[BTRFS][UBTRFS][SHELLBTRFS] Upgrade to 1.7.4 (#4417)

v1.7.4 (2020-08-23):

- Fixed issue when running compressed EXEs
- Changed build system to cmake
- Upgraded zstd to version 1.4.5
- Added support for FSCTL_GET_RETRIEVAL_POINTERS
- Miscellaneous bug fixes
This commit is contained in:
Vincent Franchomme 2022-04-28 21:33:48 +02:00 committed by Hermès BÉLUSCA - MAÏTO
parent 174dfab667
commit 0604273535
80 changed files with 11313 additions and 6442 deletions

View file

@ -1,10 +1,11 @@
/*
* Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
* Copyright (c) 2016-2020, 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.
*/
#include "zstd_ldm.h"
@ -37,8 +38,8 @@ void ZSTD_ldm_adjustParameters(ldmParams_t* params,
params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
assert(params->hashLog <= ZSTD_HASHLOG_MAX);
}
if (params->hashEveryLog == 0) {
params->hashEveryLog = params->windowLog < params->hashLog
if (params->hashRateLog == 0) {
params->hashRateLog = params->windowLog < params->hashLog
? 0
: params->windowLog - params->hashLog;
}
@ -49,9 +50,9 @@ size_t ZSTD_ldm_getTableSize(ldmParams_t params)
{
size_t const ldmHSize = ((size_t)1) << params.hashLog;
size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
size_t const ldmBucketSize =
((size_t)1) << (params.hashLog - ldmBucketSizeLog);
size_t const totalSize = ldmBucketSize + ldmHSize * sizeof(ldmEntry_t);
size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+ ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
return params.enableLdm ? totalSize : 0;
}
@ -119,20 +120,20 @@ static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
*
* Gets the small hash, checksum, and tag from the rollingHash.
*
* If the tag matches (1 << ldmParams.hashEveryLog)-1, then
* If the tag matches (1 << ldmParams.hashRateLog)-1, then
* creates an ldmEntry from the offset, and inserts it into the hash table.
*
* hBits is the length of the small hash, which is the most significant hBits
* of rollingHash. The checksum is the next 32 most significant bits, followed
* by ldmParams.hashEveryLog bits that make up the tag. */
* by ldmParams.hashRateLog bits that make up the tag. */
static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
U64 const rollingHash,
U32 const hBits,
U32 const offset,
ldmParams_t const ldmParams)
{
U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog);
U32 const tagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashRateLog);
U32 const tagMask = ((U32)1 << ldmParams.hashRateLog) - 1;
if (tag == tagMask) {
U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
@ -143,56 +144,6 @@ static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
}
}
/** ZSTD_ldm_getRollingHash() :
* Get a 64-bit hash using the first len bytes from buf.
*
* Giving bytes s = s_1, s_2, ... s_k, the hash is defined to be
* H(s) = s_1*(a^(k-1)) + s_2*(a^(k-2)) + ... + s_k*(a^0)
*
* where the constant a is defined to be prime8bytes.
*
* The implementation adds an offset to each byte, so
* H(s) = (s_1 + HASH_CHAR_OFFSET)*(a^(k-1)) + ... */
static U64 ZSTD_ldm_getRollingHash(const BYTE* buf, U32 len)
{
U64 ret = 0;
U32 i;
for (i = 0; i < len; i++) {
ret *= prime8bytes;
ret += buf[i] + LDM_HASH_CHAR_OFFSET;
}
return ret;
}
/** ZSTD_ldm_ipow() :
* Return base^exp. */
static U64 ZSTD_ldm_ipow(U64 base, U64 exp)
{
U64 ret = 1;
while (exp) {
if (exp & 1) { ret *= base; }
exp >>= 1;
base *= base;
}
return ret;
}
U64 ZSTD_ldm_getHashPower(U32 minMatchLength) {
DEBUGLOG(4, "ZSTD_ldm_getHashPower: mml=%u", minMatchLength);
assert(minMatchLength >= ZSTD_LDM_MINMATCH_MIN);
return ZSTD_ldm_ipow(prime8bytes, minMatchLength - 1);
}
/** ZSTD_ldm_updateHash() :
* Updates hash by removing toRemove and adding toAdd. */
static U64 ZSTD_ldm_updateHash(U64 hash, BYTE toRemove, BYTE toAdd, U64 hashPower)
{
hash -= ((toRemove + LDM_HASH_CHAR_OFFSET) * hashPower);
hash *= prime8bytes;
hash += toAdd + LDM_HASH_CHAR_OFFSET;
return hash;
}
/** ZSTD_ldm_countBackwardsMatch() :
* Returns the number of bytes that match backwards before pIn and pMatch.
*
@ -238,6 +189,7 @@ static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
case ZSTD_btlazy2:
case ZSTD_btopt:
case ZSTD_btultra:
case ZSTD_btultra2:
break;
default:
assert(0); /* not possible : not a valid strategy id */
@ -261,9 +213,9 @@ static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
const BYTE* cur = lastHashed + 1;
while (cur < iend) {
rollingHash = ZSTD_ldm_updateHash(rollingHash, cur[-1],
cur[ldmParams.minMatchLength-1],
state->hashPower);
rollingHash = ZSTD_rollingHash_rotate(rollingHash, cur[-1],
cur[ldmParams.minMatchLength-1],
state->hashPower);
ZSTD_ldm_makeEntryAndInsertByTag(state,
rollingHash, hBits,
(U32)(cur - base), ldmParams);
@ -272,6 +224,20 @@ static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
return rollingHash;
}
void ZSTD_ldm_fillHashTable(
ldmState_t* state, const BYTE* ip,
const BYTE* iend, ldmParams_t const* params)
{
DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
if ((size_t)(iend - ip) >= params->minMatchLength) {
U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
ZSTD_ldm_fillLdmHashTable(
state, startingHash, ip, iend - params->minMatchLength, state->window.base,
params->hashLog - params->bucketSizeLog,
*params);
}
}
/** ZSTD_ldm_limitTableUpdate() :
*
@ -297,8 +263,8 @@ static size_t ZSTD_ldm_generateSequences_internal(
U64 const hashPower = ldmState->hashPower;
U32 const hBits = params->hashLog - params->bucketSizeLog;
U32 const ldmBucketSize = 1U << params->bucketSizeLog;
U32 const hashEveryLog = params->hashEveryLog;
U32 const ldmTagMask = (1U << params->hashEveryLog) - 1;
U32 const hashRateLog = params->hashRateLog;
U32 const ldmTagMask = (1U << params->hashRateLog) - 1;
/* Prefix and extDict parameters */
U32 const dictLimit = ldmState->window.dictLimit;
U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
@ -324,16 +290,16 @@ static size_t ZSTD_ldm_generateSequences_internal(
size_t forwardMatchLength = 0, backwardMatchLength = 0;
ldmEntry_t* bestEntry = NULL;
if (ip != istart) {
rollingHash = ZSTD_ldm_updateHash(rollingHash, lastHashed[0],
lastHashed[minMatchLength],
hashPower);
rollingHash = ZSTD_rollingHash_rotate(rollingHash, lastHashed[0],
lastHashed[minMatchLength],
hashPower);
} else {
rollingHash = ZSTD_ldm_getRollingHash(ip, minMatchLength);
rollingHash = ZSTD_rollingHash_compute(ip, minMatchLength);
}
lastHashed = ip;
/* Do not insert and do not look for a match */
if (ZSTD_ldm_getTag(rollingHash, hBits, hashEveryLog) != ldmTagMask) {
if (ZSTD_ldm_getTag(rollingHash, hBits, hashRateLog) != ldmTagMask) {
ip++;
continue;
}
@ -478,7 +444,7 @@ size_t ZSTD_ldm_generateSequences(
*/
assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
/* The input could be very large (in zstdmt), so it must be broken up into
* chunks to enforce the maximmum distance and handle overflow correction.
* chunks to enforce the maximum distance and handle overflow correction.
*/
assert(sequences->pos <= sequences->size);
assert(sequences->size <= sequences->capacity);
@ -496,8 +462,10 @@ size_t ZSTD_ldm_generateSequences(
if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
U32 const ldmHSize = 1U << params->hashLog;
U32 const correction = ZSTD_window_correctOverflow(
&ldmState->window, /* cycleLog */ 0, maxDist, src);
&ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
/* invalidate dictionaries on overflow correction */
ldmState->loadedDictEnd = 0;
}
/* 2. We enforce the maximum offset allowed.
*
@ -506,8 +474,14 @@ size_t ZSTD_ldm_generateSequences(
* TODO: * Test the chunk size.
* * Try invalidation after the sequence generation and test the
* the offset against maxDist directly.
*
* NOTE: Because of dictionaries + sequence splitting we MUST make sure
* that any offset used is valid at the END of the sequence, since it may
* be split into two sequences. This condition holds when using
* ZSTD_window_enforceMaxDist(), but if we move to checking offsets
* against maxDist directly, we'll have to carefully handle that case.
*/
ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL, NULL);
ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
/* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
newLeftoverSize = ZSTD_ldm_generateSequences_internal(
ldmState, sequences, params, chunkStart, chunkSize);
@ -593,7 +567,7 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
void const* src, size_t srcSize)
{
const ZSTD_compressionParameters* const cParams = &ms->cParams;
unsigned const minMatch = cParams->searchLength;
unsigned const minMatch = cParams->minMatch;
ZSTD_blockCompressor const blockCompressor =
ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
/* Input bounds */
@ -615,14 +589,13 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
if (sequence.offset == 0)
break;
assert(sequence.offset <= (1U << cParams->windowLog));
assert(ip + sequence.litLength + sequence.matchLength <= iend);
/* Fill tables for block compressor */
ZSTD_ldm_limitTableUpdate(ms, ip);
ZSTD_ldm_fillFastTables(ms, ip);
/* Run the block compressor */
DEBUGLOG(5, "calling block compressor on segment of size %u", sequence.litLength);
DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
{
size_t const newLitLength =
blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
@ -632,7 +605,7 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
rep[i] = rep[i-1];
rep[0] = sequence.offset;
/* Store the sequence */
ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength,
ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
sequence.offset + ZSTD_REP_MOVE,
sequence.matchLength - MINMATCH);
ip += sequence.matchLength;