| 1 | /* |
| 2 | * Zlib (RFC1950 / RFC1951) compression for PuTTY. |
| 3 | * |
| 4 | * There will no doubt be criticism of my decision to reimplement |
| 5 | * Zlib compression from scratch instead of using the existing zlib |
| 6 | * code. People will cry `reinventing the wheel'; they'll claim |
| 7 | * that the `fundamental basis of OSS' is code reuse; they'll want |
| 8 | * to see a really good reason for me having chosen not to use the |
| 9 | * existing code. |
| 10 | * |
| 11 | * Well, here are my reasons. Firstly, I don't want to link the |
| 12 | * whole of zlib into the PuTTY binary; PuTTY is justifiably proud |
| 13 | * of its small size and I think zlib contains a lot of unnecessary |
| 14 | * baggage for the kind of compression that SSH requires. |
| 15 | * |
| 16 | * Secondly, I also don't like the alternative of using zlib.dll. |
| 17 | * Another thing PuTTY is justifiably proud of is its ease of |
| 18 | * installation, and the last thing I want to do is to start |
| 19 | * mandating DLLs. Not only that, but there are two _kinds_ of |
| 20 | * zlib.dll kicking around, one with C calling conventions on the |
| 21 | * exported functions and another with WINAPI conventions, and |
| 22 | * there would be a significant danger of getting the wrong one. |
| 23 | * |
| 24 | * Thirdly, there seems to be a difference of opinion on the IETF |
| 25 | * secsh mailing list about the correct way to round off a |
| 26 | * compressed packet and start the next. In particular, there's |
| 27 | * some talk of switching to a mechanism zlib isn't currently |
| 28 | * capable of supporting (see below for an explanation). Given that |
| 29 | * sort of uncertainty, I thought it might be better to have code |
| 30 | * that will support even the zlib-incompatible worst case. |
| 31 | * |
| 32 | * Fourthly, it's a _second implementation_. Second implementations |
| 33 | * are fundamentally a Good Thing in standardisation efforts. The |
| 34 | * difference of opinion mentioned above has arisen _precisely_ |
| 35 | * because there has been only one zlib implementation and |
| 36 | * everybody has used it. I don't intend that this should happen |
| 37 | * again. |
| 38 | */ |
| 39 | |
| 40 | #include <stdlib.h> |
| 41 | #include <assert.h> |
| 42 | |
| 43 | /* FIXME */ |
| 44 | #include <windows.h> |
| 45 | #include <stdio.h> |
| 46 | #include "putty.h" |
| 47 | |
| 48 | #include "ssh.h" |
| 49 | |
| 50 | /* ---------------------------------------------------------------------- |
| 51 | * Basic LZ77 code. This bit is designed modularly, so it could be |
| 52 | * ripped out and used in a different LZ77 compressor. Go to it, |
| 53 | * and good luck :-) |
| 54 | */ |
| 55 | |
| 56 | struct LZ77InternalContext; |
| 57 | struct LZ77Context { |
| 58 | struct LZ77InternalContext *ictx; |
| 59 | void *userdata; |
| 60 | void (*literal)(struct LZ77Context *ctx, unsigned char c); |
| 61 | void (*match)(struct LZ77Context *ctx, int distance, int len); |
| 62 | }; |
| 63 | |
| 64 | /* |
| 65 | * Initialise the private fields of an LZ77Context. It's up to the |
| 66 | * user to initialise the public fields. |
| 67 | */ |
| 68 | static int lz77_init(struct LZ77Context *ctx); |
| 69 | |
| 70 | /* |
| 71 | * Supply data to be compressed. Will update the private fields of |
| 72 | * the LZ77Context, and will call literal() and match() to output. |
| 73 | */ |
| 74 | static void lz77_compress(struct LZ77Context *ctx, |
| 75 | unsigned char *data, int len); |
| 76 | |
| 77 | /* |
| 78 | * Modifiable parameters. |
| 79 | */ |
| 80 | #define WINSIZE 32768 /* window size. Must be power of 2! */ |
| 81 | #define HASHMAX 2039 /* one more than max hash value */ |
| 82 | #define MAXMATCH 32 /* how many matches we track */ |
| 83 | #define HASHCHARS 3 /* how many chars make a hash */ |
| 84 | |
| 85 | /* |
| 86 | * This compressor takes a less slapdash approach than the |
| 87 | * gzip/zlib one. Rather than allowing our hash chains to fall into |
| 88 | * disuse near the far end, we keep them doubly linked so we can |
| 89 | * _find_ the far end, and then every time we add a new byte to the |
| 90 | * window (thus rolling round by one and removing the previous |
| 91 | * byte), we can carefully remove the hash chain entry. |
| 92 | */ |
| 93 | |
| 94 | #define INVALID -1 /* invalid hash _and_ invalid offset */ |
| 95 | struct WindowEntry { |
| 96 | int next, prev; /* array indices within the window */ |
| 97 | int hashval; |
| 98 | }; |
| 99 | |
| 100 | struct HashEntry { |
| 101 | int first; /* window index of first in chain */ |
| 102 | }; |
| 103 | |
| 104 | struct Match { |
| 105 | int distance, len; |
| 106 | }; |
| 107 | |
| 108 | struct LZ77InternalContext { |
| 109 | struct WindowEntry win[WINSIZE]; |
| 110 | unsigned char data[WINSIZE]; |
| 111 | int winpos; |
| 112 | struct HashEntry hashtab[HASHMAX]; |
| 113 | unsigned char pending[HASHCHARS]; |
| 114 | int npending; |
| 115 | }; |
| 116 | |
| 117 | static int lz77_hash(unsigned char *data) { |
| 118 | return (257*data[0] + 263*data[1] + 269*data[2]) % HASHMAX; |
| 119 | } |
| 120 | |
| 121 | static int lz77_init(struct LZ77Context *ctx) { |
| 122 | struct LZ77InternalContext *st; |
| 123 | int i; |
| 124 | |
| 125 | st = (struct LZ77InternalContext *)smalloc(sizeof(*st)); |
| 126 | if (!st) |
| 127 | return 0; |
| 128 | |
| 129 | ctx->ictx = st; |
| 130 | |
| 131 | for (i = 0; i < WINSIZE; i++) |
| 132 | st->win[i].next = st->win[i].prev = st->win[i].hashval = INVALID; |
| 133 | for (i = 0; i < HASHMAX; i++) |
| 134 | st->hashtab[i].first = INVALID; |
| 135 | st->winpos = 0; |
| 136 | |
| 137 | st->npending = 0; |
| 138 | |
| 139 | return 1; |
| 140 | } |
| 141 | |
| 142 | static void lz77_advance(struct LZ77InternalContext *st, |
| 143 | unsigned char c, int hash) { |
| 144 | int off; |
| 145 | |
| 146 | /* |
| 147 | * Remove the hash entry at winpos from the tail of its chain, |
| 148 | * or empty the chain if it's the only thing on the chain. |
| 149 | */ |
| 150 | if (st->win[st->winpos].prev != INVALID) { |
| 151 | st->win[st->win[st->winpos].prev].next = INVALID; |
| 152 | } else if (st->win[st->winpos].hashval != INVALID) { |
| 153 | st->hashtab[st->win[st->winpos].hashval].first = INVALID; |
| 154 | } |
| 155 | |
| 156 | /* |
| 157 | * Create a new entry at winpos and add it to the head of its |
| 158 | * hash chain. |
| 159 | */ |
| 160 | st->win[st->winpos].hashval = hash; |
| 161 | st->win[st->winpos].prev = INVALID; |
| 162 | off = st->win[st->winpos].next = st->hashtab[hash].first; |
| 163 | st->hashtab[hash].first = st->winpos; |
| 164 | if (off != INVALID) |
| 165 | st->win[off].prev = st->winpos; |
| 166 | st->data[st->winpos] = c; |
| 167 | |
| 168 | /* |
| 169 | * Advance the window pointer. |
| 170 | */ |
| 171 | st->winpos = (st->winpos + 1) & (WINSIZE-1); |
| 172 | } |
| 173 | |
| 174 | #define CHARAT(k) ( (k)<0 ? st->data[(st->winpos+k)&(WINSIZE-1)] : data[k] ) |
| 175 | |
| 176 | static void lz77_compress(struct LZ77Context *ctx, |
| 177 | unsigned char *data, int len) { |
| 178 | struct LZ77InternalContext *st = ctx->ictx; |
| 179 | int i, hash, distance, off, nmatch, matchlen, advance; |
| 180 | struct Match defermatch, matches[MAXMATCH]; |
| 181 | int deferchr; |
| 182 | |
| 183 | /* |
| 184 | * Add any pending characters from last time to the window. (We |
| 185 | * might not be able to.) |
| 186 | */ |
| 187 | for (i = 0; i < st->npending; i++) { |
| 188 | unsigned char foo[HASHCHARS]; |
| 189 | int j; |
| 190 | if (len + st->npending - i < HASHCHARS) { |
| 191 | /* Update the pending array. */ |
| 192 | for (j = i; j < st->npending; j++) |
| 193 | st->pending[j-i] = st->pending[j]; |
| 194 | break; |
| 195 | } |
| 196 | for (j = 0; j < HASHCHARS; j++) |
| 197 | foo[j] = (i + j < st->npending ? st->pending[i+j] : |
| 198 | data[i + j - st->npending]); |
| 199 | lz77_advance(st, foo[0], lz77_hash(foo)); |
| 200 | } |
| 201 | st->npending -= i; |
| 202 | |
| 203 | defermatch.len = 0; |
| 204 | while (len > 0) { |
| 205 | |
| 206 | if (len >= HASHCHARS) { |
| 207 | /* |
| 208 | * Hash the next few characters. |
| 209 | */ |
| 210 | hash = lz77_hash(data); |
| 211 | |
| 212 | /* |
| 213 | * Look the hash up in the corresponding hash chain and see |
| 214 | * what we can find. |
| 215 | */ |
| 216 | nmatch = 0; |
| 217 | for (off = st->hashtab[hash].first; |
| 218 | off != INVALID; off = st->win[off].next) { |
| 219 | /* distance = 1 if off == st->winpos-1 */ |
| 220 | /* distance = WINSIZE if off == st->winpos */ |
| 221 | distance = WINSIZE - (off + WINSIZE - st->winpos) % WINSIZE; |
| 222 | for (i = 0; i < HASHCHARS; i++) |
| 223 | if (CHARAT(i) != CHARAT(i-distance)) |
| 224 | break; |
| 225 | if (i == HASHCHARS) { |
| 226 | matches[nmatch].distance = distance; |
| 227 | matches[nmatch].len = 3; |
| 228 | if (++nmatch >= MAXMATCH) |
| 229 | break; |
| 230 | } |
| 231 | } |
| 232 | } else { |
| 233 | nmatch = 0; |
| 234 | hash = INVALID; |
| 235 | } |
| 236 | |
| 237 | if (nmatch > 0) { |
| 238 | /* |
| 239 | * We've now filled up matches[] with nmatch potential |
| 240 | * matches. Follow them down to find the longest. (We |
| 241 | * assume here that it's always worth favouring a |
| 242 | * longer match over a shorter one.) |
| 243 | */ |
| 244 | matchlen = HASHCHARS; |
| 245 | while (matchlen < len) { |
| 246 | int j; |
| 247 | for (i = j = 0; i < nmatch; i++) { |
| 248 | if (CHARAT(matchlen) == |
| 249 | CHARAT(matchlen - matches[i].distance)) { |
| 250 | matches[j++] = matches[i]; |
| 251 | } |
| 252 | } |
| 253 | if (j == 0) |
| 254 | break; |
| 255 | matchlen++; |
| 256 | nmatch = j; |
| 257 | } |
| 258 | |
| 259 | /* |
| 260 | * We've now got all the longest matches. We favour the |
| 261 | * shorter distances, which means we go with matches[0]. |
| 262 | * So see if we want to defer it or throw it away. |
| 263 | */ |
| 264 | matches[0].len = matchlen; |
| 265 | if (defermatch.len > 0) { |
| 266 | if (matches[0].len > defermatch.len + 1) { |
| 267 | /* We have a better match. Emit the deferred char, |
| 268 | * and defer this match. */ |
| 269 | ctx->literal(ctx, (unsigned char)deferchr); |
| 270 | defermatch = matches[0]; |
| 271 | deferchr = data[0]; |
| 272 | advance = 1; |
| 273 | } else { |
| 274 | /* We don't have a better match. Do the deferred one. */ |
| 275 | ctx->match(ctx, defermatch.distance, defermatch.len); |
| 276 | advance = defermatch.len - 1; |
| 277 | defermatch.len = 0; |
| 278 | } |
| 279 | } else { |
| 280 | /* There was no deferred match. Defer this one. */ |
| 281 | defermatch = matches[0]; |
| 282 | deferchr = data[0]; |
| 283 | advance = 1; |
| 284 | } |
| 285 | } else { |
| 286 | /* |
| 287 | * We found no matches. Emit the deferred match, if |
| 288 | * any; otherwise emit a literal. |
| 289 | */ |
| 290 | if (defermatch.len > 0) { |
| 291 | ctx->match(ctx, defermatch.distance, defermatch.len); |
| 292 | advance = defermatch.len - 1; |
| 293 | defermatch.len = 0; |
| 294 | } else { |
| 295 | ctx->literal(ctx, data[0]); |
| 296 | advance = 1; |
| 297 | } |
| 298 | } |
| 299 | |
| 300 | /* |
| 301 | * Now advance the position by `advance' characters, |
| 302 | * keeping the window and hash chains consistent. |
| 303 | */ |
| 304 | while (advance > 0) { |
| 305 | if (len >= HASHCHARS) { |
| 306 | lz77_advance(st, *data, lz77_hash(data)); |
| 307 | } else { |
| 308 | st->pending[st->npending++] = *data; |
| 309 | } |
| 310 | data++; |
| 311 | len--; |
| 312 | advance--; |
| 313 | } |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | /* ---------------------------------------------------------------------- |
| 318 | * Zlib compression. We always use the static Huffman tree option. |
| 319 | * Mostly this is because it's hard to scan a block in advance to |
| 320 | * work out better trees; dynamic trees are great when you're |
| 321 | * compressing a large file under no significant time constraint, |
| 322 | * but when you're compressing little bits in real time, things get |
| 323 | * hairier. |
| 324 | * |
| 325 | * I suppose it's possible that I could compute Huffman trees based |
| 326 | * on the frequencies in the _previous_ block, as a sort of |
| 327 | * heuristic, but I'm not confident that the gain would balance out |
| 328 | * having to transmit the trees. |
| 329 | */ |
| 330 | |
| 331 | static struct LZ77Context ectx; |
| 332 | |
| 333 | struct Outbuf { |
| 334 | unsigned char *outbuf; |
| 335 | int outlen, outsize; |
| 336 | unsigned long outbits; |
| 337 | int noutbits; |
| 338 | int firstblock; |
| 339 | }; |
| 340 | |
| 341 | static void outbits(struct Outbuf *out, unsigned long bits, int nbits) { |
| 342 | assert(out->noutbits + nbits <= 32); |
| 343 | out->outbits |= bits << out->noutbits; |
| 344 | out->noutbits += nbits; |
| 345 | while (out->noutbits >= 8) { |
| 346 | if (out->outlen >= out->outsize) { |
| 347 | out->outsize = out->outlen + 64; |
| 348 | out->outbuf = srealloc(out->outbuf, out->outsize); |
| 349 | } |
| 350 | out->outbuf[out->outlen++] = (unsigned char)(out->outbits & 0xFF); |
| 351 | out->outbits >>= 8; |
| 352 | out->noutbits -= 8; |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | static const unsigned char mirrorbytes[256] = { |
| 357 | 0x00, 0x80, 0x40, 0xc0, 0x20, 0xa0, 0x60, 0xe0, |
| 358 | 0x10, 0x90, 0x50, 0xd0, 0x30, 0xb0, 0x70, 0xf0, |
| 359 | 0x08, 0x88, 0x48, 0xc8, 0x28, 0xa8, 0x68, 0xe8, |
| 360 | 0x18, 0x98, 0x58, 0xd8, 0x38, 0xb8, 0x78, 0xf8, |
| 361 | 0x04, 0x84, 0x44, 0xc4, 0x24, 0xa4, 0x64, 0xe4, |
| 362 | 0x14, 0x94, 0x54, 0xd4, 0x34, 0xb4, 0x74, 0xf4, |
| 363 | 0x0c, 0x8c, 0x4c, 0xcc, 0x2c, 0xac, 0x6c, 0xec, |
| 364 | 0x1c, 0x9c, 0x5c, 0xdc, 0x3c, 0xbc, 0x7c, 0xfc, |
| 365 | 0x02, 0x82, 0x42, 0xc2, 0x22, 0xa2, 0x62, 0xe2, |
| 366 | 0x12, 0x92, 0x52, 0xd2, 0x32, 0xb2, 0x72, 0xf2, |
| 367 | 0x0a, 0x8a, 0x4a, 0xca, 0x2a, 0xaa, 0x6a, 0xea, |
| 368 | 0x1a, 0x9a, 0x5a, 0xda, 0x3a, 0xba, 0x7a, 0xfa, |
| 369 | 0x06, 0x86, 0x46, 0xc6, 0x26, 0xa6, 0x66, 0xe6, |
| 370 | 0x16, 0x96, 0x56, 0xd6, 0x36, 0xb6, 0x76, 0xf6, |
| 371 | 0x0e, 0x8e, 0x4e, 0xce, 0x2e, 0xae, 0x6e, 0xee, |
| 372 | 0x1e, 0x9e, 0x5e, 0xde, 0x3e, 0xbe, 0x7e, 0xfe, |
| 373 | 0x01, 0x81, 0x41, 0xc1, 0x21, 0xa1, 0x61, 0xe1, |
| 374 | 0x11, 0x91, 0x51, 0xd1, 0x31, 0xb1, 0x71, 0xf1, |
| 375 | 0x09, 0x89, 0x49, 0xc9, 0x29, 0xa9, 0x69, 0xe9, |
| 376 | 0x19, 0x99, 0x59, 0xd9, 0x39, 0xb9, 0x79, 0xf9, |
| 377 | 0x05, 0x85, 0x45, 0xc5, 0x25, 0xa5, 0x65, 0xe5, |
| 378 | 0x15, 0x95, 0x55, 0xd5, 0x35, 0xb5, 0x75, 0xf5, |
| 379 | 0x0d, 0x8d, 0x4d, 0xcd, 0x2d, 0xad, 0x6d, 0xed, |
| 380 | 0x1d, 0x9d, 0x5d, 0xdd, 0x3d, 0xbd, 0x7d, 0xfd, |
| 381 | 0x03, 0x83, 0x43, 0xc3, 0x23, 0xa3, 0x63, 0xe3, |
| 382 | 0x13, 0x93, 0x53, 0xd3, 0x33, 0xb3, 0x73, 0xf3, |
| 383 | 0x0b, 0x8b, 0x4b, 0xcb, 0x2b, 0xab, 0x6b, 0xeb, |
| 384 | 0x1b, 0x9b, 0x5b, 0xdb, 0x3b, 0xbb, 0x7b, 0xfb, |
| 385 | 0x07, 0x87, 0x47, 0xc7, 0x27, 0xa7, 0x67, 0xe7, |
| 386 | 0x17, 0x97, 0x57, 0xd7, 0x37, 0xb7, 0x77, 0xf7, |
| 387 | 0x0f, 0x8f, 0x4f, 0xcf, 0x2f, 0xaf, 0x6f, 0xef, |
| 388 | 0x1f, 0x9f, 0x5f, 0xdf, 0x3f, 0xbf, 0x7f, 0xff, |
| 389 | }; |
| 390 | |
| 391 | typedef struct { |
| 392 | int code, extrabits, min, max; |
| 393 | } coderecord; |
| 394 | |
| 395 | static const coderecord lencodes[] = { |
| 396 | {257, 0, 3,3}, |
| 397 | {258, 0, 4,4}, |
| 398 | {259, 0, 5,5}, |
| 399 | {260, 0, 6,6}, |
| 400 | {261, 0, 7,7}, |
| 401 | {262, 0, 8,8}, |
| 402 | {263, 0, 9,9}, |
| 403 | {264, 0, 10,10}, |
| 404 | {265, 1, 11,12}, |
| 405 | {266, 1, 13,14}, |
| 406 | {267, 1, 15,16}, |
| 407 | {268, 1, 17,18}, |
| 408 | {269, 2, 19,22}, |
| 409 | {270, 2, 23,26}, |
| 410 | {271, 2, 27,30}, |
| 411 | {272, 2, 31,34}, |
| 412 | {273, 3, 35,42}, |
| 413 | {274, 3, 43,50}, |
| 414 | {275, 3, 51,58}, |
| 415 | {276, 3, 59,66}, |
| 416 | {277, 4, 67,82}, |
| 417 | {278, 4, 83,98}, |
| 418 | {279, 4, 99,114}, |
| 419 | {280, 4, 115,130}, |
| 420 | {281, 5, 131,162}, |
| 421 | {282, 5, 163,194}, |
| 422 | {283, 5, 195,226}, |
| 423 | {284, 5, 227,257}, |
| 424 | {285, 0, 258,258}, |
| 425 | }; |
| 426 | |
| 427 | static const coderecord distcodes[] = { |
| 428 | {0, 0, 1,1}, |
| 429 | {1, 0, 2,2}, |
| 430 | {2, 0, 3,3}, |
| 431 | {3, 0, 4,4}, |
| 432 | {4, 1, 5,6}, |
| 433 | {5, 1, 7,8}, |
| 434 | {6, 2, 9,12}, |
| 435 | {7, 2, 13,16}, |
| 436 | {8, 3, 17,24}, |
| 437 | {9, 3, 25,32}, |
| 438 | {10, 4, 33,48}, |
| 439 | {11, 4, 49,64}, |
| 440 | {12, 5, 65,96}, |
| 441 | {13, 5, 97,128}, |
| 442 | {14, 6, 129,192}, |
| 443 | {15, 6, 193,256}, |
| 444 | {16, 7, 257,384}, |
| 445 | {17, 7, 385,512}, |
| 446 | {18, 8, 513,768}, |
| 447 | {19, 8, 769,1024}, |
| 448 | {20, 9, 1025,1536}, |
| 449 | {21, 9, 1537,2048}, |
| 450 | {22, 10, 2049,3072}, |
| 451 | {23, 10, 3073,4096}, |
| 452 | {24, 11, 4097,6144}, |
| 453 | {25, 11, 6145,8192}, |
| 454 | {26, 12, 8193,12288}, |
| 455 | {27, 12, 12289,16384}, |
| 456 | {28, 13, 16385,24576}, |
| 457 | {29, 13, 24577,32768}, |
| 458 | }; |
| 459 | |
| 460 | static void zlib_literal(struct LZ77Context *ectx, unsigned char c) { |
| 461 | struct Outbuf *out = (struct Outbuf *)ectx->userdata; |
| 462 | |
| 463 | if (c <= 143) { |
| 464 | /* 0 through 143 are 8 bits long starting at 00110000. */ |
| 465 | outbits(out, mirrorbytes[0x30 + c], 8); |
| 466 | } else { |
| 467 | /* 144 through 255 are 9 bits long starting at 110010000. */ |
| 468 | outbits(out, 1 + 2*mirrorbytes[0x90 - 144 + c], 9); |
| 469 | } |
| 470 | } |
| 471 | |
| 472 | static void zlib_match(struct LZ77Context *ectx, int distance, int len) { |
| 473 | const coderecord *d, *l; |
| 474 | int i, j, k; |
| 475 | struct Outbuf *out = (struct Outbuf *)ectx->userdata; |
| 476 | while (len > 0) { |
| 477 | int thislen; |
| 478 | |
| 479 | /* |
| 480 | * We can transmit matches of lengths 3 through 258 |
| 481 | * inclusive. So if len exceeds 258, we must transmit in |
| 482 | * several steps, with 258 or less in each step. |
| 483 | * |
| 484 | * Specifically: if len >= 261, we can transmit 258 and be |
| 485 | * sure of having at least 3 left for the next step. And if |
| 486 | * len <= 258, we can just transmit len. But if len == 259 |
| 487 | * or 260, we must transmit len-3. |
| 488 | */ |
| 489 | thislen = (len > 260 ? 258 : len <= 258 ? len : len-3); |
| 490 | len -= thislen; |
| 491 | |
| 492 | /* |
| 493 | * Binary-search to find which length code we're |
| 494 | * transmitting. |
| 495 | */ |
| 496 | i = -1; j = sizeof(lencodes)/sizeof(*lencodes); |
| 497 | while (j - i >= 2) { |
| 498 | k = (j+i)/2; |
| 499 | if (thislen < lencodes[k].min) |
| 500 | j = k; |
| 501 | else if (thislen > lencodes[k].max) |
| 502 | i = k; |
| 503 | else { |
| 504 | l = &lencodes[k]; |
| 505 | break; /* found it! */ |
| 506 | } |
| 507 | } |
| 508 | |
| 509 | /* |
| 510 | * Transmit the length code. 256-279 are seven bits |
| 511 | * starting at 0000000; 280-287 are eight bits starting at |
| 512 | * 11000000. |
| 513 | */ |
| 514 | if (l->code <= 279) { |
| 515 | outbits(out, mirrorbytes[(l->code-256)*2], 7); |
| 516 | } else { |
| 517 | outbits(out, mirrorbytes[0xc0 - 280 + l->code], 8); |
| 518 | } |
| 519 | |
| 520 | /* |
| 521 | * Transmit the extra bits. |
| 522 | */ |
| 523 | if (l->extrabits) |
| 524 | outbits(out, thislen - l->min, l->extrabits); |
| 525 | |
| 526 | /* |
| 527 | * Binary-search to find which distance code we're |
| 528 | * transmitting. |
| 529 | */ |
| 530 | i = -1; j = sizeof(distcodes)/sizeof(*distcodes); |
| 531 | while (j - i >= 2) { |
| 532 | k = (j+i)/2; |
| 533 | if (distance < distcodes[k].min) |
| 534 | j = k; |
| 535 | else if (distance > distcodes[k].max) |
| 536 | i = k; |
| 537 | else { |
| 538 | d = &distcodes[k]; |
| 539 | break; /* found it! */ |
| 540 | } |
| 541 | } |
| 542 | |
| 543 | /* |
| 544 | * Transmit the distance code. Five bits starting at 00000. |
| 545 | */ |
| 546 | outbits(out, mirrorbytes[d->code*8], 5); |
| 547 | |
| 548 | /* |
| 549 | * Transmit the extra bits. |
| 550 | */ |
| 551 | if (d->extrabits) |
| 552 | outbits(out, distance - d->min, d->extrabits); |
| 553 | } |
| 554 | } |
| 555 | |
| 556 | void zlib_compress_init(void) { |
| 557 | struct Outbuf *out; |
| 558 | |
| 559 | lz77_init(&ectx); |
| 560 | ectx.literal = zlib_literal; |
| 561 | ectx.match = zlib_match; |
| 562 | |
| 563 | out = smalloc(sizeof(struct Outbuf)); |
| 564 | out->outbits = out->noutbits = 0; |
| 565 | out->firstblock = 1; |
| 566 | ectx.userdata = out; |
| 567 | |
| 568 | logevent("Initialised zlib (RFC1950) compression"); |
| 569 | } |
| 570 | |
| 571 | int zlib_compress_block(unsigned char *block, int len, |
| 572 | unsigned char **outblock, int *outlen) { |
| 573 | struct Outbuf *out = (struct Outbuf *)ectx.userdata; |
| 574 | |
| 575 | out->outbuf = NULL; |
| 576 | out->outlen = out->outsize = 0; |
| 577 | |
| 578 | /* |
| 579 | * If this is the first block, output the Zlib (RFC1950) header |
| 580 | * bytes 78 9C. (Deflate compression, 32K window size, default |
| 581 | * algorithm.) |
| 582 | */ |
| 583 | if (out->firstblock) { |
| 584 | outbits(out, 0x9C78, 16); |
| 585 | out->firstblock = 0; |
| 586 | /* |
| 587 | * Start a Deflate (RFC1951) fixed-trees block. We transmit |
| 588 | * a zero bit (BFINAL=0), followed by a zero bit and a one |
| 589 | * bit (BTYPE=01). Of course these are in the wrong order |
| 590 | * (01 0). |
| 591 | */ |
| 592 | outbits(out, 2, 3); |
| 593 | } |
| 594 | |
| 595 | /* |
| 596 | * Do the compression. |
| 597 | */ |
| 598 | lz77_compress(&ectx, block, len); |
| 599 | /* |
| 600 | * End the block (by transmitting code 256, which is 0000000 in |
| 601 | * fixed-tree mode), and transmit some empty blocks to ensure |
| 602 | * we have emitted the byte containing the last piece of |
| 603 | * genuine data. There are three ways we can do this: |
| 604 | * |
| 605 | * - Minimal flush. Output end-of-block and then open a new |
| 606 | * static block. This takes 9 bits, which is guaranteed to |
| 607 | * flush out the last genuine code in the closed block; but |
| 608 | * allegedly zlib can't handle it. |
| 609 | * |
| 610 | * - Zlib partial flush. Output EOB, open and close an empty |
| 611 | * static block, and _then_ open the new block. This is the |
| 612 | * best zlib can handle. |
| 613 | * |
| 614 | * - Zlib sync flush. Output EOB, then an empty _uncompressed_ |
| 615 | * block (000, then sync to byte boundary, then send bytes |
| 616 | * 00 00 FF FF). Then open the new block. |
| 617 | * |
| 618 | * For the moment, we will use Zlib partial flush. |
| 619 | */ |
| 620 | outbits(out, 0, 7); /* close block */ |
| 621 | outbits(out, 2, 3+7); /* empty static block */ |
| 622 | outbits(out, 2, 3); /* open new block */ |
| 623 | |
| 624 | *outblock = out->outbuf; |
| 625 | *outlen = out->outlen; |
| 626 | |
| 627 | return 1; |
| 628 | } |
| 629 | |
| 630 | /* ---------------------------------------------------------------------- |
| 631 | * Zlib decompression. Of course, even though our compressor always |
| 632 | * uses static trees, our _decompressor_ has to be capable of |
| 633 | * handling dynamic trees if it sees them. |
| 634 | */ |
| 635 | |
| 636 | /* |
| 637 | * The way we work the Huffman decode is to have a table lookup on |
| 638 | * the first N bits of the input stream (in the order they arrive, |
| 639 | * of course, i.e. the first bit of the Huffman code is in bit 0). |
| 640 | * Each table entry lists the number of bits to consume, plus |
| 641 | * either an output code or a pointer to a secondary table. |
| 642 | */ |
| 643 | struct zlib_table; |
| 644 | struct zlib_tableentry; |
| 645 | |
| 646 | struct zlib_tableentry { |
| 647 | unsigned char nbits; |
| 648 | int code; |
| 649 | struct zlib_table *nexttable; |
| 650 | }; |
| 651 | |
| 652 | struct zlib_table { |
| 653 | int mask; /* mask applied to input bit stream */ |
| 654 | struct zlib_tableentry *table; |
| 655 | }; |
| 656 | |
| 657 | #define MAXCODELEN 16 |
| 658 | #define MAXSYMS 288 |
| 659 | |
| 660 | /* |
| 661 | * Build a single-level decode table for elements |
| 662 | * [minlength,maxlength) of the provided code/length tables, and |
| 663 | * recurse to build subtables. |
| 664 | */ |
| 665 | static struct zlib_table *zlib_mkonetab(int *codes, unsigned char *lengths, |
| 666 | int nsyms, |
| 667 | int pfx, int pfxbits, int bits) { |
| 668 | struct zlib_table *tab = smalloc(sizeof(struct zlib_table)); |
| 669 | int pfxmask = (1 << pfxbits) - 1; |
| 670 | int nbits, i, j, code; |
| 671 | |
| 672 | tab->table = smalloc((1 << bits) * sizeof(struct zlib_tableentry)); |
| 673 | tab->mask = (1 << bits) - 1; |
| 674 | |
| 675 | for (code = 0; code <= tab->mask; code++) { |
| 676 | tab->table[code].code = -1; |
| 677 | tab->table[code].nbits = 0; |
| 678 | tab->table[code].nexttable = NULL; |
| 679 | } |
| 680 | |
| 681 | for (i = 0; i < nsyms; i++) { |
| 682 | if (lengths[i] <= pfxbits || (codes[i] & pfxmask) != pfx) |
| 683 | continue; |
| 684 | code = (codes[i] >> pfxbits) & tab->mask; |
| 685 | for (j = code; j <= tab->mask; j += 1 << (lengths[i]-pfxbits)) { |
| 686 | tab->table[j].code = i; |
| 687 | nbits = lengths[i] - pfxbits; |
| 688 | if (tab->table[j].nbits < nbits) |
| 689 | tab->table[j].nbits = nbits; |
| 690 | } |
| 691 | } |
| 692 | for (code = 0; code <= tab->mask; code++) { |
| 693 | if (tab->table[code].nbits <= bits) |
| 694 | continue; |
| 695 | /* Generate a subtable. */ |
| 696 | tab->table[code].code = -1; |
| 697 | nbits = tab->table[code].nbits - bits; |
| 698 | if (nbits > 7) |
| 699 | nbits = 7; |
| 700 | tab->table[code].nbits = bits; |
| 701 | tab->table[code].nexttable = zlib_mkonetab(codes, lengths, nsyms, |
| 702 | pfx | (code << pfxbits), |
| 703 | pfxbits + bits, nbits); |
| 704 | } |
| 705 | |
| 706 | return tab; |
| 707 | } |
| 708 | |
| 709 | /* |
| 710 | * Build a decode table, given a set of Huffman tree lengths. |
| 711 | */ |
| 712 | static struct zlib_table *zlib_mktable(unsigned char *lengths, int nlengths) { |
| 713 | int count[MAXCODELEN], startcode[MAXCODELEN], codes[MAXSYMS]; |
| 714 | int code, maxlen; |
| 715 | int i, j; |
| 716 | |
| 717 | /* Count the codes of each length. */ |
| 718 | maxlen = 0; |
| 719 | for (i = 1; i < MAXCODELEN; i++) count[i] = 0; |
| 720 | for (i = 0; i < nlengths; i++) { |
| 721 | count[lengths[i]]++; |
| 722 | if (maxlen < lengths[i]) |
| 723 | maxlen = lengths[i]; |
| 724 | } |
| 725 | /* Determine the starting code for each length block. */ |
| 726 | code = 0; |
| 727 | for (i = 1; i < MAXCODELEN; i++) { |
| 728 | startcode[i] = code; |
| 729 | code += count[i]; |
| 730 | code <<= 1; |
| 731 | } |
| 732 | /* Determine the code for each symbol. Mirrored, of course. */ |
| 733 | for (i = 0; i < nlengths; i++) { |
| 734 | code = startcode[lengths[i]]++; |
| 735 | codes[i] = 0; |
| 736 | for (j = 0; j < lengths[i]; j++) { |
| 737 | codes[i] = (codes[i] << 1) | (code & 1); |
| 738 | code >>= 1; |
| 739 | } |
| 740 | } |
| 741 | |
| 742 | /* |
| 743 | * Now we have the complete list of Huffman codes. Build a |
| 744 | * table. |
| 745 | */ |
| 746 | return zlib_mkonetab(codes, lengths, nlengths, 0, 0, |
| 747 | maxlen < 9 ? maxlen : 9); |
| 748 | } |
| 749 | |
| 750 | static struct zlib_decompress_ctx { |
| 751 | struct zlib_table *staticlentable, *staticdisttable; |
| 752 | struct zlib_table *currlentable, *currdisttable, *lenlentable; |
| 753 | enum { |
| 754 | START, OUTSIDEBLK, |
| 755 | TREES_HDR, TREES_LENLEN, TREES_LEN, TREES_LENREP, |
| 756 | INBLK, GOTLENSYM, GOTLEN, GOTDISTSYM, |
| 757 | UNCOMP_LEN, UNCOMP_NLEN, UNCOMP_DATA |
| 758 | } state; |
| 759 | int sym, hlit, hdist, hclen, lenptr, lenextrabits, lenaddon, len, lenrep; |
| 760 | int uncomplen; |
| 761 | unsigned char lenlen[19]; |
| 762 | unsigned char lengths[286+32]; |
| 763 | unsigned long bits; |
| 764 | int nbits; |
| 765 | unsigned char window[WINSIZE]; |
| 766 | int winpos; |
| 767 | unsigned char *outblk; |
| 768 | int outlen, outsize; |
| 769 | } dctx; |
| 770 | |
| 771 | void zlib_decompress_init(void) { |
| 772 | unsigned char lengths[288]; |
| 773 | memset(lengths, 8, 144); |
| 774 | memset(lengths+144, 9, 256-144); |
| 775 | memset(lengths+256, 7, 280-256); |
| 776 | memset(lengths+280, 8, 288-280); |
| 777 | dctx.staticlentable = zlib_mktable(lengths, 288); |
| 778 | memset(lengths, 5, 32); |
| 779 | dctx.staticdisttable = zlib_mktable(lengths, 32); |
| 780 | dctx.state = START; /* even before header */ |
| 781 | dctx.currlentable = dctx.currdisttable = dctx.lenlentable = NULL; |
| 782 | dctx.bits = 0; |
| 783 | dctx.nbits = 0; |
| 784 | logevent("Initialised zlib (RFC1950) decompression"); |
| 785 | } |
| 786 | |
| 787 | int zlib_huflookup(unsigned long *bitsp, int *nbitsp, struct zlib_table *tab) { |
| 788 | unsigned long bits = *bitsp; |
| 789 | int nbits = *nbitsp; |
| 790 | while (1) { |
| 791 | struct zlib_tableentry *ent; |
| 792 | ent = &tab->table[bits & tab->mask]; |
| 793 | if (ent->nbits > nbits) |
| 794 | return -1; /* not enough data */ |
| 795 | bits >>= ent->nbits; |
| 796 | nbits -= ent->nbits; |
| 797 | if (ent->code == -1) |
| 798 | tab = ent->nexttable; |
| 799 | else { |
| 800 | *bitsp = bits; |
| 801 | *nbitsp = nbits; |
| 802 | return ent->code; |
| 803 | } |
| 804 | } |
| 805 | } |
| 806 | |
| 807 | static void zlib_emit_char(int c) { |
| 808 | dctx.window[dctx.winpos] = c; |
| 809 | dctx.winpos = (dctx.winpos + 1) & (WINSIZE-1); |
| 810 | if (dctx.outlen >= dctx.outsize) { |
| 811 | dctx.outsize = dctx.outlen + 512; |
| 812 | dctx.outblk = srealloc(dctx.outblk, dctx.outsize); |
| 813 | } |
| 814 | dctx.outblk[dctx.outlen++] = c; |
| 815 | } |
| 816 | |
| 817 | #define EATBITS(n) ( dctx.nbits -= (n), dctx.bits >>= (n) ) |
| 818 | |
| 819 | int zlib_decompress_block(unsigned char *block, int len, |
| 820 | unsigned char **outblock, int *outlen) { |
| 821 | const coderecord *rec; |
| 822 | int code, blktype, rep, dist, nlen; |
| 823 | static const unsigned char lenlenmap[] = { |
| 824 | 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 |
| 825 | }; |
| 826 | |
| 827 | dctx.outblk = NULL; |
| 828 | dctx.outsize = dctx.outlen = 0; |
| 829 | |
| 830 | while (len > 0 || dctx.nbits > 0) { |
| 831 | while (dctx.nbits < 24 && len > 0) { |
| 832 | dctx.bits |= (*block++) << dctx.nbits; |
| 833 | dctx.nbits += 8; |
| 834 | len--; |
| 835 | } |
| 836 | switch (dctx.state) { |
| 837 | case START: |
| 838 | /* Expect 16-bit zlib header, which we'll dishonourably ignore. */ |
| 839 | if (dctx.nbits < 16) |
| 840 | goto finished; /* done all we can */ |
| 841 | EATBITS(16); |
| 842 | dctx.state = OUTSIDEBLK; |
| 843 | break; |
| 844 | case OUTSIDEBLK: |
| 845 | /* Expect 3-bit block header. */ |
| 846 | if (dctx.nbits < 3) |
| 847 | goto finished; /* done all we can */ |
| 848 | EATBITS(1); |
| 849 | blktype = dctx.bits & 3; |
| 850 | EATBITS(2); |
| 851 | if (blktype == 0) { |
| 852 | int to_eat = dctx.nbits & 7; |
| 853 | dctx.state = UNCOMP_LEN; |
| 854 | EATBITS(to_eat); /* align to byte boundary */ |
| 855 | } else if (blktype == 1) { |
| 856 | dctx.currlentable = dctx.staticlentable; |
| 857 | dctx.currdisttable = dctx.staticdisttable; |
| 858 | dctx.state = INBLK; |
| 859 | } else if (blktype == 2) { |
| 860 | dctx.state = TREES_HDR; |
| 861 | } |
| 862 | break; |
| 863 | case TREES_HDR: |
| 864 | /* |
| 865 | * Dynamic block header. Five bits of HLIT, five of |
| 866 | * HDIST, four of HCLEN. |
| 867 | */ |
| 868 | if (dctx.nbits < 5+5+4) |
| 869 | goto finished; /* done all we can */ |
| 870 | dctx.hlit = 257 + (dctx.bits & 31); EATBITS(5); |
| 871 | dctx.hdist = 1 + (dctx.bits & 31); EATBITS(5); |
| 872 | dctx.hclen = 4 + (dctx.bits & 15); EATBITS(4); |
| 873 | dctx.lenptr = 0; |
| 874 | dctx.state = TREES_LENLEN; |
| 875 | memset(dctx.lenlen, 0, sizeof(dctx.lenlen)); |
| 876 | break; |
| 877 | case TREES_LENLEN: |
| 878 | if (dctx.nbits < 3) |
| 879 | goto finished; |
| 880 | while (dctx.lenptr < dctx.hclen && dctx.nbits >= 3) { |
| 881 | dctx.lenlen[lenlenmap[dctx.lenptr++]] = |
| 882 | (unsigned char)(dctx.bits & 7); |
| 883 | EATBITS(3); |
| 884 | } |
| 885 | if (dctx.lenptr == dctx.hclen) { |
| 886 | dctx.lenlentable = zlib_mktable(dctx.lenlen, 19); |
| 887 | dctx.state = TREES_LEN; |
| 888 | dctx.lenptr = 0; |
| 889 | } |
| 890 | break; |
| 891 | case TREES_LEN: |
| 892 | if (dctx.lenptr >= dctx.hlit+dctx.hdist) { |
| 893 | dctx.currlentable = zlib_mktable(dctx.lengths, dctx.hlit); |
| 894 | dctx.currdisttable = zlib_mktable(dctx.lengths + dctx.hlit, |
| 895 | dctx.hdist); |
| 896 | /* FIXME: zlib_freetable(dctx.lenlentable); */ |
| 897 | dctx.state = INBLK; |
| 898 | break; |
| 899 | } |
| 900 | code = zlib_huflookup(&dctx.bits, &dctx.nbits, dctx.lenlentable); |
| 901 | if (code == -1) |
| 902 | goto finished; |
| 903 | if (code < 16) |
| 904 | dctx.lengths[dctx.lenptr++] = code; |
| 905 | else { |
| 906 | dctx.lenextrabits = (code == 16 ? 2 : code == 17 ? 3 : 7); |
| 907 | dctx.lenaddon = (code == 18 ? 11 : 3); |
| 908 | dctx.lenrep = (code == 16 && dctx.lenptr > 0 ? |
| 909 | dctx.lengths[dctx.lenptr-1] : 0); |
| 910 | dctx.state = TREES_LENREP; |
| 911 | } |
| 912 | break; |
| 913 | case TREES_LENREP: |
| 914 | if (dctx.nbits < dctx.lenextrabits) |
| 915 | goto finished; |
| 916 | rep = dctx.lenaddon + (dctx.bits & ((1<<dctx.lenextrabits)-1)); |
| 917 | EATBITS(dctx.lenextrabits); |
| 918 | while (rep > 0 && dctx.lenptr < dctx.hlit+dctx.hdist) { |
| 919 | dctx.lengths[dctx.lenptr] = dctx.lenrep; |
| 920 | dctx.lenptr++; |
| 921 | rep--; |
| 922 | } |
| 923 | dctx.state = TREES_LEN; |
| 924 | break; |
| 925 | case INBLK: |
| 926 | code = zlib_huflookup(&dctx.bits, &dctx.nbits, dctx.currlentable); |
| 927 | if (code == -1) |
| 928 | goto finished; |
| 929 | if (code < 256) |
| 930 | zlib_emit_char(code); |
| 931 | else if (code == 256) { |
| 932 | dctx.state = OUTSIDEBLK; |
| 933 | /* FIXME: zlib_freetable(both) if not static */ |
| 934 | } else if (code < 286) { /* static tree can give >285; ignore */ |
| 935 | dctx.state = GOTLENSYM; |
| 936 | dctx.sym = code; |
| 937 | } |
| 938 | break; |
| 939 | case GOTLENSYM: |
| 940 | rec = &lencodes[dctx.sym - 257]; |
| 941 | if (dctx.nbits < rec->extrabits) |
| 942 | goto finished; |
| 943 | dctx.len = rec->min + (dctx.bits & ((1<<rec->extrabits)-1)); |
| 944 | EATBITS(rec->extrabits); |
| 945 | dctx.state = GOTLEN; |
| 946 | break; |
| 947 | case GOTLEN: |
| 948 | code = zlib_huflookup(&dctx.bits, &dctx.nbits, dctx.currdisttable); |
| 949 | if (code == -1) |
| 950 | goto finished; |
| 951 | dctx.state = GOTDISTSYM; |
| 952 | dctx.sym = code; |
| 953 | break; |
| 954 | case GOTDISTSYM: |
| 955 | rec = &distcodes[dctx.sym]; |
| 956 | if (dctx.nbits < rec->extrabits) |
| 957 | goto finished; |
| 958 | dist = rec->min + (dctx.bits & ((1<<rec->extrabits)-1)); |
| 959 | EATBITS(rec->extrabits); |
| 960 | dctx.state = INBLK; |
| 961 | while (dctx.len--) |
| 962 | zlib_emit_char(dctx.window[(dctx.winpos-dist) & (WINSIZE-1)]); |
| 963 | break; |
| 964 | case UNCOMP_LEN: |
| 965 | /* |
| 966 | * Uncompressed block. We expect to see a 16-bit LEN. |
| 967 | */ |
| 968 | if (dctx.nbits < 16) |
| 969 | goto finished; |
| 970 | dctx.uncomplen = dctx.bits & 0xFFFF; |
| 971 | EATBITS(16); |
| 972 | dctx.state = UNCOMP_NLEN; |
| 973 | break; |
| 974 | case UNCOMP_NLEN: |
| 975 | /* |
| 976 | * Uncompressed block. We expect to see a 16-bit NLEN, |
| 977 | * which should be the one's complement of the previous |
| 978 | * LEN. |
| 979 | */ |
| 980 | if (dctx.nbits < 16) |
| 981 | goto finished; |
| 982 | nlen = dctx.bits & 0xFFFF; |
| 983 | EATBITS(16); |
| 984 | dctx.state = UNCOMP_DATA; |
| 985 | break; |
| 986 | case UNCOMP_DATA: |
| 987 | if (dctx.nbits < 8) |
| 988 | goto finished; |
| 989 | zlib_emit_char(dctx.bits & 0xFF); |
| 990 | EATBITS(8); |
| 991 | if (--dctx.uncomplen == 0) |
| 992 | dctx.state = OUTSIDEBLK; /* end of uncompressed block */ |
| 993 | break; |
| 994 | } |
| 995 | } |
| 996 | |
| 997 | finished: |
| 998 | *outblock = dctx.outblk; |
| 999 | *outlen = dctx.outlen; |
| 1000 | |
| 1001 | return 1; |
| 1002 | } |
| 1003 | |
| 1004 | const struct ssh_compress ssh_zlib = { |
| 1005 | "zlib", |
| 1006 | zlib_compress_init, |
| 1007 | zlib_compress_block, |
| 1008 | zlib_decompress_init, |
| 1009 | zlib_decompress_block |
| 1010 | }; |