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;
52 static u16 range_n_bytes(const struct sw_flow_key_range *range)
54 return range->end - range->start;
57 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
58 const struct sw_flow_mask *mask)
60 const long *m = (long *)((u8 *)&mask->key + mask->range.start);
61 const long *s = (long *)((u8 *)src + mask->range.start);
62 long *d = (long *)((u8 *)dst + mask->range.start);
65 /* The memory outside of the 'mask->range' are not set since
66 * further operations on 'dst' only uses contents within
69 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
73 struct sw_flow *ovs_flow_alloc(bool percpu_stats)
78 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
80 return ERR_PTR(-ENOMEM);
85 flow->stats.is_percpu = percpu_stats;
88 flow->stats.stat = kzalloc(sizeof(*flow->stats.stat), GFP_KERNEL);
89 if (!flow->stats.stat)
92 spin_lock_init(&flow->stats.stat->lock);
94 flow->stats.cpu_stats = alloc_percpu(struct flow_stats);
95 if (!flow->stats.cpu_stats)
98 for_each_possible_cpu(cpu) {
99 struct flow_stats *cpu_stats;
101 cpu_stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
102 spin_lock_init(&cpu_stats->lock);
107 kmem_cache_free(flow_cache, flow);
108 return ERR_PTR(-ENOMEM);
111 int ovs_flow_tbl_count(struct flow_table *table)
116 static struct flex_array *alloc_buckets(unsigned int n_buckets)
118 struct flex_array *buckets;
121 buckets = flex_array_alloc(sizeof(struct hlist_head),
122 n_buckets, GFP_KERNEL);
126 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
128 flex_array_free(buckets);
132 for (i = 0; i < n_buckets; i++)
133 INIT_HLIST_HEAD((struct hlist_head *)
134 flex_array_get(buckets, i));
139 static void flow_free(struct sw_flow *flow)
141 kfree((struct sf_flow_acts __force *)flow->sf_acts);
142 if (flow->stats.is_percpu)
143 free_percpu(flow->stats.cpu_stats);
145 kfree(flow->stats.stat);
146 kmem_cache_free(flow_cache, flow);
149 static void rcu_free_flow_callback(struct rcu_head *rcu)
151 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
156 static void flow_mask_del_ref(struct sw_flow_mask *mask, bool deferred)
161 BUG_ON(!mask->ref_count);
164 if (!mask->ref_count) {
165 list_del_rcu(&mask->list);
167 kfree_rcu(mask, rcu);
173 void ovs_flow_free(struct sw_flow *flow, bool deferred)
178 flow_mask_del_ref(flow->mask, deferred);
181 call_rcu(&flow->rcu, rcu_free_flow_callback);
186 static void free_buckets(struct flex_array *buckets)
188 flex_array_free(buckets);
191 static void __table_instance_destroy(struct table_instance *ti)
198 for (i = 0; i < ti->n_buckets; i++) {
199 struct sw_flow *flow;
200 struct hlist_head *head = flex_array_get(ti->buckets, i);
201 struct hlist_node *n;
202 int ver = ti->node_ver;
204 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
205 hlist_del(&flow->hash_node[ver]);
206 ovs_flow_free(flow, false);
211 free_buckets(ti->buckets);
215 static struct table_instance *table_instance_alloc(int new_size)
217 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
222 ti->buckets = alloc_buckets(new_size);
228 ti->n_buckets = new_size;
230 ti->keep_flows = false;
231 get_random_bytes(&ti->hash_seed, sizeof(u32));
236 int ovs_flow_tbl_init(struct flow_table *table)
238 struct table_instance *ti;
240 ti = table_instance_alloc(TBL_MIN_BUCKETS);
245 rcu_assign_pointer(table->ti, ti);
246 INIT_LIST_HEAD(&table->mask_list);
247 table->last_rehash = jiffies;
252 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
254 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
256 __table_instance_destroy(ti);
259 static void table_instance_destroy(struct table_instance *ti, bool deferred)
265 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
267 __table_instance_destroy(ti);
270 void ovs_flow_tbl_destroy(struct flow_table *table)
272 struct table_instance *ti = ovsl_dereference(table->ti);
274 table_instance_destroy(ti, false);
277 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
278 u32 *bucket, u32 *last)
280 struct sw_flow *flow;
281 struct hlist_head *head;
286 while (*bucket < ti->n_buckets) {
288 head = flex_array_get(ti->buckets, *bucket);
289 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
304 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
306 hash = jhash_1word(hash, ti->hash_seed);
307 return flex_array_get(ti->buckets,
308 (hash & (ti->n_buckets - 1)));
311 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
313 struct hlist_head *head;
315 head = find_bucket(ti, flow->hash);
316 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
319 static void flow_table_copy_flows(struct table_instance *old,
320 struct table_instance *new)
325 old_ver = old->node_ver;
326 new->node_ver = !old_ver;
328 /* Insert in new table. */
329 for (i = 0; i < old->n_buckets; i++) {
330 struct sw_flow *flow;
331 struct hlist_head *head;
333 head = flex_array_get(old->buckets, i);
335 hlist_for_each_entry(flow, head, hash_node[old_ver])
336 table_instance_insert(new, flow);
339 old->keep_flows = true;
342 static struct table_instance *table_instance_rehash(struct table_instance *ti,
345 struct table_instance *new_ti;
347 new_ti = table_instance_alloc(n_buckets);
351 flow_table_copy_flows(ti, new_ti);
356 int ovs_flow_tbl_flush(struct flow_table *flow_table)
358 struct table_instance *old_ti;
359 struct table_instance *new_ti;
361 old_ti = ovsl_dereference(flow_table->ti);
362 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
366 rcu_assign_pointer(flow_table->ti, new_ti);
367 flow_table->last_rehash = jiffies;
368 flow_table->count = 0;
370 table_instance_destroy(old_ti, true);
374 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
377 u32 *hash_key = (u32 *)((u8 *)key + key_start);
378 int hash_u32s = (key_end - key_start) >> 2;
380 /* Make sure number of hash bytes are multiple of u32. */
381 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
383 return arch_fast_hash2(hash_key, hash_u32s, 0);
386 static int flow_key_start(const struct sw_flow_key *key)
388 if (key->tun_key.ipv4_dst)
391 return rounddown(offsetof(struct sw_flow_key, phy),
395 static bool cmp_key(const struct sw_flow_key *key1,
396 const struct sw_flow_key *key2,
397 int key_start, int key_end)
399 const long *cp1 = (long *)((u8 *)key1 + key_start);
400 const long *cp2 = (long *)((u8 *)key2 + key_start);
404 for (i = key_start; i < key_end; i += sizeof(long))
405 diffs |= *cp1++ ^ *cp2++;
410 static bool flow_cmp_masked_key(const struct sw_flow *flow,
411 const struct sw_flow_key *key,
412 int key_start, int key_end)
414 return cmp_key(&flow->key, key, key_start, key_end);
417 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
418 struct sw_flow_match *match)
420 struct sw_flow_key *key = match->key;
421 int key_start = flow_key_start(key);
422 int key_end = match->range.end;
424 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
427 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
428 const struct sw_flow_key *unmasked,
429 struct sw_flow_mask *mask)
431 struct sw_flow *flow;
432 struct hlist_head *head;
433 int key_start = mask->range.start;
434 int key_end = mask->range.end;
436 struct sw_flow_key masked_key;
438 ovs_flow_mask_key(&masked_key, unmasked, mask);
439 hash = flow_hash(&masked_key, key_start, key_end);
440 head = find_bucket(ti, hash);
441 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
442 if (flow->mask == mask && flow->hash == hash &&
443 flow_cmp_masked_key(flow, &masked_key,
450 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
451 const struct sw_flow_key *key,
454 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
455 struct sw_flow_mask *mask;
456 struct sw_flow *flow;
459 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
461 flow = masked_flow_lookup(ti, key, mask);
462 if (flow) /* Found */
468 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
469 const struct sw_flow_key *key)
471 u32 __always_unused n_mask_hit;
473 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
476 int ovs_flow_tbl_num_masks(const struct flow_table *table)
478 struct sw_flow_mask *mask;
481 list_for_each_entry(mask, &table->mask_list, list)
487 static struct table_instance *table_instance_expand(struct table_instance *ti)
489 return table_instance_rehash(ti, ti->n_buckets * 2);
492 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
494 struct table_instance *ti = ovsl_dereference(table->ti);
496 BUG_ON(table->count == 0);
497 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
501 static struct sw_flow_mask *mask_alloc(void)
503 struct sw_flow_mask *mask;
505 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
512 static void mask_add_ref(struct sw_flow_mask *mask)
517 static bool mask_equal(const struct sw_flow_mask *a,
518 const struct sw_flow_mask *b)
520 u8 *a_ = (u8 *)&a->key + a->range.start;
521 u8 *b_ = (u8 *)&b->key + b->range.start;
523 return (a->range.end == b->range.end)
524 && (a->range.start == b->range.start)
525 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
528 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
529 const struct sw_flow_mask *mask)
531 struct list_head *ml;
533 list_for_each(ml, &tbl->mask_list) {
534 struct sw_flow_mask *m;
535 m = container_of(ml, struct sw_flow_mask, list);
536 if (mask_equal(mask, m))
543 /* Add 'mask' into the mask list, if it is not already there. */
544 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
545 struct sw_flow_mask *new)
547 struct sw_flow_mask *mask;
548 mask = flow_mask_find(tbl, new);
550 /* Allocate a new mask if none exsits. */
554 mask->key = new->key;
555 mask->range = new->range;
556 list_add_rcu(&mask->list, &tbl->mask_list);
564 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
565 struct sw_flow_mask *mask)
567 struct table_instance *new_ti = NULL;
568 struct table_instance *ti;
571 err = flow_mask_insert(table, flow, mask);
575 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
576 flow->mask->range.end);
577 ti = ovsl_dereference(table->ti);
578 table_instance_insert(ti, flow);
581 /* Expand table, if necessary, to make room. */
582 if (table->count > ti->n_buckets)
583 new_ti = table_instance_expand(ti);
584 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
585 new_ti = table_instance_rehash(ti, ti->n_buckets);
588 rcu_assign_pointer(table->ti, new_ti);
589 table_instance_destroy(ti, true);
590 table->last_rehash = jiffies;
595 /* Initializes the flow module.
596 * Returns zero if successful or a negative error code. */
597 int ovs_flow_init(void)
599 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
600 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
602 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
604 if (flow_cache == NULL)
610 /* Uninitializes the flow module. */
611 void ovs_flow_exit(void)
613 kmem_cache_destroy(flow_cache);