xref: /linux/net/core/flow_offload.c (revision 2c7b9b936bdc6ff0a7a5f6aed8e55d27ca14807d)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #include <linux/kernel.h>
3 #include <linux/slab.h>
4 #include <net/act_api.h>
5 #include <net/flow_offload.h>
6 #include <linux/rtnetlink.h>
7 #include <linux/mutex.h>
8 #include <linux/rhashtable.h>
9 
10 struct flow_rule *flow_rule_alloc(unsigned int num_actions)
11 {
12 	struct flow_rule *rule;
13 	int i;
14 
15 	rule = kzalloc(struct_size(rule, action.entries, num_actions),
16 		       GFP_KERNEL);
17 	if (!rule)
18 		return NULL;
19 
20 	rule->action.num_entries = num_actions;
21 	/* Pre-fill each action hw_stats with DONT_CARE.
22 	 * Caller can override this if it wants stats for a given action.
23 	 */
24 	for (i = 0; i < num_actions; i++)
25 		rule->action.entries[i].hw_stats = FLOW_ACTION_HW_STATS_DONT_CARE;
26 
27 	return rule;
28 }
29 EXPORT_SYMBOL(flow_rule_alloc);
30 
31 struct flow_offload_action *offload_action_alloc(unsigned int num_actions)
32 {
33 	struct flow_offload_action *fl_action;
34 	int i;
35 
36 	fl_action = kzalloc(struct_size(fl_action, action.entries, num_actions),
37 			    GFP_KERNEL);
38 	if (!fl_action)
39 		return NULL;
40 
41 	fl_action->action.num_entries = num_actions;
42 	/* Pre-fill each action hw_stats with DONT_CARE.
43 	 * Caller can override this if it wants stats for a given action.
44 	 */
45 	for (i = 0; i < num_actions; i++)
46 		fl_action->action.entries[i].hw_stats = FLOW_ACTION_HW_STATS_DONT_CARE;
47 
48 	return fl_action;
49 }
50 
51 #define FLOW_DISSECTOR_MATCH(__rule, __type, __out)				\
52 	const struct flow_match *__m = &(__rule)->match;			\
53 	struct flow_dissector *__d = (__m)->dissector;				\
54 										\
55 	(__out)->key = skb_flow_dissector_target(__d, __type, (__m)->key);	\
56 	(__out)->mask = skb_flow_dissector_target(__d, __type, (__m)->mask);	\
57 
58 void flow_rule_match_meta(const struct flow_rule *rule,
59 			  struct flow_match_meta *out)
60 {
61 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_META, out);
62 }
63 EXPORT_SYMBOL(flow_rule_match_meta);
64 
65 void flow_rule_match_basic(const struct flow_rule *rule,
66 			   struct flow_match_basic *out)
67 {
68 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_BASIC, out);
69 }
70 EXPORT_SYMBOL(flow_rule_match_basic);
71 
72 void flow_rule_match_control(const struct flow_rule *rule,
73 			     struct flow_match_control *out)
74 {
75 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_CONTROL, out);
76 }
77 EXPORT_SYMBOL(flow_rule_match_control);
78 
79 void flow_rule_match_eth_addrs(const struct flow_rule *rule,
80 			       struct flow_match_eth_addrs *out)
81 {
82 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS, out);
83 }
84 EXPORT_SYMBOL(flow_rule_match_eth_addrs);
85 
86 void flow_rule_match_vlan(const struct flow_rule *rule,
87 			  struct flow_match_vlan *out)
88 {
89 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_VLAN, out);
90 }
91 EXPORT_SYMBOL(flow_rule_match_vlan);
92 
93 void flow_rule_match_cvlan(const struct flow_rule *rule,
94 			   struct flow_match_vlan *out)
95 {
96 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_CVLAN, out);
97 }
98 EXPORT_SYMBOL(flow_rule_match_cvlan);
99 
100 void flow_rule_match_ipv4_addrs(const struct flow_rule *rule,
101 				struct flow_match_ipv4_addrs *out)
102 {
103 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_IPV4_ADDRS, out);
104 }
105 EXPORT_SYMBOL(flow_rule_match_ipv4_addrs);
106 
107 void flow_rule_match_ipv6_addrs(const struct flow_rule *rule,
108 				struct flow_match_ipv6_addrs *out)
109 {
110 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_IPV6_ADDRS, out);
111 }
112 EXPORT_SYMBOL(flow_rule_match_ipv6_addrs);
113 
114 void flow_rule_match_ip(const struct flow_rule *rule,
115 			struct flow_match_ip *out)
116 {
117 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_IP, out);
118 }
119 EXPORT_SYMBOL(flow_rule_match_ip);
120 
121 void flow_rule_match_ports(const struct flow_rule *rule,
122 			   struct flow_match_ports *out)
123 {
124 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_PORTS, out);
125 }
126 EXPORT_SYMBOL(flow_rule_match_ports);
127 
128 void flow_rule_match_ports_range(const struct flow_rule *rule,
129 				 struct flow_match_ports_range *out)
130 {
131 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_PORTS_RANGE, out);
132 }
133 EXPORT_SYMBOL(flow_rule_match_ports_range);
134 
135 void flow_rule_match_tcp(const struct flow_rule *rule,
136 			 struct flow_match_tcp *out)
137 {
138 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_TCP, out);
139 }
140 EXPORT_SYMBOL(flow_rule_match_tcp);
141 
142 void flow_rule_match_icmp(const struct flow_rule *rule,
143 			  struct flow_match_icmp *out)
144 {
145 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ICMP, out);
146 }
147 EXPORT_SYMBOL(flow_rule_match_icmp);
148 
149 void flow_rule_match_mpls(const struct flow_rule *rule,
150 			  struct flow_match_mpls *out)
151 {
152 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_MPLS, out);
153 }
154 EXPORT_SYMBOL(flow_rule_match_mpls);
155 
156 void flow_rule_match_enc_control(const struct flow_rule *rule,
157 				 struct flow_match_control *out)
158 {
159 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_CONTROL, out);
160 }
161 EXPORT_SYMBOL(flow_rule_match_enc_control);
162 
163 void flow_rule_match_enc_ipv4_addrs(const struct flow_rule *rule,
164 				    struct flow_match_ipv4_addrs *out)
165 {
166 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS, out);
167 }
168 EXPORT_SYMBOL(flow_rule_match_enc_ipv4_addrs);
169 
170 void flow_rule_match_enc_ipv6_addrs(const struct flow_rule *rule,
171 				    struct flow_match_ipv6_addrs *out)
172 {
173 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS, out);
174 }
175 EXPORT_SYMBOL(flow_rule_match_enc_ipv6_addrs);
176 
177 void flow_rule_match_enc_ip(const struct flow_rule *rule,
178 			    struct flow_match_ip *out)
179 {
180 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_IP, out);
181 }
182 EXPORT_SYMBOL(flow_rule_match_enc_ip);
183 
184 void flow_rule_match_enc_ports(const struct flow_rule *rule,
185 			       struct flow_match_ports *out)
186 {
187 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_PORTS, out);
188 }
189 EXPORT_SYMBOL(flow_rule_match_enc_ports);
190 
191 void flow_rule_match_enc_keyid(const struct flow_rule *rule,
192 			       struct flow_match_enc_keyid *out)
193 {
194 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_KEYID, out);
195 }
196 EXPORT_SYMBOL(flow_rule_match_enc_keyid);
197 
198 void flow_rule_match_enc_opts(const struct flow_rule *rule,
199 			      struct flow_match_enc_opts *out)
200 {
201 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_ENC_OPTS, out);
202 }
203 EXPORT_SYMBOL(flow_rule_match_enc_opts);
204 
205 struct flow_action_cookie *flow_action_cookie_create(void *data,
206 						     unsigned int len,
207 						     gfp_t gfp)
208 {
209 	struct flow_action_cookie *cookie;
210 
211 	cookie = kmalloc(sizeof(*cookie) + len, gfp);
212 	if (!cookie)
213 		return NULL;
214 	cookie->cookie_len = len;
215 	memcpy(cookie->cookie, data, len);
216 	return cookie;
217 }
218 EXPORT_SYMBOL(flow_action_cookie_create);
219 
220 void flow_action_cookie_destroy(struct flow_action_cookie *cookie)
221 {
222 	kfree(cookie);
223 }
224 EXPORT_SYMBOL(flow_action_cookie_destroy);
225 
226 void flow_rule_match_ct(const struct flow_rule *rule,
227 			struct flow_match_ct *out)
228 {
229 	FLOW_DISSECTOR_MATCH(rule, FLOW_DISSECTOR_KEY_CT, out);
230 }
231 EXPORT_SYMBOL(flow_rule_match_ct);
232 
233 struct flow_block_cb *flow_block_cb_alloc(flow_setup_cb_t *cb,
234 					  void *cb_ident, void *cb_priv,
235 					  void (*release)(void *cb_priv))
236 {
237 	struct flow_block_cb *block_cb;
238 
239 	block_cb = kzalloc(sizeof(*block_cb), GFP_KERNEL);
240 	if (!block_cb)
241 		return ERR_PTR(-ENOMEM);
242 
243 	block_cb->cb = cb;
244 	block_cb->cb_ident = cb_ident;
245 	block_cb->cb_priv = cb_priv;
246 	block_cb->release = release;
247 
248 	return block_cb;
249 }
250 EXPORT_SYMBOL(flow_block_cb_alloc);
251 
252 void flow_block_cb_free(struct flow_block_cb *block_cb)
253 {
254 	if (block_cb->release)
255 		block_cb->release(block_cb->cb_priv);
256 
257 	kfree(block_cb);
258 }
259 EXPORT_SYMBOL(flow_block_cb_free);
260 
261 struct flow_block_cb *flow_block_cb_lookup(struct flow_block *block,
262 					   flow_setup_cb_t *cb, void *cb_ident)
263 {
264 	struct flow_block_cb *block_cb;
265 
266 	list_for_each_entry(block_cb, &block->cb_list, list) {
267 		if (block_cb->cb == cb &&
268 		    block_cb->cb_ident == cb_ident)
269 			return block_cb;
270 	}
271 
272 	return NULL;
273 }
274 EXPORT_SYMBOL(flow_block_cb_lookup);
275 
276 void *flow_block_cb_priv(struct flow_block_cb *block_cb)
277 {
278 	return block_cb->cb_priv;
279 }
280 EXPORT_SYMBOL(flow_block_cb_priv);
281 
282 void flow_block_cb_incref(struct flow_block_cb *block_cb)
283 {
284 	block_cb->refcnt++;
285 }
286 EXPORT_SYMBOL(flow_block_cb_incref);
287 
288 unsigned int flow_block_cb_decref(struct flow_block_cb *block_cb)
289 {
290 	return --block_cb->refcnt;
291 }
292 EXPORT_SYMBOL(flow_block_cb_decref);
293 
294 bool flow_block_cb_is_busy(flow_setup_cb_t *cb, void *cb_ident,
295 			   struct list_head *driver_block_list)
296 {
297 	struct flow_block_cb *block_cb;
298 
299 	list_for_each_entry(block_cb, driver_block_list, driver_list) {
300 		if (block_cb->cb == cb &&
301 		    block_cb->cb_ident == cb_ident)
302 			return true;
303 	}
304 
305 	return false;
306 }
307 EXPORT_SYMBOL(flow_block_cb_is_busy);
308 
309 int flow_block_cb_setup_simple(struct flow_block_offload *f,
310 			       struct list_head *driver_block_list,
311 			       flow_setup_cb_t *cb,
312 			       void *cb_ident, void *cb_priv,
313 			       bool ingress_only)
314 {
315 	struct flow_block_cb *block_cb;
316 
317 	if (ingress_only &&
318 	    f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
319 		return -EOPNOTSUPP;
320 
321 	f->driver_block_list = driver_block_list;
322 
323 	switch (f->command) {
324 	case FLOW_BLOCK_BIND:
325 		if (flow_block_cb_is_busy(cb, cb_ident, driver_block_list))
326 			return -EBUSY;
327 
328 		block_cb = flow_block_cb_alloc(cb, cb_ident, cb_priv, NULL);
329 		if (IS_ERR(block_cb))
330 			return PTR_ERR(block_cb);
331 
332 		flow_block_cb_add(block_cb, f);
333 		list_add_tail(&block_cb->driver_list, driver_block_list);
334 		return 0;
335 	case FLOW_BLOCK_UNBIND:
336 		block_cb = flow_block_cb_lookup(f->block, cb, cb_ident);
337 		if (!block_cb)
338 			return -ENOENT;
339 
340 		flow_block_cb_remove(block_cb, f);
341 		list_del(&block_cb->driver_list);
342 		return 0;
343 	default:
344 		return -EOPNOTSUPP;
345 	}
346 }
347 EXPORT_SYMBOL(flow_block_cb_setup_simple);
348 
349 static DEFINE_MUTEX(flow_indr_block_lock);
350 static LIST_HEAD(flow_block_indr_list);
351 static LIST_HEAD(flow_block_indr_dev_list);
352 static LIST_HEAD(flow_indir_dev_list);
353 
354 struct flow_indr_dev {
355 	struct list_head		list;
356 	flow_indr_block_bind_cb_t	*cb;
357 	void				*cb_priv;
358 	refcount_t			refcnt;
359 };
360 
361 static struct flow_indr_dev *flow_indr_dev_alloc(flow_indr_block_bind_cb_t *cb,
362 						 void *cb_priv)
363 {
364 	struct flow_indr_dev *indr_dev;
365 
366 	indr_dev = kmalloc(sizeof(*indr_dev), GFP_KERNEL);
367 	if (!indr_dev)
368 		return NULL;
369 
370 	indr_dev->cb		= cb;
371 	indr_dev->cb_priv	= cb_priv;
372 	refcount_set(&indr_dev->refcnt, 1);
373 
374 	return indr_dev;
375 }
376 
377 struct flow_indir_dev_info {
378 	void *data;
379 	struct net_device *dev;
380 	struct Qdisc *sch;
381 	enum tc_setup_type type;
382 	void (*cleanup)(struct flow_block_cb *block_cb);
383 	struct list_head list;
384 	enum flow_block_command command;
385 	enum flow_block_binder_type binder_type;
386 	struct list_head *cb_list;
387 };
388 
389 static void existing_qdiscs_register(flow_indr_block_bind_cb_t *cb, void *cb_priv)
390 {
391 	struct flow_block_offload bo;
392 	struct flow_indir_dev_info *cur;
393 
394 	list_for_each_entry(cur, &flow_indir_dev_list, list) {
395 		memset(&bo, 0, sizeof(bo));
396 		bo.command = cur->command;
397 		bo.binder_type = cur->binder_type;
398 		INIT_LIST_HEAD(&bo.cb_list);
399 		cb(cur->dev, cur->sch, cb_priv, cur->type, &bo, cur->data, cur->cleanup);
400 		list_splice(&bo.cb_list, cur->cb_list);
401 	}
402 }
403 
404 int flow_indr_dev_register(flow_indr_block_bind_cb_t *cb, void *cb_priv)
405 {
406 	struct flow_indr_dev *indr_dev;
407 
408 	mutex_lock(&flow_indr_block_lock);
409 	list_for_each_entry(indr_dev, &flow_block_indr_dev_list, list) {
410 		if (indr_dev->cb == cb &&
411 		    indr_dev->cb_priv == cb_priv) {
412 			refcount_inc(&indr_dev->refcnt);
413 			mutex_unlock(&flow_indr_block_lock);
414 			return 0;
415 		}
416 	}
417 
418 	indr_dev = flow_indr_dev_alloc(cb, cb_priv);
419 	if (!indr_dev) {
420 		mutex_unlock(&flow_indr_block_lock);
421 		return -ENOMEM;
422 	}
423 
424 	list_add(&indr_dev->list, &flow_block_indr_dev_list);
425 	existing_qdiscs_register(cb, cb_priv);
426 	mutex_unlock(&flow_indr_block_lock);
427 
428 	tcf_action_reoffload_cb(cb, cb_priv, true);
429 
430 	return 0;
431 }
432 EXPORT_SYMBOL(flow_indr_dev_register);
433 
434 static void __flow_block_indr_cleanup(void (*release)(void *cb_priv),
435 				      void *cb_priv,
436 				      struct list_head *cleanup_list)
437 {
438 	struct flow_block_cb *this, *next;
439 
440 	list_for_each_entry_safe(this, next, &flow_block_indr_list, indr.list) {
441 		if (this->release == release &&
442 		    this->indr.cb_priv == cb_priv)
443 			list_move(&this->indr.list, cleanup_list);
444 	}
445 }
446 
447 static void flow_block_indr_notify(struct list_head *cleanup_list)
448 {
449 	struct flow_block_cb *this, *next;
450 
451 	list_for_each_entry_safe(this, next, cleanup_list, indr.list) {
452 		list_del(&this->indr.list);
453 		this->indr.cleanup(this);
454 	}
455 }
456 
457 void flow_indr_dev_unregister(flow_indr_block_bind_cb_t *cb, void *cb_priv,
458 			      void (*release)(void *cb_priv))
459 {
460 	struct flow_indr_dev *this, *next, *indr_dev = NULL;
461 	LIST_HEAD(cleanup_list);
462 
463 	mutex_lock(&flow_indr_block_lock);
464 	list_for_each_entry_safe(this, next, &flow_block_indr_dev_list, list) {
465 		if (this->cb == cb &&
466 		    this->cb_priv == cb_priv &&
467 		    refcount_dec_and_test(&this->refcnt)) {
468 			indr_dev = this;
469 			list_del(&indr_dev->list);
470 			break;
471 		}
472 	}
473 
474 	if (!indr_dev) {
475 		mutex_unlock(&flow_indr_block_lock);
476 		return;
477 	}
478 
479 	__flow_block_indr_cleanup(release, cb_priv, &cleanup_list);
480 	mutex_unlock(&flow_indr_block_lock);
481 
482 	tcf_action_reoffload_cb(cb, cb_priv, false);
483 	flow_block_indr_notify(&cleanup_list);
484 	kfree(indr_dev);
485 }
486 EXPORT_SYMBOL(flow_indr_dev_unregister);
487 
488 static void flow_block_indr_init(struct flow_block_cb *flow_block,
489 				 struct flow_block_offload *bo,
490 				 struct net_device *dev, struct Qdisc *sch, void *data,
491 				 void *cb_priv,
492 				 void (*cleanup)(struct flow_block_cb *block_cb))
493 {
494 	flow_block->indr.binder_type = bo->binder_type;
495 	flow_block->indr.data = data;
496 	flow_block->indr.cb_priv = cb_priv;
497 	flow_block->indr.dev = dev;
498 	flow_block->indr.sch = sch;
499 	flow_block->indr.cleanup = cleanup;
500 }
501 
502 struct flow_block_cb *flow_indr_block_cb_alloc(flow_setup_cb_t *cb,
503 					       void *cb_ident, void *cb_priv,
504 					       void (*release)(void *cb_priv),
505 					       struct flow_block_offload *bo,
506 					       struct net_device *dev,
507 					       struct Qdisc *sch, void *data,
508 					       void *indr_cb_priv,
509 					       void (*cleanup)(struct flow_block_cb *block_cb))
510 {
511 	struct flow_block_cb *block_cb;
512 
513 	block_cb = flow_block_cb_alloc(cb, cb_ident, cb_priv, release);
514 	if (IS_ERR(block_cb))
515 		goto out;
516 
517 	flow_block_indr_init(block_cb, bo, dev, sch, data, indr_cb_priv, cleanup);
518 	list_add(&block_cb->indr.list, &flow_block_indr_list);
519 
520 out:
521 	return block_cb;
522 }
523 EXPORT_SYMBOL(flow_indr_block_cb_alloc);
524 
525 static struct flow_indir_dev_info *find_indir_dev(void *data)
526 {
527 	struct flow_indir_dev_info *cur;
528 
529 	list_for_each_entry(cur, &flow_indir_dev_list, list) {
530 		if (cur->data == data)
531 			return cur;
532 	}
533 	return NULL;
534 }
535 
536 static int indir_dev_add(void *data, struct net_device *dev, struct Qdisc *sch,
537 			 enum tc_setup_type type, void (*cleanup)(struct flow_block_cb *block_cb),
538 			 struct flow_block_offload *bo)
539 {
540 	struct flow_indir_dev_info *info;
541 
542 	info = find_indir_dev(data);
543 	if (info)
544 		return -EEXIST;
545 
546 	info = kzalloc(sizeof(*info), GFP_KERNEL);
547 	if (!info)
548 		return -ENOMEM;
549 
550 	info->data = data;
551 	info->dev = dev;
552 	info->sch = sch;
553 	info->type = type;
554 	info->cleanup = cleanup;
555 	info->command = bo->command;
556 	info->binder_type = bo->binder_type;
557 	info->cb_list = bo->cb_list_head;
558 
559 	list_add(&info->list, &flow_indir_dev_list);
560 	return 0;
561 }
562 
563 static int indir_dev_remove(void *data)
564 {
565 	struct flow_indir_dev_info *info;
566 
567 	info = find_indir_dev(data);
568 	if (!info)
569 		return -ENOENT;
570 
571 	list_del(&info->list);
572 
573 	kfree(info);
574 	return 0;
575 }
576 
577 int flow_indr_dev_setup_offload(struct net_device *dev,	struct Qdisc *sch,
578 				enum tc_setup_type type, void *data,
579 				struct flow_block_offload *bo,
580 				void (*cleanup)(struct flow_block_cb *block_cb))
581 {
582 	struct flow_indr_dev *this;
583 	u32 count = 0;
584 	int err;
585 
586 	mutex_lock(&flow_indr_block_lock);
587 	if (bo) {
588 		if (bo->command == FLOW_BLOCK_BIND)
589 			indir_dev_add(data, dev, sch, type, cleanup, bo);
590 		else if (bo->command == FLOW_BLOCK_UNBIND)
591 			indir_dev_remove(data);
592 	}
593 
594 	list_for_each_entry(this, &flow_block_indr_dev_list, list) {
595 		err = this->cb(dev, sch, this->cb_priv, type, bo, data, cleanup);
596 		if (!err)
597 			count++;
598 	}
599 
600 	mutex_unlock(&flow_indr_block_lock);
601 
602 	return (bo && list_empty(&bo->cb_list)) ? -EOPNOTSUPP : count;
603 }
604 EXPORT_SYMBOL(flow_indr_dev_setup_offload);
605 
606 bool flow_indr_dev_exists(void)
607 {
608 	return !list_empty(&flow_block_indr_dev_list);
609 }
610 EXPORT_SYMBOL(flow_indr_dev_exists);
611