Start refactoring hashing.

This commit moves some of the hash datastructures and code from
util.cc into Hash.cc - where it seems more appropriate.

It also starts to make more Keyed hash functions available - still
using siphash as the default 64 bit keyed hash, but also making
128 and 256 bit highway hashes available.

There already are a few other functions that are defined but not
yet implemented - these will be "static" keyed hashes - which use
an installation specific key. These will be used to, e.g., get
rid of md5 hashing for the generation of file UIDs.

src/Hash.cc

@@ -1,32 +1,58 @@
 // See the file "COPYING" in the main distribution directory for copyright.
 
-// The hash function works as follows:
-//
-// 1) For short data we have a number of universal hash functions:
-// UHASH_CW (ax + b (mod p)), H3, Dietzfelbinger and UMAC_NH (UMAC_NH is
-// not as strongly universal as the others, but probably enough). All
-// these functions require number of random bits linear to the data
-// length. And we use them for data no longer than UHASH_KEY_SIZE.
-// They are faster than HMAC/MD5 used for longer data, and most hash
-// operations are on short data.
-//
-// 2) As a fall-back, we use HMAC/MD5 (keyed MD5) for data of arbitrary
-// length. MD5 is used as a scrambling scheme so that it is difficult
-// for the adversary to construct conflicts, though I do not know if
-// HMAC/MD5 is provably universal.
-
 #include "zeek-config.h"
 
 #include "Hash.h"
+#include "digest.h"
 #include "Reporter.h"
 #include "BroString.h"
 
 #include "highwayhash/sip_hash.h"
+#include "highwayhash/highwayhash_target.h"
+#include "highwayhash/instruction_sets.h"
+
+// we use the following lines to not pull in the highwayhash headers in Hash.h - but to check the types did not change underneath us.
+static_assert(std::is_same<hash64_t, highwayhash::HHResult64>::value, "Highwayhash return values must match hash_x_t");
+static_assert(std::is_same<hash128_t, highwayhash::HHResult128>::value, "Highwayhash return values must match hash_x_t");
+static_assert(std::is_same<hash256_t, highwayhash::HHResult256>::value, "Highwayhash return values must match hash_x_t");
+
+void KeyedHash::InitializeSeeds(const std::array<uint32_t, SEED_INIT_SIZE>& seed_data)
+	{
+	static_assert(std::is_same<decltype(KeyedHash::shared_siphash_key), highwayhash::SipHashState::Key>::value, "Highwayhash Key is not unsigned long long[2]");
+	static_assert(std::is_same<decltype(KeyedHash::shared_highwayhash_key), highwayhash::HHKey>::value, "Highwayhash HHKey is not uint64_t[4]");
+	if ( seeds_initialized )
+		return;
+
+	internal_md5((const u_char*) seed_data.data(), sizeof(seed_data) - 16, shared_hmac_md5_key); // The last 128 bits of buf are for siphash
+	// yes, we use the same buffer twice to initialize two different keys. This should not really be a
+	// security problem of any kind: hmac-md5 is not really used anymore - and even if it was, the hashes
+	// should not reveal any information about their initialization vector.
+	static_assert(sizeof(shared_highwayhash_key) == SHA256_DIGEST_LENGTH);
+	calculate_digest(Hash_SHA256, (const u_char*) seed_data.data(), sizeof(seed_data) - 16, reinterpret_cast<unsigned char*>(shared_highwayhash_key));
+	memcpy(shared_siphash_key, reinterpret_cast<const char*>(seed_data.data()) + 64, 16);
+
+	seeds_initialized = true;
+	}
+
+hash64_t KeyedHash::Hash64(const void* bytes, uint64_t size)
+	{
+	return highwayhash::SipHash(shared_siphash_key, reinterpret_cast<const char *>(bytes), size);
+	}
+
+void KeyedHash::Hash128(const void* bytes, uint64_t size, hash128_t* result)
+	{
+	highwayhash::InstructionSets::Run<highwayhash::HighwayHash>(shared_highwayhash_key, reinterpret_cast<const char *>(bytes), size, result);
+	}
+
+void KeyedHash::Hash256(const void* bytes, uint64_t size, hash256_t* result)
+	{
+	highwayhash::InstructionSets::Run<highwayhash::HighwayHash>(shared_highwayhash_key, reinterpret_cast<const char *>(bytes), size, result);
+	}
 
 void init_hash_function()
 	{
 	// Make sure we have already called init_random_seed().
-	if ( ! (hmac_key_set && siphash_key_set) )
+	if ( ! KeyedHash::IsInitialized() )
 		reporter->InternalError("Zeek's hash functions aren't fully initialized");
 	}
 
@@ -156,6 +182,5 @@ void* HashKey::CopyKey(const void* k, int s) const
 
 hash_t HashKey::HashBytes(const void* bytes, int size)
 	{
-	hash_t digest = highwayhash::SipHash(shared_siphash_key, reinterpret_cast<const char *>(bytes), size);
-	return digest;
+	return KeyedHash::Hash64(bytes, size);
 	}

src/Hash.h

@@ -1,5 +1,22 @@
 // See the file "COPYING" in the main distribution directory for copyright.
 
+/***
+ * This file contains functions to generate hashes used keyed hash functions.
+ * Keyed hash functions make it difficult/impossible to find information about the
+ * output of a hash when the key is unknown to the attacker. This fact holds, even
+ * when the input value us known.
+ *
+ * We use these kinds of hashes heavily internally - e.g. for scriptland hash generation.
+ * It is important that these hashes are not easily guessable to prevent complexity attacks.
+ *
+ * The HashKey class is the actual class that is used to generate Hash keys that are used internally,
+ * e.g. for lookups in hash-tables; the Hashes are also used for connection ID generation.
+ *
+ * This means that the hashes created by most functions in this file will be different each run, unless
+ * a seed file is used. There are a few functions that create hashes that are static over runs
+ * and use an installation-wide seed value; these are specifically called out.
+ */
+
 #pragma once
 
 #include "util.h" // for bro_int_t
@@ -8,7 +25,43 @@
 
 class BroString;
 
+// to allow bro_md5_hmac access to the hmac seed
+#include "ZeekArgs.h"
+class Val;
+class Frame;
+namespace BifFunc {
+	extern Val* bro_md5_hmac(Frame* frame, const zeek::Args*);
+}
+
 typedef uint64_t hash_t;
+typedef uint64_t hash64_t;
+typedef uint64_t hash128_t[2];
+typedef uint64_t hash256_t[4];
+
+class KeyedHash {
+public:
+	constexpr static int SEED_INIT_SIZE = 20;
+	static void InitializeSeeds(const std::array<uint32_t, SEED_INIT_SIZE>& seed_data);
+	static bool IsInitialized() { return seeds_initialized; }
+
+	static hash64_t Hash64(const void* bytes, uint64_t size);
+	static void Hash128(const void* bytes, uint64_t size, hash128_t* result);
+	static void Hash256(const void* bytes, uint64_t size, hash256_t* result);
+
+	static hash64_t StaticHash64(const void* bytes, uint64_t size);
+	static void StaticHash128(const void* bytes, uint64_t size, hash128_t* result);
+	static void StaticHash256(const void* bytes, uint64_t size, hash256_t* result);
+private:
+	// actually HHKey
+	alignas(32) inline static uint64_t shared_highwayhash_key[4];
+	// actually HH_U64, which has the same type
+	alignas(16) inline static unsigned long long shared_siphash_key[2];
+	inline static uint8_t shared_hmac_md5_key[16];
+	inline static bool seeds_initialized = false;
+
+	friend void hmac_md5(size_t size, const unsigned char* bytes, unsigned char digest[16]);
+	friend Val* BifFunc::bro_md5_hmac(Frame* frame, const zeek::Args*);
+};
 
 typedef enum {
 	HASH_KEY_INT,

src/probabilistic/Hasher.cc

@@ -106,7 +106,7 @@ UHF::UHF(Hasher::seed_t arg_seed)
 // times.
 Hasher::digest UHF::hash(const void* x, size_t n) const
 	{
-	assert(sizeof(Hasher::seed_t) == 16); // siphash always needs a 128 bit seed
+	static_assert(std::is_same<highwayhash::SipHashState::Key, decltype(seed.h)>::value, "Seed value is not the same type as highwayhash key");
 	return highwayhash::SipHash(seed.h, reinterpret_cast<const char*>(x), n);
 	}
 

src/probabilistic/Hasher.h

@@ -3,6 +3,7 @@
 #pragma once
 
 #include "Hash.h"
+#include "highwayhash/sip_hash.h"
 
 #include <broker/expected.hh>
 
@@ -24,7 +25,8 @@ public:
 	typedef hash_t digest;
 	typedef std::vector<digest> digest_vector;
 	struct seed_t {
-		alignas(16) highwayhash::HH_U64 h[2];
+		// actually HH_U64, which has the same type
+		alignas(16) unsigned long long h[2];
 
 		friend seed_t operator+(seed_t lhs, const uint64_t rhs) {
 			lhs.h[0] += rhs;

src/util.cc

@@ -55,6 +55,7 @@
 #include "iosource/Manager.h"
 #include "iosource/PktSrc.h"
 #include "ConvertUTF.h"
+#include "Hash.h"
 
 #include "3rdparty/doctest.h"
 
@@ -997,27 +998,21 @@ std::string strstrip(std::string s)
 	return s;
 	}
 
-bool hmac_key_set = false;
-uint8_t shared_hmac_md5_key[16];
-
-bool siphash_key_set = false;
-alignas(16) highwayhash::HH_U64 shared_siphash_key[2];
-
 void hmac_md5(size_t size, const unsigned char* bytes, unsigned char digest[16])
 	{
-	if ( ! hmac_key_set )
+	if ( ! KeyedHash::seeds_initialized )
 		reporter->InternalError("HMAC-MD5 invoked before the HMAC key is set");
 
 	internal_md5(bytes, size, digest);
 
 	for ( int i = 0; i < 16; ++i )
-		digest[i] ^= shared_hmac_md5_key[i];
+		digest[i] ^= KeyedHash::shared_hmac_md5_key[i];
 
 	internal_md5(digest, 16, digest);
 	}
 
 static bool read_random_seeds(const char* read_file, uint32_t* seed,
-				uint32_t* buf, int bufsiz)
+				std::array<uint32_t, KeyedHash::SEED_INIT_SIZE>& buf)
 	{
 	FILE* f = nullptr;
 
@@ -1035,8 +1030,8 @@ static bool read_random_seeds(const char* read_file, uint32_t* seed,
 		return false;
 		}
 
-	// Read seeds for MD5.
-	for ( int i = 0; i < bufsiz; ++i )
+	// Read seeds for hmac-md5/siphash/highwayhash.
+	for ( int i = 0; i < KeyedHash::SEED_INIT_SIZE; ++i )
 		{
 		int tmp;
 		if ( fscanf(f, "%u", &tmp) != 1 )
@@ -1053,7 +1048,7 @@ static bool read_random_seeds(const char* read_file, uint32_t* seed,
 	}
 
 static bool write_random_seeds(const char* write_file, uint32_t seed,
-				uint32_t* buf, int bufsiz)
+				std::array<uint32_t, KeyedHash::SEED_INIT_SIZE>& buf)
 	{
 	FILE* f = nullptr;
 
@@ -1066,7 +1061,7 @@ static bool write_random_seeds(const char* write_file, uint32_t seed,
 
 	fprintf(f, "%u\n", seed);
 
-	for ( int i = 0; i < bufsiz; ++i )
+	for ( int i = 0; i < KeyedHash::SEED_INIT_SIZE; ++i )
 		fprintf(f, "%u\n", buf[i]);
 
 	fclose(f);
@@ -1096,16 +1091,14 @@ void bro_srandom(unsigned int seed)
 
 void init_random_seed(const char* read_file, const char* write_file)
 	{
-	static const int bufsiz = 20;
-	uint32_t buf[bufsiz];
-	memset(buf, 0, sizeof(buf));
-	int pos = 0;	// accumulates entropy
+	std::array<uint32_t, KeyedHash::SEED_INIT_SIZE> buf = {};
+	size_t pos = 0;	// accumulates entropy
 	bool seeds_done = false;
 	uint32_t seed = 0;
 
 	if ( read_file )
 		{
-		if ( ! read_random_seeds(read_file, &seed, buf, bufsiz) )
+		if ( ! read_random_seeds(read_file, &seed, buf) )
 			reporter->FatalError("Could not load seeds from file '%s'.\n",
 					     read_file);
 		else
@@ -1115,7 +1108,7 @@ void init_random_seed(const char* read_file, const char* write_file)
 #ifdef HAVE_GETRANDOM
 	if ( ! seeds_done )
 		{
-		ssize_t nbytes = getrandom(buf, sizeof(buf), 0);
+		ssize_t nbytes = getrandom(buf.data(), sizeof(buf), 0);
 		seeds_done = nbytes == ssize_t(sizeof(buf));
 		}
 #endif
@@ -1123,7 +1116,7 @@ void init_random_seed(const char* read_file, const char* write_file)
 	if ( ! seeds_done )
 		{
 		// Gather up some entropy.
-		gettimeofday((struct timeval *)(buf + pos), 0);
+		gettimeofday((struct timeval *)(buf.data() + pos), 0);
 		pos += sizeof(struct timeval) / sizeof(uint32_t);
 
 		// use urandom. For reasons see e.g. http://www.2uo.de/myths-about-urandom/
@@ -1137,8 +1130,8 @@ void init_random_seed(const char* read_file, const char* write_file)
 
 		if ( fd >= 0 )
 			{
-			int amt = read(fd, buf + pos,
-					sizeof(uint32_t) * (bufsiz - pos));
+			int amt = read(fd, buf.data() + pos,
+					sizeof(uint32_t) * (KeyedHash::SEED_INIT_SIZE - pos));
 			safe_close(fd);
 
 			if ( amt > 0 )
@@ -1149,12 +1142,12 @@ void init_random_seed(const char* read_file, const char* write_file)
 				errno = 0;
 			}
 
-		if ( pos < bufsiz )
-			reporter->FatalError("Could not read enough random data from /dev/urandom. Wanted %d, got %d", bufsiz, pos);
+		if ( pos < KeyedHash::SEED_INIT_SIZE )
+			reporter->FatalError("Could not read enough random data from /dev/urandom. Wanted %d, got %lu", KeyedHash::SEED_INIT_SIZE, pos);
 
 		if ( ! seed )
 			{
-			for ( int i = 0; i < pos; ++i )
+			for ( size_t i = 0; i < pos; ++i )
 				{
 				seed ^= buf[i];
 				seed = (seed << 1) | (seed >> 31);
@@ -1172,22 +1165,10 @@ void init_random_seed(const char* read_file, const char* write_file)
 		first_seed_saved = true;
 		}
 
-	if ( ! hmac_key_set )
-		{
-		assert(sizeof(buf) - 16 == 64);
-		internal_md5((const u_char*) buf, sizeof(buf) - 16, shared_hmac_md5_key); // The last 128 bits of buf are for siphash
-		hmac_key_set = true;
-		}
+	if ( ! KeyedHash::IsInitialized() )
+		KeyedHash::InitializeSeeds(buf);
 
-	if ( ! siphash_key_set )
-		{
-		assert(sizeof(buf) - 64 == 16); // siphash key length is always 128 bytes, independent of implementation
-		assert(sizeof(shared_siphash_key) == 16);
-		memcpy(shared_siphash_key, reinterpret_cast<const char*>(buf) + 64, 16);
-		siphash_key_set = true;
-		}
-
-	if ( write_file && ! write_random_seeds(write_file, seed, buf, bufsiz) )
+	if ( write_file && ! write_random_seeds(write_file, seed, buf) )
 		reporter->Error("Could not write seeds to file '%s'.\n",
 				write_file);
 	}

src/util.h

@@ -25,7 +25,6 @@
 #include <stdarg.h>
 #include <libgen.h>
 #include <memory> // std::unique_ptr
-#include "highwayhash/sip_hash.h"
 
 #include "zeek-config.h"
 
@@ -200,11 +199,6 @@ extern std::string strreplace(const std::string& s, const std::string& o, const
 // Remove all leading and trailing white space from string.
 extern std::string strstrip(std::string s);
 
-extern bool hmac_key_set;
-extern uint8_t shared_hmac_md5_key[16];
-extern bool siphash_key_set;
-extern highwayhash::HH_U64 shared_siphash_key[2];
-
 extern void hmac_md5(size_t size, const unsigned char* bytes,
 			unsigned char digest[16]);
 

src/zeek.bif

@@ -27,6 +27,7 @@
 #include "iosource/PktDumper.h"
 #include "IntrusivePtr.h"
 #include "input.h"
+#include "Hash.h"
 
 using namespace std;
 
@@ -615,7 +616,7 @@ function sha256_hash%(...%): string
 function md5_hmac%(...%): string
 	%{
 	unsigned char hmac[MD5_DIGEST_LENGTH];
-	MD5Val::hmac(@ARG@, shared_hmac_md5_key, hmac);
+	MD5Val::hmac(@ARG@, KeyedHash::shared_hmac_md5_key, hmac);
 	return new StringVal(md5_digest_print(hmac));
 	%}