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