/* * Copyright (c) 2016-present, Yann Collet, Facebook, Inc. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the * LICENSE file in the root directory of this source tree) and the GPLv2 (found * in the COPYING file in the root directory of this source tree). * You may select, at your option, one of the above-listed licenses. */ /* zstd_decompress_block : * this module takes care of decompressing _compressed_ block */ /*-******************************************************* * Dependencies *********************************************************/ #include /* memcpy, memmove, memset */ #include "compiler.h" /* prefetch */ #include "cpu.h" /* bmi2 */ #include "mem.h" /* low level memory routines */ #define FSE_STATIC_LINKING_ONLY #include "fse.h" #define HUF_STATIC_LINKING_ONLY #include "huf.h" #include "zstd_internal.h" #include "zstd_decompress_internal.h" /* ZSTD_DCtx */ #include "zstd_ddict.h" /* ZSTD_DDictDictContent */ #include "zstd_decompress_block.h" /*_******************************************************* * Macros **********************************************************/ /* These two optional macros force the use one way or another of the two * ZSTD_decompressSequences implementations. You can't force in both directions * at the same time. */ #if defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) #error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!" #endif /*_******************************************************* * Memory operations **********************************************************/ static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); } /*-************************************************************* * Block decoding ***************************************************************/ /*! ZSTD_getcBlockSize() : * Provides the size of compressed block from block header `src` */ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr) { RETURN_ERROR_IF(srcSize < ZSTD_blockHeaderSize, srcSize_wrong); { U32 const cBlockHeader = MEM_readLE24(src); U32 const cSize = cBlockHeader >> 3; bpPtr->lastBlock = cBlockHeader & 1; bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3); bpPtr->origSize = cSize; /* only useful for RLE */ if (bpPtr->blockType == bt_rle) return 1; RETURN_ERROR_IF(bpPtr->blockType == bt_reserved, corruption_detected); return cSize; } } /* Hidden declaration for fullbench */ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx, const void* src, size_t srcSize); /*! ZSTD_decodeLiteralsBlock() : * @return : nb of bytes read from src (< srcSize ) * note : symbol not declared but exposed for fullbench */ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx, const void* src, size_t srcSize) /* note : srcSize < BLOCKSIZE */ { RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected); { const BYTE* const istart = (const BYTE*) src; symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3); switch(litEncType) { case set_repeat: RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted); /* fall-through */ case set_compressed: RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3"); { size_t lhSize, litSize, litCSize; U32 singleStream=0; U32 const lhlCode = (istart[0] >> 2) & 3; U32 const lhc = MEM_readLE32(istart); size_t hufSuccess; switch(lhlCode) { case 0: case 1: default: /* note : default is impossible, since lhlCode into [0..3] */ /* 2 - 2 - 10 - 10 */ singleStream = !lhlCode; lhSize = 3; litSize = (lhc >> 4) & 0x3FF; litCSize = (lhc >> 14) & 0x3FF; break; case 2: /* 2 - 2 - 14 - 14 */ lhSize = 4; litSize = (lhc >> 4) & 0x3FFF; litCSize = lhc >> 18; break; case 3: /* 2 - 2 - 18 - 18 */ lhSize = 5; litSize = (lhc >> 4) & 0x3FFFF; litCSize = (lhc >> 22) + (istart[4] << 10); break; } RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected); RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected); /* prefetch huffman table if cold */ if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) { PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable)); } if (litEncType==set_repeat) { if (singleStream) { hufSuccess = HUF_decompress1X_usingDTable_bmi2( dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2); } else { hufSuccess = HUF_decompress4X_usingDTable_bmi2( dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2); } } else { if (singleStream) { #if defined(HUF_FORCE_DECOMPRESS_X2) hufSuccess = HUF_decompress1X_DCtx_wksp( dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->workspace, sizeof(dctx->workspace)); #else hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2( dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->workspace, sizeof(dctx->workspace), dctx->bmi2); #endif } else { hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2( dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->workspace, sizeof(dctx->workspace), dctx->bmi2); } } RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected); dctx->litPtr = dctx->litBuffer; dctx->litSize = litSize; dctx->litEntropy = 1; if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable; memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH); return litCSize + lhSize; } case set_basic: { size_t litSize, lhSize; U32 const lhlCode = ((istart[0]) >> 2) & 3; switch(lhlCode) { case 0: case 2: default: /* note : default is impossible, since lhlCode into [0..3] */ lhSize = 1; litSize = istart[0] >> 3; break; case 1: lhSize = 2; litSize = MEM_readLE16(istart) >> 4; break; case 3: lhSize = 3; litSize = MEM_readLE24(istart) >> 4; break; } if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) { /* risk reading beyond src buffer with wildcopy */ RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected); memcpy(dctx->litBuffer, istart+lhSize, litSize); dctx->litPtr = dctx->litBuffer; dctx->litSize = litSize; memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH); return lhSize+litSize; } /* direct reference into compressed stream */ dctx->litPtr = istart+lhSize; dctx->litSize = litSize; return lhSize+litSize; } case set_rle: { U32 const lhlCode = ((istart[0]) >> 2) & 3; size_t litSize, lhSize; switch(lhlCode) { case 0: case 2: default: /* note : default is impossible, since lhlCode into [0..3] */ lhSize = 1; litSize = istart[0] >> 3; break; case 1: lhSize = 2; litSize = MEM_readLE16(istart) >> 4; break; case 3: lhSize = 3; litSize = MEM_readLE24(istart) >> 4; RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4"); break; } RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected); memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH); dctx->litPtr = dctx->litBuffer; dctx->litSize = litSize; return lhSize+1; } default: RETURN_ERROR(corruption_detected, "impossible"); } } } /* Default FSE distribution tables. * These are pre-calculated FSE decoding tables using default distributions as defined in specification : * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#default-distributions * They were generated programmatically with following method : * - start from default distributions, present in /lib/common/zstd_internal.h * - generate tables normally, using ZSTD_buildFSETable() * - printout the content of tables * - pretify output, report below, test with fuzzer to ensure it's correct */ /* Default FSE distribution table for Literal Lengths */ static const ZSTD_seqSymbol LL_defaultDTable[(1<tableLog = 0; DTableH->fastMode = 0; cell->nbBits = 0; cell->nextState = 0; assert(nbAddBits < 255); cell->nbAdditionalBits = (BYTE)nbAddBits; cell->baseValue = baseValue; } /* ZSTD_buildFSETable() : * generate FSE decoding table for one symbol (ll, ml or off) * cannot fail if input is valid => * all inputs are presumed validated at this stage */ void ZSTD_buildFSETable(ZSTD_seqSymbol* dt, const short* normalizedCounter, unsigned maxSymbolValue, const U32* baseValue, const U32* nbAdditionalBits, unsigned tableLog) { ZSTD_seqSymbol* const tableDecode = dt+1; U16 symbolNext[MaxSeq+1]; U32 const maxSV1 = maxSymbolValue + 1; U32 const tableSize = 1 << tableLog; U32 highThreshold = tableSize-1; /* Sanity Checks */ assert(maxSymbolValue <= MaxSeq); assert(tableLog <= MaxFSELog); /* Init, lay down lowprob symbols */ { ZSTD_seqSymbol_header DTableH; DTableH.tableLog = tableLog; DTableH.fastMode = 1; { S16 const largeLimit= (S16)(1 << (tableLog-1)); U32 s; for (s=0; s= largeLimit) DTableH.fastMode=0; symbolNext[s] = normalizedCounter[s]; } } } memcpy(dt, &DTableH, sizeof(DTableH)); } /* Spread symbols */ { U32 const tableMask = tableSize-1; U32 const step = FSE_TABLESTEP(tableSize); U32 s, position = 0; for (s=0; s highThreshold) position = (position + step) & tableMask; /* lowprob area */ } } assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */ } /* Build Decoding table */ { U32 u; for (u=0; u max, corruption_detected); { U32 const symbol = *(const BYTE*)src; U32 const baseline = baseValue[symbol]; U32 const nbBits = nbAdditionalBits[symbol]; ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits); } *DTablePtr = DTableSpace; return 1; case set_basic : *DTablePtr = defaultTable; return 0; case set_repeat: RETURN_ERROR_IF(!flagRepeatTable, corruption_detected); /* prefetch FSE table if used */ if (ddictIsCold && (nbSeq > 24 /* heuristic */)) { const void* const pStart = *DTablePtr; size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog)); PREFETCH_AREA(pStart, pSize); } return 0; case set_compressed : { unsigned tableLog; S16 norm[MaxSeq+1]; size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize); RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected); RETURN_ERROR_IF(tableLog > maxLog, corruption_detected); ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog); *DTablePtr = DTableSpace; return headerSize; } default : assert(0); RETURN_ERROR(GENERIC, "impossible"); } } size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr, const void* src, size_t srcSize) { const BYTE* const istart = (const BYTE* const)src; const BYTE* const iend = istart + srcSize; const BYTE* ip = istart; int nbSeq; DEBUGLOG(5, "ZSTD_decodeSeqHeaders"); /* check */ RETURN_ERROR_IF(srcSize < MIN_SEQUENCES_SIZE, srcSize_wrong); /* SeqHead */ nbSeq = *ip++; if (!nbSeq) { *nbSeqPtr=0; RETURN_ERROR_IF(srcSize != 1, srcSize_wrong); return 1; } if (nbSeq > 0x7F) { if (nbSeq == 0xFF) { RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong); nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2; } else { RETURN_ERROR_IF(ip >= iend, srcSize_wrong); nbSeq = ((nbSeq-0x80)<<8) + *ip++; } } *nbSeqPtr = nbSeq; /* FSE table descriptors */ RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong); /* minimum possible size: 1 byte for symbol encoding types */ { symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6); symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3); symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3); ip++; /* Build DTables */ { size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_base, LL_bits, LL_defaultDTable, dctx->fseEntropy, dctx->ddictIsCold, nbSeq); RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected); ip += llhSize; } { size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr, OFtype, MaxOff, OffFSELog, ip, iend-ip, OF_base, OF_bits, OF_defaultDTable, dctx->fseEntropy, dctx->ddictIsCold, nbSeq); RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected); ip += ofhSize; } { size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_base, ML_bits, ML_defaultDTable, dctx->fseEntropy, dctx->ddictIsCold, nbSeq); RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected); ip += mlhSize; } } return ip-istart; } typedef struct { size_t litLength; size_t matchLength; size_t offset; const BYTE* match; } seq_t; typedef struct { size_t state; const ZSTD_seqSymbol* table; } ZSTD_fseState; typedef struct { BIT_DStream_t DStream; ZSTD_fseState stateLL; ZSTD_fseState stateOffb; ZSTD_fseState stateML; size_t prevOffset[ZSTD_REP_NUM]; const BYTE* prefixStart; const BYTE* dictEnd; size_t pos; } seqState_t; /* ZSTD_execSequenceLast7(): * exceptional case : decompress a match starting within last 7 bytes of output buffer. * requires more careful checks, to ensure there is no overflow. * performance does not matter though. * note : this case is supposed to be never generated "naturally" by reference encoder, * since in most cases it needs at least 8 bytes to look for a match. * but it's allowed by the specification. */ FORCE_NOINLINE size_t ZSTD_execSequenceLast7(BYTE* op, BYTE* const oend, seq_t sequence, const BYTE** litPtr, const BYTE* const litLimit, const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd) { BYTE* const oLitEnd = op + sequence.litLength; size_t const sequenceLength = sequence.litLength + sequence.matchLength; BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */ const BYTE* const iLitEnd = *litPtr + sequence.litLength; const BYTE* match = oLitEnd - sequence.offset; /* check */ RETURN_ERROR_IF(oMatchEnd>oend, dstSize_tooSmall, "last match must fit within dstBuffer"); RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "try to read beyond literal buffer"); /* copy literals */ while (op < oLitEnd) *op++ = *(*litPtr)++; /* copy Match */ if (sequence.offset > (size_t)(oLitEnd - base)) { /* offset beyond prefix */ RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - vBase),corruption_detected); match = dictEnd - (base-match); if (match + sequence.matchLength <= dictEnd) { memmove(oLitEnd, match, sequence.matchLength); return sequenceLength; } /* span extDict & currentPrefixSegment */ { size_t const length1 = dictEnd - match; memmove(oLitEnd, match, length1); op = oLitEnd + length1; sequence.matchLength -= length1; match = base; } } while (op < oMatchEnd) *op++ = *match++; return sequenceLength; } HINT_INLINE size_t ZSTD_execSequence(BYTE* op, BYTE* const oend, seq_t sequence, const BYTE** litPtr, const BYTE* const litLimit, const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd) { BYTE* const oLitEnd = op + sequence.litLength; size_t const sequenceLength = sequence.litLength + sequence.matchLength; BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */ BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH; const BYTE* const iLitEnd = *litPtr + sequence.litLength; const BYTE* match = oLitEnd - sequence.offset; /* check */ RETURN_ERROR_IF(oMatchEnd>oend, dstSize_tooSmall, "last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend"); RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "over-read beyond lit buffer"); if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd); /* copy Literals */ if (sequence.litLength > 8) ZSTD_wildcopy_16min(op, (*litPtr), sequence.litLength, ZSTD_no_overlap); /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */ else ZSTD_copy8(op, *litPtr); op = oLitEnd; *litPtr = iLitEnd; /* update for next sequence */ /* copy Match */ if (sequence.offset > (size_t)(oLitEnd - prefixStart)) { /* offset beyond prefix -> go into extDict */ RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected); match = dictEnd + (match - prefixStart); if (match + sequence.matchLength <= dictEnd) { memmove(oLitEnd, match, sequence.matchLength); return sequenceLength; } /* span extDict & currentPrefixSegment */ { size_t const length1 = dictEnd - match; memmove(oLitEnd, match, length1); op = oLitEnd + length1; sequence.matchLength -= length1; match = prefixStart; if (op > oend_w || sequence.matchLength < MINMATCH) { U32 i; for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i]; return sequenceLength; } } } /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */ /* match within prefix */ if (sequence.offset < 8) { /* close range match, overlap */ static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 }; /* added */ static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 }; /* subtracted */ int const sub2 = dec64table[sequence.offset]; op[0] = match[0]; op[1] = match[1]; op[2] = match[2]; op[3] = match[3]; match += dec32table[sequence.offset]; ZSTD_copy4(op+4, match); match -= sub2; } else { ZSTD_copy8(op, match); } op += 8; match += 8; if (oMatchEnd > oend-(16-MINMATCH)) { if (op < oend_w) { ZSTD_wildcopy(op, match, oend_w - op, ZSTD_overlap_src_before_dst); match += oend_w - op; op = oend_w; } while (op < oMatchEnd) *op++ = *match++; } else { ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst); /* works even if matchLength < 8 */ } return sequenceLength; } HINT_INLINE size_t ZSTD_execSequenceLong(BYTE* op, BYTE* const oend, seq_t sequence, const BYTE** litPtr, const BYTE* const litLimit, const BYTE* const prefixStart, const BYTE* const dictStart, const BYTE* const dictEnd) { BYTE* const oLitEnd = op + sequence.litLength; size_t const sequenceLength = sequence.litLength + sequence.matchLength; BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */ BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH; const BYTE* const iLitEnd = *litPtr + sequence.litLength; const BYTE* match = sequence.match; /* check */ RETURN_ERROR_IF(oMatchEnd > oend, dstSize_tooSmall, "last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend"); RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "over-read beyond lit buffer"); if (oLitEnd > oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, dictStart, dictEnd); /* copy Literals */ if (sequence.litLength > 8) ZSTD_wildcopy_16min(op, *litPtr, sequence.litLength, ZSTD_no_overlap); /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */ else ZSTD_copy8(op, *litPtr); /* note : op <= oLitEnd <= oend_w == oend - 8 */ op = oLitEnd; *litPtr = iLitEnd; /* update for next sequence */ /* copy Match */ if (sequence.offset > (size_t)(oLitEnd - prefixStart)) { /* offset beyond prefix */ RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - dictStart), corruption_detected); if (match + sequence.matchLength <= dictEnd) { memmove(oLitEnd, match, sequence.matchLength); return sequenceLength; } /* span extDict & currentPrefixSegment */ { size_t const length1 = dictEnd - match; memmove(oLitEnd, match, length1); op = oLitEnd + length1; sequence.matchLength -= length1; match = prefixStart; if (op > oend_w || sequence.matchLength < MINMATCH) { U32 i; for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i]; return sequenceLength; } } } assert(op <= oend_w); assert(sequence.matchLength >= MINMATCH); /* match within prefix */ if (sequence.offset < 8) { /* close range match, overlap */ static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 }; /* added */ static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 }; /* subtracted */ int const sub2 = dec64table[sequence.offset]; op[0] = match[0]; op[1] = match[1]; op[2] = match[2]; op[3] = match[3]; match += dec32table[sequence.offset]; ZSTD_copy4(op+4, match); match -= sub2; } else { ZSTD_copy8(op, match); } op += 8; match += 8; if (oMatchEnd > oend-(16-MINMATCH)) { if (op < oend_w) { ZSTD_wildcopy(op, match, oend_w - op, ZSTD_overlap_src_before_dst); match += oend_w - op; op = oend_w; } while (op < oMatchEnd) *op++ = *match++; } else { ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst); /* works even if matchLength < 8 */ } return sequenceLength; } static void ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt) { const void* ptr = dt; const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr; DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog); DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits", (U32)DStatePtr->state, DTableH->tableLog); BIT_reloadDStream(bitD); DStatePtr->table = dt + 1; } FORCE_INLINE_TEMPLATE void ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD) { ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state]; U32 const nbBits = DInfo.nbBits; size_t const lowBits = BIT_readBits(bitD, nbBits); DStatePtr->state = DInfo.nextState + lowBits; } /* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1) * bits before reloading. This value is the maximum number of bytes we read * after reloading when we are decoding long offsets. */ #define LONG_OFFSETS_MAX_EXTRA_BITS_32 \ (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32 \ ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32 \ : 0) typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e; #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG FORCE_INLINE_TEMPLATE seq_t ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets) { seq_t seq; U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits; U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits; U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits; U32 const totalBits = llBits+mlBits+ofBits; U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue; U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue; U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue; /* sequence */ { size_t offset; if (!ofBits) offset = 0; else { ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1); ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5); assert(ofBits <= MaxOff); if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) { U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed); offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits); BIT_reloadDStream(&seqState->DStream); if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits); assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32); /* to avoid another reload */ } else { offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */ if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); } } if (ofBits <= 1) { offset += (llBase==0); if (offset) { size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset]; temp += !temp; /* 0 is not valid; input is corrupted; force offset to 1 */ if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1]; seqState->prevOffset[1] = seqState->prevOffset[0]; seqState->prevOffset[0] = offset = temp; } else { /* offset == 0 */ offset = seqState->prevOffset[0]; } } else { seqState->prevOffset[2] = seqState->prevOffset[1]; seqState->prevOffset[1] = seqState->prevOffset[0]; seqState->prevOffset[0] = offset; } seq.offset = offset; } seq.matchLength = mlBase + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/) : 0); /* <= 16 bits */ if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32)) BIT_reloadDStream(&seqState->DStream); if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog))) BIT_reloadDStream(&seqState->DStream); /* Ensure there are enough bits to read the rest of data in 64-bit mode. */ ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64); seq.litLength = llBase + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits/*>0*/) : 0); /* <= 16 bits */ if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u", (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset); /* ANS state update */ ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream); /* <= 9 bits */ ZSTD_updateFseState(&seqState->stateML, &seqState->DStream); /* <= 9 bits */ if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */ ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream); /* <= 8 bits */ return seq; } FORCE_INLINE_TEMPLATE size_t DONT_VECTORIZE ZSTD_decompressSequences_body( ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { const BYTE* ip = (const BYTE*)seqStart; const BYTE* const iend = ip + seqSize; BYTE* const ostart = (BYTE* const)dst; BYTE* const oend = ostart + maxDstSize; BYTE* op = ostart; const BYTE* litPtr = dctx->litPtr; const BYTE* const litEnd = litPtr + dctx->litSize; const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart); const BYTE* const vBase = (const BYTE*) (dctx->virtualStart); const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd); DEBUGLOG(5, "ZSTD_decompressSequences_body"); /* Regen sequences */ if (nbSeq) { seqState_t seqState; dctx->fseEntropy = 1; { U32 i; for (i=0; ientropy.rep[i]; } RETURN_ERROR_IF( ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)), corruption_detected); ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr); ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr); ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr); ZSTD_STATIC_ASSERT( BIT_DStream_unfinished < BIT_DStream_completed && BIT_DStream_endOfBuffer < BIT_DStream_completed && BIT_DStream_completed < BIT_DStream_overflow); for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) { nbSeq--; { seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset); size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd); DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize); if (ZSTD_isError(oneSeqSize)) return oneSeqSize; op += oneSeqSize; } } /* check if reached exact end */ DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq); RETURN_ERROR_IF(nbSeq, corruption_detected); RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected); /* save reps for next block */ { U32 i; for (i=0; ientropy.rep[i] = (U32)(seqState.prevOffset[i]); } } /* last literal segment */ { size_t const lastLLSize = litEnd - litPtr; RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall); memcpy(op, litPtr, lastLLSize); op += lastLLSize; } return op-ostart; } static size_t ZSTD_decompressSequences_default(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */ #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT FORCE_INLINE_TEMPLATE seq_t ZSTD_decodeSequenceLong(seqState_t* seqState, ZSTD_longOffset_e const longOffsets) { seq_t seq; U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits; U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits; U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits; U32 const totalBits = llBits+mlBits+ofBits; U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue; U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue; U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue; /* sequence */ { size_t offset; if (!ofBits) offset = 0; else { ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1); ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5); assert(ofBits <= MaxOff); if (MEM_32bits() && longOffsets) { U32 const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN_32-1); offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits); if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream); if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits); } else { offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */ if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); } } if (ofBits <= 1) { offset += (llBase==0); if (offset) { size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset]; temp += !temp; /* 0 is not valid; input is corrupted; force offset to 1 */ if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1]; seqState->prevOffset[1] = seqState->prevOffset[0]; seqState->prevOffset[0] = offset = temp; } else { offset = seqState->prevOffset[0]; } } else { seqState->prevOffset[2] = seqState->prevOffset[1]; seqState->prevOffset[1] = seqState->prevOffset[0]; seqState->prevOffset[0] = offset; } seq.offset = offset; } seq.matchLength = mlBase + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0); /* <= 16 bits */ if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32)) BIT_reloadDStream(&seqState->DStream); if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog))) BIT_reloadDStream(&seqState->DStream); /* Verify that there is enough bits to read the rest of the data in 64-bit mode. */ ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64); seq.litLength = llBase + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0); /* <= 16 bits */ if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); { size_t const pos = seqState->pos + seq.litLength; const BYTE* const matchBase = (seq.offset > pos) ? seqState->dictEnd : seqState->prefixStart; seq.match = matchBase + pos - seq.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted. * No consequence though : no memory access will occur, overly large offset will be detected in ZSTD_execSequenceLong() */ seqState->pos = pos + seq.matchLength; } /* ANS state update */ ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream); /* <= 9 bits */ ZSTD_updateFseState(&seqState->stateML, &seqState->DStream); /* <= 9 bits */ if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */ ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream); /* <= 8 bits */ return seq; } FORCE_INLINE_TEMPLATE size_t ZSTD_decompressSequencesLong_body( ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { const BYTE* ip = (const BYTE*)seqStart; const BYTE* const iend = ip + seqSize; BYTE* const ostart = (BYTE* const)dst; BYTE* const oend = ostart + maxDstSize; BYTE* op = ostart; const BYTE* litPtr = dctx->litPtr; const BYTE* const litEnd = litPtr + dctx->litSize; const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart); const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart); const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd); /* Regen sequences */ if (nbSeq) { #define STORED_SEQS 4 #define STORED_SEQS_MASK (STORED_SEQS-1) #define ADVANCED_SEQS 4 seq_t sequences[STORED_SEQS]; int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS); seqState_t seqState; int seqNb; dctx->fseEntropy = 1; { int i; for (i=0; ientropy.rep[i]; } seqState.prefixStart = prefixStart; seqState.pos = (size_t)(op-prefixStart); seqState.dictEnd = dictEnd; assert(iend >= ip); RETURN_ERROR_IF( ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)), corruption_detected); ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr); ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr); ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr); /* prepare in advance */ for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNbentropy.rep[i] = (U32)(seqState.prevOffset[i]); } } /* last literal segment */ { size_t const lastLLSize = litEnd - litPtr; RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall); memcpy(op, litPtr, lastLLSize); op += lastLLSize; } return op-ostart; } static size_t ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */ #if DYNAMIC_BMI2 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG static TARGET_ATTRIBUTE("bmi2") size_t DONT_VECTORIZE ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */ #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT static TARGET_ATTRIBUTE("bmi2") size_t ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */ #endif /* DYNAMIC_BMI2 */ typedef size_t (*ZSTD_decompressSequences_t)( ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset); #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG static size_t ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { DEBUGLOG(5, "ZSTD_decompressSequences"); #if DYNAMIC_BMI2 if (dctx->bmi2) { return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */ #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT /* ZSTD_decompressSequencesLong() : * decompression function triggered when a minimum share of offsets is considered "long", * aka out of cache. * note : "long" definition seems overloaded here, sometimes meaning "wider than bitstream register", and sometimes meaning "farther than memory cache distance". * This function will try to mitigate main memory latency through the use of prefetching */ static size_t ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* seqStart, size_t seqSize, int nbSeq, const ZSTD_longOffset_e isLongOffset) { DEBUGLOG(5, "ZSTD_decompressSequencesLong"); #if DYNAMIC_BMI2 if (dctx->bmi2) { return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); } #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */ #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) /* ZSTD_getLongOffsetsShare() : * condition : offTable must be valid * @return : "share" of long offsets (arbitrarily defined as > (1<<23)) * compared to maximum possible of (1< 22) total += 1; } assert(tableLog <= OffFSELog); total <<= (OffFSELog - tableLog); /* scale to OffFSELog */ return total; } #endif size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, const int frame) { /* blockType == blockCompressed */ const BYTE* ip = (const BYTE*)src; /* isLongOffset must be true if there are long offsets. * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN. * We don't expect that to be the case in 64-bit mode. * In block mode, window size is not known, so we have to be conservative. * (note: but it could be evaluated from current-lowLimit) */ ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)))); DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize); RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong); /* Decode literals section */ { size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize); DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize); if (ZSTD_isError(litCSize)) return litCSize; ip += litCSize; srcSize -= litCSize; } /* Build Decoding Tables */ { /* These macros control at build-time which decompressor implementation * we use. If neither is defined, we do some inspection and dispatch at * runtime. */ #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) int usePrefetchDecoder = dctx->ddictIsCold; #endif int nbSeq; size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize); if (ZSTD_isError(seqHSize)) return seqHSize; ip += seqHSize; srcSize -= seqHSize; #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) if ( !usePrefetchDecoder && (!frame || (dctx->fParams.windowSize > (1<<24))) && (nbSeq>ADVANCED_SEQS) ) { /* could probably use a larger nbSeq limit */ U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr); U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */ usePrefetchDecoder = (shareLongOffsets >= minShare); } #endif dctx->ddictIsCold = 0; #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) if (usePrefetchDecoder) #endif #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset); #endif #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG /* else */ return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset); #endif } } size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize) { size_t dSize; ZSTD_checkContinuity(dctx, dst); dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0); dctx->previousDstEnd = (char*)dst + dSize; return dSize; }