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>
49 #define TBL_MIN_BUCKETS 1024
50 #define REHASH_INTERVAL (10 * 60 * HZ)
52 static struct kmem_cache *flow_cache;
54 static u16 range_n_bytes(const struct sw_flow_key_range *range)
56 return range->end - range->start;
59 void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
60 const struct sw_flow_mask *mask)
62 const long *m = (const long *)((const u8 *)&mask->key +
64 const long *s = (const long *)((const u8 *)src +
66 long *d = (long *)((u8 *)dst + mask->range.start);
69 /* The memory outside of the 'mask->range' are not set since
70 * further operations on 'dst' only uses contents within
73 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
77 struct sw_flow *ovs_flow_alloc(bool percpu_stats)
82 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
84 return ERR_PTR(-ENOMEM);
89 flow->stats.is_percpu = percpu_stats;
92 flow->stats.stat = kzalloc(sizeof(*flow->stats.stat), GFP_KERNEL);
93 if (!flow->stats.stat)
96 spin_lock_init(&flow->stats.stat->lock);
98 flow->stats.cpu_stats = alloc_percpu(struct flow_stats);
99 if (!flow->stats.cpu_stats)
102 for_each_possible_cpu(cpu) {
103 struct flow_stats *cpu_stats;
105 cpu_stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
106 spin_lock_init(&cpu_stats->lock);
111 kmem_cache_free(flow_cache, flow);
112 return ERR_PTR(-ENOMEM);
115 int ovs_flow_tbl_count(struct flow_table *table)
120 static struct flex_array *alloc_buckets(unsigned int n_buckets)
122 struct flex_array *buckets;
125 buckets = flex_array_alloc(sizeof(struct hlist_head),
126 n_buckets, GFP_KERNEL);
130 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
132 flex_array_free(buckets);
136 for (i = 0; i < n_buckets; i++)
137 INIT_HLIST_HEAD((struct hlist_head *)
138 flex_array_get(buckets, i));
143 static void flow_free(struct sw_flow *flow)
145 kfree((struct sf_flow_acts __force *)flow->sf_acts);
146 if (flow->stats.is_percpu)
147 free_percpu(flow->stats.cpu_stats);
149 kfree(flow->stats.stat);
150 kmem_cache_free(flow_cache, flow);
153 static void rcu_free_flow_callback(struct rcu_head *rcu)
155 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
160 static void rcu_free_sw_flow_mask_cb(struct rcu_head *rcu)
162 struct sw_flow_mask *mask = container_of(rcu, struct sw_flow_mask, rcu);
167 void ovs_flow_free(struct sw_flow *flow, bool deferred)
175 struct sw_flow_mask *mask = flow->mask;
177 BUG_ON(!mask->ref_count);
180 if (!mask->ref_count) {
181 list_del_rcu(&mask->list);
183 call_rcu(&mask->rcu, rcu_free_sw_flow_mask_cb);
190 call_rcu(&flow->rcu, rcu_free_flow_callback);
195 static void free_buckets(struct flex_array *buckets)
197 flex_array_free(buckets);
201 static void __table_instance_destroy(struct table_instance *ti)
203 free_buckets(ti->buckets);
207 static struct table_instance *table_instance_alloc(int new_size)
209 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
214 ti->buckets = alloc_buckets(new_size);
220 ti->n_buckets = new_size;
222 ti->keep_flows = false;
223 get_random_bytes(&ti->hash_seed, sizeof(u32));
228 int ovs_flow_tbl_init(struct flow_table *table)
230 struct table_instance *ti;
232 ti = table_instance_alloc(TBL_MIN_BUCKETS);
237 rcu_assign_pointer(table->ti, ti);
238 INIT_LIST_HEAD(&table->mask_list);
239 table->last_rehash = jiffies;
244 static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
246 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
248 __table_instance_destroy(ti);
251 static void table_instance_destroy(struct table_instance *ti, bool deferred)
261 for (i = 0; i < ti->n_buckets; i++) {
262 struct sw_flow *flow;
263 struct hlist_head *head = flex_array_get(ti->buckets, i);
264 struct hlist_node *n;
265 int ver = ti->node_ver;
267 hlist_for_each_entry_safe(flow, n, head, hash_node[ver]) {
268 hlist_del_rcu(&flow->hash_node[ver]);
269 ovs_flow_free(flow, deferred);
275 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
277 __table_instance_destroy(ti);
280 void ovs_flow_tbl_destroy(struct flow_table *table, bool deferred)
282 struct table_instance *ti = ovsl_dereference(table->ti);
284 table_instance_destroy(ti, deferred);
287 struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
288 u32 *bucket, u32 *last)
290 struct sw_flow *flow;
291 struct hlist_head *head;
296 while (*bucket < ti->n_buckets) {
298 head = flex_array_get(ti->buckets, *bucket);
299 hlist_for_each_entry_rcu(flow, head, hash_node[ver]) {
314 static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
316 hash = jhash_1word(hash, ti->hash_seed);
317 return flex_array_get(ti->buckets,
318 (hash & (ti->n_buckets - 1)));
321 static void table_instance_insert(struct table_instance *ti, struct sw_flow *flow)
323 struct hlist_head *head;
325 head = find_bucket(ti, flow->hash);
326 hlist_add_head_rcu(&flow->hash_node[ti->node_ver], head);
329 static void flow_table_copy_flows(struct table_instance *old,
330 struct table_instance *new)
335 old_ver = old->node_ver;
336 new->node_ver = !old_ver;
338 /* Insert in new table. */
339 for (i = 0; i < old->n_buckets; i++) {
340 struct sw_flow *flow;
341 struct hlist_head *head;
343 head = flex_array_get(old->buckets, i);
345 hlist_for_each_entry(flow, head, hash_node[old_ver])
346 table_instance_insert(new, flow);
349 old->keep_flows = true;
352 static struct table_instance *table_instance_rehash(struct table_instance *ti,
355 struct table_instance *new_ti;
357 new_ti = table_instance_alloc(n_buckets);
361 flow_table_copy_flows(ti, new_ti);
366 int ovs_flow_tbl_flush(struct flow_table *flow_table)
368 struct table_instance *old_ti;
369 struct table_instance *new_ti;
371 old_ti = ovsl_dereference(flow_table->ti);
372 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
376 rcu_assign_pointer(flow_table->ti, new_ti);
377 flow_table->last_rehash = jiffies;
378 flow_table->count = 0;
380 table_instance_destroy(old_ti, true);
384 static u32 flow_hash(const struct sw_flow_key *key, int key_start,
387 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
388 int hash_u32s = (key_end - key_start) >> 2;
390 /* Make sure number of hash bytes are multiple of u32. */
391 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
393 return arch_fast_hash2(hash_key, hash_u32s, 0);
396 static int flow_key_start(const struct sw_flow_key *key)
398 if (key->tun_key.ipv4_dst)
401 return rounddown(offsetof(struct sw_flow_key, phy),
405 static bool cmp_key(const struct sw_flow_key *key1,
406 const struct sw_flow_key *key2,
407 int key_start, int key_end)
409 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
410 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
414 for (i = key_start; i < key_end; i += sizeof(long))
415 diffs |= *cp1++ ^ *cp2++;
420 static bool flow_cmp_masked_key(const struct sw_flow *flow,
421 const struct sw_flow_key *key,
422 int key_start, int key_end)
424 return cmp_key(&flow->key, key, key_start, key_end);
427 bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
428 struct sw_flow_match *match)
430 struct sw_flow_key *key = match->key;
431 int key_start = flow_key_start(key);
432 int key_end = match->range.end;
434 return cmp_key(&flow->unmasked_key, key, key_start, key_end);
437 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
438 const struct sw_flow_key *unmasked,
439 struct sw_flow_mask *mask)
441 struct sw_flow *flow;
442 struct hlist_head *head;
443 int key_start = mask->range.start;
444 int key_end = mask->range.end;
446 struct sw_flow_key masked_key;
448 ovs_flow_mask_key(&masked_key, unmasked, mask);
449 hash = flow_hash(&masked_key, key_start, key_end);
450 head = find_bucket(ti, hash);
451 hlist_for_each_entry_rcu(flow, head, hash_node[ti->node_ver]) {
452 if (flow->mask == mask && flow->hash == hash &&
453 flow_cmp_masked_key(flow, &masked_key,
460 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
461 const struct sw_flow_key *key,
464 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
465 struct sw_flow_mask *mask;
466 struct sw_flow *flow;
469 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
471 flow = masked_flow_lookup(ti, key, mask);
472 if (flow) /* Found */
478 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
479 const struct sw_flow_key *key)
481 u32 __always_unused n_mask_hit;
483 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
486 int ovs_flow_tbl_num_masks(const struct flow_table *table)
488 struct sw_flow_mask *mask;
491 list_for_each_entry(mask, &table->mask_list, list)
497 static struct table_instance *table_instance_expand(struct table_instance *ti)
499 return table_instance_rehash(ti, ti->n_buckets * 2);
502 void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
504 struct table_instance *ti = ovsl_dereference(table->ti);
506 BUG_ON(table->count == 0);
507 hlist_del_rcu(&flow->hash_node[ti->node_ver]);
511 static struct sw_flow_mask *mask_alloc(void)
513 struct sw_flow_mask *mask;
515 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
522 static bool mask_equal(const struct sw_flow_mask *a,
523 const struct sw_flow_mask *b)
525 const u8 *a_ = (const u8 *)&a->key + a->range.start;
526 const u8 *b_ = (const u8 *)&b->key + b->range.start;
528 return (a->range.end == b->range.end)
529 && (a->range.start == b->range.start)
530 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
533 static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
534 const struct sw_flow_mask *mask)
536 struct list_head *ml;
538 list_for_each(ml, &tbl->mask_list) {
539 struct sw_flow_mask *m;
540 m = container_of(ml, struct sw_flow_mask, list);
541 if (mask_equal(mask, m))
548 /* Add 'mask' into the mask list, if it is not already there. */
549 static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
550 struct sw_flow_mask *new)
552 struct sw_flow_mask *mask;
553 mask = flow_mask_find(tbl, new);
555 /* Allocate a new mask if none exsits. */
559 mask->key = new->key;
560 mask->range = new->range;
561 list_add_rcu(&mask->list, &tbl->mask_list);
563 BUG_ON(!mask->ref_count);
571 int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
572 struct sw_flow_mask *mask)
574 struct table_instance *new_ti = NULL;
575 struct table_instance *ti;
578 err = flow_mask_insert(table, flow, mask);
582 flow->hash = flow_hash(&flow->key, flow->mask->range.start,
583 flow->mask->range.end);
584 ti = ovsl_dereference(table->ti);
585 table_instance_insert(ti, flow);
588 /* Expand table, if necessary, to make room. */
589 if (table->count > ti->n_buckets)
590 new_ti = table_instance_expand(ti);
591 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
592 new_ti = table_instance_rehash(ti, ti->n_buckets);
595 rcu_assign_pointer(table->ti, new_ti);
596 table_instance_destroy(ti, true);
597 table->last_rehash = jiffies;
602 /* Initializes the flow module.
603 * Returns zero if successful or a negative error code. */
604 int ovs_flow_init(void)
606 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
607 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
609 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
611 if (flow_cache == NULL)
617 /* Uninitializes the flow module. */
618 void ovs_flow_exit(void)
620 kmem_cache_destroy(flow_cache);