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" | |
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> | |
87545899 | 28 | #include <linux/jhash.h> |
e6445719 PS |
29 | #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 | ||
b637e498 PS |
47 | #define TBL_MIN_BUCKETS 1024 |
48 | #define REHASH_INTERVAL (10 * 60 * HZ) | |
49 | ||
e6445719 | 50 | static struct kmem_cache *flow_cache; |
63e7959c | 51 | struct kmem_cache *flow_stats_cache __read_mostly; |
e6445719 PS |
52 | |
53 | static u16 range_n_bytes(const struct sw_flow_key_range *range) | |
54 | { | |
55 | return range->end - range->start; | |
56 | } | |
57 | ||
58 | void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, | |
59 | const struct sw_flow_mask *mask) | |
60 | { | |
7085130b DDP |
61 | 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); | |
e6445719 PS |
65 | 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 | ||
23dabf88 | 76 | struct sw_flow *ovs_flow_alloc(void) |
e6445719 PS |
77 | { |
78 | struct sw_flow *flow; | |
63e7959c JR |
79 | struct flow_stats *stats; |
80 | int node; | |
e6445719 PS |
81 | |
82 | flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); | |
83 | if (!flow) | |
84 | return ERR_PTR(-ENOMEM); | |
85 | ||
e6445719 PS |
86 | flow->sf_acts = NULL; |
87 | flow->mask = NULL; | |
ca539345 PS |
88 | flow->id.unmasked_key = NULL; |
89 | flow->id.ufid_len = 0; | |
63e7959c | 90 | flow->stats_last_writer = NUMA_NO_NODE; |
e6445719 | 91 | |
63e7959c JR |
92 | /* Initialize the default stat node. */ |
93 | stats = kmem_cache_alloc_node(flow_stats_cache, | |
94 | GFP_KERNEL | __GFP_ZERO, 0); | |
95 | if (!stats) | |
23dabf88 | 96 | goto err; |
e298e505 | 97 | |
63e7959c JR |
98 | spin_lock_init(&stats->lock); |
99 | ||
100 | 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); | |
e298e505 | 105 | |
e6445719 | 106 | return flow; |
e298e505 | 107 | err: |
ece37c87 | 108 | kmem_cache_free(flow_cache, flow); |
e298e505 | 109 | return ERR_PTR(-ENOMEM); |
e6445719 PS |
110 | } |
111 | ||
12eb18f7 | 112 | int ovs_flow_tbl_count(const struct flow_table *table) |
b637e498 PS |
113 | { |
114 | return table->count; | |
115 | } | |
116 | ||
e6445719 PS |
117 | static 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 | ||
140 | static void flow_free(struct sw_flow *flow) | |
141 | { | |
63e7959c JR |
142 | int node; |
143 | ||
74ed7ab9 JS |
144 | if (ovs_identifier_is_key(&flow->id)) |
145 | kfree(flow->id.unmasked_key); | |
eb072659 | 146 | kfree((struct sw_flow_actions __force *)flow->sf_acts); |
63e7959c JR |
147 | for_each_node(node) |
148 | if (flow->stats[node]) | |
149 | kmem_cache_free(flow_stats_cache, | |
150 | (struct flow_stats __force *)flow->stats[node]); | |
e6445719 PS |
151 | kmem_cache_free(flow_cache, flow); |
152 | } | |
153 | ||
154 | static 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 | ||
e80857cc | 161 | void ovs_flow_free(struct sw_flow *flow, bool deferred) |
618ed0c8 | 162 | { |
e80857cc | 163 | if (!flow) |
618ed0c8 PS |
164 | return; |
165 | ||
e6445719 PS |
166 | if (deferred) |
167 | call_rcu(&flow->rcu, rcu_free_flow_callback); | |
168 | else | |
169 | flow_free(flow); | |
170 | } | |
171 | ||
172 | static void free_buckets(struct flex_array *buckets) | |
173 | { | |
174 | flex_array_free(buckets); | |
175 | } | |
176 | ||
e80857cc | 177 | |
b637e498 | 178 | static void __table_instance_destroy(struct table_instance *ti) |
e6445719 | 179 | { |
b637e498 PS |
180 | free_buckets(ti->buckets); |
181 | kfree(ti); | |
e6445719 PS |
182 | } |
183 | ||
b637e498 | 184 | static struct table_instance *table_instance_alloc(int new_size) |
e6445719 | 185 | { |
b637e498 | 186 | struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); |
e6445719 | 187 | |
b637e498 | 188 | if (!ti) |
e6445719 PS |
189 | return NULL; |
190 | ||
b637e498 | 191 | ti->buckets = alloc_buckets(new_size); |
e6445719 | 192 | |
b637e498 PS |
193 | if (!ti->buckets) { |
194 | kfree(ti); | |
e6445719 PS |
195 | return NULL; |
196 | } | |
b637e498 PS |
197 | ti->n_buckets = new_size; |
198 | ti->node_ver = 0; | |
199 | ti->keep_flows = false; | |
200 | get_random_bytes(&ti->hash_seed, sizeof(u32)); | |
e6445719 | 201 | |
b637e498 | 202 | return ti; |
e6445719 PS |
203 | } |
204 | ||
b637e498 | 205 | int ovs_flow_tbl_init(struct flow_table *table) |
e6445719 | 206 | { |
74ed7ab9 | 207 | struct table_instance *ti, *ufid_ti; |
e6445719 | 208 | |
b637e498 | 209 | ti = table_instance_alloc(TBL_MIN_BUCKETS); |
e6445719 | 210 | |
b637e498 PS |
211 | if (!ti) |
212 | return -ENOMEM; | |
e6445719 | 213 | |
74ed7ab9 JS |
214 | ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
215 | if (!ufid_ti) | |
216 | goto free_ti; | |
217 | ||
b637e498 | 218 | rcu_assign_pointer(table->ti, ti); |
74ed7ab9 | 219 | rcu_assign_pointer(table->ufid_ti, ufid_ti); |
b637e498 PS |
220 | INIT_LIST_HEAD(&table->mask_list); |
221 | table->last_rehash = jiffies; | |
222 | table->count = 0; | |
74ed7ab9 | 223 | table->ufid_count = 0; |
b637e498 | 224 | return 0; |
74ed7ab9 JS |
225 | |
226 | free_ti: | |
227 | __table_instance_destroy(ti); | |
228 | return -ENOMEM; | |
e6445719 PS |
229 | } |
230 | ||
231 | static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) | |
232 | { | |
b637e498 | 233 | struct table_instance *ti = container_of(rcu, struct table_instance, rcu); |
e6445719 | 234 | |
b637e498 | 235 | __table_instance_destroy(ti); |
e6445719 PS |
236 | } |
237 | ||
74ed7ab9 JS |
238 | static void table_instance_destroy(struct table_instance *ti, |
239 | struct table_instance *ufid_ti, | |
240 | bool deferred) | |
e6445719 | 241 | { |
e80857cc AZ |
242 | int i; |
243 | ||
b637e498 | 244 | if (!ti) |
e6445719 PS |
245 | return; |
246 | ||
74ed7ab9 | 247 | BUG_ON(!ufid_ti); |
e80857cc AZ |
248 | 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; | |
74ed7ab9 | 256 | int ufid_ver = ufid_ti->node_ver; |
e80857cc | 257 | |
74ed7ab9 JS |
258 | 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]); | |
e80857cc AZ |
262 | ovs_flow_free(flow, deferred); |
263 | } | |
264 | } | |
265 | ||
266 | skip_flows: | |
74ed7ab9 | 267 | if (deferred) { |
b637e498 | 268 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
74ed7ab9 JS |
269 | call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb); |
270 | } else { | |
b637e498 | 271 | __table_instance_destroy(ti); |
74ed7ab9 JS |
272 | __table_instance_destroy(ufid_ti); |
273 | } | |
b637e498 PS |
274 | } |
275 | ||
9b996e54 PS |
276 | /* No need for locking this function is called from RCU callback or |
277 | * error path. | |
278 | */ | |
279 | void ovs_flow_tbl_destroy(struct flow_table *table) | |
b637e498 | 280 | { |
9b996e54 | 281 | struct table_instance *ti = rcu_dereference_raw(table->ti); |
74ed7ab9 | 282 | struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti); |
b637e498 | 283 | |
74ed7ab9 | 284 | table_instance_destroy(ti, ufid_ti, false); |
e6445719 PS |
285 | } |
286 | ||
b637e498 | 287 | struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, |
e6445719 PS |
288 | u32 *bucket, u32 *last) |
289 | { | |
290 | struct sw_flow *flow; | |
291 | struct hlist_head *head; | |
292 | int ver; | |
293 | int i; | |
294 | ||
b637e498 PS |
295 | ver = ti->node_ver; |
296 | while (*bucket < ti->n_buckets) { | |
e6445719 | 297 | i = 0; |
b637e498 | 298 | head = flex_array_get(ti->buckets, *bucket); |
74ed7ab9 | 299 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) { |
e6445719 PS |
300 | 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 | ||
b637e498 | 314 | static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) |
e6445719 | 315 | { |
b637e498 PS |
316 | hash = jhash_1word(hash, ti->hash_seed); |
317 | return flex_array_get(ti->buckets, | |
318 | (hash & (ti->n_buckets - 1))); | |
e6445719 PS |
319 | } |
320 | ||
74ed7ab9 JS |
321 | static void table_instance_insert(struct table_instance *ti, |
322 | struct sw_flow *flow) | |
e6445719 PS |
323 | { |
324 | struct hlist_head *head; | |
325 | ||
74ed7ab9 JS |
326 | head = find_bucket(ti, flow->flow_table.hash); |
327 | hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head); | |
328 | } | |
329 | ||
330 | static 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); | |
e6445719 PS |
337 | } |
338 | ||
b637e498 | 339 | static void flow_table_copy_flows(struct table_instance *old, |
74ed7ab9 | 340 | struct table_instance *new, bool ufid) |
e6445719 PS |
341 | { |
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 | ||
74ed7ab9 JS |
355 | 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); | |
e6445719 PS |
363 | } |
364 | ||
e6445719 PS |
365 | old->keep_flows = true; |
366 | } | |
367 | ||
b637e498 | 368 | static struct table_instance *table_instance_rehash(struct table_instance *ti, |
74ed7ab9 | 369 | int n_buckets, bool ufid) |
e6445719 | 370 | { |
b637e498 | 371 | struct table_instance *new_ti; |
e6445719 | 372 | |
b637e498 PS |
373 | new_ti = table_instance_alloc(n_buckets); |
374 | if (!new_ti) | |
618ed0c8 | 375 | return NULL; |
e6445719 | 376 | |
74ed7ab9 | 377 | flow_table_copy_flows(ti, new_ti, ufid); |
e6445719 | 378 | |
b637e498 | 379 | return new_ti; |
e6445719 PS |
380 | } |
381 | ||
b637e498 | 382 | int ovs_flow_tbl_flush(struct flow_table *flow_table) |
e6445719 | 383 | { |
74ed7ab9 JS |
384 | struct table_instance *old_ti, *new_ti; |
385 | struct table_instance *old_ufid_ti, *new_ufid_ti; | |
e6445719 | 386 | |
b637e498 PS |
387 | new_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
388 | if (!new_ti) | |
389 | return -ENOMEM; | |
74ed7ab9 JS |
390 | 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); | |
b637e498 PS |
396 | |
397 | rcu_assign_pointer(flow_table->ti, new_ti); | |
74ed7ab9 | 398 | rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti); |
b637e498 PS |
399 | flow_table->last_rehash = jiffies; |
400 | flow_table->count = 0; | |
74ed7ab9 | 401 | flow_table->ufid_count = 0; |
b637e498 | 402 | |
74ed7ab9 | 403 | table_instance_destroy(old_ti, old_ufid_ti, true); |
b637e498 | 404 | return 0; |
74ed7ab9 JS |
405 | |
406 | err_free_ti: | |
407 | __table_instance_destroy(new_ti); | |
408 | return -ENOMEM; | |
e6445719 PS |
409 | } |
410 | ||
272c2cf8 JS |
411 | static u32 flow_hash(const struct sw_flow_key *key, |
412 | const struct sw_flow_key_range *range) | |
e6445719 | 413 | { |
272c2cf8 JS |
414 | int key_start = range->start; |
415 | int key_end = range->end; | |
7085130b | 416 | const u32 *hash_key = (const u32 *)((const u8 *)key + key_start); |
e6445719 PS |
417 | 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 | ||
87545899 | 422 | return jhash2(hash_key, hash_u32s, 0); |
e6445719 PS |
423 | } |
424 | ||
425 | static 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 | ||
434 | static bool cmp_key(const struct sw_flow_key *key1, | |
435 | const struct sw_flow_key *key2, | |
436 | int key_start, int key_end) | |
437 | { | |
7085130b DDP |
438 | const long *cp1 = (const long *)((const u8 *)key1 + key_start); |
439 | const long *cp2 = (const long *)((const u8 *)key2 + key_start); | |
e6445719 PS |
440 | 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 | ||
449 | static bool flow_cmp_masked_key(const struct sw_flow *flow, | |
450 | const struct sw_flow_key *key, | |
272c2cf8 | 451 | const struct sw_flow_key_range *range) |
e6445719 | 452 | { |
272c2cf8 | 453 | return cmp_key(&flow->key, key, range->start, range->end); |
e6445719 PS |
454 | } |
455 | ||
74ed7ab9 JS |
456 | static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, |
457 | const struct sw_flow_match *match) | |
e6445719 PS |
458 | { |
459 | struct sw_flow_key *key = match->key; | |
460 | int key_start = flow_key_start(key); | |
461 | int key_end = match->range.end; | |
462 | ||
74ed7ab9 JS |
463 | BUG_ON(ovs_identifier_is_ufid(&flow->id)); |
464 | return cmp_key(flow->id.unmasked_key, key, key_start, key_end); | |
e6445719 PS |
465 | } |
466 | ||
b637e498 | 467 | static struct sw_flow *masked_flow_lookup(struct table_instance *ti, |
e6445719 | 468 | const struct sw_flow_key *unmasked, |
12eb18f7 | 469 | const struct sw_flow_mask *mask) |
e6445719 PS |
470 | { |
471 | struct sw_flow *flow; | |
472 | struct hlist_head *head; | |
e6445719 PS |
473 | u32 hash; |
474 | struct sw_flow_key masked_key; | |
475 | ||
476 | ovs_flow_mask_key(&masked_key, unmasked, mask); | |
272c2cf8 | 477 | hash = flow_hash(&masked_key, &mask->range); |
b637e498 | 478 | head = find_bucket(ti, hash); |
74ed7ab9 JS |
479 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) { |
480 | if (flow->mask == mask && flow->flow_table.hash == hash && | |
272c2cf8 | 481 | flow_cmp_masked_key(flow, &masked_key, &mask->range)) |
e6445719 PS |
482 | return flow; |
483 | } | |
484 | return NULL; | |
485 | } | |
486 | ||
5bb50632 | 487 | struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl, |
1bd7116f AZ |
488 | const struct sw_flow_key *key, |
489 | u32 *n_mask_hit) | |
e6445719 | 490 | { |
663efa36 | 491 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); |
e6445719 | 492 | struct sw_flow_mask *mask; |
b637e498 | 493 | struct sw_flow *flow; |
e6445719 | 494 | |
1bd7116f | 495 | *n_mask_hit = 0; |
b637e498 | 496 | list_for_each_entry_rcu(mask, &tbl->mask_list, list) { |
1bd7116f | 497 | (*n_mask_hit)++; |
b637e498 | 498 | flow = masked_flow_lookup(ti, key, mask); |
e6445719 | 499 | if (flow) /* Found */ |
b637e498 | 500 | return flow; |
e6445719 | 501 | } |
b637e498 PS |
502 | return NULL; |
503 | } | |
e6445719 | 504 | |
5bb50632 AZ |
505 | struct 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 | ||
4a46b24e | 513 | struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl, |
12eb18f7 | 514 | const struct sw_flow_match *match) |
4a46b24e AW |
515 | { |
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); | |
74ed7ab9 JS |
523 | 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 | ||
530 | static u32 ufid_hash(const struct sw_flow_id *sfid) | |
531 | { | |
532 | return jhash(sfid->ufid, sfid->ufid_len, 0); | |
533 | } | |
534 | ||
535 | static 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 | ||
544 | bool 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 | ||
552 | struct 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)) | |
4a46b24e AW |
565 | return flow; |
566 | } | |
567 | return NULL; | |
568 | } | |
569 | ||
1bd7116f AZ |
570 | int 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 | ||
74ed7ab9 JS |
581 | static struct table_instance *table_instance_expand(struct table_instance *ti, |
582 | bool ufid) | |
b637e498 | 583 | { |
74ed7ab9 | 584 | return table_instance_rehash(ti, ti->n_buckets * 2, ufid); |
e6445719 PS |
585 | } |
586 | ||
56c19868 JR |
587 | /* Remove 'mask' from the mask list, if it is not needed any more. */ |
588 | static 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. */ | |
e6445719 PS |
606 | void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) |
607 | { | |
b637e498 | 608 | struct table_instance *ti = ovsl_dereference(table->ti); |
74ed7ab9 | 609 | struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti); |
b637e498 | 610 | |
e6445719 | 611 | BUG_ON(table->count == 0); |
74ed7ab9 | 612 | hlist_del_rcu(&flow->flow_table.node[ti->node_ver]); |
e6445719 | 613 | table->count--; |
74ed7ab9 JS |
614 | if (ovs_identifier_is_ufid(&flow->id)) { |
615 | hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]); | |
616 | table->ufid_count--; | |
617 | } | |
56c19868 JR |
618 | |
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); | |
e6445719 PS |
623 | } |
624 | ||
618ed0c8 | 625 | static struct sw_flow_mask *mask_alloc(void) |
e6445719 PS |
626 | { |
627 | struct sw_flow_mask *mask; | |
628 | ||
629 | mask = kmalloc(sizeof(*mask), GFP_KERNEL); | |
630 | if (mask) | |
e80857cc | 631 | mask->ref_count = 1; |
e6445719 PS |
632 | |
633 | return mask; | |
634 | } | |
635 | ||
e6445719 PS |
636 | static bool mask_equal(const struct sw_flow_mask *a, |
637 | const struct sw_flow_mask *b) | |
638 | { | |
7085130b DDP |
639 | const u8 *a_ = (const u8 *)&a->key + a->range.start; |
640 | const u8 *b_ = (const u8 *)&b->key + b->range.start; | |
e6445719 PS |
641 | |
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 | ||
618ed0c8 | 647 | static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl, |
e6445719 PS |
648 | const struct sw_flow_mask *mask) |
649 | { | |
650 | struct list_head *ml; | |
651 | ||
b637e498 | 652 | list_for_each(ml, &tbl->mask_list) { |
e6445719 PS |
653 | 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 | ||
d1211908 | 662 | /* Add 'mask' into the mask list, if it is not already there. */ |
618ed0c8 | 663 | static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow, |
12eb18f7 | 664 | const struct sw_flow_mask *new) |
618ed0c8 PS |
665 | { |
666 | 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); | |
e80857cc AZ |
676 | } else { |
677 | BUG_ON(!mask->ref_count); | |
678 | mask->ref_count++; | |
618ed0c8 PS |
679 | } |
680 | ||
618ed0c8 PS |
681 | flow->mask = mask; |
682 | return 0; | |
683 | } | |
684 | ||
56c19868 | 685 | /* Must be called with OVS mutex held. */ |
d29ab6f8 | 686 | static void flow_key_insert(struct flow_table *table, struct sw_flow *flow) |
e6445719 | 687 | { |
618ed0c8 PS |
688 | struct table_instance *new_ti = NULL; |
689 | struct table_instance *ti; | |
618ed0c8 | 690 | |
74ed7ab9 | 691 | flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range); |
618ed0c8 PS |
692 | 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) | |
74ed7ab9 | 698 | new_ti = table_instance_expand(ti, false); |
618ed0c8 | 699 | else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) |
74ed7ab9 | 700 | new_ti = table_instance_rehash(ti, ti->n_buckets, false); |
618ed0c8 PS |
701 | |
702 | if (new_ti) { | |
703 | rcu_assign_pointer(table->ti, new_ti); | |
74ed7ab9 | 704 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
618ed0c8 PS |
705 | table->last_rehash = jiffies; |
706 | } | |
d29ab6f8 JS |
707 | } |
708 | ||
74ed7ab9 JS |
709 | /* Must be called with OVS mutex held. */ |
710 | static 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 | ||
d29ab6f8 JS |
731 | /* Must be called with OVS mutex held. */ |
732 | int 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); | |
74ed7ab9 JS |
741 | if (ovs_identifier_is_ufid(&flow->id)) |
742 | flow_ufid_insert(table, flow); | |
d29ab6f8 | 743 | |
618ed0c8 | 744 | return 0; |
e6445719 PS |
745 | } |
746 | ||
747 | /* Initializes the flow module. | |
748 | * Returns zero if successful or a negative error code. */ | |
749 | int 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 | ||
63e7959c JR |
754 | flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow) |
755 | + (num_possible_nodes() | |
756 | * sizeof(struct flow_stats *)), | |
757 | 0, 0, NULL); | |
e6445719 PS |
758 | if (flow_cache == NULL) |
759 | return -ENOMEM; | |
760 | ||
63e7959c JR |
761 | 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 | ||
e6445719 PS |
770 | return 0; |
771 | } | |
772 | ||
773 | /* Uninitializes the flow module. */ | |
774 | void ovs_flow_exit(void) | |
775 | { | |
63e7959c | 776 | kmem_cache_destroy(flow_stats_cache); |
e6445719 PS |
777 | kmem_cache_destroy(flow_cache); |
778 | } |