| 1 | /* |
| 2 | * Copyright (c) Meta Platforms, Inc. and affiliates. |
| 3 | * All rights reserved. |
| 4 | * |
| 5 | * This source code is licensed under both the BSD-style license (found in the |
| 6 | * LICENSE file in the root directory of this source tree) and the GPLv2 (found |
| 7 | * in the COPYING file in the root directory of this source tree). |
| 8 | */ |
| 9 | |
| 10 | #include <stdlib.h> /* malloc, free */ |
| 11 | #include <limits.h> /* UINT_MAX */ |
| 12 | #include <assert.h> |
| 13 | |
| 14 | #define XXH_STATIC_LINKING_ONLY |
| 15 | #include "xxhash.h" |
| 16 | |
| 17 | #define ZSTD_STATIC_LINKING_ONLY |
| 18 | #include "zstd.h" |
| 19 | #include "zstd_errors.h" |
| 20 | #include "mem.h" |
| 21 | |
| 22 | #include "zstd_seekable.h" |
| 23 | |
| 24 | #define CHECK_Z(f) { size_t const ret = (f); if (ret != 0) return ret; } |
| 25 | |
| 26 | #undef ERROR |
| 27 | #define ERROR(name) ((size_t)-ZSTD_error_##name) |
| 28 | |
| 29 | #undef MIN |
| 30 | #undef MAX |
| 31 | #define MIN(a, b) ((a) < (b) ? (a) : (b)) |
| 32 | #define MAX(a, b) ((a) > (b) ? (a) : (b)) |
| 33 | |
| 34 | typedef struct { |
| 35 | U32 cSize; |
| 36 | U32 dSize; |
| 37 | U32 checksum; |
| 38 | } framelogEntry_t; |
| 39 | |
| 40 | struct ZSTD_frameLog_s { |
| 41 | framelogEntry_t* entries; |
| 42 | U32 size; |
| 43 | U32 capacity; |
| 44 | |
| 45 | int checksumFlag; |
| 46 | |
| 47 | /* for use when streaming out the seek table */ |
| 48 | U32 seekTablePos; |
| 49 | U32 seekTableIndex; |
| 50 | } framelog_t; |
| 51 | |
| 52 | struct ZSTD_seekable_CStream_s { |
| 53 | ZSTD_CStream* cstream; |
| 54 | ZSTD_frameLog framelog; |
| 55 | |
| 56 | U32 frameCSize; |
| 57 | U32 frameDSize; |
| 58 | |
| 59 | XXH64_state_t xxhState; |
| 60 | |
| 61 | U32 maxFrameSize; |
| 62 | |
| 63 | int writingSeekTable; |
| 64 | }; |
| 65 | |
| 66 | static size_t ZSTD_seekable_frameLog_allocVec(ZSTD_frameLog* fl) |
| 67 | { |
| 68 | /* allocate some initial space */ |
| 69 | size_t const FRAMELOG_STARTING_CAPACITY = 16; |
| 70 | fl->entries = (framelogEntry_t*)malloc( |
| 71 | sizeof(framelogEntry_t) * FRAMELOG_STARTING_CAPACITY); |
| 72 | if (fl->entries == NULL) return ERROR(memory_allocation); |
| 73 | fl->capacity = (U32)FRAMELOG_STARTING_CAPACITY; |
| 74 | return 0; |
| 75 | } |
| 76 | |
| 77 | static size_t ZSTD_seekable_frameLog_freeVec(ZSTD_frameLog* fl) |
| 78 | { |
| 79 | if (fl != NULL) free(fl->entries); |
| 80 | return 0; |
| 81 | } |
| 82 | |
| 83 | ZSTD_frameLog* ZSTD_seekable_createFrameLog(int checksumFlag) |
| 84 | { |
| 85 | ZSTD_frameLog* const fl = (ZSTD_frameLog*)malloc(sizeof(ZSTD_frameLog)); |
| 86 | if (fl == NULL) return NULL; |
| 87 | |
| 88 | if (ZSTD_isError(ZSTD_seekable_frameLog_allocVec(fl))) { |
| 89 | free(fl); |
| 90 | return NULL; |
| 91 | } |
| 92 | |
| 93 | fl->checksumFlag = checksumFlag; |
| 94 | fl->seekTablePos = 0; |
| 95 | fl->seekTableIndex = 0; |
| 96 | fl->size = 0; |
| 97 | |
| 98 | return fl; |
| 99 | } |
| 100 | |
| 101 | size_t ZSTD_seekable_freeFrameLog(ZSTD_frameLog* fl) |
| 102 | { |
| 103 | ZSTD_seekable_frameLog_freeVec(fl); |
| 104 | free(fl); |
| 105 | return 0; |
| 106 | } |
| 107 | |
| 108 | ZSTD_seekable_CStream* ZSTD_seekable_createCStream(void) |
| 109 | { |
| 110 | ZSTD_seekable_CStream* const zcs = (ZSTD_seekable_CStream*)malloc(sizeof(ZSTD_seekable_CStream)); |
| 111 | if (zcs == NULL) return NULL; |
| 112 | |
| 113 | memset(zcs, 0, sizeof(*zcs)); |
| 114 | |
| 115 | zcs->cstream = ZSTD_createCStream(); |
| 116 | if (zcs->cstream == NULL) goto failed1; |
| 117 | |
| 118 | if (ZSTD_isError(ZSTD_seekable_frameLog_allocVec(&zcs->framelog))) goto failed2; |
| 119 | |
| 120 | return zcs; |
| 121 | |
| 122 | failed2: |
| 123 | ZSTD_freeCStream(zcs->cstream); |
| 124 | failed1: |
| 125 | free(zcs); |
| 126 | return NULL; |
| 127 | } |
| 128 | |
| 129 | size_t ZSTD_seekable_freeCStream(ZSTD_seekable_CStream* zcs) |
| 130 | { |
| 131 | if (zcs == NULL) return 0; /* support free on null */ |
| 132 | ZSTD_freeCStream(zcs->cstream); |
| 133 | ZSTD_seekable_frameLog_freeVec(&zcs->framelog); |
| 134 | free(zcs); |
| 135 | return 0; |
| 136 | } |
| 137 | |
| 138 | size_t ZSTD_seekable_initCStream(ZSTD_seekable_CStream* zcs, |
| 139 | int compressionLevel, |
| 140 | int checksumFlag, |
| 141 | unsigned maxFrameSize) |
| 142 | { |
| 143 | zcs->framelog.size = 0; |
| 144 | zcs->frameCSize = 0; |
| 145 | zcs->frameDSize = 0; |
| 146 | |
| 147 | /* make sure maxFrameSize has a reasonable value */ |
| 148 | if (maxFrameSize > ZSTD_SEEKABLE_MAX_FRAME_DECOMPRESSED_SIZE) { |
| 149 | return ERROR(frameParameter_unsupported); |
| 150 | } |
| 151 | |
| 152 | zcs->maxFrameSize = maxFrameSize ? |
| 153 | maxFrameSize : ZSTD_SEEKABLE_MAX_FRAME_DECOMPRESSED_SIZE; |
| 154 | |
| 155 | zcs->framelog.checksumFlag = checksumFlag; |
| 156 | if (zcs->framelog.checksumFlag) { |
| 157 | XXH64_reset(&zcs->xxhState, 0); |
| 158 | } |
| 159 | |
| 160 | zcs->framelog.seekTablePos = 0; |
| 161 | zcs->framelog.seekTableIndex = 0; |
| 162 | zcs->writingSeekTable = 0; |
| 163 | |
| 164 | return ZSTD_initCStream(zcs->cstream, compressionLevel); |
| 165 | } |
| 166 | |
| 167 | size_t ZSTD_seekable_logFrame(ZSTD_frameLog* fl, |
| 168 | unsigned compressedSize, |
| 169 | unsigned decompressedSize, |
| 170 | unsigned checksum) |
| 171 | { |
| 172 | if (fl->size == ZSTD_SEEKABLE_MAXFRAMES) |
| 173 | return ERROR(frameIndex_tooLarge); |
| 174 | |
| 175 | /* grow the buffer if required */ |
| 176 | if (fl->size == fl->capacity) { |
| 177 | /* exponential size increase for constant amortized runtime */ |
| 178 | size_t const newCapacity = fl->capacity * 2; |
| 179 | framelogEntry_t* const newEntries = (framelogEntry_t*)realloc(fl->entries, |
| 180 | sizeof(framelogEntry_t) * newCapacity); |
| 181 | |
| 182 | if (newEntries == NULL) return ERROR(memory_allocation); |
| 183 | |
| 184 | fl->entries = newEntries; |
| 185 | assert(newCapacity <= UINT_MAX); |
| 186 | fl->capacity = (U32)newCapacity; |
| 187 | } |
| 188 | |
| 189 | fl->entries[fl->size] = (framelogEntry_t){ |
| 190 | compressedSize, decompressedSize, checksum |
| 191 | }; |
| 192 | fl->size++; |
| 193 | |
| 194 | return 0; |
| 195 | } |
| 196 | |
| 197 | size_t ZSTD_seekable_endFrame(ZSTD_seekable_CStream* zcs, ZSTD_outBuffer* output) |
| 198 | { |
| 199 | size_t const prevOutPos = output->pos; |
| 200 | /* end the frame */ |
| 201 | size_t ret = ZSTD_endStream(zcs->cstream, output); |
| 202 | |
| 203 | zcs->frameCSize += (U32)(output->pos - prevOutPos); |
| 204 | |
| 205 | /* need to flush before doing the rest */ |
| 206 | if (ret) return ret; |
| 207 | |
| 208 | /* frame done */ |
| 209 | |
| 210 | /* store the frame data for later */ |
| 211 | ret = ZSTD_seekable_logFrame( |
| 212 | &zcs->framelog, zcs->frameCSize, zcs->frameDSize, |
| 213 | zcs->framelog.checksumFlag |
| 214 | ? XXH64_digest(&zcs->xxhState) & 0xFFFFFFFFU |
| 215 | : 0); |
| 216 | if (ret) return ret; |
| 217 | |
| 218 | /* reset for the next frame */ |
| 219 | zcs->frameCSize = 0; |
| 220 | zcs->frameDSize = 0; |
| 221 | |
| 222 | ZSTD_CCtx_reset(zcs->cstream, ZSTD_reset_session_only); |
| 223 | if (zcs->framelog.checksumFlag) XXH64_reset(&zcs->xxhState, 0); |
| 224 | |
| 225 | return 0; |
| 226 | } |
| 227 | |
| 228 | size_t ZSTD_seekable_compressStream(ZSTD_seekable_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input) |
| 229 | { |
| 230 | const BYTE* const inBase = (const BYTE*) input->src + input->pos; |
| 231 | size_t inLen = input->size - input->pos; |
| 232 | |
| 233 | assert(zcs->maxFrameSize < INT_MAX); |
| 234 | ZSTD_CCtx_setParameter(zcs->cstream, ZSTD_c_srcSizeHint, (int)zcs->maxFrameSize); |
| 235 | inLen = MIN(inLen, (size_t)(zcs->maxFrameSize - zcs->frameDSize)); |
| 236 | |
| 237 | /* if we haven't finished flushing the last frame, don't start writing a new one */ |
| 238 | if (inLen > 0) { |
| 239 | ZSTD_inBuffer inTmp = { inBase, inLen, 0 }; |
| 240 | size_t const prevOutPos = output->pos; |
| 241 | |
| 242 | size_t const ret = ZSTD_compressStream(zcs->cstream, output, &inTmp); |
| 243 | |
| 244 | if (zcs->framelog.checksumFlag) { |
| 245 | XXH64_update(&zcs->xxhState, inBase, inTmp.pos); |
| 246 | } |
| 247 | |
| 248 | zcs->frameCSize += (U32)(output->pos - prevOutPos); |
| 249 | zcs->frameDSize += (U32)inTmp.pos; |
| 250 | |
| 251 | input->pos += inTmp.pos; |
| 252 | |
| 253 | if (ZSTD_isError(ret)) return ret; |
| 254 | } |
| 255 | |
| 256 | if (zcs->maxFrameSize == zcs->frameDSize) { |
| 257 | /* log the frame and start over */ |
| 258 | size_t const ret = ZSTD_seekable_endFrame(zcs, output); |
| 259 | if (ZSTD_isError(ret)) return ret; |
| 260 | |
| 261 | /* get the client ready for the next frame */ |
| 262 | return (size_t)zcs->maxFrameSize; |
| 263 | } |
| 264 | |
| 265 | return (size_t)(zcs->maxFrameSize - zcs->frameDSize); |
| 266 | } |
| 267 | |
| 268 | static inline size_t ZSTD_seekable_seekTableSize(const ZSTD_frameLog* fl) |
| 269 | { |
| 270 | size_t const sizePerFrame = 8 + (fl->checksumFlag?4:0); |
| 271 | size_t const seekTableLen = ZSTD_SKIPPABLEHEADERSIZE + |
| 272 | sizePerFrame * fl->size + |
| 273 | ZSTD_seekTableFooterSize; |
| 274 | |
| 275 | return seekTableLen; |
| 276 | } |
| 277 | |
| 278 | static inline size_t ZSTD_stwrite32(ZSTD_frameLog* fl, |
| 279 | ZSTD_outBuffer* output, U32 const value, |
| 280 | U32 const offset) |
| 281 | { |
| 282 | if (fl->seekTablePos < offset + 4) { |
| 283 | BYTE tmp[4]; /* so that we can work with buffers too small to write a whole word to */ |
| 284 | size_t const lenWrite = |
| 285 | MIN(output->size - output->pos, offset + 4 - fl->seekTablePos); |
| 286 | MEM_writeLE32(tmp, value); |
| 287 | memcpy((BYTE*)output->dst + output->pos, |
| 288 | tmp + (fl->seekTablePos - offset), lenWrite); |
| 289 | output->pos += lenWrite; |
| 290 | fl->seekTablePos += (U32)lenWrite; |
| 291 | |
| 292 | if (lenWrite < 4) return ZSTD_seekable_seekTableSize(fl) - fl->seekTablePos; |
| 293 | } |
| 294 | return 0; |
| 295 | } |
| 296 | |
| 297 | size_t ZSTD_seekable_writeSeekTable(ZSTD_frameLog* fl, ZSTD_outBuffer* output) |
| 298 | { |
| 299 | /* seekTableIndex: the current index in the table and |
| 300 | * seekTableSize: the amount of the table written so far |
| 301 | * |
| 302 | * This function is written this way so that if it has to return early |
| 303 | * because of a small buffer, it can keep going where it left off. |
| 304 | */ |
| 305 | |
| 306 | size_t const sizePerFrame = 8 + (fl->checksumFlag?4:0); |
| 307 | size_t const seekTableLen = ZSTD_seekable_seekTableSize(fl); |
| 308 | |
| 309 | CHECK_Z(ZSTD_stwrite32(fl, output, ZSTD_MAGIC_SKIPPABLE_START | 0xE, 0)); |
| 310 | assert(seekTableLen <= (size_t)UINT_MAX); |
| 311 | CHECK_Z(ZSTD_stwrite32(fl, output, (U32)seekTableLen - ZSTD_SKIPPABLEHEADERSIZE, 4)); |
| 312 | |
| 313 | while (fl->seekTableIndex < fl->size) { |
| 314 | unsigned long long const start = ZSTD_SKIPPABLEHEADERSIZE + sizePerFrame * fl->seekTableIndex; |
| 315 | assert(start + 8 <= UINT_MAX); |
| 316 | CHECK_Z(ZSTD_stwrite32(fl, output, |
| 317 | fl->entries[fl->seekTableIndex].cSize, |
| 318 | (U32)start + 0)); |
| 319 | |
| 320 | CHECK_Z(ZSTD_stwrite32(fl, output, |
| 321 | fl->entries[fl->seekTableIndex].dSize, |
| 322 | (U32)start + 4)); |
| 323 | |
| 324 | if (fl->checksumFlag) { |
| 325 | CHECK_Z(ZSTD_stwrite32( |
| 326 | fl, output, fl->entries[fl->seekTableIndex].checksum, |
| 327 | (U32)start + 8)); |
| 328 | } |
| 329 | |
| 330 | fl->seekTableIndex++; |
| 331 | } |
| 332 | |
| 333 | assert(seekTableLen <= UINT_MAX); |
| 334 | CHECK_Z(ZSTD_stwrite32(fl, output, fl->size, |
| 335 | (U32)seekTableLen - ZSTD_seekTableFooterSize)); |
| 336 | |
| 337 | if (output->size - output->pos < 1) return seekTableLen - fl->seekTablePos; |
| 338 | if (fl->seekTablePos < seekTableLen - 4) { |
| 339 | BYTE const sfd = (BYTE)((fl->checksumFlag) << 7); |
| 340 | |
| 341 | ((BYTE*)output->dst)[output->pos] = sfd; |
| 342 | output->pos++; |
| 343 | fl->seekTablePos++; |
| 344 | } |
| 345 | |
| 346 | CHECK_Z(ZSTD_stwrite32(fl, output, ZSTD_SEEKABLE_MAGICNUMBER, |
| 347 | (U32)seekTableLen - 4)); |
| 348 | |
| 349 | if (fl->seekTablePos != seekTableLen) return ERROR(GENERIC); |
| 350 | return 0; |
| 351 | } |
| 352 | |
| 353 | size_t ZSTD_seekable_endStream(ZSTD_seekable_CStream* zcs, ZSTD_outBuffer* output) |
| 354 | { |
| 355 | if (!zcs->writingSeekTable) { |
| 356 | const size_t endFrame = ZSTD_seekable_endFrame(zcs, output); |
| 357 | if (ZSTD_isError(endFrame)) return endFrame; |
| 358 | /* return an accurate size hint */ |
| 359 | if (endFrame) return endFrame + ZSTD_seekable_seekTableSize(&zcs->framelog); |
| 360 | } |
| 361 | |
| 362 | zcs->writingSeekTable = 1; |
| 363 | |
| 364 | return ZSTD_seekable_writeSeekTable(&zcs->framelog, output); |
| 365 | } |