Commit | Line | Data |
---|---|---|
e6445719 | 1 | /* |
9b996e54 | 2 | * Copyright (c) 2007-2014 Nicira, Inc. |
e6445719 PS |
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" | |
34ae932a | 21 | #include "flow_netlink.h" |
e6445719 PS |
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> | |
87545899 | 29 | #include <linux/jhash.h> |
e6445719 PS |
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/if_arp.h> | |
36 | #include <linux/ip.h> | |
37 | #include <linux/ipv6.h> | |
38 | #include <linux/sctp.h> | |
39 | #include <linux/tcp.h> | |
40 | #include <linux/udp.h> | |
41 | #include <linux/icmp.h> | |
42 | #include <linux/icmpv6.h> | |
43 | #include <linux/rculist.h> | |
44 | #include <net/ip.h> | |
45 | #include <net/ipv6.h> | |
46 | #include <net/ndisc.h> | |
47 | ||
b637e498 PS |
48 | #define TBL_MIN_BUCKETS 1024 |
49 | #define REHASH_INTERVAL (10 * 60 * HZ) | |
50 | ||
e6445719 | 51 | static struct kmem_cache *flow_cache; |
63e7959c | 52 | struct kmem_cache *flow_stats_cache __read_mostly; |
e6445719 PS |
53 | |
54 | static u16 range_n_bytes(const struct sw_flow_key_range *range) | |
55 | { | |
56 | return range->end - range->start; | |
57 | } | |
58 | ||
59 | void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, | |
60 | const struct sw_flow_mask *mask) | |
61 | { | |
7085130b DDP |
62 | const long *m = (const long *)((const u8 *)&mask->key + |
63 | mask->range.start); | |
64 | const long *s = (const long *)((const u8 *)src + | |
65 | mask->range.start); | |
e6445719 PS |
66 | long *d = (long *)((u8 *)dst + mask->range.start); |
67 | int i; | |
68 | ||
69 | /* The memory outside of the 'mask->range' are not set since | |
70 | * further operations on 'dst' only uses contents within | |
71 | * 'mask->range'. | |
72 | */ | |
73 | for (i = 0; i < range_n_bytes(&mask->range); i += sizeof(long)) | |
74 | *d++ = *s++ & *m++; | |
75 | } | |
76 | ||
23dabf88 | 77 | struct sw_flow *ovs_flow_alloc(void) |
e6445719 PS |
78 | { |
79 | struct sw_flow *flow; | |
63e7959c JR |
80 | struct flow_stats *stats; |
81 | int node; | |
e6445719 PS |
82 | |
83 | flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); | |
84 | if (!flow) | |
85 | return ERR_PTR(-ENOMEM); | |
86 | ||
e6445719 PS |
87 | flow->sf_acts = NULL; |
88 | flow->mask = NULL; | |
ca539345 PS |
89 | flow->id.unmasked_key = NULL; |
90 | flow->id.ufid_len = 0; | |
63e7959c | 91 | flow->stats_last_writer = NUMA_NO_NODE; |
e6445719 | 92 | |
63e7959c JR |
93 | /* Initialize the default stat node. */ |
94 | stats = kmem_cache_alloc_node(flow_stats_cache, | |
95 | GFP_KERNEL | __GFP_ZERO, 0); | |
96 | if (!stats) | |
23dabf88 | 97 | goto err; |
e298e505 | 98 | |
63e7959c JR |
99 | spin_lock_init(&stats->lock); |
100 | ||
101 | RCU_INIT_POINTER(flow->stats[0], stats); | |
102 | ||
103 | for_each_node(node) | |
104 | if (node != 0) | |
105 | RCU_INIT_POINTER(flow->stats[node], NULL); | |
e298e505 | 106 | |
e6445719 | 107 | return flow; |
e298e505 | 108 | err: |
ece37c87 | 109 | kmem_cache_free(flow_cache, flow); |
e298e505 | 110 | return ERR_PTR(-ENOMEM); |
e6445719 PS |
111 | } |
112 | ||
12eb18f7 | 113 | int ovs_flow_tbl_count(const struct flow_table *table) |
b637e498 PS |
114 | { |
115 | return table->count; | |
116 | } | |
117 | ||
e6445719 PS |
118 | static struct flex_array *alloc_buckets(unsigned int n_buckets) |
119 | { | |
120 | struct flex_array *buckets; | |
121 | int i, err; | |
122 | ||
123 | buckets = flex_array_alloc(sizeof(struct hlist_head), | |
124 | n_buckets, GFP_KERNEL); | |
125 | if (!buckets) | |
126 | return NULL; | |
127 | ||
128 | err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL); | |
129 | if (err) { | |
130 | flex_array_free(buckets); | |
131 | return NULL; | |
132 | } | |
133 | ||
134 | for (i = 0; i < n_buckets; i++) | |
135 | INIT_HLIST_HEAD((struct hlist_head *) | |
136 | flex_array_get(buckets, i)); | |
137 | ||
138 | return buckets; | |
139 | } | |
140 | ||
141 | static void flow_free(struct sw_flow *flow) | |
142 | { | |
63e7959c JR |
143 | int node; |
144 | ||
74ed7ab9 JS |
145 | if (ovs_identifier_is_key(&flow->id)) |
146 | kfree(flow->id.unmasked_key); | |
34ae932a TG |
147 | if (flow->sf_acts) |
148 | ovs_nla_free_flow_actions((struct sw_flow_actions __force *)flow->sf_acts); | |
63e7959c JR |
149 | for_each_node(node) |
150 | if (flow->stats[node]) | |
151 | kmem_cache_free(flow_stats_cache, | |
152 | (struct flow_stats __force *)flow->stats[node]); | |
e6445719 PS |
153 | kmem_cache_free(flow_cache, flow); |
154 | } | |
155 | ||
156 | static void rcu_free_flow_callback(struct rcu_head *rcu) | |
157 | { | |
158 | struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); | |
159 | ||
160 | flow_free(flow); | |
161 | } | |
162 | ||
e80857cc | 163 | void ovs_flow_free(struct sw_flow *flow, bool deferred) |
618ed0c8 | 164 | { |
e80857cc | 165 | if (!flow) |
618ed0c8 PS |
166 | return; |
167 | ||
e6445719 PS |
168 | if (deferred) |
169 | call_rcu(&flow->rcu, rcu_free_flow_callback); | |
170 | else | |
171 | flow_free(flow); | |
172 | } | |
173 | ||
174 | static void free_buckets(struct flex_array *buckets) | |
175 | { | |
176 | flex_array_free(buckets); | |
177 | } | |
178 | ||
e80857cc | 179 | |
b637e498 | 180 | static void __table_instance_destroy(struct table_instance *ti) |
e6445719 | 181 | { |
b637e498 PS |
182 | free_buckets(ti->buckets); |
183 | kfree(ti); | |
e6445719 PS |
184 | } |
185 | ||
b637e498 | 186 | static struct table_instance *table_instance_alloc(int new_size) |
e6445719 | 187 | { |
b637e498 | 188 | struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); |
e6445719 | 189 | |
b637e498 | 190 | if (!ti) |
e6445719 PS |
191 | return NULL; |
192 | ||
b637e498 | 193 | ti->buckets = alloc_buckets(new_size); |
e6445719 | 194 | |
b637e498 PS |
195 | if (!ti->buckets) { |
196 | kfree(ti); | |
e6445719 PS |
197 | return NULL; |
198 | } | |
b637e498 PS |
199 | ti->n_buckets = new_size; |
200 | ti->node_ver = 0; | |
201 | ti->keep_flows = false; | |
202 | get_random_bytes(&ti->hash_seed, sizeof(u32)); | |
e6445719 | 203 | |
b637e498 | 204 | return ti; |
e6445719 PS |
205 | } |
206 | ||
b637e498 | 207 | int ovs_flow_tbl_init(struct flow_table *table) |
e6445719 | 208 | { |
74ed7ab9 | 209 | struct table_instance *ti, *ufid_ti; |
e6445719 | 210 | |
b637e498 | 211 | ti = table_instance_alloc(TBL_MIN_BUCKETS); |
e6445719 | 212 | |
b637e498 PS |
213 | if (!ti) |
214 | return -ENOMEM; | |
e6445719 | 215 | |
74ed7ab9 JS |
216 | ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
217 | if (!ufid_ti) | |
218 | goto free_ti; | |
219 | ||
b637e498 | 220 | rcu_assign_pointer(table->ti, ti); |
74ed7ab9 | 221 | rcu_assign_pointer(table->ufid_ti, ufid_ti); |
b637e498 PS |
222 | INIT_LIST_HEAD(&table->mask_list); |
223 | table->last_rehash = jiffies; | |
224 | table->count = 0; | |
74ed7ab9 | 225 | table->ufid_count = 0; |
b637e498 | 226 | return 0; |
74ed7ab9 JS |
227 | |
228 | free_ti: | |
229 | __table_instance_destroy(ti); | |
230 | return -ENOMEM; | |
e6445719 PS |
231 | } |
232 | ||
233 | static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) | |
234 | { | |
b637e498 | 235 | struct table_instance *ti = container_of(rcu, struct table_instance, rcu); |
e6445719 | 236 | |
b637e498 | 237 | __table_instance_destroy(ti); |
e6445719 PS |
238 | } |
239 | ||
74ed7ab9 JS |
240 | static void table_instance_destroy(struct table_instance *ti, |
241 | struct table_instance *ufid_ti, | |
242 | bool deferred) | |
e6445719 | 243 | { |
e80857cc AZ |
244 | int i; |
245 | ||
b637e498 | 246 | if (!ti) |
e6445719 PS |
247 | return; |
248 | ||
74ed7ab9 | 249 | BUG_ON(!ufid_ti); |
e80857cc AZ |
250 | if (ti->keep_flows) |
251 | goto skip_flows; | |
252 | ||
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; | |
74ed7ab9 | 258 | int ufid_ver = ufid_ti->node_ver; |
e80857cc | 259 | |
74ed7ab9 JS |
260 | hlist_for_each_entry_safe(flow, n, head, flow_table.node[ver]) { |
261 | hlist_del_rcu(&flow->flow_table.node[ver]); | |
262 | if (ovs_identifier_is_ufid(&flow->id)) | |
263 | hlist_del_rcu(&flow->ufid_table.node[ufid_ver]); | |
e80857cc AZ |
264 | ovs_flow_free(flow, deferred); |
265 | } | |
266 | } | |
267 | ||
268 | skip_flows: | |
74ed7ab9 | 269 | if (deferred) { |
b637e498 | 270 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
74ed7ab9 JS |
271 | call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb); |
272 | } else { | |
b637e498 | 273 | __table_instance_destroy(ti); |
74ed7ab9 JS |
274 | __table_instance_destroy(ufid_ti); |
275 | } | |
b637e498 PS |
276 | } |
277 | ||
9b996e54 PS |
278 | /* No need for locking this function is called from RCU callback or |
279 | * error path. | |
280 | */ | |
281 | void ovs_flow_tbl_destroy(struct flow_table *table) | |
b637e498 | 282 | { |
9b996e54 | 283 | struct table_instance *ti = rcu_dereference_raw(table->ti); |
74ed7ab9 | 284 | struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti); |
b637e498 | 285 | |
74ed7ab9 | 286 | table_instance_destroy(ti, ufid_ti, false); |
e6445719 PS |
287 | } |
288 | ||
b637e498 | 289 | struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, |
e6445719 PS |
290 | u32 *bucket, u32 *last) |
291 | { | |
292 | struct sw_flow *flow; | |
293 | struct hlist_head *head; | |
294 | int ver; | |
295 | int i; | |
296 | ||
b637e498 PS |
297 | ver = ti->node_ver; |
298 | while (*bucket < ti->n_buckets) { | |
e6445719 | 299 | i = 0; |
b637e498 | 300 | head = flex_array_get(ti->buckets, *bucket); |
74ed7ab9 | 301 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) { |
e6445719 PS |
302 | if (i < *last) { |
303 | i++; | |
304 | continue; | |
305 | } | |
306 | *last = i + 1; | |
307 | return flow; | |
308 | } | |
309 | (*bucket)++; | |
310 | *last = 0; | |
311 | } | |
312 | ||
313 | return NULL; | |
314 | } | |
315 | ||
b637e498 | 316 | static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) |
e6445719 | 317 | { |
b637e498 PS |
318 | hash = jhash_1word(hash, ti->hash_seed); |
319 | return flex_array_get(ti->buckets, | |
320 | (hash & (ti->n_buckets - 1))); | |
e6445719 PS |
321 | } |
322 | ||
74ed7ab9 JS |
323 | static void table_instance_insert(struct table_instance *ti, |
324 | struct sw_flow *flow) | |
e6445719 PS |
325 | { |
326 | struct hlist_head *head; | |
327 | ||
74ed7ab9 JS |
328 | head = find_bucket(ti, flow->flow_table.hash); |
329 | hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head); | |
330 | } | |
331 | ||
332 | static void ufid_table_instance_insert(struct table_instance *ti, | |
333 | struct sw_flow *flow) | |
334 | { | |
335 | struct hlist_head *head; | |
336 | ||
337 | head = find_bucket(ti, flow->ufid_table.hash); | |
338 | hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head); | |
e6445719 PS |
339 | } |
340 | ||
b637e498 | 341 | static void flow_table_copy_flows(struct table_instance *old, |
74ed7ab9 | 342 | struct table_instance *new, bool ufid) |
e6445719 PS |
343 | { |
344 | int old_ver; | |
345 | int i; | |
346 | ||
347 | old_ver = old->node_ver; | |
348 | new->node_ver = !old_ver; | |
349 | ||
350 | /* Insert in new table. */ | |
351 | for (i = 0; i < old->n_buckets; i++) { | |
352 | struct sw_flow *flow; | |
353 | struct hlist_head *head; | |
354 | ||
355 | head = flex_array_get(old->buckets, i); | |
356 | ||
74ed7ab9 JS |
357 | if (ufid) |
358 | hlist_for_each_entry(flow, head, | |
359 | ufid_table.node[old_ver]) | |
360 | ufid_table_instance_insert(new, flow); | |
361 | else | |
362 | hlist_for_each_entry(flow, head, | |
363 | flow_table.node[old_ver]) | |
364 | table_instance_insert(new, flow); | |
e6445719 PS |
365 | } |
366 | ||
e6445719 PS |
367 | old->keep_flows = true; |
368 | } | |
369 | ||
b637e498 | 370 | static struct table_instance *table_instance_rehash(struct table_instance *ti, |
74ed7ab9 | 371 | int n_buckets, bool ufid) |
e6445719 | 372 | { |
b637e498 | 373 | struct table_instance *new_ti; |
e6445719 | 374 | |
b637e498 PS |
375 | new_ti = table_instance_alloc(n_buckets); |
376 | if (!new_ti) | |
618ed0c8 | 377 | return NULL; |
e6445719 | 378 | |
74ed7ab9 | 379 | flow_table_copy_flows(ti, new_ti, ufid); |
e6445719 | 380 | |
b637e498 | 381 | return new_ti; |
e6445719 PS |
382 | } |
383 | ||
b637e498 | 384 | int ovs_flow_tbl_flush(struct flow_table *flow_table) |
e6445719 | 385 | { |
74ed7ab9 JS |
386 | struct table_instance *old_ti, *new_ti; |
387 | struct table_instance *old_ufid_ti, *new_ufid_ti; | |
e6445719 | 388 | |
b637e498 PS |
389 | new_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
390 | if (!new_ti) | |
391 | return -ENOMEM; | |
74ed7ab9 JS |
392 | new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
393 | if (!new_ufid_ti) | |
394 | goto err_free_ti; | |
395 | ||
396 | old_ti = ovsl_dereference(flow_table->ti); | |
397 | old_ufid_ti = ovsl_dereference(flow_table->ufid_ti); | |
b637e498 PS |
398 | |
399 | rcu_assign_pointer(flow_table->ti, new_ti); | |
74ed7ab9 | 400 | rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti); |
b637e498 PS |
401 | flow_table->last_rehash = jiffies; |
402 | flow_table->count = 0; | |
74ed7ab9 | 403 | flow_table->ufid_count = 0; |
b637e498 | 404 | |
74ed7ab9 | 405 | table_instance_destroy(old_ti, old_ufid_ti, true); |
b637e498 | 406 | return 0; |
74ed7ab9 JS |
407 | |
408 | err_free_ti: | |
409 | __table_instance_destroy(new_ti); | |
410 | return -ENOMEM; | |
e6445719 PS |
411 | } |
412 | ||
272c2cf8 JS |
413 | static u32 flow_hash(const struct sw_flow_key *key, |
414 | const struct sw_flow_key_range *range) | |
e6445719 | 415 | { |
272c2cf8 JS |
416 | int key_start = range->start; |
417 | int key_end = range->end; | |
7085130b | 418 | const u32 *hash_key = (const u32 *)((const u8 *)key + key_start); |
e6445719 PS |
419 | int hash_u32s = (key_end - key_start) >> 2; |
420 | ||
421 | /* Make sure number of hash bytes are multiple of u32. */ | |
422 | BUILD_BUG_ON(sizeof(long) % sizeof(u32)); | |
423 | ||
87545899 | 424 | return jhash2(hash_key, hash_u32s, 0); |
e6445719 PS |
425 | } |
426 | ||
427 | static int flow_key_start(const struct sw_flow_key *key) | |
428 | { | |
c1ea5d67 | 429 | if (key->tun_key.u.ipv4.dst) |
e6445719 PS |
430 | return 0; |
431 | else | |
432 | return rounddown(offsetof(struct sw_flow_key, phy), | |
433 | sizeof(long)); | |
434 | } | |
435 | ||
436 | static bool cmp_key(const struct sw_flow_key *key1, | |
437 | const struct sw_flow_key *key2, | |
438 | int key_start, int key_end) | |
439 | { | |
7085130b DDP |
440 | const long *cp1 = (const long *)((const u8 *)key1 + key_start); |
441 | const long *cp2 = (const long *)((const u8 *)key2 + key_start); | |
e6445719 PS |
442 | long diffs = 0; |
443 | int i; | |
444 | ||
445 | for (i = key_start; i < key_end; i += sizeof(long)) | |
446 | diffs |= *cp1++ ^ *cp2++; | |
447 | ||
448 | return diffs == 0; | |
449 | } | |
450 | ||
451 | static bool flow_cmp_masked_key(const struct sw_flow *flow, | |
452 | const struct sw_flow_key *key, | |
272c2cf8 | 453 | const struct sw_flow_key_range *range) |
e6445719 | 454 | { |
272c2cf8 | 455 | return cmp_key(&flow->key, key, range->start, range->end); |
e6445719 PS |
456 | } |
457 | ||
74ed7ab9 JS |
458 | static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, |
459 | const struct sw_flow_match *match) | |
e6445719 PS |
460 | { |
461 | struct sw_flow_key *key = match->key; | |
462 | int key_start = flow_key_start(key); | |
463 | int key_end = match->range.end; | |
464 | ||
74ed7ab9 JS |
465 | BUG_ON(ovs_identifier_is_ufid(&flow->id)); |
466 | return cmp_key(flow->id.unmasked_key, key, key_start, key_end); | |
e6445719 PS |
467 | } |
468 | ||
b637e498 | 469 | static struct sw_flow *masked_flow_lookup(struct table_instance *ti, |
e6445719 | 470 | const struct sw_flow_key *unmasked, |
12eb18f7 | 471 | const struct sw_flow_mask *mask) |
e6445719 PS |
472 | { |
473 | struct sw_flow *flow; | |
474 | struct hlist_head *head; | |
e6445719 PS |
475 | u32 hash; |
476 | struct sw_flow_key masked_key; | |
477 | ||
478 | ovs_flow_mask_key(&masked_key, unmasked, mask); | |
272c2cf8 | 479 | hash = flow_hash(&masked_key, &mask->range); |
b637e498 | 480 | head = find_bucket(ti, hash); |
74ed7ab9 JS |
481 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) { |
482 | if (flow->mask == mask && flow->flow_table.hash == hash && | |
272c2cf8 | 483 | flow_cmp_masked_key(flow, &masked_key, &mask->range)) |
e6445719 PS |
484 | return flow; |
485 | } | |
486 | return NULL; | |
487 | } | |
488 | ||
5bb50632 | 489 | struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl, |
1bd7116f AZ |
490 | const struct sw_flow_key *key, |
491 | u32 *n_mask_hit) | |
e6445719 | 492 | { |
663efa36 | 493 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); |
e6445719 | 494 | struct sw_flow_mask *mask; |
b637e498 | 495 | struct sw_flow *flow; |
e6445719 | 496 | |
1bd7116f | 497 | *n_mask_hit = 0; |
b637e498 | 498 | list_for_each_entry_rcu(mask, &tbl->mask_list, list) { |
1bd7116f | 499 | (*n_mask_hit)++; |
b637e498 | 500 | flow = masked_flow_lookup(ti, key, mask); |
e6445719 | 501 | if (flow) /* Found */ |
b637e498 | 502 | return flow; |
e6445719 | 503 | } |
b637e498 PS |
504 | return NULL; |
505 | } | |
e6445719 | 506 | |
5bb50632 AZ |
507 | struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl, |
508 | const struct sw_flow_key *key) | |
509 | { | |
510 | u32 __always_unused n_mask_hit; | |
511 | ||
512 | return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit); | |
513 | } | |
514 | ||
4a46b24e | 515 | struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl, |
12eb18f7 | 516 | const struct sw_flow_match *match) |
4a46b24e AW |
517 | { |
518 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); | |
519 | struct sw_flow_mask *mask; | |
520 | struct sw_flow *flow; | |
521 | ||
522 | /* Always called under ovs-mutex. */ | |
523 | list_for_each_entry(mask, &tbl->mask_list, list) { | |
524 | flow = masked_flow_lookup(ti, match->key, mask); | |
74ed7ab9 JS |
525 | if (flow && ovs_identifier_is_key(&flow->id) && |
526 | ovs_flow_cmp_unmasked_key(flow, match)) | |
527 | return flow; | |
528 | } | |
529 | return NULL; | |
530 | } | |
531 | ||
532 | static u32 ufid_hash(const struct sw_flow_id *sfid) | |
533 | { | |
534 | return jhash(sfid->ufid, sfid->ufid_len, 0); | |
535 | } | |
536 | ||
537 | static bool ovs_flow_cmp_ufid(const struct sw_flow *flow, | |
538 | const struct sw_flow_id *sfid) | |
539 | { | |
540 | if (flow->id.ufid_len != sfid->ufid_len) | |
541 | return false; | |
542 | ||
543 | return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len); | |
544 | } | |
545 | ||
546 | bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match) | |
547 | { | |
548 | if (ovs_identifier_is_ufid(&flow->id)) | |
549 | return flow_cmp_masked_key(flow, match->key, &match->range); | |
550 | ||
551 | return ovs_flow_cmp_unmasked_key(flow, match); | |
552 | } | |
553 | ||
554 | struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl, | |
555 | const struct sw_flow_id *ufid) | |
556 | { | |
557 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti); | |
558 | struct sw_flow *flow; | |
559 | struct hlist_head *head; | |
560 | u32 hash; | |
561 | ||
562 | hash = ufid_hash(ufid); | |
563 | head = find_bucket(ti, hash); | |
564 | hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) { | |
565 | if (flow->ufid_table.hash == hash && | |
566 | ovs_flow_cmp_ufid(flow, ufid)) | |
4a46b24e AW |
567 | return flow; |
568 | } | |
569 | return NULL; | |
570 | } | |
571 | ||
1bd7116f AZ |
572 | int ovs_flow_tbl_num_masks(const struct flow_table *table) |
573 | { | |
574 | struct sw_flow_mask *mask; | |
575 | int num = 0; | |
576 | ||
577 | list_for_each_entry(mask, &table->mask_list, list) | |
578 | num++; | |
579 | ||
580 | return num; | |
581 | } | |
582 | ||
74ed7ab9 JS |
583 | static struct table_instance *table_instance_expand(struct table_instance *ti, |
584 | bool ufid) | |
b637e498 | 585 | { |
74ed7ab9 | 586 | return table_instance_rehash(ti, ti->n_buckets * 2, ufid); |
e6445719 PS |
587 | } |
588 | ||
56c19868 JR |
589 | /* Remove 'mask' from the mask list, if it is not needed any more. */ |
590 | static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask) | |
591 | { | |
592 | if (mask) { | |
593 | /* ovs-lock is required to protect mask-refcount and | |
594 | * mask list. | |
595 | */ | |
596 | ASSERT_OVSL(); | |
597 | BUG_ON(!mask->ref_count); | |
598 | mask->ref_count--; | |
599 | ||
600 | if (!mask->ref_count) { | |
601 | list_del_rcu(&mask->list); | |
602 | kfree_rcu(mask, rcu); | |
603 | } | |
604 | } | |
605 | } | |
606 | ||
607 | /* Must be called with OVS mutex held. */ | |
e6445719 PS |
608 | void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) |
609 | { | |
b637e498 | 610 | struct table_instance *ti = ovsl_dereference(table->ti); |
74ed7ab9 | 611 | struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti); |
b637e498 | 612 | |
e6445719 | 613 | BUG_ON(table->count == 0); |
74ed7ab9 | 614 | hlist_del_rcu(&flow->flow_table.node[ti->node_ver]); |
e6445719 | 615 | table->count--; |
74ed7ab9 JS |
616 | if (ovs_identifier_is_ufid(&flow->id)) { |
617 | hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]); | |
618 | table->ufid_count--; | |
619 | } | |
56c19868 JR |
620 | |
621 | /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be | |
622 | * accessible as long as the RCU read lock is held. | |
623 | */ | |
624 | flow_mask_remove(table, flow->mask); | |
e6445719 PS |
625 | } |
626 | ||
618ed0c8 | 627 | static struct sw_flow_mask *mask_alloc(void) |
e6445719 PS |
628 | { |
629 | struct sw_flow_mask *mask; | |
630 | ||
631 | mask = kmalloc(sizeof(*mask), GFP_KERNEL); | |
632 | if (mask) | |
e80857cc | 633 | mask->ref_count = 1; |
e6445719 PS |
634 | |
635 | return mask; | |
636 | } | |
637 | ||
e6445719 PS |
638 | static bool mask_equal(const struct sw_flow_mask *a, |
639 | const struct sw_flow_mask *b) | |
640 | { | |
7085130b DDP |
641 | const u8 *a_ = (const u8 *)&a->key + a->range.start; |
642 | const u8 *b_ = (const u8 *)&b->key + b->range.start; | |
e6445719 PS |
643 | |
644 | return (a->range.end == b->range.end) | |
645 | && (a->range.start == b->range.start) | |
646 | && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0); | |
647 | } | |
648 | ||
618ed0c8 | 649 | static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl, |
e6445719 PS |
650 | const struct sw_flow_mask *mask) |
651 | { | |
652 | struct list_head *ml; | |
653 | ||
b637e498 | 654 | list_for_each(ml, &tbl->mask_list) { |
e6445719 PS |
655 | struct sw_flow_mask *m; |
656 | m = container_of(ml, struct sw_flow_mask, list); | |
657 | if (mask_equal(mask, m)) | |
658 | return m; | |
659 | } | |
660 | ||
661 | return NULL; | |
662 | } | |
663 | ||
d1211908 | 664 | /* Add 'mask' into the mask list, if it is not already there. */ |
618ed0c8 | 665 | static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow, |
12eb18f7 | 666 | const struct sw_flow_mask *new) |
618ed0c8 PS |
667 | { |
668 | struct sw_flow_mask *mask; | |
669 | mask = flow_mask_find(tbl, new); | |
670 | if (!mask) { | |
671 | /* Allocate a new mask if none exsits. */ | |
672 | mask = mask_alloc(); | |
673 | if (!mask) | |
674 | return -ENOMEM; | |
675 | mask->key = new->key; | |
676 | mask->range = new->range; | |
677 | list_add_rcu(&mask->list, &tbl->mask_list); | |
e80857cc AZ |
678 | } else { |
679 | BUG_ON(!mask->ref_count); | |
680 | mask->ref_count++; | |
618ed0c8 PS |
681 | } |
682 | ||
618ed0c8 PS |
683 | flow->mask = mask; |
684 | return 0; | |
685 | } | |
686 | ||
56c19868 | 687 | /* Must be called with OVS mutex held. */ |
d29ab6f8 | 688 | static void flow_key_insert(struct flow_table *table, struct sw_flow *flow) |
e6445719 | 689 | { |
618ed0c8 PS |
690 | struct table_instance *new_ti = NULL; |
691 | struct table_instance *ti; | |
618ed0c8 | 692 | |
74ed7ab9 | 693 | flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range); |
618ed0c8 PS |
694 | ti = ovsl_dereference(table->ti); |
695 | table_instance_insert(ti, flow); | |
696 | table->count++; | |
697 | ||
698 | /* Expand table, if necessary, to make room. */ | |
699 | if (table->count > ti->n_buckets) | |
74ed7ab9 | 700 | new_ti = table_instance_expand(ti, false); |
618ed0c8 | 701 | else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) |
74ed7ab9 | 702 | new_ti = table_instance_rehash(ti, ti->n_buckets, false); |
618ed0c8 PS |
703 | |
704 | if (new_ti) { | |
705 | rcu_assign_pointer(table->ti, new_ti); | |
74ed7ab9 | 706 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
618ed0c8 PS |
707 | table->last_rehash = jiffies; |
708 | } | |
d29ab6f8 JS |
709 | } |
710 | ||
74ed7ab9 JS |
711 | /* Must be called with OVS mutex held. */ |
712 | static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow) | |
713 | { | |
714 | struct table_instance *ti; | |
715 | ||
716 | flow->ufid_table.hash = ufid_hash(&flow->id); | |
717 | ti = ovsl_dereference(table->ufid_ti); | |
718 | ufid_table_instance_insert(ti, flow); | |
719 | table->ufid_count++; | |
720 | ||
721 | /* Expand table, if necessary, to make room. */ | |
722 | if (table->ufid_count > ti->n_buckets) { | |
723 | struct table_instance *new_ti; | |
724 | ||
725 | new_ti = table_instance_expand(ti, true); | |
726 | if (new_ti) { | |
727 | rcu_assign_pointer(table->ufid_ti, new_ti); | |
728 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); | |
729 | } | |
730 | } | |
731 | } | |
732 | ||
d29ab6f8 JS |
733 | /* Must be called with OVS mutex held. */ |
734 | int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow, | |
735 | const struct sw_flow_mask *mask) | |
736 | { | |
737 | int err; | |
738 | ||
739 | err = flow_mask_insert(table, flow, mask); | |
740 | if (err) | |
741 | return err; | |
742 | flow_key_insert(table, flow); | |
74ed7ab9 JS |
743 | if (ovs_identifier_is_ufid(&flow->id)) |
744 | flow_ufid_insert(table, flow); | |
d29ab6f8 | 745 | |
618ed0c8 | 746 | return 0; |
e6445719 PS |
747 | } |
748 | ||
749 | /* Initializes the flow module. | |
750 | * Returns zero if successful or a negative error code. */ | |
751 | int ovs_flow_init(void) | |
752 | { | |
753 | BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); | |
754 | BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); | |
755 | ||
63e7959c | 756 | flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow) |
bac541e4 | 757 | + (nr_node_ids |
63e7959c JR |
758 | * sizeof(struct flow_stats *)), |
759 | 0, 0, NULL); | |
e6445719 PS |
760 | if (flow_cache == NULL) |
761 | return -ENOMEM; | |
762 | ||
63e7959c JR |
763 | flow_stats_cache |
764 | = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats), | |
765 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
766 | if (flow_stats_cache == NULL) { | |
767 | kmem_cache_destroy(flow_cache); | |
768 | flow_cache = NULL; | |
769 | return -ENOMEM; | |
770 | } | |
771 | ||
e6445719 PS |
772 | return 0; |
773 | } | |
774 | ||
775 | /* Uninitializes the flow module. */ | |
776 | void ovs_flow_exit(void) | |
777 | { | |
63e7959c | 778 | kmem_cache_destroy(flow_stats_cache); |
e6445719 PS |
779 | kmem_cache_destroy(flow_cache); |
780 | } |