]>
Commit | Line | Data |
---|---|---|
37afa456 PM |
1 | commit 5c44459c3b28a9bd3283aaceab7c615f8020c531 |
2 | Author: Mark Adler <madler@alumni.caltech.edu> | |
3 | Date: Tue Apr 17 22:09:22 2018 -0700 | |
4 | ||
5 | Fix a bug that can crash deflate on some input when using Z_FIXED. | |
6 | ||
7 | This bug was reported by Danilo Ramos of Eideticom, Inc. It has | |
8 | lain in wait 13 years before being found! The bug was introduced | |
9 | in zlib 1.2.2.2, with the addition of the Z_FIXED option. That | |
10 | option forces the use of fixed Huffman codes. For rare inputs with | |
11 | a large number of distant matches, the pending buffer into which | |
12 | the compressed data is written can overwrite the distance symbol | |
13 | table which it overlays. That results in corrupted output due to | |
14 | invalid distances, and can result in out-of-bound accesses, | |
15 | crashing the application. | |
16 | ||
17 | The fix here combines the distance buffer and literal/length | |
18 | buffers into a single symbol buffer. Now three bytes of pending | |
19 | buffer space are opened up for each literal or length/distance | |
20 | pair consumed, instead of the previous two bytes. This assures | |
21 | that the pending buffer cannot overwrite the symbol table, since | |
22 | the maximum fixed code compressed length/distance is 31 bits, and | |
23 | since there are four bytes of pending space for every three bytes | |
24 | of symbol space. | |
25 | ||
26 | diff --git a/deflate.c b/deflate.c | |
27 | index 425babc..19cba87 100644 | |
28 | --- a/deflate.c | |
29 | +++ b/deflate.c | |
30 | @@ -255,11 +255,6 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, | |
31 | int wrap = 1; | |
32 | static const char my_version[] = ZLIB_VERSION; | |
33 | ||
34 | - ushf *overlay; | |
35 | - /* We overlay pending_buf and d_buf+l_buf. This works since the average | |
36 | - * output size for (length,distance) codes is <= 24 bits. | |
37 | - */ | |
38 | - | |
39 | if (version == Z_NULL || version[0] != my_version[0] || | |
40 | stream_size != sizeof(z_stream)) { | |
41 | return Z_VERSION_ERROR; | |
42 | @@ -329,9 +324,47 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, | |
43 | ||
44 | s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ | |
45 | ||
46 | - overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2); | |
47 | - s->pending_buf = (uchf *) overlay; | |
48 | - s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); | |
49 | + /* We overlay pending_buf and sym_buf. This works since the average size | |
50 | + * for length/distance pairs over any compressed block is assured to be 31 | |
51 | + * bits or less. | |
52 | + * | |
53 | + * Analysis: The longest fixed codes are a length code of 8 bits plus 5 | |
54 | + * extra bits, for lengths 131 to 257. The longest fixed distance codes are | |
55 | + * 5 bits plus 13 extra bits, for distances 16385 to 32768. The longest | |
56 | + * possible fixed-codes length/distance pair is then 31 bits total. | |
57 | + * | |
58 | + * sym_buf starts one-fourth of the way into pending_buf. So there are | |
59 | + * three bytes in sym_buf for every four bytes in pending_buf. Each symbol | |
60 | + * in sym_buf is three bytes -- two for the distance and one for the | |
61 | + * literal/length. As each symbol is consumed, the pointer to the next | |
62 | + * sym_buf value to read moves forward three bytes. From that symbol, up to | |
63 | + * 31 bits are written to pending_buf. The closest the written pending_buf | |
64 | + * bits gets to the next sym_buf symbol to read is just before the last | |
65 | + * code is written. At that time, 31*(n-2) bits have been written, just | |
66 | + * after 24*(n-2) bits have been consumed from sym_buf. sym_buf starts at | |
67 | + * 8*n bits into pending_buf. (Note that the symbol buffer fills when n-1 | |
68 | + * symbols are written.) The closest the writing gets to what is unread is | |
69 | + * then n+14 bits. Here n is lit_bufsize, which is 16384 by default, and | |
70 | + * can range from 128 to 32768. | |
71 | + * | |
72 | + * Therefore, at a minimum, there are 142 bits of space between what is | |
73 | + * written and what is read in the overlain buffers, so the symbols cannot | |
74 | + * be overwritten by the compressed data. That space is actually 139 bits, | |
75 | + * due to the three-bit fixed-code block header. | |
76 | + * | |
77 | + * That covers the case where either Z_FIXED is specified, forcing fixed | |
78 | + * codes, or when the use of fixed codes is chosen, because that choice | |
79 | + * results in a smaller compressed block than dynamic codes. That latter | |
80 | + * condition then assures that the above analysis also covers all dynamic | |
81 | + * blocks. A dynamic-code block will only be chosen to be emitted if it has | |
82 | + * fewer bits than a fixed-code block would for the same set of symbols. | |
83 | + * Therefore its average symbol length is assured to be less than 31. So | |
84 | + * the compressed data for a dynamic block also cannot overwrite the | |
85 | + * symbols from which it is being constructed. | |
86 | + */ | |
87 | + | |
88 | + s->pending_buf = (uchf *) ZALLOC(strm, s->lit_bufsize, 4); | |
89 | + s->pending_buf_size = (ulg)s->lit_bufsize * 4; | |
90 | ||
91 | if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL || | |
92 | s->pending_buf == Z_NULL) { | |
93 | @@ -340,8 +373,12 @@ int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy, | |
94 | deflateEnd (strm); | |
95 | return Z_MEM_ERROR; | |
96 | } | |
97 | - s->d_buf = overlay + s->lit_bufsize/sizeof(ush); | |
98 | - s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; | |
99 | + s->sym_buf = s->pending_buf + s->lit_bufsize; | |
100 | + s->sym_end = (s->lit_bufsize - 1) * 3; | |
101 | + /* We avoid equality with lit_bufsize*3 because of wraparound at 64K | |
102 | + * on 16 bit machines and because stored blocks are restricted to | |
103 | + * 64K-1 bytes. | |
104 | + */ | |
105 | ||
106 | s->level = level; | |
107 | s->strategy = strategy; | |
108 | @@ -552,7 +589,7 @@ int ZEXPORT deflatePrime (strm, bits, value) | |
109 | ||
110 | if (deflateStateCheck(strm)) return Z_STREAM_ERROR; | |
111 | s = strm->state; | |
112 | - if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3)) | |
113 | + if (s->sym_buf < s->pending_out + ((Buf_size + 7) >> 3)) | |
114 | return Z_BUF_ERROR; | |
115 | do { | |
116 | put = Buf_size - s->bi_valid; | |
117 | @@ -1113,7 +1150,6 @@ int ZEXPORT deflateCopy (dest, source) | |
118 | #else | |
119 | deflate_state *ds; | |
120 | deflate_state *ss; | |
121 | - ushf *overlay; | |
122 | ||
123 | ||
124 | if (deflateStateCheck(source) || dest == Z_NULL) { | |
125 | @@ -1133,8 +1169,7 @@ int ZEXPORT deflateCopy (dest, source) | |
126 | ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte)); | |
127 | ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos)); | |
128 | ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos)); | |
129 | - overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2); | |
130 | - ds->pending_buf = (uchf *) overlay; | |
131 | + ds->pending_buf = (uchf *) ZALLOC(dest, ds->lit_bufsize, 4); | |
132 | ||
133 | if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL || | |
134 | ds->pending_buf == Z_NULL) { | |
135 | @@ -1148,8 +1183,7 @@ int ZEXPORT deflateCopy (dest, source) | |
136 | zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size); | |
137 | ||
138 | ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf); | |
139 | - ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush); | |
140 | - ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize; | |
141 | + ds->sym_buf = ds->pending_buf + ds->lit_bufsize; | |
142 | ||
143 | ds->l_desc.dyn_tree = ds->dyn_ltree; | |
144 | ds->d_desc.dyn_tree = ds->dyn_dtree; | |
145 | @@ -1925,7 +1959,7 @@ local block_state deflate_fast(s, flush) | |
146 | FLUSH_BLOCK(s, 1); | |
147 | return finish_done; | |
148 | } | |
149 | - if (s->last_lit) | |
150 | + if (s->sym_next) | |
151 | FLUSH_BLOCK(s, 0); | |
152 | return block_done; | |
153 | } | |
154 | @@ -2056,7 +2090,7 @@ local block_state deflate_slow(s, flush) | |
155 | FLUSH_BLOCK(s, 1); | |
156 | return finish_done; | |
157 | } | |
158 | - if (s->last_lit) | |
159 | + if (s->sym_next) | |
160 | FLUSH_BLOCK(s, 0); | |
161 | return block_done; | |
162 | } | |
163 | @@ -2131,7 +2165,7 @@ local block_state deflate_rle(s, flush) | |
164 | FLUSH_BLOCK(s, 1); | |
165 | return finish_done; | |
166 | } | |
167 | - if (s->last_lit) | |
168 | + if (s->sym_next) | |
169 | FLUSH_BLOCK(s, 0); | |
170 | return block_done; | |
171 | } | |
172 | @@ -2170,7 +2204,7 @@ local block_state deflate_huff(s, flush) | |
173 | FLUSH_BLOCK(s, 1); | |
174 | return finish_done; | |
175 | } | |
176 | - if (s->last_lit) | |
177 | + if (s->sym_next) | |
178 | FLUSH_BLOCK(s, 0); | |
179 | return block_done; | |
180 | } | |
181 | diff --git a/deflate.h b/deflate.h | |
182 | index 23ecdd3..d4cf1a9 100644 | |
183 | --- a/deflate.h | |
184 | +++ b/deflate.h | |
185 | @@ -217,7 +217,7 @@ typedef struct internal_state { | |
186 | /* Depth of each subtree used as tie breaker for trees of equal frequency | |
187 | */ | |
188 | ||
189 | - uchf *l_buf; /* buffer for literals or lengths */ | |
190 | + uchf *sym_buf; /* buffer for distances and literals/lengths */ | |
191 | ||
192 | uInt lit_bufsize; | |
193 | /* Size of match buffer for literals/lengths. There are 4 reasons for | |
194 | @@ -239,13 +239,8 @@ typedef struct internal_state { | |
195 | * - I can't count above 4 | |
196 | */ | |
197 | ||
198 | - uInt last_lit; /* running index in l_buf */ | |
199 | - | |
200 | - ushf *d_buf; | |
201 | - /* Buffer for distances. To simplify the code, d_buf and l_buf have | |
202 | - * the same number of elements. To use different lengths, an extra flag | |
203 | - * array would be necessary. | |
204 | - */ | |
205 | + uInt sym_next; /* running index in sym_buf */ | |
206 | + uInt sym_end; /* symbol table full when sym_next reaches this */ | |
207 | ||
208 | ulg opt_len; /* bit length of current block with optimal trees */ | |
209 | ulg static_len; /* bit length of current block with static trees */ | |
210 | @@ -325,20 +320,22 @@ void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf, | |
211 | ||
212 | # define _tr_tally_lit(s, c, flush) \ | |
213 | { uch cc = (c); \ | |
214 | - s->d_buf[s->last_lit] = 0; \ | |
215 | - s->l_buf[s->last_lit++] = cc; \ | |
216 | + s->sym_buf[s->sym_next++] = 0; \ | |
217 | + s->sym_buf[s->sym_next++] = 0; \ | |
218 | + s->sym_buf[s->sym_next++] = cc; \ | |
219 | s->dyn_ltree[cc].Freq++; \ | |
220 | - flush = (s->last_lit == s->lit_bufsize-1); \ | |
221 | + flush = (s->sym_next == s->sym_end); \ | |
222 | } | |
223 | # define _tr_tally_dist(s, distance, length, flush) \ | |
224 | { uch len = (uch)(length); \ | |
225 | ush dist = (ush)(distance); \ | |
226 | - s->d_buf[s->last_lit] = dist; \ | |
227 | - s->l_buf[s->last_lit++] = len; \ | |
228 | + s->sym_buf[s->sym_next++] = dist; \ | |
229 | + s->sym_buf[s->sym_next++] = dist >> 8; \ | |
230 | + s->sym_buf[s->sym_next++] = len; \ | |
231 | dist--; \ | |
232 | s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \ | |
233 | s->dyn_dtree[d_code(dist)].Freq++; \ | |
234 | - flush = (s->last_lit == s->lit_bufsize-1); \ | |
235 | + flush = (s->sym_next == s->sym_end); \ | |
236 | } | |
237 | #else | |
238 | # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c) | |
239 | diff --git a/trees.c b/trees.c | |
240 | index 4f4a650..decaeb7 100644 | |
241 | --- a/trees.c | |
242 | +++ b/trees.c | |
243 | @@ -416,7 +416,7 @@ local void init_block(s) | |
244 | ||
245 | s->dyn_ltree[END_BLOCK].Freq = 1; | |
246 | s->opt_len = s->static_len = 0L; | |
247 | - s->last_lit = s->matches = 0; | |
248 | + s->sym_next = s->matches = 0; | |
249 | } | |
250 | ||
251 | #define SMALLEST 1 | |
252 | @@ -948,7 +948,7 @@ void ZLIB_INTERNAL _tr_flush_block(s, buf, stored_len, last) | |
253 | ||
254 | Tracev((stderr, "\nopt %lu(%lu) stat %lu(%lu) stored %lu lit %u ", | |
255 | opt_lenb, s->opt_len, static_lenb, s->static_len, stored_len, | |
256 | - s->last_lit)); | |
257 | + s->sym_next / 3)); | |
258 | ||
259 | if (static_lenb <= opt_lenb) opt_lenb = static_lenb; | |
260 | ||
261 | @@ -1017,8 +1017,9 @@ int ZLIB_INTERNAL _tr_tally (s, dist, lc) | |
262 | unsigned dist; /* distance of matched string */ | |
263 | unsigned lc; /* match length-MIN_MATCH or unmatched char (if dist==0) */ | |
264 | { | |
265 | - s->d_buf[s->last_lit] = (ush)dist; | |
266 | - s->l_buf[s->last_lit++] = (uch)lc; | |
267 | + s->sym_buf[s->sym_next++] = dist; | |
268 | + s->sym_buf[s->sym_next++] = dist >> 8; | |
269 | + s->sym_buf[s->sym_next++] = lc; | |
270 | if (dist == 0) { | |
271 | /* lc is the unmatched char */ | |
272 | s->dyn_ltree[lc].Freq++; | |
273 | @@ -1033,30 +1034,7 @@ int ZLIB_INTERNAL _tr_tally (s, dist, lc) | |
274 | s->dyn_ltree[_length_code[lc]+LITERALS+1].Freq++; | |
275 | s->dyn_dtree[d_code(dist)].Freq++; | |
276 | } | |
277 | - | |
278 | -#ifdef TRUNCATE_BLOCK | |
279 | - /* Try to guess if it is profitable to stop the current block here */ | |
280 | - if ((s->last_lit & 0x1fff) == 0 && s->level > 2) { | |
281 | - /* Compute an upper bound for the compressed length */ | |
282 | - ulg out_length = (ulg)s->last_lit*8L; | |
283 | - ulg in_length = (ulg)((long)s->strstart - s->block_start); | |
284 | - int dcode; | |
285 | - for (dcode = 0; dcode < D_CODES; dcode++) { | |
286 | - out_length += (ulg)s->dyn_dtree[dcode].Freq * | |
287 | - (5L+extra_dbits[dcode]); | |
288 | - } | |
289 | - out_length >>= 3; | |
290 | - Tracev((stderr,"\nlast_lit %u, in %ld, out ~%ld(%ld%%) ", | |
291 | - s->last_lit, in_length, out_length, | |
292 | - 100L - out_length*100L/in_length)); | |
293 | - if (s->matches < s->last_lit/2 && out_length < in_length/2) return 1; | |
294 | - } | |
295 | -#endif | |
296 | - return (s->last_lit == s->lit_bufsize-1); | |
297 | - /* We avoid equality with lit_bufsize because of wraparound at 64K | |
298 | - * on 16 bit machines and because stored blocks are restricted to | |
299 | - * 64K-1 bytes. | |
300 | - */ | |
301 | + return (s->sym_next == s->sym_end); | |
302 | } | |
303 | ||
304 | /* =========================================================================== | |
305 | @@ -1069,13 +1047,14 @@ local void compress_block(s, ltree, dtree) | |
306 | { | |
307 | unsigned dist; /* distance of matched string */ | |
308 | int lc; /* match length or unmatched char (if dist == 0) */ | |
309 | - unsigned lx = 0; /* running index in l_buf */ | |
310 | + unsigned sx = 0; /* running index in sym_buf */ | |
311 | unsigned code; /* the code to send */ | |
312 | int extra; /* number of extra bits to send */ | |
313 | ||
314 | - if (s->last_lit != 0) do { | |
315 | - dist = s->d_buf[lx]; | |
316 | - lc = s->l_buf[lx++]; | |
317 | + if (s->sym_next != 0) do { | |
318 | + dist = s->sym_buf[sx++] & 0xff; | |
319 | + dist += (unsigned)(s->sym_buf[sx++] & 0xff) << 8; | |
320 | + lc = s->sym_buf[sx++]; | |
321 | if (dist == 0) { | |
322 | send_code(s, lc, ltree); /* send a literal byte */ | |
323 | Tracecv(isgraph(lc), (stderr," '%c' ", lc)); | |
324 | @@ -1100,11 +1079,10 @@ local void compress_block(s, ltree, dtree) | |
325 | } | |
326 | } /* literal or match pair ? */ | |
327 | ||
328 | - /* Check that the overlay between pending_buf and d_buf+l_buf is ok: */ | |
329 | - Assert((uInt)(s->pending) < s->lit_bufsize + 2*lx, | |
330 | - "pendingBuf overflow"); | |
331 | + /* Check that the overlay between pending_buf and sym_buf is ok: */ | |
332 | + Assert(s->pending < s->lit_bufsize + sx, "pendingBuf overflow"); | |
333 | ||
334 | - } while (lx < s->last_lit); | |
335 | + } while (sx < s->sym_next); | |
336 | ||
337 | send_code(s, END_BLOCK, ltree); | |
338 | } |