2 * Copyright (c) 2007-2013 Nicira, Inc.
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.
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.
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
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>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.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>
45 #include <net/ndisc.h>
47 #define TBL_MIN_BUCKETS 1024
48 #define REHASH_INTERVAL (10 * 60 * HZ)
50 static struct kmem_cache *flow_cache;
51 struct kmem_cache *flow_stats_cache __read_mostly;
53 static u16 range_n_bytes(const struct sw_flow_key_range *range)
55 return range->end - range->start;
58 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
59 const struct sw_flow_mask *mask)
61 const long *m = (const long *)((const u8 *)&mask->key +
63 const long *s = (const long *)((const u8 *)src +
65 long *d = (long *)((u8 *)dst + mask->range.start);
68 /* The memory outside of the 'mask->range' are not set since
69 * further operations on 'dst' only uses contents within
72 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
76 struct sw_flow *ovs_flow_alloc(void)
79 struct flow_stats *stats;
82 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
84 return ERR_PTR(-ENOMEM);
88 flow->stats_last_writer = NUMA_NO_NODE;
90 /* Initialize the default stat node. */
91 stats = kmem_cache_alloc_node(flow_stats_cache,
92 GFP_KERNEL | __GFP_ZERO, 0);
96 spin_lock_init(&stats->lock);
98 RCU_INIT_POINTER(flow->stats[0], stats);
102 RCU_INIT_POINTER(flow->stats[node], NULL);
106 kmem_cache_free(flow_cache, flow);
107 return ERR_PTR(-ENOMEM);
110 int ovs_flow_tbl_count(struct flow_table *table)
115 static struct flex_array *alloc_buckets(unsigned int n_buckets)
117 struct flex_array *buckets;
120 buckets = flex_array_alloc(sizeof(struct hlist_head),
121 n_buckets, GFP_KERNEL);
125 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
127 flex_array_free(buckets);
131 for (i = 0; i < n_buckets; i++)
132 INIT_HLIST_HEAD((struct hlist_head *)
133 flex_array_get(buckets, i));
138 static void flow_free(struct sw_flow *flow)
142 kfree((struct sf_flow_acts __force *)flow->sf_acts);
144 if (flow->stats[node])
145 kmem_cache_free(flow_stats_cache,
146 (struct flow_stats __force *)flow->stats[node]);
147 kmem_cache_free(flow_cache, flow);
150 static void rcu_free_flow_callback(struct rcu_head *rcu)
152 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
157 void ovs_flow_free(struct sw_flow *flow, bool deferred)
163 struct sw_flow_mask *mask = flow->mask;
165 /* ovs-lock is required to protect mask-refcount and
169 BUG_ON(!mask->ref_count);
172 if (!mask->ref_count) {
173 list_del_rcu(&mask->list);
175 kfree_rcu(mask, rcu);
182 call_rcu(&flow->rcu, rcu_free_flow_callback);
187 static void free_buckets(struct flex_array *buckets)
189 flex_array_free(buckets);
193 static void __table_instance_destroy(struct table_instance *ti)
195 free_buckets(ti->buckets);
199 static struct table_instance *table_instance_alloc(int new_size)
201 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
206 ti->buckets = alloc_buckets(new_size);
212 ti->n_buckets = new_size;
214 ti->keep_flows = false;
215 get_random_bytes(&ti->hash_seed, sizeof(u32));
220 int ovs_flow_tbl_init(struct flow_table *table)
222 struct table_instance *ti;
224 ti = table_instance_alloc(TBL_MIN_BUCKETS);
229 rcu_assign_pointer(table->ti, ti);
230 INIT_LIST_HEAD(&table->mask_list);
231 table->last_rehash = jiffies;
236 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
238 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
240 __table_instance_destroy(ti);
243 static void table_instance_destroy(struct table_instance *ti, bool deferred)
253 for (i = 0; i < ti->n_buckets; i++) {
254 struct sw_flow *flow;
255 struct hlist_head *head = flex_array_get(ti->buckets, i);
256 struct hlist_node *n;
257 int ver = ti->node_ver;
259 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
260 hlist_del_rcu(&flow->hash_node[ver]);
261 ovs_flow_free(flow, deferred);
267 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
269 __table_instance_destroy(ti);
272 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
274 struct table_instance *ti = ovsl_dereference(table->ti);
276 table_instance_destroy(ti, deferred);
279 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
280 u32 *bucket, u32 *last)
282 struct sw_flow *flow;
283 struct hlist_head *head;
288 while (*bucket < ti->n_buckets) {
290 head = flex_array_get(ti->buckets, *bucket);
291 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
306 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
308 hash = jhash_1word(hash, ti->hash_seed);
309 return flex_array_get(ti->buckets,
310 (hash & (ti->n_buckets - 1)));
313 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
315 struct hlist_head *head;
317 head = find_bucket(ti, flow->hash);
318 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
321 static void flow_table_copy_flows(struct table_instance *old,
322 struct table_instance *new)
327 old_ver = old->node_ver;
328 new->node_ver = !old_ver;
330 /* Insert in new table. */
331 for (i = 0; i < old->n_buckets; i++) {
332 struct sw_flow *flow;
333 struct hlist_head *head;
335 head = flex_array_get(old->buckets, i);
337 hlist_for_each_entry(flow, head, hash_node[old_ver])
338 table_instance_insert(new, flow);
341 old->keep_flows = true;
344 static struct table_instance *table_instance_rehash(struct table_instance *ti,
347 struct table_instance *new_ti;
349 new_ti = table_instance_alloc(n_buckets);
353 flow_table_copy_flows(ti, new_ti);
358 int ovs_flow_tbl_flush(struct flow_table *flow_table)
360 struct table_instance *old_ti;
361 struct table_instance *new_ti;
363 old_ti = ovsl_dereference(flow_table->ti);
364 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
368 rcu_assign_pointer(flow_table->ti, new_ti);
369 flow_table->last_rehash = jiffies;
370 flow_table->count = 0;
372 table_instance_destroy(old_ti, true);
376 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
379 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
380 int hash_u32s = (key_end - key_start) >> 2;
382 /* Make sure number of hash bytes are multiple of u32. */
383 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
385 return arch_fast_hash2(hash_key, hash_u32s, 0);
388 static int flow_key_start(const struct sw_flow_key *key)
390 if (key->tun_key.ipv4_dst)
393 return rounddown(offsetof(struct sw_flow_key, phy),
397 static bool cmp_key(const struct sw_flow_key *key1,
398 const struct sw_flow_key *key2,
399 int key_start, int key_end)
401 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
402 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
406 for (i = key_start; i < key_end; i += sizeof(long))
407 diffs |= *cp1++ ^ *cp2++;
412 static bool flow_cmp_masked_key(const struct sw_flow *flow,
413 const struct sw_flow_key *key,
414 int key_start, int key_end)
416 return cmp_key(&flow->key, key, key_start, key_end);
419 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
420 struct sw_flow_match *match)
422 struct sw_flow_key *key = match->key;
423 int key_start = flow_key_start(key);
424 int key_end = match->range.end;
426 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
429 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
430 const struct sw_flow_key *unmasked,
431 struct sw_flow_mask *mask)
433 struct sw_flow *flow;
434 struct hlist_head *head;
435 int key_start = mask->range.start;
436 int key_end = mask->range.end;
438 struct sw_flow_key masked_key;
440 ovs_flow_mask_key(&masked_key, unmasked, mask);
441 hash = flow_hash(&masked_key, key_start, key_end);
442 head = find_bucket(ti, hash);
443 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
444 if (flow->mask == mask && flow->hash == hash &&
445 flow_cmp_masked_key(flow, &masked_key,
452 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
453 const struct sw_flow_key *key,
456 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
457 struct sw_flow_mask *mask;
458 struct sw_flow *flow;
461 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
463 flow = masked_flow_lookup(ti, key, mask);
464 if (flow) /* Found */
470 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
471 const struct sw_flow_key *key)
473 u32 __always_unused n_mask_hit;
475 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
478 int ovs_flow_tbl_num_masks(const struct flow_table *table)
480 struct sw_flow_mask *mask;
483 list_for_each_entry(mask, &table->mask_list, list)
489 static struct table_instance *table_instance_expand(struct table_instance *ti)
491 return table_instance_rehash(ti, ti->n_buckets * 2);
494 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
496 struct table_instance *ti = ovsl_dereference(table->ti);
498 BUG_ON(table->count == 0);
499 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
503 static struct sw_flow_mask *mask_alloc(void)
505 struct sw_flow_mask *mask;
507 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
514 static bool mask_equal(const struct sw_flow_mask *a,
515 const struct sw_flow_mask *b)
517 const u8 *a_ = (const u8 *)&a->key + a->range.start;
518 const u8 *b_ = (const u8 *)&b->key + b->range.start;
520 return (a->range.end == b->range.end)
521 && (a->range.start == b->range.start)
522 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
525 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
526 const struct sw_flow_mask *mask)
528 struct list_head *ml;
530 list_for_each(ml, &tbl->mask_list) {
531 struct sw_flow_mask *m;
532 m = container_of(ml, struct sw_flow_mask, list);
533 if (mask_equal(mask, m))
540 /* Add 'mask' into the mask list, if it is not already there. */
541 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
542 struct sw_flow_mask *new)
544 struct sw_flow_mask *mask;
545 mask = flow_mask_find(tbl, new);
547 /* Allocate a new mask if none exsits. */
551 mask->key = new->key;
552 mask->range = new->range;
553 list_add_rcu(&mask->list, &tbl->mask_list);
555 BUG_ON(!mask->ref_count);
563 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
564 struct sw_flow_mask *mask)
566 struct table_instance *new_ti = NULL;
567 struct table_instance *ti;
570 err = flow_mask_insert(table, flow, mask);
574 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
575 flow->mask->range.end);
576 ti = ovsl_dereference(table->ti);
577 table_instance_insert(ti, flow);
580 /* Expand table, if necessary, to make room. */
581 if (table->count > ti->n_buckets)
582 new_ti = table_instance_expand(ti);
583 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
584 new_ti = table_instance_rehash(ti, ti->n_buckets);
587 rcu_assign_pointer(table->ti, new_ti);
588 table_instance_destroy(ti, true);
589 table->last_rehash = jiffies;
594 /* Initializes the flow module.
595 * Returns zero if successful or a negative error code. */
596 int ovs_flow_init(void)
598 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
599 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
601 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
602 + (num_possible_nodes()
603 * sizeof(struct flow_stats *)),
605 if (flow_cache == NULL)
609 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
610 0, SLAB_HWCACHE_ALIGN, NULL);
611 if (flow_stats_cache == NULL) {
612 kmem_cache_destroy(flow_cache);
620 /* Uninitializes the flow module. */
621 void ovs_flow_exit(void)
623 kmem_cache_destroy(flow_stats_cache);
624 kmem_cache_destroy(flow_cache);