#include #include #include "theft.h" #include "theft_bloom.h" #define DEFAULT_BLOOM_BITS 17 #define DEBUG_BLOOM_FILTER 0 #define LOG2_BITS_PER_BYTE 3 #define HASH_COUNT 4 static uint8_t get_bits_set_count(uint8_t counts[256], uint8_t byte); /* Initialize a bloom filter. */ struct theft_bloom *theft_bloom_init(uint8_t bit_size2) { size_t sz = 1 << (bit_size2 - LOG2_BITS_PER_BYTE); struct theft_bloom *b = malloc(sizeof(struct theft_bloom) + sz); if (b) { b->size = sz; b->bit_count = bit_size2; memset(b->bits, 0, sz); } return b; } /* Hash data and mark it in the bloom filter. */ void theft_bloom_mark(struct theft_bloom *b, uint8_t *data, size_t data_size) { uint64_t hash = theft_hash_onepass(data, data_size); uint8_t bc = b->bit_count; uint64_t mask = (1 << bc) - 1; /* Use HASH_COUNT distinct slices of MASK bits as hashes for the bloom filter. */ int bit_inc = (64 - bc) / HASH_COUNT; for (int i = 0; i < (64 - bc); i += bit_inc) { uint64_t v = (hash & (mask << i)) >> i; size_t offset = v / 8; uint8_t bit = 1 << (v & 0x07); b->bits[offset] |= bit; } } /* Check whether the data's hash is in the bloom filter. */ bool theft_bloom_check(struct theft_bloom *b, uint8_t *data, size_t data_size) { uint64_t hash = theft_hash_onepass(data, data_size); uint8_t bc = b->bit_count; uint64_t mask = (1 << bc) - 1; int bit_inc = (64 - bc) / HASH_COUNT; for (int i = 0; i < (64 - bc); i += bit_inc) { uint64_t v = (hash & (mask << i)) >> i; size_t offset = v / 8; uint8_t bit = 1 << (v & 0x07); if (0 == (b->bits[offset] & bit)) { return false; } } return true; } /* Free the bloom filter. */ void theft_bloom_free(struct theft_bloom *b) { free(b); } /* Dump the bloom filter's contents. (Debugging.) */ void theft_bloom_dump(struct theft_bloom *b) { uint8_t counts[256]; memset(counts, 0xFF, sizeof(counts)); size_t total = 0; uint16_t row_total = 0; for (size_t i = 0; i < b->size; i++) { uint8_t count = get_bits_set_count(counts, b->bits[i]); total += count; row_total += count; #if DEBUG_BLOOM_FILTER > 1 char c = (count == 0 ? '.' : '0' + count); printf("%c", c); if ((i & 63) == 0 || i == b->size - 1) { printf(" - %2.1f%%\n", (100 * row_total) / (64.0 * 8)); row_total = 0; } #endif } #if DEBUG_BLOOM_FILTER printf(" -- bloom filter: %zd of %zd bits set (%2d%%)\n", total, 8*b->size, (int)((100.0 * total)/(8.0*b->size))); #endif /* If the total number of bits set is > the number of bytes * in the table (i.e., > 1/8 full) then warn the user. */ if (total > b->size) { fprintf(stderr, "\nWARNING: bloom filter is %zd%% full, " "larger bloom_bits value recommended.\n", (size_t)((100 * total) / (8 * b->size))); } } /* Recommend a bloom filter size for a given number of trials. */ uint8_t theft_bloom_recommendation(int trials) { /* With a preferred priority of false positives under 0.1%, * the required number of bits m in the bloom filter is: * m = -lg(0.001)/(lg(2)^2) == 14.378 ~= 14, * so we want an array with 1 << ceil(log2(14*trials)) cells. * * Note: The above formula is for the *ideal* number of hash * functions, but we're using a hardcoded count. It appears to work * well enough in practice, though, and this can be adjusted later * without impacting the API. */ #define TRIAL_MULTIPILER 14 uint8_t res = DEFAULT_BLOOM_BITS; const uint8_t min = THEFT_BLOOM_BITS_MIN - LOG2_BITS_PER_BYTE; const uint8_t max = THEFT_BLOOM_BITS_MAX - LOG2_BITS_PER_BYTE; for (uint8_t i = min; i < max; i++) { int32_t v = (1 << i); if (v > (TRIAL_MULTIPILER * trials)) { res = i + LOG2_BITS_PER_BYTE; break; } } #if DEBUG_BLOOM_FILTER size_t sz = 1 << (res - LOG2_BITS_PER_BYTE); printf("Using %zd bytes for bloom filter: %d trials -> bit_size2 %u\n", sizeof(struct theft_bloom) + sz, trials, res); #endif return res; } /* Check a byte->bits set table, and lazily populate it. */ static uint8_t get_bits_set_count(uint8_t counts[256], uint8_t byte) { uint8_t v = counts[byte]; if (v != 0xFF) { return v; } uint8_t t = 0; for (uint8_t i = 0; i < 8; i++) { if (byte & (1 << i)) { t++; } } counts[byte] = t; return t; }