xref: /freebsd/sys/contrib/dpdk_rte_lpm/rte_lpm6.c (revision 537d1343)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 #include <sys/param.h>
6 #include <sys/ctype.h>
7 #include <sys/systm.h>
8 #include <sys/lock.h>
9 #include <sys/rwlock.h>
10 #include <sys/malloc.h>
11 #include <sys/mbuf.h>
12 #include <sys/socket.h>
13 #include <sys/kernel.h>
14 
15 //#include <netinet6/rte_tailq.h>
16 int errno = 0, rte_errno = 0;
17 
18 #include "rte_shim.h"
19 #include "rte_lpm6.h"
20 
21 #define RTE_LPM6_TBL24_NUM_ENTRIES        (1 << 24)
22 #define RTE_LPM6_TBL8_GROUP_NUM_ENTRIES         256
23 #define RTE_LPM6_TBL8_MAX_NUM_GROUPS      (1 << 21)
24 
25 #define RTE_LPM6_VALID_EXT_ENTRY_BITMASK 0xA0000000
26 #define RTE_LPM6_LOOKUP_SUCCESS          0x20000000
27 #define RTE_LPM6_TBL8_BITMASK            0x001FFFFF
28 
29 #define ADD_FIRST_BYTE                            3
30 #define LOOKUP_FIRST_BYTE                         4
31 #define BYTE_SIZE                                 8
32 #define BYTES2_SIZE                              16
33 
34 #define RULE_HASH_TABLE_EXTRA_SPACE              64
35 #define TBL24_IND                        UINT32_MAX
36 
37 #define lpm6_tbl8_gindex next_hop
38 
39 /** Flags for setting an entry as valid/invalid. */
40 enum valid_flag {
41 	INVALID = 0,
42 	VALID
43 };
44 
45 #if 0
46 TAILQ_HEAD(rte_lpm6_list, rte_tailq_entry);
47 
48 static struct rte_tailq_elem rte_lpm6_tailq = {
49 	.name = "RTE_LPM6",
50 };
51 EAL_REGISTER_TAILQ(rte_lpm6_tailq)
52 #endif
53 
54 /** Tbl entry structure. It is the same for both tbl24 and tbl8 */
55 struct rte_lpm6_tbl_entry {
56 	uint32_t next_hop:	21;  /**< Next hop / next table to be checked. */
57 	uint32_t depth	:8;      /**< Rule depth. */
58 
59 	/* Flags. */
60 	uint32_t valid     :1;   /**< Validation flag. */
61 	uint32_t valid_group :1; /**< Group validation flag. */
62 	uint32_t ext_entry :1;   /**< External entry. */
63 };
64 
65 /** Rules tbl entry structure. */
66 struct rte_lpm6_rule {
67 	uint8_t ip[RTE_LPM6_IPV6_ADDR_SIZE]; /**< Rule IP address. */
68 	uint32_t next_hop; /**< Rule next hop. */
69 	uint8_t depth; /**< Rule depth. */
70 };
71 
72 /** Rules tbl entry key. */
73 struct rte_lpm6_rule_key {
74 	uint8_t ip[RTE_LPM6_IPV6_ADDR_SIZE]; /**< Rule IP address. */
75 	uint8_t depth; /**< Rule depth. */
76 };
77 
78 /* Header of tbl8 */
79 struct rte_lpm_tbl8_hdr {
80 	uint32_t owner_tbl_ind; /**< owner table: TBL24_IND if owner is tbl24,
81 				  *  otherwise index of tbl8
82 				  */
83 	uint32_t owner_entry_ind; /**< index of the owner table entry where
84 				    *  pointer to the tbl8 is stored
85 				    */
86 	uint32_t ref_cnt; /**< table reference counter */
87 };
88 
89 /** LPM6 structure. */
90 struct rte_lpm6 {
91 	struct rte_lpm6_external ext;	/* Storage used by the algo wrapper */
92 	/* LPM metadata. */
93 	char name[RTE_LPM6_NAMESIZE];    /**< Name of the lpm. */
94 	uint32_t max_rules;              /**< Max number of rules. */
95 	uint32_t used_rules;             /**< Used rules so far. */
96 	uint32_t number_tbl8s;           /**< Number of tbl8s to allocate. */
97 
98 	/* LPM Tables. */
99 	//struct rte_hash *rules_tbl; /**< LPM rules. */
100 	struct rte_lpm6_tbl_entry tbl24[RTE_LPM6_TBL24_NUM_ENTRIES]
101 			__rte_cache_aligned; /**< LPM tbl24 table. */
102 
103 	uint32_t *tbl8_pool; /**< pool of indexes of free tbl8s */
104 	uint32_t tbl8_pool_pos; /**< current position in the tbl8 pool */
105 
106 	struct rte_lpm_tbl8_hdr *tbl8_hdrs; /* array of tbl8 headers */
107 
108 	struct rte_lpm6_tbl_entry tbl8[0]
109 			__rte_cache_aligned; /**< LPM tbl8 table. */
110 };
111 
112 /*
113  * Takes an array of uint8_t (IPv6 address) and masks it using the depth.
114  * It leaves untouched one bit per unit in the depth variable
115  * and set the rest to 0.
116  */
117 static inline void
ip6_mask_addr(uint8_t * ip,uint8_t depth)118 ip6_mask_addr(uint8_t *ip, uint8_t depth)
119 {
120 	int16_t part_depth, mask;
121 	int i;
122 
123 	part_depth = depth;
124 
125 	for (i = 0; i < RTE_LPM6_IPV6_ADDR_SIZE; i++) {
126 		if (part_depth < BYTE_SIZE && part_depth >= 0) {
127 			mask = (uint16_t)(~(UINT8_MAX >> part_depth));
128 			ip[i] = (uint8_t)(ip[i] & mask);
129 		} else if (part_depth < 0)
130 			ip[i] = 0;
131 
132 		part_depth -= BYTE_SIZE;
133 	}
134 }
135 
136 /* copy ipv6 address */
137 static inline void
ip6_copy_addr(uint8_t * dst,const uint8_t * src)138 ip6_copy_addr(uint8_t *dst, const uint8_t *src)
139 {
140 	rte_memcpy(dst, src, RTE_LPM6_IPV6_ADDR_SIZE);
141 }
142 
143 #if 0
144 /*
145  * LPM6 rule hash function
146  *
147  * It's used as a hash function for the rte_hash
148  *	containing rules
149  */
150 static inline uint32_t
151 rule_hash(const void *data, __rte_unused uint32_t data_len,
152 		  uint32_t init_val)
153 {
154 	return rte_jhash(data, sizeof(struct rte_lpm6_rule_key), init_val);
155 }
156 #endif
157 
158 /*
159  * Init pool of free tbl8 indexes
160  */
161 static void
tbl8_pool_init(struct rte_lpm6 * lpm)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
tbl8_get(struct rte_lpm6 * lpm,uint32_t * tbl8_ind)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
tbl8_put(struct rte_lpm6 * lpm,uint32_t tbl8_ind)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
tbl8_available(struct rte_lpm6 * lpm)206 tbl8_available(struct rte_lpm6 *lpm)
207 {
208 	return lpm->number_tbl8s - lpm->tbl8_pool_pos;
209 }
210 
211 #if 0
212 /*
213  * Init a rule key.
214  *	  note that ip must be already masked
215  */
216 static inline void
217 rule_key_init(struct rte_lpm6_rule_key *key, uint8_t *ip, uint8_t depth)
218 {
219 	ip6_copy_addr(key->ip, ip);
220 	key->depth = depth;
221 }
222 
223 /*
224  * Rebuild the entire LPM tree by reinserting all rules
225  */
226 static void
227 rebuild_lpm(struct rte_lpm6 *lpm)
228 {
229 	uint64_t next_hop;
230 	struct rte_lpm6_rule_key *rule_key;
231 	uint32_t iter = 0;
232 
233 	while (rte_hash_iterate(lpm->rules_tbl, (void *) &rule_key,
234 			(void **) &next_hop, &iter) >= 0)
235 		rte_lpm6_add(lpm, rule_key->ip, rule_key->depth,
236 			(uint32_t) next_hop);
237 }
238 #endif
239 
240 /*
241  * Allocates memory for LPM object
242  */
243 struct rte_lpm6 *
rte_lpm6_create(const char * name,int socket_id,const struct rte_lpm6_config * config)244 rte_lpm6_create(const char *name, int socket_id,
245 		const struct rte_lpm6_config *config)
246 {
247 	char mem_name[RTE_LPM6_NAMESIZE];
248 	struct rte_lpm6 *lpm = NULL;
249 	//struct rte_tailq_entry *te;
250 	uint64_t mem_size;
251 	//struct rte_lpm6_list *lpm_list;
252 	//struct rte_hash *rules_tbl = NULL;
253 	uint32_t *tbl8_pool = NULL;
254 	struct rte_lpm_tbl8_hdr *tbl8_hdrs = NULL;
255 
256 	//lpm_list = RTE_TAILQ_CAST(rte_lpm6_tailq.head, rte_lpm6_list);
257 
258 	RTE_BUILD_BUG_ON(sizeof(struct rte_lpm6_tbl_entry) != sizeof(uint32_t));
259 
260 	/* Check user arguments. */
261 	if ((name == NULL) || (socket_id < -1) || (config == NULL) ||
262 			config->number_tbl8s > RTE_LPM6_TBL8_MAX_NUM_GROUPS) {
263 		rte_errno = EINVAL;
264 		return NULL;
265 	}
266 
267 #if 0
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 		RTE_LOG(ERR, LPM, "LPM rules hash table allocation failed: %s (%d)",
285 				  rte_strerror(rte_errno), rte_errno);
286 		goto fail_wo_unlock;
287 	}
288 #endif
289 
290 	/* allocate tbl8 indexes pool */
291 	tbl8_pool = rte_malloc(NULL,
292 			sizeof(uint32_t) * config->number_tbl8s,
293 			RTE_CACHE_LINE_SIZE);
294 	if (tbl8_pool == NULL) {
295 		RTE_LOG(ERR, LPM, "LPM tbl8 pool allocation failed: %s (%d)",
296 				  rte_strerror(rte_errno), rte_errno);
297 		rte_errno = ENOMEM;
298 		goto fail_wo_unlock;
299 	}
300 
301 	/* allocate tbl8 headers */
302 	tbl8_hdrs = rte_malloc(NULL,
303 			sizeof(struct rte_lpm_tbl8_hdr) * config->number_tbl8s,
304 			RTE_CACHE_LINE_SIZE);
305 	if (tbl8_hdrs == NULL) {
306 		RTE_LOG(ERR, LPM, "LPM tbl8 headers allocation failed: %s (%d)",
307 				  rte_strerror(rte_errno), rte_errno);
308 		rte_errno = ENOMEM;
309 		goto fail_wo_unlock;
310 	}
311 
312 	snprintf(mem_name, sizeof(mem_name), "LPM_%s", name);
313 
314 	/* Determine the amount of memory to allocate. */
315 	mem_size = sizeof(*lpm) + (sizeof(lpm->tbl8[0]) *
316 			RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * config->number_tbl8s);
317 
318 #if 0
319 	rte_mcfg_tailq_write_lock();
320 
321 	/* Guarantee there's no existing */
322 	TAILQ_FOREACH(te, lpm_list, next) {
323 		lpm = (struct rte_lpm6 *) te->data;
324 		if (strncmp(name, lpm->name, RTE_LPM6_NAMESIZE) == 0)
325 			break;
326 	}
327 	lpm = NULL;
328 	if (te != NULL) {
329 		rte_errno = EEXIST;
330 		goto fail;
331 	}
332 
333 	/* allocate tailq entry */
334 	te = rte_zmalloc("LPM6_TAILQ_ENTRY", sizeof(*te), 0);
335 	if (te == NULL) {
336 		RTE_LOG(ERR, LPM, "Failed to allocate tailq entry!\n");
337 		rte_errno = ENOMEM;
338 		goto fail;
339 	}
340 #endif
341 
342 	/* Allocate memory to store the LPM data structures. */
343 	lpm = rte_zmalloc_socket(mem_name, (size_t)mem_size,
344 			RTE_CACHE_LINE_SIZE, socket_id);
345 
346 	if (lpm == NULL) {
347 		RTE_LOG(ERR, LPM, "LPM memory allocation failed\n");
348 		//rte_free(te);
349 		rte_errno = ENOMEM;
350 		goto fail;
351 	}
352 
353 	/* Save user arguments. */
354 	//lpm->max_rules = config->max_rules;
355 	lpm->number_tbl8s = config->number_tbl8s;
356 	strlcpy(lpm->name, name, sizeof(lpm->name));
357 	//lpm->rules_tbl = rules_tbl;
358 	lpm->tbl8_pool = tbl8_pool;
359 	lpm->tbl8_hdrs = tbl8_hdrs;
360 
361 	/* init the stack */
362 	tbl8_pool_init(lpm);
363 
364 	//te->data = (void *) lpm;
365 
366 	//TAILQ_INSERT_TAIL(lpm_list, te, next);
367 	rte_mcfg_tailq_write_unlock();
368 	return lpm;
369 
370 fail:
371 	rte_mcfg_tailq_write_unlock();
372 
373 fail_wo_unlock:
374 	rte_free(tbl8_hdrs);
375 	rte_free(tbl8_pool);
376 	//rte_hash_free(rules_tbl);
377 
378 	return NULL;
379 }
380 
381 #if 0
382 /*
383  * Find an existing lpm table and return a pointer to it.
384  */
385 struct rte_lpm6 *
386 rte_lpm6_find_existing(const char *name)
387 {
388 	struct rte_lpm6 *l = NULL;
389 	struct rte_tailq_entry *te;
390 	struct rte_lpm6_list *lpm_list;
391 
392 	lpm_list = RTE_TAILQ_CAST(rte_lpm6_tailq.head, rte_lpm6_list);
393 
394 	rte_mcfg_tailq_read_lock();
395 	TAILQ_FOREACH(te, lpm_list, next) {
396 		l = (struct rte_lpm6 *) te->data;
397 		if (strncmp(name, l->name, RTE_LPM6_NAMESIZE) == 0)
398 			break;
399 	}
400 	rte_mcfg_tailq_read_unlock();
401 
402 	if (te == NULL) {
403 		rte_errno = ENOENT;
404 		return NULL;
405 	}
406 
407 	return l;
408 }
409 #endif
410 
411 /*
412  * Deallocates memory for given LPM table.
413  */
414 void
rte_lpm6_free(struct rte_lpm6 * lpm)415 rte_lpm6_free(struct rte_lpm6 *lpm)
416 {
417 #if 0
418 	struct rte_lpm6_list *lpm_list;
419 	struct rte_tailq_entry *te;
420 
421 	/* Check user arguments. */
422 	if (lpm == NULL)
423 		return;
424 
425 	lpm_list = RTE_TAILQ_CAST(rte_lpm6_tailq.head, rte_lpm6_list);
426 
427 	rte_mcfg_tailq_write_lock();
428 
429 	/* find our tailq entry */
430 	TAILQ_FOREACH(te, lpm_list, next) {
431 		if (te->data == (void *) lpm)
432 			break;
433 	}
434 
435 	if (te != NULL)
436 		TAILQ_REMOVE(lpm_list, te, next);
437 
438 	rte_mcfg_tailq_write_unlock();
439 #endif
440 
441 	rte_free(lpm->tbl8_hdrs);
442 	rte_free(lpm->tbl8_pool);
443 	//rte_hash_free(lpm->rules_tbl);
444 	rte_free(lpm);
445 	//rte_free(te);
446 }
447 
448 #if 0
449 /* Find a rule */
450 static inline int
451 rule_find_with_key(struct rte_lpm6 *lpm,
452 		  const struct rte_lpm6_rule_key *rule_key,
453 		  uint32_t *next_hop)
454 {
455 	uint64_t hash_val;
456 	int ret;
457 
458 	/* lookup for a rule */
459 	ret = rte_hash_lookup_data(lpm->rules_tbl, (const void *) rule_key,
460 		(void **) &hash_val);
461 	if (ret >= 0) {
462 		*next_hop = (uint32_t) hash_val;
463 		return 1;
464 	}
465 
466 	return 0;
467 }
468 
469 /* Find a rule */
470 static int
471 rule_find(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth,
472 		  uint32_t *next_hop)
473 {
474 	struct rte_lpm6_rule_key rule_key;
475 
476 	/* init a rule key */
477 	rule_key_init(&rule_key, ip, depth);
478 
479 	return rule_find_with_key(lpm, &rule_key, next_hop);
480 }
481 
482 /*
483  * Checks if a rule already exists in the rules table and updates
484  * the nexthop if so. Otherwise it adds a new rule if enough space is available.
485  *
486  * Returns:
487  *    0 - next hop of existed rule is updated
488  *    1 - new rule successfully added
489  *   <0 - error
490  */
491 static inline int
492 rule_add(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth, uint32_t next_hop)
493 {
494 	int ret, rule_exist;
495 	struct rte_lpm6_rule_key rule_key;
496 	uint32_t unused;
497 
498 	/* init a rule key */
499 	rule_key_init(&rule_key, ip, depth);
500 
501 	/* Scan through rule list to see if rule already exists. */
502 	rule_exist = rule_find_with_key(lpm, &rule_key, &unused);
503 
504 	/*
505 	 * If rule does not exist check if there is space to add a new rule to
506 	 * this rule group. If there is no space return error.
507 	 */
508 	if (!rule_exist && lpm->used_rules == lpm->max_rules)
509 		return -ENOSPC;
510 
511 	/* add the rule or update rules next hop */
512 	ret = rte_hash_add_key_data(lpm->rules_tbl, &rule_key,
513 		(void *)(uintptr_t) next_hop);
514 	if (ret < 0)
515 		return ret;
516 
517 	/* Increment the used rules counter for this rule group. */
518 	if (!rule_exist) {
519 		lpm->used_rules++;
520 		return 1;
521 	}
522 
523 	return 0;
524 }
525 #endif
526 
527 /*
528  * Function that expands a rule across the data structure when a less-generic
529  * one has been added before. It assures that every possible combination of bits
530  * in the IP address returns a match.
531  */
532 static void
expand_rule(struct rte_lpm6 * lpm,uint32_t tbl8_gindex,uint8_t old_depth,uint8_t new_depth,uint32_t next_hop,uint8_t valid)533 expand_rule(struct rte_lpm6 *lpm, uint32_t tbl8_gindex, uint8_t old_depth,
534 		uint8_t new_depth, uint32_t next_hop, uint8_t valid)
535 {
536 	uint32_t tbl8_group_end, tbl8_gindex_next, j;
537 
538 	tbl8_group_end = tbl8_gindex + RTE_LPM6_TBL8_GROUP_NUM_ENTRIES;
539 
540 	struct rte_lpm6_tbl_entry new_tbl8_entry = {
541 		.valid = valid,
542 		.valid_group = valid,
543 		.depth = new_depth,
544 		.next_hop = next_hop,
545 		.ext_entry = 0,
546 	};
547 
548 	for (j = tbl8_gindex; j < tbl8_group_end; j++) {
549 		if (!lpm->tbl8[j].valid || (lpm->tbl8[j].ext_entry == 0
550 				&& lpm->tbl8[j].depth <= old_depth)) {
551 
552 			lpm->tbl8[j] = new_tbl8_entry;
553 
554 		} else if (lpm->tbl8[j].ext_entry == 1) {
555 
556 			tbl8_gindex_next = lpm->tbl8[j].lpm6_tbl8_gindex
557 					* RTE_LPM6_TBL8_GROUP_NUM_ENTRIES;
558 			expand_rule(lpm, tbl8_gindex_next, old_depth, new_depth,
559 					next_hop, valid);
560 		}
561 	}
562 }
563 
564 /*
565  * Init a tbl8 header
566  */
567 static inline void
init_tbl8_header(struct rte_lpm6 * lpm,uint32_t tbl_ind,uint32_t owner_tbl_ind,uint32_t owner_entry_ind)568 init_tbl8_header(struct rte_lpm6 *lpm, uint32_t tbl_ind,
569 		uint32_t owner_tbl_ind, uint32_t owner_entry_ind)
570 {
571 	struct rte_lpm_tbl8_hdr *tbl_hdr = &lpm->tbl8_hdrs[tbl_ind];
572 	tbl_hdr->owner_tbl_ind = owner_tbl_ind;
573 	tbl_hdr->owner_entry_ind = owner_entry_ind;
574 	tbl_hdr->ref_cnt = 0;
575 }
576 
577 /*
578  * Calculate index to the table based on the number and position
579  * of the bytes being inspected in this step.
580  */
581 static uint32_t
get_bitshift(const uint8_t * ip,uint8_t first_byte,uint8_t bytes)582 get_bitshift(const uint8_t *ip, uint8_t first_byte, uint8_t bytes)
583 {
584 	uint32_t entry_ind, i;
585 	int8_t bitshift;
586 
587 	entry_ind = 0;
588 	for (i = first_byte; i < (uint32_t)(first_byte + bytes); i++) {
589 		bitshift = (int8_t)((bytes - i)*BYTE_SIZE);
590 
591 		if (bitshift < 0)
592 			bitshift = 0;
593 		entry_ind = entry_ind | ip[i-1] << bitshift;
594 	}
595 
596 	return entry_ind;
597 }
598 
599 /*
600  * Simulate adding a new route to the LPM counting number
601  * of new tables that will be needed
602  *
603  * It returns 0 on success, or 1 if
604  * the process needs to be continued by calling the function again.
605  */
606 static inline int
simulate_add_step(struct rte_lpm6 * lpm,struct rte_lpm6_tbl_entry * tbl,struct rte_lpm6_tbl_entry ** next_tbl,const uint8_t * ip,uint8_t bytes,uint8_t first_byte,uint8_t depth,uint32_t * need_tbl_nb)607 simulate_add_step(struct rte_lpm6 *lpm, struct rte_lpm6_tbl_entry *tbl,
608 		struct rte_lpm6_tbl_entry **next_tbl, const uint8_t *ip,
609 		uint8_t bytes, uint8_t first_byte, uint8_t depth,
610 		uint32_t *need_tbl_nb)
611 {
612 	uint32_t entry_ind;
613 	uint8_t bits_covered;
614 	uint32_t next_tbl_ind;
615 
616 	/*
617 	 * Calculate index to the table based on the number and position
618 	 * of the bytes being inspected in this step.
619 	 */
620 	entry_ind = get_bitshift(ip, first_byte, bytes);
621 
622 	/* Number of bits covered in this step */
623 	bits_covered = (uint8_t)((bytes+first_byte-1)*BYTE_SIZE);
624 
625 	if (depth <= bits_covered) {
626 		*need_tbl_nb = 0;
627 		return 0;
628 	}
629 
630 	if (tbl[entry_ind].valid == 0 || tbl[entry_ind].ext_entry == 0) {
631 		/* from this point on a new table is needed on each level
632 		 * that is not covered yet
633 		 */
634 		depth -= bits_covered;
635 		uint32_t cnt = depth >> 3; /* depth / BYTE_SIZE */
636 		if (depth & 7) /* 0b00000111 */
637 			/* if depth % 8 > 0 then one more table is needed
638 			 * for those last bits
639 			 */
640 			cnt++;
641 
642 		*need_tbl_nb = cnt;
643 		return 0;
644 	}
645 
646 	next_tbl_ind = tbl[entry_ind].lpm6_tbl8_gindex;
647 	*next_tbl = &(lpm->tbl8[next_tbl_ind *
648 		RTE_LPM6_TBL8_GROUP_NUM_ENTRIES]);
649 	*need_tbl_nb = 0;
650 	return 1;
651 }
652 
653 /*
654  * Partially adds a new route to the data structure (tbl24+tbl8s).
655  * It returns 0 on success, a negative number on failure, or 1 if
656  * the process needs to be continued by calling the function again.
657  */
658 static inline int
add_step(struct rte_lpm6 * lpm,struct rte_lpm6_tbl_entry * tbl,uint32_t tbl_ind,struct rte_lpm6_tbl_entry ** next_tbl,uint32_t * next_tbl_ind,uint8_t * ip,uint8_t bytes,uint8_t first_byte,uint8_t depth,uint32_t next_hop,uint8_t is_new_rule)659 add_step(struct rte_lpm6 *lpm, struct rte_lpm6_tbl_entry *tbl,
660 		uint32_t tbl_ind, struct rte_lpm6_tbl_entry **next_tbl,
661 		uint32_t *next_tbl_ind, uint8_t *ip, uint8_t bytes,
662 		uint8_t first_byte, uint8_t depth, uint32_t next_hop,
663 		uint8_t is_new_rule)
664 {
665 	uint32_t entry_ind, tbl_range, tbl8_group_start, tbl8_group_end, i;
666 	uint32_t tbl8_gindex;
667 	uint8_t bits_covered;
668 	int ret;
669 
670 	/*
671 	 * Calculate index to the table based on the number and position
672 	 * of the bytes being inspected in this step.
673 	 */
674 	entry_ind = get_bitshift(ip, first_byte, bytes);
675 
676 	/* Number of bits covered in this step */
677 	bits_covered = (uint8_t)((bytes+first_byte-1)*BYTE_SIZE);
678 
679 	/*
680 	 * If depth if smaller than this number (ie this is the last step)
681 	 * expand the rule across the relevant positions in the table.
682 	 */
683 	if (depth <= bits_covered) {
684 		tbl_range = 1 << (bits_covered - depth);
685 
686 		for (i = entry_ind; i < (entry_ind + tbl_range); i++) {
687 			if (!tbl[i].valid || (tbl[i].ext_entry == 0 &&
688 					tbl[i].depth <= depth)) {
689 
690 				struct rte_lpm6_tbl_entry new_tbl_entry = {
691 					.next_hop = next_hop,
692 					.depth = depth,
693 					.valid = VALID,
694 					.valid_group = VALID,
695 					.ext_entry = 0,
696 				};
697 
698 				tbl[i] = new_tbl_entry;
699 
700 			} else if (tbl[i].ext_entry == 1) {
701 
702 				/*
703 				 * If tbl entry is valid and extended calculate the index
704 				 * into next tbl8 and expand the rule across the data structure.
705 				 */
706 				tbl8_gindex = tbl[i].lpm6_tbl8_gindex *
707 						RTE_LPM6_TBL8_GROUP_NUM_ENTRIES;
708 				expand_rule(lpm, tbl8_gindex, depth, depth,
709 						next_hop, VALID);
710 			}
711 		}
712 
713 		/* update tbl8 rule reference counter */
714 		if (tbl_ind != TBL24_IND && is_new_rule)
715 			lpm->tbl8_hdrs[tbl_ind].ref_cnt++;
716 
717 		return 0;
718 	}
719 	/*
720 	 * If this is not the last step just fill one position
721 	 * and calculate the index to the next table.
722 	 */
723 	else {
724 		/* If it's invalid a new tbl8 is needed */
725 		if (!tbl[entry_ind].valid) {
726 			/* get a new table */
727 			ret = tbl8_get(lpm, &tbl8_gindex);
728 			if (ret != 0)
729 				return -ENOSPC;
730 
731 			/* invalidate all new tbl8 entries */
732 			tbl8_group_start = tbl8_gindex *
733 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES;
734 			memset(&lpm->tbl8[tbl8_group_start], 0,
735 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES *
736 					sizeof(struct rte_lpm6_tbl_entry));
737 
738 			/* init the new table's header:
739 			 *   save the reference to the owner table
740 			 */
741 			init_tbl8_header(lpm, tbl8_gindex, tbl_ind, entry_ind);
742 
743 			/* reference to a new tbl8 */
744 			struct rte_lpm6_tbl_entry new_tbl_entry = {
745 				.lpm6_tbl8_gindex = tbl8_gindex,
746 				.depth = 0,
747 				.valid = VALID,
748 				.valid_group = VALID,
749 				.ext_entry = 1,
750 			};
751 
752 			tbl[entry_ind] = new_tbl_entry;
753 
754 			/* update the current table's reference counter */
755 			if (tbl_ind != TBL24_IND)
756 				lpm->tbl8_hdrs[tbl_ind].ref_cnt++;
757 		}
758 		/*
759 		 * If it's valid but not extended the rule that was stored
760 		 * here needs to be moved to the next table.
761 		 */
762 		else if (tbl[entry_ind].ext_entry == 0) {
763 			/* get a new tbl8 index */
764 			ret = tbl8_get(lpm, &tbl8_gindex);
765 			if (ret != 0)
766 				return -ENOSPC;
767 
768 			tbl8_group_start = tbl8_gindex *
769 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES;
770 			tbl8_group_end = tbl8_group_start +
771 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES;
772 
773 			struct rte_lpm6_tbl_entry tbl_entry = {
774 				.next_hop = tbl[entry_ind].next_hop,
775 				.depth = tbl[entry_ind].depth,
776 				.valid = VALID,
777 				.valid_group = VALID,
778 				.ext_entry = 0
779 			};
780 
781 			/* Populate new tbl8 with tbl value. */
782 			for (i = tbl8_group_start; i < tbl8_group_end; i++)
783 				lpm->tbl8[i] = tbl_entry;
784 
785 			/* init the new table's header:
786 			 *   save the reference to the owner table
787 			 */
788 			init_tbl8_header(lpm, tbl8_gindex, tbl_ind, entry_ind);
789 
790 			/*
791 			 * Update tbl entry to point to new tbl8 entry. Note: The
792 			 * ext_flag and tbl8_index need to be updated simultaneously,
793 			 * so assign whole structure in one go.
794 			 */
795 			struct rte_lpm6_tbl_entry new_tbl_entry = {
796 				.lpm6_tbl8_gindex = tbl8_gindex,
797 				.depth = 0,
798 				.valid = VALID,
799 				.valid_group = VALID,
800 				.ext_entry = 1,
801 			};
802 
803 			tbl[entry_ind] = new_tbl_entry;
804 
805 			/* update the current table's reference counter */
806 			if (tbl_ind != TBL24_IND)
807 				lpm->tbl8_hdrs[tbl_ind].ref_cnt++;
808 		}
809 
810 		*next_tbl_ind = tbl[entry_ind].lpm6_tbl8_gindex;
811 		*next_tbl = &(lpm->tbl8[*next_tbl_ind *
812 				  RTE_LPM6_TBL8_GROUP_NUM_ENTRIES]);
813 	}
814 
815 	return 1;
816 }
817 
818 /*
819  * Simulate adding a route to LPM
820  *
821  *	Returns:
822  *    0 on success
823  *    -ENOSPC not enough tbl8 left
824  */
825 static int
simulate_add(struct rte_lpm6 * lpm,const uint8_t * masked_ip,uint8_t depth)826 simulate_add(struct rte_lpm6 *lpm, const uint8_t *masked_ip, uint8_t depth)
827 {
828 	struct rte_lpm6_tbl_entry *tbl;
829 	struct rte_lpm6_tbl_entry *tbl_next = NULL;
830 	int ret, i;
831 
832 	/* number of new tables needed for a step */
833 	uint32_t need_tbl_nb;
834 	/* total number of new tables needed */
835 	uint32_t total_need_tbl_nb;
836 
837 	/* Inspect the first three bytes through tbl24 on the first step. */
838 	ret = simulate_add_step(lpm, lpm->tbl24, &tbl_next, masked_ip,
839 		ADD_FIRST_BYTE, 1, depth, &need_tbl_nb);
840 	total_need_tbl_nb = need_tbl_nb;
841 	/*
842 	 * Inspect one by one the rest of the bytes until
843 	 * the process is completed.
844 	 */
845 	for (i = ADD_FIRST_BYTE; i < RTE_LPM6_IPV6_ADDR_SIZE && ret == 1; i++) {
846 		tbl = tbl_next;
847 		ret = simulate_add_step(lpm, tbl, &tbl_next, masked_ip, 1,
848 			(uint8_t)(i + 1), depth, &need_tbl_nb);
849 		total_need_tbl_nb += need_tbl_nb;
850 	}
851 
852 	if (tbl8_available(lpm) < total_need_tbl_nb)
853 		/* not enough tbl8 to add a rule */
854 		return -ENOSPC;
855 
856 	return 0;
857 }
858 
859 /*
860  * Add a route
861  */
862 int
rte_lpm6_add(struct rte_lpm6 * lpm,const uint8_t * ip,uint8_t depth,uint32_t next_hop,int is_new_rule)863 rte_lpm6_add(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth,
864 	     uint32_t next_hop, int is_new_rule)
865 {
866 	struct rte_lpm6_tbl_entry *tbl;
867 	struct rte_lpm6_tbl_entry *tbl_next = NULL;
868 	/* init to avoid compiler warning */
869 	uint32_t tbl_next_num = 123456;
870 	int status;
871 	uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE];
872 	int i;
873 
874 	/* Check user arguments. */
875 	if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM6_MAX_DEPTH))
876 		return -EINVAL;
877 
878 	/* Copy the IP and mask it to avoid modifying user's input data. */
879 	ip6_copy_addr(masked_ip, ip);
880 	ip6_mask_addr(masked_ip, depth);
881 
882 	/* Simulate adding a new route */
883 	int ret = simulate_add(lpm, masked_ip, depth);
884 	if (ret < 0)
885 		return ret;
886 
887 #if 0
888 	/* Add the rule to the rule table. */
889 	int is_new_rule = rule_add(lpm, masked_ip, depth, next_hop);
890 	/* If there is no space available for new rule return error. */
891 	if (is_new_rule < 0)
892 		return is_new_rule;
893 #endif
894 
895 	/* Inspect the first three bytes through tbl24 on the first step. */
896 	tbl = lpm->tbl24;
897 	status = add_step(lpm, tbl, TBL24_IND, &tbl_next, &tbl_next_num,
898 		masked_ip, ADD_FIRST_BYTE, 1, depth, next_hop,
899 		is_new_rule);
900 	assert(status >= 0);
901 
902 	/*
903 	 * Inspect one by one the rest of the bytes until
904 	 * the process is completed.
905 	 */
906 	for (i = ADD_FIRST_BYTE; i < RTE_LPM6_IPV6_ADDR_SIZE && status == 1; i++) {
907 		tbl = tbl_next;
908 		status = add_step(lpm, tbl, tbl_next_num, &tbl_next,
909 			&tbl_next_num, masked_ip, 1, (uint8_t)(i + 1),
910 			depth, next_hop, is_new_rule);
911 		assert(status >= 0);
912 	}
913 
914 	return status;
915 }
916 
917 /*
918  * Takes a pointer to a table entry and inspect one level.
919  * The function returns 0 on lookup success, ENOENT if no match was found
920  * or 1 if the process needs to be continued by calling the function again.
921  */
922 static inline int
lookup_step(const struct rte_lpm6 * lpm,const struct rte_lpm6_tbl_entry * tbl,const struct rte_lpm6_tbl_entry ** tbl_next,const uint8_t * ip,uint8_t first_byte,uint32_t * next_hop)923 lookup_step(const struct rte_lpm6 *lpm, const struct rte_lpm6_tbl_entry *tbl,
924 		const struct rte_lpm6_tbl_entry **tbl_next, const uint8_t *ip,
925 		uint8_t first_byte, uint32_t *next_hop)
926 {
927 	uint32_t tbl8_index, tbl_entry;
928 
929 	/* Take the integer value from the pointer. */
930 	tbl_entry = *(const uint32_t *)tbl;
931 
932 	/* If it is valid and extended we calculate the new pointer to return. */
933 	if ((tbl_entry & RTE_LPM6_VALID_EXT_ENTRY_BITMASK) ==
934 			RTE_LPM6_VALID_EXT_ENTRY_BITMASK) {
935 
936 		tbl8_index = ip[first_byte-1] +
937 				((tbl_entry & RTE_LPM6_TBL8_BITMASK) *
938 				RTE_LPM6_TBL8_GROUP_NUM_ENTRIES);
939 
940 		*tbl_next = &lpm->tbl8[tbl8_index];
941 
942 		return 1;
943 	} else {
944 		/* If not extended then we can have a match. */
945 		*next_hop = ((uint32_t)tbl_entry & RTE_LPM6_TBL8_BITMASK);
946 		return (tbl_entry & RTE_LPM6_LOOKUP_SUCCESS) ? 0 : -ENOENT;
947 	}
948 }
949 
950 /*
951  * Looks up an IP
952  */
953 int
rte_lpm6_lookup(const struct rte_lpm6 * lpm,const uint8_t * ip,uint32_t * next_hop)954 rte_lpm6_lookup(const struct rte_lpm6 *lpm, const uint8_t *ip,
955 		uint32_t *next_hop)
956 {
957 	const struct rte_lpm6_tbl_entry *tbl;
958 	const struct rte_lpm6_tbl_entry *tbl_next = NULL;
959 	int status;
960 	uint8_t first_byte;
961 	uint32_t tbl24_index;
962 
963 	/* DEBUG: Check user input arguments. */
964 	if ((lpm == NULL) || (ip == NULL) || (next_hop == NULL))
965 		return -EINVAL;
966 
967 	first_byte = LOOKUP_FIRST_BYTE;
968 	tbl24_index = (ip[0] << BYTES2_SIZE) | (ip[1] << BYTE_SIZE) | ip[2];
969 
970 	/* Calculate pointer to the first entry to be inspected */
971 	tbl = &lpm->tbl24[tbl24_index];
972 
973 	do {
974 		/* Continue inspecting following levels until success or failure */
975 		status = lookup_step(lpm, tbl, &tbl_next, ip, first_byte++, next_hop);
976 		tbl = tbl_next;
977 	} while (status == 1);
978 
979 	return status;
980 }
981 
982 /*
983  * Looks up a group of IP addresses
984  */
985 int
rte_lpm6_lookup_bulk_func(const struct rte_lpm6 * lpm,uint8_t ips[][RTE_LPM6_IPV6_ADDR_SIZE],int32_t * next_hops,unsigned int n)986 rte_lpm6_lookup_bulk_func(const struct rte_lpm6 *lpm,
987 		uint8_t ips[][RTE_LPM6_IPV6_ADDR_SIZE],
988 		int32_t *next_hops, unsigned int n)
989 {
990 	unsigned int i;
991 	const struct rte_lpm6_tbl_entry *tbl;
992 	const struct rte_lpm6_tbl_entry *tbl_next = NULL;
993 	uint32_t tbl24_index, next_hop;
994 	uint8_t first_byte;
995 	int status;
996 
997 	/* DEBUG: Check user input arguments. */
998 	if ((lpm == NULL) || (ips == NULL) || (next_hops == NULL))
999 		return -EINVAL;
1000 
1001 	for (i = 0; i < n; i++) {
1002 		first_byte = LOOKUP_FIRST_BYTE;
1003 		tbl24_index = (ips[i][0] << BYTES2_SIZE) |
1004 				(ips[i][1] << BYTE_SIZE) | ips[i][2];
1005 
1006 		/* Calculate pointer to the first entry to be inspected */
1007 		tbl = &lpm->tbl24[tbl24_index];
1008 
1009 		do {
1010 			/* Continue inspecting following levels
1011 			 * until success or failure
1012 			 */
1013 			status = lookup_step(lpm, tbl, &tbl_next, ips[i],
1014 					first_byte++, &next_hop);
1015 			tbl = tbl_next;
1016 		} while (status == 1);
1017 
1018 		if (status < 0)
1019 			next_hops[i] = -1;
1020 		else
1021 			next_hops[i] = (int32_t)next_hop;
1022 	}
1023 
1024 	return 0;
1025 }
1026 
1027 struct rte_lpm6_rule *
fill_rule6(char * buffer,const uint8_t * ip,uint8_t depth,uint32_t next_hop)1028 fill_rule6(char *buffer, const uint8_t *ip, uint8_t depth, uint32_t next_hop)
1029 {
1030 	struct rte_lpm6_rule *rule = (struct rte_lpm6_rule *)buffer;
1031 
1032 	ip6_copy_addr((uint8_t *)&rule->ip, ip);
1033 	rule->depth = depth;
1034 	rule->next_hop = next_hop;
1035 
1036 	return (rule);
1037 }
1038 
1039 #if 0
1040 /*
1041  * Look for a rule in the high-level rules table
1042  */
1043 int
1044 rte_lpm6_is_rule_present(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth,
1045 			 uint32_t *next_hop)
1046 {
1047 	uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE];
1048 
1049 	/* Check user arguments. */
1050 	if ((lpm == NULL) || next_hop == NULL || ip == NULL ||
1051 			(depth < 1) || (depth > RTE_LPM6_MAX_DEPTH))
1052 		return -EINVAL;
1053 
1054 	/* Copy the IP and mask it to avoid modifying user's input data. */
1055 	ip6_copy_addr(masked_ip, ip);
1056 	ip6_mask_addr(masked_ip, depth);
1057 
1058 	return rule_find(lpm, masked_ip, depth, next_hop);
1059 }
1060 
1061 /*
1062  * Delete a rule from the rule table.
1063  * NOTE: Valid range for depth parameter is 1 .. 128 inclusive.
1064  * return
1065  *	  0 on success
1066  *   <0 on failure
1067  */
1068 static inline int
1069 rule_delete(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth)
1070 {
1071 	int ret;
1072 	struct rte_lpm6_rule_key rule_key;
1073 
1074 	/* init rule key */
1075 	rule_key_init(&rule_key, ip, depth);
1076 
1077 	/* delete the rule */
1078 	ret = rte_hash_del_key(lpm->rules_tbl, (void *) &rule_key);
1079 	if (ret >= 0)
1080 		lpm->used_rules--;
1081 
1082 	return ret;
1083 }
1084 
1085 /*
1086  * Deletes a group of rules
1087  *
1088  * Note that the function rebuilds the lpm table,
1089  * rather than doing incremental updates like
1090  * the regular delete function
1091  */
1092 int
1093 rte_lpm6_delete_bulk_func(struct rte_lpm6 *lpm,
1094 		uint8_t ips[][RTE_LPM6_IPV6_ADDR_SIZE], uint8_t *depths,
1095 		unsigned n)
1096 {
1097 	uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE];
1098 	unsigned i;
1099 
1100 	/* Check input arguments. */
1101 	if ((lpm == NULL) || (ips == NULL) || (depths == NULL))
1102 		return -EINVAL;
1103 
1104 	for (i = 0; i < n; i++) {
1105 		ip6_copy_addr(masked_ip, ips[i]);
1106 		ip6_mask_addr(masked_ip, depths[i]);
1107 		rule_delete(lpm, masked_ip, depths[i]);
1108 	}
1109 
1110 	/*
1111 	 * Set all the table entries to 0 (ie delete every rule
1112 	 * from the data structure.
1113 	 */
1114 	memset(lpm->tbl24, 0, sizeof(lpm->tbl24));
1115 	memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0])
1116 			* RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s);
1117 	tbl8_pool_init(lpm);
1118 
1119 	/*
1120 	 * Add every rule again (except for the ones that were removed from
1121 	 * the rules table).
1122 	 */
1123 	rebuild_lpm(lpm);
1124 
1125 	return 0;
1126 }
1127 
1128 /*
1129  * Delete all rules from the LPM table.
1130  */
1131 void
1132 rte_lpm6_delete_all(struct rte_lpm6 *lpm)
1133 {
1134 	/* Zero used rules counter. */
1135 	lpm->used_rules = 0;
1136 
1137 	/* Zero tbl24. */
1138 	memset(lpm->tbl24, 0, sizeof(lpm->tbl24));
1139 
1140 	/* Zero tbl8. */
1141 	memset(lpm->tbl8, 0, sizeof(lpm->tbl8[0]) *
1142 			RTE_LPM6_TBL8_GROUP_NUM_ENTRIES * lpm->number_tbl8s);
1143 
1144 	/* init pool of free tbl8 indexes */
1145 	tbl8_pool_init(lpm);
1146 
1147 	/* Delete all rules form the rules table. */
1148 	rte_hash_reset(lpm->rules_tbl);
1149 }
1150 #endif
1151 
1152 /*
1153  * Convert a depth to a one byte long mask
1154  *   Example: 4 will be converted to 0xF0
1155  */
1156 static uint8_t __attribute__((pure))
depth_to_mask_1b(uint8_t depth)1157 depth_to_mask_1b(uint8_t depth)
1158 {
1159 	/* To calculate a mask start with a 1 on the left hand side and right
1160 	 * shift while populating the left hand side with 1's
1161 	 */
1162 	return (signed char)0x80 >> (depth - 1);
1163 }
1164 
1165 #if 0
1166 /*
1167  * Find a less specific rule
1168  */
1169 static int
1170 rule_find_less_specific(struct rte_lpm6 *lpm, uint8_t *ip, uint8_t depth,
1171 	struct rte_lpm6_rule *rule)
1172 {
1173 	int ret;
1174 	uint32_t next_hop;
1175 	uint8_t mask;
1176 	struct rte_lpm6_rule_key rule_key;
1177 
1178 	if (depth == 1)
1179 		return 0;
1180 
1181 	rule_key_init(&rule_key, ip, depth);
1182 
1183 	while (depth > 1) {
1184 		depth--;
1185 
1186 		/* each iteration zero one more bit of the key */
1187 		mask = depth & 7; /* depth % BYTE_SIZE */
1188 		if (mask > 0)
1189 			mask = depth_to_mask_1b(mask);
1190 
1191 		rule_key.depth = depth;
1192 		rule_key.ip[depth >> 3] &= mask;
1193 
1194 		ret = rule_find_with_key(lpm, &rule_key, &next_hop);
1195 		if (ret) {
1196 			rule->depth = depth;
1197 			ip6_copy_addr(rule->ip, rule_key.ip);
1198 			rule->next_hop = next_hop;
1199 			return 1;
1200 		}
1201 	}
1202 
1203 	return 0;
1204 }
1205 #endif
1206 
1207 /*
1208  * Find range of tbl8 cells occupied by a rule
1209  */
1210 static void
rule_find_range(struct rte_lpm6 * lpm,const uint8_t * ip,uint8_t depth,struct rte_lpm6_tbl_entry ** from,struct rte_lpm6_tbl_entry ** to,uint32_t * out_tbl_ind)1211 rule_find_range(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth,
1212 		  struct rte_lpm6_tbl_entry **from,
1213 		  struct rte_lpm6_tbl_entry **to,
1214 		  uint32_t *out_tbl_ind)
1215 {
1216 	uint32_t ind;
1217 	uint32_t first_3bytes = (uint32_t)ip[0] << 16 | ip[1] << 8 | ip[2];
1218 
1219 	if (depth <= 24) {
1220 		/* rule is within the top level */
1221 		ind = first_3bytes;
1222 		*from = &lpm->tbl24[ind];
1223 		ind += (1 << (24 - depth)) - 1;
1224 		*to = &lpm->tbl24[ind];
1225 		*out_tbl_ind = TBL24_IND;
1226 	} else {
1227 		/* top level entry */
1228 		struct rte_lpm6_tbl_entry *tbl = &lpm->tbl24[first_3bytes];
1229 		assert(tbl->ext_entry == 1);
1230 		/* first tbl8 */
1231 		uint32_t tbl_ind = tbl->lpm6_tbl8_gindex;
1232 		tbl = &lpm->tbl8[tbl_ind *
1233 				RTE_LPM6_TBL8_GROUP_NUM_ENTRIES];
1234 		/* current ip byte, the top level is already behind */
1235 		uint8_t byte = 3;
1236 		/* minus top level */
1237 		depth -= 24;
1238 
1239 		/* iterate through levels (tbl8s)
1240 		 * until we reach the last one
1241 		 */
1242 		while (depth > 8) {
1243 			tbl += ip[byte];
1244 			assert(tbl->ext_entry == 1);
1245 			/* go to the next level/tbl8 */
1246 			tbl_ind = tbl->lpm6_tbl8_gindex;
1247 			tbl = &lpm->tbl8[tbl_ind *
1248 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES];
1249 			byte += 1;
1250 			depth -= 8;
1251 		}
1252 
1253 		/* last level/tbl8 */
1254 		ind = ip[byte] & depth_to_mask_1b(depth);
1255 		*from = &tbl[ind];
1256 		ind += (1 << (8 - depth)) - 1;
1257 		*to = &tbl[ind];
1258 		*out_tbl_ind = tbl_ind;
1259 	}
1260 }
1261 
1262 /*
1263  * Remove a table from the LPM tree
1264  */
1265 static void
remove_tbl(struct rte_lpm6 * lpm,struct rte_lpm_tbl8_hdr * tbl_hdr,uint32_t tbl_ind,struct rte_lpm6_rule * lsp_rule)1266 remove_tbl(struct rte_lpm6 *lpm, struct rte_lpm_tbl8_hdr *tbl_hdr,
1267 		  uint32_t tbl_ind, struct rte_lpm6_rule *lsp_rule)
1268 {
1269 	struct rte_lpm6_tbl_entry *owner_entry;
1270 
1271 	if (tbl_hdr->owner_tbl_ind == TBL24_IND)
1272 		owner_entry = &lpm->tbl24[tbl_hdr->owner_entry_ind];
1273 	else {
1274 		uint32_t owner_tbl_ind = tbl_hdr->owner_tbl_ind;
1275 		owner_entry = &lpm->tbl8[
1276 			owner_tbl_ind * RTE_LPM6_TBL8_GROUP_NUM_ENTRIES +
1277 			tbl_hdr->owner_entry_ind];
1278 
1279 		struct rte_lpm_tbl8_hdr *owner_tbl_hdr =
1280 			&lpm->tbl8_hdrs[owner_tbl_ind];
1281 		if (--owner_tbl_hdr->ref_cnt == 0)
1282 			remove_tbl(lpm, owner_tbl_hdr, owner_tbl_ind, lsp_rule);
1283 	}
1284 
1285 	assert(owner_entry->ext_entry == 1);
1286 
1287 	/* unlink the table */
1288 	if (lsp_rule != NULL) {
1289 		struct rte_lpm6_tbl_entry new_tbl_entry = {
1290 			.next_hop = lsp_rule->next_hop,
1291 			.depth = lsp_rule->depth,
1292 			.valid = VALID,
1293 			.valid_group = VALID,
1294 			.ext_entry = 0
1295 		};
1296 
1297 		*owner_entry = new_tbl_entry;
1298 	} else {
1299 		struct rte_lpm6_tbl_entry new_tbl_entry = {
1300 			.next_hop = 0,
1301 			.depth = 0,
1302 			.valid = INVALID,
1303 			.valid_group = INVALID,
1304 			.ext_entry = 0
1305 		};
1306 
1307 		*owner_entry = new_tbl_entry;
1308 	}
1309 
1310 	/* return the table to the pool */
1311 	tbl8_put(lpm, tbl_ind);
1312 }
1313 
1314 /*
1315  * Deletes a rule
1316  */
1317 int
rte_lpm6_delete(struct rte_lpm6 * lpm,const uint8_t * ip,uint8_t depth,struct rte_lpm6_rule * lsp_rule)1318 rte_lpm6_delete(struct rte_lpm6 *lpm, const uint8_t *ip, uint8_t depth,
1319     struct rte_lpm6_rule *lsp_rule)
1320 {
1321 	uint8_t masked_ip[RTE_LPM6_IPV6_ADDR_SIZE];
1322 	//struct rte_lpm6_rule lsp_rule_obj;
1323 	//struct rte_lpm6_rule *lsp_rule;
1324 	//int ret;
1325 	uint32_t tbl_ind;
1326 	struct rte_lpm6_tbl_entry *from, *to;
1327 
1328 	/* Check input arguments. */
1329 	if ((lpm == NULL) || (depth < 1) || (depth > RTE_LPM6_MAX_DEPTH))
1330 		return -EINVAL;
1331 
1332 	/* Copy the IP and mask it to avoid modifying user's input data. */
1333 	ip6_copy_addr(masked_ip, ip);
1334 	ip6_mask_addr(masked_ip, depth);
1335 
1336 #if 0
1337 	/* Delete the rule from the rule table. */
1338 	ret = rule_delete(lpm, masked_ip, depth);
1339 	if (ret < 0)
1340 		return -ENOENT;
1341 #endif
1342 
1343 	/* find rule cells */
1344 	rule_find_range(lpm, masked_ip, depth, &from, &to, &tbl_ind);
1345 
1346 #if 0
1347 	/* find a less specific rule (a rule with smaller depth)
1348 	 * note: masked_ip will be modified, don't use it anymore
1349 	 */
1350 	ret = rule_find_less_specific(lpm, masked_ip, depth,
1351 			&lsp_rule_obj);
1352 	lsp_rule = ret ? &lsp_rule_obj : NULL;
1353 #endif
1354 	/* decrement the table rule counter,
1355 	 * note that tbl24 doesn't have a header
1356 	 */
1357 	if (tbl_ind != TBL24_IND) {
1358 		struct rte_lpm_tbl8_hdr *tbl_hdr = &lpm->tbl8_hdrs[tbl_ind];
1359 		if (--tbl_hdr->ref_cnt == 0) {
1360 			/* remove the table */
1361 			remove_tbl(lpm, tbl_hdr, tbl_ind, lsp_rule);
1362 			return 0;
1363 		}
1364 	}
1365 
1366 	/* iterate rule cells */
1367 	for (; from <= to; from++)
1368 		if (from->ext_entry == 1) {
1369 			/* reference to a more specific space
1370 			 * of the prefix/rule. Entries in a more
1371 			 * specific space that are not used by
1372 			 * a more specific prefix must be occupied
1373 			 * by the prefix
1374 			 */
1375 			if (lsp_rule != NULL)
1376 				expand_rule(lpm,
1377 					from->lpm6_tbl8_gindex *
1378 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES,
1379 					depth, lsp_rule->depth,
1380 					lsp_rule->next_hop, VALID);
1381 			else
1382 				/* since the prefix has no less specific prefix,
1383 				 * its more specific space must be invalidated
1384 				 */
1385 				expand_rule(lpm,
1386 					from->lpm6_tbl8_gindex *
1387 					RTE_LPM6_TBL8_GROUP_NUM_ENTRIES,
1388 					depth, 0, 0, INVALID);
1389 		} else if (from->depth == depth) {
1390 			/* entry is not a reference and belongs to the prefix */
1391 			if (lsp_rule != NULL) {
1392 				struct rte_lpm6_tbl_entry new_tbl_entry = {
1393 					.next_hop = lsp_rule->next_hop,
1394 					.depth = lsp_rule->depth,
1395 					.valid = VALID,
1396 					.valid_group = VALID,
1397 					.ext_entry = 0
1398 				};
1399 
1400 				*from = new_tbl_entry;
1401 			} else {
1402 				struct rte_lpm6_tbl_entry new_tbl_entry = {
1403 					.next_hop = 0,
1404 					.depth = 0,
1405 					.valid = INVALID,
1406 					.valid_group = INVALID,
1407 					.ext_entry = 0
1408 				};
1409 
1410 				*from = new_tbl_entry;
1411 			}
1412 		}
1413 
1414 	return 0;
1415 }
1416