Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2013-2015 Willy Tarreau <w@1wt.eu> |
| 3 | * |
| 4 | * Permission is hereby granted, free of charge, to any person obtaining |
| 5 | * a copy of this software and associated documentation files (the |
| 6 | * "Software"), to deal in the Software without restriction, including |
| 7 | * without limitation the rights to use, copy, modify, merge, publish, |
| 8 | * distribute, sublicense, and/or sell copies of the Software, and to |
| 9 | * permit persons to whom the Software is furnished to do so, subject to |
| 10 | * the following conditions: |
| 11 | * |
| 12 | * The above copyright notice and this permission notice shall be |
| 13 | * included in all copies or substantial portions of the Software. |
| 14 | * |
| 15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, |
| 16 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES |
| 17 | * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND |
| 18 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT |
| 19 | * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, |
| 20 | * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
| 21 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
| 22 | * OTHER DEALINGS IN THE SOFTWARE. |
| 23 | */ |
| 24 | |
Willy Tarreau | 388fc25 | 2021-05-14 08:44:52 +0200 | [diff] [blame] | 25 | #include <inttypes.h> |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 26 | #include <stdio.h> |
| 27 | #include <string.h> |
| 28 | #include <import/slz.h> |
| 29 | #include <import/slz-tables.h> |
| 30 | |
| 31 | /* First, RFC1951-specific declarations and extracts from the RFC. |
| 32 | * |
| 33 | * RFC1951 - deflate stream format |
| 34 | |
| 35 | |
| 36 | * Data elements are packed into bytes in order of |
| 37 | increasing bit number within the byte, i.e., starting |
| 38 | with the least-significant bit of the byte. |
| 39 | * Data elements other than Huffman codes are packed |
| 40 | starting with the least-significant bit of the data |
| 41 | element. |
| 42 | * Huffman codes are packed starting with the most- |
| 43 | significant bit of the code. |
| 44 | |
| 45 | 3.2.3. Details of block format |
| 46 | |
| 47 | Each block of compressed data begins with 3 header bits |
| 48 | containing the following data: |
| 49 | |
| 50 | first bit BFINAL |
| 51 | next 2 bits BTYPE |
| 52 | |
| 53 | Note that the header bits do not necessarily begin on a byte |
| 54 | boundary, since a block does not necessarily occupy an integral |
| 55 | number of bytes. |
| 56 | |
| 57 | BFINAL is set if and only if this is the last block of the data |
| 58 | set. |
| 59 | |
| 60 | BTYPE specifies how the data are compressed, as follows: |
| 61 | |
| 62 | 00 - no compression |
| 63 | 01 - compressed with fixed Huffman codes |
| 64 | 10 - compressed with dynamic Huffman codes |
| 65 | 11 - reserved (error) |
| 66 | |
| 67 | 3.2.4. Non-compressed blocks (BTYPE=00) |
| 68 | |
| 69 | Any bits of input up to the next byte boundary are ignored. |
| 70 | The rest of the block consists of the following information: |
| 71 | |
| 72 | 0 1 2 3 4... |
| 73 | +---+---+---+---+================================+ |
| 74 | | LEN | NLEN |... LEN bytes of literal data...| |
| 75 | +---+---+---+---+================================+ |
| 76 | |
| 77 | LEN is the number of data bytes in the block. NLEN is the |
| 78 | one's complement of LEN. |
| 79 | |
| 80 | 3.2.5. Compressed blocks (length and distance codes) |
| 81 | |
| 82 | As noted above, encoded data blocks in the "deflate" format |
| 83 | consist of sequences of symbols drawn from three conceptually |
| 84 | distinct alphabets: either literal bytes, from the alphabet of |
| 85 | byte values (0..255), or <length, backward distance> pairs, |
| 86 | where the length is drawn from (3..258) and the distance is |
| 87 | drawn from (1..32,768). In fact, the literal and length |
| 88 | alphabets are merged into a single alphabet (0..285), where |
| 89 | values 0..255 represent literal bytes, the value 256 indicates |
| 90 | end-of-block, and values 257..285 represent length codes |
| 91 | (possibly in conjunction with extra bits following the symbol |
| 92 | code) as follows: |
| 93 | |
| 94 | Length encoding : |
| 95 | Extra Extra Extra |
| 96 | Code Bits Length(s) Code Bits Lengths Code Bits Length(s) |
| 97 | ---- ---- ------ ---- ---- ------- ---- ---- ------- |
| 98 | 257 0 3 267 1 15,16 277 4 67-82 |
| 99 | 258 0 4 268 1 17,18 278 4 83-98 |
| 100 | 259 0 5 269 2 19-22 279 4 99-114 |
| 101 | 260 0 6 270 2 23-26 280 4 115-130 |
| 102 | 261 0 7 271 2 27-30 281 5 131-162 |
| 103 | 262 0 8 272 2 31-34 282 5 163-194 |
| 104 | 263 0 9 273 3 35-42 283 5 195-226 |
| 105 | 264 0 10 274 3 43-50 284 5 227-257 |
| 106 | 265 1 11,12 275 3 51-58 285 0 258 |
| 107 | 266 1 13,14 276 3 59-66 |
| 108 | |
| 109 | Distance encoding : |
| 110 | Extra Extra Extra |
| 111 | Code Bits Dist Code Bits Dist Code Bits Distance |
| 112 | ---- ---- ---- ---- ---- ------ ---- ---- -------- |
| 113 | 0 0 1 10 4 33-48 20 9 1025-1536 |
| 114 | 1 0 2 11 4 49-64 21 9 1537-2048 |
| 115 | 2 0 3 12 5 65-96 22 10 2049-3072 |
| 116 | 3 0 4 13 5 97-128 23 10 3073-4096 |
| 117 | 4 1 5,6 14 6 129-192 24 11 4097-6144 |
| 118 | 5 1 7,8 15 6 193-256 25 11 6145-8192 |
| 119 | 6 2 9-12 16 7 257-384 26 12 8193-12288 |
| 120 | 7 2 13-16 17 7 385-512 27 12 12289-16384 |
| 121 | 8 3 17-24 18 8 513-768 28 13 16385-24576 |
| 122 | 9 3 25-32 19 8 769-1024 29 13 24577-32768 |
| 123 | |
| 124 | 3.2.6. Compression with fixed Huffman codes (BTYPE=01) |
| 125 | |
| 126 | The Huffman codes for the two alphabets are fixed, and are not |
| 127 | represented explicitly in the data. The Huffman code lengths |
| 128 | for the literal/length alphabet are: |
| 129 | |
| 130 | Lit Value Bits Codes |
| 131 | --------- ---- ----- |
| 132 | 0 - 143 8 00110000 through |
| 133 | 10111111 |
| 134 | 144 - 255 9 110010000 through |
| 135 | 111111111 |
| 136 | 256 - 279 7 0000000 through |
| 137 | 0010111 |
| 138 | 280 - 287 8 11000000 through |
| 139 | 11000111 |
| 140 | |
| 141 | The code lengths are sufficient to generate the actual codes, |
| 142 | as described above; we show the codes in the table for added |
| 143 | clarity. Literal/length values 286-287 will never actually |
| 144 | occur in the compressed data, but participate in the code |
| 145 | construction. |
| 146 | |
| 147 | Distance codes 0-31 are represented by (fixed-length) 5-bit |
| 148 | codes, with possible additional bits as shown in the table |
| 149 | shown in Paragraph 3.2.5, above. Note that distance codes 30- |
| 150 | 31 will never actually occur in the compressed data. |
| 151 | |
| 152 | */ |
| 153 | |
| 154 | /* back references, built in a way that is optimal for 32/64 bits */ |
| 155 | union ref { |
| 156 | struct { |
| 157 | uint32_t pos; |
| 158 | uint32_t word; |
| 159 | } by32; |
| 160 | uint64_t by64; |
| 161 | }; |
| 162 | |
| 163 | #if defined(USE_64BIT_QUEUE) && defined(UNALIGNED_LE_OK) |
| 164 | |
| 165 | /* enqueue code x of <xbits> bits (LSB aligned, at most 24) and copy complete |
| 166 | * 32-bit words into output buffer. X must not contain non-zero bits above |
| 167 | * xbits. |
| 168 | */ |
| 169 | static inline void enqueue24(struct slz_stream *strm, uint32_t x, uint32_t xbits) |
| 170 | { |
| 171 | uint64_t queue = strm->queue + ((uint64_t)x << strm->qbits); |
| 172 | uint32_t qbits = strm->qbits + xbits; |
| 173 | |
| 174 | if (__builtin_expect(qbits >= 32, 1)) { |
| 175 | *(uint32_t *)strm->outbuf = queue; |
| 176 | queue >>= 32; |
| 177 | qbits -= 32; |
| 178 | strm->outbuf += 4; |
| 179 | } |
| 180 | |
| 181 | strm->queue = queue; |
| 182 | strm->qbits = qbits; |
| 183 | } |
| 184 | |
| 185 | #define enqueue8 enqueue24 |
| 186 | |
| 187 | /* flush the queue and align to next byte */ |
| 188 | static inline void flush_bits(struct slz_stream *strm) |
| 189 | { |
| 190 | if (strm->qbits > 0) |
| 191 | *strm->outbuf++ = strm->queue; |
| 192 | |
| 193 | if (strm->qbits > 8) |
| 194 | *strm->outbuf++ = strm->queue >> 8; |
| 195 | |
| 196 | if (strm->qbits > 16) |
| 197 | *strm->outbuf++ = strm->queue >> 16; |
| 198 | |
| 199 | if (strm->qbits > 24) |
| 200 | *strm->outbuf++ = strm->queue >> 24; |
| 201 | |
| 202 | strm->queue = 0; |
| 203 | strm->qbits = 0; |
| 204 | } |
| 205 | |
| 206 | #else /* non-64 bit or aligned or big endian */ |
| 207 | |
| 208 | /* enqueue code x of <xbits> bits (LSB aligned, at most 24) and copy complete |
| 209 | * bytes into out buf. X must not contain non-zero bits above xbits. Prefer |
| 210 | * enqueue8() when xbits is known for being 8 or less. |
| 211 | */ |
| 212 | static void enqueue24(struct slz_stream *strm, uint32_t x, uint32_t xbits) |
| 213 | { |
| 214 | uint32_t queue = strm->queue + (x << strm->qbits); |
| 215 | uint32_t qbits = strm->qbits + xbits; |
| 216 | |
| 217 | if (qbits >= 16) { |
| 218 | #ifndef UNALIGNED_LE_OK |
| 219 | strm->outbuf[0] = queue; |
| 220 | strm->outbuf[1] = queue >> 8; |
| 221 | #else |
| 222 | *(uint16_t *)strm->outbuf = queue; |
| 223 | #endif |
| 224 | strm->outbuf += 2; |
| 225 | queue >>= 16; |
| 226 | qbits -= 16; |
| 227 | } |
| 228 | |
| 229 | if (qbits >= 8) { |
| 230 | qbits -= 8; |
| 231 | *strm->outbuf++ = queue; |
| 232 | queue >>= 8; |
| 233 | } |
| 234 | strm->qbits = qbits; |
| 235 | strm->queue = queue; |
| 236 | return; |
| 237 | } |
| 238 | |
| 239 | /* enqueue code x of <xbits> bits (at most 8) and copy complete bytes into |
| 240 | * out buf. X must not contain non-zero bits above xbits. |
| 241 | */ |
| 242 | static inline void enqueue8(struct slz_stream *strm, uint32_t x, uint32_t xbits) |
| 243 | { |
| 244 | uint32_t queue = strm->queue + (x << strm->qbits); |
| 245 | uint32_t qbits = strm->qbits + xbits; |
| 246 | |
| 247 | if (__builtin_expect((signed)(qbits - 8) >= 0, 1)) { |
| 248 | qbits -= 8; |
| 249 | *strm->outbuf++ = queue; |
| 250 | queue >>= 8; |
| 251 | } |
| 252 | |
| 253 | strm->qbits = qbits; |
| 254 | strm->queue = queue; |
| 255 | } |
| 256 | |
| 257 | /* align to next byte */ |
| 258 | static inline void flush_bits(struct slz_stream *strm) |
| 259 | { |
| 260 | if (strm->qbits > 0) |
| 261 | *strm->outbuf++ = strm->queue; |
| 262 | |
| 263 | if (strm->qbits > 8) |
| 264 | *strm->outbuf++ = strm->queue >> 8; |
| 265 | |
| 266 | strm->queue = 0; |
| 267 | strm->qbits = 0; |
| 268 | } |
| 269 | #endif |
| 270 | |
| 271 | |
| 272 | /* only valid if buffer is already aligned */ |
| 273 | static inline void copy_8b(struct slz_stream *strm, uint32_t x) |
| 274 | { |
| 275 | *strm->outbuf++ = x; |
| 276 | } |
| 277 | |
| 278 | /* only valid if buffer is already aligned */ |
| 279 | static inline void copy_16b(struct slz_stream *strm, uint32_t x) |
| 280 | { |
| 281 | strm->outbuf[0] = x; |
| 282 | strm->outbuf[1] = x >> 8; |
| 283 | strm->outbuf += 2; |
| 284 | } |
| 285 | |
| 286 | /* only valid if buffer is already aligned */ |
| 287 | static inline void copy_32b(struct slz_stream *strm, uint32_t x) |
| 288 | { |
| 289 | strm->outbuf[0] = x; |
| 290 | strm->outbuf[1] = x >> 8; |
| 291 | strm->outbuf[2] = x >> 16; |
| 292 | strm->outbuf[3] = x >> 24; |
| 293 | strm->outbuf += 4; |
| 294 | } |
| 295 | |
| 296 | static inline void send_huff(struct slz_stream *strm, uint32_t code) |
| 297 | { |
| 298 | uint32_t bits; |
| 299 | |
| 300 | code = fixed_huff[code]; |
| 301 | bits = code & 15; |
| 302 | code >>= 4; |
| 303 | enqueue24(strm, code, bits); |
| 304 | } |
| 305 | |
| 306 | static inline void send_eob(struct slz_stream *strm) |
| 307 | { |
| 308 | enqueue8(strm, 0, 7); // direct encoding of 256 = EOB (cf RFC1951) |
| 309 | } |
| 310 | |
Ilya Shipitsin | b2be9a1 | 2021-04-24 13:25:42 +0500 | [diff] [blame] | 311 | /* copies <len> literals from <buf>. <more> indicates that there are data past |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 312 | * buf + <len>. <len> must not be null. |
| 313 | */ |
| 314 | static void copy_lit(struct slz_stream *strm, const void *buf, uint32_t len, int more) |
| 315 | { |
| 316 | uint32_t len2; |
| 317 | |
| 318 | do { |
| 319 | len2 = len; |
| 320 | if (__builtin_expect(len2 > 65535, 0)) |
| 321 | len2 = 65535; |
| 322 | |
| 323 | len -= len2; |
| 324 | |
| 325 | if (strm->state != SLZ_ST_EOB) |
| 326 | send_eob(strm); |
| 327 | |
| 328 | strm->state = (more || len) ? SLZ_ST_EOB : SLZ_ST_DONE; |
| 329 | |
| 330 | enqueue8(strm, !(more || len), 3); // BFINAL = !more ; BTYPE = 00 |
| 331 | flush_bits(strm); |
| 332 | copy_16b(strm, len2); // len2 |
| 333 | copy_16b(strm, ~len2); // nlen2 |
| 334 | memcpy(strm->outbuf, buf, len2); |
| 335 | buf += len2; |
| 336 | strm->outbuf += len2; |
| 337 | } while (len); |
| 338 | } |
| 339 | |
Ilya Shipitsin | b2be9a1 | 2021-04-24 13:25:42 +0500 | [diff] [blame] | 340 | /* copies <len> literals from <buf>. <more> indicates that there are data past |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 341 | * buf + <len>. <len> must not be null. |
| 342 | */ |
| 343 | static void copy_lit_huff(struct slz_stream *strm, const unsigned char *buf, uint32_t len, int more) |
| 344 | { |
| 345 | uint32_t pos; |
| 346 | |
| 347 | /* This ugly construct limits the mount of tests and optimizes for the |
| 348 | * most common case (more > 0). |
| 349 | */ |
| 350 | if (strm->state == SLZ_ST_EOB) { |
| 351 | eob: |
| 352 | strm->state = more ? SLZ_ST_FIXED : SLZ_ST_LAST; |
| 353 | enqueue8(strm, 2 + !more, 3); // BFINAL = !more ; BTYPE = 01 |
| 354 | } |
| 355 | else if (!more) { |
| 356 | send_eob(strm); |
| 357 | goto eob; |
| 358 | } |
| 359 | |
| 360 | pos = 0; |
| 361 | do { |
| 362 | send_huff(strm, buf[pos++]); |
| 363 | } while (pos < len); |
| 364 | } |
| 365 | |
| 366 | /* format: |
| 367 | * bit0..31 = word |
| 368 | * bit32..63 = last position in buffer of similar content |
| 369 | */ |
| 370 | |
| 371 | /* This hash provides good average results on HTML contents, and is among the |
| 372 | * few which provide almost optimal results on various different pages. |
| 373 | */ |
| 374 | static inline uint32_t slz_hash(uint32_t a) |
| 375 | { |
| 376 | #if defined(__ARM_FEATURE_CRC32) |
| 377 | __asm__ volatile("crc32w %w0,%w0,%w1" : "+r"(a) : "r"(0)); |
| 378 | return a >> (32 - HASH_BITS); |
| 379 | #else |
| 380 | return ((a << 19) + (a << 6) - a) >> (32 - HASH_BITS); |
| 381 | #endif |
| 382 | } |
| 383 | |
| 384 | /* This function compares buffers <a> and <b> and reads 32 or 64 bits at a time |
| 385 | * during the approach. It makes us of unaligned little endian memory accesses |
| 386 | * on capable architectures. <max> is the maximum number of bytes that can be |
| 387 | * read, so both <a> and <b> must have at least <max> bytes ahead. <max> may |
| 388 | * safely be null or negative if that simplifies computations in the caller. |
| 389 | */ |
| 390 | static inline long memmatch(const unsigned char *a, const unsigned char *b, long max) |
| 391 | { |
| 392 | long len = 0; |
| 393 | |
| 394 | #ifdef UNALIGNED_LE_OK |
| 395 | unsigned long xor; |
| 396 | |
| 397 | while (1) { |
| 398 | if ((long)(len + 2 * sizeof(long)) > max) { |
| 399 | while (len < max) { |
| 400 | if (a[len] != b[len]) |
| 401 | break; |
| 402 | len++; |
| 403 | } |
| 404 | return len; |
| 405 | } |
| 406 | |
| 407 | xor = *(long *)&a[len] ^ *(long *)&b[len]; |
| 408 | if (xor) |
| 409 | break; |
| 410 | len += sizeof(long); |
| 411 | |
| 412 | xor = *(long *)&a[len] ^ *(long *)&b[len]; |
| 413 | if (xor) |
| 414 | break; |
| 415 | len += sizeof(long); |
| 416 | } |
| 417 | |
| 418 | #if defined(__x86_64__) || defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) |
| 419 | /* x86 has bsf. We know that xor is non-null here */ |
| 420 | asm("bsf %1,%0\n" : "=r"(xor) : "0" (xor)); |
| 421 | return len + xor / 8; |
| 422 | #else |
| 423 | if (sizeof(long) > 4 && !(xor & 0xffffffff)) { |
| 424 | /* This code is optimized out on 32-bit archs, but we still |
| 425 | * need to shift in two passes to avoid a warning. It is |
| 426 | * properly optimized out as a single shift. |
| 427 | */ |
| 428 | xor >>= 16; xor >>= 16; |
| 429 | if (xor & 0xffff) { |
| 430 | if (xor & 0xff) |
| 431 | return len + 4; |
| 432 | return len + 5; |
| 433 | } |
| 434 | if (xor & 0xffffff) |
| 435 | return len + 6; |
| 436 | return len + 7; |
| 437 | } |
| 438 | |
| 439 | if (xor & 0xffff) { |
| 440 | if (xor & 0xff) |
| 441 | return len; |
| 442 | return len + 1; |
| 443 | } |
| 444 | if (xor & 0xffffff) |
| 445 | return len + 2; |
| 446 | return len + 3; |
| 447 | #endif // x86 |
| 448 | |
| 449 | #else // UNALIGNED_LE_OK |
| 450 | /* This is the generic version for big endian or unaligned-incompatible |
| 451 | * architectures. |
| 452 | */ |
| 453 | while (len < max) { |
| 454 | if (a[len] != b[len]) |
| 455 | break; |
| 456 | len++; |
| 457 | } |
| 458 | return len; |
| 459 | |
| 460 | #endif |
| 461 | } |
| 462 | |
| 463 | /* sets <count> BYTES to -32769 in <refs> so that any uninitialized entry will |
| 464 | * verify (pos-last-1 >= 32768) and be ignored. <count> must be a multiple of |
| 465 | * 128 bytes and <refs> must be at least one count in length. It's supposed to |
| 466 | * be applied to 64-bit aligned data exclusively, which makes it slightly |
| 467 | * faster than the regular memset() since no alignment check is performed. |
| 468 | */ |
| 469 | void reset_refs(union ref *refs, long count) |
| 470 | { |
| 471 | /* avoid a shift/mask by casting to void* */ |
| 472 | union ref *end = (void *)refs + count; |
| 473 | |
| 474 | do { |
| 475 | refs[ 0].by64 = -32769; |
| 476 | refs[ 1].by64 = -32769; |
| 477 | refs[ 2].by64 = -32769; |
| 478 | refs[ 3].by64 = -32769; |
| 479 | refs[ 4].by64 = -32769; |
| 480 | refs[ 5].by64 = -32769; |
| 481 | refs[ 6].by64 = -32769; |
| 482 | refs[ 7].by64 = -32769; |
| 483 | refs[ 8].by64 = -32769; |
| 484 | refs[ 9].by64 = -32769; |
| 485 | refs[10].by64 = -32769; |
| 486 | refs[11].by64 = -32769; |
| 487 | refs[12].by64 = -32769; |
| 488 | refs[13].by64 = -32769; |
| 489 | refs[14].by64 = -32769; |
| 490 | refs[15].by64 = -32769; |
| 491 | refs += 16; |
| 492 | } while (refs < end); |
| 493 | } |
| 494 | |
| 495 | /* Compresses <ilen> bytes from <in> into <out> according to RFC1951. The |
| 496 | * output result may be up to 5 bytes larger than the input, to which 2 extra |
| 497 | * bytes may be added to send the last chunk due to BFINAL+EOB encoding (10 |
| 498 | * bits) when <more> is not set. The caller is responsible for ensuring there |
| 499 | * is enough room in the output buffer for this. The amount of output bytes is |
| 500 | * returned, and no CRC is computed. |
| 501 | */ |
| 502 | long slz_rfc1951_encode(struct slz_stream *strm, unsigned char *out, const unsigned char *in, long ilen, int more) |
| 503 | { |
| 504 | long rem = ilen; |
| 505 | unsigned long pos = 0; |
| 506 | unsigned long last; |
| 507 | uint32_t word = 0; |
| 508 | long mlen; |
| 509 | uint32_t h; |
| 510 | uint64_t ent; |
| 511 | |
| 512 | uint32_t plit = 0; |
| 513 | uint32_t bit9 = 0; |
| 514 | uint32_t dist, code; |
| 515 | union ref refs[1 << HASH_BITS]; |
| 516 | |
| 517 | if (!strm->level) { |
| 518 | /* force to send as literals (eg to preserve CPU) */ |
| 519 | strm->outbuf = out; |
| 520 | plit = pos = ilen; |
| 521 | bit9 = 52; /* force literal dump */ |
| 522 | goto final_lit_dump; |
| 523 | } |
| 524 | |
| 525 | reset_refs(refs, sizeof(refs)); |
| 526 | |
| 527 | strm->outbuf = out; |
| 528 | |
| 529 | #ifndef UNALIGNED_FASTER |
| 530 | word = ((unsigned char)in[pos] << 8) + ((unsigned char)in[pos + 1] << 16) + ((unsigned char)in[pos + 2] << 24); |
| 531 | #endif |
| 532 | while (rem >= 4) { |
| 533 | #ifndef UNALIGNED_FASTER |
| 534 | word = ((unsigned char)in[pos + 3] << 24) + (word >> 8); |
| 535 | #else |
| 536 | word = *(uint32_t *)&in[pos]; |
| 537 | #endif |
| 538 | h = slz_hash(word); |
| 539 | asm volatile ("" ::); // prevent gcc from trying to be smart with the prefetch |
| 540 | |
| 541 | if (sizeof(long) >= 8) { |
| 542 | ent = refs[h].by64; |
| 543 | last = (uint32_t)ent; |
| 544 | ent >>= 32; |
| 545 | refs[h].by64 = ((uint64_t)pos) + ((uint64_t)word << 32); |
| 546 | } else { |
| 547 | ent = refs[h].by32.word; |
| 548 | last = refs[h].by32.pos; |
| 549 | refs[h].by32.pos = pos; |
| 550 | refs[h].by32.word = word; |
| 551 | } |
| 552 | |
Willy Tarreau | df7cd97 | 2021-08-28 12:10:49 +0200 | [diff] [blame] | 553 | #ifdef FIND_OPTIMAL_MATCH |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 554 | /* Experimental code to see what could be saved with an ideal |
| 555 | * longest match lookup algorithm. This one is very slow but |
| 556 | * scans the whole window. In short, here are the savings : |
| 557 | * file orig fast(ratio) optimal(ratio) |
| 558 | * README 5185 3419 (65.9%) 3165 (61.0%) -7.5% |
| 559 | * index.html 76799 35662 (46.4%) 29875 (38.9%) -16.3% |
| 560 | * rfc1952.c 29383 13442 (45.7%) 11793 (40.1%) -12.3% |
| 561 | * |
| 562 | * Thus the savings to expect for large files is at best 16%. |
| 563 | * |
| 564 | * A non-colliding hash gives 33025 instead of 35662 (-7.4%), |
| 565 | * and keeping the last two entries gives 31724 (-11.0%). |
| 566 | */ |
| 567 | unsigned long scan; |
| 568 | int saved = 0; |
| 569 | int bestpos = 0; |
| 570 | int bestlen = 0; |
| 571 | int firstlen = 0; |
| 572 | int max_lookup = 2; // 0 = no limit |
| 573 | |
| 574 | for (scan = pos - 1; scan < pos && (unsigned long)(pos - scan - 1) < 32768; scan--) { |
| 575 | if (*(uint32_t *)(in + scan) != word) |
| 576 | continue; |
| 577 | |
| 578 | len = memmatch(in + pos, in + scan, rem); |
| 579 | if (!bestlen) |
| 580 | firstlen = len; |
| 581 | |
| 582 | if (len > bestlen) { |
| 583 | bestlen = len; |
| 584 | bestpos = scan; |
| 585 | } |
| 586 | if (!--max_lookup) |
| 587 | break; |
| 588 | } |
| 589 | if (bestlen) { |
| 590 | //printf("pos=%d last=%d bestpos=%d word=%08x ent=%08x len=%d\n", |
| 591 | // (int)pos, (int)last, (int)bestpos, (int)word, (int)ent, bestlen); |
| 592 | last = bestpos; |
| 593 | ent = word; |
| 594 | saved += bestlen - firstlen; |
| 595 | } |
| 596 | //fprintf(stderr, "first=%d best=%d saved_total=%d\n", firstlen, bestlen, saved); |
| 597 | #endif |
| 598 | |
| 599 | if ((uint32_t)ent != word) { |
| 600 | send_as_lit: |
| 601 | rem--; |
| 602 | plit++; |
| 603 | bit9 += ((unsigned char)word >= 144); |
| 604 | pos++; |
| 605 | continue; |
| 606 | } |
| 607 | |
| 608 | /* We reject pos = last and pos > last+32768 */ |
| 609 | if ((unsigned long)(pos - last - 1) >= 32768) |
| 610 | goto send_as_lit; |
| 611 | |
| 612 | /* Note: cannot encode a length larger than 258 bytes */ |
| 613 | mlen = memmatch(in + pos + 4, in + last + 4, (rem > 258 ? 258 : rem) - 4) + 4; |
| 614 | |
| 615 | /* found a matching entry */ |
| 616 | |
| 617 | if (bit9 >= 52 && mlen < 6) |
| 618 | goto send_as_lit; |
| 619 | |
| 620 | /* compute the output code, its size and the length's size in |
| 621 | * bits to know if the reference is cheaper than literals. |
| 622 | */ |
| 623 | code = len_fh[mlen]; |
| 624 | |
| 625 | /* direct mapping of dist->huffman code */ |
| 626 | dist = fh_dist_table[pos - last - 1]; |
| 627 | |
| 628 | /* if encoding the dist+length is more expensive than sending |
| 629 | * the equivalent as bytes, lets keep the literals. |
| 630 | */ |
| 631 | if ((dist & 0x1f) + (code >> 16) + 8 >= 8 * mlen + bit9) |
| 632 | goto send_as_lit; |
| 633 | |
| 634 | /* first, copy pending literals */ |
| 635 | if (plit) { |
| 636 | /* Huffman encoding requires 9 bits for octets 144..255, so this |
| 637 | * is a waste of space for binary data. Switching between Huffman |
| 638 | * and no-comp then huffman consumes 52 bits (7 for EOB + 3 for |
| 639 | * block type + 7 for alignment + 32 for LEN+NLEN + 3 for next |
| 640 | * block. Only use plain literals if there are more than 52 bits |
| 641 | * to save then. |
| 642 | */ |
| 643 | if (bit9 >= 52) |
| 644 | copy_lit(strm, in + pos - plit, plit, 1); |
| 645 | else |
| 646 | copy_lit_huff(strm, in + pos - plit, plit, 1); |
| 647 | |
| 648 | plit = 0; |
| 649 | } |
| 650 | |
| 651 | /* use mode 01 - fixed huffman */ |
| 652 | if (strm->state == SLZ_ST_EOB) { |
| 653 | strm->state = SLZ_ST_FIXED; |
| 654 | enqueue8(strm, 0x02, 3); // BTYPE = 01, BFINAL = 0 |
| 655 | } |
| 656 | |
| 657 | /* copy the length first */ |
| 658 | enqueue24(strm, code & 0xFFFF, code >> 16); |
| 659 | |
| 660 | /* in fixed huffman mode, dist is fixed 5 bits */ |
| 661 | enqueue24(strm, dist >> 5, dist & 0x1f); |
| 662 | bit9 = 0; |
| 663 | rem -= mlen; |
| 664 | pos += mlen; |
| 665 | |
| 666 | #ifndef UNALIGNED_FASTER |
| 667 | #ifdef UNALIGNED_LE_OK |
| 668 | word = *(uint32_t *)&in[pos - 1]; |
| 669 | #else |
| 670 | word = ((unsigned char)in[pos] << 8) + ((unsigned char)in[pos + 1] << 16) + ((unsigned char)in[pos + 2] << 24); |
| 671 | #endif |
| 672 | #endif |
| 673 | } |
| 674 | |
| 675 | if (__builtin_expect(rem, 0)) { |
| 676 | /* we're reading the 1..3 last bytes */ |
| 677 | plit += rem; |
| 678 | do { |
| 679 | bit9 += ((unsigned char)in[pos++] >= 144); |
| 680 | } while (--rem); |
| 681 | } |
| 682 | |
| 683 | final_lit_dump: |
| 684 | /* now copy remaining literals or mark the end */ |
| 685 | if (plit) { |
| 686 | if (bit9 >= 52) |
| 687 | copy_lit(strm, in + pos - plit, plit, more); |
| 688 | else |
| 689 | copy_lit_huff(strm, in + pos - plit, plit, more); |
| 690 | |
| 691 | plit = 0; |
| 692 | } |
| 693 | |
| 694 | strm->ilen += ilen; |
| 695 | return strm->outbuf - out; |
| 696 | } |
| 697 | |
| 698 | /* Initializes stream <strm> for use with raw deflate (rfc1951). The CRC is |
| 699 | * unused but set to zero. The compression level passed in <level> is set. This |
| 700 | * value can only be 0 (no compression) or 1 (compression) and other values |
| 701 | * will lead to unpredictable behaviour. The function always returns 0. |
| 702 | */ |
| 703 | int slz_rfc1951_init(struct slz_stream *strm, int level) |
| 704 | { |
| 705 | strm->state = SLZ_ST_EOB; // no header |
| 706 | strm->level = level; |
| 707 | strm->format = SLZ_FMT_DEFLATE; |
| 708 | strm->crc32 = 0; |
| 709 | strm->ilen = 0; |
| 710 | strm->qbits = 0; |
| 711 | strm->queue = 0; |
| 712 | return 0; |
| 713 | } |
| 714 | |
| 715 | /* Flushes any pending for stream <strm> into buffer <buf>, then sends BTYPE=1 |
| 716 | * and BFINAL=1 if needed. The stream ends in SLZ_ST_DONE. It returns the number |
| 717 | * of bytes emitted. The trailer consists in flushing the possibly pending bits |
| 718 | * from the queue (up to 7 bits), then possibly EOB (7 bits), then 3 bits, EOB, |
| 719 | * a rounding to the next byte, which amounts to a total of 4 bytes max, that |
| 720 | * the caller must ensure are available before calling the function. |
| 721 | */ |
| 722 | int slz_rfc1951_finish(struct slz_stream *strm, unsigned char *buf) |
| 723 | { |
| 724 | strm->outbuf = buf; |
| 725 | |
| 726 | if (strm->state == SLZ_ST_FIXED || strm->state == SLZ_ST_LAST) { |
| 727 | strm->state = (strm->state == SLZ_ST_LAST) ? SLZ_ST_DONE : SLZ_ST_EOB; |
| 728 | send_eob(strm); |
| 729 | } |
| 730 | |
| 731 | if (strm->state != SLZ_ST_DONE) { |
| 732 | /* send BTYPE=1, BFINAL=1 */ |
| 733 | enqueue8(strm, 3, 3); |
| 734 | send_eob(strm); |
| 735 | strm->state = SLZ_ST_DONE; |
| 736 | } |
| 737 | |
| 738 | flush_bits(strm); |
| 739 | return strm->outbuf - buf; |
| 740 | } |
| 741 | |
| 742 | /* Now RFC1952-specific declarations and extracts from RFC. |
| 743 | * From RFC1952 about the GZIP file format : |
| 744 | |
| 745 | A gzip file consists of a series of "members" ... |
| 746 | |
| 747 | 2.3. Member format |
| 748 | |
| 749 | Each member has the following structure: |
| 750 | |
| 751 | +---+---+---+---+---+---+---+---+---+---+ |
| 752 | |ID1|ID2|CM |FLG| MTIME |XFL|OS | (more-->) |
| 753 | +---+---+---+---+---+---+---+---+---+---+ |
| 754 | |
| 755 | (if FLG.FEXTRA set) |
| 756 | |
| 757 | +---+---+=================================+ |
| 758 | | XLEN |...XLEN bytes of "extra field"...| (more-->) |
| 759 | +---+---+=================================+ |
| 760 | |
| 761 | (if FLG.FNAME set) |
| 762 | |
| 763 | +=========================================+ |
| 764 | |...original file name, zero-terminated...| (more-->) |
| 765 | +=========================================+ |
| 766 | |
| 767 | (if FLG.FCOMMENT set) |
| 768 | |
| 769 | +===================================+ |
| 770 | |...file comment, zero-terminated...| (more-->) |
| 771 | +===================================+ |
| 772 | |
| 773 | (if FLG.FHCRC set) |
| 774 | |
| 775 | +---+---+ |
| 776 | | CRC16 | |
| 777 | +---+---+ |
| 778 | |
| 779 | +=======================+ |
| 780 | |...compressed blocks...| (more-->) |
| 781 | +=======================+ |
| 782 | |
| 783 | 0 1 2 3 4 5 6 7 |
| 784 | +---+---+---+---+---+---+---+---+ |
| 785 | | CRC32 | ISIZE | |
| 786 | +---+---+---+---+---+---+---+---+ |
| 787 | |
| 788 | |
| 789 | 2.3.1. Member header and trailer |
| 790 | |
| 791 | ID1 (IDentification 1) |
| 792 | ID2 (IDentification 2) |
| 793 | These have the fixed values ID1 = 31 (0x1f, \037), ID2 = 139 |
| 794 | (0x8b, \213), to identify the file as being in gzip format. |
| 795 | |
| 796 | CM (Compression Method) |
| 797 | This identifies the compression method used in the file. CM |
| 798 | = 0-7 are reserved. CM = 8 denotes the "deflate" |
| 799 | compression method, which is the one customarily used by |
| 800 | gzip and which is documented elsewhere. |
| 801 | |
| 802 | FLG (FLaGs) |
| 803 | This flag byte is divided into individual bits as follows: |
| 804 | |
| 805 | bit 0 FTEXT |
| 806 | bit 1 FHCRC |
| 807 | bit 2 FEXTRA |
| 808 | bit 3 FNAME |
| 809 | bit 4 FCOMMENT |
| 810 | bit 5 reserved |
| 811 | bit 6 reserved |
| 812 | bit 7 reserved |
| 813 | |
| 814 | Reserved FLG bits must be zero. |
| 815 | |
| 816 | MTIME (Modification TIME) |
| 817 | This gives the most recent modification time of the original |
| 818 | file being compressed. The time is in Unix format, i.e., |
| 819 | seconds since 00:00:00 GMT, Jan. 1, 1970. (Note that this |
| 820 | may cause problems for MS-DOS and other systems that use |
| 821 | local rather than Universal time.) If the compressed data |
| 822 | did not come from a file, MTIME is set to the time at which |
| 823 | compression started. MTIME = 0 means no time stamp is |
| 824 | available. |
| 825 | |
| 826 | XFL (eXtra FLags) |
| 827 | These flags are available for use by specific compression |
| 828 | methods. The "deflate" method (CM = 8) sets these flags as |
| 829 | follows: |
| 830 | |
| 831 | XFL = 2 - compressor used maximum compression, |
| 832 | slowest algorithm |
| 833 | XFL = 4 - compressor used fastest algorithm |
| 834 | |
| 835 | OS (Operating System) |
| 836 | This identifies the type of file system on which compression |
| 837 | took place. This may be useful in determining end-of-line |
| 838 | convention for text files. The currently defined values are |
| 839 | as follows: |
| 840 | |
| 841 | 0 - FAT filesystem (MS-DOS, OS/2, NT/Win32) |
| 842 | 1 - Amiga |
| 843 | 2 - VMS (or OpenVMS) |
| 844 | 3 - Unix |
| 845 | 4 - VM/CMS |
| 846 | 5 - Atari TOS |
| 847 | 6 - HPFS filesystem (OS/2, NT) |
| 848 | 7 - Macintosh |
| 849 | 8 - Z-System |
| 850 | 9 - CP/M |
| 851 | 10 - TOPS-20 |
| 852 | 11 - NTFS filesystem (NT) |
| 853 | 12 - QDOS |
| 854 | 13 - Acorn RISCOS |
| 855 | 255 - unknown |
| 856 | |
| 857 | ==> A file compressed using "gzip -1" on Unix-like systems can be : |
| 858 | |
| 859 | 1F 8B 08 00 00 00 00 00 04 03 |
| 860 | <deflate-compressed stream> |
| 861 | crc32 size32 |
| 862 | */ |
| 863 | |
| 864 | static const unsigned char gzip_hdr[] = { 0x1F, 0x8B, // ID1, ID2 |
| 865 | 0x08, 0x00, // Deflate, flags (none) |
| 866 | 0x00, 0x00, 0x00, 0x00, // mtime: none |
| 867 | 0x04, 0x03 }; // fastest comp, OS=Unix |
| 868 | |
| 869 | static inline uint32_t crc32_char(uint32_t crc, uint8_t x) |
| 870 | { |
| 871 | #if defined(__ARM_FEATURE_CRC32) |
| 872 | crc = ~crc; |
| 873 | __asm__ volatile("crc32b %w0,%w0,%w1" : "+r"(crc) : "r"(x)); |
| 874 | crc = ~crc; |
| 875 | #else |
| 876 | crc = crc32_fast[0][(crc ^ x) & 0xff] ^ (crc >> 8); |
| 877 | #endif |
| 878 | return crc; |
| 879 | } |
| 880 | |
| 881 | static inline uint32_t crc32_uint32(uint32_t data) |
| 882 | { |
| 883 | #if defined(__ARM_FEATURE_CRC32) |
| 884 | __asm__ volatile("crc32w %w0,%w0,%w1" : "+r"(data) : "r"(~0UL)); |
| 885 | data = ~data; |
| 886 | #else |
| 887 | data = crc32_fast[3][(data >> 0) & 0xff] ^ |
| 888 | crc32_fast[2][(data >> 8) & 0xff] ^ |
| 889 | crc32_fast[1][(data >> 16) & 0xff] ^ |
| 890 | crc32_fast[0][(data >> 24) & 0xff]; |
| 891 | #endif |
| 892 | return data; |
| 893 | } |
| 894 | |
| 895 | /* Modified version originally from RFC1952, working with non-inverting CRCs */ |
| 896 | uint32_t slz_crc32_by1(uint32_t crc, const unsigned char *buf, int len) |
| 897 | { |
| 898 | int n; |
| 899 | |
| 900 | for (n = 0; n < len; n++) |
| 901 | crc = crc32_char(crc, buf[n]); |
| 902 | return crc; |
| 903 | } |
| 904 | |
| 905 | /* This version computes the crc32 of <buf> over <len> bytes, doing most of it |
| 906 | * in 32-bit chunks. |
| 907 | */ |
| 908 | uint32_t slz_crc32_by4(uint32_t crc, const unsigned char *buf, int len) |
| 909 | { |
| 910 | const unsigned char *end = buf + len; |
| 911 | |
| 912 | while (buf <= end - 16) { |
| 913 | #ifdef UNALIGNED_LE_OK |
| 914 | #if defined(__ARM_FEATURE_CRC32) |
| 915 | crc = ~crc; |
| 916 | __asm__ volatile("crc32w %w0,%w0,%w1" : "+r"(crc) : "r"(*(uint32_t*)(buf))); |
| 917 | __asm__ volatile("crc32w %w0,%w0,%w1" : "+r"(crc) : "r"(*(uint32_t*)(buf + 4))); |
| 918 | __asm__ volatile("crc32w %w0,%w0,%w1" : "+r"(crc) : "r"(*(uint32_t*)(buf + 8))); |
| 919 | __asm__ volatile("crc32w %w0,%w0,%w1" : "+r"(crc) : "r"(*(uint32_t*)(buf + 12))); |
| 920 | crc = ~crc; |
| 921 | #else |
| 922 | crc ^= *(uint32_t *)buf; |
| 923 | crc = crc32_uint32(crc); |
| 924 | |
| 925 | crc ^= *(uint32_t *)(buf + 4); |
| 926 | crc = crc32_uint32(crc); |
| 927 | |
| 928 | crc ^= *(uint32_t *)(buf + 8); |
| 929 | crc = crc32_uint32(crc); |
| 930 | |
| 931 | crc ^= *(uint32_t *)(buf + 12); |
| 932 | crc = crc32_uint32(crc); |
| 933 | #endif |
| 934 | #else |
| 935 | crc = crc32_fast[3][(buf[0] ^ (crc >> 0)) & 0xff] ^ |
| 936 | crc32_fast[2][(buf[1] ^ (crc >> 8)) & 0xff] ^ |
| 937 | crc32_fast[1][(buf[2] ^ (crc >> 16)) & 0xff] ^ |
| 938 | crc32_fast[0][(buf[3] ^ (crc >> 24)) & 0xff]; |
| 939 | |
| 940 | crc = crc32_fast[3][(buf[4] ^ (crc >> 0)) & 0xff] ^ |
| 941 | crc32_fast[2][(buf[5] ^ (crc >> 8)) & 0xff] ^ |
| 942 | crc32_fast[1][(buf[6] ^ (crc >> 16)) & 0xff] ^ |
| 943 | crc32_fast[0][(buf[7] ^ (crc >> 24)) & 0xff]; |
| 944 | |
| 945 | crc = crc32_fast[3][(buf[8] ^ (crc >> 0)) & 0xff] ^ |
| 946 | crc32_fast[2][(buf[9] ^ (crc >> 8)) & 0xff] ^ |
| 947 | crc32_fast[1][(buf[10] ^ (crc >> 16)) & 0xff] ^ |
| 948 | crc32_fast[0][(buf[11] ^ (crc >> 24)) & 0xff]; |
| 949 | |
| 950 | crc = crc32_fast[3][(buf[12] ^ (crc >> 0)) & 0xff] ^ |
| 951 | crc32_fast[2][(buf[13] ^ (crc >> 8)) & 0xff] ^ |
| 952 | crc32_fast[1][(buf[14] ^ (crc >> 16)) & 0xff] ^ |
| 953 | crc32_fast[0][(buf[15] ^ (crc >> 24)) & 0xff]; |
| 954 | #endif |
| 955 | buf += 16; |
| 956 | } |
| 957 | |
| 958 | while (buf <= end - 4) { |
| 959 | #ifdef UNALIGNED_LE_OK |
| 960 | crc ^= *(uint32_t *)buf; |
| 961 | crc = crc32_uint32(crc); |
| 962 | #else |
| 963 | crc = crc32_fast[3][(buf[0] ^ (crc >> 0)) & 0xff] ^ |
| 964 | crc32_fast[2][(buf[1] ^ (crc >> 8)) & 0xff] ^ |
| 965 | crc32_fast[1][(buf[2] ^ (crc >> 16)) & 0xff] ^ |
| 966 | crc32_fast[0][(buf[3] ^ (crc >> 24)) & 0xff]; |
| 967 | #endif |
| 968 | buf += 4; |
| 969 | } |
| 970 | |
| 971 | while (buf < end) |
Willy Tarreau | 027fdcb | 2021-05-12 08:34:36 +0200 | [diff] [blame] | 972 | crc = crc32_char(crc, *buf++); |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 973 | return crc; |
| 974 | } |
| 975 | |
| 976 | /* uses the most suitable crc32 function to update crc on <buf, len> */ |
| 977 | static inline uint32_t update_crc(uint32_t crc, const void *buf, int len) |
| 978 | { |
| 979 | return slz_crc32_by4(crc, buf, len); |
| 980 | } |
| 981 | |
| 982 | /* Sends the gzip header for stream <strm> into buffer <buf>. When it's done, |
| 983 | * the stream state is updated to SLZ_ST_EOB. It returns the number of bytes |
| 984 | * emitted which is always 10. The caller is responsible for ensuring there's |
| 985 | * always enough room in the buffer. |
| 986 | */ |
| 987 | int slz_rfc1952_send_header(struct slz_stream *strm, unsigned char *buf) |
| 988 | { |
| 989 | memcpy(buf, gzip_hdr, sizeof(gzip_hdr)); |
| 990 | strm->state = SLZ_ST_EOB; |
| 991 | return sizeof(gzip_hdr); |
| 992 | } |
| 993 | |
| 994 | /* Encodes the block according to rfc1952. This means that the CRC of the input |
| 995 | * block is computed according to the CRC32 algorithm. If the header was never |
| 996 | * sent, it may be sent first. The number of output bytes is returned. |
| 997 | */ |
| 998 | long slz_rfc1952_encode(struct slz_stream *strm, unsigned char *out, const unsigned char *in, long ilen, int more) |
| 999 | { |
| 1000 | long ret = 0; |
| 1001 | |
| 1002 | if (__builtin_expect(strm->state == SLZ_ST_INIT, 0)) |
| 1003 | ret += slz_rfc1952_send_header(strm, out); |
| 1004 | |
| 1005 | strm->crc32 = update_crc(strm->crc32, in, ilen); |
| 1006 | ret += slz_rfc1951_encode(strm, out + ret, in, ilen, more); |
| 1007 | return ret; |
| 1008 | } |
| 1009 | |
| 1010 | /* Initializes stream <strm> for use with the gzip format (rfc1952). The |
| 1011 | * compression level passed in <level> is set. This value can only be 0 (no |
| 1012 | * compression) or 1 (compression) and other values will lead to unpredictable |
| 1013 | * behaviour. The function always returns 0. |
| 1014 | */ |
| 1015 | int slz_rfc1952_init(struct slz_stream *strm, int level) |
| 1016 | { |
| 1017 | strm->state = SLZ_ST_INIT; |
| 1018 | strm->level = level; |
| 1019 | strm->format = SLZ_FMT_GZIP; |
| 1020 | strm->crc32 = 0; |
| 1021 | strm->ilen = 0; |
| 1022 | strm->qbits = 0; |
| 1023 | strm->queue = 0; |
| 1024 | return 0; |
| 1025 | } |
| 1026 | |
| 1027 | /* Flushes pending bits and sends the gzip trailer for stream <strm> into |
| 1028 | * buffer <buf>. When it's done, the stream state is updated to SLZ_ST_END. It |
| 1029 | * returns the number of bytes emitted. The trailer consists in flushing the |
| 1030 | * possibly pending bits from the queue (up to 24 bits), rounding to the next |
| 1031 | * byte, then 4 bytes for the CRC and another 4 bytes for the input length. |
Ilya Shipitsin | b2be9a1 | 2021-04-24 13:25:42 +0500 | [diff] [blame] | 1032 | * That may about to 4+4+4 = 12 bytes, that the caller must ensure are |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 1033 | * available before calling the function. Note that if the initial header was |
| 1034 | * never sent, it will be sent first as well (10 extra bytes). |
| 1035 | */ |
| 1036 | int slz_rfc1952_finish(struct slz_stream *strm, unsigned char *buf) |
| 1037 | { |
| 1038 | strm->outbuf = buf; |
| 1039 | |
| 1040 | if (__builtin_expect(strm->state == SLZ_ST_INIT, 0)) |
| 1041 | strm->outbuf += slz_rfc1952_send_header(strm, strm->outbuf); |
| 1042 | |
| 1043 | slz_rfc1951_finish(strm, strm->outbuf); |
| 1044 | copy_32b(strm, strm->crc32); |
| 1045 | copy_32b(strm, strm->ilen); |
| 1046 | strm->state = SLZ_ST_END; |
| 1047 | |
| 1048 | return strm->outbuf - buf; |
| 1049 | } |
| 1050 | |
| 1051 | |
| 1052 | /* RFC1950-specific stuff. This is for the Zlib stream format. |
| 1053 | * From RFC1950 (zlib) : |
| 1054 | * |
| 1055 | |
| 1056 | 2.2. Data format |
| 1057 | |
| 1058 | A zlib stream has the following structure: |
| 1059 | |
| 1060 | 0 1 |
| 1061 | +---+---+ |
| 1062 | |CMF|FLG| (more-->) |
| 1063 | +---+---+ |
| 1064 | |
| 1065 | |
| 1066 | (if FLG.FDICT set) |
| 1067 | |
| 1068 | 0 1 2 3 |
| 1069 | +---+---+---+---+ |
| 1070 | | DICTID | (more-->) |
| 1071 | +---+---+---+---+ |
| 1072 | |
| 1073 | +=====================+---+---+---+---+ |
| 1074 | |...compressed data...| ADLER32 | |
| 1075 | +=====================+---+---+---+---+ |
| 1076 | |
| 1077 | Any data which may appear after ADLER32 are not part of the zlib |
| 1078 | stream. |
| 1079 | |
| 1080 | CMF (Compression Method and flags) |
| 1081 | This byte is divided into a 4-bit compression method and a 4- |
| 1082 | bit information field depending on the compression method. |
| 1083 | |
| 1084 | bits 0 to 3 CM Compression method |
| 1085 | bits 4 to 7 CINFO Compression info |
| 1086 | |
| 1087 | CM (Compression method) |
| 1088 | This identifies the compression method used in the file. CM = 8 |
| 1089 | denotes the "deflate" compression method with a window size up |
| 1090 | to 32K. This is the method used by gzip and PNG (see |
| 1091 | references [1] and [2] in Chapter 3, below, for the reference |
| 1092 | documents). CM = 15 is reserved. It might be used in a future |
| 1093 | version of this specification to indicate the presence of an |
| 1094 | extra field before the compressed data. |
| 1095 | |
| 1096 | CINFO (Compression info) |
| 1097 | For CM = 8, CINFO is the base-2 logarithm of the LZ77 window |
| 1098 | size, minus eight (CINFO=7 indicates a 32K window size). Values |
| 1099 | of CINFO above 7 are not allowed in this version of the |
| 1100 | specification. CINFO is not defined in this specification for |
| 1101 | CM not equal to 8. |
| 1102 | |
| 1103 | FLG (FLaGs) |
| 1104 | This flag byte is divided as follows: |
| 1105 | |
| 1106 | bits 0 to 4 FCHECK (check bits for CMF and FLG) |
| 1107 | bit 5 FDICT (preset dictionary) |
| 1108 | bits 6 to 7 FLEVEL (compression level) |
| 1109 | |
| 1110 | The FCHECK value must be such that CMF and FLG, when viewed as |
| 1111 | a 16-bit unsigned integer stored in MSB order (CMF*256 + FLG), |
| 1112 | is a multiple of 31. |
| 1113 | |
| 1114 | |
| 1115 | FDICT (Preset dictionary) |
| 1116 | If FDICT is set, a DICT dictionary identifier is present |
| 1117 | immediately after the FLG byte. The dictionary is a sequence of |
| 1118 | bytes which are initially fed to the compressor without |
| 1119 | producing any compressed output. DICT is the Adler-32 checksum |
| 1120 | of this sequence of bytes (see the definition of ADLER32 |
| 1121 | below). The decompressor can use this identifier to determine |
| 1122 | which dictionary has been used by the compressor. |
| 1123 | |
| 1124 | FLEVEL (Compression level) |
| 1125 | These flags are available for use by specific compression |
| 1126 | methods. The "deflate" method (CM = 8) sets these flags as |
| 1127 | follows: |
| 1128 | |
| 1129 | 0 - compressor used fastest algorithm |
| 1130 | 1 - compressor used fast algorithm |
| 1131 | 2 - compressor used default algorithm |
| 1132 | 3 - compressor used maximum compression, slowest algorithm |
| 1133 | |
| 1134 | The information in FLEVEL is not needed for decompression; it |
| 1135 | is there to indicate if recompression might be worthwhile. |
| 1136 | |
| 1137 | compressed data |
| 1138 | For compression method 8, the compressed data is stored in the |
| 1139 | deflate compressed data format as described in the document |
| 1140 | "DEFLATE Compressed Data Format Specification" by L. Peter |
| 1141 | Deutsch. (See reference [3] in Chapter 3, below) |
| 1142 | |
| 1143 | Other compressed data formats are not specified in this version |
| 1144 | of the zlib specification. |
| 1145 | |
| 1146 | ADLER32 (Adler-32 checksum) |
| 1147 | This contains a checksum value of the uncompressed data |
| 1148 | (excluding any dictionary data) computed according to Adler-32 |
| 1149 | algorithm. This algorithm is a 32-bit extension and improvement |
| 1150 | of the Fletcher algorithm, used in the ITU-T X.224 / ISO 8073 |
| 1151 | standard. See references [4] and [5] in Chapter 3, below) |
| 1152 | |
| 1153 | Adler-32 is composed of two sums accumulated per byte: s1 is |
| 1154 | the sum of all bytes, s2 is the sum of all s1 values. Both sums |
| 1155 | are done modulo 65521. s1 is initialized to 1, s2 to zero. The |
| 1156 | Adler-32 checksum is stored as s2*65536 + s1 in most- |
| 1157 | significant-byte first (network) order. |
| 1158 | |
| 1159 | ==> The stream can start with only 2 bytes : |
| 1160 | - CM = 0x78 : CMINFO=7 (32kB window), CM=8 (deflate) |
| 1161 | - FLG = 0x01 : FLEVEL = 0 (fastest), FDICT=0 (no dict), FCHECK=1 so |
| 1162 | that 0x7801 is a multiple of 31 (30721 = 991 * 31). |
| 1163 | |
| 1164 | ==> and it ends with only 4 bytes, the Adler-32 checksum in big-endian format. |
| 1165 | |
| 1166 | */ |
| 1167 | |
| 1168 | static const unsigned char zlib_hdr[] = { 0x78, 0x01 }; // 32k win, deflate, chk=1 |
| 1169 | |
| 1170 | |
| 1171 | /* Original version from RFC1950, verified and works OK */ |
| 1172 | uint32_t slz_adler32_by1(uint32_t crc, const unsigned char *buf, int len) |
| 1173 | { |
| 1174 | uint32_t s1 = crc & 0xffff; |
| 1175 | uint32_t s2 = (crc >> 16) & 0xffff; |
| 1176 | int n; |
| 1177 | |
| 1178 | for (n = 0; n < len; n++) { |
| 1179 | s1 = (s1 + buf[n]) % 65521; |
| 1180 | s2 = (s2 + s1) % 65521; |
| 1181 | } |
| 1182 | return (s2 << 16) + s1; |
| 1183 | } |
| 1184 | |
| 1185 | /* Computes the adler32 sum on <buf> for <len> bytes. It avoids the expensive |
| 1186 | * modulus by retrofitting the number of bytes missed between 65521 and 65536 |
| 1187 | * which is easy to count : For every sum above 65536, the modulus is offset |
| 1188 | * by (65536-65521) = 15. So for any value, we can count the accumulated extra |
| 1189 | * values by dividing the sum by 65536 and multiplying this value by |
| 1190 | * (65536-65521). That's easier with a drawing with boxes and marbles. It gives |
| 1191 | * this : |
| 1192 | * x % 65521 = (x % 65536) + (x / 65536) * (65536 - 65521) |
| 1193 | * = (x & 0xffff) + (x >> 16) * 15. |
| 1194 | */ |
| 1195 | uint32_t slz_adler32_block(uint32_t crc, const unsigned char *buf, long len) |
| 1196 | { |
| 1197 | long s1 = crc & 0xffff; |
| 1198 | long s2 = (crc >> 16); |
| 1199 | long blk; |
| 1200 | long n; |
| 1201 | |
| 1202 | do { |
| 1203 | blk = len; |
| 1204 | /* ensure we never overflow s2 (limit is about 2^((32-8)/2) */ |
| 1205 | if (blk > (1U << 12)) |
| 1206 | blk = 1U << 12; |
| 1207 | len -= blk; |
| 1208 | |
| 1209 | for (n = 0; n < blk; n++) { |
| 1210 | s1 = (s1 + buf[n]); |
| 1211 | s2 = (s2 + s1); |
| 1212 | } |
| 1213 | |
| 1214 | /* Largest value here is 2^12 * 255 = 1044480 < 2^20. We can |
| 1215 | * still overflow once, but not twice because the right hand |
| 1216 | * size is 225 max, so the total is 65761. However we also |
| 1217 | * have to take care of the values between 65521 and 65536. |
| 1218 | */ |
| 1219 | s1 = (s1 & 0xffff) + 15 * (s1 >> 16); |
| 1220 | if (s1 >= 65521) |
| 1221 | s1 -= 65521; |
| 1222 | |
| 1223 | /* For s2, the largest value is estimated to 2^32-1 for |
| 1224 | * simplicity, so the right hand side is about 15*65535 |
| 1225 | * = 983025. We can overflow twice at most. |
| 1226 | */ |
| 1227 | s2 = (s2 & 0xffff) + 15 * (s2 >> 16); |
| 1228 | s2 = (s2 & 0xffff) + 15 * (s2 >> 16); |
| 1229 | if (s2 >= 65521) |
| 1230 | s2 -= 65521; |
| 1231 | |
| 1232 | buf += blk; |
| 1233 | } while (len); |
| 1234 | return (s2 << 16) + s1; |
| 1235 | } |
| 1236 | |
| 1237 | /* Sends the zlib header for stream <strm> into buffer <buf>. When it's done, |
| 1238 | * the stream state is updated to SLZ_ST_EOB. It returns the number of bytes |
| 1239 | * emitted which is always 2. The caller is responsible for ensuring there's |
| 1240 | * always enough room in the buffer. |
| 1241 | */ |
| 1242 | int slz_rfc1950_send_header(struct slz_stream *strm, unsigned char *buf) |
| 1243 | { |
| 1244 | memcpy(buf, zlib_hdr, sizeof(zlib_hdr)); |
| 1245 | strm->state = SLZ_ST_EOB; |
| 1246 | return sizeof(zlib_hdr); |
| 1247 | } |
| 1248 | |
| 1249 | /* Encodes the block according to rfc1950. This means that the CRC of the input |
| 1250 | * block is computed according to the ADLER32 algorithm. If the header was never |
| 1251 | * sent, it may be sent first. The number of output bytes is returned. |
| 1252 | */ |
| 1253 | long slz_rfc1950_encode(struct slz_stream *strm, unsigned char *out, const unsigned char *in, long ilen, int more) |
| 1254 | { |
| 1255 | long ret = 0; |
| 1256 | |
| 1257 | if (__builtin_expect(strm->state == SLZ_ST_INIT, 0)) |
| 1258 | ret += slz_rfc1950_send_header(strm, out); |
| 1259 | |
| 1260 | strm->crc32 = slz_adler32_block(strm->crc32, in, ilen); |
| 1261 | ret += slz_rfc1951_encode(strm, out + ret, in, ilen, more); |
| 1262 | return ret; |
| 1263 | } |
| 1264 | |
| 1265 | /* Initializes stream <strm> for use with the zlib format (rfc1952). The |
| 1266 | * compression level passed in <level> is set. This value can only be 0 (no |
| 1267 | * compression) or 1 (compression) and other values will lead to unpredictable |
| 1268 | * behaviour. The function always returns 0. |
| 1269 | */ |
| 1270 | int slz_rfc1950_init(struct slz_stream *strm, int level) |
| 1271 | { |
| 1272 | strm->state = SLZ_ST_INIT; |
| 1273 | strm->level = level; |
| 1274 | strm->format = SLZ_FMT_ZLIB; |
| 1275 | strm->crc32 = 1; // rfc1950/zlib starts with initial crc=1 |
| 1276 | strm->ilen = 0; |
| 1277 | strm->qbits = 0; |
| 1278 | strm->queue = 0; |
| 1279 | return 0; |
| 1280 | } |
| 1281 | |
| 1282 | /* Flushes pending bits and sends the gzip trailer for stream <strm> into |
| 1283 | * buffer <buf>. When it's done, the stream state is updated to SLZ_ST_END. It |
| 1284 | * returns the number of bytes emitted. The trailer consists in flushing the |
| 1285 | * possibly pending bits from the queue (up to 24 bits), rounding to the next |
Ilya Shipitsin | b2be9a1 | 2021-04-24 13:25:42 +0500 | [diff] [blame] | 1286 | * byte, then 4 bytes for the CRC. That may about to 4+4 = 8 bytes, that the |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 1287 | * caller must ensure are available before calling the function. Note that if |
| 1288 | * the initial header was never sent, it will be sent first as well (2 extra |
| 1289 | * bytes). |
| 1290 | */ |
| 1291 | int slz_rfc1950_finish(struct slz_stream *strm, unsigned char *buf) |
| 1292 | { |
| 1293 | strm->outbuf = buf; |
| 1294 | |
| 1295 | if (__builtin_expect(strm->state == SLZ_ST_INIT, 0)) |
| 1296 | strm->outbuf += slz_rfc1952_send_header(strm, strm->outbuf); |
| 1297 | |
| 1298 | slz_rfc1951_finish(strm, strm->outbuf); |
| 1299 | copy_8b(strm, (strm->crc32 >> 24) & 0xff); |
| 1300 | copy_8b(strm, (strm->crc32 >> 16) & 0xff); |
| 1301 | copy_8b(strm, (strm->crc32 >> 8) & 0xff); |
| 1302 | copy_8b(strm, (strm->crc32 >> 0) & 0xff); |
| 1303 | strm->state = SLZ_ST_END; |
| 1304 | return strm->outbuf - buf; |
| 1305 | } |
| 1306 | |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 1307 | __attribute__((constructor)) |
| 1308 | static void __slz_initialize(void) |
| 1309 | { |
Willy Tarreau | 9e27428 | 2021-05-12 08:36:09 +0200 | [diff] [blame] | 1310 | #if !defined(__ARM_FEATURE_CRC32) |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 1311 | __slz_make_crc_table(); |
Willy Tarreau | 9e27428 | 2021-05-12 08:36:09 +0200 | [diff] [blame] | 1312 | #endif |
Willy Tarreau | ab2b782 | 2021-04-22 14:09:44 +0200 | [diff] [blame] | 1313 | __slz_prepare_dist_table(); |
| 1314 | } |