]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/flow_table.c
compat: Clean up gre_calc_hlen
[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/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 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 int ovs_flow_tbl_init(struct flow_table *table)
238 {
239 struct table_instance *ti, *ufid_ti;
240 struct mask_array *ma;
241
242 table->mask_cache = __alloc_percpu(sizeof(struct mask_cache_entry) *
243 MC_HASH_ENTRIES, __alignof__(struct mask_cache_entry));
244 if (!table->mask_cache)
245 return -ENOMEM;
246
247 ma = tbl_mask_array_alloc(MASK_ARRAY_SIZE_MIN);
248 if (!ma)
249 goto free_mask_cache;
250
251 ti = table_instance_alloc(TBL_MIN_BUCKETS);
252 if (!ti)
253 goto free_mask_array;
254
255 ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
256 if (!ufid_ti)
257 goto free_ti;
258
259 rcu_assign_pointer(table->ti, ti);
260 rcu_assign_pointer(table->ufid_ti, ufid_ti);
261 rcu_assign_pointer(table->mask_array, ma);
262 table->last_rehash = jiffies;
263 table->count = 0;
264 table->ufid_count = 0;
265 return 0;
266
267 free_ti:
268 __table_instance_destroy(ti);
269 free_mask_array:
270 kfree(ma);
271 free_mask_cache:
272 free_percpu(table->mask_cache);
273 return -ENOMEM;
274 }
275
276 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
277 {
278 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
279
280 __table_instance_destroy(ti);
281 }
282
283 static void table_instance_destroy(struct table_instance *ti,
284 struct table_instance *ufid_ti,
285 bool deferred)
286 {
287 int i;
288
289 if (!ti)
290 return;
291
292 BUG_ON(!ufid_ti);
293 if (ti->keep_flows)
294 goto skip_flows;
295
296 for (i = 0; i < ti->n_buckets; i++) {
297 struct sw_flow *flow;
298 struct hlist_head *head = &ti->buckets[i];
299 struct hlist_node *n;
300 int ver = ti->node_ver;
301 int ufid_ver = ufid_ti->node_ver;
302
303 hlist_for_each_entry_safe(flow, n, head, flow_table.node[ver]) {
304 hlist_del_rcu(&flow->flow_table.node[ver]);
305 if (ovs_identifier_is_ufid(&flow->id))
306 hlist_del_rcu(&flow->ufid_table.node[ufid_ver]);
307 ovs_flow_free(flow, deferred);
308 }
309 }
310
311 skip_flows:
312 if (deferred) {
313 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
314 call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb);
315 } else {
316 __table_instance_destroy(ti);
317 __table_instance_destroy(ufid_ti);
318 }
319 }
320
321 /* No need for locking this function is called from RCU callback or
322 * error path.
323 */
324 void ovs_flow_tbl_destroy(struct flow_table *table)
325 {
326 struct table_instance *ti = rcu_dereference_raw(table->ti);
327 struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti);
328
329 free_percpu(table->mask_cache);
330 kfree(rcu_dereference_raw(table->mask_array));
331 table_instance_destroy(ti, ufid_ti, false);
332 }
333
334 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
335 u32 *bucket, u32 *last)
336 {
337 struct sw_flow *flow;
338 struct hlist_head *head;
339 int ver;
340 int i;
341
342 ver = ti->node_ver;
343 while (*bucket < ti->n_buckets) {
344 i = 0;
345 head = &ti->buckets[*bucket];
346 hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) {
347 if (i < *last) {
348 i++;
349 continue;
350 }
351 *last = i + 1;
352 return flow;
353 }
354 (*bucket)++;
355 *last = 0;
356 }
357
358 return NULL;
359 }
360
361 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
362 {
363 hash = jhash_1word(hash, ti->hash_seed);
364 return &ti->buckets[hash & (ti->n_buckets - 1)];
365 }
366
367 static void table_instance_insert(struct table_instance *ti,
368 struct sw_flow *flow)
369 {
370 struct hlist_head *head;
371
372 head = find_bucket(ti, flow->flow_table.hash);
373 hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head);
374 }
375
376 static void ufid_table_instance_insert(struct table_instance *ti,
377 struct sw_flow *flow)
378 {
379 struct hlist_head *head;
380
381 head = find_bucket(ti, flow->ufid_table.hash);
382 hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head);
383 }
384
385 static void flow_table_copy_flows(struct table_instance *old,
386 struct table_instance *new, bool ufid)
387 {
388 int old_ver;
389 int i;
390
391 old_ver = old->node_ver;
392 new->node_ver = !old_ver;
393
394 /* Insert in new table. */
395 for (i = 0; i < old->n_buckets; i++) {
396 struct sw_flow *flow;
397 struct hlist_head *head = &old->buckets[i];
398
399 if (ufid)
400 hlist_for_each_entry(flow, head,
401 ufid_table.node[old_ver])
402 ufid_table_instance_insert(new, flow);
403 else
404 hlist_for_each_entry(flow, head,
405 flow_table.node[old_ver])
406 table_instance_insert(new, flow);
407 }
408
409 old->keep_flows = true;
410 }
411
412 static struct table_instance *table_instance_rehash(struct table_instance *ti,
413 int n_buckets, bool ufid)
414 {
415 struct table_instance *new_ti;
416
417 new_ti = table_instance_alloc(n_buckets);
418 if (!new_ti)
419 return NULL;
420
421 flow_table_copy_flows(ti, new_ti, ufid);
422
423 return new_ti;
424 }
425
426 int ovs_flow_tbl_flush(struct flow_table *flow_table)
427 {
428 struct table_instance *old_ti, *new_ti;
429 struct table_instance *old_ufid_ti, *new_ufid_ti;
430
431 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
432 if (!new_ti)
433 return -ENOMEM;
434 new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
435 if (!new_ufid_ti)
436 goto err_free_ti;
437
438 old_ti = ovsl_dereference(flow_table->ti);
439 old_ufid_ti = ovsl_dereference(flow_table->ufid_ti);
440
441 rcu_assign_pointer(flow_table->ti, new_ti);
442 rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti);
443 flow_table->last_rehash = jiffies;
444 flow_table->count = 0;
445 flow_table->ufid_count = 0;
446
447 table_instance_destroy(old_ti, old_ufid_ti, true);
448 return 0;
449
450 err_free_ti:
451 __table_instance_destroy(new_ti);
452 return -ENOMEM;
453 }
454
455 static u32 flow_hash(const struct sw_flow_key *key,
456 const struct sw_flow_key_range *range)
457 {
458 int key_start = range->start;
459 int key_end = range->end;
460 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
461 int hash_u32s = (key_end - key_start) >> 2;
462
463 /* Make sure number of hash bytes are multiple of u32. */
464 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
465
466 return jhash2(hash_key, hash_u32s, 0);
467 }
468
469 static int flow_key_start(const struct sw_flow_key *key)
470 {
471 if (key->tun_proto)
472 return 0;
473 else
474 return rounddown(offsetof(struct sw_flow_key, phy),
475 sizeof(long));
476 }
477
478 static bool cmp_key(const struct sw_flow_key *key1,
479 const struct sw_flow_key *key2,
480 int key_start, int key_end)
481 {
482 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
483 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
484 long diffs = 0;
485 int i;
486
487 for (i = key_start; i < key_end; i += sizeof(long))
488 diffs |= *cp1++ ^ *cp2++;
489
490 return diffs == 0;
491 }
492
493 static bool flow_cmp_masked_key(const struct sw_flow *flow,
494 const struct sw_flow_key *key,
495 const struct sw_flow_key_range *range)
496 {
497 return cmp_key(&flow->key, key, range->start, range->end);
498 }
499
500 static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
501 const struct sw_flow_match *match)
502 {
503 struct sw_flow_key *key = match->key;
504 int key_start = flow_key_start(key);
505 int key_end = match->range.end;
506
507 BUG_ON(ovs_identifier_is_ufid(&flow->id));
508 return cmp_key(flow->id.unmasked_key, key, key_start, key_end);
509 }
510
511 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
512 const struct sw_flow_key *unmasked,
513 const struct sw_flow_mask *mask,
514 u32 *n_mask_hit)
515 {
516 struct sw_flow *flow;
517 struct hlist_head *head;
518 u32 hash;
519 struct sw_flow_key masked_key;
520
521 ovs_flow_mask_key(&masked_key, unmasked, false, mask);
522 hash = flow_hash(&masked_key, &mask->range);
523 head = find_bucket(ti, hash);
524 (*n_mask_hit)++;
525 hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) {
526 if (flow->mask == mask && flow->flow_table.hash == hash &&
527 flow_cmp_masked_key(flow, &masked_key, &mask->range))
528 return flow;
529 }
530 return NULL;
531 }
532
533 /* Flow lookup does full lookup on flow table. It starts with
534 * mask from index passed in *index.
535 */
536 static struct sw_flow *flow_lookup(struct flow_table *tbl,
537 struct table_instance *ti,
538 const struct mask_array *ma,
539 const struct sw_flow_key *key,
540 u32 *n_mask_hit,
541 u32 *index)
542 {
543 struct sw_flow_mask *mask;
544 struct sw_flow *flow;
545 int i;
546
547 if (*index < ma->max) {
548 mask = rcu_dereference_ovsl(ma->masks[*index]);
549 if (mask) {
550 flow = masked_flow_lookup(ti, key, mask, n_mask_hit);
551 if (flow)
552 return flow;
553 }
554 }
555
556 for (i = 0; i < ma->max; i++) {
557
558 if (i == *index)
559 continue;
560
561 mask = rcu_dereference_ovsl(ma->masks[i]);
562 if (!mask)
563 continue;
564
565 flow = masked_flow_lookup(ti, key, mask, n_mask_hit);
566 if (flow) { /* Found */
567 *index = i;
568 return flow;
569 }
570 }
571
572 return NULL;
573 }
574
575 /*
576 * mask_cache maps flow to probable mask. This cache is not tightly
577 * coupled cache, It means updates to mask list can result in inconsistent
578 * cache entry in mask cache.
579 * This is per cpu cache and is divided in MC_HASH_SEGS segments.
580 * In case of a hash collision the entry is hashed in next segment.
581 */
582 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
583 const struct sw_flow_key *key,
584 u32 skb_hash,
585 u32 *n_mask_hit)
586 {
587 struct mask_array *ma = rcu_dereference(tbl->mask_array);
588 struct table_instance *ti = rcu_dereference(tbl->ti);
589 struct mask_cache_entry *entries, *ce;
590 struct sw_flow *flow;
591 u32 hash;
592 int seg;
593
594 *n_mask_hit = 0;
595 if (unlikely(!skb_hash)) {
596 u32 mask_index = 0;
597
598 return flow_lookup(tbl, ti, ma, key, n_mask_hit, &mask_index);
599 }
600
601 /* Pre and post recirulation flows usually have the same skb_hash
602 * value. To avoid hash collisions, rehash the 'skb_hash' with
603 * 'recirc_id'. */
604 if (key->recirc_id)
605 skb_hash = jhash_1word(skb_hash, key->recirc_id);
606
607 ce = NULL;
608 hash = skb_hash;
609 entries = this_cpu_ptr(tbl->mask_cache);
610
611 /* Find the cache entry 'ce' to operate on. */
612 for (seg = 0; seg < MC_HASH_SEGS; seg++) {
613 int index = hash & (MC_HASH_ENTRIES - 1);
614 struct mask_cache_entry *e;
615
616 e = &entries[index];
617 if (e->skb_hash == skb_hash) {
618 flow = flow_lookup(tbl, ti, ma, key, n_mask_hit,
619 &e->mask_index);
620 if (!flow)
621 e->skb_hash = 0;
622 return flow;
623 }
624
625 if (!ce || e->skb_hash < ce->skb_hash)
626 ce = e; /* A better replacement cache candidate. */
627
628 hash >>= MC_HASH_SHIFT;
629 }
630
631 /* Cache miss, do full lookup. */
632 flow = flow_lookup(tbl, ti, ma, key, n_mask_hit, &ce->mask_index);
633 if (flow)
634 ce->skb_hash = skb_hash;
635
636 return flow;
637 }
638
639 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
640 const struct sw_flow_key *key)
641 {
642 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
643 struct mask_array *ma = rcu_dereference_ovsl(tbl->mask_array);
644 u32 __always_unused n_mask_hit;
645 u32 index = 0;
646
647 return flow_lookup(tbl, ti, ma, key, &n_mask_hit, &index);
648 }
649
650 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
651 const struct sw_flow_match *match)
652 {
653 struct mask_array *ma = ovsl_dereference(tbl->mask_array);
654 int i;
655
656 /* Always called under ovs-mutex. */
657 for (i = 0; i < ma->max; i++) {
658 struct table_instance *ti = ovsl_dereference(tbl->ti);
659 u32 __always_unused n_mask_hit;
660 struct sw_flow_mask *mask;
661 struct sw_flow *flow;
662
663 mask = ovsl_dereference(ma->masks[i]);
664 if (!mask)
665 continue;
666 flow = masked_flow_lookup(ti, match->key, mask, &n_mask_hit);
667 if (flow && ovs_identifier_is_key(&flow->id) &&
668 ovs_flow_cmp_unmasked_key(flow, match))
669 return flow;
670 }
671 return NULL;
672 }
673
674 static u32 ufid_hash(const struct sw_flow_id *sfid)
675 {
676 return jhash(sfid->ufid, sfid->ufid_len, 0);
677 }
678
679 static bool ovs_flow_cmp_ufid(const struct sw_flow *flow,
680 const struct sw_flow_id *sfid)
681 {
682 if (flow->id.ufid_len != sfid->ufid_len)
683 return false;
684
685 return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len);
686 }
687
688 bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match)
689 {
690 if (ovs_identifier_is_ufid(&flow->id))
691 return flow_cmp_masked_key(flow, match->key, &match->range);
692
693 return ovs_flow_cmp_unmasked_key(flow, match);
694 }
695
696 struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl,
697 const struct sw_flow_id *ufid)
698 {
699 struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti);
700 struct sw_flow *flow;
701 struct hlist_head *head;
702 u32 hash;
703
704 hash = ufid_hash(ufid);
705 head = find_bucket(ti, hash);
706 hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) {
707 if (flow->ufid_table.hash == hash &&
708 ovs_flow_cmp_ufid(flow, ufid))
709 return flow;
710 }
711 return NULL;
712 }
713
714 int ovs_flow_tbl_num_masks(const struct flow_table *table)
715 {
716 struct mask_array *ma;
717
718 ma = rcu_dereference_ovsl(table->mask_array);
719 return ma->count;
720 }
721
722 static struct table_instance *table_instance_expand(struct table_instance *ti,
723 bool ufid)
724 {
725 return table_instance_rehash(ti, ti->n_buckets * 2, ufid);
726 }
727
728 static void tbl_mask_array_delete_mask(struct mask_array *ma,
729 struct sw_flow_mask *mask)
730 {
731 int i;
732
733 /* Remove the deleted mask pointers from the array */
734 for (i = 0; i < ma->max; i++) {
735 if (mask == ovsl_dereference(ma->masks[i])) {
736 RCU_INIT_POINTER(ma->masks[i], NULL);
737 ma->count--;
738 kfree_rcu(mask, rcu);
739 return;
740 }
741 }
742 BUG();
743 }
744
745 /* Remove 'mask' from the mask list, if it is not needed any more. */
746 static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
747 {
748 if (mask) {
749 /* ovs-lock is required to protect mask-refcount and
750 * mask list.
751 */
752 ASSERT_OVSL();
753 BUG_ON(!mask->ref_count);
754 mask->ref_count--;
755
756 if (!mask->ref_count) {
757 struct mask_array *ma;
758
759 ma = ovsl_dereference(tbl->mask_array);
760 tbl_mask_array_delete_mask(ma, mask);
761
762 /* Shrink the mask array if necessary. */
763 if (ma->max >= (MASK_ARRAY_SIZE_MIN * 2) &&
764 ma->count <= (ma->max / 3))
765 tbl_mask_array_realloc(tbl, ma->max / 2);
766
767 }
768 }
769 }
770
771 /* Must be called with OVS mutex held. */
772 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
773 {
774 struct table_instance *ti = ovsl_dereference(table->ti);
775 struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti);
776
777 BUG_ON(table->count == 0);
778 hlist_del_rcu(&flow->flow_table.node[ti->node_ver]);
779 table->count--;
780 if (ovs_identifier_is_ufid(&flow->id)) {
781 hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]);
782 table->ufid_count--;
783 }
784
785 /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
786 * accessible as long as the RCU read lock is held.
787 */
788 flow_mask_remove(table, flow->mask);
789 }
790
791 static struct sw_flow_mask *mask_alloc(void)
792 {
793 struct sw_flow_mask *mask;
794
795 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
796 if (mask)
797 mask->ref_count = 1;
798
799 return mask;
800 }
801
802 static bool mask_equal(const struct sw_flow_mask *a,
803 const struct sw_flow_mask *b)
804 {
805 const u8 *a_ = (const u8 *)&a->key + a->range.start;
806 const u8 *b_ = (const u8 *)&b->key + b->range.start;
807
808 return (a->range.end == b->range.end)
809 && (a->range.start == b->range.start)
810 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
811 }
812
813 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
814 const struct sw_flow_mask *mask)
815 {
816 struct mask_array *ma;
817 int i;
818
819 ma = ovsl_dereference(tbl->mask_array);
820 for (i = 0; i < ma->max; i++) {
821 struct sw_flow_mask *t;
822
823 t = ovsl_dereference(ma->masks[i]);
824 if (t && mask_equal(mask, t))
825 return t;
826 }
827
828 return NULL;
829 }
830
831 /* Add 'mask' into the mask list, if it is not already there. */
832 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
833 const struct sw_flow_mask *new)
834 {
835 struct sw_flow_mask *mask;
836
837 mask = flow_mask_find(tbl, new);
838 if (!mask) {
839 struct mask_array *ma;
840 int i;
841
842 /* Allocate a new mask if none exsits. */
843 mask = mask_alloc();
844 if (!mask)
845 return -ENOMEM;
846
847 mask->key = new->key;
848 mask->range = new->range;
849
850 /* Add mask to mask-list. */
851 ma = ovsl_dereference(tbl->mask_array);
852 if (ma->count >= ma->max) {
853 int err;
854
855 err = tbl_mask_array_realloc(tbl, ma->max +
856 MASK_ARRAY_SIZE_MIN);
857 if (err) {
858 kfree(mask);
859 return err;
860 }
861 ma = ovsl_dereference(tbl->mask_array);
862 }
863
864 for (i = 0; i < ma->max; i++) {
865 struct sw_flow_mask *t;
866
867 t = ovsl_dereference(ma->masks[i]);
868 if (!t) {
869 rcu_assign_pointer(ma->masks[i], mask);
870 ma->count++;
871 break;
872 }
873 }
874
875 } else {
876 BUG_ON(!mask->ref_count);
877 mask->ref_count++;
878 }
879
880 flow->mask = mask;
881 return 0;
882 }
883
884 /* Must be called with OVS mutex held. */
885 static void flow_key_insert(struct flow_table *table, struct sw_flow *flow)
886 {
887 struct table_instance *new_ti = NULL;
888 struct table_instance *ti;
889
890 flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range);
891 ti = ovsl_dereference(table->ti);
892 table_instance_insert(ti, flow);
893 table->count++;
894
895 /* Expand table, if necessary, to make room. */
896 if (table->count > ti->n_buckets)
897 new_ti = table_instance_expand(ti, false);
898 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
899 new_ti = table_instance_rehash(ti, ti->n_buckets, false);
900
901 if (new_ti) {
902 rcu_assign_pointer(table->ti, new_ti);
903 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
904 table->last_rehash = jiffies;
905 }
906 }
907
908 /* Must be called with OVS mutex held. */
909 static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow)
910 {
911 struct table_instance *ti;
912
913 flow->ufid_table.hash = ufid_hash(&flow->id);
914 ti = ovsl_dereference(table->ufid_ti);
915 ufid_table_instance_insert(ti, flow);
916 table->ufid_count++;
917
918 /* Expand table, if necessary, to make room. */
919 if (table->ufid_count > ti->n_buckets) {
920 struct table_instance *new_ti;
921
922 new_ti = table_instance_expand(ti, true);
923 if (new_ti) {
924 rcu_assign_pointer(table->ufid_ti, new_ti);
925 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
926 }
927 }
928 }
929
930 /* Must be called with OVS mutex held. */
931 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
932 const struct sw_flow_mask *mask)
933 {
934 int err;
935
936 err = flow_mask_insert(table, flow, mask);
937 if (err)
938 return err;
939 flow_key_insert(table, flow);
940 if (ovs_identifier_is_ufid(&flow->id))
941 flow_ufid_insert(table, flow);
942
943 return 0;
944 }
945
946 /* Initializes the flow module.
947 * Returns zero if successful or a negative error code.
948 */
949 int ovs_flow_init(void)
950 {
951 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
952 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
953
954 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
955 + (nr_cpu_ids
956 * sizeof(struct flow_stats *)),
957 0, 0, NULL);
958 if (flow_cache == NULL)
959 return -ENOMEM;
960
961 flow_stats_cache
962 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
963 0, SLAB_HWCACHE_ALIGN, NULL);
964 if (flow_stats_cache == NULL) {
965 kmem_cache_destroy(flow_cache);
966 flow_cache = NULL;
967 return -ENOMEM;
968 }
969
970 return 0;
971 }
972
973 /* Uninitializes the flow module. */
974 void ovs_flow_exit(void)
975 {
976 kmem_cache_destroy(flow_stats_cache);
977 kmem_cache_destroy(flow_cache);
978 }