Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 1 | /** |
| 2 | * @file hash_table.c |
| 3 | * @author Radek Krejci <rkrejci@cesnet.cz> |
| 4 | * @brief libyang dictionary for storing strings and generic hash table |
| 5 | * |
| 6 | * Copyright (c) 2015 - 2018 CESNET, z.s.p.o. |
| 7 | * |
| 8 | * This source code is licensed under BSD 3-Clause License (the "License"). |
| 9 | * You may not use this file except in compliance with the License. |
| 10 | * You may obtain a copy of the License at |
| 11 | * |
| 12 | * https://opensource.org/licenses/BSD-3-Clause |
| 13 | */ |
| 14 | |
Radek Krejci | 535ea9f | 2020-05-29 16:01:05 +0200 | [diff] [blame] | 15 | #include "hash_table.h" |
Radek Krejci | e7b9509 | 2019-05-15 11:03:07 +0200 | [diff] [blame] | 16 | |
Michal Vasko | 6973015 | 2020-10-09 16:30:07 +0200 | [diff] [blame] | 17 | #include <assert.h> |
| 18 | #include <pthread.h> |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 19 | #include <stdint.h> |
| 20 | #include <stdlib.h> |
Michal Vasko | 6973015 | 2020-10-09 16:30:07 +0200 | [diff] [blame] | 21 | #include <string.h> |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 22 | |
Radek Krejci | 535ea9f | 2020-05-29 16:01:05 +0200 | [diff] [blame] | 23 | #include "common.h" |
Michal Vasko | 6973015 | 2020-10-09 16:30:07 +0200 | [diff] [blame] | 24 | #include "compat.h" |
Radek Krejci | 535ea9f | 2020-05-29 16:01:05 +0200 | [diff] [blame] | 25 | #include "dict.h" |
Radek Krejci | 857189e | 2020-09-01 13:26:36 +0200 | [diff] [blame] | 26 | #include "log.h" |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 27 | |
Radek Krejci | f13b87b | 2020-12-01 22:02:17 +0100 | [diff] [blame] | 28 | #define LYDICT_MIN_SIZE 1024 |
| 29 | |
Radek Krejci | 857189e | 2020-09-01 13:26:36 +0200 | [diff] [blame] | 30 | /** |
| 31 | * @brief Comparison callback for dictionary's hash table |
| 32 | * |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 33 | * Implementation of ::lyht_value_equal_cb. |
Radek Krejci | 857189e | 2020-09-01 13:26:36 +0200 | [diff] [blame] | 34 | */ |
| 35 | static ly_bool |
| 36 | lydict_val_eq(void *val1_p, void *val2_p, ly_bool UNUSED(mod), void *cb_data) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 37 | { |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 38 | LY_CHECK_ARG_RET(NULL, val1_p, val2_p, cb_data, 0); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 39 | |
| 40 | const char *str1 = ((struct dict_rec *)val1_p)->value; |
| 41 | const char *str2 = ((struct dict_rec *)val2_p)->value; |
| 42 | |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 43 | LY_CHECK_ERR_RET(!str1, LOGARG(NULL, val1_p), 0); |
| 44 | LY_CHECK_ERR_RET(!str2, LOGARG(NULL, val2_p), 0); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 45 | |
| 46 | if (strncmp(str1, str2, *(size_t *)cb_data) == 0) { |
| 47 | return 1; |
| 48 | } |
| 49 | |
| 50 | return 0; |
| 51 | } |
| 52 | |
| 53 | void |
| 54 | lydict_init(struct dict_table *dict) |
| 55 | { |
Michal Vasko | d989ba0 | 2020-08-24 10:59:24 +0200 | [diff] [blame] | 56 | LY_CHECK_ARG_RET(NULL, dict, ); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 57 | |
Radek Krejci | f13b87b | 2020-12-01 22:02:17 +0100 | [diff] [blame] | 58 | dict->hash_tab = lyht_new(LYDICT_MIN_SIZE, sizeof(struct dict_rec), lydict_val_eq, NULL, 1); |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 59 | LY_CHECK_ERR_RET(!dict->hash_tab, LOGINT(NULL), ); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 60 | pthread_mutex_init(&dict->lock, NULL); |
| 61 | } |
| 62 | |
| 63 | void |
| 64 | lydict_clean(struct dict_table *dict) |
| 65 | { |
Michal Vasko | 44f3d2c | 2020-08-24 09:49:38 +0200 | [diff] [blame] | 66 | struct dict_rec *dict_rec = NULL; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 67 | struct ht_rec *rec = NULL; |
| 68 | |
Michal Vasko | d989ba0 | 2020-08-24 10:59:24 +0200 | [diff] [blame] | 69 | LY_CHECK_ARG_RET(NULL, dict, ); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 70 | |
Radek Krejci | 1deb5be | 2020-08-26 16:43:36 +0200 | [diff] [blame] | 71 | for (uint32_t i = 0; i < dict->hash_tab->size; i++) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 72 | /* get ith record */ |
| 73 | rec = (struct ht_rec *)&dict->hash_tab->recs[i * dict->hash_tab->rec_size]; |
| 74 | if (rec->hits == 1) { |
| 75 | /* |
| 76 | * this should not happen, all records inserted into |
| 77 | * dictionary are supposed to be removed using lydict_remove() |
| 78 | * before calling lydict_clean() |
| 79 | */ |
Michal Vasko | 44f3d2c | 2020-08-24 09:49:38 +0200 | [diff] [blame] | 80 | dict_rec = (struct dict_rec *)rec->val; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 81 | LOGWRN(NULL, "String \"%s\" not freed from the dictionary, refcount %d", dict_rec->value, dict_rec->refcount); |
| 82 | /* if record wasn't removed before free string allocated for that record */ |
| 83 | #ifdef NDEBUG |
| 84 | free(dict_rec->value); |
| 85 | #endif |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | /* free table and destroy mutex */ |
| 90 | lyht_free(dict->hash_tab); |
| 91 | pthread_mutex_destroy(&dict->lock); |
| 92 | } |
| 93 | |
| 94 | /* |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 95 | * Usage: |
| 96 | * - init hash to 0 |
| 97 | * - repeatedly call dict_hash_multi(), provide hash from the last call |
| 98 | * - call dict_hash_multi() with key_part = NULL to finish the hash |
| 99 | */ |
| 100 | uint32_t |
| 101 | dict_hash_multi(uint32_t hash, const char *key_part, size_t len) |
| 102 | { |
| 103 | uint32_t i; |
| 104 | |
aPiecek | 4f07c3e | 2021-06-11 10:53:07 +0200 | [diff] [blame] | 105 | if (key_part && len) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 106 | for (i = 0; i < len; ++i) { |
| 107 | hash += key_part[i]; |
| 108 | hash += (hash << 10); |
| 109 | hash ^= (hash >> 6); |
| 110 | } |
| 111 | } else { |
| 112 | hash += (hash << 3); |
| 113 | hash ^= (hash >> 11); |
| 114 | hash += (hash << 15); |
| 115 | } |
| 116 | |
| 117 | return hash; |
| 118 | } |
| 119 | |
Radek Krejci | f2dc4c5 | 2018-11-08 09:04:13 +0100 | [diff] [blame] | 120 | /* |
| 121 | * Bob Jenkin's one-at-a-time hash |
| 122 | * http://www.burtleburtle.net/bob/hash/doobs.html |
| 123 | * |
| 124 | * Spooky hash is faster, but it works only for little endian architectures. |
| 125 | */ |
| 126 | uint32_t |
| 127 | dict_hash(const char *key, size_t len) |
| 128 | { |
| 129 | uint32_t hash; |
| 130 | |
| 131 | hash = dict_hash_multi(0, key, len); |
| 132 | return dict_hash_multi(hash, NULL, len); |
| 133 | } |
| 134 | |
Michal Vasko | 5bcc33b | 2020-10-06 15:33:44 +0200 | [diff] [blame] | 135 | static ly_bool |
| 136 | lydict_resize_val_eq(void *val1_p, void *val2_p, ly_bool mod, void *cb_data) |
| 137 | { |
| 138 | LY_CHECK_ARG_RET(NULL, val1_p, val2_p, 0); |
| 139 | |
| 140 | const char *str1 = ((struct dict_rec *)val1_p)->value; |
| 141 | const char *str2 = ((struct dict_rec *)val2_p)->value; |
| 142 | |
| 143 | LY_CHECK_ERR_RET(!str1, LOGARG(NULL, val1_p), 0); |
| 144 | LY_CHECK_ERR_RET(!str2, LOGARG(NULL, val2_p), 0); |
| 145 | |
| 146 | if (mod) { |
| 147 | /* used when inserting new values */ |
| 148 | if (strcmp(str1, str2) == 0) { |
| 149 | return 1; |
| 150 | } |
| 151 | } else { |
| 152 | /* used when finding the original value again in the resized table */ |
| 153 | return lydict_val_eq(val1_p, val2_p, mod, cb_data); |
| 154 | } |
| 155 | |
| 156 | return 0; |
| 157 | } |
| 158 | |
Jan Kundrát | c53a7ec | 2021-12-09 16:01:19 +0100 | [diff] [blame] | 159 | LIBYANG_API_DEF LY_ERR |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 160 | lydict_remove(const struct ly_ctx *ctx, const char *value) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 161 | { |
Michal Vasko | e180ed0 | 2021-02-05 16:31:20 +0100 | [diff] [blame] | 162 | LY_ERR ret = LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 163 | size_t len; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 164 | uint32_t hash; |
| 165 | struct dict_rec rec, *match = NULL; |
| 166 | char *val_p; |
| 167 | |
Michal Vasko | 016569c | 2021-03-04 15:50:02 +0100 | [diff] [blame] | 168 | if (!ctx || !value) { |
Michal Vasko | e180ed0 | 2021-02-05 16:31:20 +0100 | [diff] [blame] | 169 | return LY_SUCCESS; |
Michal Vasko | 0f6b3e2 | 2018-09-07 12:18:12 +0200 | [diff] [blame] | 170 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 171 | |
Michal Vasko | e180ed0 | 2021-02-05 16:31:20 +0100 | [diff] [blame] | 172 | LOGDBG(LY_LDGDICT, "removing \"%s\"", value); |
| 173 | |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 174 | len = strlen(value); |
| 175 | hash = dict_hash(value, len); |
| 176 | |
| 177 | /* create record for lyht_find call */ |
| 178 | rec.value = (char *)value; |
| 179 | rec.refcount = 0; |
| 180 | |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 181 | pthread_mutex_lock((pthread_mutex_t *)&ctx->dict.lock); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 182 | /* set len as data for compare callback */ |
| 183 | lyht_set_cb_data(ctx->dict.hash_tab, (void *)&len); |
| 184 | /* check if value is already inserted */ |
| 185 | ret = lyht_find(ctx->dict.hash_tab, &rec, hash, (void **)&match); |
| 186 | |
Radek Krejci | 1deb5be | 2020-08-26 16:43:36 +0200 | [diff] [blame] | 187 | if (ret == LY_SUCCESS) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 188 | LY_CHECK_ERR_GOTO(!match, LOGINT(ctx), finish); |
| 189 | |
| 190 | /* if value is already in dictionary, decrement reference counter */ |
| 191 | match->refcount--; |
| 192 | if (match->refcount == 0) { |
| 193 | /* |
| 194 | * remove record |
| 195 | * save pointer to stored string before lyht_remove to |
| 196 | * free it after it is removed from hash table |
| 197 | */ |
| 198 | val_p = match->value; |
Michal Vasko | 5bcc33b | 2020-10-06 15:33:44 +0200 | [diff] [blame] | 199 | ret = lyht_remove_with_resize_cb(ctx->dict.hash_tab, &rec, hash, lydict_resize_val_eq); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 200 | free(val_p); |
| 201 | LY_CHECK_ERR_GOTO(ret, LOGINT(ctx), finish); |
| 202 | } |
Michal Vasko | e180ed0 | 2021-02-05 16:31:20 +0100 | [diff] [blame] | 203 | } else if (ret == LY_ENOTFOUND) { |
| 204 | LOGERR(ctx, LY_ENOTFOUND, "Value \"%s\" was not found in the dictionary.", value); |
| 205 | } else { |
| 206 | LOGINT(ctx); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 207 | } |
| 208 | |
| 209 | finish: |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 210 | pthread_mutex_unlock((pthread_mutex_t *)&ctx->dict.lock); |
Michal Vasko | e180ed0 | 2021-02-05 16:31:20 +0100 | [diff] [blame] | 211 | return ret; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 212 | } |
| 213 | |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 214 | LY_ERR |
| 215 | dict_insert(const struct ly_ctx *ctx, char *value, size_t len, ly_bool zerocopy, const char **str_p) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 216 | { |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 217 | LY_ERR ret = LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 218 | struct dict_rec *match = NULL, rec; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 219 | uint32_t hash; |
| 220 | |
Radek Krejci | 422afb1 | 2021-03-04 16:38:16 +0100 | [diff] [blame] | 221 | LOGDBG(LY_LDGDICT, "inserting \"%.*s\"", (int)len, value); |
Michal Vasko | e180ed0 | 2021-02-05 16:31:20 +0100 | [diff] [blame] | 222 | |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 223 | hash = dict_hash(value, len); |
| 224 | /* set len as data for compare callback */ |
| 225 | lyht_set_cb_data(ctx->dict.hash_tab, (void *)&len); |
| 226 | /* create record for lyht_insert */ |
| 227 | rec.value = value; |
| 228 | rec.refcount = 1; |
| 229 | |
Michal Vasko | 5bcc33b | 2020-10-06 15:33:44 +0200 | [diff] [blame] | 230 | ret = lyht_insert_with_resize_cb(ctx->dict.hash_tab, (void *)&rec, hash, lydict_resize_val_eq, (void **)&match); |
Michal Vasko | fcc96b9 | 2018-09-12 11:12:01 +0200 | [diff] [blame] | 231 | if (ret == LY_EEXIST) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 232 | match->refcount++; |
| 233 | if (zerocopy) { |
| 234 | free(value); |
| 235 | } |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 236 | ret = LY_SUCCESS; |
Michal Vasko | fcc96b9 | 2018-09-12 11:12:01 +0200 | [diff] [blame] | 237 | } else if (ret == LY_SUCCESS) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 238 | if (!zerocopy) { |
| 239 | /* |
| 240 | * allocate string for new record |
| 241 | * record is already inserted in hash table |
| 242 | */ |
| 243 | match->value = malloc(sizeof *match->value * (len + 1)); |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 244 | LY_CHECK_ERR_RET(!match->value, LOGMEM(ctx), LY_EMEM); |
Michal Vasko | 08e9b11 | 2021-06-11 15:41:17 +0200 | [diff] [blame] | 245 | if (len) { |
| 246 | memcpy(match->value, value, len); |
| 247 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 248 | match->value[len] = '\0'; |
| 249 | } |
| 250 | } else { |
| 251 | /* lyht_insert returned error */ |
David Sedlák | 234a91f | 2019-08-15 13:16:43 +0200 | [diff] [blame] | 252 | if (zerocopy) { |
| 253 | free(value); |
| 254 | } |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 255 | return ret; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 256 | } |
| 257 | |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 258 | if (str_p) { |
| 259 | *str_p = match->value; |
| 260 | } |
| 261 | |
| 262 | return ret; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 263 | } |
| 264 | |
Jan Kundrát | c53a7ec | 2021-12-09 16:01:19 +0100 | [diff] [blame] | 265 | LIBYANG_API_DEF LY_ERR |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 266 | lydict_insert(const struct ly_ctx *ctx, const char *value, size_t len, const char **str_p) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 267 | { |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 268 | LY_ERR result; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 269 | |
Michal Vasko | be27c91 | 2021-02-24 17:07:37 +0100 | [diff] [blame] | 270 | LY_CHECK_ARG_RET(ctx, ctx, str_p, LY_EINVAL); |
| 271 | |
| 272 | if (!value) { |
| 273 | *str_p = NULL; |
| 274 | return LY_SUCCESS; |
| 275 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 276 | |
Radek Krejci | 0ae092d | 2018-09-20 16:43:19 +0200 | [diff] [blame] | 277 | if (!len) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 278 | len = strlen(value); |
| 279 | } |
| 280 | |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 281 | pthread_mutex_lock((pthread_mutex_t *)&ctx->dict.lock); |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 282 | result = dict_insert(ctx, (char *)value, len, 0, str_p); |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 283 | pthread_mutex_unlock((pthread_mutex_t *)&ctx->dict.lock); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 284 | |
| 285 | return result; |
| 286 | } |
| 287 | |
Jan Kundrát | c53a7ec | 2021-12-09 16:01:19 +0100 | [diff] [blame] | 288 | LIBYANG_API_DEF LY_ERR |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 289 | lydict_insert_zc(const struct ly_ctx *ctx, char *value, const char **str_p) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 290 | { |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 291 | LY_ERR result; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 292 | |
Michal Vasko | be27c91 | 2021-02-24 17:07:37 +0100 | [diff] [blame] | 293 | LY_CHECK_ARG_RET(ctx, ctx, str_p, LY_EINVAL); |
| 294 | |
| 295 | if (!value) { |
| 296 | *str_p = NULL; |
| 297 | return LY_SUCCESS; |
| 298 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 299 | |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 300 | pthread_mutex_lock((pthread_mutex_t *)&ctx->dict.lock); |
Radek Krejci | 011e4aa | 2020-09-04 15:22:31 +0200 | [diff] [blame] | 301 | result = dict_insert(ctx, value, strlen(value), 1, str_p); |
Michal Vasko | 52927e2 | 2020-03-16 17:26:14 +0100 | [diff] [blame] | 302 | pthread_mutex_unlock((pthread_mutex_t *)&ctx->dict.lock); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 303 | |
| 304 | return result; |
| 305 | } |
| 306 | |
Radek Krejci | 2d7a47b | 2019-05-16 13:34:10 +0200 | [diff] [blame] | 307 | struct ht_rec * |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 308 | lyht_get_rec(unsigned char *recs, uint16_t rec_size, uint32_t idx) |
| 309 | { |
| 310 | return (struct ht_rec *)&recs[idx * rec_size]; |
| 311 | } |
| 312 | |
| 313 | struct hash_table * |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 314 | lyht_new(uint32_t size, uint16_t val_size, lyht_value_equal_cb val_equal, void *cb_data, uint16_t resize) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 315 | { |
| 316 | struct hash_table *ht; |
| 317 | |
| 318 | /* check that 2^x == size (power of 2) */ |
| 319 | assert(size && !(size & (size - 1))); |
| 320 | assert(val_equal && val_size); |
| 321 | assert(resize == 0 || resize == 1); |
| 322 | |
| 323 | if (size < LYHT_MIN_SIZE) { |
| 324 | size = LYHT_MIN_SIZE; |
| 325 | } |
| 326 | |
| 327 | ht = malloc(sizeof *ht); |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 328 | LY_CHECK_ERR_RET(!ht, LOGMEM(NULL), NULL); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 329 | |
| 330 | ht->used = 0; |
| 331 | ht->size = size; |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 332 | ht->invalid = 0; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 333 | ht->val_equal = val_equal; |
| 334 | ht->cb_data = cb_data; |
Radek Krejci | 1deb5be | 2020-08-26 16:43:36 +0200 | [diff] [blame] | 335 | ht->resize = resize; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 336 | |
| 337 | ht->rec_size = (sizeof(struct ht_rec) - 1) + val_size; |
| 338 | /* allocate the records correctly */ |
| 339 | ht->recs = calloc(size, ht->rec_size); |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 340 | LY_CHECK_ERR_RET(!ht->recs, free(ht); LOGMEM(NULL), NULL); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 341 | |
| 342 | return ht; |
| 343 | } |
| 344 | |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 345 | lyht_value_equal_cb |
| 346 | lyht_set_cb(struct hash_table *ht, lyht_value_equal_cb new_val_equal) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 347 | { |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 348 | lyht_value_equal_cb prev; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 349 | |
| 350 | prev = ht->val_equal; |
| 351 | ht->val_equal = new_val_equal; |
| 352 | return prev; |
| 353 | } |
| 354 | |
| 355 | void * |
| 356 | lyht_set_cb_data(struct hash_table *ht, void *new_cb_data) |
| 357 | { |
| 358 | void *prev; |
| 359 | |
| 360 | prev = ht->cb_data; |
| 361 | ht->cb_data = new_cb_data; |
| 362 | return prev; |
| 363 | } |
| 364 | |
| 365 | struct hash_table * |
| 366 | lyht_dup(const struct hash_table *orig) |
| 367 | { |
| 368 | struct hash_table *ht; |
| 369 | |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 370 | LY_CHECK_ARG_RET(NULL, orig, NULL); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 371 | |
| 372 | ht = lyht_new(orig->size, orig->rec_size - (sizeof(struct ht_rec) - 1), orig->val_equal, orig->cb_data, orig->resize ? 1 : 0); |
| 373 | if (!ht) { |
| 374 | return NULL; |
| 375 | } |
| 376 | |
Radek Krejci | d78fd46 | 2021-06-02 21:29:20 +0200 | [diff] [blame] | 377 | memcpy(ht->recs, orig->recs, (size_t)orig->used * (size_t)orig->rec_size); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 378 | ht->used = orig->used; |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 379 | ht->invalid = orig->invalid; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 380 | return ht; |
| 381 | } |
| 382 | |
| 383 | void |
| 384 | lyht_free(struct hash_table *ht) |
| 385 | { |
| 386 | if (ht) { |
| 387 | free(ht->recs); |
| 388 | free(ht); |
| 389 | } |
| 390 | } |
| 391 | |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 392 | /** |
| 393 | * @brief Resize a hash table. |
| 394 | * |
| 395 | * @param[in] ht Hash table to resize. |
| 396 | * @param[in] operation Operation to perform. 1 to enlarge, -1 to shrink, 0 to only rehash all records. |
| 397 | * @return LY_ERR value. |
| 398 | */ |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 399 | static LY_ERR |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 400 | lyht_resize(struct hash_table *ht, int operation) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 401 | { |
| 402 | struct ht_rec *rec; |
| 403 | unsigned char *old_recs; |
| 404 | uint32_t i, old_size; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 405 | |
| 406 | old_recs = ht->recs; |
| 407 | old_size = ht->size; |
| 408 | |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 409 | if (operation > 0) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 410 | /* double the size */ |
| 411 | ht->size <<= 1; |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 412 | } else if (operation < 0) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 413 | /* half the size */ |
| 414 | ht->size >>= 1; |
| 415 | } |
| 416 | |
| 417 | ht->recs = calloc(ht->size, ht->rec_size); |
Michal Vasko | b3d0d6b | 2018-09-07 10:17:33 +0200 | [diff] [blame] | 418 | LY_CHECK_ERR_RET(!ht->recs, LOGMEM(NULL); ht->recs = old_recs; ht->size = old_size, LY_EMEM); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 419 | |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 420 | /* reset used and invalid, it will increase again */ |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 421 | ht->used = 0; |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 422 | ht->invalid = 0; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 423 | |
| 424 | /* add all the old records into the new records array */ |
| 425 | for (i = 0; i < old_size; ++i) { |
| 426 | rec = lyht_get_rec(old_recs, ht->rec_size, i); |
| 427 | if (rec->hits > 0) { |
Radek Krejci | 1deb5be | 2020-08-26 16:43:36 +0200 | [diff] [blame] | 428 | LY_ERR ret = lyht_insert(ht, rec->val, rec->hash, NULL); |
Michal Vasko | 26bbb27 | 2022-08-02 14:54:33 +0200 | [diff] [blame] | 429 | |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 430 | assert(!ret); |
| 431 | (void)ret; |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | /* final touches */ |
| 436 | free(old_recs); |
| 437 | return LY_SUCCESS; |
| 438 | } |
| 439 | |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 440 | /** |
| 441 | * @brief Search for the first match. |
| 442 | * |
| 443 | * @param[in] ht Hash table to search in. |
| 444 | * @param[in] hash Hash to find. |
| 445 | * @param[out] rec_p Optional found record. |
| 446 | * @return LY_SUCCESS hash found, returned its record, |
| 447 | * @return LY_ENOTFOUND hash not found, returned the record where it would be inserted. |
| 448 | */ |
| 449 | static LY_ERR |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 450 | lyht_find_first(struct hash_table *ht, uint32_t hash, struct ht_rec **rec_p) |
| 451 | { |
| 452 | struct ht_rec *rec; |
| 453 | uint32_t i, idx; |
| 454 | |
| 455 | if (rec_p) { |
| 456 | *rec_p = NULL; |
| 457 | } |
| 458 | |
| 459 | idx = i = hash & (ht->size - 1); |
| 460 | rec = lyht_get_rec(ht->recs, ht->rec_size, idx); |
| 461 | |
| 462 | /* skip through overflow and deleted records */ |
| 463 | while ((rec->hits != 0) && ((rec->hits == -1) || ((rec->hash & (ht->size - 1)) != idx))) { |
| 464 | if ((rec->hits == -1) && rec_p && !(*rec_p)) { |
| 465 | /* remember this record for return */ |
| 466 | *rec_p = rec; |
| 467 | } |
| 468 | i = (i + 1) % ht->size; |
| 469 | if (i == idx) { |
| 470 | /* we went through all the records (very unlikely, but possible when many records are invalid), |
| 471 | * just return not found */ |
| 472 | assert(!rec_p || *rec_p); |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 473 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 474 | } |
| 475 | rec = lyht_get_rec(ht->recs, ht->rec_size, i); |
| 476 | } |
| 477 | if (rec->hits == 0) { |
| 478 | /* we could not find the value */ |
| 479 | if (rec_p && !*rec_p) { |
| 480 | *rec_p = rec; |
| 481 | } |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 482 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 483 | } |
| 484 | |
| 485 | /* we have found a record with equal (shortened) hash */ |
| 486 | if (rec_p) { |
| 487 | *rec_p = rec; |
| 488 | } |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 489 | return LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 490 | } |
| 491 | |
| 492 | /** |
| 493 | * @brief Search for the next collision. |
| 494 | * |
| 495 | * @param[in] ht Hash table to search in. |
| 496 | * @param[in,out] last Last returned collision record. |
| 497 | * @param[in] first First collision record (hits > 1). |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 498 | * @return LY_SUCCESS when hash collision found, \p last points to this next collision, |
| 499 | * @return LY_ENOTFOUND when hash collision not found, \p last points to the record where it would be inserted. |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 500 | */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 501 | static LY_ERR |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 502 | lyht_find_collision(struct hash_table *ht, struct ht_rec **last, struct ht_rec *first) |
| 503 | { |
| 504 | struct ht_rec *empty = NULL; |
| 505 | uint32_t i, idx; |
| 506 | |
| 507 | assert(last && *last); |
| 508 | |
| 509 | idx = (*last)->hash & (ht->size - 1); |
| 510 | i = (((unsigned char *)*last) - ht->recs) / ht->rec_size; |
| 511 | |
| 512 | do { |
| 513 | i = (i + 1) % ht->size; |
| 514 | *last = lyht_get_rec(ht->recs, ht->rec_size, i); |
| 515 | if (*last == first) { |
| 516 | /* we went through all the records (very unlikely, but possible when many records are invalid), |
| 517 | * just return an invalid record */ |
| 518 | assert(empty); |
| 519 | *last = empty; |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 520 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 521 | } |
| 522 | |
| 523 | if (((*last)->hits == -1) && !empty) { |
| 524 | empty = *last; |
| 525 | } |
| 526 | } while (((*last)->hits != 0) && (((*last)->hits == -1) || (((*last)->hash & (ht->size - 1)) != idx))); |
| 527 | |
| 528 | if ((*last)->hits > 0) { |
| 529 | /* we found a collision */ |
| 530 | assert((*last)->hits == 1); |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 531 | return LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 532 | } |
| 533 | |
| 534 | /* no next collision found, return the record where it would be inserted */ |
| 535 | if (empty) { |
| 536 | *last = empty; |
| 537 | } /* else (*last)->hits == 0, it is already correct */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 538 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 539 | } |
| 540 | |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 541 | /** |
| 542 | * @brief Search for a record with specific value and hash. |
| 543 | * |
| 544 | * @param[in] ht Hash table to search in. |
| 545 | * @param[in] val_p Pointer to the value to find. |
| 546 | * @param[in] hash Hash to find. |
Radek Krejci | 857189e | 2020-09-01 13:26:36 +0200 | [diff] [blame] | 547 | * @param[in] mod Whether the operation modifies the hash table (insert or remove) or not (find). |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 548 | * @param[out] crec_p Optional found first record. |
| 549 | * @param[out] col Optional collision number of @p rec_p, 0 for no collision. |
| 550 | * @param[out] rec_p Found exact matching record, may be a collision of @p crec_p. |
| 551 | * @return LY_ENOTFOUND if no record found, |
| 552 | * @return LY_SUCCESS if record was found. |
| 553 | */ |
| 554 | static LY_ERR |
Radek Krejci | 857189e | 2020-09-01 13:26:36 +0200 | [diff] [blame] | 555 | lyht_find_rec(struct hash_table *ht, void *val_p, uint32_t hash, ly_bool mod, struct ht_rec **crec_p, uint32_t *col, |
Radek Krejci | 0f96988 | 2020-08-21 16:56:47 +0200 | [diff] [blame] | 556 | struct ht_rec **rec_p) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 557 | { |
| 558 | struct ht_rec *rec, *crec; |
| 559 | uint32_t i, c; |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 560 | LY_ERR r; |
| 561 | |
| 562 | if (crec_p) { |
| 563 | *crec_p = NULL; |
| 564 | } |
| 565 | if (col) { |
| 566 | *col = 0; |
| 567 | } |
| 568 | *rec_p = NULL; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 569 | |
| 570 | if (lyht_find_first(ht, hash, &rec)) { |
| 571 | /* not found */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 572 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 573 | } |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 574 | if ((rec->hash == hash) && ht->val_equal(val_p, &rec->val, mod, ht->cb_data)) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 575 | /* even the value matches */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 576 | if (crec_p) { |
| 577 | *crec_p = rec; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 578 | } |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 579 | if (col) { |
| 580 | *col = 0; |
| 581 | } |
| 582 | *rec_p = rec; |
| 583 | return LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 584 | } |
| 585 | |
| 586 | /* some collisions, we need to go through them, too */ |
| 587 | crec = rec; |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 588 | c = crec->hits; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 589 | for (i = 1; i < c; ++i) { |
| 590 | r = lyht_find_collision(ht, &rec, crec); |
| 591 | assert(!r); |
| 592 | (void)r; |
| 593 | |
| 594 | /* compare values */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 595 | if ((rec->hash == hash) && ht->val_equal(val_p, &rec->val, mod, ht->cb_data)) { |
| 596 | if (crec_p) { |
| 597 | *crec_p = crec; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 598 | } |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 599 | if (col) { |
| 600 | *col = i; |
| 601 | } |
| 602 | *rec_p = rec; |
| 603 | return LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 604 | } |
| 605 | } |
| 606 | |
| 607 | /* not found even in collisions */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 608 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 609 | } |
| 610 | |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 611 | LY_ERR |
| 612 | lyht_find(struct hash_table *ht, void *val_p, uint32_t hash, void **match_p) |
| 613 | { |
| 614 | struct ht_rec *rec; |
| 615 | |
| 616 | lyht_find_rec(ht, val_p, hash, 0, NULL, NULL, &rec); |
| 617 | |
| 618 | if (rec && match_p) { |
| 619 | *match_p = rec->val; |
| 620 | } |
| 621 | return rec ? LY_SUCCESS : LY_ENOTFOUND; |
| 622 | } |
| 623 | |
| 624 | LY_ERR |
Michal Vasko | 6374de2 | 2022-09-05 15:48:48 +0200 | [diff] [blame^] | 625 | lyht_find_next_with_collision_cb(struct hash_table *ht, void *val_p, uint32_t hash, |
| 626 | lyht_value_equal_cb collision_val_equal, void **match_p) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 627 | { |
| 628 | struct ht_rec *rec, *crec; |
| 629 | uint32_t i, c; |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 630 | LY_ERR r; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 631 | |
Michal Vasko | 6374de2 | 2022-09-05 15:48:48 +0200 | [diff] [blame^] | 632 | /* find the record of the previously found value */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 633 | if (lyht_find_rec(ht, val_p, hash, 1, &crec, &i, &rec)) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 634 | /* not found, cannot happen */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 635 | LOGINT_RET(NULL); |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 636 | } |
| 637 | |
Michal Vasko | 6374de2 | 2022-09-05 15:48:48 +0200 | [diff] [blame^] | 638 | /* go through collisions and find the next one after the previous one */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 639 | c = crec->hits; |
| 640 | for (++i; i < c; ++i) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 641 | r = lyht_find_collision(ht, &rec, crec); |
| 642 | assert(!r); |
| 643 | (void)r; |
| 644 | |
Michal Vasko | 6374de2 | 2022-09-05 15:48:48 +0200 | [diff] [blame^] | 645 | if (rec->hash != hash) { |
| 646 | continue; |
| 647 | } |
| 648 | |
| 649 | if (collision_val_equal) { |
| 650 | if (collision_val_equal(val_p, &rec->val, 0, ht->cb_data)) { |
| 651 | /* even the value matches */ |
| 652 | if (match_p) { |
| 653 | *match_p = rec->val; |
| 654 | } |
| 655 | return LY_SUCCESS; |
| 656 | } |
| 657 | } else if (ht->val_equal(val_p, &rec->val, 0, ht->cb_data)) { |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 658 | /* even the value matches */ |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 659 | if (match_p) { |
| 660 | *match_p = rec->val; |
| 661 | } |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 662 | return LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 663 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 664 | } |
| 665 | |
| 666 | /* the last equal value was already returned */ |
Michal Vasko | da85903 | 2020-07-14 12:20:14 +0200 | [diff] [blame] | 667 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 668 | } |
| 669 | |
| 670 | LY_ERR |
Michal Vasko | 6374de2 | 2022-09-05 15:48:48 +0200 | [diff] [blame^] | 671 | lyht_find_next(struct hash_table *ht, void *val_p, uint32_t hash, void **match_p) |
| 672 | { |
| 673 | return lyht_find_next_with_collision_cb(ht, val_p, hash, NULL, match_p); |
| 674 | } |
| 675 | |
| 676 | LY_ERR |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 677 | lyht_insert_with_resize_cb(struct hash_table *ht, void *val_p, uint32_t hash, lyht_value_equal_cb resize_val_equal, |
| 678 | void **match_p) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 679 | { |
Radek Krejci | 1deb5be | 2020-08-26 16:43:36 +0200 | [diff] [blame] | 680 | LY_ERR r, ret = LY_SUCCESS; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 681 | struct ht_rec *rec, *crec = NULL; |
| 682 | int32_t i; |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 683 | lyht_value_equal_cb old_val_equal = NULL; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 684 | |
| 685 | if (!lyht_find_first(ht, hash, &rec)) { |
| 686 | /* we found matching shortened hash */ |
| 687 | if ((rec->hash == hash) && ht->val_equal(val_p, &rec->val, 1, ht->cb_data)) { |
| 688 | /* even the value matches */ |
| 689 | if (match_p) { |
| 690 | *match_p = (void *)&rec->val; |
| 691 | } |
| 692 | return LY_EEXIST; |
| 693 | } |
| 694 | |
| 695 | /* some collisions, we need to go through them, too */ |
| 696 | crec = rec; |
| 697 | for (i = 1; i < crec->hits; ++i) { |
| 698 | r = lyht_find_collision(ht, &rec, crec); |
| 699 | assert(!r); |
| 700 | |
| 701 | /* compare values */ |
| 702 | if ((rec->hash == hash) && ht->val_equal(val_p, &rec->val, 1, ht->cb_data)) { |
| 703 | if (match_p) { |
| 704 | *match_p = (void *)&rec->val; |
| 705 | } |
| 706 | return LY_EEXIST; |
| 707 | } |
| 708 | } |
| 709 | |
| 710 | /* value not found, get the record where it will be inserted */ |
| 711 | r = lyht_find_collision(ht, &rec, crec); |
| 712 | assert(r); |
| 713 | } |
| 714 | |
| 715 | /* insert it into the returned record */ |
| 716 | assert(rec->hits < 1); |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 717 | if (rec->hits < 0) { |
| 718 | --ht->invalid; |
| 719 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 720 | rec->hash = hash; |
| 721 | rec->hits = 1; |
| 722 | memcpy(&rec->val, val_p, ht->rec_size - (sizeof(struct ht_rec) - 1)); |
| 723 | if (match_p) { |
| 724 | *match_p = (void *)&rec->val; |
| 725 | } |
| 726 | |
| 727 | if (crec) { |
| 728 | /* there was a collision, increase hits */ |
| 729 | if (crec->hits == INT32_MAX) { |
| 730 | LOGINT(NULL); |
| 731 | } |
| 732 | ++crec->hits; |
| 733 | } |
| 734 | |
| 735 | /* check size & enlarge if needed */ |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 736 | ++ht->used; |
| 737 | if (ht->resize) { |
Radek Krejci | f13b87b | 2020-12-01 22:02:17 +0100 | [diff] [blame] | 738 | r = (ht->used * LYHT_HUNDRED_PERCENTAGE) / ht->size; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 739 | if ((ht->resize == 1) && (r >= LYHT_FIRST_SHRINK_PERCENTAGE)) { |
| 740 | /* enable shrinking */ |
| 741 | ht->resize = 2; |
| 742 | } |
| 743 | if ((ht->resize == 2) && (r >= LYHT_ENLARGE_PERCENTAGE)) { |
| 744 | if (resize_val_equal) { |
| 745 | old_val_equal = lyht_set_cb(ht, resize_val_equal); |
| 746 | } |
| 747 | |
| 748 | /* enlarge */ |
| 749 | ret = lyht_resize(ht, 1); |
| 750 | /* if hash_table was resized, we need to find new matching value */ |
Michal Vasko | 6973015 | 2020-10-09 16:30:07 +0200 | [diff] [blame] | 751 | if ((ret == LY_SUCCESS) && match_p) { |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 752 | lyht_find(ht, val_p, hash, match_p); |
| 753 | } |
| 754 | |
| 755 | if (resize_val_equal) { |
| 756 | lyht_set_cb(ht, old_val_equal); |
| 757 | } |
| 758 | } |
| 759 | } |
| 760 | return ret; |
| 761 | } |
| 762 | |
Radek Krejci | 0ae092d | 2018-09-20 16:43:19 +0200 | [diff] [blame] | 763 | LY_ERR |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 764 | lyht_insert(struct hash_table *ht, void *val_p, uint32_t hash, void **match_p) |
| 765 | { |
| 766 | return lyht_insert_with_resize_cb(ht, val_p, hash, NULL, match_p); |
| 767 | } |
| 768 | |
| 769 | LY_ERR |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 770 | lyht_remove_with_resize_cb(struct hash_table *ht, void *val_p, uint32_t hash, lyht_value_equal_cb resize_val_equal) |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 771 | { |
| 772 | struct ht_rec *rec, *crec; |
| 773 | int32_t i; |
Radek Krejci | 857189e | 2020-09-01 13:26:36 +0200 | [diff] [blame] | 774 | ly_bool first_matched = 0; |
Radek Krejci | 1deb5be | 2020-08-26 16:43:36 +0200 | [diff] [blame] | 775 | LY_ERR r, ret = LY_SUCCESS; |
Michal Vasko | 62524a9 | 2021-02-26 10:08:50 +0100 | [diff] [blame] | 776 | lyht_value_equal_cb old_val_equal; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 777 | |
Michal Vasko | 4a4c7ed | 2020-07-17 09:30:12 +0200 | [diff] [blame] | 778 | LY_CHECK_ERR_RET(lyht_find_first(ht, hash, &rec), LOGARG(NULL, hash), LY_ENOTFOUND); /* hash not found */ |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 779 | |
| 780 | if ((rec->hash == hash) && ht->val_equal(val_p, &rec->val, 1, ht->cb_data)) { |
| 781 | /* even the value matches */ |
| 782 | first_matched = 1; |
| 783 | } |
| 784 | |
| 785 | /* we always need to go through collisions */ |
| 786 | crec = rec; |
| 787 | for (i = 1; i < crec->hits; ++i) { |
| 788 | r = lyht_find_collision(ht, &rec, crec); |
| 789 | assert(!r); |
| 790 | |
| 791 | /* compare values */ |
| 792 | if (!first_matched && (rec->hash == hash) && ht->val_equal(val_p, &rec->val, 1, ht->cb_data)) { |
| 793 | break; |
| 794 | } |
| 795 | } |
| 796 | |
| 797 | if (i < crec->hits) { |
| 798 | /* one of collisions matched, reduce collision count, remove the record */ |
| 799 | assert(!first_matched); |
| 800 | --crec->hits; |
| 801 | rec->hits = -1; |
| 802 | } else if (first_matched) { |
| 803 | /* the first record matches */ |
| 804 | if (crec != rec) { |
| 805 | /* ... so put the last collision in its place */ |
| 806 | rec->hits = crec->hits - 1; |
| 807 | memcpy(crec, rec, ht->rec_size); |
| 808 | } |
| 809 | rec->hits = -1; |
| 810 | } else { |
| 811 | /* value not found even in collisions */ |
Michal Vasko | 4a4c7ed | 2020-07-17 09:30:12 +0200 | [diff] [blame] | 812 | return LY_ENOTFOUND; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 813 | } |
| 814 | |
| 815 | /* check size & shrink if needed */ |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 816 | --ht->used; |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 817 | ++ht->invalid; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 818 | if (ht->resize == 2) { |
Radek Krejci | f13b87b | 2020-12-01 22:02:17 +0100 | [diff] [blame] | 819 | r = (ht->used * LYHT_HUNDRED_PERCENTAGE) / ht->size; |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 820 | if ((r < LYHT_SHRINK_PERCENTAGE) && (ht->size > LYHT_MIN_SIZE)) { |
Michal Vasko | 5bcc33b | 2020-10-06 15:33:44 +0200 | [diff] [blame] | 821 | if (resize_val_equal) { |
| 822 | old_val_equal = lyht_set_cb(ht, resize_val_equal); |
| 823 | } |
| 824 | |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 825 | /* shrink */ |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 826 | ret = lyht_resize(ht, -1); |
Michal Vasko | 5bcc33b | 2020-10-06 15:33:44 +0200 | [diff] [blame] | 827 | |
| 828 | if (resize_val_equal) { |
| 829 | lyht_set_cb(ht, old_val_equal); |
| 830 | } |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 831 | } |
| 832 | } |
| 833 | |
Michal Vasko | dc95d9c | 2021-04-12 15:11:48 +0200 | [diff] [blame] | 834 | /* rehash all records if needed */ |
| 835 | r = ((ht->size - ht->used - ht->invalid) * 100) / ht->size; |
| 836 | if (r < LYHT_REHASH_PERCENTAGE) { |
| 837 | if (resize_val_equal) { |
| 838 | old_val_equal = lyht_set_cb(ht, resize_val_equal); |
| 839 | } |
| 840 | |
| 841 | /* rehash */ |
| 842 | ret = lyht_resize(ht, 0); |
| 843 | |
| 844 | if (resize_val_equal) { |
| 845 | lyht_set_cb(ht, old_val_equal); |
| 846 | } |
| 847 | } |
| 848 | |
Radek Krejci | 5aeea3a | 2018-09-05 13:29:36 +0200 | [diff] [blame] | 849 | return ret; |
| 850 | } |
Michal Vasko | 5bcc33b | 2020-10-06 15:33:44 +0200 | [diff] [blame] | 851 | |
| 852 | LY_ERR |
| 853 | lyht_remove(struct hash_table *ht, void *val_p, uint32_t hash) |
| 854 | { |
| 855 | return lyht_remove_with_resize_cb(ht, val_p, hash, NULL); |
| 856 | } |
Michal Vasko | 626196f | 2022-08-05 12:49:52 +0200 | [diff] [blame] | 857 | |
| 858 | uint32_t |
| 859 | lyht_get_fixed_size(uint32_t item_count) |
| 860 | { |
| 861 | uint32_t i, size = 0; |
| 862 | |
| 863 | /* detect number of upper zero bits in the items' counter value ... */ |
| 864 | for (i = (sizeof item_count * CHAR_BIT) - 1; i > 0; i--) { |
| 865 | size = item_count << i; |
| 866 | size = size >> i; |
| 867 | if (size == item_count) { |
| 868 | break; |
| 869 | } |
| 870 | } |
| 871 | assert(i); |
| 872 | |
| 873 | /* ... and then we convert it to the position of the highest non-zero bit ... */ |
| 874 | i = (sizeof item_count * CHAR_BIT) - i; |
| 875 | |
| 876 | /* ... and by using it to shift 1 to the left we get the closest sufficient hash table size */ |
| 877 | size = 1 << i; |
| 878 | |
| 879 | return size; |
| 880 | } |