]> git.proxmox.com Git - ovs.git/blob - datapath/flow_table.c
dpctl: Fix dpctl process command parameter error.
[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/jhash.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/cpumask.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 #include "flow_netlink.h"
49
50 #define TBL_MIN_BUCKETS 1024
51 #define MASK_ARRAY_SIZE_MIN 16
52 #define REHASH_INTERVAL (10 * 60 * HZ)
53
54 #define MC_HASH_SHIFT 8
55 #define MC_HASH_ENTRIES (1u << MC_HASH_SHIFT)
56 #define MC_HASH_SEGS ((sizeof(uint32_t) * 8) / MC_HASH_SHIFT)
57
58 static struct kmem_cache *flow_cache;
59 struct kmem_cache *flow_stats_cache __read_mostly;
60
61 static u16 range_n_bytes(const struct sw_flow_key_range *range)
62 {
63 return range->end - range->start;
64 }
65
66 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
67 bool full, const struct sw_flow_mask *mask)
68 {
69 int start = full ? 0 : mask->range.start;
70 int len = full ? sizeof *dst : range_n_bytes(&mask->range);
71 const long *m = (const long *)((const u8 *)&mask->key + start);
72 const long *s = (const long *)((const u8 *)src + start);
73 long *d = (long *)((u8 *)dst + start);
74 int i;
75
76 /* If 'full' is true then all of 'dst' is fully initialized. Otherwise,
77 * if 'full' is false the memory outside of the 'mask->range' is left
78 * uninitialized. This can be used as an optimization when further
79 * operations on 'dst' only use contents within 'mask->range'.
80 */
81 for (i = 0; i < len; i += sizeof(long))
82 *d++ = *s++ & *m++;
83 }
84
85 struct sw_flow *ovs_flow_alloc(void)
86 {
87 struct sw_flow *flow;
88 struct sw_flow_stats *stats;
89
90 flow = kmem_cache_zalloc(flow_cache, GFP_KERNEL);
91 if (!flow)
92 return ERR_PTR(-ENOMEM);
93
94 flow->stats_last_writer = -1;
95
96 /* Initialize the default stat node. */
97 stats = kmem_cache_alloc_node(flow_stats_cache,
98 GFP_KERNEL | __GFP_ZERO,
99 node_online(0) ? 0 : NUMA_NO_NODE);
100 if (!stats)
101 goto err;
102
103 spin_lock_init(&stats->lock);
104
105 RCU_INIT_POINTER(flow->stats[0], stats);
106
107 cpumask_set_cpu(0, &flow->cpu_used_mask);
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 void flow_free(struct sw_flow *flow)
121 {
122 int cpu;
123
124 if (ovs_identifier_is_key(&flow->id))
125 kfree(flow->id.unmasked_key);
126 if (flow->sf_acts)
127 ovs_nla_free_flow_actions((struct sw_flow_actions __force *)flow->sf_acts);
128 /* We open code this to make sure cpu 0 is always considered */
129 for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask))
130 if (flow->stats[cpu])
131 kmem_cache_free(flow_stats_cache,
132 rcu_dereference_raw(flow->stats[cpu]));
133 kmem_cache_free(flow_cache, flow);
134 }
135
136 static void rcu_free_flow_callback(struct rcu_head *rcu)
137 {
138 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
139
140 flow_free(flow);
141 }
142
143 void ovs_flow_free(struct sw_flow *flow, bool deferred)
144 {
145 if (!flow)
146 return;
147
148 if (deferred)
149 call_rcu(&flow->rcu, rcu_free_flow_callback);
150 else
151 flow_free(flow);
152 }
153
154 static void __table_instance_destroy(struct table_instance *ti)
155 {
156 kvfree(ti->buckets);
157 kfree(ti);
158 }
159
160 static struct table_instance *table_instance_alloc(int new_size)
161 {
162 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
163 int i;
164
165 if (!ti)
166 return NULL;
167
168 ti->buckets = kvmalloc_array(new_size, sizeof(struct hlist_head),
169 GFP_KERNEL);
170 if (!ti->buckets) {
171 kfree(ti);
172 return NULL;
173 }
174
175 for (i = 0; i < new_size; i++)
176 INIT_HLIST_HEAD(&ti->buckets[i]);
177
178 ti->n_buckets = new_size;
179 ti->node_ver = 0;
180 ti->keep_flows = false;
181 get_random_bytes(&ti->hash_seed, sizeof(u32));
182
183 return ti;
184 }
185
186 static void mask_array_rcu_cb(struct rcu_head *rcu)
187 {
188 struct mask_array *ma = container_of(rcu, struct mask_array, rcu);
189
190 kfree(ma);
191 }
192
193 static struct mask_array *tbl_mask_array_alloc(int size)
194 {
195 struct mask_array *new;
196
197 size = max(MASK_ARRAY_SIZE_MIN, size);
198 new = kzalloc(sizeof(struct mask_array) +
199 sizeof(struct sw_flow_mask *) * size, GFP_KERNEL);
200 if (!new)
201 return NULL;
202
203 new->count = 0;
204 new->max = size;
205
206 return new;
207 }
208
209 static int tbl_mask_array_realloc(struct flow_table *tbl, int size)
210 {
211 struct mask_array *old;
212 struct mask_array *new;
213
214 new = tbl_mask_array_alloc(size);
215 if (!new)
216 return -ENOMEM;
217
218 old = ovsl_dereference(tbl->mask_array);
219 if (old) {
220 int i, count = 0;
221
222 for (i = 0; i < old->max; i++) {
223 if (ovsl_dereference(old->masks[i]))
224 new->masks[count++] = old->masks[i];
225 }
226
227 new->count = count;
228 }
229 rcu_assign_pointer(tbl->mask_array, new);
230
231 if (old)
232 call_rcu(&old->rcu, mask_array_rcu_cb);
233
234 return 0;
235 }
236
237 static int tbl_mask_array_add_mask(struct flow_table *tbl,
238 struct sw_flow_mask *new)
239 {
240 struct mask_array *ma = ovsl_dereference(tbl->mask_array);
241 int err, ma_count = READ_ONCE(ma->count);
242
243 if (ma_count >= ma->max) {
244 err = tbl_mask_array_realloc(tbl, ma->max +
245 MASK_ARRAY_SIZE_MIN);
246 if (err)
247 return err;
248
249 ma = ovsl_dereference(tbl->mask_array);
250 }
251
252 BUG_ON(ovsl_dereference(ma->masks[ma_count]));
253
254 rcu_assign_pointer(ma->masks[ma_count], new);
255 WRITE_ONCE(ma->count, ma_count +1);
256
257 return 0;
258 }
259
260 static void tbl_mask_array_del_mask(struct flow_table *tbl,
261 struct sw_flow_mask *mask)
262 {
263 struct mask_array *ma = ovsl_dereference(tbl->mask_array);
264 int i, ma_count = READ_ONCE(ma->count);
265
266 /* Remove the deleted mask pointers from the array */
267 for (i = 0; i < ma_count; i++) {
268 if (mask == ovsl_dereference(ma->masks[i]))
269 goto found;
270 }
271
272 BUG();
273 return;
274
275 found:
276 WRITE_ONCE(ma->count, ma_count -1);
277
278 rcu_assign_pointer(ma->masks[i], ma->masks[ma_count -1]);
279 RCU_INIT_POINTER(ma->masks[ma_count -1], NULL);
280
281 kfree_rcu(mask, rcu);
282
283 /* Shrink the mask array if necessary. */
284 if (ma->max >= (MASK_ARRAY_SIZE_MIN * 2) &&
285 ma_count <= (ma->max / 3))
286 tbl_mask_array_realloc(tbl, ma->max / 2);
287 }
288
289 /* Remove 'mask' from the mask list, if it is not needed any more. */
290 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
291 {
292 if (mask) {
293 /* ovs-lock is required to protect mask-refcount and
294 * mask list.
295 */
296 ASSERT_OVSL();
297 BUG_ON(!mask->ref_count);
298 mask->ref_count--;
299
300 if (!mask->ref_count)
301 tbl_mask_array_del_mask(tbl, mask);
302 }
303 }
304
305 int ovs_flow_tbl_init(struct flow_table *table)
306 {
307 struct table_instance *ti, *ufid_ti;
308 struct mask_array *ma;
309
310 table->mask_cache = __alloc_percpu(sizeof(struct mask_cache_entry) *
311 MC_HASH_ENTRIES, __alignof__(struct mask_cache_entry));
312 if (!table->mask_cache)
313 return -ENOMEM;
314
315 ma = tbl_mask_array_alloc(MASK_ARRAY_SIZE_MIN);
316 if (!ma)
317 goto free_mask_cache;
318
319 ti = table_instance_alloc(TBL_MIN_BUCKETS);
320 if (!ti)
321 goto free_mask_array;
322
323 ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
324 if (!ufid_ti)
325 goto free_ti;
326
327 rcu_assign_pointer(table->ti, ti);
328 rcu_assign_pointer(table->ufid_ti, ufid_ti);
329 rcu_assign_pointer(table->mask_array, ma);
330 table->last_rehash = jiffies;
331 table->count = 0;
332 table->ufid_count = 0;
333 return 0;
334
335 free_ti:
336 __table_instance_destroy(ti);
337 free_mask_array:
338 kfree(ma);
339 free_mask_cache:
340 free_percpu(table->mask_cache);
341 return -ENOMEM;
342 }
343
344 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
345 {
346 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
347
348 __table_instance_destroy(ti);
349 }
350
351 static void table_instance_flow_free(struct flow_table *table,
352 struct table_instance *ti,
353 struct table_instance *ufid_ti,
354 struct sw_flow *flow,
355 bool count)
356 {
357 hlist_del_rcu(&flow->flow_table.node[ti->node_ver]);
358 if (count)
359 table->count--;
360
361 if (ovs_identifier_is_ufid(&flow->id)) {
362 hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]);
363
364 if (count)
365 table->ufid_count--;
366 }
367
368 flow_mask_remove(table, flow->mask);
369 }
370
371 static void table_instance_destroy(struct flow_table *table,
372 struct table_instance *ti,
373 struct table_instance *ufid_ti,
374 bool deferred)
375 {
376 int i;
377
378 if (!ti)
379 return;
380
381 BUG_ON(!ufid_ti);
382 if (ti->keep_flows)
383 goto skip_flows;
384
385 for (i = 0; i < ti->n_buckets; i++) {
386 struct sw_flow *flow;
387 struct hlist_head *head = &ti->buckets[i];
388 struct hlist_node *n;
389
390 hlist_for_each_entry_safe(flow, n, head,
391 flow_table.node[ti->node_ver]) {
392
393 table_instance_flow_free(table, ti, ufid_ti,
394 flow, false);
395 ovs_flow_free(flow, deferred);
396 }
397 }
398
399 skip_flows:
400 if (deferred) {
401 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
402 call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb);
403 } else {
404 __table_instance_destroy(ti);
405 __table_instance_destroy(ufid_ti);
406 }
407 }
408
409 /* No need for locking this function is called from RCU callback or
410 * error path.
411 */
412 void ovs_flow_tbl_destroy(struct flow_table *table)
413 {
414 struct table_instance *ti = rcu_dereference_raw(table->ti);
415 struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti);
416
417 free_percpu(table->mask_cache);
418 kfree(rcu_dereference_raw(table->mask_array));
419 table_instance_destroy(table, ti, ufid_ti, false);
420 }
421
422 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
423 u32 *bucket, u32 *last)
424 {
425 struct sw_flow *flow;
426 struct hlist_head *head;
427 int ver;
428 int i;
429
430 ver = ti->node_ver;
431 while (*bucket < ti->n_buckets) {
432 i = 0;
433 head = &ti->buckets[*bucket];
434 hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) {
435 if (i < *last) {
436 i++;
437 continue;
438 }
439 *last = i + 1;
440 return flow;
441 }
442 (*bucket)++;
443 *last = 0;
444 }
445
446 return NULL;
447 }
448
449 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
450 {
451 hash = jhash_1word(hash, ti->hash_seed);
452 return &ti->buckets[hash & (ti->n_buckets - 1)];
453 }
454
455 static void table_instance_insert(struct table_instance *ti,
456 struct sw_flow *flow)
457 {
458 struct hlist_head *head;
459
460 head = find_bucket(ti, flow->flow_table.hash);
461 hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head);
462 }
463
464 static void ufid_table_instance_insert(struct table_instance *ti,
465 struct sw_flow *flow)
466 {
467 struct hlist_head *head;
468
469 head = find_bucket(ti, flow->ufid_table.hash);
470 hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head);
471 }
472
473 static void flow_table_copy_flows(struct table_instance *old,
474 struct table_instance *new, bool ufid)
475 {
476 int old_ver;
477 int i;
478
479 old_ver = old->node_ver;
480 new->node_ver = !old_ver;
481
482 /* Insert in new table. */
483 for (i = 0; i < old->n_buckets; i++) {
484 struct sw_flow *flow;
485 struct hlist_head *head = &old->buckets[i];
486
487 if (ufid)
488 hlist_for_each_entry_rcu(flow, head,
489 ufid_table.node[old_ver])
490 ufid_table_instance_insert(new, flow);
491 else
492 hlist_for_each_entry_rcu(flow, head,
493 flow_table.node[old_ver])
494 table_instance_insert(new, flow);
495 }
496
497 old->keep_flows = true;
498 }
499
500 static struct table_instance *table_instance_rehash(struct table_instance *ti,
501 int n_buckets, bool ufid)
502 {
503 struct table_instance *new_ti;
504
505 new_ti = table_instance_alloc(n_buckets);
506 if (!new_ti)
507 return NULL;
508
509 flow_table_copy_flows(ti, new_ti, ufid);
510
511 return new_ti;
512 }
513
514 int ovs_flow_tbl_flush(struct flow_table *flow_table)
515 {
516 struct table_instance *old_ti, *new_ti;
517 struct table_instance *old_ufid_ti, *new_ufid_ti;
518
519 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
520 if (!new_ti)
521 return -ENOMEM;
522 new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
523 if (!new_ufid_ti)
524 goto err_free_ti;
525
526 old_ti = ovsl_dereference(flow_table->ti);
527 old_ufid_ti = ovsl_dereference(flow_table->ufid_ti);
528
529 rcu_assign_pointer(flow_table->ti, new_ti);
530 rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti);
531 flow_table->last_rehash = jiffies;
532 flow_table->count = 0;
533 flow_table->ufid_count = 0;
534
535 table_instance_destroy(flow_table, old_ti, old_ufid_ti, true);
536 return 0;
537
538 err_free_ti:
539 __table_instance_destroy(new_ti);
540 return -ENOMEM;
541 }
542
543 static u32 flow_hash(const struct sw_flow_key *key,
544 const struct sw_flow_key_range *range)
545 {
546 const u32 *hash_key = (const u32 *)((const u8 *)key + range->start);
547
548 /* Make sure number of hash bytes are multiple of u32. */
549 int hash_u32s = range_n_bytes(range) >> 2;
550
551 return jhash2(hash_key, hash_u32s, 0);
552 }
553
554 static int flow_key_start(const struct sw_flow_key *key)
555 {
556 if (key->tun_proto)
557 return 0;
558 else
559 return rounddown(offsetof(struct sw_flow_key, phy),
560 sizeof(long));
561 }
562
563 static bool cmp_key(const struct sw_flow_key *key1,
564 const struct sw_flow_key *key2,
565 int key_start, int key_end)
566 {
567 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
568 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
569 long diffs = 0;
570 int i;
571
572 for (i = key_start; i < key_end; i += sizeof(long))
573 diffs |= *cp1++ ^ *cp2++;
574
575 return diffs == 0;
576 }
577
578 static bool flow_cmp_masked_key(const struct sw_flow *flow,
579 const struct sw_flow_key *key,
580 const struct sw_flow_key_range *range)
581 {
582 return cmp_key(&flow->key, key, range->start, range->end);
583 }
584
585 static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
586 const struct sw_flow_match *match)
587 {
588 struct sw_flow_key *key = match->key;
589 int key_start = flow_key_start(key);
590 int key_end = match->range.end;
591
592 BUG_ON(ovs_identifier_is_ufid(&flow->id));
593 return cmp_key(flow->id.unmasked_key, key, key_start, key_end);
594 }
595
596 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
597 const struct sw_flow_key *unmasked,
598 const struct sw_flow_mask *mask,
599 u32 *n_mask_hit)
600 {
601 struct sw_flow *flow;
602 struct hlist_head *head;
603 u32 hash;
604 struct sw_flow_key masked_key;
605
606 ovs_flow_mask_key(&masked_key, unmasked, false, mask);
607 hash = flow_hash(&masked_key, &mask->range);
608 head = find_bucket(ti, hash);
609 (*n_mask_hit)++;
610 hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) {
611 if (flow->mask == mask && flow->flow_table.hash == hash &&
612 flow_cmp_masked_key(flow, &masked_key, &mask->range))
613 return flow;
614 }
615 return NULL;
616 }
617
618 /* Flow lookup does full lookup on flow table. It starts with
619 * mask from index passed in *index.
620 */
621 static struct sw_flow *flow_lookup(struct flow_table *tbl,
622 struct table_instance *ti,
623 const struct mask_array *ma,
624 const struct sw_flow_key *key,
625 u32 *n_mask_hit,
626 u32 *index)
627 {
628 struct sw_flow *flow;
629 struct sw_flow_mask *mask;
630 int i;
631
632 if (likely(*index < ma->max)) {
633 mask = rcu_dereference_ovsl(ma->masks[*index]);
634 if (mask) {
635 flow = masked_flow_lookup(ti, key, mask, n_mask_hit);
636 if (flow)
637 return flow;
638 }
639 }
640
641 for (i = 0; i < ma->max; i++) {
642
643 if (i == *index)
644 continue;
645
646 mask = rcu_dereference_ovsl(ma->masks[i]);
647 if (unlikely(!mask))
648 break;
649
650 flow = masked_flow_lookup(ti, key, mask, n_mask_hit);
651 if (flow) { /* Found */
652 *index = i;
653 return flow;
654 }
655 }
656
657 return NULL;
658 }
659
660 /*
661 * mask_cache maps flow to probable mask. This cache is not tightly
662 * coupled cache, It means updates to mask list can result in inconsistent
663 * cache entry in mask cache.
664 * This is per cpu cache and is divided in MC_HASH_SEGS segments.
665 * In case of a hash collision the entry is hashed in next segment.
666 */
667 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
668 const struct sw_flow_key *key,
669 u32 skb_hash,
670 u32 *n_mask_hit)
671 {
672 struct mask_array *ma = rcu_dereference(tbl->mask_array);
673 struct table_instance *ti = rcu_dereference(tbl->ti);
674 struct mask_cache_entry *entries, *ce;
675 struct sw_flow *flow;
676 u32 hash;
677 int seg;
678
679 *n_mask_hit = 0;
680 if (unlikely(!skb_hash)) {
681 u32 mask_index = 0;
682
683 return flow_lookup(tbl, ti, ma, key, n_mask_hit, &mask_index);
684 }
685
686 /* Pre and post recirulation flows usually have the same skb_hash
687 * value. To avoid hash collisions, rehash the 'skb_hash' with
688 * 'recirc_id'. */
689 if (key->recirc_id)
690 skb_hash = jhash_1word(skb_hash, key->recirc_id);
691
692 ce = NULL;
693 hash = skb_hash;
694 entries = this_cpu_ptr(tbl->mask_cache);
695
696 /* Find the cache entry 'ce' to operate on. */
697 for (seg = 0; seg < MC_HASH_SEGS; seg++) {
698 int index = hash & (MC_HASH_ENTRIES - 1);
699 struct mask_cache_entry *e;
700
701 e = &entries[index];
702 if (e->skb_hash == skb_hash) {
703 flow = flow_lookup(tbl, ti, ma, key, n_mask_hit,
704 &e->mask_index);
705 if (!flow)
706 e->skb_hash = 0;
707 return flow;
708 }
709
710 if (!ce || e->skb_hash < ce->skb_hash)
711 ce = e; /* A better replacement cache candidate. */
712
713 hash >>= MC_HASH_SHIFT;
714 }
715
716 /* Cache miss, do full lookup. */
717 flow = flow_lookup(tbl, ti, ma, key, n_mask_hit, &ce->mask_index);
718 if (flow)
719 ce->skb_hash = skb_hash;
720
721 return flow;
722 }
723
724 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
725 const struct sw_flow_key *key)
726 {
727 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
728 struct mask_array *ma = rcu_dereference_ovsl(tbl->mask_array);
729 u32 __always_unused n_mask_hit;
730 u32 index = 0;
731
732 return flow_lookup(tbl, ti, ma, key, &n_mask_hit, &index);
733 }
734
735 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
736 const struct sw_flow_match *match)
737 {
738 struct mask_array *ma = ovsl_dereference(tbl->mask_array);
739 int i;
740
741 /* Always called under ovs-mutex. */
742 for (i = 0; i < ma->max; i++) {
743 struct table_instance *ti = ovsl_dereference(tbl->ti);
744 u32 __always_unused n_mask_hit;
745 struct sw_flow_mask *mask;
746 struct sw_flow *flow;
747
748 mask = ovsl_dereference(ma->masks[i]);
749 if (!mask)
750 continue;
751 flow = masked_flow_lookup(ti, match->key, mask, &n_mask_hit);
752 if (flow && ovs_identifier_is_key(&flow->id) &&
753 ovs_flow_cmp_unmasked_key(flow, match))
754 return flow;
755 }
756 return NULL;
757 }
758
759 static u32 ufid_hash(const struct sw_flow_id *sfid)
760 {
761 return jhash(sfid->ufid, sfid->ufid_len, 0);
762 }
763
764 static bool ovs_flow_cmp_ufid(const struct sw_flow *flow,
765 const struct sw_flow_id *sfid)
766 {
767 if (flow->id.ufid_len != sfid->ufid_len)
768 return false;
769
770 return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len);
771 }
772
773 bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match)
774 {
775 if (ovs_identifier_is_ufid(&flow->id))
776 return flow_cmp_masked_key(flow, match->key, &match->range);
777
778 return ovs_flow_cmp_unmasked_key(flow, match);
779 }
780
781 struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl,
782 const struct sw_flow_id *ufid)
783 {
784 struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti);
785 struct sw_flow *flow;
786 struct hlist_head *head;
787 u32 hash;
788
789 hash = ufid_hash(ufid);
790 head = find_bucket(ti, hash);
791 hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) {
792 if (flow->ufid_table.hash == hash &&
793 ovs_flow_cmp_ufid(flow, ufid))
794 return flow;
795 }
796 return NULL;
797 }
798
799 int ovs_flow_tbl_num_masks(const struct flow_table *table)
800 {
801 struct mask_array *ma;
802
803 ma = rcu_dereference_ovsl(table->mask_array);
804 return READ_ONCE(ma->count);
805 }
806
807 static struct table_instance *table_instance_expand(struct table_instance *ti,
808 bool ufid)
809 {
810 return table_instance_rehash(ti, ti->n_buckets * 2, ufid);
811 }
812
813 /* Must be called with OVS mutex held. */
814 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
815 {
816 struct table_instance *ti = ovsl_dereference(table->ti);
817 struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti);
818
819 BUG_ON(table->count == 0);
820 table_instance_flow_free(table, ti, ufid_ti, flow, true);
821 }
822
823 static struct sw_flow_mask *mask_alloc(void)
824 {
825 struct sw_flow_mask *mask;
826
827 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
828 if (mask)
829 mask->ref_count = 1;
830
831 return mask;
832 }
833
834 static bool mask_equal(const struct sw_flow_mask *a,
835 const struct sw_flow_mask *b)
836 {
837 const u8 *a_ = (const u8 *)&a->key + a->range.start;
838 const u8 *b_ = (const u8 *)&b->key + b->range.start;
839
840 return (a->range.end == b->range.end)
841 && (a->range.start == b->range.start)
842 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
843 }
844
845 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
846 const struct sw_flow_mask *mask)
847 {
848 struct mask_array *ma;
849 int i;
850
851 ma = ovsl_dereference(tbl->mask_array);
852 for (i = 0; i < ma->max; i++) {
853 struct sw_flow_mask *t;
854
855 t = ovsl_dereference(ma->masks[i]);
856 if (t && mask_equal(mask, t))
857 return t;
858 }
859
860 return NULL;
861 }
862
863 /* Add 'mask' into the mask list, if it is not already there. */
864 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
865 const struct sw_flow_mask *new)
866 {
867 struct sw_flow_mask *mask;
868
869 mask = flow_mask_find(tbl, new);
870 if (!mask) {
871 /* Allocate a new mask if none exsits. */
872 mask = mask_alloc();
873 if (!mask)
874 return -ENOMEM;
875
876 mask->key = new->key;
877 mask->range = new->range;
878
879 /* Add mask to mask-list. */
880 if (tbl_mask_array_add_mask(tbl, mask)) {
881 kfree(mask);
882 return -ENOMEM;
883 }
884
885 } else {
886 BUG_ON(!mask->ref_count);
887 mask->ref_count++;
888 }
889
890 flow->mask = mask;
891 return 0;
892 }
893
894 /* Must be called with OVS mutex held. */
895 static void flow_key_insert(struct flow_table *table, struct sw_flow *flow)
896 {
897 struct table_instance *new_ti = NULL;
898 struct table_instance *ti;
899
900 flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range);
901 ti = ovsl_dereference(table->ti);
902 table_instance_insert(ti, flow);
903 table->count++;
904
905 /* Expand table, if necessary, to make room. */
906 if (table->count > ti->n_buckets)
907 new_ti = table_instance_expand(ti, false);
908 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
909 new_ti = table_instance_rehash(ti, ti->n_buckets, false);
910
911 if (new_ti) {
912 rcu_assign_pointer(table->ti, new_ti);
913 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
914 table->last_rehash = jiffies;
915 }
916 }
917
918 /* Must be called with OVS mutex held. */
919 static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow)
920 {
921 struct table_instance *ti;
922
923 flow->ufid_table.hash = ufid_hash(&flow->id);
924 ti = ovsl_dereference(table->ufid_ti);
925 ufid_table_instance_insert(ti, flow);
926 table->ufid_count++;
927
928 /* Expand table, if necessary, to make room. */
929 if (table->ufid_count > ti->n_buckets) {
930 struct table_instance *new_ti;
931
932 new_ti = table_instance_expand(ti, true);
933 if (new_ti) {
934 rcu_assign_pointer(table->ufid_ti, new_ti);
935 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
936 }
937 }
938 }
939
940 /* Must be called with OVS mutex held. */
941 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
942 const struct sw_flow_mask *mask)
943 {
944 int err;
945
946 err = flow_mask_insert(table, flow, mask);
947 if (err)
948 return err;
949 flow_key_insert(table, flow);
950 if (ovs_identifier_is_ufid(&flow->id))
951 flow_ufid_insert(table, flow);
952
953 return 0;
954 }
955
956 /* Initializes the flow module.
957 * Returns zero if successful or a negative error code.
958 */
959 int ovs_flow_init(void)
960 {
961 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
962 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
963
964 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
965 + (nr_cpu_ids
966 * sizeof(struct sw_flow_stats *)),
967 0, 0, NULL);
968 if (flow_cache == NULL)
969 return -ENOMEM;
970
971 flow_stats_cache
972 = kmem_cache_create("sw_flow_stats", sizeof(struct sw_flow_stats),
973 0, SLAB_HWCACHE_ALIGN, NULL);
974 if (flow_stats_cache == NULL) {
975 kmem_cache_destroy(flow_cache);
976 flow_cache = NULL;
977 return -ENOMEM;
978 }
979
980 return 0;
981 }
982
983 /* Uninitializes the flow module. */
984 void ovs_flow_exit(void)
985 {
986 kmem_cache_destroy(flow_stats_cache);
987 kmem_cache_destroy(flow_cache);
988 }