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, | |
ae5f2fb1 | 60 | bool full, const struct sw_flow_mask *mask) |
e6445719 | 61 | { |
ae5f2fb1 JG |
62 | int start = full ? 0 : mask->range.start; |
63 | int len = full ? sizeof *dst : range_n_bytes(&mask->range); | |
64 | const long *m = (const long *)((const u8 *)&mask->key + start); | |
65 | const long *s = (const long *)((const u8 *)src + start); | |
66 | long *d = (long *)((u8 *)dst + start); | |
e6445719 PS |
67 | int i; |
68 | ||
ae5f2fb1 JG |
69 | /* If 'full' is true then all of 'dst' is fully initialized. Otherwise, |
70 | * if 'full' is false the memory outside of the 'mask->range' is left | |
71 | * uninitialized. This can be used as an optimization when further | |
72 | * operations on 'dst' only use contents within 'mask->range'. | |
e6445719 | 73 | */ |
ae5f2fb1 | 74 | for (i = 0; i < len; i += sizeof(long)) |
e6445719 PS |
75 | *d++ = *s++ & *m++; |
76 | } | |
77 | ||
23dabf88 | 78 | struct sw_flow *ovs_flow_alloc(void) |
e6445719 PS |
79 | { |
80 | struct sw_flow *flow; | |
63e7959c JR |
81 | struct flow_stats *stats; |
82 | int node; | |
e6445719 PS |
83 | |
84 | flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); | |
85 | if (!flow) | |
86 | return ERR_PTR(-ENOMEM); | |
87 | ||
e6445719 PS |
88 | flow->sf_acts = NULL; |
89 | flow->mask = NULL; | |
ca539345 PS |
90 | flow->id.unmasked_key = NULL; |
91 | flow->id.ufid_len = 0; | |
63e7959c | 92 | flow->stats_last_writer = NUMA_NO_NODE; |
e6445719 | 93 | |
63e7959c JR |
94 | /* Initialize the default stat node. */ |
95 | stats = kmem_cache_alloc_node(flow_stats_cache, | |
598c12d0 KK |
96 | GFP_KERNEL | __GFP_ZERO, |
97 | node_online(0) ? 0 : NUMA_NO_NODE); | |
63e7959c | 98 | if (!stats) |
23dabf88 | 99 | goto err; |
e298e505 | 100 | |
63e7959c JR |
101 | spin_lock_init(&stats->lock); |
102 | ||
103 | RCU_INIT_POINTER(flow->stats[0], stats); | |
104 | ||
105 | for_each_node(node) | |
106 | if (node != 0) | |
107 | RCU_INIT_POINTER(flow->stats[node], NULL); | |
e298e505 | 108 | |
e6445719 | 109 | return flow; |
e298e505 | 110 | err: |
ece37c87 | 111 | kmem_cache_free(flow_cache, flow); |
e298e505 | 112 | return ERR_PTR(-ENOMEM); |
e6445719 PS |
113 | } |
114 | ||
12eb18f7 | 115 | int ovs_flow_tbl_count(const struct flow_table *table) |
b637e498 PS |
116 | { |
117 | return table->count; | |
118 | } | |
119 | ||
e6445719 PS |
120 | static struct flex_array *alloc_buckets(unsigned int n_buckets) |
121 | { | |
122 | struct flex_array *buckets; | |
123 | int i, err; | |
124 | ||
125 | buckets = flex_array_alloc(sizeof(struct hlist_head), | |
126 | n_buckets, GFP_KERNEL); | |
127 | if (!buckets) | |
128 | return NULL; | |
129 | ||
130 | err = flex_array_prealloc(buckets, 0, n_buckets, GFP_KERNEL); | |
131 | if (err) { | |
132 | flex_array_free(buckets); | |
133 | return NULL; | |
134 | } | |
135 | ||
136 | for (i = 0; i < n_buckets; i++) | |
137 | INIT_HLIST_HEAD((struct hlist_head *) | |
138 | flex_array_get(buckets, i)); | |
139 | ||
140 | return buckets; | |
141 | } | |
142 | ||
143 | static void flow_free(struct sw_flow *flow) | |
144 | { | |
63e7959c JR |
145 | int node; |
146 | ||
74ed7ab9 JS |
147 | if (ovs_identifier_is_key(&flow->id)) |
148 | kfree(flow->id.unmasked_key); | |
34ae932a TG |
149 | if (flow->sf_acts) |
150 | ovs_nla_free_flow_actions((struct sw_flow_actions __force *)flow->sf_acts); | |
63e7959c JR |
151 | for_each_node(node) |
152 | if (flow->stats[node]) | |
153 | kmem_cache_free(flow_stats_cache, | |
154 | (struct flow_stats __force *)flow->stats[node]); | |
e6445719 PS |
155 | kmem_cache_free(flow_cache, flow); |
156 | } | |
157 | ||
158 | static void rcu_free_flow_callback(struct rcu_head *rcu) | |
159 | { | |
160 | struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); | |
161 | ||
162 | flow_free(flow); | |
163 | } | |
164 | ||
e80857cc | 165 | void ovs_flow_free(struct sw_flow *flow, bool deferred) |
618ed0c8 | 166 | { |
e80857cc | 167 | if (!flow) |
618ed0c8 PS |
168 | return; |
169 | ||
e6445719 PS |
170 | if (deferred) |
171 | call_rcu(&flow->rcu, rcu_free_flow_callback); | |
172 | else | |
173 | flow_free(flow); | |
174 | } | |
175 | ||
176 | static void free_buckets(struct flex_array *buckets) | |
177 | { | |
178 | flex_array_free(buckets); | |
179 | } | |
180 | ||
e80857cc | 181 | |
b637e498 | 182 | static void __table_instance_destroy(struct table_instance *ti) |
e6445719 | 183 | { |
b637e498 PS |
184 | free_buckets(ti->buckets); |
185 | kfree(ti); | |
e6445719 PS |
186 | } |
187 | ||
b637e498 | 188 | static struct table_instance *table_instance_alloc(int new_size) |
e6445719 | 189 | { |
b637e498 | 190 | struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); |
e6445719 | 191 | |
b637e498 | 192 | if (!ti) |
e6445719 PS |
193 | return NULL; |
194 | ||
b637e498 | 195 | ti->buckets = alloc_buckets(new_size); |
e6445719 | 196 | |
b637e498 PS |
197 | if (!ti->buckets) { |
198 | kfree(ti); | |
e6445719 PS |
199 | return NULL; |
200 | } | |
b637e498 PS |
201 | ti->n_buckets = new_size; |
202 | ti->node_ver = 0; | |
203 | ti->keep_flows = false; | |
204 | get_random_bytes(&ti->hash_seed, sizeof(u32)); | |
e6445719 | 205 | |
b637e498 | 206 | return ti; |
e6445719 PS |
207 | } |
208 | ||
b637e498 | 209 | int ovs_flow_tbl_init(struct flow_table *table) |
e6445719 | 210 | { |
74ed7ab9 | 211 | struct table_instance *ti, *ufid_ti; |
e6445719 | 212 | |
b637e498 | 213 | ti = table_instance_alloc(TBL_MIN_BUCKETS); |
e6445719 | 214 | |
b637e498 PS |
215 | if (!ti) |
216 | return -ENOMEM; | |
e6445719 | 217 | |
74ed7ab9 JS |
218 | ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
219 | if (!ufid_ti) | |
220 | goto free_ti; | |
221 | ||
b637e498 | 222 | rcu_assign_pointer(table->ti, ti); |
74ed7ab9 | 223 | rcu_assign_pointer(table->ufid_ti, ufid_ti); |
b637e498 PS |
224 | INIT_LIST_HEAD(&table->mask_list); |
225 | table->last_rehash = jiffies; | |
226 | table->count = 0; | |
74ed7ab9 | 227 | table->ufid_count = 0; |
b637e498 | 228 | return 0; |
74ed7ab9 JS |
229 | |
230 | free_ti: | |
231 | __table_instance_destroy(ti); | |
232 | return -ENOMEM; | |
e6445719 PS |
233 | } |
234 | ||
235 | static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) | |
236 | { | |
b637e498 | 237 | struct table_instance *ti = container_of(rcu, struct table_instance, rcu); |
e6445719 | 238 | |
b637e498 | 239 | __table_instance_destroy(ti); |
e6445719 PS |
240 | } |
241 | ||
74ed7ab9 JS |
242 | static void table_instance_destroy(struct table_instance *ti, |
243 | struct table_instance *ufid_ti, | |
244 | bool deferred) | |
e6445719 | 245 | { |
e80857cc AZ |
246 | int i; |
247 | ||
b637e498 | 248 | if (!ti) |
e6445719 PS |
249 | return; |
250 | ||
74ed7ab9 | 251 | BUG_ON(!ufid_ti); |
e80857cc AZ |
252 | if (ti->keep_flows) |
253 | goto skip_flows; | |
254 | ||
255 | for (i = 0; i < ti->n_buckets; i++) { | |
256 | struct sw_flow *flow; | |
257 | struct hlist_head *head = flex_array_get(ti->buckets, i); | |
258 | struct hlist_node *n; | |
259 | int ver = ti->node_ver; | |
74ed7ab9 | 260 | int ufid_ver = ufid_ti->node_ver; |
e80857cc | 261 | |
74ed7ab9 JS |
262 | hlist_for_each_entry_safe(flow, n, head, flow_table.node[ver]) { |
263 | hlist_del_rcu(&flow->flow_table.node[ver]); | |
264 | if (ovs_identifier_is_ufid(&flow->id)) | |
265 | hlist_del_rcu(&flow->ufid_table.node[ufid_ver]); | |
e80857cc AZ |
266 | ovs_flow_free(flow, deferred); |
267 | } | |
268 | } | |
269 | ||
270 | skip_flows: | |
74ed7ab9 | 271 | if (deferred) { |
b637e498 | 272 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
74ed7ab9 JS |
273 | call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb); |
274 | } else { | |
b637e498 | 275 | __table_instance_destroy(ti); |
74ed7ab9 JS |
276 | __table_instance_destroy(ufid_ti); |
277 | } | |
b637e498 PS |
278 | } |
279 | ||
9b996e54 PS |
280 | /* No need for locking this function is called from RCU callback or |
281 | * error path. | |
282 | */ | |
283 | void ovs_flow_tbl_destroy(struct flow_table *table) | |
b637e498 | 284 | { |
9b996e54 | 285 | struct table_instance *ti = rcu_dereference_raw(table->ti); |
74ed7ab9 | 286 | struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti); |
b637e498 | 287 | |
74ed7ab9 | 288 | table_instance_destroy(ti, ufid_ti, false); |
e6445719 PS |
289 | } |
290 | ||
b637e498 | 291 | struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, |
e6445719 PS |
292 | u32 *bucket, u32 *last) |
293 | { | |
294 | struct sw_flow *flow; | |
295 | struct hlist_head *head; | |
296 | int ver; | |
297 | int i; | |
298 | ||
b637e498 PS |
299 | ver = ti->node_ver; |
300 | while (*bucket < ti->n_buckets) { | |
e6445719 | 301 | i = 0; |
b637e498 | 302 | head = flex_array_get(ti->buckets, *bucket); |
74ed7ab9 | 303 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) { |
e6445719 PS |
304 | if (i < *last) { |
305 | i++; | |
306 | continue; | |
307 | } | |
308 | *last = i + 1; | |
309 | return flow; | |
310 | } | |
311 | (*bucket)++; | |
312 | *last = 0; | |
313 | } | |
314 | ||
315 | return NULL; | |
316 | } | |
317 | ||
b637e498 | 318 | static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) |
e6445719 | 319 | { |
b637e498 PS |
320 | hash = jhash_1word(hash, ti->hash_seed); |
321 | return flex_array_get(ti->buckets, | |
322 | (hash & (ti->n_buckets - 1))); | |
e6445719 PS |
323 | } |
324 | ||
74ed7ab9 JS |
325 | static void table_instance_insert(struct table_instance *ti, |
326 | struct sw_flow *flow) | |
e6445719 PS |
327 | { |
328 | struct hlist_head *head; | |
329 | ||
74ed7ab9 JS |
330 | head = find_bucket(ti, flow->flow_table.hash); |
331 | hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head); | |
332 | } | |
333 | ||
334 | static void ufid_table_instance_insert(struct table_instance *ti, | |
335 | struct sw_flow *flow) | |
336 | { | |
337 | struct hlist_head *head; | |
338 | ||
339 | head = find_bucket(ti, flow->ufid_table.hash); | |
340 | hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head); | |
e6445719 PS |
341 | } |
342 | ||
b637e498 | 343 | static void flow_table_copy_flows(struct table_instance *old, |
74ed7ab9 | 344 | struct table_instance *new, bool ufid) |
e6445719 PS |
345 | { |
346 | int old_ver; | |
347 | int i; | |
348 | ||
349 | old_ver = old->node_ver; | |
350 | new->node_ver = !old_ver; | |
351 | ||
352 | /* Insert in new table. */ | |
353 | for (i = 0; i < old->n_buckets; i++) { | |
354 | struct sw_flow *flow; | |
355 | struct hlist_head *head; | |
356 | ||
357 | head = flex_array_get(old->buckets, i); | |
358 | ||
74ed7ab9 JS |
359 | if (ufid) |
360 | hlist_for_each_entry(flow, head, | |
361 | ufid_table.node[old_ver]) | |
362 | ufid_table_instance_insert(new, flow); | |
363 | else | |
364 | hlist_for_each_entry(flow, head, | |
365 | flow_table.node[old_ver]) | |
366 | table_instance_insert(new, flow); | |
e6445719 PS |
367 | } |
368 | ||
e6445719 PS |
369 | old->keep_flows = true; |
370 | } | |
371 | ||
b637e498 | 372 | static struct table_instance *table_instance_rehash(struct table_instance *ti, |
74ed7ab9 | 373 | int n_buckets, bool ufid) |
e6445719 | 374 | { |
b637e498 | 375 | struct table_instance *new_ti; |
e6445719 | 376 | |
b637e498 PS |
377 | new_ti = table_instance_alloc(n_buckets); |
378 | if (!new_ti) | |
618ed0c8 | 379 | return NULL; |
e6445719 | 380 | |
74ed7ab9 | 381 | flow_table_copy_flows(ti, new_ti, ufid); |
e6445719 | 382 | |
b637e498 | 383 | return new_ti; |
e6445719 PS |
384 | } |
385 | ||
b637e498 | 386 | int ovs_flow_tbl_flush(struct flow_table *flow_table) |
e6445719 | 387 | { |
74ed7ab9 JS |
388 | struct table_instance *old_ti, *new_ti; |
389 | struct table_instance *old_ufid_ti, *new_ufid_ti; | |
e6445719 | 390 | |
b637e498 PS |
391 | new_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
392 | if (!new_ti) | |
393 | return -ENOMEM; | |
74ed7ab9 JS |
394 | new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
395 | if (!new_ufid_ti) | |
396 | goto err_free_ti; | |
397 | ||
398 | old_ti = ovsl_dereference(flow_table->ti); | |
399 | old_ufid_ti = ovsl_dereference(flow_table->ufid_ti); | |
b637e498 PS |
400 | |
401 | rcu_assign_pointer(flow_table->ti, new_ti); | |
74ed7ab9 | 402 | rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti); |
b637e498 PS |
403 | flow_table->last_rehash = jiffies; |
404 | flow_table->count = 0; | |
74ed7ab9 | 405 | flow_table->ufid_count = 0; |
b637e498 | 406 | |
74ed7ab9 | 407 | table_instance_destroy(old_ti, old_ufid_ti, true); |
b637e498 | 408 | return 0; |
74ed7ab9 JS |
409 | |
410 | err_free_ti: | |
411 | __table_instance_destroy(new_ti); | |
412 | return -ENOMEM; | |
e6445719 PS |
413 | } |
414 | ||
272c2cf8 JS |
415 | static u32 flow_hash(const struct sw_flow_key *key, |
416 | const struct sw_flow_key_range *range) | |
e6445719 | 417 | { |
272c2cf8 JS |
418 | int key_start = range->start; |
419 | int key_end = range->end; | |
7085130b | 420 | const u32 *hash_key = (const u32 *)((const u8 *)key + key_start); |
e6445719 PS |
421 | int hash_u32s = (key_end - key_start) >> 2; |
422 | ||
423 | /* Make sure number of hash bytes are multiple of u32. */ | |
424 | BUILD_BUG_ON(sizeof(long) % sizeof(u32)); | |
425 | ||
87545899 | 426 | return jhash2(hash_key, hash_u32s, 0); |
e6445719 PS |
427 | } |
428 | ||
429 | static int flow_key_start(const struct sw_flow_key *key) | |
430 | { | |
00a93bab | 431 | if (key->tun_proto) |
e6445719 PS |
432 | return 0; |
433 | else | |
434 | return rounddown(offsetof(struct sw_flow_key, phy), | |
435 | sizeof(long)); | |
436 | } | |
437 | ||
438 | static bool cmp_key(const struct sw_flow_key *key1, | |
439 | const struct sw_flow_key *key2, | |
440 | int key_start, int key_end) | |
441 | { | |
7085130b DDP |
442 | const long *cp1 = (const long *)((const u8 *)key1 + key_start); |
443 | const long *cp2 = (const long *)((const u8 *)key2 + key_start); | |
e6445719 PS |
444 | long diffs = 0; |
445 | int i; | |
446 | ||
447 | for (i = key_start; i < key_end; i += sizeof(long)) | |
448 | diffs |= *cp1++ ^ *cp2++; | |
449 | ||
450 | return diffs == 0; | |
451 | } | |
452 | ||
453 | static bool flow_cmp_masked_key(const struct sw_flow *flow, | |
454 | const struct sw_flow_key *key, | |
272c2cf8 | 455 | const struct sw_flow_key_range *range) |
e6445719 | 456 | { |
272c2cf8 | 457 | return cmp_key(&flow->key, key, range->start, range->end); |
e6445719 PS |
458 | } |
459 | ||
74ed7ab9 JS |
460 | static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, |
461 | const struct sw_flow_match *match) | |
e6445719 PS |
462 | { |
463 | struct sw_flow_key *key = match->key; | |
464 | int key_start = flow_key_start(key); | |
465 | int key_end = match->range.end; | |
466 | ||
74ed7ab9 JS |
467 | BUG_ON(ovs_identifier_is_ufid(&flow->id)); |
468 | return cmp_key(flow->id.unmasked_key, key, key_start, key_end); | |
e6445719 PS |
469 | } |
470 | ||
b637e498 | 471 | static struct sw_flow *masked_flow_lookup(struct table_instance *ti, |
e6445719 | 472 | const struct sw_flow_key *unmasked, |
12eb18f7 | 473 | const struct sw_flow_mask *mask) |
e6445719 PS |
474 | { |
475 | struct sw_flow *flow; | |
476 | struct hlist_head *head; | |
e6445719 PS |
477 | u32 hash; |
478 | struct sw_flow_key masked_key; | |
479 | ||
ae5f2fb1 | 480 | ovs_flow_mask_key(&masked_key, unmasked, false, mask); |
272c2cf8 | 481 | hash = flow_hash(&masked_key, &mask->range); |
b637e498 | 482 | head = find_bucket(ti, hash); |
74ed7ab9 JS |
483 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) { |
484 | if (flow->mask == mask && flow->flow_table.hash == hash && | |
272c2cf8 | 485 | flow_cmp_masked_key(flow, &masked_key, &mask->range)) |
e6445719 PS |
486 | return flow; |
487 | } | |
488 | return NULL; | |
489 | } | |
490 | ||
5bb50632 | 491 | struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl, |
1bd7116f AZ |
492 | const struct sw_flow_key *key, |
493 | u32 *n_mask_hit) | |
e6445719 | 494 | { |
663efa36 | 495 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); |
e6445719 | 496 | struct sw_flow_mask *mask; |
b637e498 | 497 | struct sw_flow *flow; |
e6445719 | 498 | |
1bd7116f | 499 | *n_mask_hit = 0; |
b637e498 | 500 | list_for_each_entry_rcu(mask, &tbl->mask_list, list) { |
1bd7116f | 501 | (*n_mask_hit)++; |
b637e498 | 502 | flow = masked_flow_lookup(ti, key, mask); |
e6445719 | 503 | if (flow) /* Found */ |
b637e498 | 504 | return flow; |
e6445719 | 505 | } |
b637e498 PS |
506 | return NULL; |
507 | } | |
e6445719 | 508 | |
5bb50632 AZ |
509 | struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl, |
510 | const struct sw_flow_key *key) | |
511 | { | |
512 | u32 __always_unused n_mask_hit; | |
513 | ||
514 | return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit); | |
515 | } | |
516 | ||
4a46b24e | 517 | struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl, |
12eb18f7 | 518 | const struct sw_flow_match *match) |
4a46b24e AW |
519 | { |
520 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); | |
521 | struct sw_flow_mask *mask; | |
522 | struct sw_flow *flow; | |
523 | ||
524 | /* Always called under ovs-mutex. */ | |
525 | list_for_each_entry(mask, &tbl->mask_list, list) { | |
526 | flow = masked_flow_lookup(ti, match->key, mask); | |
74ed7ab9 JS |
527 | if (flow && ovs_identifier_is_key(&flow->id) && |
528 | ovs_flow_cmp_unmasked_key(flow, match)) | |
529 | return flow; | |
530 | } | |
531 | return NULL; | |
532 | } | |
533 | ||
534 | static u32 ufid_hash(const struct sw_flow_id *sfid) | |
535 | { | |
536 | return jhash(sfid->ufid, sfid->ufid_len, 0); | |
537 | } | |
538 | ||
539 | static bool ovs_flow_cmp_ufid(const struct sw_flow *flow, | |
540 | const struct sw_flow_id *sfid) | |
541 | { | |
542 | if (flow->id.ufid_len != sfid->ufid_len) | |
543 | return false; | |
544 | ||
545 | return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len); | |
546 | } | |
547 | ||
548 | bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match) | |
549 | { | |
550 | if (ovs_identifier_is_ufid(&flow->id)) | |
551 | return flow_cmp_masked_key(flow, match->key, &match->range); | |
552 | ||
553 | return ovs_flow_cmp_unmasked_key(flow, match); | |
554 | } | |
555 | ||
556 | struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl, | |
557 | const struct sw_flow_id *ufid) | |
558 | { | |
559 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti); | |
560 | struct sw_flow *flow; | |
561 | struct hlist_head *head; | |
562 | u32 hash; | |
563 | ||
564 | hash = ufid_hash(ufid); | |
565 | head = find_bucket(ti, hash); | |
566 | hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) { | |
567 | if (flow->ufid_table.hash == hash && | |
568 | ovs_flow_cmp_ufid(flow, ufid)) | |
4a46b24e AW |
569 | return flow; |
570 | } | |
571 | return NULL; | |
572 | } | |
573 | ||
1bd7116f AZ |
574 | int ovs_flow_tbl_num_masks(const struct flow_table *table) |
575 | { | |
576 | struct sw_flow_mask *mask; | |
577 | int num = 0; | |
578 | ||
579 | list_for_each_entry(mask, &table->mask_list, list) | |
580 | num++; | |
581 | ||
582 | return num; | |
583 | } | |
584 | ||
74ed7ab9 JS |
585 | static struct table_instance *table_instance_expand(struct table_instance *ti, |
586 | bool ufid) | |
b637e498 | 587 | { |
74ed7ab9 | 588 | return table_instance_rehash(ti, ti->n_buckets * 2, ufid); |
e6445719 PS |
589 | } |
590 | ||
56c19868 JR |
591 | /* Remove 'mask' from the mask list, if it is not needed any more. */ |
592 | static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask) | |
593 | { | |
594 | if (mask) { | |
595 | /* ovs-lock is required to protect mask-refcount and | |
596 | * mask list. | |
597 | */ | |
598 | ASSERT_OVSL(); | |
599 | BUG_ON(!mask->ref_count); | |
600 | mask->ref_count--; | |
601 | ||
602 | if (!mask->ref_count) { | |
603 | list_del_rcu(&mask->list); | |
604 | kfree_rcu(mask, rcu); | |
605 | } | |
606 | } | |
607 | } | |
608 | ||
609 | /* Must be called with OVS mutex held. */ | |
e6445719 PS |
610 | void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) |
611 | { | |
b637e498 | 612 | struct table_instance *ti = ovsl_dereference(table->ti); |
74ed7ab9 | 613 | struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti); |
b637e498 | 614 | |
e6445719 | 615 | BUG_ON(table->count == 0); |
74ed7ab9 | 616 | hlist_del_rcu(&flow->flow_table.node[ti->node_ver]); |
e6445719 | 617 | table->count--; |
74ed7ab9 JS |
618 | if (ovs_identifier_is_ufid(&flow->id)) { |
619 | hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]); | |
620 | table->ufid_count--; | |
621 | } | |
56c19868 JR |
622 | |
623 | /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be | |
624 | * accessible as long as the RCU read lock is held. | |
625 | */ | |
626 | flow_mask_remove(table, flow->mask); | |
e6445719 PS |
627 | } |
628 | ||
618ed0c8 | 629 | static struct sw_flow_mask *mask_alloc(void) |
e6445719 PS |
630 | { |
631 | struct sw_flow_mask *mask; | |
632 | ||
633 | mask = kmalloc(sizeof(*mask), GFP_KERNEL); | |
634 | if (mask) | |
e80857cc | 635 | mask->ref_count = 1; |
e6445719 PS |
636 | |
637 | return mask; | |
638 | } | |
639 | ||
e6445719 PS |
640 | static bool mask_equal(const struct sw_flow_mask *a, |
641 | const struct sw_flow_mask *b) | |
642 | { | |
7085130b DDP |
643 | const u8 *a_ = (const u8 *)&a->key + a->range.start; |
644 | const u8 *b_ = (const u8 *)&b->key + b->range.start; | |
e6445719 PS |
645 | |
646 | return (a->range.end == b->range.end) | |
647 | && (a->range.start == b->range.start) | |
648 | && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0); | |
649 | } | |
650 | ||
618ed0c8 | 651 | static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl, |
e6445719 PS |
652 | const struct sw_flow_mask *mask) |
653 | { | |
654 | struct list_head *ml; | |
655 | ||
b637e498 | 656 | list_for_each(ml, &tbl->mask_list) { |
e6445719 PS |
657 | struct sw_flow_mask *m; |
658 | m = container_of(ml, struct sw_flow_mask, list); | |
659 | if (mask_equal(mask, m)) | |
660 | return m; | |
661 | } | |
662 | ||
663 | return NULL; | |
664 | } | |
665 | ||
d1211908 | 666 | /* Add 'mask' into the mask list, if it is not already there. */ |
618ed0c8 | 667 | static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow, |
12eb18f7 | 668 | const struct sw_flow_mask *new) |
618ed0c8 PS |
669 | { |
670 | struct sw_flow_mask *mask; | |
671 | mask = flow_mask_find(tbl, new); | |
672 | if (!mask) { | |
673 | /* Allocate a new mask if none exsits. */ | |
674 | mask = mask_alloc(); | |
675 | if (!mask) | |
676 | return -ENOMEM; | |
677 | mask->key = new->key; | |
678 | mask->range = new->range; | |
679 | list_add_rcu(&mask->list, &tbl->mask_list); | |
e80857cc AZ |
680 | } else { |
681 | BUG_ON(!mask->ref_count); | |
682 | mask->ref_count++; | |
618ed0c8 PS |
683 | } |
684 | ||
618ed0c8 PS |
685 | flow->mask = mask; |
686 | return 0; | |
687 | } | |
688 | ||
56c19868 | 689 | /* Must be called with OVS mutex held. */ |
d29ab6f8 | 690 | static void flow_key_insert(struct flow_table *table, struct sw_flow *flow) |
e6445719 | 691 | { |
618ed0c8 PS |
692 | struct table_instance *new_ti = NULL; |
693 | struct table_instance *ti; | |
618ed0c8 | 694 | |
74ed7ab9 | 695 | flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range); |
618ed0c8 PS |
696 | ti = ovsl_dereference(table->ti); |
697 | table_instance_insert(ti, flow); | |
698 | table->count++; | |
699 | ||
700 | /* Expand table, if necessary, to make room. */ | |
701 | if (table->count > ti->n_buckets) | |
74ed7ab9 | 702 | new_ti = table_instance_expand(ti, false); |
618ed0c8 | 703 | else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) |
74ed7ab9 | 704 | new_ti = table_instance_rehash(ti, ti->n_buckets, false); |
618ed0c8 PS |
705 | |
706 | if (new_ti) { | |
707 | rcu_assign_pointer(table->ti, new_ti); | |
74ed7ab9 | 708 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
618ed0c8 PS |
709 | table->last_rehash = jiffies; |
710 | } | |
d29ab6f8 JS |
711 | } |
712 | ||
74ed7ab9 JS |
713 | /* Must be called with OVS mutex held. */ |
714 | static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow) | |
715 | { | |
716 | struct table_instance *ti; | |
717 | ||
718 | flow->ufid_table.hash = ufid_hash(&flow->id); | |
719 | ti = ovsl_dereference(table->ufid_ti); | |
720 | ufid_table_instance_insert(ti, flow); | |
721 | table->ufid_count++; | |
722 | ||
723 | /* Expand table, if necessary, to make room. */ | |
724 | if (table->ufid_count > ti->n_buckets) { | |
725 | struct table_instance *new_ti; | |
726 | ||
727 | new_ti = table_instance_expand(ti, true); | |
728 | if (new_ti) { | |
729 | rcu_assign_pointer(table->ufid_ti, new_ti); | |
730 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); | |
731 | } | |
732 | } | |
733 | } | |
734 | ||
d29ab6f8 JS |
735 | /* Must be called with OVS mutex held. */ |
736 | int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow, | |
737 | const struct sw_flow_mask *mask) | |
738 | { | |
739 | int err; | |
740 | ||
741 | err = flow_mask_insert(table, flow, mask); | |
742 | if (err) | |
743 | return err; | |
744 | flow_key_insert(table, flow); | |
74ed7ab9 JS |
745 | if (ovs_identifier_is_ufid(&flow->id)) |
746 | flow_ufid_insert(table, flow); | |
d29ab6f8 | 747 | |
618ed0c8 | 748 | return 0; |
e6445719 PS |
749 | } |
750 | ||
751 | /* Initializes the flow module. | |
752 | * Returns zero if successful or a negative error code. */ | |
753 | int ovs_flow_init(void) | |
754 | { | |
755 | BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); | |
756 | BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); | |
757 | ||
63e7959c | 758 | flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow) |
bac541e4 | 759 | + (nr_node_ids |
63e7959c JR |
760 | * sizeof(struct flow_stats *)), |
761 | 0, 0, NULL); | |
e6445719 PS |
762 | if (flow_cache == NULL) |
763 | return -ENOMEM; | |
764 | ||
63e7959c JR |
765 | flow_stats_cache |
766 | = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats), | |
767 | 0, SLAB_HWCACHE_ALIGN, NULL); | |
768 | if (flow_stats_cache == NULL) { | |
769 | kmem_cache_destroy(flow_cache); | |
770 | flow_cache = NULL; | |
771 | return -ENOMEM; | |
772 | } | |
773 | ||
e6445719 PS |
774 | return 0; |
775 | } | |
776 | ||
777 | /* Uninitializes the flow module. */ | |
778 | void ovs_flow_exit(void) | |
779 | { | |
63e7959c | 780 | kmem_cache_destroy(flow_stats_cache); |
e6445719 PS |
781 | kmem_cache_destroy(flow_cache); |
782 | } |