (guaranteed to have
287       at most one remaining entry, since if the code is incomplete, the
288       maximum code length that was allowed to get this far is one bit) */
289    if (huff != 0) {
290        here.op = (unsigned char)64;            /* invalid code marker */
291        here.bits = (unsigned char)(len - drop);
292        here.val = (unsigned short)0;
293        next[huff] = here;
294    }
295
296    /* set return parameters */
297    *table += used;
298    *bits = root;
299    return 0;
300}
301