Commit | Line | Data |
---|---|---|
c9422999 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
e6445719 | 2 | /* |
9b996e54 | 3 | * Copyright (c) 2007-2014 Nicira, Inc. |
e6445719 PS |
4 | */ |
5 | ||
6 | #include "flow.h" | |
7 | #include "datapath.h" | |
34ae932a | 8 | #include "flow_netlink.h" |
e6445719 PS |
9 | #include <linux/uaccess.h> |
10 | #include <linux/netdevice.h> | |
11 | #include <linux/etherdevice.h> | |
12 | #include <linux/if_ether.h> | |
13 | #include <linux/if_vlan.h> | |
14 | #include <net/llc_pdu.h> | |
15 | #include <linux/kernel.h> | |
87545899 | 16 | #include <linux/jhash.h> |
e6445719 PS |
17 | #include <linux/jiffies.h> |
18 | #include <linux/llc.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/in.h> | |
21 | #include <linux/rcupdate.h> | |
db74a333 | 22 | #include <linux/cpumask.h> |
e6445719 PS |
23 | #include <linux/if_arp.h> |
24 | #include <linux/ip.h> | |
25 | #include <linux/ipv6.h> | |
26 | #include <linux/sctp.h> | |
27 | #include <linux/tcp.h> | |
28 | #include <linux/udp.h> | |
29 | #include <linux/icmp.h> | |
30 | #include <linux/icmpv6.h> | |
31 | #include <linux/rculist.h> | |
32 | #include <net/ip.h> | |
33 | #include <net/ipv6.h> | |
34 | #include <net/ndisc.h> | |
35 | ||
b637e498 PS |
36 | #define TBL_MIN_BUCKETS 1024 |
37 | #define REHASH_INTERVAL (10 * 60 * HZ) | |
38 | ||
e6445719 | 39 | static struct kmem_cache *flow_cache; |
63e7959c | 40 | struct kmem_cache *flow_stats_cache __read_mostly; |
e6445719 PS |
41 | |
42 | static u16 range_n_bytes(const struct sw_flow_key_range *range) | |
43 | { | |
44 | return range->end - range->start; | |
45 | } | |
46 | ||
47 | void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, | |
ae5f2fb1 | 48 | bool full, const struct sw_flow_mask *mask) |
e6445719 | 49 | { |
ae5f2fb1 JG |
50 | int start = full ? 0 : mask->range.start; |
51 | int len = full ? sizeof *dst : range_n_bytes(&mask->range); | |
52 | const long *m = (const long *)((const u8 *)&mask->key + start); | |
53 | const long *s = (const long *)((const u8 *)src + start); | |
54 | long *d = (long *)((u8 *)dst + start); | |
e6445719 PS |
55 | int i; |
56 | ||
ae5f2fb1 JG |
57 | /* If 'full' is true then all of 'dst' is fully initialized. Otherwise, |
58 | * if 'full' is false the memory outside of the 'mask->range' is left | |
59 | * uninitialized. This can be used as an optimization when further | |
60 | * operations on 'dst' only use contents within 'mask->range'. | |
e6445719 | 61 | */ |
ae5f2fb1 | 62 | for (i = 0; i < len; i += sizeof(long)) |
e6445719 PS |
63 | *d++ = *s++ & *m++; |
64 | } | |
65 | ||
23dabf88 | 66 | struct sw_flow *ovs_flow_alloc(void) |
e6445719 PS |
67 | { |
68 | struct sw_flow *flow; | |
aef833c5 | 69 | struct sw_flow_stats *stats; |
e6445719 | 70 | |
db74a333 | 71 | flow = kmem_cache_zalloc(flow_cache, GFP_KERNEL); |
e6445719 PS |
72 | if (!flow) |
73 | return ERR_PTR(-ENOMEM); | |
74 | ||
db74a333 | 75 | flow->stats_last_writer = -1; |
e6445719 | 76 | |
63e7959c JR |
77 | /* Initialize the default stat node. */ |
78 | stats = kmem_cache_alloc_node(flow_stats_cache, | |
598c12d0 KK |
79 | GFP_KERNEL | __GFP_ZERO, |
80 | node_online(0) ? 0 : NUMA_NO_NODE); | |
63e7959c | 81 | if (!stats) |
23dabf88 | 82 | goto err; |
e298e505 | 83 | |
63e7959c JR |
84 | spin_lock_init(&stats->lock); |
85 | ||
86 | RCU_INIT_POINTER(flow->stats[0], stats); | |
87 | ||
c4b2bf6b TZ |
88 | cpumask_set_cpu(0, &flow->cpu_used_mask); |
89 | ||
e6445719 | 90 | return flow; |
e298e505 | 91 | err: |
ece37c87 | 92 | kmem_cache_free(flow_cache, flow); |
e298e505 | 93 | return ERR_PTR(-ENOMEM); |
e6445719 PS |
94 | } |
95 | ||
12eb18f7 | 96 | int ovs_flow_tbl_count(const struct flow_table *table) |
b637e498 PS |
97 | { |
98 | return table->count; | |
99 | } | |
100 | ||
e6445719 PS |
101 | static void flow_free(struct sw_flow *flow) |
102 | { | |
db74a333 | 103 | int cpu; |
63e7959c | 104 | |
74ed7ab9 JS |
105 | if (ovs_identifier_is_key(&flow->id)) |
106 | kfree(flow->id.unmasked_key); | |
34ae932a TG |
107 | if (flow->sf_acts) |
108 | ovs_nla_free_flow_actions((struct sw_flow_actions __force *)flow->sf_acts); | |
db74a333 | 109 | /* We open code this to make sure cpu 0 is always considered */ |
c4b2bf6b | 110 | for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) |
db74a333 | 111 | if (flow->stats[cpu]) |
63e7959c | 112 | kmem_cache_free(flow_stats_cache, |
aef833c5 | 113 | (struct sw_flow_stats __force *)flow->stats[cpu]); |
e6445719 PS |
114 | kmem_cache_free(flow_cache, flow); |
115 | } | |
116 | ||
117 | static void rcu_free_flow_callback(struct rcu_head *rcu) | |
118 | { | |
119 | struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); | |
120 | ||
121 | flow_free(flow); | |
122 | } | |
123 | ||
e80857cc | 124 | void ovs_flow_free(struct sw_flow *flow, bool deferred) |
618ed0c8 | 125 | { |
e80857cc | 126 | if (!flow) |
618ed0c8 PS |
127 | return; |
128 | ||
e6445719 PS |
129 | if (deferred) |
130 | call_rcu(&flow->rcu, rcu_free_flow_callback); | |
131 | else | |
132 | flow_free(flow); | |
133 | } | |
134 | ||
b637e498 | 135 | static void __table_instance_destroy(struct table_instance *ti) |
e6445719 | 136 | { |
ee9c5e67 | 137 | kvfree(ti->buckets); |
b637e498 | 138 | kfree(ti); |
e6445719 PS |
139 | } |
140 | ||
b637e498 | 141 | static struct table_instance *table_instance_alloc(int new_size) |
e6445719 | 142 | { |
b637e498 | 143 | struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); |
ee9c5e67 | 144 | int i; |
e6445719 | 145 | |
b637e498 | 146 | if (!ti) |
e6445719 PS |
147 | return NULL; |
148 | ||
ee9c5e67 KO |
149 | ti->buckets = kvmalloc_array(new_size, sizeof(struct hlist_head), |
150 | GFP_KERNEL); | |
b637e498 PS |
151 | if (!ti->buckets) { |
152 | kfree(ti); | |
e6445719 PS |
153 | return NULL; |
154 | } | |
ee9c5e67 KO |
155 | |
156 | for (i = 0; i < new_size; i++) | |
157 | INIT_HLIST_HEAD(&ti->buckets[i]); | |
158 | ||
b637e498 PS |
159 | ti->n_buckets = new_size; |
160 | ti->node_ver = 0; | |
161 | ti->keep_flows = false; | |
162 | get_random_bytes(&ti->hash_seed, sizeof(u32)); | |
e6445719 | 163 | |
b637e498 | 164 | return ti; |
e6445719 PS |
165 | } |
166 | ||
b637e498 | 167 | int ovs_flow_tbl_init(struct flow_table *table) |
e6445719 | 168 | { |
74ed7ab9 | 169 | struct table_instance *ti, *ufid_ti; |
e6445719 | 170 | |
b637e498 | 171 | ti = table_instance_alloc(TBL_MIN_BUCKETS); |
e6445719 | 172 | |
b637e498 PS |
173 | if (!ti) |
174 | return -ENOMEM; | |
e6445719 | 175 | |
74ed7ab9 JS |
176 | ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
177 | if (!ufid_ti) | |
178 | goto free_ti; | |
179 | ||
b637e498 | 180 | rcu_assign_pointer(table->ti, ti); |
74ed7ab9 | 181 | rcu_assign_pointer(table->ufid_ti, ufid_ti); |
b637e498 PS |
182 | INIT_LIST_HEAD(&table->mask_list); |
183 | table->last_rehash = jiffies; | |
184 | table->count = 0; | |
74ed7ab9 | 185 | table->ufid_count = 0; |
b637e498 | 186 | return 0; |
74ed7ab9 JS |
187 | |
188 | free_ti: | |
189 | __table_instance_destroy(ti); | |
190 | return -ENOMEM; | |
e6445719 PS |
191 | } |
192 | ||
193 | static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) | |
194 | { | |
b637e498 | 195 | struct table_instance *ti = container_of(rcu, struct table_instance, rcu); |
e6445719 | 196 | |
b637e498 | 197 | __table_instance_destroy(ti); |
e6445719 PS |
198 | } |
199 | ||
74ed7ab9 JS |
200 | static void table_instance_destroy(struct table_instance *ti, |
201 | struct table_instance *ufid_ti, | |
202 | bool deferred) | |
e6445719 | 203 | { |
e80857cc AZ |
204 | int i; |
205 | ||
b637e498 | 206 | if (!ti) |
e6445719 PS |
207 | return; |
208 | ||
74ed7ab9 | 209 | BUG_ON(!ufid_ti); |
e80857cc AZ |
210 | if (ti->keep_flows) |
211 | goto skip_flows; | |
212 | ||
213 | for (i = 0; i < ti->n_buckets; i++) { | |
214 | struct sw_flow *flow; | |
ee9c5e67 | 215 | struct hlist_head *head = &ti->buckets[i]; |
e80857cc AZ |
216 | struct hlist_node *n; |
217 | int ver = ti->node_ver; | |
74ed7ab9 | 218 | int ufid_ver = ufid_ti->node_ver; |
e80857cc | 219 | |
74ed7ab9 JS |
220 | hlist_for_each_entry_safe(flow, n, head, flow_table.node[ver]) { |
221 | hlist_del_rcu(&flow->flow_table.node[ver]); | |
222 | if (ovs_identifier_is_ufid(&flow->id)) | |
223 | hlist_del_rcu(&flow->ufid_table.node[ufid_ver]); | |
e80857cc AZ |
224 | ovs_flow_free(flow, deferred); |
225 | } | |
226 | } | |
227 | ||
228 | skip_flows: | |
74ed7ab9 | 229 | if (deferred) { |
b637e498 | 230 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
74ed7ab9 JS |
231 | call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb); |
232 | } else { | |
b637e498 | 233 | __table_instance_destroy(ti); |
74ed7ab9 JS |
234 | __table_instance_destroy(ufid_ti); |
235 | } | |
b637e498 PS |
236 | } |
237 | ||
9b996e54 PS |
238 | /* No need for locking this function is called from RCU callback or |
239 | * error path. | |
240 | */ | |
241 | void ovs_flow_tbl_destroy(struct flow_table *table) | |
b637e498 | 242 | { |
9b996e54 | 243 | struct table_instance *ti = rcu_dereference_raw(table->ti); |
74ed7ab9 | 244 | struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti); |
b637e498 | 245 | |
74ed7ab9 | 246 | table_instance_destroy(ti, ufid_ti, false); |
e6445719 PS |
247 | } |
248 | ||
b637e498 | 249 | struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, |
e6445719 PS |
250 | u32 *bucket, u32 *last) |
251 | { | |
252 | struct sw_flow *flow; | |
253 | struct hlist_head *head; | |
254 | int ver; | |
255 | int i; | |
256 | ||
b637e498 PS |
257 | ver = ti->node_ver; |
258 | while (*bucket < ti->n_buckets) { | |
e6445719 | 259 | i = 0; |
ee9c5e67 | 260 | head = &ti->buckets[*bucket]; |
74ed7ab9 | 261 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) { |
e6445719 PS |
262 | if (i < *last) { |
263 | i++; | |
264 | continue; | |
265 | } | |
266 | *last = i + 1; | |
267 | return flow; | |
268 | } | |
269 | (*bucket)++; | |
270 | *last = 0; | |
271 | } | |
272 | ||
273 | return NULL; | |
274 | } | |
275 | ||
b637e498 | 276 | static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) |
e6445719 | 277 | { |
b637e498 | 278 | hash = jhash_1word(hash, ti->hash_seed); |
ee9c5e67 | 279 | return &ti->buckets[hash & (ti->n_buckets - 1)]; |
e6445719 PS |
280 | } |
281 | ||
74ed7ab9 JS |
282 | static void table_instance_insert(struct table_instance *ti, |
283 | struct sw_flow *flow) | |
e6445719 PS |
284 | { |
285 | struct hlist_head *head; | |
286 | ||
74ed7ab9 JS |
287 | head = find_bucket(ti, flow->flow_table.hash); |
288 | hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head); | |
289 | } | |
290 | ||
291 | static void ufid_table_instance_insert(struct table_instance *ti, | |
292 | struct sw_flow *flow) | |
293 | { | |
294 | struct hlist_head *head; | |
295 | ||
296 | head = find_bucket(ti, flow->ufid_table.hash); | |
297 | hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head); | |
e6445719 PS |
298 | } |
299 | ||
b637e498 | 300 | static void flow_table_copy_flows(struct table_instance *old, |
74ed7ab9 | 301 | struct table_instance *new, bool ufid) |
e6445719 PS |
302 | { |
303 | int old_ver; | |
304 | int i; | |
305 | ||
306 | old_ver = old->node_ver; | |
307 | new->node_ver = !old_ver; | |
308 | ||
309 | /* Insert in new table. */ | |
310 | for (i = 0; i < old->n_buckets; i++) { | |
311 | struct sw_flow *flow; | |
ee9c5e67 | 312 | struct hlist_head *head = &old->buckets[i]; |
e6445719 | 313 | |
74ed7ab9 JS |
314 | if (ufid) |
315 | hlist_for_each_entry(flow, head, | |
316 | ufid_table.node[old_ver]) | |
317 | ufid_table_instance_insert(new, flow); | |
318 | else | |
319 | hlist_for_each_entry(flow, head, | |
320 | flow_table.node[old_ver]) | |
321 | table_instance_insert(new, flow); | |
e6445719 PS |
322 | } |
323 | ||
e6445719 PS |
324 | old->keep_flows = true; |
325 | } | |
326 | ||
b637e498 | 327 | static struct table_instance *table_instance_rehash(struct table_instance *ti, |
74ed7ab9 | 328 | int n_buckets, bool ufid) |
e6445719 | 329 | { |
b637e498 | 330 | struct table_instance *new_ti; |
e6445719 | 331 | |
b637e498 PS |
332 | new_ti = table_instance_alloc(n_buckets); |
333 | if (!new_ti) | |
618ed0c8 | 334 | return NULL; |
e6445719 | 335 | |
74ed7ab9 | 336 | flow_table_copy_flows(ti, new_ti, ufid); |
e6445719 | 337 | |
b637e498 | 338 | return new_ti; |
e6445719 PS |
339 | } |
340 | ||
b637e498 | 341 | int ovs_flow_tbl_flush(struct flow_table *flow_table) |
e6445719 | 342 | { |
74ed7ab9 JS |
343 | struct table_instance *old_ti, *new_ti; |
344 | struct table_instance *old_ufid_ti, *new_ufid_ti; | |
e6445719 | 345 | |
b637e498 PS |
346 | new_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
347 | if (!new_ti) | |
348 | return -ENOMEM; | |
74ed7ab9 JS |
349 | new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
350 | if (!new_ufid_ti) | |
351 | goto err_free_ti; | |
352 | ||
353 | old_ti = ovsl_dereference(flow_table->ti); | |
354 | old_ufid_ti = ovsl_dereference(flow_table->ufid_ti); | |
b637e498 PS |
355 | |
356 | rcu_assign_pointer(flow_table->ti, new_ti); | |
74ed7ab9 | 357 | rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti); |
b637e498 PS |
358 | flow_table->last_rehash = jiffies; |
359 | flow_table->count = 0; | |
74ed7ab9 | 360 | flow_table->ufid_count = 0; |
b637e498 | 361 | |
74ed7ab9 | 362 | table_instance_destroy(old_ti, old_ufid_ti, true); |
b637e498 | 363 | return 0; |
74ed7ab9 JS |
364 | |
365 | err_free_ti: | |
366 | __table_instance_destroy(new_ti); | |
367 | return -ENOMEM; | |
e6445719 PS |
368 | } |
369 | ||
272c2cf8 JS |
370 | static u32 flow_hash(const struct sw_flow_key *key, |
371 | const struct sw_flow_key_range *range) | |
e6445719 | 372 | { |
272c2cf8 JS |
373 | int key_start = range->start; |
374 | int key_end = range->end; | |
7085130b | 375 | const u32 *hash_key = (const u32 *)((const u8 *)key + key_start); |
e6445719 PS |
376 | int hash_u32s = (key_end - key_start) >> 2; |
377 | ||
378 | /* Make sure number of hash bytes are multiple of u32. */ | |
379 | BUILD_BUG_ON(sizeof(long) % sizeof(u32)); | |
380 | ||
87545899 | 381 | return jhash2(hash_key, hash_u32s, 0); |
e6445719 PS |
382 | } |
383 | ||
384 | static int flow_key_start(const struct sw_flow_key *key) | |
385 | { | |
00a93bab | 386 | if (key->tun_proto) |
e6445719 PS |
387 | return 0; |
388 | else | |
389 | return rounddown(offsetof(struct sw_flow_key, phy), | |
390 | sizeof(long)); | |
391 | } | |
392 | ||
393 | static bool cmp_key(const struct sw_flow_key *key1, | |
394 | const struct sw_flow_key *key2, | |
395 | int key_start, int key_end) | |
396 | { | |
7085130b DDP |
397 | const long *cp1 = (const long *)((const u8 *)key1 + key_start); |
398 | const long *cp2 = (const long *)((const u8 *)key2 + key_start); | |
e6445719 PS |
399 | long diffs = 0; |
400 | int i; | |
401 | ||
402 | for (i = key_start; i < key_end; i += sizeof(long)) | |
403 | diffs |= *cp1++ ^ *cp2++; | |
404 | ||
405 | return diffs == 0; | |
406 | } | |
407 | ||
408 | static bool flow_cmp_masked_key(const struct sw_flow *flow, | |
409 | const struct sw_flow_key *key, | |
272c2cf8 | 410 | const struct sw_flow_key_range *range) |
e6445719 | 411 | { |
272c2cf8 | 412 | return cmp_key(&flow->key, key, range->start, range->end); |
e6445719 PS |
413 | } |
414 | ||
74ed7ab9 JS |
415 | static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, |
416 | const struct sw_flow_match *match) | |
e6445719 PS |
417 | { |
418 | struct sw_flow_key *key = match->key; | |
419 | int key_start = flow_key_start(key); | |
420 | int key_end = match->range.end; | |
421 | ||
74ed7ab9 JS |
422 | BUG_ON(ovs_identifier_is_ufid(&flow->id)); |
423 | return cmp_key(flow->id.unmasked_key, key, key_start, key_end); | |
e6445719 PS |
424 | } |
425 | ||
b637e498 | 426 | static struct sw_flow *masked_flow_lookup(struct table_instance *ti, |
e6445719 | 427 | const struct sw_flow_key *unmasked, |
12eb18f7 | 428 | const struct sw_flow_mask *mask) |
e6445719 PS |
429 | { |
430 | struct sw_flow *flow; | |
431 | struct hlist_head *head; | |
e6445719 PS |
432 | u32 hash; |
433 | struct sw_flow_key masked_key; | |
434 | ||
ae5f2fb1 | 435 | ovs_flow_mask_key(&masked_key, unmasked, false, mask); |
272c2cf8 | 436 | hash = flow_hash(&masked_key, &mask->range); |
b637e498 | 437 | head = find_bucket(ti, hash); |
74ed7ab9 JS |
438 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) { |
439 | if (flow->mask == mask && flow->flow_table.hash == hash && | |
272c2cf8 | 440 | flow_cmp_masked_key(flow, &masked_key, &mask->range)) |
e6445719 PS |
441 | return flow; |
442 | } | |
443 | return NULL; | |
444 | } | |
445 | ||
5bb50632 | 446 | struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl, |
1bd7116f AZ |
447 | const struct sw_flow_key *key, |
448 | u32 *n_mask_hit) | |
e6445719 | 449 | { |
663efa36 | 450 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); |
e6445719 | 451 | struct sw_flow_mask *mask; |
b637e498 | 452 | struct sw_flow *flow; |
e6445719 | 453 | |
1bd7116f | 454 | *n_mask_hit = 0; |
b637e498 | 455 | list_for_each_entry_rcu(mask, &tbl->mask_list, list) { |
1bd7116f | 456 | (*n_mask_hit)++; |
b637e498 | 457 | flow = masked_flow_lookup(ti, key, mask); |
e6445719 | 458 | if (flow) /* Found */ |
b637e498 | 459 | return flow; |
e6445719 | 460 | } |
b637e498 PS |
461 | return NULL; |
462 | } | |
e6445719 | 463 | |
5bb50632 AZ |
464 | struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl, |
465 | const struct sw_flow_key *key) | |
466 | { | |
467 | u32 __always_unused n_mask_hit; | |
468 | ||
469 | return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit); | |
470 | } | |
471 | ||
4a46b24e | 472 | struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl, |
12eb18f7 | 473 | const struct sw_flow_match *match) |
4a46b24e AW |
474 | { |
475 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); | |
476 | struct sw_flow_mask *mask; | |
477 | struct sw_flow *flow; | |
478 | ||
479 | /* Always called under ovs-mutex. */ | |
480 | list_for_each_entry(mask, &tbl->mask_list, list) { | |
481 | flow = masked_flow_lookup(ti, match->key, mask); | |
74ed7ab9 JS |
482 | if (flow && ovs_identifier_is_key(&flow->id) && |
483 | ovs_flow_cmp_unmasked_key(flow, match)) | |
484 | return flow; | |
485 | } | |
486 | return NULL; | |
487 | } | |
488 | ||
489 | static u32 ufid_hash(const struct sw_flow_id *sfid) | |
490 | { | |
491 | return jhash(sfid->ufid, sfid->ufid_len, 0); | |
492 | } | |
493 | ||
494 | static bool ovs_flow_cmp_ufid(const struct sw_flow *flow, | |
495 | const struct sw_flow_id *sfid) | |
496 | { | |
497 | if (flow->id.ufid_len != sfid->ufid_len) | |
498 | return false; | |
499 | ||
500 | return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len); | |
501 | } | |
502 | ||
503 | bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match) | |
504 | { | |
505 | if (ovs_identifier_is_ufid(&flow->id)) | |
506 | return flow_cmp_masked_key(flow, match->key, &match->range); | |
507 | ||
508 | return ovs_flow_cmp_unmasked_key(flow, match); | |
509 | } | |
510 | ||
511 | struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl, | |
512 | const struct sw_flow_id *ufid) | |
513 | { | |
514 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti); | |
515 | struct sw_flow *flow; | |
516 | struct hlist_head *head; | |
517 | u32 hash; | |
518 | ||
519 | hash = ufid_hash(ufid); | |
520 | head = find_bucket(ti, hash); | |
521 | hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver]) { | |
522 | if (flow->ufid_table.hash == hash && | |
523 | ovs_flow_cmp_ufid(flow, ufid)) | |
4a46b24e AW |
524 | return flow; |
525 | } | |
526 | return NULL; | |
527 | } | |
528 | ||
1bd7116f AZ |
529 | int ovs_flow_tbl_num_masks(const struct flow_table *table) |
530 | { | |
531 | struct sw_flow_mask *mask; | |
532 | int num = 0; | |
533 | ||
534 | list_for_each_entry(mask, &table->mask_list, list) | |
535 | num++; | |
536 | ||
537 | return num; | |
538 | } | |
539 | ||
74ed7ab9 JS |
540 | static struct table_instance *table_instance_expand(struct table_instance *ti, |
541 | bool ufid) | |
b637e498 | 542 | { |
74ed7ab9 | 543 | return table_instance_rehash(ti, ti->n_buckets * 2, ufid); |
e6445719 PS |
544 | } |
545 | ||
56c19868 JR |
546 | /* Remove 'mask' from the mask list, if it is not needed any more. */ |
547 | static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask) | |
548 | { | |
549 | if (mask) { | |
550 | /* ovs-lock is required to protect mask-refcount and | |
551 | * mask list. | |
552 | */ | |
553 | ASSERT_OVSL(); | |
554 | BUG_ON(!mask->ref_count); | |
555 | mask->ref_count--; | |
556 | ||
557 | if (!mask->ref_count) { | |
558 | list_del_rcu(&mask->list); | |
559 | kfree_rcu(mask, rcu); | |
560 | } | |
561 | } | |
562 | } | |
563 | ||
564 | /* Must be called with OVS mutex held. */ | |
e6445719 PS |
565 | void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) |
566 | { | |
b637e498 | 567 | struct table_instance *ti = ovsl_dereference(table->ti); |
74ed7ab9 | 568 | struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti); |
b637e498 | 569 | |
e6445719 | 570 | BUG_ON(table->count == 0); |
74ed7ab9 | 571 | hlist_del_rcu(&flow->flow_table.node[ti->node_ver]); |
e6445719 | 572 | table->count--; |
74ed7ab9 JS |
573 | if (ovs_identifier_is_ufid(&flow->id)) { |
574 | hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]); | |
575 | table->ufid_count--; | |
576 | } | |
56c19868 JR |
577 | |
578 | /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be | |
579 | * accessible as long as the RCU read lock is held. | |
580 | */ | |
581 | flow_mask_remove(table, flow->mask); | |
e6445719 PS |
582 | } |
583 | ||
618ed0c8 | 584 | static struct sw_flow_mask *mask_alloc(void) |
e6445719 PS |
585 | { |
586 | struct sw_flow_mask *mask; | |
587 | ||
588 | mask = kmalloc(sizeof(*mask), GFP_KERNEL); | |
589 | if (mask) | |
e80857cc | 590 | mask->ref_count = 1; |
e6445719 PS |
591 | |
592 | return mask; | |
593 | } | |
594 | ||
e6445719 PS |
595 | static bool mask_equal(const struct sw_flow_mask *a, |
596 | const struct sw_flow_mask *b) | |
597 | { | |
7085130b DDP |
598 | const u8 *a_ = (const u8 *)&a->key + a->range.start; |
599 | const u8 *b_ = (const u8 *)&b->key + b->range.start; | |
e6445719 PS |
600 | |
601 | return (a->range.end == b->range.end) | |
602 | && (a->range.start == b->range.start) | |
603 | && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0); | |
604 | } | |
605 | ||
618ed0c8 | 606 | static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl, |
e6445719 PS |
607 | const struct sw_flow_mask *mask) |
608 | { | |
609 | struct list_head *ml; | |
610 | ||
b637e498 | 611 | list_for_each(ml, &tbl->mask_list) { |
e6445719 PS |
612 | struct sw_flow_mask *m; |
613 | m = container_of(ml, struct sw_flow_mask, list); | |
614 | if (mask_equal(mask, m)) | |
615 | return m; | |
616 | } | |
617 | ||
618 | return NULL; | |
619 | } | |
620 | ||
d1211908 | 621 | /* Add 'mask' into the mask list, if it is not already there. */ |
618ed0c8 | 622 | static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow, |
12eb18f7 | 623 | const struct sw_flow_mask *new) |
618ed0c8 PS |
624 | { |
625 | struct sw_flow_mask *mask; | |
626 | mask = flow_mask_find(tbl, new); | |
627 | if (!mask) { | |
628 | /* Allocate a new mask if none exsits. */ | |
629 | mask = mask_alloc(); | |
630 | if (!mask) | |
631 | return -ENOMEM; | |
632 | mask->key = new->key; | |
633 | mask->range = new->range; | |
634 | list_add_rcu(&mask->list, &tbl->mask_list); | |
e80857cc AZ |
635 | } else { |
636 | BUG_ON(!mask->ref_count); | |
637 | mask->ref_count++; | |
618ed0c8 PS |
638 | } |
639 | ||
618ed0c8 PS |
640 | flow->mask = mask; |
641 | return 0; | |
642 | } | |
643 | ||
56c19868 | 644 | /* Must be called with OVS mutex held. */ |
d29ab6f8 | 645 | static void flow_key_insert(struct flow_table *table, struct sw_flow *flow) |
e6445719 | 646 | { |
618ed0c8 PS |
647 | struct table_instance *new_ti = NULL; |
648 | struct table_instance *ti; | |
618ed0c8 | 649 | |
74ed7ab9 | 650 | flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range); |
618ed0c8 PS |
651 | ti = ovsl_dereference(table->ti); |
652 | table_instance_insert(ti, flow); | |
653 | table->count++; | |
654 | ||
655 | /* Expand table, if necessary, to make room. */ | |
656 | if (table->count > ti->n_buckets) | |
74ed7ab9 | 657 | new_ti = table_instance_expand(ti, false); |
618ed0c8 | 658 | else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) |
74ed7ab9 | 659 | new_ti = table_instance_rehash(ti, ti->n_buckets, false); |
618ed0c8 PS |
660 | |
661 | if (new_ti) { | |
662 | rcu_assign_pointer(table->ti, new_ti); | |
74ed7ab9 | 663 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
618ed0c8 PS |
664 | table->last_rehash = jiffies; |
665 | } | |
d29ab6f8 JS |
666 | } |
667 | ||
74ed7ab9 JS |
668 | /* Must be called with OVS mutex held. */ |
669 | static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow) | |
670 | { | |
671 | struct table_instance *ti; | |
672 | ||
673 | flow->ufid_table.hash = ufid_hash(&flow->id); | |
674 | ti = ovsl_dereference(table->ufid_ti); | |
675 | ufid_table_instance_insert(ti, flow); | |
676 | table->ufid_count++; | |
677 | ||
678 | /* Expand table, if necessary, to make room. */ | |
679 | if (table->ufid_count > ti->n_buckets) { | |
680 | struct table_instance *new_ti; | |
681 | ||
682 | new_ti = table_instance_expand(ti, true); | |
683 | if (new_ti) { | |
684 | rcu_assign_pointer(table->ufid_ti, new_ti); | |
685 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); | |
686 | } | |
687 | } | |
688 | } | |
689 | ||
d29ab6f8 JS |
690 | /* Must be called with OVS mutex held. */ |
691 | int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow, | |
692 | const struct sw_flow_mask *mask) | |
693 | { | |
694 | int err; | |
695 | ||
696 | err = flow_mask_insert(table, flow, mask); | |
697 | if (err) | |
698 | return err; | |
699 | flow_key_insert(table, flow); | |
74ed7ab9 JS |
700 | if (ovs_identifier_is_ufid(&flow->id)) |
701 | flow_ufid_insert(table, flow); | |
d29ab6f8 | 702 | |
618ed0c8 | 703 | return 0; |
e6445719 PS |
704 | } |
705 | ||
706 | /* Initializes the flow module. | |
707 | * Returns zero if successful or a negative error code. */ | |
708 | int ovs_flow_init(void) | |
709 | { | |
710 | BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); | |
711 | BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); | |
712 | ||
63e7959c | 713 | flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow) |
db74a333 | 714 | + (nr_cpu_ids |
aef833c5 | 715 | * sizeof(struct sw_flow_stats *)), |
63e7959c | 716 | 0, 0, NULL); |
e6445719 PS |
717 | if (flow_cache == NULL) |
718 | return -ENOMEM; | |
719 | ||
63e7959c | 720 | flow_stats_cache |
aef833c5 | 721 | = kmem_cache_create("sw_flow_stats", sizeof(struct sw_flow_stats), |
63e7959c JR |
722 | 0, SLAB_HWCACHE_ALIGN, NULL); |
723 | if (flow_stats_cache == NULL) { | |
724 | kmem_cache_destroy(flow_cache); | |
725 | flow_cache = NULL; | |
726 | return -ENOMEM; | |
727 | } | |
728 | ||
e6445719 PS |
729 | return 0; |
730 | } | |
731 | ||
732 | /* Uninitializes the flow module. */ | |
733 | void ovs_flow_exit(void) | |
734 | { | |
63e7959c | 735 | kmem_cache_destroy(flow_stats_cache); |
e6445719 PS |
736 | kmem_cache_destroy(flow_cache); |
737 | } |