xref: /openbmc/linux/net/bridge/br_mrp.c (revision 808643ea)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 
3 #include <linux/mrp_bridge.h>
4 #include "br_private_mrp.h"
5 
6 static const u8 mrp_test_dmac[ETH_ALEN] = { 0x1, 0x15, 0x4e, 0x0, 0x0, 0x1 };
7 static const u8 mrp_in_test_dmac[ETH_ALEN] = { 0x1, 0x15, 0x4e, 0x0, 0x0, 0x3 };
8 
9 static int br_mrp_process(struct net_bridge_port *p, struct sk_buff *skb);
10 
11 static struct br_frame_type mrp_frame_type __read_mostly = {
12 	.type = cpu_to_be16(ETH_P_MRP),
13 	.frame_handler = br_mrp_process,
14 };
15 
16 static bool br_mrp_is_ring_port(struct net_bridge_port *p_port,
17 				struct net_bridge_port *s_port,
18 				struct net_bridge_port *port)
19 {
20 	if (port == p_port ||
21 	    port == s_port)
22 		return true;
23 
24 	return false;
25 }
26 
27 static bool br_mrp_is_in_port(struct net_bridge_port *i_port,
28 			      struct net_bridge_port *port)
29 {
30 	if (port == i_port)
31 		return true;
32 
33 	return false;
34 }
35 
36 static struct net_bridge_port *br_mrp_get_port(struct net_bridge *br,
37 					       u32 ifindex)
38 {
39 	struct net_bridge_port *res = NULL;
40 	struct net_bridge_port *port;
41 
42 	list_for_each_entry(port, &br->port_list, list) {
43 		if (port->dev->ifindex == ifindex) {
44 			res = port;
45 			break;
46 		}
47 	}
48 
49 	return res;
50 }
51 
52 static struct br_mrp *br_mrp_find_id(struct net_bridge *br, u32 ring_id)
53 {
54 	struct br_mrp *res = NULL;
55 	struct br_mrp *mrp;
56 
57 	hlist_for_each_entry_rcu(mrp, &br->mrp_list, list,
58 				 lockdep_rtnl_is_held()) {
59 		if (mrp->ring_id == ring_id) {
60 			res = mrp;
61 			break;
62 		}
63 	}
64 
65 	return res;
66 }
67 
68 static struct br_mrp *br_mrp_find_in_id(struct net_bridge *br, u32 in_id)
69 {
70 	struct br_mrp *res = NULL;
71 	struct br_mrp *mrp;
72 
73 	hlist_for_each_entry_rcu(mrp, &br->mrp_list, list,
74 				 lockdep_rtnl_is_held()) {
75 		if (mrp->in_id == in_id) {
76 			res = mrp;
77 			break;
78 		}
79 	}
80 
81 	return res;
82 }
83 
84 static bool br_mrp_unique_ifindex(struct net_bridge *br, u32 ifindex)
85 {
86 	struct br_mrp *mrp;
87 
88 	hlist_for_each_entry_rcu(mrp, &br->mrp_list, list,
89 				 lockdep_rtnl_is_held()) {
90 		struct net_bridge_port *p;
91 
92 		p = rtnl_dereference(mrp->p_port);
93 		if (p && p->dev->ifindex == ifindex)
94 			return false;
95 
96 		p = rtnl_dereference(mrp->s_port);
97 		if (p && p->dev->ifindex == ifindex)
98 			return false;
99 
100 		p = rtnl_dereference(mrp->i_port);
101 		if (p && p->dev->ifindex == ifindex)
102 			return false;
103 	}
104 
105 	return true;
106 }
107 
108 static struct br_mrp *br_mrp_find_port(struct net_bridge *br,
109 				       struct net_bridge_port *p)
110 {
111 	struct br_mrp *res = NULL;
112 	struct br_mrp *mrp;
113 
114 	hlist_for_each_entry_rcu(mrp, &br->mrp_list, list,
115 				 lockdep_rtnl_is_held()) {
116 		if (rcu_access_pointer(mrp->p_port) == p ||
117 		    rcu_access_pointer(mrp->s_port) == p ||
118 		    rcu_access_pointer(mrp->i_port) == p) {
119 			res = mrp;
120 			break;
121 		}
122 	}
123 
124 	return res;
125 }
126 
127 static int br_mrp_next_seq(struct br_mrp *mrp)
128 {
129 	mrp->seq_id++;
130 	return mrp->seq_id;
131 }
132 
133 static struct sk_buff *br_mrp_skb_alloc(struct net_bridge_port *p,
134 					const u8 *src, const u8 *dst)
135 {
136 	struct ethhdr *eth_hdr;
137 	struct sk_buff *skb;
138 	__be16 *version;
139 
140 	skb = dev_alloc_skb(MRP_MAX_FRAME_LENGTH);
141 	if (!skb)
142 		return NULL;
143 
144 	skb->dev = p->dev;
145 	skb->protocol = htons(ETH_P_MRP);
146 	skb->priority = MRP_FRAME_PRIO;
147 	skb_reserve(skb, sizeof(*eth_hdr));
148 
149 	eth_hdr = skb_push(skb, sizeof(*eth_hdr));
150 	ether_addr_copy(eth_hdr->h_dest, dst);
151 	ether_addr_copy(eth_hdr->h_source, src);
152 	eth_hdr->h_proto = htons(ETH_P_MRP);
153 
154 	version = skb_put(skb, sizeof(*version));
155 	*version = cpu_to_be16(MRP_VERSION);
156 
157 	return skb;
158 }
159 
160 static void br_mrp_skb_tlv(struct sk_buff *skb,
161 			   enum br_mrp_tlv_header_type type,
162 			   u8 length)
163 {
164 	struct br_mrp_tlv_hdr *hdr;
165 
166 	hdr = skb_put(skb, sizeof(*hdr));
167 	hdr->type = type;
168 	hdr->length = length;
169 }
170 
171 static void br_mrp_skb_common(struct sk_buff *skb, struct br_mrp *mrp)
172 {
173 	struct br_mrp_common_hdr *hdr;
174 
175 	br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_COMMON, sizeof(*hdr));
176 
177 	hdr = skb_put(skb, sizeof(*hdr));
178 	hdr->seq_id = cpu_to_be16(br_mrp_next_seq(mrp));
179 	memset(hdr->domain, 0xff, MRP_DOMAIN_UUID_LENGTH);
180 }
181 
182 static struct sk_buff *br_mrp_alloc_test_skb(struct br_mrp *mrp,
183 					     struct net_bridge_port *p,
184 					     enum br_mrp_port_role_type port_role)
185 {
186 	struct br_mrp_ring_test_hdr *hdr = NULL;
187 	struct sk_buff *skb = NULL;
188 
189 	if (!p)
190 		return NULL;
191 
192 	skb = br_mrp_skb_alloc(p, p->dev->dev_addr, mrp_test_dmac);
193 	if (!skb)
194 		return NULL;
195 
196 	br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_RING_TEST, sizeof(*hdr));
197 	hdr = skb_put(skb, sizeof(*hdr));
198 
199 	hdr->prio = cpu_to_be16(mrp->prio);
200 	ether_addr_copy(hdr->sa, p->br->dev->dev_addr);
201 	hdr->port_role = cpu_to_be16(port_role);
202 	hdr->state = cpu_to_be16(mrp->ring_state);
203 	hdr->transitions = cpu_to_be16(mrp->ring_transitions);
204 	hdr->timestamp = cpu_to_be32(jiffies_to_msecs(jiffies));
205 
206 	br_mrp_skb_common(skb, mrp);
207 
208 	/* In case the node behaves as MRA then the Test frame needs to have
209 	 * an Option TLV which includes eventually a sub-option TLV that has
210 	 * the type AUTO_MGR
211 	 */
212 	if (mrp->ring_role == BR_MRP_RING_ROLE_MRA) {
213 		struct br_mrp_sub_option1_hdr *sub_opt = NULL;
214 		struct br_mrp_tlv_hdr *sub_tlv = NULL;
215 		struct br_mrp_oui_hdr *oui = NULL;
216 		u8 length;
217 
218 		length = sizeof(*sub_opt) + sizeof(*sub_tlv) + sizeof(oui) +
219 			MRP_OPT_PADDING;
220 		br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_OPTION, length);
221 
222 		oui = skb_put(skb, sizeof(*oui));
223 		memset(oui, 0x0, sizeof(*oui));
224 		sub_opt = skb_put(skb, sizeof(*sub_opt));
225 		memset(sub_opt, 0x0, sizeof(*sub_opt));
226 
227 		sub_tlv = skb_put(skb, sizeof(*sub_tlv));
228 		sub_tlv->type = BR_MRP_SUB_TLV_HEADER_TEST_AUTO_MGR;
229 
230 		/* 32 bit alligment shall be ensured therefore add 2 bytes */
231 		skb_put(skb, MRP_OPT_PADDING);
232 	}
233 
234 	br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_END, 0x0);
235 
236 	return skb;
237 }
238 
239 static struct sk_buff *br_mrp_alloc_in_test_skb(struct br_mrp *mrp,
240 						struct net_bridge_port *p,
241 						enum br_mrp_port_role_type port_role)
242 {
243 	struct br_mrp_in_test_hdr *hdr = NULL;
244 	struct sk_buff *skb = NULL;
245 
246 	if (!p)
247 		return NULL;
248 
249 	skb = br_mrp_skb_alloc(p, p->dev->dev_addr, mrp_in_test_dmac);
250 	if (!skb)
251 		return NULL;
252 
253 	br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_IN_TEST, sizeof(*hdr));
254 	hdr = skb_put(skb, sizeof(*hdr));
255 
256 	hdr->id = cpu_to_be16(mrp->in_id);
257 	ether_addr_copy(hdr->sa, p->br->dev->dev_addr);
258 	hdr->port_role = cpu_to_be16(port_role);
259 	hdr->state = cpu_to_be16(mrp->in_state);
260 	hdr->transitions = cpu_to_be16(mrp->in_transitions);
261 	hdr->timestamp = cpu_to_be32(jiffies_to_msecs(jiffies));
262 
263 	br_mrp_skb_common(skb, mrp);
264 	br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_END, 0x0);
265 
266 	return skb;
267 }
268 
269 /* This function is continuously called in the following cases:
270  * - when node role is MRM, in this case test_monitor is always set to false
271  *   because it needs to notify the userspace that the ring is open and needs to
272  *   send MRP_Test frames
273  * - when node role is MRA, there are 2 subcases:
274  *     - when MRA behaves as MRM, in this case is similar with MRM role
275  *     - when MRA behaves as MRC, in this case test_monitor is set to true,
276  *       because it needs to detect when it stops seeing MRP_Test frames
277  *       from MRM node but it doesn't need to send MRP_Test frames.
278  */
279 static void br_mrp_test_work_expired(struct work_struct *work)
280 {
281 	struct delayed_work *del_work = to_delayed_work(work);
282 	struct br_mrp *mrp = container_of(del_work, struct br_mrp, test_work);
283 	struct net_bridge_port *p;
284 	bool notify_open = false;
285 	struct sk_buff *skb;
286 
287 	if (time_before_eq(mrp->test_end, jiffies))
288 		return;
289 
290 	if (mrp->test_count_miss < mrp->test_max_miss) {
291 		mrp->test_count_miss++;
292 	} else {
293 		/* Notify that the ring is open only if the ring state is
294 		 * closed, otherwise it would continue to notify at every
295 		 * interval.
296 		 * Also notify that the ring is open when the node has the
297 		 * role MRA and behaves as MRC. The reason is that the
298 		 * userspace needs to know when the MRM stopped sending
299 		 * MRP_Test frames so that the current node to try to take
300 		 * the role of a MRM.
301 		 */
302 		if (mrp->ring_state == BR_MRP_RING_STATE_CLOSED ||
303 		    mrp->test_monitor)
304 			notify_open = true;
305 	}
306 
307 	rcu_read_lock();
308 
309 	p = rcu_dereference(mrp->p_port);
310 	if (p) {
311 		if (!mrp->test_monitor) {
312 			skb = br_mrp_alloc_test_skb(mrp, p,
313 						    BR_MRP_PORT_ROLE_PRIMARY);
314 			if (!skb)
315 				goto out;
316 
317 			skb_reset_network_header(skb);
318 			dev_queue_xmit(skb);
319 		}
320 
321 		if (notify_open && !mrp->ring_role_offloaded)
322 			br_mrp_ring_port_open(p->dev, true);
323 	}
324 
325 	p = rcu_dereference(mrp->s_port);
326 	if (p) {
327 		if (!mrp->test_monitor) {
328 			skb = br_mrp_alloc_test_skb(mrp, p,
329 						    BR_MRP_PORT_ROLE_SECONDARY);
330 			if (!skb)
331 				goto out;
332 
333 			skb_reset_network_header(skb);
334 			dev_queue_xmit(skb);
335 		}
336 
337 		if (notify_open && !mrp->ring_role_offloaded)
338 			br_mrp_ring_port_open(p->dev, true);
339 	}
340 
341 out:
342 	rcu_read_unlock();
343 
344 	queue_delayed_work(system_wq, &mrp->test_work,
345 			   usecs_to_jiffies(mrp->test_interval));
346 }
347 
348 /* This function is continuously called when the node has the interconnect role
349  * MIM. It would generate interconnect test frames and will send them on all 3
350  * ports. But will also check if it stop receiving interconnect test frames.
351  */
352 static void br_mrp_in_test_work_expired(struct work_struct *work)
353 {
354 	struct delayed_work *del_work = to_delayed_work(work);
355 	struct br_mrp *mrp = container_of(del_work, struct br_mrp, in_test_work);
356 	struct net_bridge_port *p;
357 	bool notify_open = false;
358 	struct sk_buff *skb;
359 
360 	if (time_before_eq(mrp->in_test_end, jiffies))
361 		return;
362 
363 	if (mrp->in_test_count_miss < mrp->in_test_max_miss) {
364 		mrp->in_test_count_miss++;
365 	} else {
366 		/* Notify that the interconnect ring is open only if the
367 		 * interconnect ring state is closed, otherwise it would
368 		 * continue to notify at every interval.
369 		 */
370 		if (mrp->in_state == BR_MRP_IN_STATE_CLOSED)
371 			notify_open = true;
372 	}
373 
374 	rcu_read_lock();
375 
376 	p = rcu_dereference(mrp->p_port);
377 	if (p) {
378 		skb = br_mrp_alloc_in_test_skb(mrp, p,
379 					       BR_MRP_PORT_ROLE_PRIMARY);
380 		if (!skb)
381 			goto out;
382 
383 		skb_reset_network_header(skb);
384 		dev_queue_xmit(skb);
385 
386 		if (notify_open && !mrp->in_role_offloaded)
387 			br_mrp_in_port_open(p->dev, true);
388 	}
389 
390 	p = rcu_dereference(mrp->s_port);
391 	if (p) {
392 		skb = br_mrp_alloc_in_test_skb(mrp, p,
393 					       BR_MRP_PORT_ROLE_SECONDARY);
394 		if (!skb)
395 			goto out;
396 
397 		skb_reset_network_header(skb);
398 		dev_queue_xmit(skb);
399 
400 		if (notify_open && !mrp->in_role_offloaded)
401 			br_mrp_in_port_open(p->dev, true);
402 	}
403 
404 	p = rcu_dereference(mrp->i_port);
405 	if (p) {
406 		skb = br_mrp_alloc_in_test_skb(mrp, p,
407 					       BR_MRP_PORT_ROLE_INTER);
408 		if (!skb)
409 			goto out;
410 
411 		skb_reset_network_header(skb);
412 		dev_queue_xmit(skb);
413 
414 		if (notify_open && !mrp->in_role_offloaded)
415 			br_mrp_in_port_open(p->dev, true);
416 	}
417 
418 out:
419 	rcu_read_unlock();
420 
421 	queue_delayed_work(system_wq, &mrp->in_test_work,
422 			   usecs_to_jiffies(mrp->in_test_interval));
423 }
424 
425 /* Deletes the MRP instance.
426  * note: called under rtnl_lock
427  */
428 static void br_mrp_del_impl(struct net_bridge *br, struct br_mrp *mrp)
429 {
430 	struct net_bridge_port *p;
431 	u8 state;
432 
433 	/* Stop sending MRP_Test frames */
434 	cancel_delayed_work_sync(&mrp->test_work);
435 	br_mrp_switchdev_send_ring_test(br, mrp, 0, 0, 0, 0);
436 
437 	/* Stop sending MRP_InTest frames if has an interconnect role */
438 	cancel_delayed_work_sync(&mrp->in_test_work);
439 	br_mrp_switchdev_send_in_test(br, mrp, 0, 0, 0);
440 
441 	/* Disable the roles */
442 	br_mrp_switchdev_set_ring_role(br, mrp, BR_MRP_RING_ROLE_DISABLED);
443 	p = rtnl_dereference(mrp->i_port);
444 	if (p)
445 		br_mrp_switchdev_set_in_role(br, mrp, mrp->in_id, mrp->ring_id,
446 					     BR_MRP_IN_ROLE_DISABLED);
447 
448 	br_mrp_switchdev_del(br, mrp);
449 
450 	/* Reset the ports */
451 	p = rtnl_dereference(mrp->p_port);
452 	if (p) {
453 		spin_lock_bh(&br->lock);
454 		state = netif_running(br->dev) ?
455 				BR_STATE_FORWARDING : BR_STATE_DISABLED;
456 		p->state = state;
457 		p->flags &= ~BR_MRP_AWARE;
458 		spin_unlock_bh(&br->lock);
459 		br_mrp_port_switchdev_set_state(p, state);
460 		rcu_assign_pointer(mrp->p_port, NULL);
461 	}
462 
463 	p = rtnl_dereference(mrp->s_port);
464 	if (p) {
465 		spin_lock_bh(&br->lock);
466 		state = netif_running(br->dev) ?
467 				BR_STATE_FORWARDING : BR_STATE_DISABLED;
468 		p->state = state;
469 		p->flags &= ~BR_MRP_AWARE;
470 		spin_unlock_bh(&br->lock);
471 		br_mrp_port_switchdev_set_state(p, state);
472 		rcu_assign_pointer(mrp->s_port, NULL);
473 	}
474 
475 	p = rtnl_dereference(mrp->i_port);
476 	if (p) {
477 		spin_lock_bh(&br->lock);
478 		state = netif_running(br->dev) ?
479 				BR_STATE_FORWARDING : BR_STATE_DISABLED;
480 		p->state = state;
481 		p->flags &= ~BR_MRP_AWARE;
482 		spin_unlock_bh(&br->lock);
483 		br_mrp_port_switchdev_set_state(p, state);
484 		rcu_assign_pointer(mrp->i_port, NULL);
485 	}
486 
487 	hlist_del_rcu(&mrp->list);
488 	kfree_rcu(mrp, rcu);
489 
490 	if (hlist_empty(&br->mrp_list))
491 		br_del_frame(br, &mrp_frame_type);
492 }
493 
494 /* Adds a new MRP instance.
495  * note: called under rtnl_lock
496  */
497 int br_mrp_add(struct net_bridge *br, struct br_mrp_instance *instance)
498 {
499 	struct net_bridge_port *p;
500 	struct br_mrp *mrp;
501 	int err;
502 
503 	/* If the ring exists, it is not possible to create another one with the
504 	 * same ring_id
505 	 */
506 	mrp = br_mrp_find_id(br, instance->ring_id);
507 	if (mrp)
508 		return -EINVAL;
509 
510 	if (!br_mrp_get_port(br, instance->p_ifindex) ||
511 	    !br_mrp_get_port(br, instance->s_ifindex))
512 		return -EINVAL;
513 
514 	/* It is not possible to have the same port part of multiple rings */
515 	if (!br_mrp_unique_ifindex(br, instance->p_ifindex) ||
516 	    !br_mrp_unique_ifindex(br, instance->s_ifindex))
517 		return -EINVAL;
518 
519 	mrp = kzalloc(sizeof(*mrp), GFP_KERNEL);
520 	if (!mrp)
521 		return -ENOMEM;
522 
523 	mrp->ring_id = instance->ring_id;
524 	mrp->prio = instance->prio;
525 
526 	p = br_mrp_get_port(br, instance->p_ifindex);
527 	spin_lock_bh(&br->lock);
528 	p->state = BR_STATE_FORWARDING;
529 	p->flags |= BR_MRP_AWARE;
530 	spin_unlock_bh(&br->lock);
531 	rcu_assign_pointer(mrp->p_port, p);
532 
533 	p = br_mrp_get_port(br, instance->s_ifindex);
534 	spin_lock_bh(&br->lock);
535 	p->state = BR_STATE_FORWARDING;
536 	p->flags |= BR_MRP_AWARE;
537 	spin_unlock_bh(&br->lock);
538 	rcu_assign_pointer(mrp->s_port, p);
539 
540 	if (hlist_empty(&br->mrp_list))
541 		br_add_frame(br, &mrp_frame_type);
542 
543 	INIT_DELAYED_WORK(&mrp->test_work, br_mrp_test_work_expired);
544 	INIT_DELAYED_WORK(&mrp->in_test_work, br_mrp_in_test_work_expired);
545 	hlist_add_tail_rcu(&mrp->list, &br->mrp_list);
546 
547 	err = br_mrp_switchdev_add(br, mrp);
548 	if (err)
549 		goto delete_mrp;
550 
551 	return 0;
552 
553 delete_mrp:
554 	br_mrp_del_impl(br, mrp);
555 
556 	return err;
557 }
558 
559 /* Deletes the MRP instance from which the port is part of
560  * note: called under rtnl_lock
561  */
562 void br_mrp_port_del(struct net_bridge *br, struct net_bridge_port *p)
563 {
564 	struct br_mrp *mrp = br_mrp_find_port(br, p);
565 
566 	/* If the port is not part of a MRP instance just bail out */
567 	if (!mrp)
568 		return;
569 
570 	br_mrp_del_impl(br, mrp);
571 }
572 
573 /* Deletes existing MRP instance based on ring_id
574  * note: called under rtnl_lock
575  */
576 int br_mrp_del(struct net_bridge *br, struct br_mrp_instance *instance)
577 {
578 	struct br_mrp *mrp = br_mrp_find_id(br, instance->ring_id);
579 
580 	if (!mrp)
581 		return -EINVAL;
582 
583 	br_mrp_del_impl(br, mrp);
584 
585 	return 0;
586 }
587 
588 /* Set port state, port state can be forwarding, blocked or disabled
589  * note: already called with rtnl_lock
590  */
591 int br_mrp_set_port_state(struct net_bridge_port *p,
592 			  enum br_mrp_port_state_type state)
593 {
594 	u32 port_state;
595 
596 	if (!p || !(p->flags & BR_MRP_AWARE))
597 		return -EINVAL;
598 
599 	spin_lock_bh(&p->br->lock);
600 
601 	if (state == BR_MRP_PORT_STATE_FORWARDING)
602 		port_state = BR_STATE_FORWARDING;
603 	else
604 		port_state = BR_STATE_BLOCKING;
605 
606 	p->state = port_state;
607 	spin_unlock_bh(&p->br->lock);
608 
609 	br_mrp_port_switchdev_set_state(p, port_state);
610 
611 	return 0;
612 }
613 
614 /* Set port role, port role can be primary or secondary
615  * note: already called with rtnl_lock
616  */
617 int br_mrp_set_port_role(struct net_bridge_port *p,
618 			 enum br_mrp_port_role_type role)
619 {
620 	struct br_mrp *mrp;
621 
622 	if (!p || !(p->flags & BR_MRP_AWARE))
623 		return -EINVAL;
624 
625 	mrp = br_mrp_find_port(p->br, p);
626 
627 	if (!mrp)
628 		return -EINVAL;
629 
630 	switch (role) {
631 	case BR_MRP_PORT_ROLE_PRIMARY:
632 		rcu_assign_pointer(mrp->p_port, p);
633 		break;
634 	case BR_MRP_PORT_ROLE_SECONDARY:
635 		rcu_assign_pointer(mrp->s_port, p);
636 		break;
637 	default:
638 		return -EINVAL;
639 	}
640 
641 	br_mrp_port_switchdev_set_role(p, role);
642 
643 	return 0;
644 }
645 
646 /* Set ring state, ring state can be only Open or Closed
647  * note: already called with rtnl_lock
648  */
649 int br_mrp_set_ring_state(struct net_bridge *br,
650 			  struct br_mrp_ring_state *state)
651 {
652 	struct br_mrp *mrp = br_mrp_find_id(br, state->ring_id);
653 
654 	if (!mrp)
655 		return -EINVAL;
656 
657 	if (mrp->ring_state != state->ring_state)
658 		mrp->ring_transitions++;
659 
660 	mrp->ring_state = state->ring_state;
661 
662 	br_mrp_switchdev_set_ring_state(br, mrp, state->ring_state);
663 
664 	return 0;
665 }
666 
667 /* Set ring role, ring role can be only MRM(Media Redundancy Manager) or
668  * MRC(Media Redundancy Client).
669  * note: already called with rtnl_lock
670  */
671 int br_mrp_set_ring_role(struct net_bridge *br,
672 			 struct br_mrp_ring_role *role)
673 {
674 	struct br_mrp *mrp = br_mrp_find_id(br, role->ring_id);
675 	enum br_mrp_hw_support support;
676 
677 	if (!mrp)
678 		return -EINVAL;
679 
680 	mrp->ring_role = role->ring_role;
681 
682 	/* If there is an error just bailed out */
683 	support = br_mrp_switchdev_set_ring_role(br, mrp, role->ring_role);
684 	if (support == BR_MRP_NONE)
685 		return -EOPNOTSUPP;
686 
687 	/* Now detect if the HW actually applied the role or not. If the HW
688 	 * applied the role it means that the SW will not to do those operations
689 	 * anymore. For example if the role ir MRM then the HW will notify the
690 	 * SW when ring is open, but if the is not pushed to the HW the SW will
691 	 * need to detect when the ring is open
692 	 */
693 	mrp->ring_role_offloaded = support == BR_MRP_SW ? 0 : 1;
694 
695 	return 0;
696 }
697 
698 /* Start to generate or monitor MRP test frames, the frames are generated by
699  * HW and if it fails, they are generated by the SW.
700  * note: already called with rtnl_lock
701  */
702 int br_mrp_start_test(struct net_bridge *br,
703 		      struct br_mrp_start_test *test)
704 {
705 	struct br_mrp *mrp = br_mrp_find_id(br, test->ring_id);
706 	enum br_mrp_hw_support support;
707 
708 	if (!mrp)
709 		return -EINVAL;
710 
711 	/* Try to push it to the HW and if it fails then continue with SW
712 	 * implementation and if that also fails then return error.
713 	 */
714 	support = br_mrp_switchdev_send_ring_test(br, mrp, test->interval,
715 						  test->max_miss, test->period,
716 						  test->monitor);
717 	if (support == BR_MRP_NONE)
718 		return -EOPNOTSUPP;
719 
720 	if (support == BR_MRP_HW)
721 		return 0;
722 
723 	mrp->test_interval = test->interval;
724 	mrp->test_end = jiffies + usecs_to_jiffies(test->period);
725 	mrp->test_max_miss = test->max_miss;
726 	mrp->test_monitor = test->monitor;
727 	mrp->test_count_miss = 0;
728 	queue_delayed_work(system_wq, &mrp->test_work,
729 			   usecs_to_jiffies(test->interval));
730 
731 	return 0;
732 }
733 
734 /* Set in state, int state can be only Open or Closed
735  * note: already called with rtnl_lock
736  */
737 int br_mrp_set_in_state(struct net_bridge *br, struct br_mrp_in_state *state)
738 {
739 	struct br_mrp *mrp = br_mrp_find_in_id(br, state->in_id);
740 
741 	if (!mrp)
742 		return -EINVAL;
743 
744 	if (mrp->in_state != state->in_state)
745 		mrp->in_transitions++;
746 
747 	mrp->in_state = state->in_state;
748 
749 	br_mrp_switchdev_set_in_state(br, mrp, state->in_state);
750 
751 	return 0;
752 }
753 
754 /* Set in role, in role can be only MIM(Media Interconnection Manager) or
755  * MIC(Media Interconnection Client).
756  * note: already called with rtnl_lock
757  */
758 int br_mrp_set_in_role(struct net_bridge *br, struct br_mrp_in_role *role)
759 {
760 	struct br_mrp *mrp = br_mrp_find_id(br, role->ring_id);
761 	enum br_mrp_hw_support support;
762 	struct net_bridge_port *p;
763 
764 	if (!mrp)
765 		return -EINVAL;
766 
767 	if (!br_mrp_get_port(br, role->i_ifindex))
768 		return -EINVAL;
769 
770 	if (role->in_role == BR_MRP_IN_ROLE_DISABLED) {
771 		u8 state;
772 
773 		/* It is not allowed to disable a port that doesn't exist */
774 		p = rtnl_dereference(mrp->i_port);
775 		if (!p)
776 			return -EINVAL;
777 
778 		/* Stop the generating MRP_InTest frames */
779 		cancel_delayed_work_sync(&mrp->in_test_work);
780 		br_mrp_switchdev_send_in_test(br, mrp, 0, 0, 0);
781 
782 		/* Remove the port */
783 		spin_lock_bh(&br->lock);
784 		state = netif_running(br->dev) ?
785 				BR_STATE_FORWARDING : BR_STATE_DISABLED;
786 		p->state = state;
787 		p->flags &= ~BR_MRP_AWARE;
788 		spin_unlock_bh(&br->lock);
789 		br_mrp_port_switchdev_set_state(p, state);
790 		rcu_assign_pointer(mrp->i_port, NULL);
791 
792 		mrp->in_role = role->in_role;
793 		mrp->in_id = 0;
794 
795 		return 0;
796 	}
797 
798 	/* It is not possible to have the same port part of multiple rings */
799 	if (!br_mrp_unique_ifindex(br, role->i_ifindex))
800 		return -EINVAL;
801 
802 	/* It is not allowed to set a different interconnect port if the mrp
803 	 * instance has already one. First it needs to be disabled and after
804 	 * that set the new port
805 	 */
806 	if (rcu_access_pointer(mrp->i_port))
807 		return -EINVAL;
808 
809 	p = br_mrp_get_port(br, role->i_ifindex);
810 	spin_lock_bh(&br->lock);
811 	p->state = BR_STATE_FORWARDING;
812 	p->flags |= BR_MRP_AWARE;
813 	spin_unlock_bh(&br->lock);
814 	rcu_assign_pointer(mrp->i_port, p);
815 
816 	mrp->in_role = role->in_role;
817 	mrp->in_id = role->in_id;
818 
819 	/* If there is an error just bailed out */
820 	support = br_mrp_switchdev_set_in_role(br, mrp, role->in_id,
821 					       role->ring_id, role->in_role);
822 	if (support == BR_MRP_NONE)
823 		return -EOPNOTSUPP;
824 
825 	/* Now detect if the HW actually applied the role or not. If the HW
826 	 * applied the role it means that the SW will not to do those operations
827 	 * anymore. For example if the role is MIM then the HW will notify the
828 	 * SW when interconnect ring is open, but if the is not pushed to the HW
829 	 * the SW will need to detect when the interconnect ring is open.
830 	 */
831 	mrp->in_role_offloaded = support == BR_MRP_SW ? 0 : 1;
832 
833 	return 0;
834 }
835 
836 /* Start to generate MRP_InTest frames, the frames are generated by
837  * HW and if it fails, they are generated by the SW.
838  * note: already called with rtnl_lock
839  */
840 int br_mrp_start_in_test(struct net_bridge *br,
841 			 struct br_mrp_start_in_test *in_test)
842 {
843 	struct br_mrp *mrp = br_mrp_find_in_id(br, in_test->in_id);
844 	enum br_mrp_hw_support support;
845 
846 	if (!mrp)
847 		return -EINVAL;
848 
849 	if (mrp->in_role != BR_MRP_IN_ROLE_MIM)
850 		return -EINVAL;
851 
852 	/* Try to push it to the HW and if it fails then continue with SW
853 	 * implementation and if that also fails then return error.
854 	 */
855 	support =  br_mrp_switchdev_send_in_test(br, mrp, in_test->interval,
856 						 in_test->max_miss,
857 						 in_test->period);
858 	if (support == BR_MRP_NONE)
859 		return -EOPNOTSUPP;
860 
861 	if (support == BR_MRP_HW)
862 		return 0;
863 
864 	mrp->in_test_interval = in_test->interval;
865 	mrp->in_test_end = jiffies + usecs_to_jiffies(in_test->period);
866 	mrp->in_test_max_miss = in_test->max_miss;
867 	mrp->in_test_count_miss = 0;
868 	queue_delayed_work(system_wq, &mrp->in_test_work,
869 			   usecs_to_jiffies(in_test->interval));
870 
871 	return 0;
872 }
873 
874 /* Determine if the frame type is a ring frame */
875 static bool br_mrp_ring_frame(struct sk_buff *skb)
876 {
877 	const struct br_mrp_tlv_hdr *hdr;
878 	struct br_mrp_tlv_hdr _hdr;
879 
880 	hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr);
881 	if (!hdr)
882 		return false;
883 
884 	if (hdr->type == BR_MRP_TLV_HEADER_RING_TEST ||
885 	    hdr->type == BR_MRP_TLV_HEADER_RING_TOPO ||
886 	    hdr->type == BR_MRP_TLV_HEADER_RING_LINK_DOWN ||
887 	    hdr->type == BR_MRP_TLV_HEADER_RING_LINK_UP ||
888 	    hdr->type == BR_MRP_TLV_HEADER_OPTION)
889 		return true;
890 
891 	return false;
892 }
893 
894 /* Determine if the frame type is an interconnect frame */
895 static bool br_mrp_in_frame(struct sk_buff *skb)
896 {
897 	const struct br_mrp_tlv_hdr *hdr;
898 	struct br_mrp_tlv_hdr _hdr;
899 
900 	hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr);
901 	if (!hdr)
902 		return false;
903 
904 	if (hdr->type == BR_MRP_TLV_HEADER_IN_TEST ||
905 	    hdr->type == BR_MRP_TLV_HEADER_IN_TOPO ||
906 	    hdr->type == BR_MRP_TLV_HEADER_IN_LINK_DOWN ||
907 	    hdr->type == BR_MRP_TLV_HEADER_IN_LINK_UP ||
908 	    hdr->type == BR_MRP_TLV_HEADER_IN_LINK_STATUS)
909 		return true;
910 
911 	return false;
912 }
913 
914 /* Process only MRP Test frame. All the other MRP frames are processed by
915  * userspace application
916  * note: already called with rcu_read_lock
917  */
918 static void br_mrp_mrm_process(struct br_mrp *mrp, struct net_bridge_port *port,
919 			       struct sk_buff *skb)
920 {
921 	const struct br_mrp_tlv_hdr *hdr;
922 	struct br_mrp_tlv_hdr _hdr;
923 
924 	/* Each MRP header starts with a version field which is 16 bits.
925 	 * Therefore skip the version and get directly the TLV header.
926 	 */
927 	hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr);
928 	if (!hdr)
929 		return;
930 
931 	if (hdr->type != BR_MRP_TLV_HEADER_RING_TEST)
932 		return;
933 
934 	mrp->test_count_miss = 0;
935 
936 	/* Notify the userspace that the ring is closed only when the ring is
937 	 * not closed
938 	 */
939 	if (mrp->ring_state != BR_MRP_RING_STATE_CLOSED)
940 		br_mrp_ring_port_open(port->dev, false);
941 }
942 
943 /* Determine if the test hdr has a better priority than the node */
944 static bool br_mrp_test_better_than_own(struct br_mrp *mrp,
945 					struct net_bridge *br,
946 					const struct br_mrp_ring_test_hdr *hdr)
947 {
948 	u16 prio = be16_to_cpu(hdr->prio);
949 
950 	if (prio < mrp->prio ||
951 	    (prio == mrp->prio &&
952 	    ether_addr_to_u64(hdr->sa) < ether_addr_to_u64(br->dev->dev_addr)))
953 		return true;
954 
955 	return false;
956 }
957 
958 /* Process only MRP Test frame. All the other MRP frames are processed by
959  * userspace application
960  * note: already called with rcu_read_lock
961  */
962 static void br_mrp_mra_process(struct br_mrp *mrp, struct net_bridge *br,
963 			       struct net_bridge_port *port,
964 			       struct sk_buff *skb)
965 {
966 	const struct br_mrp_ring_test_hdr *test_hdr;
967 	struct br_mrp_ring_test_hdr _test_hdr;
968 	const struct br_mrp_tlv_hdr *hdr;
969 	struct br_mrp_tlv_hdr _hdr;
970 
971 	/* Each MRP header starts with a version field which is 16 bits.
972 	 * Therefore skip the version and get directly the TLV header.
973 	 */
974 	hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr);
975 	if (!hdr)
976 		return;
977 
978 	if (hdr->type != BR_MRP_TLV_HEADER_RING_TEST)
979 		return;
980 
981 	test_hdr = skb_header_pointer(skb, sizeof(uint16_t) + sizeof(_hdr),
982 				      sizeof(_test_hdr), &_test_hdr);
983 	if (!test_hdr)
984 		return;
985 
986 	/* Only frames that have a better priority than the node will
987 	 * clear the miss counter because otherwise the node will need to behave
988 	 * as MRM.
989 	 */
990 	if (br_mrp_test_better_than_own(mrp, br, test_hdr))
991 		mrp->test_count_miss = 0;
992 }
993 
994 /* Process only MRP InTest frame. All the other MRP frames are processed by
995  * userspace application
996  * note: already called with rcu_read_lock
997  */
998 static bool br_mrp_mim_process(struct br_mrp *mrp, struct net_bridge_port *port,
999 			       struct sk_buff *skb)
1000 {
1001 	const struct br_mrp_in_test_hdr *in_hdr;
1002 	struct br_mrp_in_test_hdr _in_hdr;
1003 	const struct br_mrp_tlv_hdr *hdr;
1004 	struct br_mrp_tlv_hdr _hdr;
1005 
1006 	/* Each MRP header starts with a version field which is 16 bits.
1007 	 * Therefore skip the version and get directly the TLV header.
1008 	 */
1009 	hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr);
1010 	if (!hdr)
1011 		return false;
1012 
1013 	/* The check for InTest frame type was already done */
1014 	in_hdr = skb_header_pointer(skb, sizeof(uint16_t) + sizeof(_hdr),
1015 				    sizeof(_in_hdr), &_in_hdr);
1016 	if (!in_hdr)
1017 		return false;
1018 
1019 	/* It needs to process only it's own InTest frames. */
1020 	if (mrp->in_id != ntohs(in_hdr->id))
1021 		return false;
1022 
1023 	mrp->in_test_count_miss = 0;
1024 
1025 	/* Notify the userspace that the ring is closed only when the ring is
1026 	 * not closed
1027 	 */
1028 	if (mrp->in_state != BR_MRP_IN_STATE_CLOSED)
1029 		br_mrp_in_port_open(port->dev, false);
1030 
1031 	return true;
1032 }
1033 
1034 /* Get the MRP frame type
1035  * note: already called with rcu_read_lock
1036  */
1037 static u8 br_mrp_get_frame_type(struct sk_buff *skb)
1038 {
1039 	const struct br_mrp_tlv_hdr *hdr;
1040 	struct br_mrp_tlv_hdr _hdr;
1041 
1042 	/* Each MRP header starts with a version field which is 16 bits.
1043 	 * Therefore skip the version and get directly the TLV header.
1044 	 */
1045 	hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr);
1046 	if (!hdr)
1047 		return 0xff;
1048 
1049 	return hdr->type;
1050 }
1051 
1052 static bool br_mrp_mrm_behaviour(struct br_mrp *mrp)
1053 {
1054 	if (mrp->ring_role == BR_MRP_RING_ROLE_MRM ||
1055 	    (mrp->ring_role == BR_MRP_RING_ROLE_MRA && !mrp->test_monitor))
1056 		return true;
1057 
1058 	return false;
1059 }
1060 
1061 static bool br_mrp_mrc_behaviour(struct br_mrp *mrp)
1062 {
1063 	if (mrp->ring_role == BR_MRP_RING_ROLE_MRC ||
1064 	    (mrp->ring_role == BR_MRP_RING_ROLE_MRA && mrp->test_monitor))
1065 		return true;
1066 
1067 	return false;
1068 }
1069 
1070 /* This will just forward the frame to the other mrp ring ports, depending on
1071  * the frame type, ring role and interconnect role
1072  * note: already called with rcu_read_lock
1073  */
1074 static int br_mrp_rcv(struct net_bridge_port *p,
1075 		      struct sk_buff *skb, struct net_device *dev)
1076 {
1077 	struct net_bridge_port *p_port, *s_port, *i_port = NULL;
1078 	struct net_bridge_port *p_dst, *s_dst, *i_dst = NULL;
1079 	struct net_bridge *br;
1080 	struct br_mrp *mrp;
1081 
1082 	/* If port is disabled don't accept any frames */
1083 	if (p->state == BR_STATE_DISABLED)
1084 		return 0;
1085 
1086 	br = p->br;
1087 	mrp =  br_mrp_find_port(br, p);
1088 	if (unlikely(!mrp))
1089 		return 0;
1090 
1091 	p_port = rcu_dereference(mrp->p_port);
1092 	if (!p_port)
1093 		return 0;
1094 	p_dst = p_port;
1095 
1096 	s_port = rcu_dereference(mrp->s_port);
1097 	if (!s_port)
1098 		return 0;
1099 	s_dst = s_port;
1100 
1101 	/* If the frame is a ring frame then it is not required to check the
1102 	 * interconnect role and ports to process or forward the frame
1103 	 */
1104 	if (br_mrp_ring_frame(skb)) {
1105 		/* If the role is MRM then don't forward the frames */
1106 		if (mrp->ring_role == BR_MRP_RING_ROLE_MRM) {
1107 			br_mrp_mrm_process(mrp, p, skb);
1108 			goto no_forward;
1109 		}
1110 
1111 		/* If the role is MRA then don't forward the frames if it
1112 		 * behaves as MRM node
1113 		 */
1114 		if (mrp->ring_role == BR_MRP_RING_ROLE_MRA) {
1115 			if (!mrp->test_monitor) {
1116 				br_mrp_mrm_process(mrp, p, skb);
1117 				goto no_forward;
1118 			}
1119 
1120 			br_mrp_mra_process(mrp, br, p, skb);
1121 		}
1122 
1123 		goto forward;
1124 	}
1125 
1126 	if (br_mrp_in_frame(skb)) {
1127 		u8 in_type = br_mrp_get_frame_type(skb);
1128 
1129 		i_port = rcu_dereference(mrp->i_port);
1130 		i_dst = i_port;
1131 
1132 		/* If the ring port is in block state it should not forward
1133 		 * In_Test frames
1134 		 */
1135 		if (br_mrp_is_ring_port(p_port, s_port, p) &&
1136 		    p->state == BR_STATE_BLOCKING &&
1137 		    in_type == BR_MRP_TLV_HEADER_IN_TEST)
1138 			goto no_forward;
1139 
1140 		/* Nodes that behaves as MRM needs to stop forwarding the
1141 		 * frames in case the ring is closed, otherwise will be a loop.
1142 		 * In this case the frame is no forward between the ring ports.
1143 		 */
1144 		if (br_mrp_mrm_behaviour(mrp) &&
1145 		    br_mrp_is_ring_port(p_port, s_port, p) &&
1146 		    (s_port->state != BR_STATE_FORWARDING ||
1147 		     p_port->state != BR_STATE_FORWARDING)) {
1148 			p_dst = NULL;
1149 			s_dst = NULL;
1150 		}
1151 
1152 		/* A node that behaves as MRC and doesn't have a interconnect
1153 		 * role then it should forward all frames between the ring ports
1154 		 * because it doesn't have an interconnect port
1155 		 */
1156 		if (br_mrp_mrc_behaviour(mrp) &&
1157 		    mrp->in_role == BR_MRP_IN_ROLE_DISABLED)
1158 			goto forward;
1159 
1160 		if (mrp->in_role == BR_MRP_IN_ROLE_MIM) {
1161 			if (in_type == BR_MRP_TLV_HEADER_IN_TEST) {
1162 				/* MIM should not forward it's own InTest
1163 				 * frames
1164 				 */
1165 				if (br_mrp_mim_process(mrp, p, skb)) {
1166 					goto no_forward;
1167 				} else {
1168 					if (br_mrp_is_ring_port(p_port, s_port,
1169 								p))
1170 						i_dst = NULL;
1171 
1172 					if (br_mrp_is_in_port(i_port, p))
1173 						goto no_forward;
1174 				}
1175 			} else {
1176 				/* MIM should forward IntLinkChange/Status and
1177 				 * IntTopoChange between ring ports but MIM
1178 				 * should not forward IntLinkChange/Status and
1179 				 * IntTopoChange if the frame was received at
1180 				 * the interconnect port
1181 				 */
1182 				if (br_mrp_is_ring_port(p_port, s_port, p))
1183 					i_dst = NULL;
1184 
1185 				if (br_mrp_is_in_port(i_port, p))
1186 					goto no_forward;
1187 			}
1188 		}
1189 
1190 		if (mrp->in_role == BR_MRP_IN_ROLE_MIC) {
1191 			/* MIC should forward InTest frames on all ports
1192 			 * regardless of the received port
1193 			 */
1194 			if (in_type == BR_MRP_TLV_HEADER_IN_TEST)
1195 				goto forward;
1196 
1197 			/* MIC should forward IntLinkChange frames only if they
1198 			 * are received on ring ports to all the ports
1199 			 */
1200 			if (br_mrp_is_ring_port(p_port, s_port, p) &&
1201 			    (in_type == BR_MRP_TLV_HEADER_IN_LINK_UP ||
1202 			     in_type == BR_MRP_TLV_HEADER_IN_LINK_DOWN))
1203 				goto forward;
1204 
1205 			/* MIC should forward IntLinkStatus frames only to
1206 			 * interconnect port if it was received on a ring port.
1207 			 * If it is received on interconnect port then, it
1208 			 * should be forward on both ring ports
1209 			 */
1210 			if (br_mrp_is_ring_port(p_port, s_port, p) &&
1211 			    in_type == BR_MRP_TLV_HEADER_IN_LINK_STATUS) {
1212 				p_dst = NULL;
1213 				s_dst = NULL;
1214 			}
1215 
1216 			/* Should forward the InTopo frames only between the
1217 			 * ring ports
1218 			 */
1219 			if (in_type == BR_MRP_TLV_HEADER_IN_TOPO) {
1220 				i_dst = NULL;
1221 				goto forward;
1222 			}
1223 
1224 			/* In all the other cases don't forward the frames */
1225 			goto no_forward;
1226 		}
1227 	}
1228 
1229 forward:
1230 	if (p_dst)
1231 		br_forward(p_dst, skb, true, false);
1232 	if (s_dst)
1233 		br_forward(s_dst, skb, true, false);
1234 	if (i_dst)
1235 		br_forward(i_dst, skb, true, false);
1236 
1237 no_forward:
1238 	return 1;
1239 }
1240 
1241 /* Check if the frame was received on a port that is part of MRP ring
1242  * and if the frame has MRP eth. In that case process the frame otherwise do
1243  * normal forwarding.
1244  * note: already called with rcu_read_lock
1245  */
1246 static int br_mrp_process(struct net_bridge_port *p, struct sk_buff *skb)
1247 {
1248 	/* If there is no MRP instance do normal forwarding */
1249 	if (likely(!(p->flags & BR_MRP_AWARE)))
1250 		goto out;
1251 
1252 	return br_mrp_rcv(p, skb, p->dev);
1253 out:
1254 	return 0;
1255 }
1256 
1257 bool br_mrp_enabled(struct net_bridge *br)
1258 {
1259 	return !hlist_empty(&br->mrp_list);
1260 }
1261