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 | 36 | #define TBL_MIN_BUCKETS 1024 |
4bc63b1b | 37 | #define MASK_ARRAY_SIZE_MIN 16 |
b637e498 PS |
38 | #define REHASH_INTERVAL (10 * 60 * HZ) |
39 | ||
04b7d136 TZ |
40 | #define MC_HASH_SHIFT 8 |
41 | #define MC_HASH_ENTRIES (1u << MC_HASH_SHIFT) | |
42 | #define MC_HASH_SEGS ((sizeof(uint32_t) * 8) / MC_HASH_SHIFT) | |
43 | ||
e6445719 | 44 | static struct kmem_cache *flow_cache; |
63e7959c | 45 | struct kmem_cache *flow_stats_cache __read_mostly; |
e6445719 PS |
46 | |
47 | static u16 range_n_bytes(const struct sw_flow_key_range *range) | |
48 | { | |
49 | return range->end - range->start; | |
50 | } | |
51 | ||
52 | void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, | |
ae5f2fb1 | 53 | bool full, const struct sw_flow_mask *mask) |
e6445719 | 54 | { |
ae5f2fb1 JG |
55 | int start = full ? 0 : mask->range.start; |
56 | int len = full ? sizeof *dst : range_n_bytes(&mask->range); | |
57 | const long *m = (const long *)((const u8 *)&mask->key + start); | |
58 | const long *s = (const long *)((const u8 *)src + start); | |
59 | long *d = (long *)((u8 *)dst + start); | |
e6445719 PS |
60 | int i; |
61 | ||
ae5f2fb1 JG |
62 | /* If 'full' is true then all of 'dst' is fully initialized. Otherwise, |
63 | * if 'full' is false the memory outside of the 'mask->range' is left | |
64 | * uninitialized. This can be used as an optimization when further | |
65 | * operations on 'dst' only use contents within 'mask->range'. | |
e6445719 | 66 | */ |
ae5f2fb1 | 67 | for (i = 0; i < len; i += sizeof(long)) |
e6445719 PS |
68 | *d++ = *s++ & *m++; |
69 | } | |
70 | ||
23dabf88 | 71 | struct sw_flow *ovs_flow_alloc(void) |
e6445719 PS |
72 | { |
73 | struct sw_flow *flow; | |
aef833c5 | 74 | struct sw_flow_stats *stats; |
e6445719 | 75 | |
db74a333 | 76 | flow = kmem_cache_zalloc(flow_cache, GFP_KERNEL); |
e6445719 PS |
77 | if (!flow) |
78 | return ERR_PTR(-ENOMEM); | |
79 | ||
db74a333 | 80 | flow->stats_last_writer = -1; |
e6445719 | 81 | |
63e7959c JR |
82 | /* Initialize the default stat node. */ |
83 | stats = kmem_cache_alloc_node(flow_stats_cache, | |
598c12d0 KK |
84 | GFP_KERNEL | __GFP_ZERO, |
85 | node_online(0) ? 0 : NUMA_NO_NODE); | |
63e7959c | 86 | if (!stats) |
23dabf88 | 87 | goto err; |
e298e505 | 88 | |
63e7959c JR |
89 | spin_lock_init(&stats->lock); |
90 | ||
91 | RCU_INIT_POINTER(flow->stats[0], stats); | |
92 | ||
c4b2bf6b TZ |
93 | cpumask_set_cpu(0, &flow->cpu_used_mask); |
94 | ||
e6445719 | 95 | return flow; |
e298e505 | 96 | err: |
ece37c87 | 97 | kmem_cache_free(flow_cache, flow); |
e298e505 | 98 | return ERR_PTR(-ENOMEM); |
e6445719 PS |
99 | } |
100 | ||
12eb18f7 | 101 | int ovs_flow_tbl_count(const struct flow_table *table) |
b637e498 PS |
102 | { |
103 | return table->count; | |
104 | } | |
105 | ||
e6445719 PS |
106 | static void flow_free(struct sw_flow *flow) |
107 | { | |
db74a333 | 108 | int cpu; |
63e7959c | 109 | |
74ed7ab9 JS |
110 | if (ovs_identifier_is_key(&flow->id)) |
111 | kfree(flow->id.unmasked_key); | |
34ae932a TG |
112 | if (flow->sf_acts) |
113 | ovs_nla_free_flow_actions((struct sw_flow_actions __force *)flow->sf_acts); | |
db74a333 | 114 | /* We open code this to make sure cpu 0 is always considered */ |
c4b2bf6b | 115 | for (cpu = 0; cpu < nr_cpu_ids; cpu = cpumask_next(cpu, &flow->cpu_used_mask)) |
db74a333 | 116 | if (flow->stats[cpu]) |
63e7959c | 117 | kmem_cache_free(flow_stats_cache, |
aef833c5 | 118 | (struct sw_flow_stats __force *)flow->stats[cpu]); |
e6445719 PS |
119 | kmem_cache_free(flow_cache, flow); |
120 | } | |
121 | ||
122 | static void rcu_free_flow_callback(struct rcu_head *rcu) | |
123 | { | |
124 | struct sw_flow *flow = container_of(rcu, struct sw_flow, rcu); | |
125 | ||
126 | flow_free(flow); | |
127 | } | |
128 | ||
e80857cc | 129 | void ovs_flow_free(struct sw_flow *flow, bool deferred) |
618ed0c8 | 130 | { |
e80857cc | 131 | if (!flow) |
618ed0c8 PS |
132 | return; |
133 | ||
e6445719 PS |
134 | if (deferred) |
135 | call_rcu(&flow->rcu, rcu_free_flow_callback); | |
136 | else | |
137 | flow_free(flow); | |
138 | } | |
139 | ||
b637e498 | 140 | static void __table_instance_destroy(struct table_instance *ti) |
e6445719 | 141 | { |
ee9c5e67 | 142 | kvfree(ti->buckets); |
b637e498 | 143 | kfree(ti); |
e6445719 PS |
144 | } |
145 | ||
b637e498 | 146 | static struct table_instance *table_instance_alloc(int new_size) |
e6445719 | 147 | { |
b637e498 | 148 | struct table_instance *ti = kmalloc(sizeof(*ti), GFP_KERNEL); |
ee9c5e67 | 149 | int i; |
e6445719 | 150 | |
b637e498 | 151 | if (!ti) |
e6445719 PS |
152 | return NULL; |
153 | ||
ee9c5e67 KO |
154 | ti->buckets = kvmalloc_array(new_size, sizeof(struct hlist_head), |
155 | GFP_KERNEL); | |
b637e498 PS |
156 | if (!ti->buckets) { |
157 | kfree(ti); | |
e6445719 PS |
158 | return NULL; |
159 | } | |
ee9c5e67 KO |
160 | |
161 | for (i = 0; i < new_size; i++) | |
162 | INIT_HLIST_HEAD(&ti->buckets[i]); | |
163 | ||
b637e498 PS |
164 | ti->n_buckets = new_size; |
165 | ti->node_ver = 0; | |
166 | ti->keep_flows = false; | |
167 | get_random_bytes(&ti->hash_seed, sizeof(u32)); | |
e6445719 | 168 | |
b637e498 | 169 | return ti; |
e6445719 PS |
170 | } |
171 | ||
4bc63b1b TZ |
172 | static struct mask_array *tbl_mask_array_alloc(int size) |
173 | { | |
174 | struct mask_array *new; | |
175 | ||
176 | size = max(MASK_ARRAY_SIZE_MIN, size); | |
177 | new = kzalloc(sizeof(struct mask_array) + | |
178 | sizeof(struct sw_flow_mask *) * size, GFP_KERNEL); | |
179 | if (!new) | |
180 | return NULL; | |
181 | ||
182 | new->count = 0; | |
183 | new->max = size; | |
184 | ||
185 | return new; | |
186 | } | |
187 | ||
188 | static int tbl_mask_array_realloc(struct flow_table *tbl, int size) | |
189 | { | |
190 | struct mask_array *old; | |
191 | struct mask_array *new; | |
192 | ||
193 | new = tbl_mask_array_alloc(size); | |
194 | if (!new) | |
195 | return -ENOMEM; | |
196 | ||
197 | old = ovsl_dereference(tbl->mask_array); | |
198 | if (old) { | |
199 | int i; | |
200 | ||
201 | for (i = 0; i < old->max; i++) { | |
202 | if (ovsl_dereference(old->masks[i])) | |
203 | new->masks[new->count++] = old->masks[i]; | |
204 | } | |
205 | } | |
206 | ||
207 | rcu_assign_pointer(tbl->mask_array, new); | |
208 | kfree_rcu(old, rcu); | |
209 | ||
210 | return 0; | |
211 | } | |
212 | ||
50b0e61b TZ |
213 | static int tbl_mask_array_add_mask(struct flow_table *tbl, |
214 | struct sw_flow_mask *new) | |
215 | { | |
216 | struct mask_array *ma = ovsl_dereference(tbl->mask_array); | |
217 | int err, ma_count = READ_ONCE(ma->count); | |
218 | ||
219 | if (ma_count >= ma->max) { | |
220 | err = tbl_mask_array_realloc(tbl, ma->max + | |
221 | MASK_ARRAY_SIZE_MIN); | |
222 | if (err) | |
223 | return err; | |
224 | ||
225 | ma = ovsl_dereference(tbl->mask_array); | |
226 | } | |
227 | ||
228 | BUG_ON(ovsl_dereference(ma->masks[ma_count])); | |
229 | ||
230 | rcu_assign_pointer(ma->masks[ma_count], new); | |
231 | WRITE_ONCE(ma->count, ma_count +1); | |
232 | ||
233 | return 0; | |
234 | } | |
235 | ||
236 | static void tbl_mask_array_del_mask(struct flow_table *tbl, | |
237 | struct sw_flow_mask *mask) | |
238 | { | |
239 | struct mask_array *ma = ovsl_dereference(tbl->mask_array); | |
240 | int i, ma_count = READ_ONCE(ma->count); | |
241 | ||
242 | /* Remove the deleted mask pointers from the array */ | |
243 | for (i = 0; i < ma_count; i++) { | |
244 | if (mask == ovsl_dereference(ma->masks[i])) | |
245 | goto found; | |
246 | } | |
247 | ||
248 | BUG(); | |
249 | return; | |
250 | ||
251 | found: | |
252 | WRITE_ONCE(ma->count, ma_count -1); | |
253 | ||
254 | rcu_assign_pointer(ma->masks[i], ma->masks[ma_count -1]); | |
255 | RCU_INIT_POINTER(ma->masks[ma_count -1], NULL); | |
256 | ||
257 | kfree_rcu(mask, rcu); | |
258 | ||
259 | /* Shrink the mask array if necessary. */ | |
260 | if (ma->max >= (MASK_ARRAY_SIZE_MIN * 2) && | |
261 | ma_count <= (ma->max / 3)) | |
262 | tbl_mask_array_realloc(tbl, ma->max / 2); | |
263 | } | |
264 | ||
265 | /* Remove 'mask' from the mask list, if it is not needed any more. */ | |
266 | static void flow_mask_remove(struct flow_table *tbl, struct sw_flow_mask *mask) | |
267 | { | |
268 | if (mask) { | |
269 | /* ovs-lock is required to protect mask-refcount and | |
270 | * mask list. | |
271 | */ | |
272 | ASSERT_OVSL(); | |
273 | BUG_ON(!mask->ref_count); | |
274 | mask->ref_count--; | |
275 | ||
276 | if (!mask->ref_count) | |
277 | tbl_mask_array_del_mask(tbl, mask); | |
278 | } | |
279 | } | |
280 | ||
b637e498 | 281 | int ovs_flow_tbl_init(struct flow_table *table) |
e6445719 | 282 | { |
74ed7ab9 | 283 | struct table_instance *ti, *ufid_ti; |
4bc63b1b | 284 | struct mask_array *ma; |
e6445719 | 285 | |
04b7d136 TZ |
286 | table->mask_cache = __alloc_percpu(sizeof(struct mask_cache_entry) * |
287 | MC_HASH_ENTRIES, | |
288 | __alignof__(struct mask_cache_entry)); | |
289 | if (!table->mask_cache) | |
290 | return -ENOMEM; | |
e6445719 | 291 | |
4bc63b1b TZ |
292 | ma = tbl_mask_array_alloc(MASK_ARRAY_SIZE_MIN); |
293 | if (!ma) | |
294 | goto free_mask_cache; | |
295 | ||
04b7d136 | 296 | ti = table_instance_alloc(TBL_MIN_BUCKETS); |
b637e498 | 297 | if (!ti) |
4bc63b1b | 298 | goto free_mask_array; |
e6445719 | 299 | |
74ed7ab9 JS |
300 | ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
301 | if (!ufid_ti) | |
302 | goto free_ti; | |
303 | ||
b637e498 | 304 | rcu_assign_pointer(table->ti, ti); |
74ed7ab9 | 305 | rcu_assign_pointer(table->ufid_ti, ufid_ti); |
4bc63b1b | 306 | rcu_assign_pointer(table->mask_array, ma); |
b637e498 PS |
307 | table->last_rehash = jiffies; |
308 | table->count = 0; | |
74ed7ab9 | 309 | table->ufid_count = 0; |
b637e498 | 310 | return 0; |
74ed7ab9 JS |
311 | |
312 | free_ti: | |
313 | __table_instance_destroy(ti); | |
4bc63b1b TZ |
314 | free_mask_array: |
315 | kfree(ma); | |
04b7d136 TZ |
316 | free_mask_cache: |
317 | free_percpu(table->mask_cache); | |
74ed7ab9 | 318 | return -ENOMEM; |
e6445719 PS |
319 | } |
320 | ||
321 | static void flow_tbl_destroy_rcu_cb(struct rcu_head *rcu) | |
322 | { | |
b637e498 | 323 | struct table_instance *ti = container_of(rcu, struct table_instance, rcu); |
e6445719 | 324 | |
b637e498 | 325 | __table_instance_destroy(ti); |
e6445719 PS |
326 | } |
327 | ||
50b0e61b TZ |
328 | static void table_instance_flow_free(struct flow_table *table, |
329 | struct table_instance *ti, | |
330 | struct table_instance *ufid_ti, | |
331 | struct sw_flow *flow, | |
332 | bool count) | |
333 | { | |
334 | hlist_del_rcu(&flow->flow_table.node[ti->node_ver]); | |
335 | if (count) | |
336 | table->count--; | |
337 | ||
338 | if (ovs_identifier_is_ufid(&flow->id)) { | |
339 | hlist_del_rcu(&flow->ufid_table.node[ufid_ti->node_ver]); | |
340 | ||
341 | if (count) | |
342 | table->ufid_count--; | |
343 | } | |
344 | ||
345 | flow_mask_remove(table, flow->mask); | |
346 | } | |
347 | ||
348 | static void table_instance_destroy(struct flow_table *table, | |
349 | struct table_instance *ti, | |
74ed7ab9 JS |
350 | struct table_instance *ufid_ti, |
351 | bool deferred) | |
e6445719 | 352 | { |
e80857cc AZ |
353 | int i; |
354 | ||
b637e498 | 355 | if (!ti) |
e6445719 PS |
356 | return; |
357 | ||
74ed7ab9 | 358 | BUG_ON(!ufid_ti); |
e80857cc AZ |
359 | if (ti->keep_flows) |
360 | goto skip_flows; | |
361 | ||
362 | for (i = 0; i < ti->n_buckets; i++) { | |
363 | struct sw_flow *flow; | |
ee9c5e67 | 364 | struct hlist_head *head = &ti->buckets[i]; |
e80857cc | 365 | struct hlist_node *n; |
e80857cc | 366 | |
50b0e61b TZ |
367 | hlist_for_each_entry_safe(flow, n, head, |
368 | flow_table.node[ti->node_ver]) { | |
369 | ||
370 | table_instance_flow_free(table, ti, ufid_ti, | |
371 | flow, false); | |
e80857cc AZ |
372 | ovs_flow_free(flow, deferred); |
373 | } | |
374 | } | |
375 | ||
376 | skip_flows: | |
74ed7ab9 | 377 | if (deferred) { |
b637e498 | 378 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
74ed7ab9 JS |
379 | call_rcu(&ufid_ti->rcu, flow_tbl_destroy_rcu_cb); |
380 | } else { | |
b637e498 | 381 | __table_instance_destroy(ti); |
74ed7ab9 JS |
382 | __table_instance_destroy(ufid_ti); |
383 | } | |
b637e498 PS |
384 | } |
385 | ||
9b996e54 PS |
386 | /* No need for locking this function is called from RCU callback or |
387 | * error path. | |
388 | */ | |
389 | void ovs_flow_tbl_destroy(struct flow_table *table) | |
b637e498 | 390 | { |
9b996e54 | 391 | struct table_instance *ti = rcu_dereference_raw(table->ti); |
74ed7ab9 | 392 | struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti); |
b637e498 | 393 | |
04b7d136 | 394 | free_percpu(table->mask_cache); |
4bc63b1b | 395 | kfree_rcu(rcu_dereference_raw(table->mask_array), rcu); |
50b0e61b | 396 | table_instance_destroy(table, ti, ufid_ti, false); |
e6445719 PS |
397 | } |
398 | ||
b637e498 | 399 | struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *ti, |
e6445719 PS |
400 | u32 *bucket, u32 *last) |
401 | { | |
402 | struct sw_flow *flow; | |
403 | struct hlist_head *head; | |
404 | int ver; | |
405 | int i; | |
406 | ||
b637e498 PS |
407 | ver = ti->node_ver; |
408 | while (*bucket < ti->n_buckets) { | |
e6445719 | 409 | i = 0; |
ee9c5e67 | 410 | head = &ti->buckets[*bucket]; |
74ed7ab9 | 411 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ver]) { |
e6445719 PS |
412 | if (i < *last) { |
413 | i++; | |
414 | continue; | |
415 | } | |
416 | *last = i + 1; | |
417 | return flow; | |
418 | } | |
419 | (*bucket)++; | |
420 | *last = 0; | |
421 | } | |
422 | ||
423 | return NULL; | |
424 | } | |
425 | ||
b637e498 | 426 | static struct hlist_head *find_bucket(struct table_instance *ti, u32 hash) |
e6445719 | 427 | { |
b637e498 | 428 | hash = jhash_1word(hash, ti->hash_seed); |
ee9c5e67 | 429 | return &ti->buckets[hash & (ti->n_buckets - 1)]; |
e6445719 PS |
430 | } |
431 | ||
74ed7ab9 JS |
432 | static void table_instance_insert(struct table_instance *ti, |
433 | struct sw_flow *flow) | |
e6445719 PS |
434 | { |
435 | struct hlist_head *head; | |
436 | ||
74ed7ab9 JS |
437 | head = find_bucket(ti, flow->flow_table.hash); |
438 | hlist_add_head_rcu(&flow->flow_table.node[ti->node_ver], head); | |
439 | } | |
440 | ||
441 | static void ufid_table_instance_insert(struct table_instance *ti, | |
442 | struct sw_flow *flow) | |
443 | { | |
444 | struct hlist_head *head; | |
445 | ||
446 | head = find_bucket(ti, flow->ufid_table.hash); | |
447 | hlist_add_head_rcu(&flow->ufid_table.node[ti->node_ver], head); | |
e6445719 PS |
448 | } |
449 | ||
b637e498 | 450 | static void flow_table_copy_flows(struct table_instance *old, |
74ed7ab9 | 451 | struct table_instance *new, bool ufid) |
e6445719 PS |
452 | { |
453 | int old_ver; | |
454 | int i; | |
455 | ||
456 | old_ver = old->node_ver; | |
457 | new->node_ver = !old_ver; | |
458 | ||
459 | /* Insert in new table. */ | |
460 | for (i = 0; i < old->n_buckets; i++) { | |
461 | struct sw_flow *flow; | |
ee9c5e67 | 462 | struct hlist_head *head = &old->buckets[i]; |
e6445719 | 463 | |
74ed7ab9 | 464 | if (ufid) |
64948427 TZ |
465 | hlist_for_each_entry_rcu(flow, head, |
466 | ufid_table.node[old_ver], | |
467 | lockdep_ovsl_is_held()) | |
74ed7ab9 JS |
468 | ufid_table_instance_insert(new, flow); |
469 | else | |
64948427 TZ |
470 | hlist_for_each_entry_rcu(flow, head, |
471 | flow_table.node[old_ver], | |
472 | lockdep_ovsl_is_held()) | |
74ed7ab9 | 473 | table_instance_insert(new, flow); |
e6445719 PS |
474 | } |
475 | ||
e6445719 PS |
476 | old->keep_flows = true; |
477 | } | |
478 | ||
b637e498 | 479 | static struct table_instance *table_instance_rehash(struct table_instance *ti, |
74ed7ab9 | 480 | int n_buckets, bool ufid) |
e6445719 | 481 | { |
b637e498 | 482 | struct table_instance *new_ti; |
e6445719 | 483 | |
b637e498 PS |
484 | new_ti = table_instance_alloc(n_buckets); |
485 | if (!new_ti) | |
618ed0c8 | 486 | return NULL; |
e6445719 | 487 | |
74ed7ab9 | 488 | flow_table_copy_flows(ti, new_ti, ufid); |
e6445719 | 489 | |
b637e498 | 490 | return new_ti; |
e6445719 PS |
491 | } |
492 | ||
b637e498 | 493 | int ovs_flow_tbl_flush(struct flow_table *flow_table) |
e6445719 | 494 | { |
74ed7ab9 JS |
495 | struct table_instance *old_ti, *new_ti; |
496 | struct table_instance *old_ufid_ti, *new_ufid_ti; | |
e6445719 | 497 | |
b637e498 PS |
498 | new_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
499 | if (!new_ti) | |
500 | return -ENOMEM; | |
74ed7ab9 JS |
501 | new_ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS); |
502 | if (!new_ufid_ti) | |
503 | goto err_free_ti; | |
504 | ||
505 | old_ti = ovsl_dereference(flow_table->ti); | |
506 | old_ufid_ti = ovsl_dereference(flow_table->ufid_ti); | |
b637e498 PS |
507 | |
508 | rcu_assign_pointer(flow_table->ti, new_ti); | |
74ed7ab9 | 509 | rcu_assign_pointer(flow_table->ufid_ti, new_ufid_ti); |
b637e498 PS |
510 | flow_table->last_rehash = jiffies; |
511 | flow_table->count = 0; | |
74ed7ab9 | 512 | flow_table->ufid_count = 0; |
b637e498 | 513 | |
50b0e61b | 514 | table_instance_destroy(flow_table, old_ti, old_ufid_ti, true); |
b637e498 | 515 | return 0; |
74ed7ab9 JS |
516 | |
517 | err_free_ti: | |
518 | __table_instance_destroy(new_ti); | |
519 | return -ENOMEM; | |
e6445719 PS |
520 | } |
521 | ||
272c2cf8 JS |
522 | static u32 flow_hash(const struct sw_flow_key *key, |
523 | const struct sw_flow_key_range *range) | |
e6445719 | 524 | { |
515b65a4 | 525 | const u32 *hash_key = (const u32 *)((const u8 *)key + range->start); |
e6445719 PS |
526 | |
527 | /* Make sure number of hash bytes are multiple of u32. */ | |
515b65a4 | 528 | int hash_u32s = range_n_bytes(range) >> 2; |
e6445719 | 529 | |
87545899 | 530 | return jhash2(hash_key, hash_u32s, 0); |
e6445719 PS |
531 | } |
532 | ||
533 | static int flow_key_start(const struct sw_flow_key *key) | |
534 | { | |
00a93bab | 535 | if (key->tun_proto) |
e6445719 PS |
536 | return 0; |
537 | else | |
538 | return rounddown(offsetof(struct sw_flow_key, phy), | |
539 | sizeof(long)); | |
540 | } | |
541 | ||
542 | static bool cmp_key(const struct sw_flow_key *key1, | |
543 | const struct sw_flow_key *key2, | |
544 | int key_start, int key_end) | |
545 | { | |
7085130b DDP |
546 | const long *cp1 = (const long *)((const u8 *)key1 + key_start); |
547 | const long *cp2 = (const long *)((const u8 *)key2 + key_start); | |
e6445719 PS |
548 | long diffs = 0; |
549 | int i; | |
550 | ||
551 | for (i = key_start; i < key_end; i += sizeof(long)) | |
552 | diffs |= *cp1++ ^ *cp2++; | |
553 | ||
554 | return diffs == 0; | |
555 | } | |
556 | ||
557 | static bool flow_cmp_masked_key(const struct sw_flow *flow, | |
558 | const struct sw_flow_key *key, | |
272c2cf8 | 559 | const struct sw_flow_key_range *range) |
e6445719 | 560 | { |
272c2cf8 | 561 | return cmp_key(&flow->key, key, range->start, range->end); |
e6445719 PS |
562 | } |
563 | ||
74ed7ab9 JS |
564 | static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow, |
565 | const struct sw_flow_match *match) | |
e6445719 PS |
566 | { |
567 | struct sw_flow_key *key = match->key; | |
568 | int key_start = flow_key_start(key); | |
569 | int key_end = match->range.end; | |
570 | ||
74ed7ab9 JS |
571 | BUG_ON(ovs_identifier_is_ufid(&flow->id)); |
572 | return cmp_key(flow->id.unmasked_key, key, key_start, key_end); | |
e6445719 PS |
573 | } |
574 | ||
b637e498 | 575 | static struct sw_flow *masked_flow_lookup(struct table_instance *ti, |
e6445719 | 576 | const struct sw_flow_key *unmasked, |
04b7d136 TZ |
577 | const struct sw_flow_mask *mask, |
578 | u32 *n_mask_hit) | |
e6445719 PS |
579 | { |
580 | struct sw_flow *flow; | |
581 | struct hlist_head *head; | |
e6445719 PS |
582 | u32 hash; |
583 | struct sw_flow_key masked_key; | |
584 | ||
ae5f2fb1 | 585 | ovs_flow_mask_key(&masked_key, unmasked, false, mask); |
272c2cf8 | 586 | hash = flow_hash(&masked_key, &mask->range); |
b637e498 | 587 | head = find_bucket(ti, hash); |
04b7d136 TZ |
588 | (*n_mask_hit)++; |
589 | ||
a2cfb96c MB |
590 | hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver], |
591 | lockdep_ovsl_is_held()) { | |
74ed7ab9 | 592 | if (flow->mask == mask && flow->flow_table.hash == hash && |
272c2cf8 | 593 | flow_cmp_masked_key(flow, &masked_key, &mask->range)) |
e6445719 PS |
594 | return flow; |
595 | } | |
596 | return NULL; | |
597 | } | |
598 | ||
a7f35e78 TZ |
599 | /* Flow lookup does full lookup on flow table. It starts with |
600 | * mask from index passed in *index. | |
601 | */ | |
04b7d136 TZ |
602 | static struct sw_flow *flow_lookup(struct flow_table *tbl, |
603 | struct table_instance *ti, | |
4bc63b1b | 604 | struct mask_array *ma, |
04b7d136 | 605 | const struct sw_flow_key *key, |
4bc63b1b TZ |
606 | u32 *n_mask_hit, |
607 | u32 *index) | |
e6445719 | 608 | { |
b637e498 | 609 | struct sw_flow *flow; |
57f7d7b9 | 610 | struct sw_flow_mask *mask; |
4bc63b1b | 611 | int i; |
e6445719 | 612 | |
0a3e0137 | 613 | if (likely(*index < ma->max)) { |
a7f35e78 | 614 | mask = rcu_dereference_ovsl(ma->masks[*index]); |
4bc63b1b TZ |
615 | if (mask) { |
616 | flow = masked_flow_lookup(ti, key, mask, n_mask_hit); | |
a7f35e78 | 617 | if (flow) |
4bc63b1b | 618 | return flow; |
a7f35e78 TZ |
619 | } |
620 | } | |
621 | ||
622 | for (i = 0; i < ma->max; i++) { | |
623 | ||
624 | if (i == *index) | |
625 | continue; | |
626 | ||
627 | mask = rcu_dereference_ovsl(ma->masks[i]); | |
0a3e0137 | 628 | if (unlikely(!mask)) |
57f7d7b9 | 629 | break; |
a7f35e78 TZ |
630 | |
631 | flow = masked_flow_lookup(ti, key, mask, n_mask_hit); | |
632 | if (flow) { /* Found */ | |
633 | *index = i; | |
634 | return flow; | |
4bc63b1b | 635 | } |
e6445719 | 636 | } |
4bc63b1b | 637 | |
b637e498 PS |
638 | return NULL; |
639 | } | |
e6445719 | 640 | |
04b7d136 TZ |
641 | /* |
642 | * mask_cache maps flow to probable mask. This cache is not tightly | |
643 | * coupled cache, It means updates to mask list can result in inconsistent | |
644 | * cache entry in mask cache. | |
645 | * This is per cpu cache and is divided in MC_HASH_SEGS segments. | |
646 | * In case of a hash collision the entry is hashed in next segment. | |
647 | * */ | |
648 | struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl, | |
649 | const struct sw_flow_key *key, | |
650 | u32 skb_hash, | |
651 | u32 *n_mask_hit) | |
652 | { | |
a7f35e78 TZ |
653 | struct mask_array *ma = rcu_dereference(tbl->mask_array); |
654 | struct table_instance *ti = rcu_dereference(tbl->ti); | |
655 | struct mask_cache_entry *entries, *ce; | |
04b7d136 | 656 | struct sw_flow *flow; |
a7f35e78 | 657 | u32 hash; |
04b7d136 TZ |
658 | int seg; |
659 | ||
660 | *n_mask_hit = 0; | |
4bc63b1b | 661 | if (unlikely(!skb_hash)) { |
a7f35e78 | 662 | u32 mask_index = 0; |
4bc63b1b TZ |
663 | |
664 | return flow_lookup(tbl, ti, ma, key, n_mask_hit, &mask_index); | |
665 | } | |
04b7d136 | 666 | |
a7f35e78 TZ |
667 | /* Pre and post recirulation flows usually have the same skb_hash |
668 | * value. To avoid hash collisions, rehash the 'skb_hash' with | |
669 | * 'recirc_id'. */ | |
670 | if (key->recirc_id) | |
671 | skb_hash = jhash_1word(skb_hash, key->recirc_id); | |
672 | ||
673 | ce = NULL; | |
674 | hash = skb_hash; | |
04b7d136 TZ |
675 | entries = this_cpu_ptr(tbl->mask_cache); |
676 | ||
a7f35e78 | 677 | /* Find the cache entry 'ce' to operate on. */ |
04b7d136 | 678 | for (seg = 0; seg < MC_HASH_SEGS; seg++) { |
a7f35e78 TZ |
679 | int index = hash & (MC_HASH_ENTRIES - 1); |
680 | struct mask_cache_entry *e; | |
681 | ||
682 | e = &entries[index]; | |
683 | if (e->skb_hash == skb_hash) { | |
684 | flow = flow_lookup(tbl, ti, ma, key, n_mask_hit, | |
685 | &e->mask_index); | |
686 | if (!flow) | |
687 | e->skb_hash = 0; | |
688 | return flow; | |
04b7d136 TZ |
689 | } |
690 | ||
a7f35e78 TZ |
691 | if (!ce || e->skb_hash < ce->skb_hash) |
692 | ce = e; /* A better replacement cache candidate. */ | |
04b7d136 TZ |
693 | |
694 | hash >>= MC_HASH_SHIFT; | |
695 | } | |
696 | ||
a7f35e78 TZ |
697 | /* Cache miss, do full lookup. */ |
698 | flow = flow_lookup(tbl, ti, ma, key, n_mask_hit, &ce->mask_index); | |
4bc63b1b | 699 | if (flow) |
a7f35e78 | 700 | ce->skb_hash = skb_hash; |
04b7d136 TZ |
701 | |
702 | return flow; | |
703 | } | |
704 | ||
5bb50632 AZ |
705 | struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl, |
706 | const struct sw_flow_key *key) | |
707 | { | |
04b7d136 | 708 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); |
4bc63b1b | 709 | struct mask_array *ma = rcu_dereference_ovsl(tbl->mask_array); |
5bb50632 | 710 | u32 __always_unused n_mask_hit; |
a7f35e78 | 711 | u32 index = 0; |
5bb50632 | 712 | |
4bc63b1b | 713 | return flow_lookup(tbl, ti, ma, key, &n_mask_hit, &index); |
5bb50632 AZ |
714 | } |
715 | ||
4a46b24e | 716 | struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl, |
12eb18f7 | 717 | const struct sw_flow_match *match) |
4a46b24e | 718 | { |
4bc63b1b TZ |
719 | struct mask_array *ma = ovsl_dereference(tbl->mask_array); |
720 | int i; | |
4a46b24e AW |
721 | |
722 | /* Always called under ovs-mutex. */ | |
4bc63b1b TZ |
723 | for (i = 0; i < ma->max; i++) { |
724 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ti); | |
725 | u32 __always_unused n_mask_hit; | |
726 | struct sw_flow_mask *mask; | |
727 | struct sw_flow *flow; | |
728 | ||
729 | mask = ovsl_dereference(ma->masks[i]); | |
730 | if (!mask) | |
731 | continue; | |
732 | ||
04b7d136 | 733 | flow = masked_flow_lookup(ti, match->key, mask, &n_mask_hit); |
74ed7ab9 | 734 | if (flow && ovs_identifier_is_key(&flow->id) && |
4bc63b1b | 735 | ovs_flow_cmp_unmasked_key(flow, match)) { |
74ed7ab9 | 736 | return flow; |
4bc63b1b | 737 | } |
74ed7ab9 | 738 | } |
4bc63b1b | 739 | |
74ed7ab9 JS |
740 | return NULL; |
741 | } | |
742 | ||
743 | static u32 ufid_hash(const struct sw_flow_id *sfid) | |
744 | { | |
745 | return jhash(sfid->ufid, sfid->ufid_len, 0); | |
746 | } | |
747 | ||
748 | static bool ovs_flow_cmp_ufid(const struct sw_flow *flow, | |
749 | const struct sw_flow_id *sfid) | |
750 | { | |
751 | if (flow->id.ufid_len != sfid->ufid_len) | |
752 | return false; | |
753 | ||
754 | return !memcmp(flow->id.ufid, sfid->ufid, sfid->ufid_len); | |
755 | } | |
756 | ||
757 | bool ovs_flow_cmp(const struct sw_flow *flow, const struct sw_flow_match *match) | |
758 | { | |
759 | if (ovs_identifier_is_ufid(&flow->id)) | |
760 | return flow_cmp_masked_key(flow, match->key, &match->range); | |
761 | ||
762 | return ovs_flow_cmp_unmasked_key(flow, match); | |
763 | } | |
764 | ||
765 | struct sw_flow *ovs_flow_tbl_lookup_ufid(struct flow_table *tbl, | |
766 | const struct sw_flow_id *ufid) | |
767 | { | |
768 | struct table_instance *ti = rcu_dereference_ovsl(tbl->ufid_ti); | |
769 | struct sw_flow *flow; | |
770 | struct hlist_head *head; | |
771 | u32 hash; | |
772 | ||
773 | hash = ufid_hash(ufid); | |
774 | head = find_bucket(ti, hash); | |
a2cfb96c MB |
775 | hlist_for_each_entry_rcu(flow, head, ufid_table.node[ti->node_ver], |
776 | lockdep_ovsl_is_held()) { | |
74ed7ab9 JS |
777 | if (flow->ufid_table.hash == hash && |
778 | ovs_flow_cmp_ufid(flow, ufid)) | |
4a46b24e AW |
779 | return flow; |
780 | } | |
781 | return NULL; | |
782 | } | |
783 | ||
1bd7116f AZ |
784 | int ovs_flow_tbl_num_masks(const struct flow_table *table) |
785 | { | |
4bc63b1b | 786 | struct mask_array *ma = rcu_dereference_ovsl(table->mask_array); |
57f7d7b9 | 787 | return READ_ONCE(ma->count); |
1bd7116f AZ |
788 | } |
789 | ||
74ed7ab9 JS |
790 | static struct table_instance *table_instance_expand(struct table_instance *ti, |
791 | bool ufid) | |
b637e498 | 792 | { |
74ed7ab9 | 793 | return table_instance_rehash(ti, ti->n_buckets * 2, ufid); |
e6445719 PS |
794 | } |
795 | ||
56c19868 | 796 | /* Must be called with OVS mutex held. */ |
e6445719 PS |
797 | void ovs_flow_tbl_remove(struct flow_table *table, struct sw_flow *flow) |
798 | { | |
b637e498 | 799 | struct table_instance *ti = ovsl_dereference(table->ti); |
74ed7ab9 | 800 | struct table_instance *ufid_ti = ovsl_dereference(table->ufid_ti); |
b637e498 | 801 | |
e6445719 | 802 | BUG_ON(table->count == 0); |
50b0e61b | 803 | table_instance_flow_free(table, ti, ufid_ti, flow, true); |
e6445719 PS |
804 | } |
805 | ||
618ed0c8 | 806 | static struct sw_flow_mask *mask_alloc(void) |
e6445719 PS |
807 | { |
808 | struct sw_flow_mask *mask; | |
809 | ||
810 | mask = kmalloc(sizeof(*mask), GFP_KERNEL); | |
811 | if (mask) | |
e80857cc | 812 | mask->ref_count = 1; |
e6445719 PS |
813 | |
814 | return mask; | |
815 | } | |
816 | ||
e6445719 PS |
817 | static bool mask_equal(const struct sw_flow_mask *a, |
818 | const struct sw_flow_mask *b) | |
819 | { | |
7085130b DDP |
820 | const u8 *a_ = (const u8 *)&a->key + a->range.start; |
821 | const u8 *b_ = (const u8 *)&b->key + b->range.start; | |
e6445719 PS |
822 | |
823 | return (a->range.end == b->range.end) | |
824 | && (a->range.start == b->range.start) | |
825 | && (memcmp(a_, b_, range_n_bytes(&a->range)) == 0); | |
826 | } | |
827 | ||
618ed0c8 | 828 | static struct sw_flow_mask *flow_mask_find(const struct flow_table *tbl, |
e6445719 PS |
829 | const struct sw_flow_mask *mask) |
830 | { | |
4bc63b1b TZ |
831 | struct mask_array *ma; |
832 | int i; | |
e6445719 | 833 | |
4bc63b1b TZ |
834 | ma = ovsl_dereference(tbl->mask_array); |
835 | for (i = 0; i < ma->max; i++) { | |
836 | struct sw_flow_mask *t; | |
837 | t = ovsl_dereference(ma->masks[i]); | |
838 | ||
839 | if (t && mask_equal(mask, t)) | |
840 | return t; | |
e6445719 PS |
841 | } |
842 | ||
843 | return NULL; | |
844 | } | |
845 | ||
d1211908 | 846 | /* Add 'mask' into the mask list, if it is not already there. */ |
618ed0c8 | 847 | static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow, |
12eb18f7 | 848 | const struct sw_flow_mask *new) |
618ed0c8 PS |
849 | { |
850 | struct sw_flow_mask *mask; | |
4bc63b1b | 851 | |
618ed0c8 PS |
852 | mask = flow_mask_find(tbl, new); |
853 | if (!mask) { | |
854 | /* Allocate a new mask if none exsits. */ | |
855 | mask = mask_alloc(); | |
856 | if (!mask) | |
857 | return -ENOMEM; | |
858 | mask->key = new->key; | |
859 | mask->range = new->range; | |
4bc63b1b TZ |
860 | |
861 | /* Add mask to mask-list. */ | |
57f7d7b9 TZ |
862 | if (tbl_mask_array_add_mask(tbl, mask)) { |
863 | kfree(mask); | |
864 | return -ENOMEM; | |
4bc63b1b | 865 | } |
e80857cc AZ |
866 | } else { |
867 | BUG_ON(!mask->ref_count); | |
868 | mask->ref_count++; | |
618ed0c8 PS |
869 | } |
870 | ||
618ed0c8 PS |
871 | flow->mask = mask; |
872 | return 0; | |
873 | } | |
874 | ||
56c19868 | 875 | /* Must be called with OVS mutex held. */ |
d29ab6f8 | 876 | static void flow_key_insert(struct flow_table *table, struct sw_flow *flow) |
e6445719 | 877 | { |
618ed0c8 PS |
878 | struct table_instance *new_ti = NULL; |
879 | struct table_instance *ti; | |
618ed0c8 | 880 | |
74ed7ab9 | 881 | flow->flow_table.hash = flow_hash(&flow->key, &flow->mask->range); |
618ed0c8 PS |
882 | ti = ovsl_dereference(table->ti); |
883 | table_instance_insert(ti, flow); | |
884 | table->count++; | |
885 | ||
886 | /* Expand table, if necessary, to make room. */ | |
887 | if (table->count > ti->n_buckets) | |
74ed7ab9 | 888 | new_ti = table_instance_expand(ti, false); |
618ed0c8 | 889 | else if (time_after(jiffies, table->last_rehash + REHASH_INTERVAL)) |
74ed7ab9 | 890 | new_ti = table_instance_rehash(ti, ti->n_buckets, false); |
618ed0c8 PS |
891 | |
892 | if (new_ti) { | |
893 | rcu_assign_pointer(table->ti, new_ti); | |
74ed7ab9 | 894 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); |
618ed0c8 PS |
895 | table->last_rehash = jiffies; |
896 | } | |
d29ab6f8 JS |
897 | } |
898 | ||
74ed7ab9 JS |
899 | /* Must be called with OVS mutex held. */ |
900 | static void flow_ufid_insert(struct flow_table *table, struct sw_flow *flow) | |
901 | { | |
902 | struct table_instance *ti; | |
903 | ||
904 | flow->ufid_table.hash = ufid_hash(&flow->id); | |
905 | ti = ovsl_dereference(table->ufid_ti); | |
906 | ufid_table_instance_insert(ti, flow); | |
907 | table->ufid_count++; | |
908 | ||
909 | /* Expand table, if necessary, to make room. */ | |
910 | if (table->ufid_count > ti->n_buckets) { | |
911 | struct table_instance *new_ti; | |
912 | ||
913 | new_ti = table_instance_expand(ti, true); | |
914 | if (new_ti) { | |
915 | rcu_assign_pointer(table->ufid_ti, new_ti); | |
916 | call_rcu(&ti->rcu, flow_tbl_destroy_rcu_cb); | |
917 | } | |
918 | } | |
919 | } | |
920 | ||
d29ab6f8 JS |
921 | /* Must be called with OVS mutex held. */ |
922 | int ovs_flow_tbl_insert(struct flow_table *table, struct sw_flow *flow, | |
923 | const struct sw_flow_mask *mask) | |
924 | { | |
925 | int err; | |
926 | ||
927 | err = flow_mask_insert(table, flow, mask); | |
928 | if (err) | |
929 | return err; | |
930 | flow_key_insert(table, flow); | |
74ed7ab9 JS |
931 | if (ovs_identifier_is_ufid(&flow->id)) |
932 | flow_ufid_insert(table, flow); | |
d29ab6f8 | 933 | |
618ed0c8 | 934 | return 0; |
e6445719 PS |
935 | } |
936 | ||
937 | /* Initializes the flow module. | |
938 | * Returns zero if successful or a negative error code. */ | |
939 | int ovs_flow_init(void) | |
940 | { | |
941 | BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); | |
942 | BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); | |
943 | ||
63e7959c | 944 | flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow) |
db74a333 | 945 | + (nr_cpu_ids |
aef833c5 | 946 | * sizeof(struct sw_flow_stats *)), |
63e7959c | 947 | 0, 0, NULL); |
e6445719 PS |
948 | if (flow_cache == NULL) |
949 | return -ENOMEM; | |
950 | ||
63e7959c | 951 | flow_stats_cache |
aef833c5 | 952 | = kmem_cache_create("sw_flow_stats", sizeof(struct sw_flow_stats), |
63e7959c JR |
953 | 0, SLAB_HWCACHE_ALIGN, NULL); |
954 | if (flow_stats_cache == NULL) { | |
955 | kmem_cache_destroy(flow_cache); | |
956 | flow_cache = NULL; | |
957 | return -ENOMEM; | |
958 | } | |
959 | ||
e6445719 PS |
960 | return 0; |
961 | } | |
962 | ||
963 | /* Uninitializes the flow module. */ | |
964 | void ovs_flow_exit(void) | |
965 | { | |
63e7959c | 966 | kmem_cache_destroy(flow_stats_cache); |
e6445719 PS |
967 | kmem_cache_destroy(flow_cache); |
968 | } |