Skip to content

Commit

Permalink
Auto merge of #5026 - str4d:bloom-filter-backports, r=<try>
Browse files Browse the repository at this point in the history 
Backport bloom filter improvements

Cherry-picked from the following upstream PRs:

- bitcoin/bitcoin#7113
- bitcoin/bitcoin#7818
  - Only the second commit (to resolve conflicts).
- bitcoin/bitcoin#7934
- bitcoin/bitcoin#8655
  - Partial backport to help resolve conflicts.
- bitcoin/bitcoin#9060
- bitcoin/bitcoin#9223
- bitcoin/bitcoin#9644
  - Partial backport to help resolve conflicts.
- bitcoin/bitcoin#9916
- bitcoin/bitcoin#9750
- bitcoin/bitcoin#13176
- bitcoin/bitcoin#13948
- bitcoin/bitcoin#16073
- bitcoin/bitcoin#18670
- bitcoin/bitcoin#18806
  - Reveals upstream's covert fix for CVE-2013-5700.
- bitcoin/bitcoin#19968
  • Loading branch information
@zkbot
zkbot committed Mar 5, 2021
2 parents 2aa9fbb + bb2463a commit be45961
Show file tree
Hide file tree
Showing 21 changed files with 260 additions and 250 deletions.
9 changes: 9 additions & 0 deletions src/bench/rollingbloom.cpp
View file
Expand Up @@ -39,4 +39,13 @@ static void RollingBloom(benchmark::State& state)
}
}

static void RollingBloomReset(benchmark::State& state)
{
CRollingBloomFilter filter(120000, 0.000001);
while (state.KeepRunning()) {
filter.reset();
}
}

BENCHMARK(RollingBloom);
BENCHMARK(RollingBloomReset); // To-backport: 20000
179 changes: 92 additions & 87 deletions src/bloom.cpp
View file
Expand Up @@ -14,82 +14,65 @@
#include <math.h>
#include <stdlib.h>

#include <algorithm>

#define LN2SQUARED 0.4804530139182014246671025263266649717305529515945455
#define LN2 0.6931471805599453094172321214581765680755001343602552

using namespace std;

CBloomFilter::CBloomFilter(unsigned int nElements, double nFPRate, unsigned int nTweakIn, unsigned char nFlagsIn) :
CBloomFilter::CBloomFilter(const unsigned int nElements, const double nFPRate, const unsigned int nTweakIn, unsigned char nFlagsIn) :
/**
* The ideal size for a bloom filter with a given number of elements and false positive rate is:
* - nElements * log(fp rate) / ln(2)^2
* We ignore filter parameters which will create a bloom filter larger than the protocol limits
*/
vData(min((unsigned int)(-1 / LN2SQUARED * nElements * log(nFPRate)), MAX_BLOOM_FILTER_SIZE * 8) / 8),
vData(std::min((unsigned int)(-1 / LN2SQUARED * nElements * log(nFPRate)), MAX_BLOOM_FILTER_SIZE * 8) / 8),
/**
* The ideal number of hash functions is filter size * ln(2) / number of elements
* Again, we ignore filter parameters which will create a bloom filter with more hash functions than the protocol limits
* See https://en.wikipedia.org/wiki/Bloom_filter for an explanation of these formulas
*/
isFull(false),
isEmpty(false),
nHashFuncs(min((unsigned int)(vData.size() * 8 / nElements * LN2), MAX_HASH_FUNCS)),
nHashFuncs(std::min((unsigned int)(vData.size() * 8 / nElements * LN2), MAX_HASH_FUNCS)),
nTweak(nTweakIn),
nFlags(nFlagsIn)
{
}

// Private constructor used by CRollingBloomFilter
CBloomFilter::CBloomFilter(unsigned int nElements, double nFPRate, unsigned int nTweakIn) :
vData((unsigned int)(-1 / LN2SQUARED * nElements * log(nFPRate)) / 8),
isFull(false),
isEmpty(true),
nHashFuncs((unsigned int)(vData.size() * 8 / nElements * LN2)),
nTweak(nTweakIn),
nFlags(BLOOM_UPDATE_NONE)
{
}

inline unsigned int CBloomFilter::Hash(unsigned int nHashNum, const std::vector<unsigned char>& vDataToHash) const
{
// 0xFBA4C795 chosen as it guarantees a reasonable bit difference between nHashNum values.
return MurmurHash3(nHashNum * 0xFBA4C795 + nTweak, vDataToHash) % (vData.size() * 8);
}

void CBloomFilter::insert(const vector<unsigned char>& vKey)
void CBloomFilter::insert(const std::vector<unsigned char>& vKey)
{
if (isFull)
if (vData.empty()) // Avoid divide-by-zero (CVE-2013-5700)
return;
for (unsigned int i = 0; i < nHashFuncs; i++)
{
unsigned int nIndex = Hash(i, vKey);
// Sets bit nIndex of vData
vData[nIndex >> 3] |= (1 << (7 & nIndex));
}
isEmpty = false;
}

void CBloomFilter::insert(const COutPoint& outpoint)
{
CDataStream stream(SER_NETWORK, PROTOCOL_VERSION);
stream << outpoint;
vector<unsigned char> data(stream.begin(), stream.end());
std::vector<unsigned char> data(stream.begin(), stream.end());
insert(data);
}

void CBloomFilter::insert(const uint256& hash)
{
vector<unsigned char> data(hash.begin(), hash.end());
std::vector<unsigned char> data(hash.begin(), hash.end());
insert(data);
}

bool CBloomFilter::contains(const vector<unsigned char>& vKey) const
bool CBloomFilter::contains(const std::vector<unsigned char>& vKey) const
{
if (isFull)
if (vData.empty()) // Avoid divide-by-zero (CVE-2013-5700)
return true;
if (isEmpty)
return false;
for (unsigned int i = 0; i < nHashFuncs; i++)
{
unsigned int nIndex = Hash(i, vKey);
Expand All @@ -104,29 +87,16 @@ bool CBloomFilter::contains(const COutPoint& outpoint) const
{
CDataStream stream(SER_NETWORK, PROTOCOL_VERSION);
stream << outpoint;
vector<unsigned char> data(stream.begin(), stream.end());
std::vector<unsigned char> data(stream.begin(), stream.end());
return contains(data);
}

bool CBloomFilter::contains(const uint256& hash) const
{
vector<unsigned char> data(hash.begin(), hash.end());
std::vector<unsigned char> data(hash.begin(), hash.end());
return contains(data);
}

void CBloomFilter::clear()
{
vData.assign(vData.size(),0);
isFull = false;
isEmpty = true;
}

void CBloomFilter::reset(unsigned int nNewTweak)
{
clear();
nTweak = nNewTweak;
}

bool CBloomFilter::IsWithinSizeConstraints() const
{
return vData.size() <= MAX_BLOOM_FILTER_SIZE && nHashFuncs <= MAX_HASH_FUNCS;
Expand All @@ -137,10 +107,8 @@ bool CBloomFilter::IsRelevantAndUpdate(const CTransaction& tx)
bool fFound = false;
// Match if the filter contains the hash of tx
// for finding tx when they appear in a block
if (isFull)
if (vData.empty()) // zero-size = "match-all" filter
return true;
if (isEmpty)
return false;
const uint256& hash = tx.GetHash();
if (contains(hash))
fFound = true;
Expand All @@ -153,7 +121,7 @@ bool CBloomFilter::IsRelevantAndUpdate(const CTransaction& tx)
// This means clients don't have to update the filter themselves when a new relevant tx
// is discovered in order to find spending transactions, which avoids round-tripping and race conditions.
CScript::const_iterator pc = txout.scriptPubKey.begin();
vector<unsigned char> data;
std::vector<unsigned char> data;
while (pc < txout.scriptPubKey.end())
{
opcodetype opcode;
Expand All @@ -167,7 +135,7 @@ bool CBloomFilter::IsRelevantAndUpdate(const CTransaction& tx)
else if ((nFlags & BLOOM_UPDATE_MASK) == BLOOM_UPDATE_P2PUBKEY_ONLY)
{
txnouttype type;
vector<vector<unsigned char> > vSolutions;
std::vector<std::vector<unsigned char> > vSolutions;
if (Solver(txout.scriptPubKey, type, vSolutions) &&
(type == TX_PUBKEY || type == TX_MULTISIG))
insert(COutPoint(hash, i));
Expand All @@ -188,7 +156,7 @@ bool CBloomFilter::IsRelevantAndUpdate(const CTransaction& tx)

// Match if the filter contains any arbitrary script data element in any scriptSig in tx
CScript::const_iterator pc = txin.scriptSig.begin();
vector<unsigned char> data;
std::vector<unsigned char> data;
while (pc < txin.scriptSig.end())
{
opcodetype opcode;
Expand All @@ -202,70 +170,107 @@ bool CBloomFilter::IsRelevantAndUpdate(const CTransaction& tx)
return false;
}

void CBloomFilter::UpdateEmptyFull()
CRollingBloomFilter::CRollingBloomFilter(const unsigned int nElements, const double fpRate)
{
bool full = true;
bool empty = true;
for (unsigned int i = 0; i < vData.size(); i++)
{
full &= vData[i] == 0xff;
empty &= vData[i] == 0;
}
isFull = full;
isEmpty = empty;
double logFpRate = log(fpRate);
/* The optimal number of hash functions is log(fpRate) / log(0.5), but
* restrict it to the range 1-50. */
nHashFuncs = std::max(1, std::min((int)round(logFpRate / log(0.5)), 50));
/* In this rolling bloom filter, we'll store between 2 and 3 generations of nElements / 2 entries. */
nEntriesPerGeneration = (nElements + 1) / 2;
uint32_t nMaxElements = nEntriesPerGeneration * 3;
/* The maximum fpRate = pow(1.0 - exp(-nHashFuncs * nMaxElements / nFilterBits), nHashFuncs)
* => pow(fpRate, 1.0 / nHashFuncs) = 1.0 - exp(-nHashFuncs * nMaxElements / nFilterBits)
* => 1.0 - pow(fpRate, 1.0 / nHashFuncs) = exp(-nHashFuncs * nMaxElements / nFilterBits)
* => log(1.0 - pow(fpRate, 1.0 / nHashFuncs)) = -nHashFuncs * nMaxElements / nFilterBits
* => nFilterBits = -nHashFuncs * nMaxElements / log(1.0 - pow(fpRate, 1.0 / nHashFuncs))
* => nFilterBits = -nHashFuncs * nMaxElements / log(1.0 - exp(logFpRate / nHashFuncs))
*/
uint32_t nFilterBits = (uint32_t)ceil(-1.0 * nHashFuncs * nMaxElements / log(1.0 - exp(logFpRate / nHashFuncs)));
data.clear();
/* For each data element we need to store 2 bits. If both bits are 0, the
* bit is treated as unset. If the bits are (01), (10), or (11), the bit is
* treated as set in generation 1, 2, or 3 respectively.
* These bits are stored in separate integers: position P corresponds to bit
* (P & 63) of the integers data[(P >> 6) * 2] and data[(P >> 6) * 2 + 1]. */
data.resize(((nFilterBits + 63) / 64) << 1);
reset();
}

CRollingBloomFilter::CRollingBloomFilter(unsigned int nElements, double fpRate) :
b1(nElements * 2, fpRate, 0), b2(nElements * 2, fpRate, 0)
{
// Implemented using two bloom filters of 2 * nElements each.
// We fill them up, and clear them, staggered, every nElements
// inserted, so at least one always contains the last nElements
// inserted.
nInsertions = 0;
nBloomSize = nElements * 2;
/* Similar to CBloomFilter::Hash */
static inline uint32_t RollingBloomHash(unsigned int nHashNum, uint32_t nTweak, const std::vector<unsigned char>& vDataToHash) {
return MurmurHash3(nHashNum * 0xFBA4C795 + nTweak, vDataToHash);
}

reset();

// A replacement for x % n. This assumes that x and n are 32bit integers, and x is a uniformly random distributed 32bit value
// which should be the case for a good hash.
// See https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/
static inline uint32_t FastMod(uint32_t x, size_t n) {
return ((uint64_t)x * (uint64_t)n) >> 32;
}

void CRollingBloomFilter::insert(const std::vector<unsigned char>& vKey)
{
if (nInsertions == 0) {
b1.clear();
} else if (nInsertions == nBloomSize / 2) {
b2.clear();
if (nEntriesThisGeneration == nEntriesPerGeneration) {
nEntriesThisGeneration = 0;
nGeneration++;
if (nGeneration == 4) {
nGeneration = 1;
}
uint64_t nGenerationMask1 = 0 - (uint64_t)(nGeneration & 1);
uint64_t nGenerationMask2 = 0 - (uint64_t)(nGeneration >> 1);
/* Wipe old entries that used this generation number. */
for (uint32_t p = 0; p < data.size(); p += 2) {
uint64_t p1 = data[p], p2 = data[p + 1];
uint64_t mask = (p1 ^ nGenerationMask1) | (p2 ^ nGenerationMask2);
data[p] = p1 & mask;
data[p + 1] = p2 & mask;
}
}
b1.insert(vKey);
b2.insert(vKey);
if (++nInsertions == nBloomSize) {
nInsertions = 0;
nEntriesThisGeneration++;

for (int n = 0; n < nHashFuncs; n++) {
uint32_t h = RollingBloomHash(n, nTweak, vKey);
int bit = h & 0x3F;
/* FastMod works with the upper bits of h, so it is safe to ignore that the lower bits of h are already used for bit. */
uint32_t pos = FastMod(h, data.size());
/* The lowest bit of pos is ignored, and set to zero for the first bit, and to one for the second. */
data[pos & ~1] = (data[pos & ~1] & ~(((uint64_t)1) << bit)) | ((uint64_t)(nGeneration & 1)) << bit;
data[pos | 1] = (data[pos | 1] & ~(((uint64_t)1) << bit)) | ((uint64_t)(nGeneration >> 1)) << bit;
}
}

void CRollingBloomFilter::insert(const uint256& hash)
{
vector<unsigned char> data(hash.begin(), hash.end());
insert(data);
std::vector<unsigned char> vData(hash.begin(), hash.end());
insert(vData);
}

bool CRollingBloomFilter::contains(const std::vector<unsigned char>& vKey) const
{
if (nInsertions < nBloomSize / 2) {
return b2.contains(vKey);
for (int n = 0; n < nHashFuncs; n++) {
uint32_t h = RollingBloomHash(n, nTweak, vKey);
int bit = h & 0x3F;
uint32_t pos = FastMod(h, data.size());
/* If the relevant bit is not set in either data[pos & ~1] or data[pos | 1], the filter does not contain vKey */
if (!(((data[pos & ~1] | data[pos | 1]) >> bit) & 1)) {
return false;
}
}
return b1.contains(vKey);
return true;
}

bool CRollingBloomFilter::contains(const uint256& hash) const
{
vector<unsigned char> data(hash.begin(), hash.end());
return contains(data);
std::vector<unsigned char> vData(hash.begin(), hash.end());
return contains(vData);
}

void CRollingBloomFilter::reset()
{
unsigned int nNewTweak = GetRand(std::numeric_limits<unsigned int>::max());
b1.reset(nNewTweak);
b2.reset(nNewTweak);
nInsertions = 0;
nTweak = GetRand(std::numeric_limits<unsigned int>::max());
nEntriesThisGeneration = 0;
nGeneration = 1;
std::fill(data.begin(), data.end(), 0);
}

0 comments on commit be45961

Please sign in to comment.