xref: /linux/drivers/net/ethernet/mellanox/mlxsw/spectrum_mr.c (revision e3b9f1e81de2083f359bacd2a94bf1c024f2ede0)
1 /*
2  * drivers/net/ethernet/mellanox/mlxsw/spectrum_mr.c
3  * Copyright (c) 2017 Mellanox Technologies. All rights reserved.
4  * Copyright (c) 2017 Yotam Gigi <yotamg@mellanox.com>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions are met:
8  *
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. Neither the names of the copyright holders nor the names of its
15  *    contributors may be used to endorse or promote products derived from
16  *    this software without specific prior written permission.
17  *
18  * Alternatively, this software may be distributed under the terms of the
19  * GNU General Public License ("GPL") version 2 as published by the Free
20  * Software Foundation.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
23  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
26  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <linux/rhashtable.h>
36 
37 #include "spectrum_mr.h"
38 #include "spectrum_router.h"
39 
40 struct mlxsw_sp_mr {
41 	const struct mlxsw_sp_mr_ops *mr_ops;
42 	void *catchall_route_priv;
43 	struct delayed_work stats_update_dw;
44 	struct list_head table_list;
45 #define MLXSW_SP_MR_ROUTES_COUNTER_UPDATE_INTERVAL 5000 /* ms */
46 	unsigned long priv[0];
47 	/* priv has to be always the last item */
48 };
49 
50 struct mlxsw_sp_mr_vif {
51 	struct net_device *dev;
52 	const struct mlxsw_sp_rif *rif;
53 	unsigned long vif_flags;
54 
55 	/* A list of route_vif_entry structs that point to routes that the VIF
56 	 * instance is used as one of the egress VIFs
57 	 */
58 	struct list_head route_evif_list;
59 
60 	/* A list of route_vif_entry structs that point to routes that the VIF
61 	 * instance is used as an ingress VIF
62 	 */
63 	struct list_head route_ivif_list;
64 };
65 
66 struct mlxsw_sp_mr_route_vif_entry {
67 	struct list_head vif_node;
68 	struct list_head route_node;
69 	struct mlxsw_sp_mr_vif *mr_vif;
70 	struct mlxsw_sp_mr_route *mr_route;
71 };
72 
73 struct mlxsw_sp_mr_table {
74 	struct list_head node;
75 	enum mlxsw_sp_l3proto proto;
76 	struct mlxsw_sp *mlxsw_sp;
77 	u32 vr_id;
78 	struct mlxsw_sp_mr_vif vifs[MAXVIFS];
79 	struct list_head route_list;
80 	struct rhashtable route_ht;
81 	char catchall_route_priv[0];
82 	/* catchall_route_priv has to be always the last item */
83 };
84 
85 struct mlxsw_sp_mr_route {
86 	struct list_head node;
87 	struct rhash_head ht_node;
88 	struct mlxsw_sp_mr_route_key key;
89 	enum mlxsw_sp_mr_route_action route_action;
90 	u16 min_mtu;
91 	struct mfc_cache *mfc4;
92 	void *route_priv;
93 	const struct mlxsw_sp_mr_table *mr_table;
94 	/* A list of route_vif_entry structs that point to the egress VIFs */
95 	struct list_head evif_list;
96 	/* A route_vif_entry struct that point to the ingress VIF */
97 	struct mlxsw_sp_mr_route_vif_entry ivif;
98 };
99 
100 static const struct rhashtable_params mlxsw_sp_mr_route_ht_params = {
101 	.key_len = sizeof(struct mlxsw_sp_mr_route_key),
102 	.key_offset = offsetof(struct mlxsw_sp_mr_route, key),
103 	.head_offset = offsetof(struct mlxsw_sp_mr_route, ht_node),
104 	.automatic_shrinking = true,
105 };
106 
107 static bool mlxsw_sp_mr_vif_regular(const struct mlxsw_sp_mr_vif *vif)
108 {
109 	return !(vif->vif_flags & (VIFF_TUNNEL | VIFF_REGISTER));
110 }
111 
112 static bool mlxsw_sp_mr_vif_valid(const struct mlxsw_sp_mr_vif *vif)
113 {
114 	return mlxsw_sp_mr_vif_regular(vif) && vif->dev && vif->rif;
115 }
116 
117 static bool mlxsw_sp_mr_vif_exists(const struct mlxsw_sp_mr_vif *vif)
118 {
119 	return vif->dev;
120 }
121 
122 static bool
123 mlxsw_sp_mr_route_ivif_in_evifs(const struct mlxsw_sp_mr_route *mr_route)
124 {
125 	vifi_t ivif;
126 
127 	switch (mr_route->mr_table->proto) {
128 	case MLXSW_SP_L3_PROTO_IPV4:
129 		ivif = mr_route->mfc4->mfc_parent;
130 		return mr_route->mfc4->mfc_un.res.ttls[ivif] != 255;
131 	case MLXSW_SP_L3_PROTO_IPV6:
132 		/* fall through */
133 	default:
134 		WARN_ON_ONCE(1);
135 	}
136 	return false;
137 }
138 
139 static int
140 mlxsw_sp_mr_route_valid_evifs_num(const struct mlxsw_sp_mr_route *mr_route)
141 {
142 	struct mlxsw_sp_mr_route_vif_entry *rve;
143 	int valid_evifs;
144 
145 	valid_evifs = 0;
146 	list_for_each_entry(rve, &mr_route->evif_list, route_node)
147 		if (mlxsw_sp_mr_vif_valid(rve->mr_vif))
148 			valid_evifs++;
149 	return valid_evifs;
150 }
151 
152 static bool mlxsw_sp_mr_route_starg(const struct mlxsw_sp_mr_route *mr_route)
153 {
154 	switch (mr_route->mr_table->proto) {
155 	case MLXSW_SP_L3_PROTO_IPV4:
156 		return mr_route->key.source_mask.addr4 == htonl(INADDR_ANY);
157 	case MLXSW_SP_L3_PROTO_IPV6:
158 		/* fall through */
159 	default:
160 		WARN_ON_ONCE(1);
161 	}
162 	return false;
163 }
164 
165 static enum mlxsw_sp_mr_route_action
166 mlxsw_sp_mr_route_action(const struct mlxsw_sp_mr_route *mr_route)
167 {
168 	struct mlxsw_sp_mr_route_vif_entry *rve;
169 
170 	/* If the ingress port is not regular and resolved, trap the route */
171 	if (!mlxsw_sp_mr_vif_valid(mr_route->ivif.mr_vif))
172 		return MLXSW_SP_MR_ROUTE_ACTION_TRAP;
173 
174 	/* The kernel does not match a (*,G) route that the ingress interface is
175 	 * not one of the egress interfaces, so trap these kind of routes.
176 	 */
177 	if (mlxsw_sp_mr_route_starg(mr_route) &&
178 	    !mlxsw_sp_mr_route_ivif_in_evifs(mr_route))
179 		return MLXSW_SP_MR_ROUTE_ACTION_TRAP;
180 
181 	/* If the route has no valid eVIFs, trap it. */
182 	if (!mlxsw_sp_mr_route_valid_evifs_num(mr_route))
183 		return MLXSW_SP_MR_ROUTE_ACTION_TRAP;
184 
185 	/* If one of the eVIFs has no RIF, trap-and-forward the route as there
186 	 * is some more routing to do in software too.
187 	 */
188 	list_for_each_entry(rve, &mr_route->evif_list, route_node)
189 		if (mlxsw_sp_mr_vif_exists(rve->mr_vif) && !rve->mr_vif->rif)
190 			return MLXSW_SP_MR_ROUTE_ACTION_TRAP_AND_FORWARD;
191 
192 	return MLXSW_SP_MR_ROUTE_ACTION_FORWARD;
193 }
194 
195 static enum mlxsw_sp_mr_route_prio
196 mlxsw_sp_mr_route_prio(const struct mlxsw_sp_mr_route *mr_route)
197 {
198 	return mlxsw_sp_mr_route_starg(mr_route) ?
199 		MLXSW_SP_MR_ROUTE_PRIO_STARG : MLXSW_SP_MR_ROUTE_PRIO_SG;
200 }
201 
202 static void mlxsw_sp_mr_route4_key(struct mlxsw_sp_mr_table *mr_table,
203 				   struct mlxsw_sp_mr_route_key *key,
204 				   const struct mfc_cache *mfc)
205 {
206 	bool starg = (mfc->mfc_origin == htonl(INADDR_ANY));
207 
208 	memset(key, 0, sizeof(*key));
209 	key->vrid = mr_table->vr_id;
210 	key->proto = mr_table->proto;
211 	key->group.addr4 = mfc->mfc_mcastgrp;
212 	key->group_mask.addr4 = htonl(0xffffffff);
213 	key->source.addr4 = mfc->mfc_origin;
214 	key->source_mask.addr4 = htonl(starg ? 0 : 0xffffffff);
215 }
216 
217 static int mlxsw_sp_mr_route_evif_link(struct mlxsw_sp_mr_route *mr_route,
218 				       struct mlxsw_sp_mr_vif *mr_vif)
219 {
220 	struct mlxsw_sp_mr_route_vif_entry *rve;
221 
222 	rve = kzalloc(sizeof(*rve), GFP_KERNEL);
223 	if (!rve)
224 		return -ENOMEM;
225 	rve->mr_route = mr_route;
226 	rve->mr_vif = mr_vif;
227 	list_add_tail(&rve->route_node, &mr_route->evif_list);
228 	list_add_tail(&rve->vif_node, &mr_vif->route_evif_list);
229 	return 0;
230 }
231 
232 static void
233 mlxsw_sp_mr_route_evif_unlink(struct mlxsw_sp_mr_route_vif_entry *rve)
234 {
235 	list_del(&rve->route_node);
236 	list_del(&rve->vif_node);
237 	kfree(rve);
238 }
239 
240 static void mlxsw_sp_mr_route_ivif_link(struct mlxsw_sp_mr_route *mr_route,
241 					struct mlxsw_sp_mr_vif *mr_vif)
242 {
243 	mr_route->ivif.mr_route = mr_route;
244 	mr_route->ivif.mr_vif = mr_vif;
245 	list_add_tail(&mr_route->ivif.vif_node, &mr_vif->route_ivif_list);
246 }
247 
248 static void mlxsw_sp_mr_route_ivif_unlink(struct mlxsw_sp_mr_route *mr_route)
249 {
250 	list_del(&mr_route->ivif.vif_node);
251 }
252 
253 static int
254 mlxsw_sp_mr_route_info_create(struct mlxsw_sp_mr_table *mr_table,
255 			      struct mlxsw_sp_mr_route *mr_route,
256 			      struct mlxsw_sp_mr_route_info *route_info)
257 {
258 	struct mlxsw_sp_mr_route_vif_entry *rve;
259 	u16 *erif_indices;
260 	u16 irif_index;
261 	u16 erif = 0;
262 
263 	erif_indices = kmalloc_array(MAXVIFS, sizeof(*erif_indices),
264 				     GFP_KERNEL);
265 	if (!erif_indices)
266 		return -ENOMEM;
267 
268 	list_for_each_entry(rve, &mr_route->evif_list, route_node) {
269 		if (mlxsw_sp_mr_vif_valid(rve->mr_vif)) {
270 			u16 rifi = mlxsw_sp_rif_index(rve->mr_vif->rif);
271 
272 			erif_indices[erif++] = rifi;
273 		}
274 	}
275 
276 	if (mlxsw_sp_mr_vif_valid(mr_route->ivif.mr_vif))
277 		irif_index = mlxsw_sp_rif_index(mr_route->ivif.mr_vif->rif);
278 	else
279 		irif_index = 0;
280 
281 	route_info->irif_index = irif_index;
282 	route_info->erif_indices = erif_indices;
283 	route_info->min_mtu = mr_route->min_mtu;
284 	route_info->route_action = mr_route->route_action;
285 	route_info->erif_num = erif;
286 	return 0;
287 }
288 
289 static void
290 mlxsw_sp_mr_route_info_destroy(struct mlxsw_sp_mr_route_info *route_info)
291 {
292 	kfree(route_info->erif_indices);
293 }
294 
295 static int mlxsw_sp_mr_route_write(struct mlxsw_sp_mr_table *mr_table,
296 				   struct mlxsw_sp_mr_route *mr_route,
297 				   bool replace)
298 {
299 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
300 	struct mlxsw_sp_mr_route_info route_info;
301 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
302 	int err;
303 
304 	err = mlxsw_sp_mr_route_info_create(mr_table, mr_route, &route_info);
305 	if (err)
306 		return err;
307 
308 	if (!replace) {
309 		struct mlxsw_sp_mr_route_params route_params;
310 
311 		mr_route->route_priv = kzalloc(mr->mr_ops->route_priv_size,
312 					       GFP_KERNEL);
313 		if (!mr_route->route_priv) {
314 			err = -ENOMEM;
315 			goto out;
316 		}
317 
318 		route_params.key = mr_route->key;
319 		route_params.value = route_info;
320 		route_params.prio = mlxsw_sp_mr_route_prio(mr_route);
321 		err = mr->mr_ops->route_create(mlxsw_sp, mr->priv,
322 					       mr_route->route_priv,
323 					       &route_params);
324 		if (err)
325 			kfree(mr_route->route_priv);
326 	} else {
327 		err = mr->mr_ops->route_update(mlxsw_sp, mr_route->route_priv,
328 					       &route_info);
329 	}
330 out:
331 	mlxsw_sp_mr_route_info_destroy(&route_info);
332 	return err;
333 }
334 
335 static void mlxsw_sp_mr_route_erase(struct mlxsw_sp_mr_table *mr_table,
336 				    struct mlxsw_sp_mr_route *mr_route)
337 {
338 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
339 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
340 
341 	mr->mr_ops->route_destroy(mlxsw_sp, mr->priv, mr_route->route_priv);
342 	kfree(mr_route->route_priv);
343 }
344 
345 static struct mlxsw_sp_mr_route *
346 mlxsw_sp_mr_route4_create(struct mlxsw_sp_mr_table *mr_table,
347 			  struct mfc_cache *mfc)
348 {
349 	struct mlxsw_sp_mr_route_vif_entry *rve, *tmp;
350 	struct mlxsw_sp_mr_route *mr_route;
351 	int err = 0;
352 	int i;
353 
354 	/* Allocate and init a new route and fill it with parameters */
355 	mr_route = kzalloc(sizeof(*mr_route), GFP_KERNEL);
356 	if (!mr_route)
357 		return ERR_PTR(-ENOMEM);
358 	INIT_LIST_HEAD(&mr_route->evif_list);
359 	mlxsw_sp_mr_route4_key(mr_table, &mr_route->key, mfc);
360 
361 	/* Find min_mtu and link iVIF and eVIFs */
362 	mr_route->min_mtu = ETH_MAX_MTU;
363 	ipmr_cache_hold(mfc);
364 	mr_route->mfc4 = mfc;
365 	mr_route->mr_table = mr_table;
366 	for (i = 0; i < MAXVIFS; i++) {
367 		if (mfc->mfc_un.res.ttls[i] != 255) {
368 			err = mlxsw_sp_mr_route_evif_link(mr_route,
369 							  &mr_table->vifs[i]);
370 			if (err)
371 				goto err;
372 			if (mr_table->vifs[i].dev &&
373 			    mr_table->vifs[i].dev->mtu < mr_route->min_mtu)
374 				mr_route->min_mtu = mr_table->vifs[i].dev->mtu;
375 		}
376 	}
377 	mlxsw_sp_mr_route_ivif_link(mr_route, &mr_table->vifs[mfc->mfc_parent]);
378 
379 	mr_route->route_action = mlxsw_sp_mr_route_action(mr_route);
380 	return mr_route;
381 err:
382 	ipmr_cache_put(mfc);
383 	list_for_each_entry_safe(rve, tmp, &mr_route->evif_list, route_node)
384 		mlxsw_sp_mr_route_evif_unlink(rve);
385 	kfree(mr_route);
386 	return ERR_PTR(err);
387 }
388 
389 static void mlxsw_sp_mr_route4_destroy(struct mlxsw_sp_mr_table *mr_table,
390 				       struct mlxsw_sp_mr_route *mr_route)
391 {
392 	struct mlxsw_sp_mr_route_vif_entry *rve, *tmp;
393 
394 	mlxsw_sp_mr_route_ivif_unlink(mr_route);
395 	ipmr_cache_put(mr_route->mfc4);
396 	list_for_each_entry_safe(rve, tmp, &mr_route->evif_list, route_node)
397 		mlxsw_sp_mr_route_evif_unlink(rve);
398 	kfree(mr_route);
399 }
400 
401 static void mlxsw_sp_mr_route_destroy(struct mlxsw_sp_mr_table *mr_table,
402 				      struct mlxsw_sp_mr_route *mr_route)
403 {
404 	switch (mr_table->proto) {
405 	case MLXSW_SP_L3_PROTO_IPV4:
406 		mlxsw_sp_mr_route4_destroy(mr_table, mr_route);
407 		break;
408 	case MLXSW_SP_L3_PROTO_IPV6:
409 		/* fall through */
410 	default:
411 		WARN_ON_ONCE(1);
412 	}
413 }
414 
415 static void mlxsw_sp_mr_mfc_offload_set(struct mlxsw_sp_mr_route *mr_route,
416 					bool offload)
417 {
418 	switch (mr_route->mr_table->proto) {
419 	case MLXSW_SP_L3_PROTO_IPV4:
420 		if (offload)
421 			mr_route->mfc4->mfc_flags |= MFC_OFFLOAD;
422 		else
423 			mr_route->mfc4->mfc_flags &= ~MFC_OFFLOAD;
424 		break;
425 	case MLXSW_SP_L3_PROTO_IPV6:
426 		/* fall through */
427 	default:
428 		WARN_ON_ONCE(1);
429 	}
430 }
431 
432 static void mlxsw_sp_mr_mfc_offload_update(struct mlxsw_sp_mr_route *mr_route)
433 {
434 	bool offload;
435 
436 	offload = mr_route->route_action != MLXSW_SP_MR_ROUTE_ACTION_TRAP;
437 	mlxsw_sp_mr_mfc_offload_set(mr_route, offload);
438 }
439 
440 static void __mlxsw_sp_mr_route_del(struct mlxsw_sp_mr_table *mr_table,
441 				    struct mlxsw_sp_mr_route *mr_route)
442 {
443 	mlxsw_sp_mr_mfc_offload_set(mr_route, false);
444 	mlxsw_sp_mr_route_erase(mr_table, mr_route);
445 	rhashtable_remove_fast(&mr_table->route_ht, &mr_route->ht_node,
446 			       mlxsw_sp_mr_route_ht_params);
447 	list_del(&mr_route->node);
448 	mlxsw_sp_mr_route_destroy(mr_table, mr_route);
449 }
450 
451 int mlxsw_sp_mr_route4_add(struct mlxsw_sp_mr_table *mr_table,
452 			   struct mfc_cache *mfc, bool replace)
453 {
454 	struct mlxsw_sp_mr_route *mr_orig_route = NULL;
455 	struct mlxsw_sp_mr_route *mr_route;
456 	int err;
457 
458 	/* If the route is a (*,*) route, abort, as these kind of routes are
459 	 * used for proxy routes.
460 	 */
461 	if (mfc->mfc_origin == htonl(INADDR_ANY) &&
462 	    mfc->mfc_mcastgrp == htonl(INADDR_ANY)) {
463 		dev_warn(mr_table->mlxsw_sp->bus_info->dev,
464 			 "Offloading proxy routes is not supported.\n");
465 		return -EINVAL;
466 	}
467 
468 	/* Create a new route */
469 	mr_route = mlxsw_sp_mr_route4_create(mr_table, mfc);
470 	if (IS_ERR(mr_route))
471 		return PTR_ERR(mr_route);
472 
473 	/* Find any route with a matching key */
474 	mr_orig_route = rhashtable_lookup_fast(&mr_table->route_ht,
475 					       &mr_route->key,
476 					       mlxsw_sp_mr_route_ht_params);
477 	if (replace) {
478 		/* On replace case, make the route point to the new route_priv.
479 		 */
480 		if (WARN_ON(!mr_orig_route)) {
481 			err = -ENOENT;
482 			goto err_no_orig_route;
483 		}
484 		mr_route->route_priv = mr_orig_route->route_priv;
485 	} else if (mr_orig_route) {
486 		/* On non replace case, if another route with the same key was
487 		 * found, abort, as duplicate routes are used for proxy routes.
488 		 */
489 		dev_warn(mr_table->mlxsw_sp->bus_info->dev,
490 			 "Offloading proxy routes is not supported.\n");
491 		err = -EINVAL;
492 		goto err_duplicate_route;
493 	}
494 
495 	/* Put it in the table data-structures */
496 	list_add_tail(&mr_route->node, &mr_table->route_list);
497 	err = rhashtable_insert_fast(&mr_table->route_ht,
498 				     &mr_route->ht_node,
499 				     mlxsw_sp_mr_route_ht_params);
500 	if (err)
501 		goto err_rhashtable_insert;
502 
503 	/* Write the route to the hardware */
504 	err = mlxsw_sp_mr_route_write(mr_table, mr_route, replace);
505 	if (err)
506 		goto err_mr_route_write;
507 
508 	/* Destroy the original route */
509 	if (replace) {
510 		rhashtable_remove_fast(&mr_table->route_ht,
511 				       &mr_orig_route->ht_node,
512 				       mlxsw_sp_mr_route_ht_params);
513 		list_del(&mr_orig_route->node);
514 		mlxsw_sp_mr_route4_destroy(mr_table, mr_orig_route);
515 	}
516 
517 	mlxsw_sp_mr_mfc_offload_update(mr_route);
518 	return 0;
519 
520 err_mr_route_write:
521 	rhashtable_remove_fast(&mr_table->route_ht, &mr_route->ht_node,
522 			       mlxsw_sp_mr_route_ht_params);
523 err_rhashtable_insert:
524 	list_del(&mr_route->node);
525 err_no_orig_route:
526 err_duplicate_route:
527 	mlxsw_sp_mr_route4_destroy(mr_table, mr_route);
528 	return err;
529 }
530 
531 void mlxsw_sp_mr_route4_del(struct mlxsw_sp_mr_table *mr_table,
532 			    struct mfc_cache *mfc)
533 {
534 	struct mlxsw_sp_mr_route *mr_route;
535 	struct mlxsw_sp_mr_route_key key;
536 
537 	mlxsw_sp_mr_route4_key(mr_table, &key, mfc);
538 	mr_route = rhashtable_lookup_fast(&mr_table->route_ht, &key,
539 					  mlxsw_sp_mr_route_ht_params);
540 	if (mr_route)
541 		__mlxsw_sp_mr_route_del(mr_table, mr_route);
542 }
543 
544 /* Should be called after the VIF struct is updated */
545 static int
546 mlxsw_sp_mr_route_ivif_resolve(struct mlxsw_sp_mr_table *mr_table,
547 			       struct mlxsw_sp_mr_route_vif_entry *rve)
548 {
549 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
550 	enum mlxsw_sp_mr_route_action route_action;
551 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
552 	u16 irif_index;
553 	int err;
554 
555 	route_action = mlxsw_sp_mr_route_action(rve->mr_route);
556 	if (route_action == MLXSW_SP_MR_ROUTE_ACTION_TRAP)
557 		return 0;
558 
559 	/* rve->mr_vif->rif is guaranteed to be valid at this stage */
560 	irif_index = mlxsw_sp_rif_index(rve->mr_vif->rif);
561 	err = mr->mr_ops->route_irif_update(mlxsw_sp, rve->mr_route->route_priv,
562 					    irif_index);
563 	if (err)
564 		return err;
565 
566 	err = mr->mr_ops->route_action_update(mlxsw_sp,
567 					      rve->mr_route->route_priv,
568 					      route_action);
569 	if (err)
570 		/* No need to rollback here because the iRIF change only takes
571 		 * place after the action has been updated.
572 		 */
573 		return err;
574 
575 	rve->mr_route->route_action = route_action;
576 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
577 	return 0;
578 }
579 
580 static void
581 mlxsw_sp_mr_route_ivif_unresolve(struct mlxsw_sp_mr_table *mr_table,
582 				 struct mlxsw_sp_mr_route_vif_entry *rve)
583 {
584 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
585 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
586 
587 	mr->mr_ops->route_action_update(mlxsw_sp, rve->mr_route->route_priv,
588 					MLXSW_SP_MR_ROUTE_ACTION_TRAP);
589 	rve->mr_route->route_action = MLXSW_SP_MR_ROUTE_ACTION_TRAP;
590 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
591 }
592 
593 /* Should be called after the RIF struct is updated */
594 static int
595 mlxsw_sp_mr_route_evif_resolve(struct mlxsw_sp_mr_table *mr_table,
596 			       struct mlxsw_sp_mr_route_vif_entry *rve)
597 {
598 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
599 	enum mlxsw_sp_mr_route_action route_action;
600 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
601 	u16 erif_index = 0;
602 	int err;
603 
604 	/* Update the route action, as the new eVIF can be a tunnel or a pimreg
605 	 * device which will require updating the action.
606 	 */
607 	route_action = mlxsw_sp_mr_route_action(rve->mr_route);
608 	if (route_action != rve->mr_route->route_action) {
609 		err = mr->mr_ops->route_action_update(mlxsw_sp,
610 						      rve->mr_route->route_priv,
611 						      route_action);
612 		if (err)
613 			return err;
614 	}
615 
616 	/* Add the eRIF */
617 	if (mlxsw_sp_mr_vif_valid(rve->mr_vif)) {
618 		erif_index = mlxsw_sp_rif_index(rve->mr_vif->rif);
619 		err = mr->mr_ops->route_erif_add(mlxsw_sp,
620 						 rve->mr_route->route_priv,
621 						 erif_index);
622 		if (err)
623 			goto err_route_erif_add;
624 	}
625 
626 	/* Update the minimum MTU */
627 	if (rve->mr_vif->dev->mtu < rve->mr_route->min_mtu) {
628 		rve->mr_route->min_mtu = rve->mr_vif->dev->mtu;
629 		err = mr->mr_ops->route_min_mtu_update(mlxsw_sp,
630 						       rve->mr_route->route_priv,
631 						       rve->mr_route->min_mtu);
632 		if (err)
633 			goto err_route_min_mtu_update;
634 	}
635 
636 	rve->mr_route->route_action = route_action;
637 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
638 	return 0;
639 
640 err_route_min_mtu_update:
641 	if (mlxsw_sp_mr_vif_valid(rve->mr_vif))
642 		mr->mr_ops->route_erif_del(mlxsw_sp, rve->mr_route->route_priv,
643 					   erif_index);
644 err_route_erif_add:
645 	if (route_action != rve->mr_route->route_action)
646 		mr->mr_ops->route_action_update(mlxsw_sp,
647 						rve->mr_route->route_priv,
648 						rve->mr_route->route_action);
649 	return err;
650 }
651 
652 /* Should be called before the RIF struct is updated */
653 static void
654 mlxsw_sp_mr_route_evif_unresolve(struct mlxsw_sp_mr_table *mr_table,
655 				 struct mlxsw_sp_mr_route_vif_entry *rve)
656 {
657 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
658 	enum mlxsw_sp_mr_route_action route_action;
659 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
660 	u16 rifi;
661 
662 	/* If the unresolved RIF was not valid, no need to delete it */
663 	if (!mlxsw_sp_mr_vif_valid(rve->mr_vif))
664 		return;
665 
666 	/* Update the route action: if there is only one valid eVIF in the
667 	 * route, set the action to trap as the VIF deletion will lead to zero
668 	 * valid eVIFs. On any other case, use the mlxsw_sp_mr_route_action to
669 	 * determine the route action.
670 	 */
671 	if (mlxsw_sp_mr_route_valid_evifs_num(rve->mr_route) == 1)
672 		route_action = MLXSW_SP_MR_ROUTE_ACTION_TRAP;
673 	else
674 		route_action = mlxsw_sp_mr_route_action(rve->mr_route);
675 	if (route_action != rve->mr_route->route_action)
676 		mr->mr_ops->route_action_update(mlxsw_sp,
677 						rve->mr_route->route_priv,
678 						route_action);
679 
680 	/* Delete the erif from the route */
681 	rifi = mlxsw_sp_rif_index(rve->mr_vif->rif);
682 	mr->mr_ops->route_erif_del(mlxsw_sp, rve->mr_route->route_priv, rifi);
683 	rve->mr_route->route_action = route_action;
684 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
685 }
686 
687 static int mlxsw_sp_mr_vif_resolve(struct mlxsw_sp_mr_table *mr_table,
688 				   struct net_device *dev,
689 				   struct mlxsw_sp_mr_vif *mr_vif,
690 				   unsigned long vif_flags,
691 				   const struct mlxsw_sp_rif *rif)
692 {
693 	struct mlxsw_sp_mr_route_vif_entry *irve, *erve;
694 	int err;
695 
696 	/* Update the VIF */
697 	mr_vif->dev = dev;
698 	mr_vif->rif = rif;
699 	mr_vif->vif_flags = vif_flags;
700 
701 	/* Update all routes where this VIF is used as an unresolved iRIF */
702 	list_for_each_entry(irve, &mr_vif->route_ivif_list, vif_node) {
703 		err = mlxsw_sp_mr_route_ivif_resolve(mr_table, irve);
704 		if (err)
705 			goto err_irif_unresolve;
706 	}
707 
708 	/* Update all routes where this VIF is used as an unresolved eRIF */
709 	list_for_each_entry(erve, &mr_vif->route_evif_list, vif_node) {
710 		err = mlxsw_sp_mr_route_evif_resolve(mr_table, erve);
711 		if (err)
712 			goto err_erif_unresolve;
713 	}
714 	return 0;
715 
716 err_erif_unresolve:
717 	list_for_each_entry_from_reverse(erve, &mr_vif->route_evif_list,
718 					 vif_node)
719 		mlxsw_sp_mr_route_evif_unresolve(mr_table, erve);
720 err_irif_unresolve:
721 	list_for_each_entry_from_reverse(irve, &mr_vif->route_ivif_list,
722 					 vif_node)
723 		mlxsw_sp_mr_route_ivif_unresolve(mr_table, irve);
724 	mr_vif->rif = NULL;
725 	return err;
726 }
727 
728 static void mlxsw_sp_mr_vif_unresolve(struct mlxsw_sp_mr_table *mr_table,
729 				      struct net_device *dev,
730 				      struct mlxsw_sp_mr_vif *mr_vif)
731 {
732 	struct mlxsw_sp_mr_route_vif_entry *rve;
733 
734 	/* Update all routes where this VIF is used as an unresolved eRIF */
735 	list_for_each_entry(rve, &mr_vif->route_evif_list, vif_node)
736 		mlxsw_sp_mr_route_evif_unresolve(mr_table, rve);
737 
738 	/* Update all routes where this VIF is used as an unresolved iRIF */
739 	list_for_each_entry(rve, &mr_vif->route_ivif_list, vif_node)
740 		mlxsw_sp_mr_route_ivif_unresolve(mr_table, rve);
741 
742 	/* Update the VIF */
743 	mr_vif->dev = dev;
744 	mr_vif->rif = NULL;
745 }
746 
747 int mlxsw_sp_mr_vif_add(struct mlxsw_sp_mr_table *mr_table,
748 			struct net_device *dev, vifi_t vif_index,
749 			unsigned long vif_flags, const struct mlxsw_sp_rif *rif)
750 {
751 	struct mlxsw_sp_mr_vif *mr_vif = &mr_table->vifs[vif_index];
752 
753 	if (WARN_ON(vif_index >= MAXVIFS))
754 		return -EINVAL;
755 	if (mr_vif->dev)
756 		return -EEXIST;
757 	return mlxsw_sp_mr_vif_resolve(mr_table, dev, mr_vif, vif_flags, rif);
758 }
759 
760 void mlxsw_sp_mr_vif_del(struct mlxsw_sp_mr_table *mr_table, vifi_t vif_index)
761 {
762 	struct mlxsw_sp_mr_vif *mr_vif = &mr_table->vifs[vif_index];
763 
764 	if (WARN_ON(vif_index >= MAXVIFS))
765 		return;
766 	if (WARN_ON(!mr_vif->dev))
767 		return;
768 	mlxsw_sp_mr_vif_unresolve(mr_table, NULL, mr_vif);
769 }
770 
771 static struct mlxsw_sp_mr_vif *
772 mlxsw_sp_mr_dev_vif_lookup(struct mlxsw_sp_mr_table *mr_table,
773 			   const struct net_device *dev)
774 {
775 	vifi_t vif_index;
776 
777 	for (vif_index = 0; vif_index < MAXVIFS; vif_index++)
778 		if (mr_table->vifs[vif_index].dev == dev)
779 			return &mr_table->vifs[vif_index];
780 	return NULL;
781 }
782 
783 int mlxsw_sp_mr_rif_add(struct mlxsw_sp_mr_table *mr_table,
784 			const struct mlxsw_sp_rif *rif)
785 {
786 	const struct net_device *rif_dev = mlxsw_sp_rif_dev(rif);
787 	struct mlxsw_sp_mr_vif *mr_vif;
788 
789 	if (!rif_dev)
790 		return 0;
791 
792 	mr_vif = mlxsw_sp_mr_dev_vif_lookup(mr_table, rif_dev);
793 	if (!mr_vif)
794 		return 0;
795 	return mlxsw_sp_mr_vif_resolve(mr_table, mr_vif->dev, mr_vif,
796 				       mr_vif->vif_flags, rif);
797 }
798 
799 void mlxsw_sp_mr_rif_del(struct mlxsw_sp_mr_table *mr_table,
800 			 const struct mlxsw_sp_rif *rif)
801 {
802 	const struct net_device *rif_dev = mlxsw_sp_rif_dev(rif);
803 	struct mlxsw_sp_mr_vif *mr_vif;
804 
805 	if (!rif_dev)
806 		return;
807 
808 	mr_vif = mlxsw_sp_mr_dev_vif_lookup(mr_table, rif_dev);
809 	if (!mr_vif)
810 		return;
811 	mlxsw_sp_mr_vif_unresolve(mr_table, mr_vif->dev, mr_vif);
812 }
813 
814 void mlxsw_sp_mr_rif_mtu_update(struct mlxsw_sp_mr_table *mr_table,
815 				const struct mlxsw_sp_rif *rif, int mtu)
816 {
817 	const struct net_device *rif_dev = mlxsw_sp_rif_dev(rif);
818 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
819 	struct mlxsw_sp_mr_route_vif_entry *rve;
820 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
821 	struct mlxsw_sp_mr_vif *mr_vif;
822 
823 	if (!rif_dev)
824 		return;
825 
826 	/* Search for a VIF that use that RIF */
827 	mr_vif = mlxsw_sp_mr_dev_vif_lookup(mr_table, rif_dev);
828 	if (!mr_vif)
829 		return;
830 
831 	/* Update all the routes that uses that VIF as eVIF */
832 	list_for_each_entry(rve, &mr_vif->route_evif_list, vif_node) {
833 		if (mtu < rve->mr_route->min_mtu) {
834 			rve->mr_route->min_mtu = mtu;
835 			mr->mr_ops->route_min_mtu_update(mlxsw_sp,
836 							 rve->mr_route->route_priv,
837 							 mtu);
838 		}
839 	}
840 }
841 
842 struct mlxsw_sp_mr_table *mlxsw_sp_mr_table_create(struct mlxsw_sp *mlxsw_sp,
843 						   u32 vr_id,
844 						   enum mlxsw_sp_l3proto proto)
845 {
846 	struct mlxsw_sp_mr_route_params catchall_route_params = {
847 		.prio = MLXSW_SP_MR_ROUTE_PRIO_CATCHALL,
848 		.key = {
849 			.vrid = vr_id,
850 		},
851 		.value = {
852 			.route_action = MLXSW_SP_MR_ROUTE_ACTION_TRAP,
853 		}
854 	};
855 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
856 	struct mlxsw_sp_mr_table *mr_table;
857 	int err;
858 	int i;
859 
860 	mr_table = kzalloc(sizeof(*mr_table) + mr->mr_ops->route_priv_size,
861 			   GFP_KERNEL);
862 	if (!mr_table)
863 		return ERR_PTR(-ENOMEM);
864 
865 	mr_table->vr_id = vr_id;
866 	mr_table->mlxsw_sp = mlxsw_sp;
867 	mr_table->proto = proto;
868 	INIT_LIST_HEAD(&mr_table->route_list);
869 
870 	err = rhashtable_init(&mr_table->route_ht,
871 			      &mlxsw_sp_mr_route_ht_params);
872 	if (err)
873 		goto err_route_rhashtable_init;
874 
875 	for (i = 0; i < MAXVIFS; i++) {
876 		INIT_LIST_HEAD(&mr_table->vifs[i].route_evif_list);
877 		INIT_LIST_HEAD(&mr_table->vifs[i].route_ivif_list);
878 	}
879 
880 	err = mr->mr_ops->route_create(mlxsw_sp, mr->priv,
881 				       mr_table->catchall_route_priv,
882 				       &catchall_route_params);
883 	if (err)
884 		goto err_ops_route_create;
885 	list_add_tail(&mr_table->node, &mr->table_list);
886 	return mr_table;
887 
888 err_ops_route_create:
889 	rhashtable_destroy(&mr_table->route_ht);
890 err_route_rhashtable_init:
891 	kfree(mr_table);
892 	return ERR_PTR(err);
893 }
894 
895 void mlxsw_sp_mr_table_destroy(struct mlxsw_sp_mr_table *mr_table)
896 {
897 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
898 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
899 
900 	WARN_ON(!mlxsw_sp_mr_table_empty(mr_table));
901 	list_del(&mr_table->node);
902 	mr->mr_ops->route_destroy(mlxsw_sp, mr->priv,
903 				  &mr_table->catchall_route_priv);
904 	rhashtable_destroy(&mr_table->route_ht);
905 	kfree(mr_table);
906 }
907 
908 void mlxsw_sp_mr_table_flush(struct mlxsw_sp_mr_table *mr_table)
909 {
910 	struct mlxsw_sp_mr_route *mr_route, *tmp;
911 	int i;
912 
913 	list_for_each_entry_safe(mr_route, tmp, &mr_table->route_list, node)
914 		__mlxsw_sp_mr_route_del(mr_table, mr_route);
915 
916 	for (i = 0; i < MAXVIFS; i++) {
917 		mr_table->vifs[i].dev = NULL;
918 		mr_table->vifs[i].rif = NULL;
919 	}
920 }
921 
922 bool mlxsw_sp_mr_table_empty(const struct mlxsw_sp_mr_table *mr_table)
923 {
924 	int i;
925 
926 	for (i = 0; i < MAXVIFS; i++)
927 		if (mr_table->vifs[i].dev)
928 			return false;
929 	return list_empty(&mr_table->route_list);
930 }
931 
932 static void mlxsw_sp_mr_route_stats_update(struct mlxsw_sp *mlxsw_sp,
933 					   struct mlxsw_sp_mr_route *mr_route)
934 {
935 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
936 	u64 packets, bytes;
937 
938 	if (mr_route->route_action == MLXSW_SP_MR_ROUTE_ACTION_TRAP)
939 		return;
940 
941 	mr->mr_ops->route_stats(mlxsw_sp, mr_route->route_priv, &packets,
942 				&bytes);
943 
944 	switch (mr_route->mr_table->proto) {
945 	case MLXSW_SP_L3_PROTO_IPV4:
946 		if (mr_route->mfc4->mfc_un.res.pkt != packets)
947 			mr_route->mfc4->mfc_un.res.lastuse = jiffies;
948 		mr_route->mfc4->mfc_un.res.pkt = packets;
949 		mr_route->mfc4->mfc_un.res.bytes = bytes;
950 		break;
951 	case MLXSW_SP_L3_PROTO_IPV6:
952 		/* fall through */
953 	default:
954 		WARN_ON_ONCE(1);
955 	}
956 }
957 
958 static void mlxsw_sp_mr_stats_update(struct work_struct *work)
959 {
960 	struct mlxsw_sp_mr *mr = container_of(work, struct mlxsw_sp_mr,
961 					      stats_update_dw.work);
962 	struct mlxsw_sp_mr_table *mr_table;
963 	struct mlxsw_sp_mr_route *mr_route;
964 	unsigned long interval;
965 
966 	rtnl_lock();
967 	list_for_each_entry(mr_table, &mr->table_list, node)
968 		list_for_each_entry(mr_route, &mr_table->route_list, node)
969 			mlxsw_sp_mr_route_stats_update(mr_table->mlxsw_sp,
970 						       mr_route);
971 	rtnl_unlock();
972 
973 	interval = msecs_to_jiffies(MLXSW_SP_MR_ROUTES_COUNTER_UPDATE_INTERVAL);
974 	mlxsw_core_schedule_dw(&mr->stats_update_dw, interval);
975 }
976 
977 int mlxsw_sp_mr_init(struct mlxsw_sp *mlxsw_sp,
978 		     const struct mlxsw_sp_mr_ops *mr_ops)
979 {
980 	struct mlxsw_sp_mr *mr;
981 	unsigned long interval;
982 	int err;
983 
984 	mr = kzalloc(sizeof(*mr) + mr_ops->priv_size, GFP_KERNEL);
985 	if (!mr)
986 		return -ENOMEM;
987 	mr->mr_ops = mr_ops;
988 	mlxsw_sp->mr = mr;
989 	INIT_LIST_HEAD(&mr->table_list);
990 
991 	err = mr_ops->init(mlxsw_sp, mr->priv);
992 	if (err)
993 		goto err;
994 
995 	/* Create the delayed work for counter updates */
996 	INIT_DELAYED_WORK(&mr->stats_update_dw, mlxsw_sp_mr_stats_update);
997 	interval = msecs_to_jiffies(MLXSW_SP_MR_ROUTES_COUNTER_UPDATE_INTERVAL);
998 	mlxsw_core_schedule_dw(&mr->stats_update_dw, interval);
999 	return 0;
1000 err:
1001 	kfree(mr);
1002 	return err;
1003 }
1004 
1005 void mlxsw_sp_mr_fini(struct mlxsw_sp *mlxsw_sp)
1006 {
1007 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
1008 
1009 	cancel_delayed_work_sync(&mr->stats_update_dw);
1010 	mr->mr_ops->fini(mr->priv);
1011 	kfree(mr);
1012 }
1013