1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2014 Intel Corporation 3 */ 4 #include <string.h> 5 #include <stdalign.h> 6 #include <stdint.h> 7 #include <errno.h> 8 #include <stdio.h> 9 #include <sys/queue.h> 10 11 #include <rte_log.h> 12 #include <rte_common.h> 13 #include <rte_malloc.h> 14 #include <rte_memcpy.h> 15 #include <rte_eal_memconfig.h> 16 #include <rte_string_fns.h> 17 #include <rte_errno.h> 18 #include <rte_hash.h> 19 #include <assert.h> 20 #include <rte_jhash.h> 21 #include <rte_tailq.h> 22 23 #include "rte_lpm6.h" 24 #include "lpm_log.h" 25 26 #define RTE_LPM6_TBL24_NUM_ENTRIES (1 << 24) 27 #define RTE_LPM6_TBL8_GROUP_NUM_ENTRIES 256 28 #define RTE_LPM6_TBL8_MAX_NUM_GROUPS (1 << 21) 29 30 #define RTE_LPM6_VALID_EXT_ENTRY_BITMASK 0xA0000000 31 #define RTE_LPM6_LOOKUP_SUCCESS 0x20000000 32 #define RTE_LPM6_TBL8_BITMASK 0x001FFFFF 33 34 #define ADD_FIRST_BYTE 3 35 #define LOOKUP_FIRST_BYTE 4 36 #define BYTE_SIZE 8 37 #define BYTES2_SIZE 16 38 39 #define RULE_HASH_TABLE_EXTRA_SPACE 64 40 #define TBL24_IND UINT32_MAX 41 42 #define lpm6_tbl8_gindex next_hop 43 44 /** Flags for setting an entry as valid/invalid. */ 45 enum valid_flag { 46 INVALID = 0, 47 VALID 48 }; 49 50 TAILQ_HEAD(rte_lpm6_list, rte_tailq_entry); 51 52 static struct rte_tailq_elem rte_lpm6_tailq = { 53 .name = "RTE_LPM6", 54 }; 55 EAL_REGISTER_TAILQ(rte_lpm6_tailq) 56 57 /** Tbl entry structure. It is the same for both tbl24 and tbl8 */ 58 struct rte_lpm6_tbl_entry { 59 uint32_t next_hop: 21; /**< Next hop / next table to be checked. */ 60 uint32_t depth :8; /**< Rule depth. */ 61 62 /* Flags. */ 63 uint32_t valid :1; /**< Validation flag. */ 64 uint32_t valid_group :1; /**< Group validation flag. */ 65 uint32_t ext_entry :1; /**< External entry. */ 66 }; 67 68 /** Rules tbl entry structure. */ 69 struct rte_lpm6_rule { 70 uint8_t ip[RTE_LPM6_IPV6_ADDR_SIZE]; /**< Rule IP address. */ 71 uint32_t next_hop; /**< Rule next hop. */ 72 uint8_t depth; /**< Rule depth. */ 73 }; 74 75 /** Rules tbl entry key. */ 76 struct rte_lpm6_rule_key { 77 uint8_t ip[RTE_LPM6_IPV6_ADDR_SIZE]; /**< Rule IP address. */ 78 uint32_t depth; /**< Rule depth. */ 79 }; 80 81 /* Header of tbl8 */ 82 struct rte_lpm_tbl8_hdr { 83 uint32_t owner_tbl_ind; /**< owner table: TBL24_IND if owner is tbl24, 84 * otherwise index of tbl8 85 */ 86 uint32_t owner_entry_ind; /**< index of the owner table entry where 87 * pointer to the tbl8 is stored 88 */ 89 uint32_t ref_cnt; /**< table reference counter */ 90 }; 91 92 /** LPM6 structure. */ 93 struct rte_lpm6 { 94 /* LPM metadata. */ 95 char name[RTE_LPM6_NAMESIZE]; /**< Name of the lpm. */ 96 uint32_t max_rules; /**< Max number of rules. */ 97 uint32_t used_rules; /**< Used rules so far. */ 98 uint32_t number_tbl8s; /**< Number of tbl8s to allocate. */ 99 100 /* LPM Tables. */ 101 struct rte_hash *rules_tbl; /**< LPM rules. */ 102 alignas(RTE_CACHE_LINE_SIZE) struct rte_lpm6_tbl_entry tbl24[RTE_LPM6_TBL24_NUM_ENTRIES]; 103 /**< LPM tbl24 table. */ 104 105 uint32_t *tbl8_pool; /**< pool of indexes of free tbl8s */ 106 uint32_t tbl8_pool_pos; /**< current position in the tbl8 pool */ 107 108 struct rte_lpm_tbl8_hdr *tbl8_hdrs; /* array of tbl8 headers */ 109 110 alignas(RTE_CACHE_LINE_SIZE) struct rte_lpm6_tbl_entry tbl8[0]; 111 /**< LPM tbl8 table. */ 112 }; 113 114 /* 115 * Takes an array of uint8_t (IPv6 address) and masks it using the depth. 116 * It leaves untouched one bit per unit in the depth variable 117 * and set the rest to 0. 118 */ 119 static inline void 120 ip6_mask_addr(uint8_t *ip, uint8_t depth) 121 { 122 int16_t part_depth, mask; 123 int i; 124 125 part_depth = depth; 126 127 for (i = 0; i < RTE_LPM6_IPV6_ADDR_SIZE; i++) { 128 if (part_depth < BYTE_SIZE && part_depth >= 0) { 129 mask = (uint16_t)(~(UINT8_MAX >> part_depth)); 130 ip[i] = (uint8_t)(ip[i] & mask); 131 } else if (part_depth < 0) 132 ip[i] = 0; 133 134 part_depth -= BYTE_SIZE; 135 } 136 } 137 138 /* copy ipv6 address */ 139 static inline void 140 ip6_copy_addr(uint8_t *dst, const uint8_t *src) 141 { 142 rte_memcpy(dst, src, RTE_LPM6_IPV6_ADDR_SIZE); 143 } 144 145 /* 146 * LPM6 rule hash function 147 * 148 * It's used as a hash function for the rte_hash 149 * containing rules 150 */ 151 static inline uint32_t 152 rule_hash(const void *data, __rte_unused uint32_t data_len, 153 uint32_t init_val) 154 { 155 return rte_jhash(data, sizeof(struct rte_lpm6_rule_key), init_val); 156 } 157 158 /* 159 * Init pool of free tbl8 indexes 160 */ 161 static void 162 tbl8_pool_init(struct rte_lpm6 *lpm) 163 { 164 uint32_t i; 165 166 /* put entire range of indexes to the tbl8 pool */ 167 for (i = 0; i < lpm->number_tbl8s; i++) 168 lpm->tbl8_pool[i] = i; 169 170 lpm->tbl8_pool_pos = 0; 171 } 172 173 /* 174 * Get an index of a free tbl8 from the pool 175 */ 176 static inline uint32_t 177 tbl8_get(struct rte_lpm6 *lpm, uint32_t *tbl8_ind) 178 { 179 if (lpm->tbl8_pool_pos == lpm->number_tbl8s) 180 /* no more free tbl8 */ 181 return -ENOSPC; 182 183 /* next index */ 184 *tbl8_ind = lpm->tbl8_pool[lpm->tbl8_pool_pos++]; 185 return 0; 186 } 187 188 /* 189 * Put an index of a free tbl8 back to the pool 190 */ 191 static inline uint32_t 192 tbl8_put(struct rte_lpm6 *lpm, uint32_t tbl8_ind) 193 { 194 if (lpm->tbl8_pool_pos == 0) 195 /* pool is full */ 196 return -ENOSPC; 197 198 lpm->tbl8_pool[--lpm->tbl8_pool_pos] = tbl8_ind; 199 return 0; 200 } 201 202 /* 203 * Returns number of tbl8s available in the pool 204 */ 205 static inline uint32_t 206 tbl8_available(struct rte_lpm6 *lpm) 207 { 208 return lpm->number_tbl8s - lpm->tbl8_pool_pos; 209 } 210 211 /* 212 * Init a rule key. 213 * note that ip must be already masked 214 */ 215 static inline void 216 rule_key_init(struct rte_lpm6_rule_key *key, uint8_t *ip, uint8_t depth) 217 { 218 ip6_copy_addr(key->ip, ip); 219 key->depth = depth; 220 } 221 222 /* 223 * Rebuild the entire LPM tree by reinserting all rules 224 */ 225 static void 226 rebuild_lpm(struct rte_lpm6 *lpm) 227 { 228 uint64_t next_hop; 229 struct rte_lpm6_rule_key *rule_key; 230 uint32_t iter = 0; 231 232 while (rte_hash_iterate(lpm->rules_tbl, (void *) &rule_key, 233 (void **) &next_hop, &iter) >= 0) 234 rte_lpm6_add(lpm, rule_key->ip, rule_key->depth, 235 (uint32_t) next_hop); 236 } 237 238 /* 239 * Allocates memory for LPM object 240 */ 241 struct rte_lpm6 * 242 rte_lpm6_create(const char *name, int socket_id, 243 const struct rte_lpm6_config *config) 244 { 245 char mem_name[RTE_LPM6_NAMESIZE]; 246 struct rte_lpm6 *lpm = NULL; 247 struct rte_tailq_entry *te; 248 uint64_t mem_size; 249 struct rte_lpm6_list *lpm_list; 250 struct rte_hash *rules_tbl = NULL; 251 uint32_t *tbl8_pool = NULL; 252 struct rte_lpm_tbl8_hdr *tbl8_hdrs = NULL; 253 254 lpm_list = RTE_TAILQ_CAST(rte_lpm6_tailq.head, rte_lpm6_list); 255 256 RTE_BUILD_BUG_ON(sizeof(struct rte_lpm6_tbl_entry) != sizeof(uint32_t)); 257 RTE_BUILD_BUG_ON(sizeof(struct rte_lpm6_rule_key) % 258 sizeof(uint32_t) != 0); 259 260 /* Check user arguments. */ 261 if ((name == NULL) || (socket_id < -1) || (config == NULL) || 262 (config->max_rules == 0) || 263 config->number_tbl8s > RTE_LPM6_TBL8_MAX_NUM_GROUPS) { 264 rte_errno = EINVAL; 265 return NULL; 266 } 267 268 /* create rules hash table */ 269 snprintf(mem_name, sizeof(mem_name), "LRH_%s", name); 270 struct rte_hash_parameters rule_hash_tbl_params = { 271 .entries = config->max_rules * 1.2 + 272 RULE_HASH_TABLE_EXTRA_SPACE, 273 .key_len = sizeof(struct rte_lpm6_rule_key), 274 .hash_func = rule_hash, 275 .hash_func_init_val = 0, 276 .name = mem_name, 277 .reserved = 0, 278 .socket_id = socket_id, 279 .extra_flag = 0 280 }; 281 282 rules_tbl = rte_hash_create(&rule_hash_tbl_params); 283 if (rules_tbl == NULL) { 284 LPM_LOG(ERR, "LPM rules hash table allocation failed: %s (%d)", 285 rte_strerror(rte_errno), rte_errno); 286 goto fail_wo_unlock; 287 } 288 289 /* allocate tbl8 indexes pool */ 290 tbl8_pool = rte_malloc(NULL, 291 sizeof(uint32_t) * config->number_tbl8s, 292 RTE_CACHE_LINE_SIZE); 293 if (tbl8_pool == NULL) { 294 LPM_LOG(ERR, "LPM tbl8 pool allocation failed: %s (%d)", 295 rte_strerror(rte_errno), rte_errno); 296 rte_errno = ENOMEM; 297 goto fail_wo_unlock; 298 } 299 300 /* allocate tbl8 headers */ 301 tbl8_hdrs = rte_malloc(NULL, 302 sizeof(struct rte_lpm_tbl8_hdr) * config->number_tbl8s, 303 RTE_CACHE_LINE_SIZE); 304 if (tbl8_hdrs == NULL) { 305 LPM_LOG(ERR, "LPM tbl8 headers allocation failed: %s (%d)", 306 rte_strerror(rte_errno), rte_errno); 307 rte_errno = ENOMEM; 308 goto fail_wo_unlock; 309 } 310 311 snprintf(mem_name, sizeof(mem_name), "LPM_%s", name); 312 313 /* Determine the amount of memory to allocate. */ 314 mem_size = sizeof(*lpm) + (sizeof(lpm->tbl8[0]) * 315 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * config->number_tbl8s); 316 317 rte_mcfg_tailq_write_lock(); 318 319 /* Guarantee there's no existing */ 320 TAILQ_FOREACH(te, lpm_list, next) { 321 lpm = (struct rte_lpm6 *) te->data; 322 if (strncmp(name, lpm->name, RTE_LPM6_NAMESIZE) == 0) 323 break; 324 } 325 lpm = NULL; 326 if (te != NULL) { 327 rte_errno = EEXIST; 328 goto fail; 329 } 330 331 /* allocate tailq entry */ 332 te = rte_zmalloc("LPM6_TAILQ_ENTRY", sizeof(*te), 0); 333 if (te == NULL) { 334 LPM_LOG(ERR, "Failed to allocate tailq entry!"); 335 rte_errno = ENOMEM; 336 goto fail; 337 } 338 339 /* Allocate memory to store the LPM data structures. */ 340 lpm = rte_zmalloc_socket(mem_name, (size_t)mem_size, 341 RTE_CACHE_LINE_SIZE, socket_id); 342 343 if (lpm == NULL) { 344 LPM_LOG(ERR, "LPM memory allocation failed"); 345 rte_free(te); 346 rte_errno = ENOMEM; 347 goto fail; 348 } 349 350 /* Save user arguments. */ 351 lpm->max_rules = config->max_rules; 352 lpm->number_tbl8s = config->number_tbl8s; 353 strlcpy(lpm->name, name, sizeof(lpm->name)); 354 lpm->rules_tbl = rules_tbl; 355 lpm->tbl8_pool = tbl8_pool; 356 lpm->tbl8_hdrs = tbl8_hdrs; 357 358 /* init the stack */ 359 tbl8_pool_init(lpm); 360 361 te->data = (void *) lpm; 362 363 TAILQ_INSERT_TAIL(lpm_list, te, next); 364 rte_mcfg_tailq_write_unlock(); 365 return lpm; 366 367 fail: 368 rte_mcfg_tailq_write_unlock(); 369 370 fail_wo_unlock: 371 rte_free(tbl8_hdrs); 372 rte_free(tbl8_pool); 373 rte_hash_free(rules_tbl); 374 375 return NULL; 376 } 377 378 /* 379 * Find an existing lpm table and return a pointer to it. 380 */ 381 struct rte_lpm6 * 382 rte_lpm6_find_existing(const char *name) 383 { 384 struct rte_lpm6 *l = NULL; 385 struct rte_tailq_entry *te; 386 struct rte_lpm6_list *lpm_list; 387 388 lpm_list = RTE_TAILQ_CAST(rte_lpm6_tailq.head, rte_lpm6_list); 389 390 rte_mcfg_tailq_read_lock(); 391 TAILQ_FOREACH(te, lpm_list, next) { 392 l = (struct rte_lpm6 *) te->data; 393 if (strncmp(name, l->name, RTE_LPM6_NAMESIZE) == 0) 394 break; 395 } 396 rte_mcfg_tailq_read_unlock(); 397 398 if (te == NULL) { 399 rte_errno = ENOENT; 400 return NULL; 401 } 402 403 return l; 404 } 405 406 /* 407 * Deallocates memory for given LPM table. 408 */ 409 void 410 rte_lpm6_free(struct rte_lpm6 *lpm) 411 { 412 struct rte_lpm6_list *lpm_list; 413 struct rte_tailq_entry *te; 414 415 /* Check user arguments. */ 416 if (lpm == NULL) 417 return; 418 419 lpm_list = RTE_TAILQ_CAST(rte_lpm6_tailq.head, rte_lpm6_list); 420 421 rte_mcfg_tailq_write_lock(); 422 423 /* find our tailq entry */ 424 TAILQ_FOREACH(te, lpm_list, next) { 425 if (te->data == (void *) lpm) 426 break; 427 } 428 429 if (te != NULL) 430 TAILQ_REMOVE(lpm_list, te, next); 431 432 rte_mcfg_tailq_write_unlock(); 433 434 rte_free(lpm->tbl8_hdrs); 435 rte_free(lpm->tbl8_pool); 436 rte_hash_free(lpm->rules_tbl); 437 rte_free(lpm); 438 rte_free(te); 439 } 440 441 /* Find a rule */ 442 static inline int 443 rule_find_with_key(struct rte_lpm6 *lpm, 444 const struct rte_lpm6_rule_key *rule_key, 445 uint32_t *next_hop) 446 { 447 uint64_t hash_val; 448 int ret; 449 450 /* lookup for a rule */ 451 ret = rte_hash_lookup_data(lpm->rules_tbl, (const void *) rule_key, 452 (void **) &hash_val); 453 if (ret >= 0) { 454 *next_hop = (uint32_t) hash_val; 455 return 1; 456 } 457 458 return 0; 459 } 460 461 /* Find a rule */ 462 static int 463 rule_find(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth, 464 uint32_t *next_hop) 465 { 466 struct rte_lpm6_rule_key rule_key; 467 468 /* init a rule key */ 469 rule_key_init(&rule_key, ip, depth); 470 471 return rule_find_with_key(lpm, &rule_key, next_hop); 472 } 473 474 /* 475 * Checks if a rule already exists in the rules table and updates 476 * the nexthop if so. Otherwise it adds a new rule if enough space is available. 477 * 478 * Returns: 479 * 0 - next hop of existed rule is updated 480 * 1 - new rule successfully added 481 * <0 - error 482 */ 483 static inline int 484 rule_add(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth, uint32_t next_hop) 485 { 486 int ret, rule_exist; 487 struct rte_lpm6_rule_key rule_key; 488 uint32_t unused; 489 490 /* init a rule key */ 491 rule_key_init(&rule_key, ip, depth); 492 493 /* Scan through rule list to see if rule already exists. */ 494 rule_exist = rule_find_with_key(lpm, &rule_key, &unused); 495 496 /* 497 * If rule does not exist check if there is space to add a new rule to 498 * this rule group. If there is no space return error. 499 */ 500 if (!rule_exist && lpm->used_rules == lpm->max_rules) 501 return -ENOSPC; 502 503 /* add the rule or update rules next hop */ 504 ret = rte_hash_add_key_data(lpm->rules_tbl, &rule_key, 505 (void *)(uintptr_t) next_hop); 506 if (ret < 0) 507 return ret; 508 509 /* Increment the used rules counter for this rule group. */ 510 if (!rule_exist) { 511 lpm->used_rules++; 512 return 1; 513 } 514 515 return 0; 516 } 517 518 /* 519 * Function that expands a rule across the data structure when a less-generic 520 * one has been added before. It assures that every possible combination of bits 521 * in the IP address returns a match. 522 */ 523 static void 524 expand_rule(struct rte_lpm6 *lpm, uint32_t tbl8_gindex, uint8_t old_depth, 525 uint8_t new_depth, uint32_t next_hop, uint8_t valid) 526 { 527 uint32_t tbl8_group_end, tbl8_gindex_next, j; 528 529 tbl8_group_end = tbl8_gindex + RTE_LPM6_TBL8_GROUP_NUM_ENTRIES; 530 531 struct rte_lpm6_tbl_entry new_tbl8_entry = { 532 .valid = valid, 533 .valid_group = valid, 534 .depth = new_depth, 535 .next_hop = next_hop, 536 .ext_entry = 0, 537 }; 538 539 for (j = tbl8_gindex; j < tbl8_group_end; j++) { 540 if (!lpm->tbl8[j].valid || (lpm->tbl8[j].ext_entry == 0 541 && lpm->tbl8[j].depth <= old_depth)) { 542 543 lpm->tbl8[j] = new_tbl8_entry; 544 545 } else if (lpm->tbl8[j].ext_entry == 1) { 546 547 tbl8_gindex_next = lpm->tbl8[j].lpm6_tbl8_gindex 548 * RTE_LPM6_TBL8_GROUP_NUM_ENTRIES; 549 expand_rule(lpm, tbl8_gindex_next, old_depth, new_depth, 550 next_hop, valid); 551 } 552 } 553 } 554 555 /* 556 * Init a tbl8 header 557 */ 558 static inline void 559 init_tbl8_header(struct rte_lpm6 *lpm, uint32_t tbl_ind, 560 uint32_t owner_tbl_ind, uint32_t owner_entry_ind) 561 { 562 struct rte_lpm_tbl8_hdr *tbl_hdr = &lpm->tbl8_hdrs[tbl_ind]; 563 tbl_hdr->owner_tbl_ind = owner_tbl_ind; 564 tbl_hdr->owner_entry_ind = owner_entry_ind; 565 tbl_hdr->ref_cnt = 0; 566 } 567 568 /* 569 * Calculate index to the table based on the number and position 570 * of the bytes being inspected in this step. 571 */ 572 static uint32_t 573 get_bitshift(const uint8_t *ip, uint8_t first_byte, uint8_t bytes) 574 { 575 uint32_t entry_ind, i; 576 int8_t bitshift; 577 578 entry_ind = 0; 579 for (i = first_byte; i < (uint32_t)(first_byte + bytes); i++) { 580 bitshift = (int8_t)((bytes - i)*BYTE_SIZE); 581 582 if (bitshift < 0) 583 bitshift = 0; 584 entry_ind = entry_ind | ip[i-1] << bitshift; 585 } 586 587 return entry_ind; 588 } 589 590 /* 591 * Simulate adding a new route to the LPM counting number 592 * of new tables that will be needed 593 * 594 * It returns 0 on success, or 1 if 595 * the process needs to be continued by calling the function again. 596 */ 597 static inline int 598 simulate_add_step(struct rte_lpm6 *lpm, struct rte_lpm6_tbl_entry *tbl, 599 struct rte_lpm6_tbl_entry **next_tbl, const uint8_t *ip, 600 uint8_t bytes, uint8_t first_byte, uint8_t depth, 601 uint32_t *need_tbl_nb) 602 { 603 uint32_t entry_ind; 604 uint8_t bits_covered; 605 uint32_t next_tbl_ind; 606 607 /* 608 * Calculate index to the table based on the number and position 609 * of the bytes being inspected in this step. 610 */ 611 entry_ind = get_bitshift(ip, first_byte, bytes); 612 613 /* Number of bits covered in this step */ 614 bits_covered = (uint8_t)((bytes+first_byte-1)*BYTE_SIZE); 615 616 if (depth <= bits_covered) { 617 *need_tbl_nb = 0; 618 return 0; 619 } 620 621 if (tbl[entry_ind].valid == 0 || tbl[entry_ind].ext_entry == 0) { 622 /* from this point on a new table is needed on each level 623 * that is not covered yet 624 */ 625 depth -= bits_covered; 626 uint32_t cnt = depth >> 3; /* depth / BYTE_SIZE */ 627 if (depth & 7) /* 0b00000111 */ 628 /* if depth % 8 > 0 then one more table is needed 629 * for those last bits 630 */ 631 cnt++; 632 633 *need_tbl_nb = cnt; 634 return 0; 635 } 636 637 next_tbl_ind = tbl[entry_ind].lpm6_tbl8_gindex; 638 *next_tbl = &(lpm->tbl8[next_tbl_ind * 639 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES]); 640 *need_tbl_nb = 0; 641 return 1; 642 } 643 644 /* 645 * Partially adds a new route to the data structure (tbl24+tbl8s). 646 * It returns 0 on success, a negative number on failure, or 1 if 647 * the process needs to be continued by calling the function again. 648 */ 649 static inline int 650 add_step(struct rte_lpm6 *lpm, struct rte_lpm6_tbl_entry *tbl, 651 uint32_t tbl_ind, struct rte_lpm6_tbl_entry **next_tbl, 652 uint32_t *next_tbl_ind, uint8_t *ip, uint8_t bytes, 653 uint8_t first_byte, uint8_t depth, uint32_t next_hop, 654 uint8_t is_new_rule) 655 { 656 uint32_t entry_ind, tbl_range, tbl8_group_start, tbl8_group_end, i; 657 uint32_t tbl8_gindex; 658 uint8_t bits_covered; 659 int ret; 660 661 /* 662 * Calculate index to the table based on the number and position 663 * of the bytes being inspected in this step. 664 */ 665 entry_ind = get_bitshift(ip, first_byte, bytes); 666 667 /* Number of bits covered in this step */ 668 bits_covered = (uint8_t)((bytes+first_byte-1)*BYTE_SIZE); 669 670 /* 671 * If depth if smaller than this number (ie this is the last step) 672 * expand the rule across the relevant positions in the table. 673 */ 674 if (depth <= bits_covered) { 675 tbl_range = 1 << (bits_covered - depth); 676 677 for (i = entry_ind; i < (entry_ind + tbl_range); i++) { 678 if (!tbl[i].valid || (tbl[i].ext_entry == 0 && 679 tbl[i].depth <= depth)) { 680 681 struct rte_lpm6_tbl_entry new_tbl_entry = { 682 .next_hop = next_hop, 683 .depth = depth, 684 .valid = VALID, 685 .valid_group = VALID, 686 .ext_entry = 0, 687 }; 688 689 tbl[i] = new_tbl_entry; 690 691 } else if (tbl[i].ext_entry == 1) { 692 693 /* 694 * If tbl entry is valid and extended calculate the index 695 * into next tbl8 and expand the rule across the data structure. 696 */ 697 tbl8_gindex = tbl[i].lpm6_tbl8_gindex * 698 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES; 699 expand_rule(lpm, tbl8_gindex, depth, depth, 700 next_hop, VALID); 701 } 702 } 703 704 /* update tbl8 rule reference counter */ 705 if (tbl_ind != TBL24_IND && is_new_rule) 706 lpm->tbl8_hdrs[tbl_ind].ref_cnt++; 707 708 return 0; 709 } 710 /* 711 * If this is not the last step just fill one position 712 * and calculate the index to the next table. 713 */ 714 else { 715 /* If it's invalid a new tbl8 is needed */ 716 if (!tbl[entry_ind].valid) { 717 /* get a new table */ 718 ret = tbl8_get(lpm, &tbl8_gindex); 719 if (ret != 0) 720 return -ENOSPC; 721 722 /* invalidate all new tbl8 entries */ 723 tbl8_group_start = tbl8_gindex * 724 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES; 725 memset(&lpm->tbl8[tbl8_group_start], 0, 726 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * 727 sizeof(struct rte_lpm6_tbl_entry)); 728 729 /* init the new table's header: 730 * save the reference to the owner table 731 */ 732 init_tbl8_header(lpm, tbl8_gindex, tbl_ind, entry_ind); 733 734 /* reference to a new tbl8 */ 735 struct rte_lpm6_tbl_entry new_tbl_entry = { 736 .lpm6_tbl8_gindex = tbl8_gindex, 737 .depth = 0, 738 .valid = VALID, 739 .valid_group = VALID, 740 .ext_entry = 1, 741 }; 742 743 tbl[entry_ind] = new_tbl_entry; 744 745 /* update the current table's reference counter */ 746 if (tbl_ind != TBL24_IND) 747 lpm->tbl8_hdrs[tbl_ind].ref_cnt++; 748 } 749 /* 750 * If it's valid but not extended the rule that was stored 751 * here needs to be moved to the next table. 752 */ 753 else if (tbl[entry_ind].ext_entry == 0) { 754 /* get a new tbl8 index */ 755 ret = tbl8_get(lpm, &tbl8_gindex); 756 if (ret != 0) 757 return -ENOSPC; 758 759 tbl8_group_start = tbl8_gindex * 760 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES; 761 tbl8_group_end = tbl8_group_start + 762 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES; 763 764 struct rte_lpm6_tbl_entry tbl_entry = { 765 .next_hop = tbl[entry_ind].next_hop, 766 .depth = tbl[entry_ind].depth, 767 .valid = VALID, 768 .valid_group = VALID, 769 .ext_entry = 0 770 }; 771 772 /* Populate new tbl8 with tbl value. */ 773 for (i = tbl8_group_start; i < tbl8_group_end; i++) 774 lpm->tbl8[i] = tbl_entry; 775 776 /* init the new table's header: 777 * save the reference to the owner table 778 */ 779 init_tbl8_header(lpm, tbl8_gindex, tbl_ind, entry_ind); 780 781 /* 782 * Update tbl entry to point to new tbl8 entry. Note: The 783 * ext_flag and tbl8_index need to be updated simultaneously, 784 * so assign whole structure in one go. 785 */ 786 struct rte_lpm6_tbl_entry new_tbl_entry = { 787 .lpm6_tbl8_gindex = tbl8_gindex, 788 .depth = 0, 789 .valid = VALID, 790 .valid_group = VALID, 791 .ext_entry = 1, 792 }; 793 794 tbl[entry_ind] = new_tbl_entry; 795 796 /* update the current table's reference counter */ 797 if (tbl_ind != TBL24_IND) 798 lpm->tbl8_hdrs[tbl_ind].ref_cnt++; 799 } 800 801 *next_tbl_ind = tbl[entry_ind].lpm6_tbl8_gindex; 802 *next_tbl = &(lpm->tbl8[*next_tbl_ind * 803 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES]); 804 } 805 806 return 1; 807 } 808 809 /* 810 * Simulate adding a route to LPM 811 * 812 * Returns: 813 * 0 on success 814 * -ENOSPC not enough tbl8 left 815 */ 816 static int 817 simulate_add(struct rte_lpm6 *lpm, const uint8_t *masked_ip, uint8_t depth) 818 { 819 struct rte_lpm6_tbl_entry *tbl; 820 struct rte_lpm6_tbl_entry *tbl_next = NULL; 821 int ret, i; 822 823 /* number of new tables needed for a step */ 824 uint32_t need_tbl_nb; 825 /* total number of new tables needed */ 826 uint32_t total_need_tbl_nb; 827 828 /* Inspect the first three bytes through tbl24 on the first step. */ 829 ret = simulate_add_step(lpm, lpm->tbl24, &tbl_next, masked_ip, 830 ADD_FIRST_BYTE, 1, depth, &need_tbl_nb); 831 total_need_tbl_nb = need_tbl_nb; 832 /* 833 * Inspect one by one the rest of the bytes until 834 * the process is completed. 835 */ 836 for (i = ADD_FIRST_BYTE; i < RTE_LPM6_IPV6_ADDR_SIZE && ret == 1; i++) { 837 tbl = tbl_next; 838 ret = simulate_add_step(lpm, tbl, &tbl_next, masked_ip, 1, 839 (uint8_t)(i + 1), depth, &need_tbl_nb); 840 total_need_tbl_nb += need_tbl_nb; 841 } 842 843 if (tbl8_available(lpm) < total_need_tbl_nb) 844 /* not enough tbl8 to add a rule */ 845 return -ENOSPC; 846 847 return 0; 848 } 849 850 /* 851 * Add a route 852 */ 853 int 854 rte_lpm6_add(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth, 855 uint32_t next_hop) 856 { 857 struct rte_lpm6_tbl_entry *tbl; 858 struct rte_lpm6_tbl_entry *tbl_next = NULL; 859 /* init to avoid compiler warning */ 860 uint32_t tbl_next_num = 123456; 861 int status; 862 uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE]; 863 int i; 864 865 /* Check user arguments. */ 866 if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM6_MAX_DEPTH)) 867 return -EINVAL; 868 869 /* Copy the IP and mask it to avoid modifying user's input data. */ 870 ip6_copy_addr(masked_ip, ip); 871 ip6_mask_addr(masked_ip, depth); 872 873 /* Simulate adding a new route */ 874 int ret = simulate_add(lpm, masked_ip, depth); 875 if (ret < 0) 876 return ret; 877 878 /* Add the rule to the rule table. */ 879 int is_new_rule = rule_add(lpm, masked_ip, depth, next_hop); 880 /* If there is no space available for new rule return error. */ 881 if (is_new_rule < 0) 882 return is_new_rule; 883 884 /* Inspect the first three bytes through tbl24 on the first step. */ 885 tbl = lpm->tbl24; 886 status = add_step(lpm, tbl, TBL24_IND, &tbl_next, &tbl_next_num, 887 masked_ip, ADD_FIRST_BYTE, 1, depth, next_hop, 888 is_new_rule); 889 assert(status >= 0); 890 891 /* 892 * Inspect one by one the rest of the bytes until 893 * the process is completed. 894 */ 895 for (i = ADD_FIRST_BYTE; i < RTE_LPM6_IPV6_ADDR_SIZE && status == 1; i++) { 896 tbl = tbl_next; 897 status = add_step(lpm, tbl, tbl_next_num, &tbl_next, 898 &tbl_next_num, masked_ip, 1, (uint8_t)(i + 1), 899 depth, next_hop, is_new_rule); 900 assert(status >= 0); 901 } 902 903 return status; 904 } 905 906 /* 907 * Takes a pointer to a table entry and inspect one level. 908 * The function returns 0 on lookup success, ENOENT if no match was found 909 * or 1 if the process needs to be continued by calling the function again. 910 */ 911 static inline int 912 lookup_step(const struct rte_lpm6 *lpm, const struct rte_lpm6_tbl_entry *tbl, 913 const struct rte_lpm6_tbl_entry **tbl_next, const uint8_t *ip, 914 uint8_t first_byte, uint32_t *next_hop) 915 { 916 uint32_t tbl8_index, tbl_entry; 917 918 /* Take the integer value from the pointer. */ 919 tbl_entry = *(const uint32_t *)tbl; 920 921 /* If it is valid and extended we calculate the new pointer to return. */ 922 if ((tbl_entry & RTE_LPM6_VALID_EXT_ENTRY_BITMASK) == 923 RTE_LPM6_VALID_EXT_ENTRY_BITMASK) { 924 925 tbl8_index = ip[first_byte-1] + 926 ((tbl_entry & RTE_LPM6_TBL8_BITMASK) * 927 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES); 928 929 *tbl_next = &lpm->tbl8[tbl8_index]; 930 931 return 1; 932 } else { 933 /* If not extended then we can have a match. */ 934 *next_hop = ((uint32_t)tbl_entry & RTE_LPM6_TBL8_BITMASK); 935 return (tbl_entry & RTE_LPM6_LOOKUP_SUCCESS) ? 0 : -ENOENT; 936 } 937 } 938 939 /* 940 * Looks up an IP 941 */ 942 int 943 rte_lpm6_lookup(const struct rte_lpm6 *lpm, const uint8_t *ip, 944 uint32_t *next_hop) 945 { 946 const struct rte_lpm6_tbl_entry *tbl; 947 const struct rte_lpm6_tbl_entry *tbl_next = NULL; 948 int status; 949 uint8_t first_byte; 950 uint32_t tbl24_index; 951 952 /* DEBUG: Check user input arguments. */ 953 if ((lpm == NULL) || (ip == NULL) || (next_hop == NULL)) 954 return -EINVAL; 955 956 first_byte = LOOKUP_FIRST_BYTE; 957 tbl24_index = (ip[0] << BYTES2_SIZE) | (ip[1] << BYTE_SIZE) | ip[2]; 958 959 /* Calculate pointer to the first entry to be inspected */ 960 tbl = &lpm->tbl24[tbl24_index]; 961 962 do { 963 /* Continue inspecting following levels until success or failure */ 964 status = lookup_step(lpm, tbl, &tbl_next, ip, first_byte++, next_hop); 965 tbl = tbl_next; 966 } while (status == 1); 967 968 return status; 969 } 970 971 /* 972 * Looks up a group of IP addresses 973 */ 974 int 975 rte_lpm6_lookup_bulk_func(const struct rte_lpm6 *lpm, 976 uint8_t ips[][RTE_LPM6_IPV6_ADDR_SIZE], 977 int32_t *next_hops, unsigned int n) 978 { 979 unsigned int i; 980 const struct rte_lpm6_tbl_entry *tbl; 981 const struct rte_lpm6_tbl_entry *tbl_next = NULL; 982 uint32_t tbl24_index, next_hop; 983 uint8_t first_byte; 984 int status; 985 986 /* DEBUG: Check user input arguments. */ 987 if ((lpm == NULL) || (ips == NULL) || (next_hops == NULL)) 988 return -EINVAL; 989 990 for (i = 0; i < n; i++) { 991 first_byte = LOOKUP_FIRST_BYTE; 992 tbl24_index = (ips[i][0] << BYTES2_SIZE) | 993 (ips[i][1] << BYTE_SIZE) | ips[i][2]; 994 995 /* Calculate pointer to the first entry to be inspected */ 996 tbl = &lpm->tbl24[tbl24_index]; 997 998 do { 999 /* Continue inspecting following levels 1000 * until success or failure 1001 */ 1002 status = lookup_step(lpm, tbl, &tbl_next, ips[i], 1003 first_byte++, &next_hop); 1004 tbl = tbl_next; 1005 } while (status == 1); 1006 1007 if (status < 0) 1008 next_hops[i] = -1; 1009 else 1010 next_hops[i] = (int32_t)next_hop; 1011 } 1012 1013 return 0; 1014 } 1015 1016 /* 1017 * Look for a rule in the high-level rules table 1018 */ 1019 int 1020 rte_lpm6_is_rule_present(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth, 1021 uint32_t *next_hop) 1022 { 1023 uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE]; 1024 1025 /* Check user arguments. */ 1026 if ((lpm == NULL) || next_hop == NULL || ip == NULL || 1027 (depth < 1) || (depth > RTE_LPM6_MAX_DEPTH)) 1028 return -EINVAL; 1029 1030 /* Copy the IP and mask it to avoid modifying user's input data. */ 1031 ip6_copy_addr(masked_ip, ip); 1032 ip6_mask_addr(masked_ip, depth); 1033 1034 return rule_find(lpm, masked_ip, depth, next_hop); 1035 } 1036 1037 /* 1038 * Delete a rule from the rule table. 1039 * NOTE: Valid range for depth parameter is 1 .. 128 inclusive. 1040 * return 1041 * 0 on success 1042 * <0 on failure 1043 */ 1044 static inline int 1045 rule_delete(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth) 1046 { 1047 int ret; 1048 struct rte_lpm6_rule_key rule_key; 1049 1050 /* init rule key */ 1051 rule_key_init(&rule_key, ip, depth); 1052 1053 /* delete the rule */ 1054 ret = rte_hash_del_key(lpm->rules_tbl, (void *) &rule_key); 1055 if (ret >= 0) 1056 lpm->used_rules--; 1057 1058 return ret; 1059 } 1060 1061 /* 1062 * Deletes a group of rules 1063 * 1064 * Note that the function rebuilds the lpm table, 1065 * rather than doing incremental updates like 1066 * the regular delete function 1067 */ 1068 int 1069 rte_lpm6_delete_bulk_func(struct rte_lpm6 *lpm, 1070 uint8_t ips[][RTE_LPM6_IPV6_ADDR_SIZE], uint8_t *depths, 1071 unsigned n) 1072 { 1073 uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE]; 1074 unsigned i; 1075 1076 /* Check input arguments. */ 1077 if ((lpm == NULL) || (ips == NULL) || (depths == NULL)) 1078 return -EINVAL; 1079 1080 for (i = 0; i < n; i++) { 1081 ip6_copy_addr(masked_ip, ips[i]); 1082 ip6_mask_addr(masked_ip, depths[i]); 1083 rule_delete(lpm, masked_ip, depths[i]); 1084 } 1085 1086 /* 1087 * Set all the table entries to 0 (ie delete every rule 1088 * from the data structure. 1089 */ 1090 memset(lpm->tbl24, 0, sizeof(lpm->tbl24)); 1091 memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0]) 1092 * RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s); 1093 tbl8_pool_init(lpm); 1094 1095 /* 1096 * Add every rule again (except for the ones that were removed from 1097 * the rules table). 1098 */ 1099 rebuild_lpm(lpm); 1100 1101 return 0; 1102 } 1103 1104 /* 1105 * Delete all rules from the LPM table. 1106 */ 1107 void 1108 rte_lpm6_delete_all(struct rte_lpm6 *lpm) 1109 { 1110 /* Zero used rules counter. */ 1111 lpm->used_rules = 0; 1112 1113 /* Zero tbl24. */ 1114 memset(lpm->tbl24, 0, sizeof(lpm->tbl24)); 1115 1116 /* Zero tbl8. */ 1117 memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0]) * 1118 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s); 1119 1120 /* init pool of free tbl8 indexes */ 1121 tbl8_pool_init(lpm); 1122 1123 /* Delete all rules form the rules table. */ 1124 rte_hash_reset(lpm->rules_tbl); 1125 } 1126 1127 /* 1128 * Convert a depth to a one byte long mask 1129 * Example: 4 will be converted to 0xF0 1130 */ 1131 static uint8_t __attribute__((pure)) 1132 depth_to_mask_1b(uint8_t depth) 1133 { 1134 /* To calculate a mask start with a 1 on the left hand side and right 1135 * shift while populating the left hand side with 1's 1136 */ 1137 return (signed char)0x80 >> (depth - 1); 1138 } 1139 1140 /* 1141 * Find a less specific rule 1142 */ 1143 static int 1144 rule_find_less_specific(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth, 1145 struct rte_lpm6_rule *rule) 1146 { 1147 int ret; 1148 uint32_t next_hop; 1149 uint8_t mask; 1150 struct rte_lpm6_rule_key rule_key; 1151 1152 if (depth == 1) 1153 return 0; 1154 1155 rule_key_init(&rule_key, ip, depth); 1156 1157 while (depth > 1) { 1158 depth--; 1159 1160 /* each iteration zero one more bit of the key */ 1161 mask = depth & 7; /* depth % BYTE_SIZE */ 1162 if (mask > 0) 1163 mask = depth_to_mask_1b(mask); 1164 1165 rule_key.depth = depth; 1166 rule_key.ip[depth >> 3] &= mask; 1167 1168 ret = rule_find_with_key(lpm, &rule_key, &next_hop); 1169 if (ret) { 1170 rule->depth = depth; 1171 ip6_copy_addr(rule->ip, rule_key.ip); 1172 rule->next_hop = next_hop; 1173 return 1; 1174 } 1175 } 1176 1177 return 0; 1178 } 1179 1180 /* 1181 * Find range of tbl8 cells occupied by a rule 1182 */ 1183 static void 1184 rule_find_range(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth, 1185 struct rte_lpm6_tbl_entry **from, 1186 struct rte_lpm6_tbl_entry **to, 1187 uint32_t *out_tbl_ind) 1188 { 1189 uint32_t ind; 1190 uint32_t first_3bytes = (uint32_t)ip[0] << 16 | ip[1] << 8 | ip[2]; 1191 1192 if (depth <= 24) { 1193 /* rule is within the top level */ 1194 ind = first_3bytes; 1195 *from = &lpm->tbl24[ind]; 1196 ind += (1 << (24 - depth)) - 1; 1197 *to = &lpm->tbl24[ind]; 1198 *out_tbl_ind = TBL24_IND; 1199 } else { 1200 /* top level entry */ 1201 struct rte_lpm6_tbl_entry *tbl = &lpm->tbl24[first_3bytes]; 1202 assert(tbl->ext_entry == 1); 1203 /* first tbl8 */ 1204 uint32_t tbl_ind = tbl->lpm6_tbl8_gindex; 1205 tbl = &lpm->tbl8[tbl_ind * 1206 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES]; 1207 /* current ip byte, the top level is already behind */ 1208 uint8_t byte = 3; 1209 /* minus top level */ 1210 depth -= 24; 1211 1212 /* iterate through levels (tbl8s) 1213 * until we reach the last one 1214 */ 1215 while (depth > 8) { 1216 tbl += ip[byte]; 1217 assert(tbl->ext_entry == 1); 1218 /* go to the next level/tbl8 */ 1219 tbl_ind = tbl->lpm6_tbl8_gindex; 1220 tbl = &lpm->tbl8[tbl_ind * 1221 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES]; 1222 byte += 1; 1223 depth -= 8; 1224 } 1225 1226 /* last level/tbl8 */ 1227 ind = ip[byte] & depth_to_mask_1b(depth); 1228 *from = &tbl[ind]; 1229 ind += (1 << (8 - depth)) - 1; 1230 *to = &tbl[ind]; 1231 *out_tbl_ind = tbl_ind; 1232 } 1233 } 1234 1235 /* 1236 * Remove a table from the LPM tree 1237 */ 1238 static void 1239 remove_tbl(struct rte_lpm6 *lpm, struct rte_lpm_tbl8_hdr *tbl_hdr, 1240 uint32_t tbl_ind, struct rte_lpm6_rule *lsp_rule) 1241 { 1242 struct rte_lpm6_tbl_entry *owner_entry; 1243 1244 if (tbl_hdr->owner_tbl_ind == TBL24_IND) 1245 owner_entry = &lpm->tbl24[tbl_hdr->owner_entry_ind]; 1246 else { 1247 uint32_t owner_tbl_ind = tbl_hdr->owner_tbl_ind; 1248 owner_entry = &lpm->tbl8[ 1249 owner_tbl_ind * RTE_LPM6_TBL8_GROUP_NUM_ENTRIES + 1250 tbl_hdr->owner_entry_ind]; 1251 1252 struct rte_lpm_tbl8_hdr *owner_tbl_hdr = 1253 &lpm->tbl8_hdrs[owner_tbl_ind]; 1254 if (--owner_tbl_hdr->ref_cnt == 0) 1255 remove_tbl(lpm, owner_tbl_hdr, owner_tbl_ind, lsp_rule); 1256 } 1257 1258 assert(owner_entry->ext_entry == 1); 1259 1260 /* unlink the table */ 1261 if (lsp_rule != NULL) { 1262 struct rte_lpm6_tbl_entry new_tbl_entry = { 1263 .next_hop = lsp_rule->next_hop, 1264 .depth = lsp_rule->depth, 1265 .valid = VALID, 1266 .valid_group = VALID, 1267 .ext_entry = 0 1268 }; 1269 1270 *owner_entry = new_tbl_entry; 1271 } else { 1272 struct rte_lpm6_tbl_entry new_tbl_entry = { 1273 .next_hop = 0, 1274 .depth = 0, 1275 .valid = INVALID, 1276 .valid_group = INVALID, 1277 .ext_entry = 0 1278 }; 1279 1280 *owner_entry = new_tbl_entry; 1281 } 1282 1283 /* return the table to the pool */ 1284 tbl8_put(lpm, tbl_ind); 1285 } 1286 1287 /* 1288 * Deletes a rule 1289 */ 1290 int 1291 rte_lpm6_delete(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth) 1292 { 1293 uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE]; 1294 struct rte_lpm6_rule lsp_rule_obj; 1295 struct rte_lpm6_rule *lsp_rule; 1296 int ret; 1297 uint32_t tbl_ind; 1298 struct rte_lpm6_tbl_entry *from, *to; 1299 1300 /* Check input arguments. */ 1301 if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM6_MAX_DEPTH)) 1302 return -EINVAL; 1303 1304 /* Copy the IP and mask it to avoid modifying user's input data. */ 1305 ip6_copy_addr(masked_ip, ip); 1306 ip6_mask_addr(masked_ip, depth); 1307 1308 /* Delete the rule from the rule table. */ 1309 ret = rule_delete(lpm, masked_ip, depth); 1310 if (ret < 0) 1311 return -ENOENT; 1312 1313 /* find rule cells */ 1314 rule_find_range(lpm, masked_ip, depth, &from, &to, &tbl_ind); 1315 1316 /* find a less specific rule (a rule with smaller depth) 1317 * note: masked_ip will be modified, don't use it anymore 1318 */ 1319 ret = rule_find_less_specific(lpm, masked_ip, depth, 1320 &lsp_rule_obj); 1321 lsp_rule = ret ? &lsp_rule_obj : NULL; 1322 1323 /* decrement the table rule counter, 1324 * note that tbl24 doesn't have a header 1325 */ 1326 if (tbl_ind != TBL24_IND) { 1327 struct rte_lpm_tbl8_hdr *tbl_hdr = &lpm->tbl8_hdrs[tbl_ind]; 1328 if (--tbl_hdr->ref_cnt == 0) { 1329 /* remove the table */ 1330 remove_tbl(lpm, tbl_hdr, tbl_ind, lsp_rule); 1331 return 0; 1332 } 1333 } 1334 1335 /* iterate rule cells */ 1336 for (; from <= to; from++) 1337 if (from->ext_entry == 1) { 1338 /* reference to a more specific space 1339 * of the prefix/rule. Entries in a more 1340 * specific space that are not used by 1341 * a more specific prefix must be occupied 1342 * by the prefix 1343 */ 1344 if (lsp_rule != NULL) 1345 expand_rule(lpm, 1346 from->lpm6_tbl8_gindex * 1347 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES, 1348 depth, lsp_rule->depth, 1349 lsp_rule->next_hop, VALID); 1350 else 1351 /* since the prefix has no less specific prefix, 1352 * its more specific space must be invalidated 1353 */ 1354 expand_rule(lpm, 1355 from->lpm6_tbl8_gindex * 1356 RTE_LPM6_TBL8_GROUP_NUM_ENTRIES, 1357 depth, 0, 0, INVALID); 1358 } else if (from->depth == depth) { 1359 /* entry is not a reference and belongs to the prefix */ 1360 if (lsp_rule != NULL) { 1361 struct rte_lpm6_tbl_entry new_tbl_entry = { 1362 .next_hop = lsp_rule->next_hop, 1363 .depth = lsp_rule->depth, 1364 .valid = VALID, 1365 .valid_group = VALID, 1366 .ext_entry = 0 1367 }; 1368 1369 *from = new_tbl_entry; 1370 } else { 1371 struct rte_lpm6_tbl_entry new_tbl_entry = { 1372 .next_hop = 0, 1373 .depth = 0, 1374 .valid = INVALID, 1375 .valid_group = INVALID, 1376 .ext_entry = 0 1377 }; 1378 1379 *from = new_tbl_entry; 1380 } 1381 } 1382 1383 return 0; 1384 } 1385