xref: /openbmc/linux/drivers/net/ethernet/mellanox/mlxsw/spectrum_mr.c (revision 28efb0046512e8a13ed9f9bdf0d68d10bbfbe9cf)
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 == 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 == 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 = 0xffffffff;
213 	key->source.addr4 = mfc->mfc_origin;
214 	key->source_mask.addr4 = 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 == INADDR_ANY && mfc->mfc_mcastgrp == INADDR_ANY) {
462 		dev_warn(mr_table->mlxsw_sp->bus_info->dev,
463 			 "Offloading proxy routes is not supported.\n");
464 		return -EINVAL;
465 	}
466 
467 	/* Create a new route */
468 	mr_route = mlxsw_sp_mr_route4_create(mr_table, mfc);
469 	if (IS_ERR(mr_route))
470 		return PTR_ERR(mr_route);
471 
472 	/* Find any route with a matching key */
473 	mr_orig_route = rhashtable_lookup_fast(&mr_table->route_ht,
474 					       &mr_route->key,
475 					       mlxsw_sp_mr_route_ht_params);
476 	if (replace) {
477 		/* On replace case, make the route point to the new route_priv.
478 		 */
479 		if (WARN_ON(!mr_orig_route)) {
480 			err = -ENOENT;
481 			goto err_no_orig_route;
482 		}
483 		mr_route->route_priv = mr_orig_route->route_priv;
484 	} else if (mr_orig_route) {
485 		/* On non replace case, if another route with the same key was
486 		 * found, abort, as duplicate routes are used for proxy routes.
487 		 */
488 		dev_warn(mr_table->mlxsw_sp->bus_info->dev,
489 			 "Offloading proxy routes is not supported.\n");
490 		err = -EINVAL;
491 		goto err_duplicate_route;
492 	}
493 
494 	/* Put it in the table data-structures */
495 	list_add_tail(&mr_route->node, &mr_table->route_list);
496 	err = rhashtable_insert_fast(&mr_table->route_ht,
497 				     &mr_route->ht_node,
498 				     mlxsw_sp_mr_route_ht_params);
499 	if (err)
500 		goto err_rhashtable_insert;
501 
502 	/* Write the route to the hardware */
503 	err = mlxsw_sp_mr_route_write(mr_table, mr_route, replace);
504 	if (err)
505 		goto err_mr_route_write;
506 
507 	/* Destroy the original route */
508 	if (replace) {
509 		rhashtable_remove_fast(&mr_table->route_ht,
510 				       &mr_orig_route->ht_node,
511 				       mlxsw_sp_mr_route_ht_params);
512 		list_del(&mr_orig_route->node);
513 		mlxsw_sp_mr_route4_destroy(mr_table, mr_orig_route);
514 	}
515 
516 	mlxsw_sp_mr_mfc_offload_update(mr_route);
517 	return 0;
518 
519 err_mr_route_write:
520 	rhashtable_remove_fast(&mr_table->route_ht, &mr_route->ht_node,
521 			       mlxsw_sp_mr_route_ht_params);
522 err_rhashtable_insert:
523 	list_del(&mr_route->node);
524 err_no_orig_route:
525 err_duplicate_route:
526 	mlxsw_sp_mr_route4_destroy(mr_table, mr_route);
527 	return err;
528 }
529 
530 void mlxsw_sp_mr_route4_del(struct mlxsw_sp_mr_table *mr_table,
531 			    struct mfc_cache *mfc)
532 {
533 	struct mlxsw_sp_mr_route *mr_route;
534 	struct mlxsw_sp_mr_route_key key;
535 
536 	mlxsw_sp_mr_route4_key(mr_table, &key, mfc);
537 	mr_route = rhashtable_lookup_fast(&mr_table->route_ht, &key,
538 					  mlxsw_sp_mr_route_ht_params);
539 	if (mr_route)
540 		__mlxsw_sp_mr_route_del(mr_table, mr_route);
541 }
542 
543 /* Should be called after the VIF struct is updated */
544 static int
545 mlxsw_sp_mr_route_ivif_resolve(struct mlxsw_sp_mr_table *mr_table,
546 			       struct mlxsw_sp_mr_route_vif_entry *rve)
547 {
548 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
549 	enum mlxsw_sp_mr_route_action route_action;
550 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
551 	u16 irif_index;
552 	int err;
553 
554 	route_action = mlxsw_sp_mr_route_action(rve->mr_route);
555 	if (route_action == MLXSW_SP_MR_ROUTE_ACTION_TRAP)
556 		return 0;
557 
558 	/* rve->mr_vif->rif is guaranteed to be valid at this stage */
559 	irif_index = mlxsw_sp_rif_index(rve->mr_vif->rif);
560 	err = mr->mr_ops->route_irif_update(mlxsw_sp, rve->mr_route->route_priv,
561 					    irif_index);
562 	if (err)
563 		return err;
564 
565 	err = mr->mr_ops->route_action_update(mlxsw_sp,
566 					      rve->mr_route->route_priv,
567 					      route_action);
568 	if (err)
569 		/* No need to rollback here because the iRIF change only takes
570 		 * place after the action has been updated.
571 		 */
572 		return err;
573 
574 	rve->mr_route->route_action = route_action;
575 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
576 	return 0;
577 }
578 
579 static void
580 mlxsw_sp_mr_route_ivif_unresolve(struct mlxsw_sp_mr_table *mr_table,
581 				 struct mlxsw_sp_mr_route_vif_entry *rve)
582 {
583 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
584 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
585 
586 	mr->mr_ops->route_action_update(mlxsw_sp, rve->mr_route->route_priv,
587 					MLXSW_SP_MR_ROUTE_ACTION_TRAP);
588 	rve->mr_route->route_action = MLXSW_SP_MR_ROUTE_ACTION_TRAP;
589 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
590 }
591 
592 /* Should be called after the RIF struct is updated */
593 static int
594 mlxsw_sp_mr_route_evif_resolve(struct mlxsw_sp_mr_table *mr_table,
595 			       struct mlxsw_sp_mr_route_vif_entry *rve)
596 {
597 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
598 	enum mlxsw_sp_mr_route_action route_action;
599 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
600 	u16 erif_index = 0;
601 	int err;
602 
603 	/* Update the route action, as the new eVIF can be a tunnel or a pimreg
604 	 * device which will require updating the action.
605 	 */
606 	route_action = mlxsw_sp_mr_route_action(rve->mr_route);
607 	if (route_action != rve->mr_route->route_action) {
608 		err = mr->mr_ops->route_action_update(mlxsw_sp,
609 						      rve->mr_route->route_priv,
610 						      route_action);
611 		if (err)
612 			return err;
613 	}
614 
615 	/* Add the eRIF */
616 	if (mlxsw_sp_mr_vif_valid(rve->mr_vif)) {
617 		erif_index = mlxsw_sp_rif_index(rve->mr_vif->rif);
618 		err = mr->mr_ops->route_erif_add(mlxsw_sp,
619 						 rve->mr_route->route_priv,
620 						 erif_index);
621 		if (err)
622 			goto err_route_erif_add;
623 	}
624 
625 	/* Update the minimum MTU */
626 	if (rve->mr_vif->dev->mtu < rve->mr_route->min_mtu) {
627 		rve->mr_route->min_mtu = rve->mr_vif->dev->mtu;
628 		err = mr->mr_ops->route_min_mtu_update(mlxsw_sp,
629 						       rve->mr_route->route_priv,
630 						       rve->mr_route->min_mtu);
631 		if (err)
632 			goto err_route_min_mtu_update;
633 	}
634 
635 	rve->mr_route->route_action = route_action;
636 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
637 	return 0;
638 
639 err_route_min_mtu_update:
640 	if (mlxsw_sp_mr_vif_valid(rve->mr_vif))
641 		mr->mr_ops->route_erif_del(mlxsw_sp, rve->mr_route->route_priv,
642 					   erif_index);
643 err_route_erif_add:
644 	if (route_action != rve->mr_route->route_action)
645 		mr->mr_ops->route_action_update(mlxsw_sp,
646 						rve->mr_route->route_priv,
647 						rve->mr_route->route_action);
648 	return err;
649 }
650 
651 /* Should be called before the RIF struct is updated */
652 static void
653 mlxsw_sp_mr_route_evif_unresolve(struct mlxsw_sp_mr_table *mr_table,
654 				 struct mlxsw_sp_mr_route_vif_entry *rve)
655 {
656 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
657 	enum mlxsw_sp_mr_route_action route_action;
658 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
659 	u16 rifi;
660 
661 	/* If the unresolved RIF was not valid, no need to delete it */
662 	if (!mlxsw_sp_mr_vif_valid(rve->mr_vif))
663 		return;
664 
665 	/* Update the route action: if there is only one valid eVIF in the
666 	 * route, set the action to trap as the VIF deletion will lead to zero
667 	 * valid eVIFs. On any other case, use the mlxsw_sp_mr_route_action to
668 	 * determine the route action.
669 	 */
670 	if (mlxsw_sp_mr_route_valid_evifs_num(rve->mr_route) == 1)
671 		route_action = MLXSW_SP_MR_ROUTE_ACTION_TRAP;
672 	else
673 		route_action = mlxsw_sp_mr_route_action(rve->mr_route);
674 	if (route_action != rve->mr_route->route_action)
675 		mr->mr_ops->route_action_update(mlxsw_sp,
676 						rve->mr_route->route_priv,
677 						route_action);
678 
679 	/* Delete the erif from the route */
680 	rifi = mlxsw_sp_rif_index(rve->mr_vif->rif);
681 	mr->mr_ops->route_erif_del(mlxsw_sp, rve->mr_route->route_priv, rifi);
682 	rve->mr_route->route_action = route_action;
683 	mlxsw_sp_mr_mfc_offload_update(rve->mr_route);
684 }
685 
686 static int mlxsw_sp_mr_vif_resolve(struct mlxsw_sp_mr_table *mr_table,
687 				   struct net_device *dev,
688 				   struct mlxsw_sp_mr_vif *mr_vif,
689 				   unsigned long vif_flags,
690 				   const struct mlxsw_sp_rif *rif)
691 {
692 	struct mlxsw_sp_mr_route_vif_entry *irve, *erve;
693 	int err;
694 
695 	/* Update the VIF */
696 	mr_vif->dev = dev;
697 	mr_vif->rif = rif;
698 	mr_vif->vif_flags = vif_flags;
699 
700 	/* Update all routes where this VIF is used as an unresolved iRIF */
701 	list_for_each_entry(irve, &mr_vif->route_ivif_list, vif_node) {
702 		err = mlxsw_sp_mr_route_ivif_resolve(mr_table, irve);
703 		if (err)
704 			goto err_irif_unresolve;
705 	}
706 
707 	/* Update all routes where this VIF is used as an unresolved eRIF */
708 	list_for_each_entry(erve, &mr_vif->route_evif_list, vif_node) {
709 		err = mlxsw_sp_mr_route_evif_resolve(mr_table, erve);
710 		if (err)
711 			goto err_erif_unresolve;
712 	}
713 	return 0;
714 
715 err_erif_unresolve:
716 	list_for_each_entry_from_reverse(erve, &mr_vif->route_evif_list,
717 					 vif_node)
718 		mlxsw_sp_mr_route_evif_unresolve(mr_table, erve);
719 err_irif_unresolve:
720 	list_for_each_entry_from_reverse(irve, &mr_vif->route_ivif_list,
721 					 vif_node)
722 		mlxsw_sp_mr_route_ivif_unresolve(mr_table, irve);
723 	mr_vif->rif = NULL;
724 	return err;
725 }
726 
727 static void mlxsw_sp_mr_vif_unresolve(struct mlxsw_sp_mr_table *mr_table,
728 				      struct net_device *dev,
729 				      struct mlxsw_sp_mr_vif *mr_vif)
730 {
731 	struct mlxsw_sp_mr_route_vif_entry *rve;
732 
733 	/* Update all routes where this VIF is used as an unresolved eRIF */
734 	list_for_each_entry(rve, &mr_vif->route_evif_list, vif_node)
735 		mlxsw_sp_mr_route_evif_unresolve(mr_table, rve);
736 
737 	/* Update all routes where this VIF is used as an unresolved iRIF */
738 	list_for_each_entry(rve, &mr_vif->route_ivif_list, vif_node)
739 		mlxsw_sp_mr_route_ivif_unresolve(mr_table, rve);
740 
741 	/* Update the VIF */
742 	mr_vif->dev = dev;
743 	mr_vif->rif = NULL;
744 }
745 
746 int mlxsw_sp_mr_vif_add(struct mlxsw_sp_mr_table *mr_table,
747 			struct net_device *dev, vifi_t vif_index,
748 			unsigned long vif_flags, const struct mlxsw_sp_rif *rif)
749 {
750 	struct mlxsw_sp_mr_vif *mr_vif = &mr_table->vifs[vif_index];
751 
752 	if (WARN_ON(vif_index >= MAXVIFS))
753 		return -EINVAL;
754 	if (mr_vif->dev)
755 		return -EEXIST;
756 	return mlxsw_sp_mr_vif_resolve(mr_table, dev, mr_vif, vif_flags, rif);
757 }
758 
759 void mlxsw_sp_mr_vif_del(struct mlxsw_sp_mr_table *mr_table, vifi_t vif_index)
760 {
761 	struct mlxsw_sp_mr_vif *mr_vif = &mr_table->vifs[vif_index];
762 
763 	if (WARN_ON(vif_index >= MAXVIFS))
764 		return;
765 	if (WARN_ON(!mr_vif->dev))
766 		return;
767 	mlxsw_sp_mr_vif_unresolve(mr_table, NULL, mr_vif);
768 }
769 
770 struct mlxsw_sp_mr_vif *
771 mlxsw_sp_mr_dev_vif_lookup(struct mlxsw_sp_mr_table *mr_table,
772 			   const struct net_device *dev)
773 {
774 	vifi_t vif_index;
775 
776 	for (vif_index = 0; vif_index < MAXVIFS; vif_index++)
777 		if (mr_table->vifs[vif_index].dev == dev)
778 			return &mr_table->vifs[vif_index];
779 	return NULL;
780 }
781 
782 int mlxsw_sp_mr_rif_add(struct mlxsw_sp_mr_table *mr_table,
783 			const struct mlxsw_sp_rif *rif)
784 {
785 	const struct net_device *rif_dev = mlxsw_sp_rif_dev(rif);
786 	struct mlxsw_sp_mr_vif *mr_vif;
787 
788 	if (!rif_dev)
789 		return 0;
790 
791 	mr_vif = mlxsw_sp_mr_dev_vif_lookup(mr_table, rif_dev);
792 	if (!mr_vif)
793 		return 0;
794 	return mlxsw_sp_mr_vif_resolve(mr_table, mr_vif->dev, mr_vif,
795 				       mr_vif->vif_flags, rif);
796 }
797 
798 void mlxsw_sp_mr_rif_del(struct mlxsw_sp_mr_table *mr_table,
799 			 const struct mlxsw_sp_rif *rif)
800 {
801 	const struct net_device *rif_dev = mlxsw_sp_rif_dev(rif);
802 	struct mlxsw_sp_mr_vif *mr_vif;
803 
804 	if (!rif_dev)
805 		return;
806 
807 	mr_vif = mlxsw_sp_mr_dev_vif_lookup(mr_table, rif_dev);
808 	if (!mr_vif)
809 		return;
810 	mlxsw_sp_mr_vif_unresolve(mr_table, mr_vif->dev, mr_vif);
811 }
812 
813 void mlxsw_sp_mr_rif_mtu_update(struct mlxsw_sp_mr_table *mr_table,
814 				const struct mlxsw_sp_rif *rif, int mtu)
815 {
816 	const struct net_device *rif_dev = mlxsw_sp_rif_dev(rif);
817 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
818 	struct mlxsw_sp_mr_route_vif_entry *rve;
819 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
820 	struct mlxsw_sp_mr_vif *mr_vif;
821 
822 	if (!rif_dev)
823 		return;
824 
825 	/* Search for a VIF that use that RIF */
826 	mr_vif = mlxsw_sp_mr_dev_vif_lookup(mr_table, rif_dev);
827 	if (!mr_vif)
828 		return;
829 
830 	/* Update all the routes that uses that VIF as eVIF */
831 	list_for_each_entry(rve, &mr_vif->route_evif_list, vif_node) {
832 		if (mtu < rve->mr_route->min_mtu) {
833 			rve->mr_route->min_mtu = mtu;
834 			mr->mr_ops->route_min_mtu_update(mlxsw_sp,
835 							 rve->mr_route->route_priv,
836 							 mtu);
837 		}
838 	}
839 }
840 
841 struct mlxsw_sp_mr_table *mlxsw_sp_mr_table_create(struct mlxsw_sp *mlxsw_sp,
842 						   u32 vr_id,
843 						   enum mlxsw_sp_l3proto proto)
844 {
845 	struct mlxsw_sp_mr_route_params catchall_route_params = {
846 		.prio = MLXSW_SP_MR_ROUTE_PRIO_CATCHALL,
847 		.key = {
848 			.vrid = vr_id,
849 		},
850 		.value = {
851 			.route_action = MLXSW_SP_MR_ROUTE_ACTION_TRAP,
852 		}
853 	};
854 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
855 	struct mlxsw_sp_mr_table *mr_table;
856 	int err;
857 	int i;
858 
859 	mr_table = kzalloc(sizeof(*mr_table) + mr->mr_ops->route_priv_size,
860 			   GFP_KERNEL);
861 	if (!mr_table)
862 		return ERR_PTR(-ENOMEM);
863 
864 	mr_table->vr_id = vr_id;
865 	mr_table->mlxsw_sp = mlxsw_sp;
866 	mr_table->proto = proto;
867 	INIT_LIST_HEAD(&mr_table->route_list);
868 
869 	err = rhashtable_init(&mr_table->route_ht,
870 			      &mlxsw_sp_mr_route_ht_params);
871 	if (err)
872 		goto err_route_rhashtable_init;
873 
874 	for (i = 0; i < MAXVIFS; i++) {
875 		INIT_LIST_HEAD(&mr_table->vifs[i].route_evif_list);
876 		INIT_LIST_HEAD(&mr_table->vifs[i].route_ivif_list);
877 	}
878 
879 	err = mr->mr_ops->route_create(mlxsw_sp, mr->priv,
880 				       mr_table->catchall_route_priv,
881 				       &catchall_route_params);
882 	if (err)
883 		goto err_ops_route_create;
884 	list_add_tail(&mr_table->node, &mr->table_list);
885 	return mr_table;
886 
887 err_ops_route_create:
888 	rhashtable_destroy(&mr_table->route_ht);
889 err_route_rhashtable_init:
890 	kfree(mr_table);
891 	return ERR_PTR(err);
892 }
893 
894 void mlxsw_sp_mr_table_destroy(struct mlxsw_sp_mr_table *mr_table)
895 {
896 	struct mlxsw_sp *mlxsw_sp = mr_table->mlxsw_sp;
897 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
898 
899 	WARN_ON(!mlxsw_sp_mr_table_empty(mr_table));
900 	list_del(&mr_table->node);
901 	mr->mr_ops->route_destroy(mlxsw_sp, mr->priv,
902 				  &mr_table->catchall_route_priv);
903 	rhashtable_destroy(&mr_table->route_ht);
904 	kfree(mr_table);
905 }
906 
907 void mlxsw_sp_mr_table_flush(struct mlxsw_sp_mr_table *mr_table)
908 {
909 	struct mlxsw_sp_mr_route *mr_route, *tmp;
910 	int i;
911 
912 	list_for_each_entry_safe(mr_route, tmp, &mr_table->route_list, node)
913 		__mlxsw_sp_mr_route_del(mr_table, mr_route);
914 
915 	for (i = 0; i < MAXVIFS; i++) {
916 		mr_table->vifs[i].dev = NULL;
917 		mr_table->vifs[i].rif = NULL;
918 	}
919 }
920 
921 bool mlxsw_sp_mr_table_empty(const struct mlxsw_sp_mr_table *mr_table)
922 {
923 	int i;
924 
925 	for (i = 0; i < MAXVIFS; i++)
926 		if (mr_table->vifs[i].dev)
927 			return false;
928 	return list_empty(&mr_table->route_list);
929 }
930 
931 static void mlxsw_sp_mr_route_stats_update(struct mlxsw_sp *mlxsw_sp,
932 					   struct mlxsw_sp_mr_route *mr_route)
933 {
934 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
935 	u64 packets, bytes;
936 
937 	if (mr_route->route_action == MLXSW_SP_MR_ROUTE_ACTION_TRAP)
938 		return;
939 
940 	mr->mr_ops->route_stats(mlxsw_sp, mr_route->route_priv, &packets,
941 				&bytes);
942 
943 	switch (mr_route->mr_table->proto) {
944 	case MLXSW_SP_L3_PROTO_IPV4:
945 		if (mr_route->mfc4->mfc_un.res.pkt != packets)
946 			mr_route->mfc4->mfc_un.res.lastuse = jiffies;
947 		mr_route->mfc4->mfc_un.res.pkt = packets;
948 		mr_route->mfc4->mfc_un.res.bytes = bytes;
949 		break;
950 	case MLXSW_SP_L3_PROTO_IPV6:
951 		/* fall through */
952 	default:
953 		WARN_ON_ONCE(1);
954 	}
955 }
956 
957 static void mlxsw_sp_mr_stats_update(struct work_struct *work)
958 {
959 	struct mlxsw_sp_mr *mr = container_of(work, struct mlxsw_sp_mr,
960 					      stats_update_dw.work);
961 	struct mlxsw_sp_mr_table *mr_table;
962 	struct mlxsw_sp_mr_route *mr_route;
963 	unsigned long interval;
964 
965 	rtnl_lock();
966 	list_for_each_entry(mr_table, &mr->table_list, node)
967 		list_for_each_entry(mr_route, &mr_table->route_list, node)
968 			mlxsw_sp_mr_route_stats_update(mr_table->mlxsw_sp,
969 						       mr_route);
970 	rtnl_unlock();
971 
972 	interval = msecs_to_jiffies(MLXSW_SP_MR_ROUTES_COUNTER_UPDATE_INTERVAL);
973 	mlxsw_core_schedule_dw(&mr->stats_update_dw, interval);
974 }
975 
976 int mlxsw_sp_mr_init(struct mlxsw_sp *mlxsw_sp,
977 		     const struct mlxsw_sp_mr_ops *mr_ops)
978 {
979 	struct mlxsw_sp_mr *mr;
980 	unsigned long interval;
981 	int err;
982 
983 	mr = kzalloc(sizeof(*mr) + mr_ops->priv_size, GFP_KERNEL);
984 	if (!mr)
985 		return -ENOMEM;
986 	mr->mr_ops = mr_ops;
987 	mlxsw_sp->mr = mr;
988 	INIT_LIST_HEAD(&mr->table_list);
989 
990 	err = mr_ops->init(mlxsw_sp, mr->priv);
991 	if (err)
992 		goto err;
993 
994 	/* Create the delayed work for counter updates */
995 	INIT_DELAYED_WORK(&mr->stats_update_dw, mlxsw_sp_mr_stats_update);
996 	interval = msecs_to_jiffies(MLXSW_SP_MR_ROUTES_COUNTER_UPDATE_INTERVAL);
997 	mlxsw_core_schedule_dw(&mr->stats_update_dw, interval);
998 	return 0;
999 err:
1000 	kfree(mr);
1001 	return err;
1002 }
1003 
1004 void mlxsw_sp_mr_fini(struct mlxsw_sp *mlxsw_sp)
1005 {
1006 	struct mlxsw_sp_mr *mr = mlxsw_sp->mr;
1007 
1008 	cancel_delayed_work_sync(&mr->stats_update_dw);
1009 	mr->mr_ops->fini(mr->priv);
1010 	kfree(mr);
1011 }
1012