use siphash to replace hash_string (#472)

* use siphash to replace hash_string

* add siphash algorithm

Author: dreamwind1985

include/xquic/xquic.h

@@ -1094,6 +1094,8 @@ typedef struct xqc_config_s {
 
     /** bucket size of connection hash table in engine */
     size_t          conns_hash_bucket_size;
+    /* for warning when the number of elements in one bucket exceeds the value of hash_conflict_threshold*/
+    uint32_t        hash_conflict_threshold;
 
     /** capacity of connection priority queue in engine */
     size_t          conns_active_pq_capacity;

src/common/xqc_siphash.h

@@ -0,0 +1,190 @@
+/**
+ * @copyright Copyright (c) 2025, Alibaba Group Holding Limited
+ */
+
+#ifndef _XQC_SIPHASH_H_INCLUDED_
+#define _XQC_SIPHASH_H_INCLUDED_
+
+#include <inttypes.h>
+#include <stdint.h>
+#include <string.h>
+#include "src/common/xqc_common.h"
+
+#define XQC_SIPHASH_KEY_SIZE        16
+#define XQC_SIPHASH_C_ROUNDS        2
+#define XQC_SIPHASH_D_ROUNDS        4
+#define XQC_DEFAULT_HASH_SIZE       8
+
+/* save siphash context */
+typedef struct xqc_siphash_ctx {
+    /* v0 v1 v2 v3  */
+    uint64_t v0;
+    uint64_t v1;
+    uint64_t v2;
+    uint64_t v3;
+    int hash_size; /* save sizeof(hash), only 8 or 16 */
+    /*  SipHash-2-4  */
+    int crounds; 
+    int drounds;
+} xqc_siphash_ctx_t;
+
+#define ROTL(x, b) (uint64_t)(((x) << (b)) | ((x) >> (64 - (b))))
+
+#define U32TO8_LE(p, v)                                                        \
+    (p)[0] = (uint8_t)((v));                                                   \
+    (p)[1] = (uint8_t)((v) >> 8);                                              \
+    (p)[2] = (uint8_t)((v) >> 16);                                             \
+    (p)[3] = (uint8_t)((v) >> 24);
+
+#define U64TO8_LE(p, v)                                                        \
+    U32TO8_LE((p), (uint32_t)((v)));                                           \
+    U32TO8_LE((p) + 4, (uint32_t)((v) >> 32));
+
+#define U8TO64_LE(p)                                                           \
+    (((uint64_t)((p)[0])) | ((uint64_t)((p)[1]) << 8) |                        \
+     ((uint64_t)((p)[2]) << 16) | ((uint64_t)((p)[3]) << 24) |                 \
+     ((uint64_t)((p)[4]) << 32) | ((uint64_t)((p)[5]) << 40) |                 \
+     ((uint64_t)((p)[6]) << 48) | ((uint64_t)((p)[7]) << 56))
+
+#define SIPROUND                                                               \
+    do {                                                                       \
+        v0 += v1;                                                              \
+        v1 = ROTL(v1, 13);                                                     \
+        v1 ^= v0;                                                              \
+        v0 = ROTL(v0, 32);                                                     \
+        v2 += v3;                                                              \
+        v3 = ROTL(v3, 16);                                                     \
+        v3 ^= v2;                                                              \
+        v0 += v3;                                                              \
+        v3 = ROTL(v3, 21);                                                     \
+        v3 ^= v0;                                                              \
+        v2 += v1;                                                              \
+        v1 = ROTL(v1, 17);                                                     \
+        v1 ^= v2;                                                              \
+        v2 = ROTL(v2, 32);                                                     \
+    } while (0)
+
+static inline int 
+xqc_siphash_init(xqc_siphash_ctx_t *ctx, const unsigned char *k, size_t key_len,
+    size_t hash_size, int crounds, int drounds)
+{
+    uint64_t k0, k1;
+    if (key_len != XQC_SIPHASH_KEY_SIZE) {
+        return XQC_ERROR;
+    }
+    /* hash_size must be 8 or 16 */
+    if (hash_size != 8 && hash_size != 16) {
+        return XQC_ERROR;
+    }
+    k0 = U8TO64_LE(k);
+    k1 = U8TO64_LE(k + 8);
+    
+    ctx->v0 = 0x736f6d6570736575ULL ^ k0;
+    ctx->v1 = 0x646f72616e646f6dULL ^ k1;
+    ctx->v2 = 0x6c7967656e657261ULL ^ k0;
+    ctx->v3 = 0x7465646279746573ULL ^ k1;
+
+    ctx->hash_size = hash_size;
+    if (hash_size == 16) {
+        ctx->v1 ^= 0xee;
+    }
+    /* default: SipHash-2-4 */
+    if (crounds == 0) {
+        ctx->crounds = XQC_SIPHASH_C_ROUNDS;
+    } else {
+        ctx->crounds = crounds;
+    }
+    if (drounds == 0) {
+        ctx->drounds = XQC_SIPHASH_D_ROUNDS;
+    } else {
+        ctx->drounds = drounds;
+    }
+    return XQC_OK;
+}
+
+
+/*
+    Computes a SipHash value
+    *ctx: point to siphash context 
+    *in: pointer to input data (read-only)
+    inlen: input data length in bytes (any size_t value)
+    *out: pointer to output data (write-only), outlen bytes must be allocated
+    outlen: length of the output in bytes, must be 8 or 16
+*/
+
+static inline int
+xqc_siphash(xqc_siphash_ctx_t *ctx, const uint8_t *in, size_t inlen, uint8_t *out, size_t outlen)
+{
+    uint64_t b = (uint64_t)inlen << 56, m;
+    const uint8_t *pi = in, *end = in + inlen - (inlen % sizeof(uint64_t));
+    int left = inlen & 7;
+    int i = 0;
+    uint64_t v0 = ctx->v0;
+    uint64_t v1 = ctx->v1;
+    uint64_t v2 = ctx->v2;
+    uint64_t v3 = ctx->v3;
+
+    if (outlen != ctx->hash_size) {
+        return XQC_ERROR;
+    }
+
+    for(; pi != end; pi += 8) {
+        m = U8TO64_LE(pi);
+        v3 ^= m;
+        for (i = 0; i < ctx->crounds; i++) {
+            SIPROUND; 
+        }
+        v0 ^= m;
+    }
+
+    switch (left) {
+    case 7:
+        b |= ((uint64_t)pi[6]) << 48;
+    case 6:
+        b |= ((uint64_t)pi[5]) << 40;
+    case 5:
+        b |= ((uint64_t)pi[4]) << 32;
+    case 4:
+        b |= ((uint64_t)pi[3]) << 24;
+    case 3:
+        b |= ((uint64_t)pi[2]) << 16;
+    case 2:
+        b |= ((uint64_t)pi[1]) << 8;
+    case 1:
+        b |= ((uint64_t)pi[0]);
+        break;
+    case 0:
+        break;
+    }
+
+    v3 ^= b;
+    for (i = 0; i < ctx->crounds; ++i)
+        SIPROUND;
+    v0 ^= b;
+
+    if (outlen == 16) {
+        v2 ^= 0xee;
+    } else {
+        v2 ^= 0xff;
+    }
+
+    for (i = 0; i < ctx->drounds; ++i)
+        SIPROUND;
+
+    b = v0 ^ v1 ^ v2  ^ v3;
+    U64TO8_LE(out, b);
+    if (outlen == 8) {
+        return XQC_OK;
+    }
+    v1 ^= 0xdd;
+    for (i = 0; i < ctx->drounds; ++i)
+        SIPROUND;
+    b = v0 ^ v1 ^ v2  ^ v3;
+    U64TO8_LE(out + 8, b);
+    return XQC_OK;
+}
+
+
+
+
+#endif /* _XQC_SIPHASH_H_INCLUDED_ */

src/common/xqc_str_hash.h

@@ -6,8 +6,21 @@
 #define _XQC_STR_HASH_INCLUDED_
 
 #include <stdint.h>
+#include <time.h>
 
 #include "src/common/xqc_str.h"
+#include "src/common/xqc_siphash.h"
+#include "src/common/xqc_log.h"
+
+/*
+ * default log threshold of number of element in on bucket
+ * test result of max conflict (default 1024*1024 hash buckets) in one bucket:
+ * 100000 connections, the max conflict is 5
+ * 1000000 connections, the max conflict is 24
+ */
+#define XQC_HASH_DEFAULT_CONFLICT_THRESHOLD     50
+/*10 second, log interval must not less then 10 second*/
+#define XQC_HASH_CONFLICT_LOG_INTERVAL  10
 
 typedef struct xqc_str_hash_element_s {
     uint64_t    hash;
@@ -24,20 +37,79 @@ typedef struct xqc_str_hash_table_s {
     xqc_str_hash_node_t   **list;
     size_t                  count;
     xqc_allocator_t         allocator;  /* memory allocator */
+    uint8_t                *conflict_stat;  /* statistic the number of elements in every bucket */
+    xqc_siphash_ctx_t       siphash_ctx;    /* siphash context */
+    uint32_t                conflict_thres; /* conflict threshold in one bucket, warning if exceeded */
+    time_t                  last_log_time;  /* last timestamp(second) for logging the max conflict value*/
+    xqc_log_t              *log;            /* pointer to engine's log*/
 } xqc_str_hash_table_t;
 
+/* calculate the hash value using the siphash algorithm */
+static inline uint64_t
+xqc_siphash_get_hash(xqc_siphash_ctx_t *ctx, const uint8_t *data, size_t len)
+{
+    uint64_t hash_value;
+    if (xqc_siphash(ctx, data, len, (uint8_t *)(&hash_value), sizeof(hash_value)) == XQC_OK) {
+        return hash_value;
+    }
+    /*
+     * impossible, we set hash_size value 8 when we call xqc_siphash_init, sizeof(hash_value) is always 8
+     * xqc_siphash return XQC_ERROR only when hash_size not equal sizeof(hash_value), it is impossible here
+     */
+    return 0;
+}
 
 static inline int
-xqc_str_hash_init(xqc_str_hash_table_t *hash_tab,  xqc_allocator_t allocator, size_t bucket_num)
+xqc_str_hash_init(xqc_str_hash_table_t *hash_tab,
+                  xqc_allocator_t allocator, size_t bucket_num,
+                  uint32_t conflict_thres, uint8_t *key,
+                  size_t key_len, xqc_log_t *log)
 {
+    if (bucket_num == 0) { /* impossible */
+        return XQC_ERROR;
+    }
+    if (key_len != XQC_SIPHASH_KEY_SIZE) { /* siphash key length must be 16 */
+        return XQC_ERROR;
+    }
+    if (log == NULL) {
+        return XQC_ERROR;
+    }
+    xqc_memzero(hash_tab, sizeof(xqc_str_hash_table_t));
     hash_tab->allocator = allocator;
     hash_tab->list = allocator.malloc(allocator.opaque, sizeof(xqc_str_hash_node_t *) * bucket_num);
     if (hash_tab->list == NULL) {
         return XQC_ERROR;
     }
     xqc_memzero(hash_tab->list, sizeof(xqc_str_hash_node_t *) * bucket_num);
     hash_tab->count = bucket_num;
+    hash_tab->conflict_stat = allocator.malloc(allocator.opaque, sizeof(uint8_t) * bucket_num);
+    if (hash_tab->conflict_stat == NULL) {
+        goto fail;
+    }
+    xqc_memzero(hash_tab->conflict_stat, sizeof(uint8_t) * bucket_num);
+    if (conflict_thres > 0) {
+        hash_tab->conflict_thres = conflict_thres;
+    } else {
+        hash_tab->conflict_thres = XQC_HASH_DEFAULT_CONFLICT_THRESHOLD;
+    }
+    hash_tab->last_log_time = 0;
+    hash_tab->log = log;
+    if (xqc_siphash_init(&hash_tab->siphash_ctx, key, key_len,
+                         XQC_DEFAULT_HASH_SIZE, XQC_SIPHASH_C_ROUNDS,
+                         XQC_SIPHASH_D_ROUNDS) != XQC_OK)
+    {
+        goto fail;
+    }
     return XQC_OK;
+
+fail:
+    if (hash_tab->list) {
+        allocator.free(allocator.opaque, hash_tab->list);
+    }
+    if (hash_tab->conflict_stat) {
+        allocator.free(allocator.opaque, hash_tab->conflict_stat);
+    }
+    return XQC_ERROR;
 }
 
 static inline void
@@ -53,6 +125,7 @@ xqc_str_hash_release(xqc_str_hash_table_t *hash_tab)
         }
     }
     a->free(a->opaque, hash_tab->list);
+    a->free(a->opaque, hash_tab->conflict_stat);
 }
 
 static inline void *
@@ -90,6 +163,16 @@ xqc_str_hash_add(xqc_str_hash_table_t *hash_tab, xqc_str_hash_element_t e)
 
     node->next = hash_tab->list[index];
     hash_tab->list[index] = node;
+    hash_tab->conflict_stat[index] += 1;
+    if (hash_tab->conflict_stat[index] > hash_tab->conflict_thres) {
+        time_t now_sec = time(NULL);
+        if (now_sec >= hash_tab->last_log_time + XQC_HASH_CONFLICT_LOG_INTERVAL) {
+            xqc_log(hash_tab->log, XQC_LOG_WARN,
+                    "|xqc conn hash conflict exceed|index:%ui, number of elements:%d|",
+                    index, hash_tab->conflict_stat[index]);
+            hash_tab->last_log_time = now_sec;
+        }
+    }
 
     return XQC_OK;
 }
@@ -104,6 +187,9 @@ xqc_str_hash_delete(xqc_str_hash_table_t *hash_tab, uint64_t hash, xqc_str_t str
     while (node) {
         if (node->element.hash == hash && xqc_str_equal(str, node->element.str)) {
             *pp = node->next;
+            if (hash_tab->conflict_stat[index] > 0) {
+                hash_tab->conflict_stat[index] -= 1;
+            }
             a->free(a->opaque, node->element.str.data);
             a->free(a->opaque, node);
             return XQC_OK;