• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* Copyright (c) 2020 Mellanox Technologies. */
3 
4 #include <net/dst_metadata.h>
5 #include <linux/netdevice.h>
6 #include <linux/list.h>
7 #include <linux/rculist.h>
8 #include <linux/rtnetlink.h>
9 #include <linux/workqueue.h>
10 #include <linux/spinlock.h>
11 #include "tc.h"
12 #include "neigh.h"
13 #include "en_rep.h"
14 #include "eswitch.h"
15 #include "lib/fs_chains.h"
16 #include "en/tc_ct.h"
17 #include "en/mapping.h"
18 #include "en/tc_tun.h"
19 #include "lib/port_tun.h"
20 #include "en/tc/sample.h"
21 
22 struct mlx5e_rep_indr_block_priv {
23 	struct net_device *netdev;
24 	struct mlx5e_rep_priv *rpriv;
25 
26 	struct list_head list;
27 };
28 
mlx5e_rep_encap_entry_attach(struct mlx5e_priv * priv,struct mlx5e_encap_entry * e,struct mlx5e_neigh * m_neigh,struct net_device * neigh_dev)29 int mlx5e_rep_encap_entry_attach(struct mlx5e_priv *priv,
30 				 struct mlx5e_encap_entry *e,
31 				 struct mlx5e_neigh *m_neigh,
32 				 struct net_device *neigh_dev)
33 {
34 	struct mlx5e_rep_priv *rpriv = priv->ppriv;
35 	struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
36 	struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
37 	struct mlx5e_neigh_hash_entry *nhe;
38 	int err;
39 
40 	err = mlx5_tun_entropy_refcount_inc(tun_entropy, e->reformat_type);
41 	if (err)
42 		return err;
43 
44 	mutex_lock(&rpriv->neigh_update.encap_lock);
45 	nhe = mlx5e_rep_neigh_entry_lookup(priv, m_neigh);
46 	if (!nhe) {
47 		err = mlx5e_rep_neigh_entry_create(priv, m_neigh, neigh_dev, &nhe);
48 		if (err) {
49 			mutex_unlock(&rpriv->neigh_update.encap_lock);
50 			mlx5_tun_entropy_refcount_dec(tun_entropy,
51 						      e->reformat_type);
52 			return err;
53 		}
54 	}
55 
56 	e->nhe = nhe;
57 	spin_lock(&nhe->encap_list_lock);
58 	list_add_rcu(&e->encap_list, &nhe->encap_list);
59 	spin_unlock(&nhe->encap_list_lock);
60 
61 	mutex_unlock(&rpriv->neigh_update.encap_lock);
62 
63 	return 0;
64 }
65 
mlx5e_rep_encap_entry_detach(struct mlx5e_priv * priv,struct mlx5e_encap_entry * e)66 void mlx5e_rep_encap_entry_detach(struct mlx5e_priv *priv,
67 				  struct mlx5e_encap_entry *e)
68 {
69 	struct mlx5e_rep_priv *rpriv = priv->ppriv;
70 	struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
71 	struct mlx5_tun_entropy *tun_entropy = &uplink_priv->tun_entropy;
72 
73 	if (!e->nhe)
74 		return;
75 
76 	spin_lock(&e->nhe->encap_list_lock);
77 	list_del_rcu(&e->encap_list);
78 	spin_unlock(&e->nhe->encap_list_lock);
79 
80 	mlx5e_rep_neigh_entry_release(e->nhe);
81 	e->nhe = NULL;
82 	mlx5_tun_entropy_refcount_dec(tun_entropy, e->reformat_type);
83 }
84 
mlx5e_rep_update_flows(struct mlx5e_priv * priv,struct mlx5e_encap_entry * e,bool neigh_connected,unsigned char ha[ETH_ALEN])85 void mlx5e_rep_update_flows(struct mlx5e_priv *priv,
86 			    struct mlx5e_encap_entry *e,
87 			    bool neigh_connected,
88 			    unsigned char ha[ETH_ALEN])
89 {
90 	struct ethhdr *eth = (struct ethhdr *)e->encap_header;
91 	struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
92 	bool encap_connected;
93 	LIST_HEAD(flow_list);
94 
95 	ASSERT_RTNL();
96 
97 	mutex_lock(&esw->offloads.encap_tbl_lock);
98 	encap_connected = !!(e->flags & MLX5_ENCAP_ENTRY_VALID);
99 	if (encap_connected == neigh_connected && ether_addr_equal(e->h_dest, ha))
100 		goto unlock;
101 
102 	mlx5e_take_all_encap_flows(e, &flow_list);
103 
104 	if ((e->flags & MLX5_ENCAP_ENTRY_VALID) &&
105 	    (!neigh_connected || !ether_addr_equal(e->h_dest, ha)))
106 		mlx5e_tc_encap_flows_del(priv, e, &flow_list);
107 
108 	if (neigh_connected && !(e->flags & MLX5_ENCAP_ENTRY_VALID)) {
109 		struct net_device *route_dev;
110 
111 		ether_addr_copy(e->h_dest, ha);
112 		ether_addr_copy(eth->h_dest, ha);
113 		/* Update the encap source mac, in case that we delete
114 		 * the flows when encap source mac changed.
115 		 */
116 		route_dev = __dev_get_by_index(dev_net(priv->netdev), e->route_dev_ifindex);
117 		if (route_dev)
118 			ether_addr_copy(eth->h_source, route_dev->dev_addr);
119 
120 		mlx5e_tc_encap_flows_add(priv, e, &flow_list);
121 	}
122 unlock:
123 	mutex_unlock(&esw->offloads.encap_tbl_lock);
124 	mlx5e_put_flow_list(priv, &flow_list);
125 }
126 
127 static int
mlx5e_rep_setup_tc_cls_flower(struct mlx5e_priv * priv,struct flow_cls_offload * cls_flower,int flags)128 mlx5e_rep_setup_tc_cls_flower(struct mlx5e_priv *priv,
129 			      struct flow_cls_offload *cls_flower, int flags)
130 {
131 	switch (cls_flower->command) {
132 	case FLOW_CLS_REPLACE:
133 		return mlx5e_configure_flower(priv->netdev, priv, cls_flower,
134 					      flags);
135 	case FLOW_CLS_DESTROY:
136 		return mlx5e_delete_flower(priv->netdev, priv, cls_flower,
137 					   flags);
138 	case FLOW_CLS_STATS:
139 		return mlx5e_stats_flower(priv->netdev, priv, cls_flower,
140 					  flags);
141 	default:
142 		return -EOPNOTSUPP;
143 	}
144 }
145 
146 static
mlx5e_rep_setup_tc_cls_matchall(struct mlx5e_priv * priv,struct tc_cls_matchall_offload * ma)147 int mlx5e_rep_setup_tc_cls_matchall(struct mlx5e_priv *priv,
148 				    struct tc_cls_matchall_offload *ma)
149 {
150 	switch (ma->command) {
151 	case TC_CLSMATCHALL_REPLACE:
152 		return mlx5e_tc_configure_matchall(priv, ma);
153 	case TC_CLSMATCHALL_DESTROY:
154 		return mlx5e_tc_delete_matchall(priv, ma);
155 	case TC_CLSMATCHALL_STATS:
156 		mlx5e_tc_stats_matchall(priv, ma);
157 		return 0;
158 	default:
159 		return -EOPNOTSUPP;
160 	}
161 }
162 
mlx5e_rep_setup_tc_cb(enum tc_setup_type type,void * type_data,void * cb_priv)163 static int mlx5e_rep_setup_tc_cb(enum tc_setup_type type, void *type_data,
164 				 void *cb_priv)
165 {
166 	unsigned long flags = MLX5_TC_FLAG(INGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
167 	struct mlx5e_priv *priv = cb_priv;
168 
169 	if (!priv->netdev || !netif_device_present(priv->netdev))
170 		return -EOPNOTSUPP;
171 
172 	switch (type) {
173 	case TC_SETUP_CLSFLOWER:
174 		return mlx5e_rep_setup_tc_cls_flower(priv, type_data, flags);
175 	case TC_SETUP_CLSMATCHALL:
176 		return mlx5e_rep_setup_tc_cls_matchall(priv, type_data);
177 	default:
178 		return -EOPNOTSUPP;
179 	}
180 }
181 
mlx5e_rep_setup_ft_cb(enum tc_setup_type type,void * type_data,void * cb_priv)182 static int mlx5e_rep_setup_ft_cb(enum tc_setup_type type, void *type_data,
183 				 void *cb_priv)
184 {
185 	struct flow_cls_offload tmp, *f = type_data;
186 	struct mlx5e_priv *priv = cb_priv;
187 	struct mlx5_eswitch *esw;
188 	unsigned long flags;
189 	int err;
190 
191 	flags = MLX5_TC_FLAG(INGRESS) |
192 		MLX5_TC_FLAG(ESW_OFFLOAD) |
193 		MLX5_TC_FLAG(FT_OFFLOAD);
194 	esw = priv->mdev->priv.eswitch;
195 
196 	switch (type) {
197 	case TC_SETUP_CLSFLOWER:
198 		memcpy(&tmp, f, sizeof(*f));
199 
200 		if (!mlx5_chains_prios_supported(esw_chains(esw)))
201 			return -EOPNOTSUPP;
202 
203 		/* Re-use tc offload path by moving the ft flow to the
204 		 * reserved ft chain.
205 		 *
206 		 * FT offload can use prio range [0, INT_MAX], so we normalize
207 		 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
208 		 * as with tc, where prio 0 isn't supported.
209 		 *
210 		 * We only support chain 0 of FT offload.
211 		 */
212 		if (tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)))
213 			return -EOPNOTSUPP;
214 		if (tmp.common.chain_index != 0)
215 			return -EOPNOTSUPP;
216 
217 		tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
218 		tmp.common.prio++;
219 		err = mlx5e_rep_setup_tc_cls_flower(priv, &tmp, flags);
220 		memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
221 		return err;
222 	default:
223 		return -EOPNOTSUPP;
224 	}
225 }
226 
227 static LIST_HEAD(mlx5e_rep_block_tc_cb_list);
228 static LIST_HEAD(mlx5e_rep_block_ft_cb_list);
mlx5e_rep_setup_tc(struct net_device * dev,enum tc_setup_type type,void * type_data)229 int mlx5e_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
230 		       void *type_data)
231 {
232 	struct mlx5e_priv *priv = netdev_priv(dev);
233 	struct flow_block_offload *f = type_data;
234 
235 	f->unlocked_driver_cb = true;
236 
237 	switch (type) {
238 	case TC_SETUP_BLOCK:
239 		return flow_block_cb_setup_simple(type_data,
240 						  &mlx5e_rep_block_tc_cb_list,
241 						  mlx5e_rep_setup_tc_cb,
242 						  priv, priv, true);
243 	case TC_SETUP_FT:
244 		return flow_block_cb_setup_simple(type_data,
245 						  &mlx5e_rep_block_ft_cb_list,
246 						  mlx5e_rep_setup_ft_cb,
247 						  priv, priv, true);
248 	default:
249 		return -EOPNOTSUPP;
250 	}
251 }
252 
mlx5e_rep_tc_init(struct mlx5e_rep_priv * rpriv)253 int mlx5e_rep_tc_init(struct mlx5e_rep_priv *rpriv)
254 {
255 	struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
256 	int err;
257 
258 	mutex_init(&uplink_priv->unready_flows_lock);
259 	INIT_LIST_HEAD(&uplink_priv->unready_flows);
260 
261 	/* init shared tc flow table */
262 	err = mlx5e_tc_esw_init(&uplink_priv->tc_ht);
263 	return err;
264 }
265 
mlx5e_rep_tc_cleanup(struct mlx5e_rep_priv * rpriv)266 void mlx5e_rep_tc_cleanup(struct mlx5e_rep_priv *rpriv)
267 {
268 	/* delete shared tc flow table */
269 	mlx5e_tc_esw_cleanup(&rpriv->uplink_priv.tc_ht);
270 	mutex_destroy(&rpriv->uplink_priv.unready_flows_lock);
271 }
272 
mlx5e_rep_tc_enable(struct mlx5e_priv * priv)273 void mlx5e_rep_tc_enable(struct mlx5e_priv *priv)
274 {
275 	struct mlx5e_rep_priv *rpriv = priv->ppriv;
276 
277 	INIT_WORK(&rpriv->uplink_priv.reoffload_flows_work,
278 		  mlx5e_tc_reoffload_flows_work);
279 }
280 
mlx5e_rep_tc_disable(struct mlx5e_priv * priv)281 void mlx5e_rep_tc_disable(struct mlx5e_priv *priv)
282 {
283 	struct mlx5e_rep_priv *rpriv = priv->ppriv;
284 
285 	cancel_work_sync(&rpriv->uplink_priv.reoffload_flows_work);
286 }
287 
mlx5e_rep_tc_event_port_affinity(struct mlx5e_priv * priv)288 int mlx5e_rep_tc_event_port_affinity(struct mlx5e_priv *priv)
289 {
290 	struct mlx5e_rep_priv *rpriv = priv->ppriv;
291 
292 	queue_work(priv->wq, &rpriv->uplink_priv.reoffload_flows_work);
293 
294 	return NOTIFY_OK;
295 }
296 
297 static struct mlx5e_rep_indr_block_priv *
mlx5e_rep_indr_block_priv_lookup(struct mlx5e_rep_priv * rpriv,struct net_device * netdev)298 mlx5e_rep_indr_block_priv_lookup(struct mlx5e_rep_priv *rpriv,
299 				 struct net_device *netdev)
300 {
301 	struct mlx5e_rep_indr_block_priv *cb_priv;
302 
303 	list_for_each_entry(cb_priv,
304 			    &rpriv->uplink_priv.tc_indr_block_priv_list,
305 			    list)
306 		if (cb_priv->netdev == netdev)
307 			return cb_priv;
308 
309 	return NULL;
310 }
311 
312 static int
mlx5e_rep_indr_offload(struct net_device * netdev,struct flow_cls_offload * flower,struct mlx5e_rep_indr_block_priv * indr_priv,unsigned long flags)313 mlx5e_rep_indr_offload(struct net_device *netdev,
314 		       struct flow_cls_offload *flower,
315 		       struct mlx5e_rep_indr_block_priv *indr_priv,
316 		       unsigned long flags)
317 {
318 	struct mlx5e_priv *priv = netdev_priv(indr_priv->rpriv->netdev);
319 	int err = 0;
320 
321 	if (!netif_device_present(indr_priv->rpriv->netdev))
322 		return -EOPNOTSUPP;
323 
324 	switch (flower->command) {
325 	case FLOW_CLS_REPLACE:
326 		err = mlx5e_configure_flower(netdev, priv, flower, flags);
327 		break;
328 	case FLOW_CLS_DESTROY:
329 		err = mlx5e_delete_flower(netdev, priv, flower, flags);
330 		break;
331 	case FLOW_CLS_STATS:
332 		err = mlx5e_stats_flower(netdev, priv, flower, flags);
333 		break;
334 	default:
335 		err = -EOPNOTSUPP;
336 	}
337 
338 	return err;
339 }
340 
mlx5e_rep_indr_setup_tc_cb(enum tc_setup_type type,void * type_data,void * indr_priv)341 static int mlx5e_rep_indr_setup_tc_cb(enum tc_setup_type type,
342 				      void *type_data, void *indr_priv)
343 {
344 	unsigned long flags = MLX5_TC_FLAG(EGRESS) | MLX5_TC_FLAG(ESW_OFFLOAD);
345 	struct mlx5e_rep_indr_block_priv *priv = indr_priv;
346 
347 	switch (type) {
348 	case TC_SETUP_CLSFLOWER:
349 		return mlx5e_rep_indr_offload(priv->netdev, type_data, priv,
350 					      flags);
351 	default:
352 		return -EOPNOTSUPP;
353 	}
354 }
355 
mlx5e_rep_indr_setup_ft_cb(enum tc_setup_type type,void * type_data,void * indr_priv)356 static int mlx5e_rep_indr_setup_ft_cb(enum tc_setup_type type,
357 				      void *type_data, void *indr_priv)
358 {
359 	struct mlx5e_rep_indr_block_priv *priv = indr_priv;
360 	struct flow_cls_offload *f = type_data;
361 	struct flow_cls_offload tmp;
362 	struct mlx5e_priv *mpriv;
363 	struct mlx5_eswitch *esw;
364 	unsigned long flags;
365 	int err;
366 
367 	mpriv = netdev_priv(priv->rpriv->netdev);
368 	esw = mpriv->mdev->priv.eswitch;
369 
370 	flags = MLX5_TC_FLAG(EGRESS) |
371 		MLX5_TC_FLAG(ESW_OFFLOAD) |
372 		MLX5_TC_FLAG(FT_OFFLOAD);
373 
374 	switch (type) {
375 	case TC_SETUP_CLSFLOWER:
376 		memcpy(&tmp, f, sizeof(*f));
377 
378 		/* Re-use tc offload path by moving the ft flow to the
379 		 * reserved ft chain.
380 		 *
381 		 * FT offload can use prio range [0, INT_MAX], so we normalize
382 		 * it to range [1, mlx5_esw_chains_get_prio_range(esw)]
383 		 * as with tc, where prio 0 isn't supported.
384 		 *
385 		 * We only support chain 0 of FT offload.
386 		 */
387 		if (!mlx5_chains_prios_supported(esw_chains(esw)) ||
388 		    tmp.common.prio >= mlx5_chains_get_prio_range(esw_chains(esw)) ||
389 		    tmp.common.chain_index)
390 			return -EOPNOTSUPP;
391 
392 		tmp.common.chain_index = mlx5_chains_get_nf_ft_chain(esw_chains(esw));
393 		tmp.common.prio++;
394 		err = mlx5e_rep_indr_offload(priv->netdev, &tmp, priv, flags);
395 		memcpy(&f->stats, &tmp.stats, sizeof(f->stats));
396 		return err;
397 	default:
398 		return -EOPNOTSUPP;
399 	}
400 }
401 
mlx5e_rep_indr_block_unbind(void * cb_priv)402 static void mlx5e_rep_indr_block_unbind(void *cb_priv)
403 {
404 	struct mlx5e_rep_indr_block_priv *indr_priv = cb_priv;
405 
406 	list_del(&indr_priv->list);
407 	kfree(indr_priv);
408 }
409 
410 static LIST_HEAD(mlx5e_block_cb_list);
411 
412 static int
mlx5e_rep_indr_setup_block(struct net_device * netdev,struct Qdisc * sch,struct mlx5e_rep_priv * rpriv,struct flow_block_offload * f,flow_setup_cb_t * setup_cb,void * data,void (* cleanup)(struct flow_block_cb * block_cb))413 mlx5e_rep_indr_setup_block(struct net_device *netdev, struct Qdisc *sch,
414 			   struct mlx5e_rep_priv *rpriv,
415 			   struct flow_block_offload *f,
416 			   flow_setup_cb_t *setup_cb,
417 			   void *data,
418 			   void (*cleanup)(struct flow_block_cb *block_cb))
419 {
420 	struct mlx5e_priv *priv = netdev_priv(rpriv->netdev);
421 	struct mlx5e_rep_indr_block_priv *indr_priv;
422 	struct flow_block_cb *block_cb;
423 
424 	if (!mlx5e_tc_tun_device_to_offload(priv, netdev) &&
425 	    !(is_vlan_dev(netdev) && vlan_dev_real_dev(netdev) == rpriv->netdev))
426 		return -EOPNOTSUPP;
427 
428 	if (f->binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
429 		return -EOPNOTSUPP;
430 
431 	f->unlocked_driver_cb = true;
432 	f->driver_block_list = &mlx5e_block_cb_list;
433 
434 	switch (f->command) {
435 	case FLOW_BLOCK_BIND:
436 		indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
437 		if (indr_priv)
438 			return -EEXIST;
439 
440 		indr_priv = kmalloc(sizeof(*indr_priv), GFP_KERNEL);
441 		if (!indr_priv)
442 			return -ENOMEM;
443 
444 		indr_priv->netdev = netdev;
445 		indr_priv->rpriv = rpriv;
446 		list_add(&indr_priv->list,
447 			 &rpriv->uplink_priv.tc_indr_block_priv_list);
448 
449 		block_cb = flow_indr_block_cb_alloc(setup_cb, indr_priv, indr_priv,
450 						    mlx5e_rep_indr_block_unbind,
451 						    f, netdev, sch, data, rpriv,
452 						    cleanup);
453 		if (IS_ERR(block_cb)) {
454 			list_del(&indr_priv->list);
455 			kfree(indr_priv);
456 			return PTR_ERR(block_cb);
457 		}
458 		flow_block_cb_add(block_cb, f);
459 		list_add_tail(&block_cb->driver_list, &mlx5e_block_cb_list);
460 
461 		return 0;
462 	case FLOW_BLOCK_UNBIND:
463 		indr_priv = mlx5e_rep_indr_block_priv_lookup(rpriv, netdev);
464 		if (!indr_priv)
465 			return -ENOENT;
466 
467 		block_cb = flow_block_cb_lookup(f->block, setup_cb, indr_priv);
468 		if (!block_cb)
469 			return -ENOENT;
470 
471 		flow_indr_block_cb_remove(block_cb, f);
472 		list_del(&block_cb->driver_list);
473 		return 0;
474 	default:
475 		return -EOPNOTSUPP;
476 	}
477 	return 0;
478 }
479 
480 static
mlx5e_rep_indr_setup_cb(struct net_device * netdev,struct Qdisc * sch,void * cb_priv,enum tc_setup_type type,void * type_data,void * data,void (* cleanup)(struct flow_block_cb * block_cb))481 int mlx5e_rep_indr_setup_cb(struct net_device *netdev, struct Qdisc *sch, void *cb_priv,
482 			    enum tc_setup_type type, void *type_data,
483 			    void *data,
484 			    void (*cleanup)(struct flow_block_cb *block_cb))
485 {
486 	switch (type) {
487 	case TC_SETUP_BLOCK:
488 		return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
489 						  mlx5e_rep_indr_setup_tc_cb,
490 						  data, cleanup);
491 	case TC_SETUP_FT:
492 		return mlx5e_rep_indr_setup_block(netdev, sch, cb_priv, type_data,
493 						  mlx5e_rep_indr_setup_ft_cb,
494 						  data, cleanup);
495 	default:
496 		return -EOPNOTSUPP;
497 	}
498 }
499 
mlx5e_rep_tc_netdevice_event_register(struct mlx5e_rep_priv * rpriv)500 int mlx5e_rep_tc_netdevice_event_register(struct mlx5e_rep_priv *rpriv)
501 {
502 	struct mlx5_rep_uplink_priv *uplink_priv = &rpriv->uplink_priv;
503 
504 	/* init indirect block notifications */
505 	INIT_LIST_HEAD(&uplink_priv->tc_indr_block_priv_list);
506 
507 	return flow_indr_dev_register(mlx5e_rep_indr_setup_cb, rpriv);
508 }
509 
mlx5e_rep_tc_netdevice_event_unregister(struct mlx5e_rep_priv * rpriv)510 void mlx5e_rep_tc_netdevice_event_unregister(struct mlx5e_rep_priv *rpriv)
511 {
512 	flow_indr_dev_unregister(mlx5e_rep_indr_setup_cb, rpriv,
513 				 mlx5e_rep_indr_block_unbind);
514 }
515 
mlx5e_restore_tunnel(struct mlx5e_priv * priv,struct sk_buff * skb,struct mlx5e_tc_update_priv * tc_priv,u32 tunnel_id)516 static bool mlx5e_restore_tunnel(struct mlx5e_priv *priv, struct sk_buff *skb,
517 				 struct mlx5e_tc_update_priv *tc_priv,
518 				 u32 tunnel_id)
519 {
520 	struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
521 	struct tunnel_match_enc_opts enc_opts = {};
522 	struct mlx5_rep_uplink_priv *uplink_priv;
523 	struct mlx5e_rep_priv *uplink_rpriv;
524 	struct metadata_dst *tun_dst;
525 	struct tunnel_match_key key;
526 	u32 tun_id, enc_opts_id;
527 	struct net_device *dev;
528 	int err;
529 
530 	enc_opts_id = tunnel_id & ENC_OPTS_BITS_MASK;
531 	tun_id = tunnel_id >> ENC_OPTS_BITS;
532 
533 	if (!tun_id)
534 		return true;
535 
536 	uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
537 	uplink_priv = &uplink_rpriv->uplink_priv;
538 
539 	err = mapping_find(uplink_priv->tunnel_mapping, tun_id, &key);
540 	if (err) {
541 		WARN_ON_ONCE(true);
542 		netdev_dbg(priv->netdev,
543 			   "Couldn't find tunnel for tun_id: %d, err: %d\n",
544 			   tun_id, err);
545 		return false;
546 	}
547 
548 	if (enc_opts_id) {
549 		err = mapping_find(uplink_priv->tunnel_enc_opts_mapping,
550 				   enc_opts_id, &enc_opts);
551 		if (err) {
552 			netdev_dbg(priv->netdev,
553 				   "Couldn't find tunnel (opts) for tun_id: %d, err: %d\n",
554 				   enc_opts_id, err);
555 			return false;
556 		}
557 	}
558 
559 	if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
560 		tun_dst = __ip_tun_set_dst(key.enc_ipv4.src, key.enc_ipv4.dst,
561 					   key.enc_ip.tos, key.enc_ip.ttl,
562 					   key.enc_tp.dst, TUNNEL_KEY,
563 					   key32_to_tunnel_id(key.enc_key_id.keyid),
564 					   enc_opts.key.len);
565 	} else if (key.enc_control.addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
566 		tun_dst = __ipv6_tun_set_dst(&key.enc_ipv6.src, &key.enc_ipv6.dst,
567 					     key.enc_ip.tos, key.enc_ip.ttl,
568 					     key.enc_tp.dst, 0, TUNNEL_KEY,
569 					     key32_to_tunnel_id(key.enc_key_id.keyid),
570 					     enc_opts.key.len);
571 	} else {
572 		netdev_dbg(priv->netdev,
573 			   "Couldn't restore tunnel, unsupported addr_type: %d\n",
574 			   key.enc_control.addr_type);
575 		return false;
576 	}
577 
578 	if (!tun_dst) {
579 		netdev_dbg(priv->netdev, "Couldn't restore tunnel, no tun_dst\n");
580 		return false;
581 	}
582 
583 	tun_dst->u.tun_info.key.tp_src = key.enc_tp.src;
584 
585 	if (enc_opts.key.len)
586 		ip_tunnel_info_opts_set(&tun_dst->u.tun_info,
587 					enc_opts.key.data,
588 					enc_opts.key.len,
589 					enc_opts.key.dst_opt_type);
590 
591 	skb_dst_set(skb, (struct dst_entry *)tun_dst);
592 	dev = dev_get_by_index(&init_net, key.filter_ifindex);
593 	if (!dev) {
594 		netdev_dbg(priv->netdev,
595 			   "Couldn't find tunnel device with ifindex: %d\n",
596 			   key.filter_ifindex);
597 		return false;
598 	}
599 
600 	/* Set tun_dev so we do dev_put() after datapath */
601 	tc_priv->tun_dev = dev;
602 
603 	skb->dev = dev;
604 
605 	return true;
606 }
607 
mlx5e_restore_skb_chain(struct sk_buff * skb,u32 chain,u32 reg_c1,struct mlx5e_tc_update_priv * tc_priv)608 static bool mlx5e_restore_skb_chain(struct sk_buff *skb, u32 chain, u32 reg_c1,
609 				    struct mlx5e_tc_update_priv *tc_priv)
610 {
611 	struct mlx5e_priv *priv = netdev_priv(skb->dev);
612 	u32 tunnel_id = (reg_c1 >> ESW_TUN_OFFSET) & TUNNEL_ID_MASK;
613 
614 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
615 	if (chain) {
616 		struct mlx5_rep_uplink_priv *uplink_priv;
617 		struct mlx5e_rep_priv *uplink_rpriv;
618 		struct tc_skb_ext *tc_skb_ext;
619 		struct mlx5_eswitch *esw;
620 		u32 zone_restore_id;
621 
622 		tc_skb_ext = tc_skb_ext_alloc(skb);
623 		if (!tc_skb_ext) {
624 			WARN_ON(1);
625 			return false;
626 		}
627 		tc_skb_ext->chain = chain;
628 		zone_restore_id = reg_c1 & ESW_ZONE_ID_MASK;
629 		esw = priv->mdev->priv.eswitch;
630 		uplink_rpriv = mlx5_eswitch_get_uplink_priv(esw, REP_ETH);
631 		uplink_priv = &uplink_rpriv->uplink_priv;
632 		if (!mlx5e_tc_ct_restore_flow(uplink_priv->ct_priv, skb,
633 					      zone_restore_id))
634 			return false;
635 	}
636 #endif /* CONFIG_NET_TC_SKB_EXT */
637 
638 	return mlx5e_restore_tunnel(priv, skb, tc_priv, tunnel_id);
639 }
640 
mlx5e_restore_skb_sample(struct mlx5e_priv * priv,struct sk_buff * skb,struct mlx5_mapped_obj * mapped_obj,struct mlx5e_tc_update_priv * tc_priv)641 static void mlx5e_restore_skb_sample(struct mlx5e_priv *priv, struct sk_buff *skb,
642 				     struct mlx5_mapped_obj *mapped_obj,
643 				     struct mlx5e_tc_update_priv *tc_priv)
644 {
645 	if (!mlx5e_restore_tunnel(priv, skb, tc_priv, mapped_obj->sample.tunnel_id)) {
646 		netdev_dbg(priv->netdev,
647 			   "Failed to restore tunnel info for sampled packet\n");
648 		return;
649 	}
650 	mlx5e_tc_sample_skb(skb, mapped_obj);
651 	mlx5_rep_tc_post_napi_receive(tc_priv);
652 }
653 
mlx5e_rep_tc_update_skb(struct mlx5_cqe64 * cqe,struct sk_buff * skb,struct mlx5e_tc_update_priv * tc_priv)654 bool mlx5e_rep_tc_update_skb(struct mlx5_cqe64 *cqe,
655 			     struct sk_buff *skb,
656 			     struct mlx5e_tc_update_priv *tc_priv)
657 {
658 	struct mlx5_mapped_obj mapped_obj;
659 	struct mlx5_eswitch *esw;
660 	struct mlx5e_priv *priv;
661 	u32 reg_c0;
662 	int err;
663 
664 	reg_c0 = (be32_to_cpu(cqe->sop_drop_qpn) & MLX5E_TC_FLOW_ID_MASK);
665 	if (!reg_c0 || reg_c0 == MLX5_FS_DEFAULT_FLOW_TAG)
666 		return true;
667 
668 	/* If reg_c0 is not equal to the default flow tag then skb->mark
669 	 * is not supported and must be reset back to 0.
670 	 */
671 	skb->mark = 0;
672 
673 	priv = netdev_priv(skb->dev);
674 	esw = priv->mdev->priv.eswitch;
675 	err = mapping_find(esw->offloads.reg_c0_obj_pool, reg_c0, &mapped_obj);
676 	if (err) {
677 		netdev_dbg(priv->netdev,
678 			   "Couldn't find mapped object for reg_c0: %d, err: %d\n",
679 			   reg_c0, err);
680 		return false;
681 	}
682 
683 	if (mapped_obj.type == MLX5_MAPPED_OBJ_CHAIN) {
684 		u32 reg_c1 = be32_to_cpu(cqe->ft_metadata);
685 
686 		return mlx5e_restore_skb_chain(skb, mapped_obj.chain, reg_c1, tc_priv);
687 	} else if (mapped_obj.type == MLX5_MAPPED_OBJ_SAMPLE) {
688 		mlx5e_restore_skb_sample(priv, skb, &mapped_obj, tc_priv);
689 		return false;
690 	} else {
691 		netdev_dbg(priv->netdev, "Invalid mapped object type: %d\n", mapped_obj.type);
692 		return false;
693 	}
694 
695 	return true;
696 }
697 
mlx5_rep_tc_post_napi_receive(struct mlx5e_tc_update_priv * tc_priv)698 void mlx5_rep_tc_post_napi_receive(struct mlx5e_tc_update_priv *tc_priv)
699 {
700 	if (tc_priv->tun_dev)
701 		dev_put(tc_priv->tun_dev);
702 }
703