]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/flow_table.c
datapath: fix sparse warning in function tbl_mask_array_delete_mask()
[mirror_ovs.git] / datapath / flow_table.c
1 /*
2 * Copyright (c) 2007-2013 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 <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/hash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
32 #include <linux/in.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
35 #include <linux/ip.h>
36 #include <linux/ipv6.h>
37 #include <linux/sctp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
43 #include <net/ip.h>
44 #include <net/ipv6.h>
45 #include <net/ndisc.h>
46
47 #include "vlan.h"
48
49 #define TBL_MIN_BUCKETS 1024
50 #define MASK_ARRAY_SIZE_MIN 16
51 #define REHASH_INTERVAL (10 * 60 * HZ)
52
53 #define MC_HASH_SHIFT 8
54 #define MC_HASH_ENTRIES (1u << MC_HASH_SHIFT)
55 #define MC_HASH_SEGS ((sizeof(uint32_t) * 8) / MC_HASH_SHIFT)
56
57 static struct kmem_cache *flow_cache;
58 struct kmem_cache *flow_stats_cache __read_mostly;
59
60 static u16 range_n_bytes(const struct sw_flow_key_range *range)
61 {
62 return range->end - range->start;
63 }
64
65 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
66 const struct sw_flow_mask *mask)
67 {
68 const long *m = (const long *)((const u8 *)&mask->key +
69 mask->range.start);
70 const long *s = (const long *)((const u8 *)src +
71 mask->range.start);
72 long *d = (long *)((u8 *)dst + mask->range.start);
73 int i;
74
75 /* The memory outside of the 'mask->range' are not set since
76 * further operations on 'dst' only uses contents within
77 * 'mask->range'.
78 */
79 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
80 *d++ = *s++ & *m++;
81 }
82
83 struct sw_flow *ovs_flow_alloc(void)
84 {
85 struct sw_flow *flow;
86 struct flow_stats *stats;
87 int node;
88
89 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
90 if (!flow)
91 return ERR_PTR(-ENOMEM);
92
93 flow->sf_acts = NULL;
94 flow->mask = NULL;
95 flow->stats_last_writer = NUMA_NO_NODE;
96
97 /* Initialize the default stat node. */
98 stats = kmem_cache_alloc_node(flow_stats_cache,
99 GFP_KERNEL | __GFP_ZERO, 0);
100 if (!stats)
101 goto err;
102
103 spin_lock_init(&stats->lock);
104
105 RCU_INIT_POINTER(flow->stats[0], stats);
106
107 for_each_node(node)
108 if (node != 0)
109 RCU_INIT_POINTER(flow->stats[node], NULL);
110
111 return flow;
112 err:
113 kmem_cache_free(flow_cache, flow);
114 return ERR_PTR(-ENOMEM);
115 }
116
117 int ovs_flow_tbl_count(struct flow_table *table)
118 {
119 return table->count;
120 }
121
122 static struct flex_array *alloc_buckets(unsigned int n_buckets)
123 {
124 struct flex_array *buckets;
125 int i, err;
126
127 buckets = flex_array_alloc(sizeof(struct hlist_head),
128 n_buckets, GFP_KERNEL);
129 if (!buckets)
130 return NULL;
131
132 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
133 if (err) {
134 flex_array_free(buckets);
135 return NULL;
136 }
137
138 for (i = 0; i < n_buckets; i++)
139 INIT_HLIST_HEAD((struct hlist_head *)
140 flex_array_get(buckets, i));
141
142 return buckets;
143 }
144
145 static void flow_free(struct sw_flow *flow)
146 {
147 int node;
148
149 kfree((struct sw_flow_actions __force *)flow->sf_acts);
150 for_each_node(node)
151 if (flow->stats[node])
152 kmem_cache_free(flow_stats_cache,
153 (struct flow_stats __force *)flow->stats[node]);
154 kmem_cache_free(flow_cache, flow);
155 }
156
157 static void rcu_free_flow_callback(struct rcu_head *rcu)
158 {
159 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
160
161 flow_free(flow);
162 }
163
164 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
165 {
166 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
167
168 kfree(mask);
169 }
170
171 void ovs_flow_free(struct sw_flow *flow, bool deferred)
172 {
173 if (!flow)
174 return;
175
176 if (deferred)
177 call_rcu(&flow->rcu, rcu_free_flow_callback);
178 else
179 flow_free(flow);
180 }
181
182 static void free_buckets(struct flex_array *buckets)
183 {
184 flex_array_free(buckets);
185 }
186
187
188 static void __table_instance_destroy(struct table_instance *ti)
189 {
190 free_buckets(ti->buckets);
191 kfree(ti);
192 }
193
194 static struct table_instance *table_instance_alloc(int new_size)
195 {
196 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
197
198 if (!ti)
199 return NULL;
200
201 ti->buckets = alloc_buckets(new_size);
202
203 if (!ti->buckets) {
204 kfree(ti);
205 return NULL;
206 }
207 ti->n_buckets = new_size;
208 ti->node_ver = 0;
209 ti->keep_flows = false;
210 get_random_bytes(&ti->hash_seed, sizeof(u32));
211
212 return ti;
213 }
214
215 static void mask_array_rcu_cb(struct rcu_head *rcu)
216 {
217 struct mask_array *ma = container_of(rcu, struct mask_array, rcu);
218
219 kfree(ma);
220 }
221
222 static struct mask_array *tbl_mask_array_alloc(int size)
223 {
224 struct mask_array *new;
225
226 new = kzalloc(sizeof(struct mask_array) +
227 sizeof(struct sw_flow_mask *) * size, GFP_KERNEL);
228 if (!new)
229 return NULL;
230
231 new->count = 0;
232 new->max = size;
233
234 return new;
235 }
236
237 static int tbl_mask_array_realloc(struct flow_table *tbl, int size)
238 {
239 struct mask_array *old;
240 struct mask_array *new;
241
242 new = tbl_mask_array_alloc(size);
243 if (!new)
244 return -ENOMEM;
245
246 old = ovsl_dereference(tbl->mask_array);
247 if (old) {
248 int i;
249
250 for (i = 0; i < old->max; i++)
251 new->masks[i] = old->masks[i];
252
253 new->count = old->count;
254 }
255 rcu_assign_pointer(tbl->mask_array, new);
256
257 if (old)
258 call_rcu(&old->rcu, mask_array_rcu_cb);
259
260 return 0;
261 }
262
263 static void tbl_mask_array_delete_mask(struct mask_array *ma,
264 const struct sw_flow_mask *mask)
265 {
266 int i = 0;
267
268 /* Delete a mask pointer from the valid section.
269 *
270 * Also move the last entry in its place, so there is no
271 * whole in the valid section.
272 *
273 * Notice the last entry still points to the original mask.
274 *
275 * <Note>: there is a small race window that may cause a mask
276 * to be missed in a search. Imaging a core is
277 * walking through the array, passing the index of deleting mask.
278 * But before reaching the last entry, it is overwritten,
279 * by another core that is adding a new mask, now the last entry
280 * will point to the new mask. In this case, the moved up last
281 * entry can be missed by the core walking the mask array.
282 *
283 * In case this missed mask would have led to successful
284 * lookup, Hitting the race window could cause a packet to miss
285 * kernel flow cache, and be sent to the user space.
286 * </Note>
287 */
288 for (i = 0; i < ma->count; i++)
289 if (mask == ovsl_dereference(ma->masks[i])) {
290 struct sw_flow_mask *last;
291
292 last = ovsl_dereference(ma->masks[ma->count - 1]);
293 rcu_assign_pointer(ma->masks[i], last);
294 ma->count--;
295 break;
296 }
297
298 /* Remove the deleted mask pointers from the invalid section. */
299 for (i = ma->count; i < ma->max; i++)
300 if (mask == ovsl_dereference(ma->masks[i]))
301 RCU_INIT_POINTER(ma->masks[i], NULL);
302 }
303
304 int ovs_flow_tbl_init(struct flow_table *table)
305 {
306 struct table_instance *ti;
307 struct mask_array *ma;
308
309 table->mask_cache = __alloc_percpu(sizeof(struct mask_cache_entry) *
310 MC_HASH_ENTRIES, __alignof__(struct mask_cache_entry));
311 if (!table->mask_cache)
312 return -ENOMEM;
313
314 ma = tbl_mask_array_alloc(MASK_ARRAY_SIZE_MIN);
315 if (!ma)
316 goto free_mask_cache;
317
318 ti = table_instance_alloc(TBL_MIN_BUCKETS);
319 if (!ti)
320 goto free_mask_array;
321
322 rcu_assign_pointer(table->ti, ti);
323 rcu_assign_pointer(table->mask_array, ma);
324 table->last_rehash = jiffies;
325 table->count = 0;
326 return 0;
327
328 free_mask_array:
329 kfree((struct mask_array __force *)table->mask_array);
330 free_mask_cache:
331 free_percpu(table->mask_cache);
332 return -ENOMEM;
333 }
334
335 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
336 {
337 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
338
339 __table_instance_destroy(ti);
340 }
341
342 static void table_instance_destroy(struct table_instance *ti, bool deferred)
343 {
344 int i;
345
346 if (!ti)
347 return;
348
349 if (ti->keep_flows)
350 goto skip_flows;
351
352 for (i = 0; i < ti->n_buckets; i++) {
353 struct sw_flow *flow;
354 struct hlist_head *head = flex_array_get(ti->buckets, i);
355 struct hlist_node *n;
356 int ver = ti->node_ver;
357
358 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
359 hlist_del_rcu(&flow->hash_node[ver]);
360 ovs_flow_free(flow, deferred);
361 }
362 }
363
364 skip_flows:
365 if (deferred)
366 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
367 else
368 __table_instance_destroy(ti);
369 }
370
371 /* No need for locking this function is called from RCU callback or
372 * error path. */
373 void ovs_flow_tbl_destroy(struct flow_table *table)
374 {
375 struct table_instance *ti = (struct table_instance __force *)table->ti;
376
377 free_percpu(table->mask_cache);
378 kfree((struct mask_array __force *)table->mask_array);
379 table_instance_destroy(ti, false);
380 }
381
382 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
383 u32 *bucket, u32 *last)
384 {
385 struct sw_flow *flow;
386 struct hlist_head *head;
387 int ver;
388 int i;
389
390 ver = ti->node_ver;
391 while (*bucket < ti->n_buckets) {
392 i = 0;
393 head = flex_array_get(ti->buckets, *bucket);
394 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
395 if (i < *last) {
396 i++;
397 continue;
398 }
399 *last = i + 1;
400 return flow;
401 }
402 (*bucket)++;
403 *last = 0;
404 }
405
406 return NULL;
407 }
408
409 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
410 {
411 hash = jhash_1word(hash, ti->hash_seed);
412 return flex_array_get(ti->buckets,
413 (hash & (ti->n_buckets - 1)));
414 }
415
416 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
417 {
418 struct hlist_head *head;
419
420 head = find_bucket(ti, flow->hash);
421 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
422 }
423
424 static void flow_table_copy_flows(struct table_instance *old,
425 struct table_instance *new)
426 {
427 int old_ver;
428 int i;
429
430 old_ver = old->node_ver;
431 new->node_ver = !old_ver;
432
433 /* Insert in new table. */
434 for (i = 0; i < old->n_buckets; i++) {
435 struct sw_flow *flow;
436 struct hlist_head *head;
437
438 head = flex_array_get(old->buckets, i);
439
440 hlist_for_each_entry(flow, head, hash_node[old_ver])
441 table_instance_insert(new, flow);
442 }
443
444 old->keep_flows = true;
445 }
446
447 static struct table_instance *table_instance_rehash(struct table_instance *ti,
448 int n_buckets)
449 {
450 struct table_instance *new_ti;
451
452 new_ti = table_instance_alloc(n_buckets);
453 if (!new_ti)
454 return NULL;
455
456 flow_table_copy_flows(ti, new_ti);
457
458 return new_ti;
459 }
460
461 int ovs_flow_tbl_flush(struct flow_table *flow_table)
462 {
463 struct table_instance *old_ti;
464 struct table_instance *new_ti;
465
466 old_ti = ovsl_dereference(flow_table->ti);
467 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
468 if (!new_ti)
469 return -ENOMEM;
470
471 rcu_assign_pointer(flow_table->ti, new_ti);
472 flow_table->last_rehash = jiffies;
473 flow_table->count = 0;
474
475 table_instance_destroy(old_ti, true);
476 return 0;
477 }
478
479 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
480 int key_end)
481 {
482 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
483 int hash_u32s = (key_end - key_start) >> 2;
484
485 /* Make sure number of hash bytes are multiple of u32. */
486 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
487
488 return arch_fast_hash2(hash_key, hash_u32s, 0);
489 }
490
491 static int flow_key_start(const struct sw_flow_key *key)
492 {
493 if (key->tun_key.ipv4_dst)
494 return 0;
495 else
496 return rounddown(offsetof(struct sw_flow_key, phy),
497 sizeof(long));
498 }
499
500 static bool cmp_key(const struct sw_flow_key *key1,
501 const struct sw_flow_key *key2,
502 int key_start, int key_end)
503 {
504 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
505 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
506 long diffs = 0;
507 int i;
508
509 for (i = key_start; i < key_end; i += sizeof(long))
510 diffs |= *cp1++ ^ *cp2++;
511
512 return diffs == 0;
513 }
514
515 static bool flow_cmp_masked_key(const struct sw_flow *flow,
516 const struct sw_flow_key *key,
517 int key_start, int key_end)
518 {
519 return cmp_key(&flow->key, key, key_start, key_end);
520 }
521
522 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
523 struct sw_flow_match *match)
524 {
525 struct sw_flow_key *key = match->key;
526 int key_start = flow_key_start(key);
527 int key_end = match->range.end;
528
529 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
530 }
531
532 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
533 const struct sw_flow_key *unmasked,
534 struct sw_flow_mask *mask,
535 u32 *n_mask_hit)
536 {
537 struct sw_flow *flow;
538 struct hlist_head *head;
539 int key_start = mask->range.start;
540 int key_end = mask->range.end;
541 u32 hash;
542 struct sw_flow_key masked_key;
543
544 ovs_flow_mask_key(&masked_key, unmasked, mask);
545 hash = flow_hash(&masked_key, key_start, key_end);
546 head = find_bucket(ti, hash);
547 (*n_mask_hit)++;
548 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
549 if (flow->mask == mask && flow->hash == hash &&
550 flow_cmp_masked_key(flow, &masked_key,
551 key_start, key_end))
552 return flow;
553 }
554 return NULL;
555 }
556
557 static struct sw_flow *flow_lookup(struct flow_table *tbl,
558 struct table_instance *ti,
559 struct mask_array *ma,
560 const struct sw_flow_key *key,
561 u32 *n_mask_hit,
562 u32 *index)
563 {
564 struct sw_flow *flow;
565 int i;
566
567 for (i = 0; i < ma->count; i++) {
568 struct sw_flow_mask *mask;
569
570 mask = rcu_dereference_ovsl(ma->masks[i]);
571 if (mask) {
572 flow = masked_flow_lookup(ti, key, mask, n_mask_hit);
573 if (flow) { /* Found */
574 *index = i;
575 return flow;
576 }
577 }
578 }
579
580 return NULL;
581 }
582
583 /* If the the cache index is outside of the valid region, update the index
584 * in case cache entry was moved up. */
585 static void fixup_cache_entry_index(struct mask_cache_entry *e,
586 const struct mask_array *ma,
587 const struct sw_flow_mask *cache)
588 {
589 int i;
590
591 for (i = 0; i < ma->count; i++)
592 if (cache == ovsl_dereference(ma->masks[i])) {
593 e->mask_index = i;
594 return;
595 }
596 }
597
598 /*
599 * mask_cache maps flow to probable mask. This cache is not tightly
600 * coupled cache, It means updates to mask list can result in inconsistent
601 * cache entry in mask cache.
602 * This is per cpu cache and is divided in MC_HASH_SEGS segments.
603 * In case of a hash collision the entry is hashed in next segment.
604 * */
605 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
606 const struct sw_flow_key *key,
607 u32 skb_hash,
608 u32 *n_mask_hit)
609 {
610 struct mask_array *ma = rcu_dereference_ovsl(tbl->mask_array);
611 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
612 struct mask_cache_entry *entries, *ce;
613 struct sw_flow *flow;
614 u32 hash = skb_hash;
615 int seg;
616
617 *n_mask_hit = 0;
618 if (unlikely(!skb_hash)) {
619 u32 __always_unused mask_index;
620
621 return flow_lookup(tbl, ti, ma, key, n_mask_hit, &mask_index);
622 }
623
624 ce = NULL;
625 entries = this_cpu_ptr(tbl->mask_cache);
626
627 /* Find the cache entry 'ce' to operate on. */
628 for (seg = 0; seg < MC_HASH_SEGS; seg++) {
629 int index = hash & (MC_HASH_ENTRIES - 1);
630 struct mask_cache_entry *e;
631
632 e = &entries[index];
633 if (e->skb_hash == skb_hash) {
634 struct sw_flow_mask *cache;
635 int i = e->mask_index;
636
637 if (likely(i < ma->count)) {
638 cache = rcu_dereference_ovsl(ma->masks[i]);
639 flow = masked_flow_lookup(ti, key, cache,
640 n_mask_hit);
641 } else if (i < ma->max) {
642 flow = NULL;
643 cache = rcu_dereference_ovsl(ma->masks[i]);
644 if (cache) {
645 fixup_cache_entry_index(e, ma, cache);
646 flow = masked_flow_lookup(ti, key,
647 cache, n_mask_hit);
648 }
649 } else {
650 flow = NULL;
651 }
652
653 if (flow) /* Cache hit. */
654 return flow;
655
656 /* Cache miss. This is the best cache
657 * replacement candidate. */
658 e->skb_hash = 0;
659 ce = e;
660 break;
661 }
662
663 if (!ce || e->skb_hash < ce->skb_hash)
664 ce = e; /* A better replacement cache candidate. */
665
666 hash >>= MC_HASH_SHIFT;
667 }
668
669 /* Cache miss, do full lookup. */
670 flow = flow_lookup(tbl, ti, ma, key, n_mask_hit, &ce->mask_index);
671 if (flow)
672 ce->skb_hash = skb_hash;
673
674 return flow;
675 }
676
677 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
678 const struct sw_flow_key *key)
679 {
680 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
681 struct mask_array *ma = rcu_dereference_ovsl(tbl->mask_array);
682 u32 __always_unused n_mask_hit;
683 u32 __always_unused index;
684
685 n_mask_hit = 0;
686 return flow_lookup(tbl, ti, ma, key, &n_mask_hit, &index);
687 }
688
689 int ovs_flow_tbl_num_masks(const struct flow_table *table)
690 {
691 struct mask_array *ma;
692
693 ma = rcu_dereference_ovsl(table->mask_array);
694 return ma->count;
695 }
696
697 static struct table_instance *table_instance_expand(struct table_instance *ti)
698 {
699 return table_instance_rehash(ti, ti->n_buckets * 2);
700 }
701
702 /* Remove 'mask' from the mask list, if it is not needed any more. */
703 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
704 {
705 if (mask) {
706 /* ovs-lock is required to protect mask-refcount and
707 * mask list.
708 */
709 ASSERT_OVSL();
710 BUG_ON(!mask->ref_count);
711 mask->ref_count--;
712
713 if (!mask->ref_count) {
714 struct mask_array *ma;
715
716 ma = ovsl_dereference(tbl->mask_array);
717 /* Shrink the mask array if necessary. */
718 if (ma->max > MASK_ARRAY_SIZE_MIN * 2
719 && ma->count <= ma->max / 4) {
720
721 tbl_mask_array_realloc(tbl, ma->max / 2);
722 ma = ovsl_dereference(tbl->mask_array);
723 }
724
725 tbl_mask_array_delete_mask(ma, mask);
726 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
727 }
728 }
729 }
730
731 /* Must be called with OVS mutex held. */
732 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
733 {
734 struct table_instance *ti = ovsl_dereference(table->ti);
735
736 BUG_ON(table->count == 0);
737 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
738 table->count--;
739
740 /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
741 * accessible as long as the RCU read lock is held. */
742 flow_mask_remove(table, flow->mask);
743 }
744
745 static struct sw_flow_mask *mask_alloc(void)
746 {
747 struct sw_flow_mask *mask;
748
749 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
750 if (mask)
751 mask->ref_count = 1;
752
753 return mask;
754 }
755
756 static bool mask_equal(const struct sw_flow_mask *a,
757 const struct sw_flow_mask *b)
758 {
759 const u8 *a_ = (const u8 *)&a->key + a->range.start;
760 const u8 *b_ = (const u8 *)&b->key + b->range.start;
761
762 return (a->range.end == b->range.end)
763 && (a->range.start == b->range.start)
764 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
765 }
766
767 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
768 const struct sw_flow_mask *mask)
769 {
770 struct mask_array *ma;
771 int i;
772
773 ma = ovsl_dereference(tbl->mask_array);
774 for (i = 0; i < ma->count; i++) {
775 struct sw_flow_mask *t;
776
777 t = ovsl_dereference(ma->masks[i]);
778 if (t && mask_equal(mask, t))
779 return t;
780 }
781
782 return NULL;
783 }
784
785 /* Add 'mask' into the mask list, if it is not already there. */
786 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
787 struct sw_flow_mask *new)
788 {
789 struct sw_flow_mask *mask;
790
791 mask = flow_mask_find(tbl, new);
792 if (!mask) {
793 struct mask_array *ma;
794
795 /* Allocate a new mask if none exsits. */
796 mask = mask_alloc();
797 if (!mask)
798 return -ENOMEM;
799
800 mask->key = new->key;
801 mask->range = new->range;
802
803 /* Add mask to mask-list. */
804 ma = ovsl_dereference(tbl->mask_array);
805 if (ma->count >= ma->max) {
806 int err;
807
808 err = tbl_mask_array_realloc(tbl, ma->max +
809 MASK_ARRAY_SIZE_MIN);
810 if (err) {
811 kfree(mask);
812 return err;
813 }
814 ma = ovsl_dereference(tbl->mask_array);
815 }
816
817 rcu_assign_pointer(ma->masks[ma->count], mask);
818 ma->count++;
819 } else {
820 BUG_ON(!mask->ref_count);
821 mask->ref_count++;
822 }
823
824 flow->mask = mask;
825 return 0;
826 }
827
828 /* Must be called with OVS mutex held. */
829 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
830 struct sw_flow_mask *mask)
831 {
832 struct table_instance *new_ti = NULL;
833 struct table_instance *ti;
834 int err;
835
836 err = flow_mask_insert(table, flow, mask);
837 if (err)
838 return err;
839
840 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
841 flow->mask->range.end);
842 ti = ovsl_dereference(table->ti);
843 table_instance_insert(ti, flow);
844 table->count++;
845
846 /* Expand table, if necessary, to make room. */
847 if (table->count > ti->n_buckets)
848 new_ti = table_instance_expand(ti);
849 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
850 new_ti = table_instance_rehash(ti, ti->n_buckets);
851
852 if (new_ti) {
853 rcu_assign_pointer(table->ti, new_ti);
854 table_instance_destroy(ti, true);
855 table->last_rehash = jiffies;
856 }
857 return 0;
858 }
859
860 /* Initializes the flow module.
861 * Returns zero if successful or a negative error code. */
862 int ovs_flow_init(void)
863 {
864 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
865 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
866
867 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
868 + (num_possible_nodes()
869 * sizeof(struct flow_stats *)),
870 0, 0, NULL);
871 if (flow_cache == NULL)
872 return -ENOMEM;
873
874 flow_stats_cache
875 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
876 0, SLAB_HWCACHE_ALIGN, NULL);
877 if (flow_stats_cache == NULL) {
878 kmem_cache_destroy(flow_cache);
879 flow_cache = NULL;
880 return -ENOMEM;
881 }
882
883 return 0;
884 }
885
886 /* Uninitializes the flow module. */
887 void ovs_flow_exit(void)
888 {
889 kmem_cache_destroy(flow_stats_cache);
890 kmem_cache_destroy(flow_cache);
891 }