xref: /linux/net/openvswitch/flow_table.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  * Copyright (c) 2007-2014 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #include "flow.h"
20 #include "datapath.h"
21 #include "flow_netlink.h"
22 #include <linux/uaccess.h>
23 #include <linux/netdevice.h>
24 #include <linux/etherdevice.h>
25 #include <linux/if_ether.h>
26 #include <linux/if_vlan.h>
27 #include <net/llc_pdu.h>
28 #include <linux/kernel.h>
29 #include <linux/jhash.h>
30 #include <linux/jiffies.h>
31 #include <linux/llc.h>
32 #include <linux/module.h>
33 #include <linux/in.h>
34 #include <linux/rcupdate.h>
35 #include <linux/if_arp.h>
36 #include <linux/ip.h>
37 #include <linux/ipv6.h>
38 #include <linux/sctp.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/icmp.h>
42 #include <linux/icmpv6.h>
43 #include <linux/rculist.h>
44 #include <net/ip.h>
45 #include <net/ipv6.h>
46 #include <net/ndisc.h>
47 
48 #define TBL_MIN_BUCKETS		1024
49 #define REHASH_INTERVAL		(10 * 60 * HZ)
50 
51 static struct kmem_cache *flow_cache;
52 struct kmem_cache *flow_stats_cache __read_mostly;
53 
54 static u16 range_n_bytes(const struct sw_flow_key_range *range)
55 {
56 	return range->end - range->start;
57 }
58 
59 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
60 		       bool full, const struct sw_flow_mask *mask)
61 {
62 	int start = full ? 0 : mask->range.start;
63 	int len = full ? sizeof *dst : range_n_bytes(&mask->range);
64 	const long *m = (const long *)((const u8 *)&mask->key + start);
65 	const long *s = (const long *)((const u8 *)src + start);
66 	long *d = (long *)((u8 *)dst + start);
67 	int i;
68 
69 	/* If 'full' is true then all of 'dst' is fully initialized. Otherwise,
70 	 * if 'full' is false the memory outside of the 'mask->range' is left
71 	 * uninitialized. This can be used as an optimization when further
72 	 * operations on 'dst' only use contents within 'mask->range'.
73 	 */
74 	for (i = 0; i < len; i += sizeof(long))
75 		*d++ = *s++ & *m++;
76 }
77 
78 struct sw_flow *ovs_flow_alloc(void)
79 {
80 	struct sw_flow *flow;
81 	struct flow_stats *stats;
82 	int node;
83 
84 	flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
85 	if (!flow)
86 		return ERR_PTR(-ENOMEM);
87 
88 	flow->sf_acts = NULL;
89 	flow->mask = NULL;
90 	flow->id.unmasked_key = NULL;
91 	flow->id.ufid_len = 0;
92 	flow->stats_last_writer = NUMA_NO_NODE;
93 
94 	/* Initialize the default stat node. */
95 	stats = kmem_cache_alloc_node(flow_stats_cache,
96 				      GFP_KERNEL | __GFP_ZERO,
97 				      node_online(0) ? 0 : NUMA_NO_NODE);
98 	if (!stats)
99 		goto err;
100 
101 	spin_lock_init(&stats->lock);
102 
103 	RCU_INIT_POINTER(flow->stats[0], stats);
104 
105 	for_each_node(node)
106 		if (node != 0)
107 			RCU_INIT_POINTER(flow->stats[node], NULL);
108 
109 	return flow;
110 err:
111 	kmem_cache_free(flow_cache, flow);
112 	return ERR_PTR(-ENOMEM);
113 }
114 
115 int ovs_flow_tbl_count(const struct flow_table *table)
116 {
117 	return table->count;
118 }
119 
120 static struct flex_array *alloc_buckets(unsigned int n_buckets)
121 {
122 	struct flex_array *buckets;
123 	int i, err;
124 
125 	buckets = flex_array_alloc(sizeof(struct hlist_head),
126 				   n_buckets, GFP_KERNEL);
127 	if (!buckets)
128 		return NULL;
129 
130 	err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
131 	if (err) {
132 		flex_array_free(buckets);
133 		return NULL;
134 	}
135 
136 	for (i = 0; i < n_buckets; i++)
137 		INIT_HLIST_HEAD((struct hlist_head *)
138 					flex_array_get(buckets, i));
139 
140 	return buckets;
141 }
142 
143 static void flow_free(struct sw_flow *flow)
144 {
145 	int node;
146 
147 	if (ovs_identifier_is_key(&flow->id))
148 		kfree(flow->id.unmasked_key);
149 	if (flow->sf_acts)
150 		ovs_nla_free_flow_actions((struct sw_flow_actions __force *)flow->sf_acts);
151 	for_each_node(node)
152 		if (flow->stats[node])
153 			kmem_cache_free(flow_stats_cache,
154 					(struct flow_stats __force *)flow->stats[node]);
155 	kmem_cache_free(flow_cache, flow);
156 }
157 
158 static void rcu_free_flow_callback(struct rcu_head *rcu)
159 {
160 	struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
161 
162 	flow_free(flow);
163 }
164 
165 void ovs_flow_free(struct sw_flow *flow, bool deferred)
166 {
167 	if (!flow)
168 		return;
169 
170 	if (deferred)
171 		call_rcu(&flow->rcu, rcu_free_flow_callback);
172 	else
173 		flow_free(flow);
174 }
175 
176 static void free_buckets(struct flex_array *buckets)
177 {
178 	flex_array_free(buckets);
179 }
180 
181 
182 static void __table_instance_destroy(struct table_instance *ti)
183 {
184 	free_buckets(ti->buckets);
185 	kfree(ti);
186 }
187 
188 static struct table_instance *table_instance_alloc(int new_size)
189 {
190 	struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
191 
192 	if (!ti)
193 		return NULL;
194 
195 	ti->buckets = alloc_buckets(new_size);
196 
197 	if (!ti->buckets) {
198 		kfree(ti);
199 		return NULL;
200 	}
201 	ti->n_buckets = new_size;
202 	ti->node_ver = 0;
203 	ti->keep_flows = false;
204 	get_random_bytes(&ti->hash_seed, sizeof(u32));
205 
206 	return ti;
207 }
208 
209 int ovs_flow_tbl_init(struct flow_table *table)
210 {
211 	struct table_instance *ti, *ufid_ti;
212 
213 	ti = table_instance_alloc(TBL_MIN_BUCKETS);
214 
215 	if (!ti)
216 		return -ENOMEM;
217 
218 	ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
219 	if (!ufid_ti)
220 		goto free_ti;
221 
222 	rcu_assign_pointer(table->ti, ti);
223 	rcu_assign_pointer(table->ufid_ti, ufid_ti);
224 	INIT_LIST_HEAD(&table->mask_list);
225 	table->last_rehash = jiffies;
226 	table->count = 0;
227 	table->ufid_count = 0;
228 	return 0;
229 
230 free_ti:
231 	__table_instance_destroy(ti);
232 	return -ENOMEM;
233 }
234 
235 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
236 {
237 	struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
238 
239 	__table_instance_destroy(ti);
240 }
241 
242 static void table_instance_destroy(struct table_instance *ti,
243 				   struct table_instance *ufid_ti,
244 				   bool deferred)
245 {
246 	int i;
247 
248 	if (!ti)
249 		return;
250 
251 	BUG_ON(!ufid_ti);
252 	if (ti->keep_flows)
253 		goto skip_flows;
254 
255 	for (i = 0; i < ti->n_buckets; i++) {
256 		struct sw_flow *flow;
257 		struct hlist_head *head = flex_array_get(ti->buckets, i);
258 		struct hlist_node *n;
259 		int ver = ti->node_ver;
260 		int ufid_ver = ufid_ti->node_ver;
261 
262 		hlist_for_each_entry_safe(flow, n, head, flow_table.node[ver]) {
263 			hlist_del_rcu(&flow->flow_table.node[ver]);
264 			if (ovs_identifier_is_ufid(&flow->id))
265 				hlist_del_rcu(&flow->ufid_table.node[ufid_ver]);
266 			ovs_flow_free(flow, deferred);
267 		}
268 	}
269 
270 skip_flows:
271 	if (deferred) {
272 		call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
273 		call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb);
274 	} else {
275 		__table_instance_destroy(ti);
276 		__table_instance_destroy(ufid_ti);
277 	}
278 }
279 
280 /* No need for locking this function is called from RCU callback or
281  * error path.
282  */
283 void ovs_flow_tbl_destroy(struct flow_table *table)
284 {
285 	struct table_instance *ti = rcu_dereference_raw(table->ti);
286 	struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti);
287 
288 	table_instance_destroy(ti, ufid_ti, false);
289 }
290 
291 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
292 				       u32 *bucket, u32 *last)
293 {
294 	struct sw_flow *flow;
295 	struct hlist_head *head;
296 	int ver;
297 	int i;
298 
299 	ver = ti->node_ver;
300 	while (*bucket < ti->n_buckets) {
301 		i = 0;
302 		head = flex_array_get(ti->buckets, *bucket);
303 		hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) {
304 			if (i < *last) {
305 				i++;
306 				continue;
307 			}
308 			*last = i + 1;
309 			return flow;
310 		}
311 		(*bucket)++;
312 		*last = 0;
313 	}
314 
315 	return NULL;
316 }
317 
318 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
319 {
320 	hash = jhash_1word(hash, ti->hash_seed);
321 	return flex_array_get(ti->buckets,
322 				(hash & (ti->n_buckets - 1)));
323 }
324 
325 static void table_instance_insert(struct table_instance *ti,
326 				  struct sw_flow *flow)
327 {
328 	struct hlist_head *head;
329 
330 	head = find_bucket(ti, flow->flow_table.hash);
331 	hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head);
332 }
333 
334 static void ufid_table_instance_insert(struct table_instance *ti,
335 				       struct sw_flow *flow)
336 {
337 	struct hlist_head *head;
338 
339 	head = find_bucket(ti, flow->ufid_table.hash);
340 	hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head);
341 }
342 
343 static void flow_table_copy_flows(struct table_instance *old,
344 				  struct table_instance *new, bool ufid)
345 {
346 	int old_ver;
347 	int i;
348 
349 	old_ver = old->node_ver;
350 	new->node_ver = !old_ver;
351 
352 	/* Insert in new table. */
353 	for (i = 0; i < old->n_buckets; i++) {
354 		struct sw_flow *flow;
355 		struct hlist_head *head;
356 
357 		head = flex_array_get(old->buckets, i);
358 
359 		if (ufid)
360 			hlist_for_each_entry(flow, head,
361 					     ufid_table.node[old_ver])
362 				ufid_table_instance_insert(new, flow);
363 		else
364 			hlist_for_each_entry(flow, head,
365 					     flow_table.node[old_ver])
366 				table_instance_insert(new, flow);
367 	}
368 
369 	old->keep_flows = true;
370 }
371 
372 static struct table_instance *table_instance_rehash(struct table_instance *ti,
373 						    int n_buckets, bool ufid)
374 {
375 	struct table_instance *new_ti;
376 
377 	new_ti = table_instance_alloc(n_buckets);
378 	if (!new_ti)
379 		return NULL;
380 
381 	flow_table_copy_flows(ti, new_ti, ufid);
382 
383 	return new_ti;
384 }
385 
386 int ovs_flow_tbl_flush(struct flow_table *flow_table)
387 {
388 	struct table_instance *old_ti, *new_ti;
389 	struct table_instance *old_ufid_ti, *new_ufid_ti;
390 
391 	new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
392 	if (!new_ti)
393 		return -ENOMEM;
394 	new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
395 	if (!new_ufid_ti)
396 		goto err_free_ti;
397 
398 	old_ti = ovsl_dereference(flow_table->ti);
399 	old_ufid_ti = ovsl_dereference(flow_table->ufid_ti);
400 
401 	rcu_assign_pointer(flow_table->ti, new_ti);
402 	rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti);
403 	flow_table->last_rehash = jiffies;
404 	flow_table->count = 0;
405 	flow_table->ufid_count = 0;
406 
407 	table_instance_destroy(old_ti, old_ufid_ti, true);
408 	return 0;
409 
410 err_free_ti:
411 	__table_instance_destroy(new_ti);
412 	return -ENOMEM;
413 }
414 
415 static u32 flow_hash(const struct sw_flow_key *key,
416 		     const struct sw_flow_key_range *range)
417 {
418 	int key_start = range->start;
419 	int key_end = range->end;
420 	const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
421 	int hash_u32s = (key_end - key_start) >> 2;
422 
423 	/* Make sure number of hash bytes are multiple of u32. */
424 	BUILD_BUG_ON(sizeof(long) % sizeof(u32));
425 
426 	return jhash2(hash_key, hash_u32s, 0);
427 }
428 
429 static int flow_key_start(const struct sw_flow_key *key)
430 {
431 	if (key->tun_key.u.ipv4.dst)
432 		return 0;
433 	else
434 		return rounddown(offsetof(struct sw_flow_key, phy),
435 					  sizeof(long));
436 }
437 
438 static bool cmp_key(const struct sw_flow_key *key1,
439 		    const struct sw_flow_key *key2,
440 		    int key_start, int key_end)
441 {
442 	const long *cp1 = (const long *)((const u8 *)key1 + key_start);
443 	const long *cp2 = (const long *)((const u8 *)key2 + key_start);
444 	long diffs = 0;
445 	int i;
446 
447 	for (i = key_start; i < key_end;  i += sizeof(long))
448 		diffs |= *cp1++ ^ *cp2++;
449 
450 	return diffs == 0;
451 }
452 
453 static bool flow_cmp_masked_key(const struct sw_flow *flow,
454 				const struct sw_flow_key *key,
455 				const struct sw_flow_key_range *range)
456 {
457 	return cmp_key(&flow->key, key, range->start, range->end);
458 }
459 
460 static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
461 				      const struct sw_flow_match *match)
462 {
463 	struct sw_flow_key *key = match->key;
464 	int key_start = flow_key_start(key);
465 	int key_end = match->range.end;
466 
467 	BUG_ON(ovs_identifier_is_ufid(&flow->id));
468 	return cmp_key(flow->id.unmasked_key, key, key_start, key_end);
469 }
470 
471 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
472 					  const struct sw_flow_key *unmasked,
473 					  const struct sw_flow_mask *mask)
474 {
475 	struct sw_flow *flow;
476 	struct hlist_head *head;
477 	u32 hash;
478 	struct sw_flow_key masked_key;
479 
480 	ovs_flow_mask_key(&masked_key, unmasked, false, mask);
481 	hash = flow_hash(&masked_key, &mask->range);
482 	head = find_bucket(ti, hash);
483 	hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) {
484 		if (flow->mask == mask && flow->flow_table.hash == hash &&
485 		    flow_cmp_masked_key(flow, &masked_key, &mask->range))
486 			return flow;
487 	}
488 	return NULL;
489 }
490 
491 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
492 				    const struct sw_flow_key *key,
493 				    u32 *n_mask_hit)
494 {
495 	struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
496 	struct sw_flow_mask *mask;
497 	struct sw_flow *flow;
498 
499 	*n_mask_hit = 0;
500 	list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
501 		(*n_mask_hit)++;
502 		flow = masked_flow_lookup(ti, key, mask);
503 		if (flow)  /* Found */
504 			return flow;
505 	}
506 	return NULL;
507 }
508 
509 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
510 				    const struct sw_flow_key *key)
511 {
512 	u32 __always_unused n_mask_hit;
513 
514 	return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
515 }
516 
517 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
518 					  const struct sw_flow_match *match)
519 {
520 	struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
521 	struct sw_flow_mask *mask;
522 	struct sw_flow *flow;
523 
524 	/* Always called under ovs-mutex. */
525 	list_for_each_entry(mask, &tbl->mask_list, list) {
526 		flow = masked_flow_lookup(ti, match->key, mask);
527 		if (flow && ovs_identifier_is_key(&flow->id) &&
528 		    ovs_flow_cmp_unmasked_key(flow, match))
529 			return flow;
530 	}
531 	return NULL;
532 }
533 
534 static u32 ufid_hash(const struct sw_flow_id *sfid)
535 {
536 	return jhash(sfid->ufid, sfid->ufid_len, 0);
537 }
538 
539 static bool ovs_flow_cmp_ufid(const struct sw_flow *flow,
540 			      const struct sw_flow_id *sfid)
541 {
542 	if (flow->id.ufid_len != sfid->ufid_len)
543 		return false;
544 
545 	return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len);
546 }
547 
548 bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match)
549 {
550 	if (ovs_identifier_is_ufid(&flow->id))
551 		return flow_cmp_masked_key(flow, match->key, &match->range);
552 
553 	return ovs_flow_cmp_unmasked_key(flow, match);
554 }
555 
556 struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl,
557 					 const struct sw_flow_id *ufid)
558 {
559 	struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti);
560 	struct sw_flow *flow;
561 	struct hlist_head *head;
562 	u32 hash;
563 
564 	hash = ufid_hash(ufid);
565 	head = find_bucket(ti, hash);
566 	hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) {
567 		if (flow->ufid_table.hash == hash &&
568 		    ovs_flow_cmp_ufid(flow, ufid))
569 			return flow;
570 	}
571 	return NULL;
572 }
573 
574 int ovs_flow_tbl_num_masks(const struct flow_table *table)
575 {
576 	struct sw_flow_mask *mask;
577 	int num = 0;
578 
579 	list_for_each_entry(mask, &table->mask_list, list)
580 		num++;
581 
582 	return num;
583 }
584 
585 static struct table_instance *table_instance_expand(struct table_instance *ti,
586 						    bool ufid)
587 {
588 	return table_instance_rehash(ti, ti->n_buckets * 2, ufid);
589 }
590 
591 /* Remove 'mask' from the mask list, if it is not needed any more. */
592 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
593 {
594 	if (mask) {
595 		/* ovs-lock is required to protect mask-refcount and
596 		 * mask list.
597 		 */
598 		ASSERT_OVSL();
599 		BUG_ON(!mask->ref_count);
600 		mask->ref_count--;
601 
602 		if (!mask->ref_count) {
603 			list_del_rcu(&mask->list);
604 			kfree_rcu(mask, rcu);
605 		}
606 	}
607 }
608 
609 /* Must be called with OVS mutex held. */
610 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
611 {
612 	struct table_instance *ti = ovsl_dereference(table->ti);
613 	struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti);
614 
615 	BUG_ON(table->count == 0);
616 	hlist_del_rcu(&flow->flow_table.node[ti->node_ver]);
617 	table->count--;
618 	if (ovs_identifier_is_ufid(&flow->id)) {
619 		hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]);
620 		table->ufid_count--;
621 	}
622 
623 	/* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
624 	 * accessible as long as the RCU read lock is held.
625 	 */
626 	flow_mask_remove(table, flow->mask);
627 }
628 
629 static struct sw_flow_mask *mask_alloc(void)
630 {
631 	struct sw_flow_mask *mask;
632 
633 	mask = kmalloc(sizeof(*mask), GFP_KERNEL);
634 	if (mask)
635 		mask->ref_count = 1;
636 
637 	return mask;
638 }
639 
640 static bool mask_equal(const struct sw_flow_mask *a,
641 		       const struct sw_flow_mask *b)
642 {
643 	const u8 *a_ = (const u8 *)&a->key + a->range.start;
644 	const u8 *b_ = (const u8 *)&b->key + b->range.start;
645 
646 	return  (a->range.end == b->range.end)
647 		&& (a->range.start == b->range.start)
648 		&& (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
649 }
650 
651 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
652 					   const struct sw_flow_mask *mask)
653 {
654 	struct list_head *ml;
655 
656 	list_for_each(ml, &tbl->mask_list) {
657 		struct sw_flow_mask *m;
658 		m = container_of(ml, struct sw_flow_mask, list);
659 		if (mask_equal(mask, m))
660 			return m;
661 	}
662 
663 	return NULL;
664 }
665 
666 /* Add 'mask' into the mask list, if it is not already there. */
667 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
668 			    const struct sw_flow_mask *new)
669 {
670 	struct sw_flow_mask *mask;
671 	mask = flow_mask_find(tbl, new);
672 	if (!mask) {
673 		/* Allocate a new mask if none exsits. */
674 		mask = mask_alloc();
675 		if (!mask)
676 			return -ENOMEM;
677 		mask->key = new->key;
678 		mask->range = new->range;
679 		list_add_rcu(&mask->list, &tbl->mask_list);
680 	} else {
681 		BUG_ON(!mask->ref_count);
682 		mask->ref_count++;
683 	}
684 
685 	flow->mask = mask;
686 	return 0;
687 }
688 
689 /* Must be called with OVS mutex held. */
690 static void flow_key_insert(struct flow_table *table, struct sw_flow *flow)
691 {
692 	struct table_instance *new_ti = NULL;
693 	struct table_instance *ti;
694 
695 	flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range);
696 	ti = ovsl_dereference(table->ti);
697 	table_instance_insert(ti, flow);
698 	table->count++;
699 
700 	/* Expand table, if necessary, to make room. */
701 	if (table->count > ti->n_buckets)
702 		new_ti = table_instance_expand(ti, false);
703 	else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
704 		new_ti = table_instance_rehash(ti, ti->n_buckets, false);
705 
706 	if (new_ti) {
707 		rcu_assign_pointer(table->ti, new_ti);
708 		call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
709 		table->last_rehash = jiffies;
710 	}
711 }
712 
713 /* Must be called with OVS mutex held. */
714 static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow)
715 {
716 	struct table_instance *ti;
717 
718 	flow->ufid_table.hash = ufid_hash(&flow->id);
719 	ti = ovsl_dereference(table->ufid_ti);
720 	ufid_table_instance_insert(ti, flow);
721 	table->ufid_count++;
722 
723 	/* Expand table, if necessary, to make room. */
724 	if (table->ufid_count > ti->n_buckets) {
725 		struct table_instance *new_ti;
726 
727 		new_ti = table_instance_expand(ti, true);
728 		if (new_ti) {
729 			rcu_assign_pointer(table->ufid_ti, new_ti);
730 			call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
731 		}
732 	}
733 }
734 
735 /* Must be called with OVS mutex held. */
736 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
737 			const struct sw_flow_mask *mask)
738 {
739 	int err;
740 
741 	err = flow_mask_insert(table, flow, mask);
742 	if (err)
743 		return err;
744 	flow_key_insert(table, flow);
745 	if (ovs_identifier_is_ufid(&flow->id))
746 		flow_ufid_insert(table, flow);
747 
748 	return 0;
749 }
750 
751 /* Initializes the flow module.
752  * Returns zero if successful or a negative error code. */
753 int ovs_flow_init(void)
754 {
755 	BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
756 	BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
757 
758 	flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
759 				       + (nr_node_ids
760 					  * sizeof(struct flow_stats *)),
761 				       0, 0, NULL);
762 	if (flow_cache == NULL)
763 		return -ENOMEM;
764 
765 	flow_stats_cache
766 		= kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
767 				    0, SLAB_HWCACHE_ALIGN, NULL);
768 	if (flow_stats_cache == NULL) {
769 		kmem_cache_destroy(flow_cache);
770 		flow_cache = NULL;
771 		return -ENOMEM;
772 	}
773 
774 	return 0;
775 }
776 
777 /* Uninitializes the flow module. */
778 void ovs_flow_exit(void)
779 {
780 	kmem_cache_destroy(flow_stats_cache);
781 	kmem_cache_destroy(flow_cache);
782 }
783