blob: 4613df8c82900e32a4e0188688c9e0354022484d [file] [log] [blame]
Pravin B Shelare6445712013-10-03 18:16:47 -07001/*
Pravin B Shelar9b996e52014-05-06 18:41:20 -07002 * Copyright (c) 2007-2014 Nicira, Inc.
Pravin B Shelare6445712013-10-03 18:16:47 -07003 *
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>
Daniel Borkmann87545892014-12-10 16:33:11 +010028#include <linux/jhash.h>
Pravin B Shelare6445712013-10-03 18:16:47 -070029#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/if_arp.h>
35#include <linux/ip.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>
43#include <net/ip.h>
44#include <net/ipv6.h>
45#include <net/ndisc.h>
46
Pravin B Shelarb637e492013-10-04 00:14:23 -070047#define TBL_MIN_BUCKETS 1024
48#define REHASH_INTERVAL (10 * 60 * HZ)
49
Pravin B Shelare6445712013-10-03 18:16:47 -070050static struct kmem_cache *flow_cache;
Jarno Rajahalme63e79592014-03-27 12:42:54 -070051struct kmem_cache *flow_stats_cache __read_mostly;
Pravin B Shelare6445712013-10-03 18:16:47 -070052
53static u16 range_n_bytes(const struct sw_flow_key_range *range)
54{
55 return range->end - range->start;
56}
57
58void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
59 const struct sw_flow_mask *mask)
60{
Daniele Di Proietto70851302014-01-23 10:56:49 -080061 const long *m = (const long *)((const u8 *)&mask->key +
62 mask->range.start);
63 const long *s = (const long *)((const u8 *)src +
64 mask->range.start);
Pravin B Shelare6445712013-10-03 18:16:47 -070065 long *d = (long *)((u8 *)dst + mask->range.start);
66 int i;
67
68 /* The memory outside of the 'mask->range' are not set since
69 * further operations on 'dst' only uses contents within
70 * 'mask->range'.
71 */
72 for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long))
73 *d++ = *s++ & *m++;
74}
75
Jarno Rajahalme23dabf82014-03-27 12:35:23 -070076struct sw_flow *ovs_flow_alloc(void)
Pravin B Shelare6445712013-10-03 18:16:47 -070077{
78 struct sw_flow *flow;
Jarno Rajahalme63e79592014-03-27 12:42:54 -070079 struct flow_stats *stats;
80 int node;
Pravin B Shelare6445712013-10-03 18:16:47 -070081
82 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
83 if (!flow)
84 return ERR_PTR(-ENOMEM);
85
Pravin B Shelare6445712013-10-03 18:16:47 -070086 flow->sf_acts = NULL;
87 flow->mask = NULL;
Pravin B Shelarca539342015-02-06 11:17:13 -080088 flow->id.unmasked_key = NULL;
89 flow->id.ufid_len = 0;
Jarno Rajahalme63e79592014-03-27 12:42:54 -070090 flow->stats_last_writer = NUMA_NO_NODE;
Pravin B Shelare6445712013-10-03 18:16:47 -070091
Jarno Rajahalme63e79592014-03-27 12:42:54 -070092 /* Initialize the default stat node. */
93 stats = kmem_cache_alloc_node(flow_stats_cache,
94 GFP_KERNEL | __GFP_ZERO, 0);
95 if (!stats)
Jarno Rajahalme23dabf82014-03-27 12:35:23 -070096 goto err;
Pravin B Shelare298e502013-10-29 17:22:21 -070097
Jarno Rajahalme63e79592014-03-27 12:42:54 -070098 spin_lock_init(&stats->lock);
Pravin B Shelare298e502013-10-29 17:22:21 -070099
Jarno Rajahalme63e79592014-03-27 12:42:54 -0700100 RCU_INIT_POINTER(flow->stats[0], stats);
101
102 for_each_node(node)
103 if (node != 0)
104 RCU_INIT_POINTER(flow->stats[node], NULL);
105
Pravin B Shelare6445712013-10-03 18:16:47 -0700106 return flow;
Pravin B Shelare298e502013-10-29 17:22:21 -0700107err:
Wei Yongjunece37c82014-01-08 18:13:14 +0800108 kmem_cache_free(flow_cache, flow);
Pravin B Shelare298e502013-10-29 17:22:21 -0700109 return ERR_PTR(-ENOMEM);
Pravin B Shelare6445712013-10-03 18:16:47 -0700110}
111
Thomas Graf12eb18f2014-11-06 06:58:52 -0800112int ovs_flow_tbl_count(const struct flow_table *table)
Pravin B Shelarb637e492013-10-04 00:14:23 -0700113{
114 return table->count;
115}
116
Pravin B Shelare6445712013-10-03 18:16:47 -0700117static struct flex_array *alloc_buckets(unsigned int n_buckets)
118{
119 struct flex_array *buckets;
120 int i, err;
121
122 buckets = flex_array_alloc(sizeof(struct hlist_head),
123 n_buckets, GFP_KERNEL);
124 if (!buckets)
125 return NULL;
126
127 err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL);
128 if (err) {
129 flex_array_free(buckets);
130 return NULL;
131 }
132
133 for (i = 0; i < n_buckets; i++)
134 INIT_HLIST_HEAD((struct hlist_head *)
135 flex_array_get(buckets, i));
136
137 return buckets;
138}
139
140static void flow_free(struct sw_flow *flow)
141{
Jarno Rajahalme63e79592014-03-27 12:42:54 -0700142 int node;
143
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800144 if (ovs_identifier_is_key(&flow->id))
145 kfree(flow->id.unmasked_key);
Jarno Rajahalmeeb072652014-05-05 14:15:18 -0700146 kfree((struct sw_flow_actions __force *)flow->sf_acts);
Jarno Rajahalme63e79592014-03-27 12:42:54 -0700147 for_each_node(node)
148 if (flow->stats[node])
149 kmem_cache_free(flow_stats_cache,
150 (struct flow_stats __force *)flow->stats[node]);
Pravin B Shelare6445712013-10-03 18:16:47 -0700151 kmem_cache_free(flow_cache, flow);
152}
153
154static void rcu_free_flow_callback(struct rcu_head *rcu)
155{
156 struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu);
157
158 flow_free(flow);
159}
160
161void ovs_flow_free(struct sw_flow *flow, bool deferred)
162{
163 if (!flow)
164 return;
165
Pravin B Shelare6445712013-10-03 18:16:47 -0700166 if (deferred)
167 call_rcu(&flow->rcu, rcu_free_flow_callback);
168 else
169 flow_free(flow);
170}
171
172static void free_buckets(struct flex_array *buckets)
173{
174 flex_array_free(buckets);
175}
176
Andy Zhoue80857c2014-01-21 09:31:04 -0800177
Pravin B Shelarb637e492013-10-04 00:14:23 -0700178static void __table_instance_destroy(struct table_instance *ti)
Pravin B Shelare6445712013-10-03 18:16:47 -0700179{
Pravin B Shelarb637e492013-10-04 00:14:23 -0700180 free_buckets(ti->buckets);
181 kfree(ti);
Pravin B Shelare6445712013-10-03 18:16:47 -0700182}
183
Pravin B Shelarb637e492013-10-04 00:14:23 -0700184static struct table_instance *table_instance_alloc(int new_size)
Pravin B Shelare6445712013-10-03 18:16:47 -0700185{
Pravin B Shelarb637e492013-10-04 00:14:23 -0700186 struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL);
Pravin B Shelare6445712013-10-03 18:16:47 -0700187
Pravin B Shelarb637e492013-10-04 00:14:23 -0700188 if (!ti)
Pravin B Shelare6445712013-10-03 18:16:47 -0700189 return NULL;
190
Pravin B Shelarb637e492013-10-04 00:14:23 -0700191 ti->buckets = alloc_buckets(new_size);
Pravin B Shelare6445712013-10-03 18:16:47 -0700192
Pravin B Shelarb637e492013-10-04 00:14:23 -0700193 if (!ti->buckets) {
194 kfree(ti);
Pravin B Shelare6445712013-10-03 18:16:47 -0700195 return NULL;
196 }
Pravin B Shelarb637e492013-10-04 00:14:23 -0700197 ti->n_buckets = new_size;
198 ti->node_ver = 0;
199 ti->keep_flows = false;
200 get_random_bytes(&ti->hash_seed, sizeof(u32));
201
202 return ti;
203}
204
205int ovs_flow_tbl_init(struct flow_table *table)
206{
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800207 struct table_instance *ti, *ufid_ti;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700208
209 ti = table_instance_alloc(TBL_MIN_BUCKETS);
210
211 if (!ti)
212 return -ENOMEM;
213
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800214 ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
215 if (!ufid_ti)
216 goto free_ti;
217
Pravin B Shelarb637e492013-10-04 00:14:23 -0700218 rcu_assign_pointer(table->ti, ti);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800219 rcu_assign_pointer(table->ufid_ti, ufid_ti);
Pravin B Shelarb637e492013-10-04 00:14:23 -0700220 INIT_LIST_HEAD(&table->mask_list);
221 table->last_rehash = jiffies;
Pravin B Shelare6445712013-10-03 18:16:47 -0700222 table->count = 0;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800223 table->ufid_count = 0;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700224 return 0;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800225
226free_ti:
227 __table_instance_destroy(ti);
228 return -ENOMEM;
Pravin B Shelare6445712013-10-03 18:16:47 -0700229}
230
231static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu)
232{
Pravin B Shelarb637e492013-10-04 00:14:23 -0700233 struct table_instance *ti = container_of(rcu, struct table_instance, rcu);
Pravin B Shelare6445712013-10-03 18:16:47 -0700234
Pravin B Shelarb637e492013-10-04 00:14:23 -0700235 __table_instance_destroy(ti);
236}
237
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800238static void table_instance_destroy(struct table_instance *ti,
239 struct table_instance *ufid_ti,
240 bool deferred)
Pravin B Shelarb637e492013-10-04 00:14:23 -0700241{
Andy Zhoue80857c2014-01-21 09:31:04 -0800242 int i;
243
Pravin B Shelarb637e492013-10-04 00:14:23 -0700244 if (!ti)
245 return;
246
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800247 BUG_ON(!ufid_ti);
Andy Zhoue80857c2014-01-21 09:31:04 -0800248 if (ti->keep_flows)
249 goto skip_flows;
250
251 for (i = 0; i < ti->n_buckets; i++) {
252 struct sw_flow *flow;
253 struct hlist_head *head = flex_array_get(ti->buckets, i);
254 struct hlist_node *n;
255 int ver = ti->node_ver;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800256 int ufid_ver = ufid_ti->node_ver;
Andy Zhoue80857c2014-01-21 09:31:04 -0800257
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800258 hlist_for_each_entry_safe(flow, n, head, flow_table.node[ver]) {
259 hlist_del_rcu(&flow->flow_table.node[ver]);
260 if (ovs_identifier_is_ufid(&flow->id))
261 hlist_del_rcu(&flow->ufid_table.node[ufid_ver]);
Andy Zhoue80857c2014-01-21 09:31:04 -0800262 ovs_flow_free(flow, deferred);
263 }
264 }
265
266skip_flows:
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800267 if (deferred) {
Pravin B Shelarb637e492013-10-04 00:14:23 -0700268 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800269 call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb);
270 } else {
Pravin B Shelarb637e492013-10-04 00:14:23 -0700271 __table_instance_destroy(ti);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800272 __table_instance_destroy(ufid_ti);
273 }
Pravin B Shelare6445712013-10-03 18:16:47 -0700274}
275
Pravin B Shelar9b996e52014-05-06 18:41:20 -0700276/* No need for locking this function is called from RCU callback or
277 * error path.
278 */
279void ovs_flow_tbl_destroy(struct flow_table *table)
Pravin B Shelare6445712013-10-03 18:16:47 -0700280{
Pravin B Shelar9b996e52014-05-06 18:41:20 -0700281 struct table_instance *ti = rcu_dereference_raw(table->ti);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800282 struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti);
Pravin B Shelare6445712013-10-03 18:16:47 -0700283
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800284 table_instance_destroy(ti, ufid_ti, false);
Pravin B Shelare6445712013-10-03 18:16:47 -0700285}
286
Pravin B Shelarb637e492013-10-04 00:14:23 -0700287struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti,
Pravin B Shelare6445712013-10-03 18:16:47 -0700288 u32 *bucket, u32 *last)
289{
290 struct sw_flow *flow;
291 struct hlist_head *head;
292 int ver;
293 int i;
294
Pravin B Shelarb637e492013-10-04 00:14:23 -0700295 ver = ti->node_ver;
296 while (*bucket < ti->n_buckets) {
Pravin B Shelare6445712013-10-03 18:16:47 -0700297 i = 0;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700298 head = flex_array_get(ti->buckets, *bucket);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800299 hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) {
Pravin B Shelare6445712013-10-03 18:16:47 -0700300 if (i < *last) {
301 i++;
302 continue;
303 }
304 *last = i + 1;
305 return flow;
306 }
307 (*bucket)++;
308 *last = 0;
309 }
310
311 return NULL;
312}
313
Pravin B Shelarb637e492013-10-04 00:14:23 -0700314static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash)
Pravin B Shelare6445712013-10-03 18:16:47 -0700315{
Pravin B Shelarb637e492013-10-04 00:14:23 -0700316 hash = jhash_1word(hash, ti->hash_seed);
317 return flex_array_get(ti->buckets,
318 (hash & (ti->n_buckets - 1)));
Pravin B Shelare6445712013-10-03 18:16:47 -0700319}
320
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800321static void table_instance_insert(struct table_instance *ti,
322 struct sw_flow *flow)
Pravin B Shelare6445712013-10-03 18:16:47 -0700323{
324 struct hlist_head *head;
325
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800326 head = find_bucket(ti, flow->flow_table.hash);
327 hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head);
328}
329
330static void ufid_table_instance_insert(struct table_instance *ti,
331 struct sw_flow *flow)
332{
333 struct hlist_head *head;
334
335 head = find_bucket(ti, flow->ufid_table.hash);
336 hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head);
Pravin B Shelare6445712013-10-03 18:16:47 -0700337}
338
Pravin B Shelarb637e492013-10-04 00:14:23 -0700339static void flow_table_copy_flows(struct table_instance *old,
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800340 struct table_instance *new, bool ufid)
Pravin B Shelare6445712013-10-03 18:16:47 -0700341{
342 int old_ver;
343 int i;
344
345 old_ver = old->node_ver;
346 new->node_ver = !old_ver;
347
348 /* Insert in new table. */
349 for (i = 0; i < old->n_buckets; i++) {
350 struct sw_flow *flow;
351 struct hlist_head *head;
352
353 head = flex_array_get(old->buckets, i);
354
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800355 if (ufid)
356 hlist_for_each_entry(flow, head,
357 ufid_table.node[old_ver])
358 ufid_table_instance_insert(new, flow);
359 else
360 hlist_for_each_entry(flow, head,
361 flow_table.node[old_ver])
362 table_instance_insert(new, flow);
Pravin B Shelare6445712013-10-03 18:16:47 -0700363 }
364
Pravin B Shelare6445712013-10-03 18:16:47 -0700365 old->keep_flows = true;
366}
367
Pravin B Shelarb637e492013-10-04 00:14:23 -0700368static struct table_instance *table_instance_rehash(struct table_instance *ti,
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800369 int n_buckets, bool ufid)
Pravin B Shelare6445712013-10-03 18:16:47 -0700370{
Pravin B Shelarb637e492013-10-04 00:14:23 -0700371 struct table_instance *new_ti;
Pravin B Shelare6445712013-10-03 18:16:47 -0700372
Pravin B Shelarb637e492013-10-04 00:14:23 -0700373 new_ti = table_instance_alloc(n_buckets);
374 if (!new_ti)
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700375 return NULL;
Pravin B Shelare6445712013-10-03 18:16:47 -0700376
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800377 flow_table_copy_flows(ti, new_ti, ufid);
Pravin B Shelare6445712013-10-03 18:16:47 -0700378
Pravin B Shelarb637e492013-10-04 00:14:23 -0700379 return new_ti;
Pravin B Shelare6445712013-10-03 18:16:47 -0700380}
381
Pravin B Shelarb637e492013-10-04 00:14:23 -0700382int ovs_flow_tbl_flush(struct flow_table *flow_table)
Pravin B Shelare6445712013-10-03 18:16:47 -0700383{
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800384 struct table_instance *old_ti, *new_ti;
385 struct table_instance *old_ufid_ti, *new_ufid_ti;
Pravin B Shelare6445712013-10-03 18:16:47 -0700386
Pravin B Shelarb637e492013-10-04 00:14:23 -0700387 new_ti = table_instance_alloc(TBL_MIN_BUCKETS);
388 if (!new_ti)
389 return -ENOMEM;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800390 new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
391 if (!new_ufid_ti)
392 goto err_free_ti;
393
394 old_ti = ovsl_dereference(flow_table->ti);
395 old_ufid_ti = ovsl_dereference(flow_table->ufid_ti);
Pravin B Shelarb637e492013-10-04 00:14:23 -0700396
397 rcu_assign_pointer(flow_table->ti, new_ti);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800398 rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti);
Pravin B Shelarb637e492013-10-04 00:14:23 -0700399 flow_table->last_rehash = jiffies;
400 flow_table->count = 0;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800401 flow_table->ufid_count = 0;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700402
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800403 table_instance_destroy(old_ti, old_ufid_ti, true);
Pravin B Shelarb637e492013-10-04 00:14:23 -0700404 return 0;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800405
406err_free_ti:
407 __table_instance_destroy(new_ti);
408 return -ENOMEM;
Pravin B Shelare6445712013-10-03 18:16:47 -0700409}
410
Joe Stringer272c2cf2015-01-21 16:42:50 -0800411static u32 flow_hash(const struct sw_flow_key *key,
412 const struct sw_flow_key_range *range)
Pravin B Shelare6445712013-10-03 18:16:47 -0700413{
Joe Stringer272c2cf2015-01-21 16:42:50 -0800414 int key_start = range->start;
415 int key_end = range->end;
Daniele Di Proietto70851302014-01-23 10:56:49 -0800416 const u32 *hash_key = (const u32 *)((const u8 *)key + key_start);
Pravin B Shelare6445712013-10-03 18:16:47 -0700417 int hash_u32s = (key_end - key_start) >> 2;
418
419 /* Make sure number of hash bytes are multiple of u32. */
420 BUILD_BUG_ON(sizeof(long) % sizeof(u32));
421
Daniel Borkmann87545892014-12-10 16:33:11 +0100422 return jhash2(hash_key, hash_u32s, 0);
Pravin B Shelare6445712013-10-03 18:16:47 -0700423}
424
425static int flow_key_start(const struct sw_flow_key *key)
426{
427 if (key->tun_key.ipv4_dst)
428 return 0;
429 else
430 return rounddown(offsetof(struct sw_flow_key, phy),
431 sizeof(long));
432}
433
434static bool cmp_key(const struct sw_flow_key *key1,
435 const struct sw_flow_key *key2,
436 int key_start, int key_end)
437{
Daniele Di Proietto70851302014-01-23 10:56:49 -0800438 const long *cp1 = (const long *)((const u8 *)key1 + key_start);
439 const long *cp2 = (const long *)((const u8 *)key2 + key_start);
Pravin B Shelare6445712013-10-03 18:16:47 -0700440 long diffs = 0;
441 int i;
442
443 for (i = key_start; i < key_end; i += sizeof(long))
444 diffs |= *cp1++ ^ *cp2++;
445
446 return diffs == 0;
447}
448
449static bool flow_cmp_masked_key(const struct sw_flow *flow,
450 const struct sw_flow_key *key,
Joe Stringer272c2cf2015-01-21 16:42:50 -0800451 const struct sw_flow_key_range *range)
Pravin B Shelare6445712013-10-03 18:16:47 -0700452{
Joe Stringer272c2cf2015-01-21 16:42:50 -0800453 return cmp_key(&flow->key, key, range->start, range->end);
Pravin B Shelare6445712013-10-03 18:16:47 -0700454}
455
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800456static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
457 const struct sw_flow_match *match)
Pravin B Shelare6445712013-10-03 18:16:47 -0700458{
459 struct sw_flow_key *key = match->key;
460 int key_start = flow_key_start(key);
461 int key_end = match->range.end;
462
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800463 BUG_ON(ovs_identifier_is_ufid(&flow->id));
464 return cmp_key(flow->id.unmasked_key, key, key_start, key_end);
Pravin B Shelare6445712013-10-03 18:16:47 -0700465}
466
Pravin B Shelarb637e492013-10-04 00:14:23 -0700467static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
Pravin B Shelare6445712013-10-03 18:16:47 -0700468 const struct sw_flow_key *unmasked,
Thomas Graf12eb18f2014-11-06 06:58:52 -0800469 const struct sw_flow_mask *mask)
Pravin B Shelare6445712013-10-03 18:16:47 -0700470{
471 struct sw_flow *flow;
472 struct hlist_head *head;
Pravin B Shelare6445712013-10-03 18:16:47 -0700473 u32 hash;
474 struct sw_flow_key masked_key;
475
476 ovs_flow_mask_key(&masked_key, unmasked, mask);
Joe Stringer272c2cf2015-01-21 16:42:50 -0800477 hash = flow_hash(&masked_key, &mask->range);
Pravin B Shelarb637e492013-10-04 00:14:23 -0700478 head = find_bucket(ti, hash);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800479 hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) {
480 if (flow->mask == mask && flow->flow_table.hash == hash &&
Joe Stringer272c2cf2015-01-21 16:42:50 -0800481 flow_cmp_masked_key(flow, &masked_key, &mask->range))
Pravin B Shelare6445712013-10-03 18:16:47 -0700482 return flow;
483 }
484 return NULL;
485}
486
Andy Zhou5bb50632013-11-25 10:42:46 -0800487struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
Andy Zhou1bd71162013-10-22 10:42:46 -0700488 const struct sw_flow_key *key,
489 u32 *n_mask_hit)
Pravin B Shelare6445712013-10-03 18:16:47 -0700490{
Jesse Gross663efa32013-12-03 10:58:53 -0800491 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
Pravin B Shelare6445712013-10-03 18:16:47 -0700492 struct sw_flow_mask *mask;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700493 struct sw_flow *flow;
Pravin B Shelare6445712013-10-03 18:16:47 -0700494
Andy Zhou1bd71162013-10-22 10:42:46 -0700495 *n_mask_hit = 0;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700496 list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
Andy Zhou1bd71162013-10-22 10:42:46 -0700497 (*n_mask_hit)++;
Pravin B Shelarb637e492013-10-04 00:14:23 -0700498 flow = masked_flow_lookup(ti, key, mask);
Pravin B Shelare6445712013-10-03 18:16:47 -0700499 if (flow) /* Found */
Pravin B Shelarb637e492013-10-04 00:14:23 -0700500 return flow;
Pravin B Shelare6445712013-10-03 18:16:47 -0700501 }
Pravin B Shelarb637e492013-10-04 00:14:23 -0700502 return NULL;
503}
Pravin B Shelare6445712013-10-03 18:16:47 -0700504
Andy Zhou5bb50632013-11-25 10:42:46 -0800505struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
506 const struct sw_flow_key *key)
507{
508 u32 __always_unused n_mask_hit;
509
510 return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
511}
512
Alex Wang4a46b242014-06-30 20:30:29 -0700513struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
Thomas Graf12eb18f2014-11-06 06:58:52 -0800514 const struct sw_flow_match *match)
Alex Wang4a46b242014-06-30 20:30:29 -0700515{
516 struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
517 struct sw_flow_mask *mask;
518 struct sw_flow *flow;
519
520 /* Always called under ovs-mutex. */
521 list_for_each_entry(mask, &tbl->mask_list, list) {
522 flow = masked_flow_lookup(ti, match->key, mask);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800523 if (flow && ovs_identifier_is_key(&flow->id) &&
524 ovs_flow_cmp_unmasked_key(flow, match))
525 return flow;
526 }
527 return NULL;
528}
529
530static u32 ufid_hash(const struct sw_flow_id *sfid)
531{
532 return jhash(sfid->ufid, sfid->ufid_len, 0);
533}
534
535static bool ovs_flow_cmp_ufid(const struct sw_flow *flow,
536 const struct sw_flow_id *sfid)
537{
538 if (flow->id.ufid_len != sfid->ufid_len)
539 return false;
540
541 return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len);
542}
543
544bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match)
545{
546 if (ovs_identifier_is_ufid(&flow->id))
547 return flow_cmp_masked_key(flow, match->key, &match->range);
548
549 return ovs_flow_cmp_unmasked_key(flow, match);
550}
551
552struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl,
553 const struct sw_flow_id *ufid)
554{
555 struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti);
556 struct sw_flow *flow;
557 struct hlist_head *head;
558 u32 hash;
559
560 hash = ufid_hash(ufid);
561 head = find_bucket(ti, hash);
562 hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) {
563 if (flow->ufid_table.hash == hash &&
564 ovs_flow_cmp_ufid(flow, ufid))
Alex Wang4a46b242014-06-30 20:30:29 -0700565 return flow;
566 }
567 return NULL;
568}
569
Andy Zhou1bd71162013-10-22 10:42:46 -0700570int ovs_flow_tbl_num_masks(const struct flow_table *table)
571{
572 struct sw_flow_mask *mask;
573 int num = 0;
574
575 list_for_each_entry(mask, &table->mask_list, list)
576 num++;
577
578 return num;
579}
580
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800581static struct table_instance *table_instance_expand(struct table_instance *ti,
582 bool ufid)
Pravin B Shelarb637e492013-10-04 00:14:23 -0700583{
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800584 return table_instance_rehash(ti, ti->n_buckets * 2, ufid);
Pravin B Shelare6445712013-10-03 18:16:47 -0700585}
586
Jarno Rajahalme56c19862014-05-05 13:24:53 -0700587/* Remove 'mask' from the mask list, if it is not needed any more. */
588static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask)
589{
590 if (mask) {
591 /* ovs-lock is required to protect mask-refcount and
592 * mask list.
593 */
594 ASSERT_OVSL();
595 BUG_ON(!mask->ref_count);
596 mask->ref_count--;
597
598 if (!mask->ref_count) {
599 list_del_rcu(&mask->list);
600 kfree_rcu(mask, rcu);
601 }
602 }
603}
604
605/* Must be called with OVS mutex held. */
Pravin B Shelare6445712013-10-03 18:16:47 -0700606void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow)
607{
Pravin B Shelarb637e492013-10-04 00:14:23 -0700608 struct table_instance *ti = ovsl_dereference(table->ti);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800609 struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti);
Pravin B Shelarb637e492013-10-04 00:14:23 -0700610
Pravin B Shelare6445712013-10-03 18:16:47 -0700611 BUG_ON(table->count == 0);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800612 hlist_del_rcu(&flow->flow_table.node[ti->node_ver]);
Pravin B Shelare6445712013-10-03 18:16:47 -0700613 table->count--;
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800614 if (ovs_identifier_is_ufid(&flow->id)) {
615 hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]);
616 table->ufid_count--;
617 }
Jarno Rajahalme56c19862014-05-05 13:24:53 -0700618
619 /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
620 * accessible as long as the RCU read lock is held.
621 */
622 flow_mask_remove(table, flow->mask);
Pravin B Shelare6445712013-10-03 18:16:47 -0700623}
624
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700625static struct sw_flow_mask *mask_alloc(void)
Pravin B Shelare6445712013-10-03 18:16:47 -0700626{
627 struct sw_flow_mask *mask;
628
629 mask = kmalloc(sizeof(*mask), GFP_KERNEL);
630 if (mask)
Andy Zhoue80857c2014-01-21 09:31:04 -0800631 mask->ref_count = 1;
Pravin B Shelare6445712013-10-03 18:16:47 -0700632
633 return mask;
634}
635
Pravin B Shelare6445712013-10-03 18:16:47 -0700636static bool mask_equal(const struct sw_flow_mask *a,
637 const struct sw_flow_mask *b)
638{
Daniele Di Proietto70851302014-01-23 10:56:49 -0800639 const u8 *a_ = (const u8 *)&a->key + a->range.start;
640 const u8 *b_ = (const u8 *)&b->key + b->range.start;
Pravin B Shelare6445712013-10-03 18:16:47 -0700641
642 return (a->range.end == b->range.end)
643 && (a->range.start == b->range.start)
644 && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0);
645}
646
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700647static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl,
Pravin B Shelare6445712013-10-03 18:16:47 -0700648 const struct sw_flow_mask *mask)
649{
650 struct list_head *ml;
651
Pravin B Shelarb637e492013-10-04 00:14:23 -0700652 list_for_each(ml, &tbl->mask_list) {
Pravin B Shelare6445712013-10-03 18:16:47 -0700653 struct sw_flow_mask *m;
654 m = container_of(ml, struct sw_flow_mask, list);
655 if (mask_equal(mask, m))
656 return m;
657 }
658
659 return NULL;
660}
661
Ben Pfaffd1211902013-11-25 10:40:51 -0800662/* Add 'mask' into the mask list, if it is not already there. */
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700663static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
Thomas Graf12eb18f2014-11-06 06:58:52 -0800664 const struct sw_flow_mask *new)
Pravin B Shelare6445712013-10-03 18:16:47 -0700665{
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700666 struct sw_flow_mask *mask;
667 mask = flow_mask_find(tbl, new);
668 if (!mask) {
669 /* Allocate a new mask if none exsits. */
670 mask = mask_alloc();
671 if (!mask)
672 return -ENOMEM;
673 mask->key = new->key;
674 mask->range = new->range;
675 list_add_rcu(&mask->list, &tbl->mask_list);
Andy Zhoue80857c2014-01-21 09:31:04 -0800676 } else {
677 BUG_ON(!mask->ref_count);
678 mask->ref_count++;
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700679 }
680
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700681 flow->mask = mask;
682 return 0;
683}
684
Jarno Rajahalme56c19862014-05-05 13:24:53 -0700685/* Must be called with OVS mutex held. */
Joe Stringerd29ab6f2015-01-21 16:42:49 -0800686static void flow_key_insert(struct flow_table *table, struct sw_flow *flow)
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700687{
688 struct table_instance *new_ti = NULL;
689 struct table_instance *ti;
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700690
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800691 flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range);
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700692 ti = ovsl_dereference(table->ti);
693 table_instance_insert(ti, flow);
694 table->count++;
695
696 /* Expand table, if necessary, to make room. */
697 if (table->count > ti->n_buckets)
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800698 new_ti = table_instance_expand(ti, false);
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700699 else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL))
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800700 new_ti = table_instance_rehash(ti, ti->n_buckets, false);
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700701
702 if (new_ti) {
703 rcu_assign_pointer(table->ti, new_ti);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800704 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700705 table->last_rehash = jiffies;
706 }
Joe Stringerd29ab6f2015-01-21 16:42:49 -0800707}
708
709/* Must be called with OVS mutex held. */
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800710static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow)
711{
712 struct table_instance *ti;
713
714 flow->ufid_table.hash = ufid_hash(&flow->id);
715 ti = ovsl_dereference(table->ufid_ti);
716 ufid_table_instance_insert(ti, flow);
717 table->ufid_count++;
718
719 /* Expand table, if necessary, to make room. */
720 if (table->ufid_count > ti->n_buckets) {
721 struct table_instance *new_ti;
722
723 new_ti = table_instance_expand(ti, true);
724 if (new_ti) {
725 rcu_assign_pointer(table->ufid_ti, new_ti);
726 call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb);
727 }
728 }
729}
730
731/* Must be called with OVS mutex held. */
Joe Stringerd29ab6f2015-01-21 16:42:49 -0800732int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow,
733 const struct sw_flow_mask *mask)
734{
735 int err;
736
737 err = flow_mask_insert(table, flow, mask);
738 if (err)
739 return err;
740 flow_key_insert(table, flow);
Joe Stringer74ed7ab2015-01-21 16:42:52 -0800741 if (ovs_identifier_is_ufid(&flow->id))
742 flow_ufid_insert(table, flow);
Joe Stringerd29ab6f2015-01-21 16:42:49 -0800743
Pravin B Shelar618ed0c2013-10-04 00:17:42 -0700744 return 0;
Pravin B Shelare6445712013-10-03 18:16:47 -0700745}
746
747/* Initializes the flow module.
748 * Returns zero if successful or a negative error code. */
749int ovs_flow_init(void)
750{
751 BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
752 BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
753
Jarno Rajahalme63e79592014-03-27 12:42:54 -0700754 flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
755 + (num_possible_nodes()
756 * sizeof(struct flow_stats *)),
757 0, 0, NULL);
Pravin B Shelare6445712013-10-03 18:16:47 -0700758 if (flow_cache == NULL)
759 return -ENOMEM;
760
Jarno Rajahalme63e79592014-03-27 12:42:54 -0700761 flow_stats_cache
762 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
763 0, SLAB_HWCACHE_ALIGN, NULL);
764 if (flow_stats_cache == NULL) {
765 kmem_cache_destroy(flow_cache);
766 flow_cache = NULL;
767 return -ENOMEM;
768 }
769
Pravin B Shelare6445712013-10-03 18:16:47 -0700770 return 0;
771}
772
773/* Uninitializes the flow module. */
774void ovs_flow_exit(void)
775{
Jarno Rajahalme63e79592014-03-27 12:42:54 -0700776 kmem_cache_destroy(flow_stats_cache);
Pravin B Shelare6445712013-10-03 18:16:47 -0700777 kmem_cache_destroy(flow_cache);
778}