Changing the Bloom filter hashing so that it's independent of

CompositeHash.

We do this by hashing values added to a BloomFilter another time more
with a stable hash seeded only by either the filter's name or the
global_hash_seed (or Bro's random() seed if neither is defined).

I'm also adding a new bif bloomfilter_internal_state() that returns a
string representation of a Bloom filter's current internal state. This
is solely for writing tests that check that the filters end up
consistent when seeded with the same value.
This commit is contained in:
Robin Sommer 2013-07-31 17:14:02 -07:00
parent 6c197fbebf
commit 2a0790c231
14 changed files with 157 additions and 59 deletions

View file

@ -3042,6 +3042,11 @@ module GLOBAL;
## Number of bytes per packet to capture from live interfaces.
const snaplen = 8192 &redef;
## Seed for hashes computed internally for probabilistic data structures. Using
## the same value here will make the hashes compatible between independent Bro
## instances. If left unset, Bro will use a temporary local seed.
const global_hash_seed: string = "" &redef;
# Load BiFs defined by plugins.
@load base/bif/plugins

View file

@ -566,14 +566,14 @@ BroType* BloomFilterVal::Type() const
void BloomFilterVal::Add(const Val* val)
{
HashKey* key = hash->ComputeHash(val, 1);
bloom_filter->Add(key->Hash());
bloom_filter->Add(key);
delete key;
}
size_t BloomFilterVal::Count(const Val* val) const
{
HashKey* key = hash->ComputeHash(val, 1);
size_t cnt = bloom_filter->Count(key->Hash());
size_t cnt = bloom_filter->Count(key);
delete key;
return cnt;
}
@ -588,6 +588,11 @@ bool BloomFilterVal::Empty() const
return bloom_filter->Empty();
}
string BloomFilterVal::InternalState() const
{
return bloom_filter->InternalState();
}
BloomFilterVal* BloomFilterVal::Merge(const BloomFilterVal* x,
const BloomFilterVal* y)
{

View file

@ -127,6 +127,7 @@ public:
size_t Count(const Val* val) const;
void Clear();
bool Empty() const;
string InternalState() const;
static BloomFilterVal* Merge(const BloomFilterVal* x,
const BloomFilterVal* y);

View file

@ -490,6 +490,16 @@ BitVector::size_type BitVector::FindNext(size_type i) const
return block ? bi * bits_per_block + lowest_bit(block) : find_from(bi + 1);
}
size_t BitVector::Hash() const
{
size_t hash = 0;
for ( size_type i = 0; i < Blocks(); ++i )
hash += bits[i];
return hash;
}
BitVector::size_type BitVector::lowest_bit(block_type block)
{
block_type x = block - (block & (block - 1));

View file

@ -276,6 +276,13 @@ public:
*/
size_type FindNext(size_type i) const;
/** Computes a hash value of the internal representation.
* This is mainly for debugging/testing purposes.
*
* @return The hash.
*/
size_t Hash() const;
/**
* Serializes the bit vector.
*

View file

@ -9,6 +9,8 @@
#include "CounterVector.h"
#include "Serializer.h"
#include "../util.h"
using namespace probabilistic;
BloomFilter::BloomFilter()
@ -107,6 +109,11 @@ BasicBloomFilter* BasicBloomFilter::Clone() const
return copy;
}
std::string BasicBloomFilter::InternalState() const
{
return fmt("%" PRIu64, (uint64_t)bits->Hash());
}
BasicBloomFilter::BasicBloomFilter()
{
bits = 0;
@ -133,14 +140,18 @@ bool BasicBloomFilter::DoUnserialize(UnserialInfo* info)
return (bits != 0);
}
void BasicBloomFilter::AddImpl(const Hasher::digest_vector& h)
void BasicBloomFilter::Add(const HashKey* key)
{
Hasher::digest_vector h = hasher->Hash(key);
for ( size_t i = 0; i < h.size(); ++i )
bits->Set(h[i] % bits->Size());
}
size_t BasicBloomFilter::CountImpl(const Hasher::digest_vector& h) const
size_t BasicBloomFilter::Count(const HashKey* key) const
{
Hasher::digest_vector h = hasher->Hash(key);
for ( size_t i = 0; i < h.size(); ++i )
{
if ( ! (*bits)[h[i] % bits->Size()] )
@ -206,6 +217,11 @@ CountingBloomFilter* CountingBloomFilter::Clone() const
return copy;
}
string CountingBloomFilter::InternalState() const
{
return fmt("%" PRIu64, (uint64_t)cells->Hash());
}
IMPLEMENT_SERIAL(CountingBloomFilter, SER_COUNTINGBLOOMFILTER)
bool CountingBloomFilter::DoSerialize(SerialInfo* info) const
@ -222,14 +238,18 @@ bool CountingBloomFilter::DoUnserialize(UnserialInfo* info)
}
// TODO: Use partitioning in add/count to allow for reusing CMS bounds.
void CountingBloomFilter::AddImpl(const Hasher::digest_vector& h)
void CountingBloomFilter::Add(const HashKey* key)
{
Hasher::digest_vector h = hasher->Hash(key);
for ( size_t i = 0; i < h.size(); ++i )
cells->Increment(h[i] % cells->Size());
}
size_t CountingBloomFilter::CountImpl(const Hasher::digest_vector& h) const
size_t CountingBloomFilter::Count(const HashKey* key) const
{
Hasher::digest_vector h = hasher->Hash(key);
CounterVector::size_type min =
std::numeric_limits<CounterVector::size_type>::max();

View file

@ -22,27 +22,20 @@ public:
virtual ~BloomFilter();
/**
* Adds an element of type T to the Bloom filter.
* @param x The element to add
* Adds an element to the Bloom filter.
*
* @param key The key associated with the element to add.
*/
template <typename T>
void Add(const T& x)
{
AddImpl((*hasher)(x));
}
virtual void Add(const HashKey* key) = 0;
/**
* Retrieves the associated count of a given value.
*
* @param x The value of type `T` to check.
* @param key The key associated with the element to check.
*
* @return The counter associated with *x*.
* @return The counter associated with *key*.
*/
template <typename T>
size_t Count(const T& x) const
{
return CountImpl((*hasher)(x));
}
virtual size_t Count(const HashKey* key) const = 0;
/**
* Checks whether the Bloom filter is empty.
@ -72,6 +65,12 @@ public:
*/
virtual BloomFilter* Clone() const = 0;
/**
* Returns a string with a representation of the Bloom filter's
* internal state. This is for debugging/testing purposes only.
*/
virtual string InternalState() const = 0;
/**
* Serializes the Bloom filter.
*
@ -106,25 +105,6 @@ protected:
*/
BloomFilter(const Hasher* hasher);
/**
* Abstract method for implementinng the *Add* operation.
*
* @param hashes A set of *k* hashes for the item to add, computed by
* the internal hasher object.
*
*/
virtual void AddImpl(const Hasher::digest_vector& hashes) = 0;
/**
* Abstract method for implementing the *Count* operation.
*
* @param hashes A set of *k* hashes for the item to add, computed by
* the internal hasher object.
*
* @return Returns the counter associated with the hashed element.
*/
virtual size_t CountImpl(const Hasher::digest_vector& hashes) const = 0;
const Hasher* hasher;
};
@ -177,6 +157,7 @@ public:
virtual void Clear();
virtual bool Merge(const BloomFilter* other);
virtual BasicBloomFilter* Clone() const;
virtual string InternalState() const;
protected:
DECLARE_SERIAL(BasicBloomFilter);
@ -187,8 +168,8 @@ protected:
BasicBloomFilter();
// Overridden from BloomFilter.
virtual void AddImpl(const Hasher::digest_vector& h);
virtual size_t CountImpl(const Hasher::digest_vector& h) const;
virtual void Add(const HashKey* key);
virtual size_t Count(const HashKey* key) const;
private:
BitVector* bits;
@ -216,6 +197,7 @@ public:
virtual void Clear();
virtual bool Merge(const BloomFilter* other);
virtual CountingBloomFilter* Clone() const;
virtual string InternalState() const;
protected:
DECLARE_SERIAL(CountingBloomFilter);
@ -226,8 +208,8 @@ protected:
CountingBloomFilter();
// Overridden from BloomFilter.
virtual void AddImpl(const Hasher::digest_vector& h);
virtual size_t CountImpl(const Hasher::digest_vector& h) const;
virtual void Add(const HashKey* key);
virtual size_t Count(const HashKey* key) const;
private:
CounterVector* cells;

View file

@ -153,6 +153,11 @@ CounterVector operator|(const CounterVector& x, const CounterVector& y)
}
size_t CounterVector::Hash() const
{
return bits->Hash();
}
bool CounterVector::Serialize(SerialInfo* info) const
{
return SerialObj::Serialize(info);

View file

@ -126,6 +126,13 @@ public:
*/
CounterVector& operator|=(const CounterVector& other);
/** Computes a hash value of the internal representation.
* This is mainly for debugging/testing purposes.
*
* @return The hash.
*/
size_t Hash() const;
/**
* Serializes the bit vector.
*

View file

@ -31,6 +31,11 @@ size_t Hasher::MakeSeed(const void* data, size_t size)
return *reinterpret_cast<size_t*>(buf); // Use the first bytes as seed.
}
Hasher::digest_vector Hasher::Hash(const HashKey* key) const
{
return Hash(key->Key(), key->Size());
}
bool Hasher::Serialize(SerialInfo* info) const
{
return SerialObj::Serialize(info);
@ -77,7 +82,6 @@ Hasher::Hasher(size_t arg_k, size_t arg_seed)
seed = arg_seed;
}
UHF::UHF(size_t seed)
: h(seed)
{

View file

@ -50,6 +50,15 @@ public:
return Hash(&x, sizeof(T));
}
/**
* Computes hash values for an element.
*
* @param x The key of the value to hash.
*
* @return Vector of *k* hash values.
*/
digest_vector Hash(const HashKey* key) const;
/**
* Computes the hashes for a set of bytes.
*

View file

@ -20,11 +20,6 @@ module GLOBAL;
## Creates a basic Bloom filter.
##
## .. note:: A Bloom filter can have a name associated with it. In the future,
## Bloom filters with the same name will be compatible across indepedent Bro
## instances, i.e., it will be possible to merge them. Currently, however, that is
## not yet supported.
##
## fp: The desired false-positive rate.
##
## capacity: the maximum number of elements that guarantees a false-positive
@ -61,11 +56,6 @@ function bloomfilter_basic_init%(fp: double, capacity: count,
## alternative to bloomfilter_basic_init where the user has full control over
## the number of hash functions and cells in the underlying bit vector.
##
## .. note:: A Bloom filter can have a name associated with it. In the future,
## Bloom filters with the same name will be compatible across indepedent Bro
## instances, i.e., it will be possible to merge them. Currently, however, that is
## not yet supported.
##
## k: The number of hash functions to use.
##
## cells: The number of cells of the underlying bit vector.
@ -102,11 +92,6 @@ function bloomfilter_basic_init2%(k: count, cells: count,
## Creates a counting Bloom filter.
##
## .. note:: A Bloom filter can have a name associated with it. In the future,
## Bloom filters with the same name will be compatible across indepedent Bro
## instances, i.e., it will be possible to merge them. Currently, however, that is
## not yet supported.
##
## k: The number of hash functions to use.
##
## cells: The number of cells of the underlying counter vector. As there's no
@ -250,3 +235,13 @@ function bloomfilter_merge%(bf1: opaque of bloomfilter,
return BloomFilterVal::Merge(bfv1, bfv2);
%}
## Returns a string with a representation of a Bloom filter's internal
## state. This is for debugging/testing purposes only.
##
## bf: The Bloom filter handle.
function bloomfilter_internal_state%(bf: opaque of bloomfilter%): string
%{
BloomFilterVal* bfv = static_cast<BloomFilterVal*>(bf);
return new StringVal(bfv->InternalState());
%}

View file

@ -0,0 +1,8 @@
bf1, global_seed, 1
bf2, global_seed, 5
bf3, my_seed, 5
bf4, my_seed, 6
bf1, global_seed, 5
bf2, global_seed, 6
bf3, my_seed, 5
bf4, my_seed, 6

View file

@ -0,0 +1,40 @@
# @TEST-EXEC: bro -b %INPUT global_hash_seed="foo" >>output
# @TEST-EXEC: bro -b %INPUT global_hash_seed="my_seed" >>output
# @TEST-EXEC: btest-diff output
type Foo: record
{
a: count;
b: string;
};
function test_bloom_filter()
{
local bf1 = bloomfilter_basic_init(0.9, 10);
bloomfilter_add(bf1, "foo");
bloomfilter_add(bf1, "bar");
local bf2 = bloomfilter_basic_init(0.9, 10);
bloomfilter_add(bf2, Foo($a=1, $b="xx"));
bloomfilter_add(bf2, Foo($a=2, $b="yy"));
local bf3 = bloomfilter_basic_init(0.9, 10, "my_seed");
bloomfilter_add(bf3, "foo");
bloomfilter_add(bf3, "bar");
local bf4 = bloomfilter_basic_init(0.9, 10, "my_seed");
bloomfilter_add(bf4, Foo($a=1, $b="xx"));
bloomfilter_add(bf4, Foo($a=2, $b="yy"));
print "bf1, global_seed", bloomfilter_internal_state(bf1);
print "bf2, global_seed", bloomfilter_internal_state(bf2);
print "bf3, my_seed", bloomfilter_internal_state(bf3);
print "bf4, my_seed", bloomfilter_internal_state(bf4);
}
event bro_init()
{
test_bloom_filter();
}