1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2017-2018 Netronome Systems, Inc. */
3 
4 #include <linux/etherdevice.h>
5 #include <linux/inetdevice.h>
6 #include <net/netevent.h>
7 #include <linux/idr.h>
8 #include <net/dst_metadata.h>
9 #include <net/arp.h>
10 
11 #include "cmsg.h"
12 #include "main.h"
13 #include "../nfp_net_repr.h"
14 #include "../nfp_net.h"
15 
16 #define NFP_FL_MAX_ROUTES               32
17 
18 #define NFP_TUN_PRE_TUN_RULE_LIMIT	32
19 #define NFP_TUN_PRE_TUN_RULE_DEL	0x1
20 #define NFP_TUN_PRE_TUN_IDX_BIT		0x8
21 
22 /**
23  * struct nfp_tun_pre_run_rule - rule matched before decap
24  * @flags:		options for the rule offset
25  * @port_idx:		index of destination MAC address for the rule
26  * @vlan_tci:		VLAN info associated with MAC
27  * @host_ctx_id:	stats context of rule to update
28  */
29 struct nfp_tun_pre_tun_rule {
30 	__be32 flags;
31 	__be16 port_idx;
32 	__be16 vlan_tci;
33 	__be32 host_ctx_id;
34 };
35 
36 /**
37  * struct nfp_tun_active_tuns - periodic message of active tunnels
38  * @seq:		sequence number of the message
39  * @count:		number of tunnels report in message
40  * @flags:		options part of the request
41  * @tun_info.ipv4:		dest IPv4 address of active route
42  * @tun_info.egress_port:	port the encapsulated packet egressed
43  * @tun_info.extra:		reserved for future use
44  * @tun_info:		tunnels that have sent traffic in reported period
45  */
46 struct nfp_tun_active_tuns {
47 	__be32 seq;
48 	__be32 count;
49 	__be32 flags;
50 	struct route_ip_info {
51 		__be32 ipv4;
52 		__be32 egress_port;
53 		__be32 extra[2];
54 	} tun_info[];
55 };
56 
57 /**
58  * struct nfp_tun_neigh - neighbour/route entry on the NFP
59  * @dst_ipv4:	destination IPv4 address
60  * @src_ipv4:	source IPv4 address
61  * @dst_addr:	destination MAC address
62  * @src_addr:	source MAC address
63  * @port_id:	NFP port to output packet on - associated with source IPv4
64  */
65 struct nfp_tun_neigh {
66 	__be32 dst_ipv4;
67 	__be32 src_ipv4;
68 	u8 dst_addr[ETH_ALEN];
69 	u8 src_addr[ETH_ALEN];
70 	__be32 port_id;
71 };
72 
73 /**
74  * struct nfp_tun_req_route_ipv4 - NFP requests a route/neighbour lookup
75  * @ingress_port:	ingress port of packet that signalled request
76  * @ipv4_addr:		destination ipv4 address for route
77  * @reserved:		reserved for future use
78  */
79 struct nfp_tun_req_route_ipv4 {
80 	__be32 ingress_port;
81 	__be32 ipv4_addr;
82 	__be32 reserved[2];
83 };
84 
85 /**
86  * struct nfp_ipv4_route_entry - routes that are offloaded to the NFP
87  * @ipv4_addr:	destination of route
88  * @list:	list pointer
89  */
90 struct nfp_ipv4_route_entry {
91 	__be32 ipv4_addr;
92 	struct list_head list;
93 };
94 
95 #define NFP_FL_IPV4_ADDRS_MAX        32
96 
97 /**
98  * struct nfp_tun_ipv4_addr - set the IP address list on the NFP
99  * @count:	number of IPs populated in the array
100  * @ipv4_addr:	array of IPV4_ADDRS_MAX 32 bit IPv4 addresses
101  */
102 struct nfp_tun_ipv4_addr {
103 	__be32 count;
104 	__be32 ipv4_addr[NFP_FL_IPV4_ADDRS_MAX];
105 };
106 
107 /**
108  * struct nfp_ipv4_addr_entry - cached IPv4 addresses
109  * @ipv4_addr:	IP address
110  * @ref_count:	number of rules currently using this IP
111  * @list:	list pointer
112  */
113 struct nfp_ipv4_addr_entry {
114 	__be32 ipv4_addr;
115 	int ref_count;
116 	struct list_head list;
117 };
118 
119 #define NFP_TUN_MAC_OFFLOAD_DEL_FLAG	0x2
120 
121 /**
122  * struct nfp_tun_mac_addr_offload - configure MAC address of tunnel EP on NFP
123  * @flags:	MAC address offload options
124  * @count:	number of MAC addresses in the message (should be 1)
125  * @index:	index of MAC address in the lookup table
126  * @addr:	interface MAC address
127  */
128 struct nfp_tun_mac_addr_offload {
129 	__be16 flags;
130 	__be16 count;
131 	__be16 index;
132 	u8 addr[ETH_ALEN];
133 };
134 
135 enum nfp_flower_mac_offload_cmd {
136 	NFP_TUNNEL_MAC_OFFLOAD_ADD =		0,
137 	NFP_TUNNEL_MAC_OFFLOAD_DEL =		1,
138 	NFP_TUNNEL_MAC_OFFLOAD_MOD =		2,
139 };
140 
141 #define NFP_MAX_MAC_INDEX       0xff
142 
143 /**
144  * struct nfp_tun_offloaded_mac - hashtable entry for an offloaded MAC
145  * @ht_node:		Hashtable entry
146  * @addr:		Offloaded MAC address
147  * @index:		Offloaded index for given MAC address
148  * @ref_count:		Number of devs using this MAC address
149  * @repr_list:		List of reprs sharing this MAC address
150  * @bridge_count:	Number of bridge/internal devs with MAC
151  */
152 struct nfp_tun_offloaded_mac {
153 	struct rhash_head ht_node;
154 	u8 addr[ETH_ALEN];
155 	u16 index;
156 	int ref_count;
157 	struct list_head repr_list;
158 	int bridge_count;
159 };
160 
161 static const struct rhashtable_params offloaded_macs_params = {
162 	.key_offset	= offsetof(struct nfp_tun_offloaded_mac, addr),
163 	.head_offset	= offsetof(struct nfp_tun_offloaded_mac, ht_node),
164 	.key_len	= ETH_ALEN,
165 	.automatic_shrinking	= true,
166 };
167 
168 void nfp_tunnel_keep_alive(struct nfp_app *app, struct sk_buff *skb)
169 {
170 	struct nfp_tun_active_tuns *payload;
171 	struct net_device *netdev;
172 	int count, i, pay_len;
173 	struct neighbour *n;
174 	__be32 ipv4_addr;
175 	u32 port;
176 
177 	payload = nfp_flower_cmsg_get_data(skb);
178 	count = be32_to_cpu(payload->count);
179 	if (count > NFP_FL_MAX_ROUTES) {
180 		nfp_flower_cmsg_warn(app, "Tunnel keep-alive request exceeds max routes.\n");
181 		return;
182 	}
183 
184 	pay_len = nfp_flower_cmsg_get_data_len(skb);
185 	if (pay_len != struct_size(payload, tun_info, count)) {
186 		nfp_flower_cmsg_warn(app, "Corruption in tunnel keep-alive message.\n");
187 		return;
188 	}
189 
190 	rcu_read_lock();
191 	for (i = 0; i < count; i++) {
192 		ipv4_addr = payload->tun_info[i].ipv4;
193 		port = be32_to_cpu(payload->tun_info[i].egress_port);
194 		netdev = nfp_app_dev_get(app, port, NULL);
195 		if (!netdev)
196 			continue;
197 
198 		n = neigh_lookup(&arp_tbl, &ipv4_addr, netdev);
199 		if (!n)
200 			continue;
201 
202 		/* Update the used timestamp of neighbour */
203 		neigh_event_send(n, NULL);
204 		neigh_release(n);
205 	}
206 	rcu_read_unlock();
207 }
208 
209 static int
210 nfp_flower_xmit_tun_conf(struct nfp_app *app, u8 mtype, u16 plen, void *pdata,
211 			 gfp_t flag)
212 {
213 	struct sk_buff *skb;
214 	unsigned char *msg;
215 
216 	skb = nfp_flower_cmsg_alloc(app, plen, mtype, flag);
217 	if (!skb)
218 		return -ENOMEM;
219 
220 	msg = nfp_flower_cmsg_get_data(skb);
221 	memcpy(msg, pdata, nfp_flower_cmsg_get_data_len(skb));
222 
223 	nfp_ctrl_tx(app->ctrl, skb);
224 	return 0;
225 }
226 
227 static bool nfp_tun_has_route(struct nfp_app *app, __be32 ipv4_addr)
228 {
229 	struct nfp_flower_priv *priv = app->priv;
230 	struct nfp_ipv4_route_entry *entry;
231 	struct list_head *ptr, *storage;
232 
233 	spin_lock_bh(&priv->tun.neigh_off_lock);
234 	list_for_each_safe(ptr, storage, &priv->tun.neigh_off_list) {
235 		entry = list_entry(ptr, struct nfp_ipv4_route_entry, list);
236 		if (entry->ipv4_addr == ipv4_addr) {
237 			spin_unlock_bh(&priv->tun.neigh_off_lock);
238 			return true;
239 		}
240 	}
241 	spin_unlock_bh(&priv->tun.neigh_off_lock);
242 	return false;
243 }
244 
245 static void nfp_tun_add_route_to_cache(struct nfp_app *app, __be32 ipv4_addr)
246 {
247 	struct nfp_flower_priv *priv = app->priv;
248 	struct nfp_ipv4_route_entry *entry;
249 	struct list_head *ptr, *storage;
250 
251 	spin_lock_bh(&priv->tun.neigh_off_lock);
252 	list_for_each_safe(ptr, storage, &priv->tun.neigh_off_list) {
253 		entry = list_entry(ptr, struct nfp_ipv4_route_entry, list);
254 		if (entry->ipv4_addr == ipv4_addr) {
255 			spin_unlock_bh(&priv->tun.neigh_off_lock);
256 			return;
257 		}
258 	}
259 	entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
260 	if (!entry) {
261 		spin_unlock_bh(&priv->tun.neigh_off_lock);
262 		nfp_flower_cmsg_warn(app, "Mem error when storing new route.\n");
263 		return;
264 	}
265 
266 	entry->ipv4_addr = ipv4_addr;
267 	list_add_tail(&entry->list, &priv->tun.neigh_off_list);
268 	spin_unlock_bh(&priv->tun.neigh_off_lock);
269 }
270 
271 static void nfp_tun_del_route_from_cache(struct nfp_app *app, __be32 ipv4_addr)
272 {
273 	struct nfp_flower_priv *priv = app->priv;
274 	struct nfp_ipv4_route_entry *entry;
275 	struct list_head *ptr, *storage;
276 
277 	spin_lock_bh(&priv->tun.neigh_off_lock);
278 	list_for_each_safe(ptr, storage, &priv->tun.neigh_off_list) {
279 		entry = list_entry(ptr, struct nfp_ipv4_route_entry, list);
280 		if (entry->ipv4_addr == ipv4_addr) {
281 			list_del(&entry->list);
282 			kfree(entry);
283 			break;
284 		}
285 	}
286 	spin_unlock_bh(&priv->tun.neigh_off_lock);
287 }
288 
289 static void
290 nfp_tun_write_neigh(struct net_device *netdev, struct nfp_app *app,
291 		    struct flowi4 *flow, struct neighbour *neigh, gfp_t flag)
292 {
293 	struct nfp_tun_neigh payload;
294 	u32 port_id;
295 
296 	port_id = nfp_flower_get_port_id_from_netdev(app, netdev);
297 	if (!port_id)
298 		return;
299 
300 	memset(&payload, 0, sizeof(struct nfp_tun_neigh));
301 	payload.dst_ipv4 = flow->daddr;
302 
303 	/* If entry has expired send dst IP with all other fields 0. */
304 	if (!(neigh->nud_state & NUD_VALID) || neigh->dead) {
305 		nfp_tun_del_route_from_cache(app, payload.dst_ipv4);
306 		/* Trigger ARP to verify invalid neighbour state. */
307 		neigh_event_send(neigh, NULL);
308 		goto send_msg;
309 	}
310 
311 	/* Have a valid neighbour so populate rest of entry. */
312 	payload.src_ipv4 = flow->saddr;
313 	ether_addr_copy(payload.src_addr, netdev->dev_addr);
314 	neigh_ha_snapshot(payload.dst_addr, neigh, netdev);
315 	payload.port_id = cpu_to_be32(port_id);
316 	/* Add destination of new route to NFP cache. */
317 	nfp_tun_add_route_to_cache(app, payload.dst_ipv4);
318 
319 send_msg:
320 	nfp_flower_xmit_tun_conf(app, NFP_FLOWER_CMSG_TYPE_TUN_NEIGH,
321 				 sizeof(struct nfp_tun_neigh),
322 				 (unsigned char *)&payload, flag);
323 }
324 
325 static int
326 nfp_tun_neigh_event_handler(struct notifier_block *nb, unsigned long event,
327 			    void *ptr)
328 {
329 	struct nfp_flower_priv *app_priv;
330 	struct netevent_redirect *redir;
331 	struct flowi4 flow = {};
332 	struct neighbour *n;
333 	struct nfp_app *app;
334 	struct rtable *rt;
335 	int err;
336 
337 	switch (event) {
338 	case NETEVENT_REDIRECT:
339 		redir = (struct netevent_redirect *)ptr;
340 		n = redir->neigh;
341 		break;
342 	case NETEVENT_NEIGH_UPDATE:
343 		n = (struct neighbour *)ptr;
344 		break;
345 	default:
346 		return NOTIFY_DONE;
347 	}
348 
349 	flow.daddr = *(__be32 *)n->primary_key;
350 
351 	app_priv = container_of(nb, struct nfp_flower_priv, tun.neigh_nb);
352 	app = app_priv->app;
353 
354 	if (!nfp_netdev_is_nfp_repr(n->dev) &&
355 	    !nfp_flower_internal_port_can_offload(app, n->dev))
356 		return NOTIFY_DONE;
357 
358 	/* Only concerned with changes to routes already added to NFP. */
359 	if (!nfp_tun_has_route(app, flow.daddr))
360 		return NOTIFY_DONE;
361 
362 #if IS_ENABLED(CONFIG_INET)
363 	/* Do a route lookup to populate flow data. */
364 	rt = ip_route_output_key(dev_net(n->dev), &flow);
365 	err = PTR_ERR_OR_ZERO(rt);
366 	if (err)
367 		return NOTIFY_DONE;
368 
369 	ip_rt_put(rt);
370 #else
371 	return NOTIFY_DONE;
372 #endif
373 
374 	flow.flowi4_proto = IPPROTO_UDP;
375 	nfp_tun_write_neigh(n->dev, app, &flow, n, GFP_ATOMIC);
376 
377 	return NOTIFY_OK;
378 }
379 
380 void nfp_tunnel_request_route(struct nfp_app *app, struct sk_buff *skb)
381 {
382 	struct nfp_tun_req_route_ipv4 *payload;
383 	struct net_device *netdev;
384 	struct flowi4 flow = {};
385 	struct neighbour *n;
386 	struct rtable *rt;
387 	int err;
388 
389 	payload = nfp_flower_cmsg_get_data(skb);
390 
391 	rcu_read_lock();
392 	netdev = nfp_app_dev_get(app, be32_to_cpu(payload->ingress_port), NULL);
393 	if (!netdev)
394 		goto fail_rcu_unlock;
395 
396 	flow.daddr = payload->ipv4_addr;
397 	flow.flowi4_proto = IPPROTO_UDP;
398 
399 #if IS_ENABLED(CONFIG_INET)
400 	/* Do a route lookup on same namespace as ingress port. */
401 	rt = ip_route_output_key(dev_net(netdev), &flow);
402 	err = PTR_ERR_OR_ZERO(rt);
403 	if (err)
404 		goto fail_rcu_unlock;
405 #else
406 	goto fail_rcu_unlock;
407 #endif
408 
409 	/* Get the neighbour entry for the lookup */
410 	n = dst_neigh_lookup(&rt->dst, &flow.daddr);
411 	ip_rt_put(rt);
412 	if (!n)
413 		goto fail_rcu_unlock;
414 	nfp_tun_write_neigh(n->dev, app, &flow, n, GFP_ATOMIC);
415 	neigh_release(n);
416 	rcu_read_unlock();
417 	return;
418 
419 fail_rcu_unlock:
420 	rcu_read_unlock();
421 	nfp_flower_cmsg_warn(app, "Requested route not found.\n");
422 }
423 
424 static void nfp_tun_write_ipv4_list(struct nfp_app *app)
425 {
426 	struct nfp_flower_priv *priv = app->priv;
427 	struct nfp_ipv4_addr_entry *entry;
428 	struct nfp_tun_ipv4_addr payload;
429 	struct list_head *ptr, *storage;
430 	int count;
431 
432 	memset(&payload, 0, sizeof(struct nfp_tun_ipv4_addr));
433 	mutex_lock(&priv->tun.ipv4_off_lock);
434 	count = 0;
435 	list_for_each_safe(ptr, storage, &priv->tun.ipv4_off_list) {
436 		if (count >= NFP_FL_IPV4_ADDRS_MAX) {
437 			mutex_unlock(&priv->tun.ipv4_off_lock);
438 			nfp_flower_cmsg_warn(app, "IPv4 offload exceeds limit.\n");
439 			return;
440 		}
441 		entry = list_entry(ptr, struct nfp_ipv4_addr_entry, list);
442 		payload.ipv4_addr[count++] = entry->ipv4_addr;
443 	}
444 	payload.count = cpu_to_be32(count);
445 	mutex_unlock(&priv->tun.ipv4_off_lock);
446 
447 	nfp_flower_xmit_tun_conf(app, NFP_FLOWER_CMSG_TYPE_TUN_IPS,
448 				 sizeof(struct nfp_tun_ipv4_addr),
449 				 &payload, GFP_KERNEL);
450 }
451 
452 void nfp_tunnel_add_ipv4_off(struct nfp_app *app, __be32 ipv4)
453 {
454 	struct nfp_flower_priv *priv = app->priv;
455 	struct nfp_ipv4_addr_entry *entry;
456 	struct list_head *ptr, *storage;
457 
458 	mutex_lock(&priv->tun.ipv4_off_lock);
459 	list_for_each_safe(ptr, storage, &priv->tun.ipv4_off_list) {
460 		entry = list_entry(ptr, struct nfp_ipv4_addr_entry, list);
461 		if (entry->ipv4_addr == ipv4) {
462 			entry->ref_count++;
463 			mutex_unlock(&priv->tun.ipv4_off_lock);
464 			return;
465 		}
466 	}
467 
468 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
469 	if (!entry) {
470 		mutex_unlock(&priv->tun.ipv4_off_lock);
471 		nfp_flower_cmsg_warn(app, "Mem error when offloading IP address.\n");
472 		return;
473 	}
474 	entry->ipv4_addr = ipv4;
475 	entry->ref_count = 1;
476 	list_add_tail(&entry->list, &priv->tun.ipv4_off_list);
477 	mutex_unlock(&priv->tun.ipv4_off_lock);
478 
479 	nfp_tun_write_ipv4_list(app);
480 }
481 
482 void nfp_tunnel_del_ipv4_off(struct nfp_app *app, __be32 ipv4)
483 {
484 	struct nfp_flower_priv *priv = app->priv;
485 	struct nfp_ipv4_addr_entry *entry;
486 	struct list_head *ptr, *storage;
487 
488 	mutex_lock(&priv->tun.ipv4_off_lock);
489 	list_for_each_safe(ptr, storage, &priv->tun.ipv4_off_list) {
490 		entry = list_entry(ptr, struct nfp_ipv4_addr_entry, list);
491 		if (entry->ipv4_addr == ipv4) {
492 			entry->ref_count--;
493 			if (!entry->ref_count) {
494 				list_del(&entry->list);
495 				kfree(entry);
496 			}
497 			break;
498 		}
499 	}
500 	mutex_unlock(&priv->tun.ipv4_off_lock);
501 
502 	nfp_tun_write_ipv4_list(app);
503 }
504 
505 static int
506 __nfp_tunnel_offload_mac(struct nfp_app *app, u8 *mac, u16 idx, bool del)
507 {
508 	struct nfp_tun_mac_addr_offload payload;
509 
510 	memset(&payload, 0, sizeof(payload));
511 
512 	if (del)
513 		payload.flags = cpu_to_be16(NFP_TUN_MAC_OFFLOAD_DEL_FLAG);
514 
515 	/* FW supports multiple MACs per cmsg but restrict to single. */
516 	payload.count = cpu_to_be16(1);
517 	payload.index = cpu_to_be16(idx);
518 	ether_addr_copy(payload.addr, mac);
519 
520 	return nfp_flower_xmit_tun_conf(app, NFP_FLOWER_CMSG_TYPE_TUN_MAC,
521 					sizeof(struct nfp_tun_mac_addr_offload),
522 					&payload, GFP_KERNEL);
523 }
524 
525 static bool nfp_tunnel_port_is_phy_repr(int port)
526 {
527 	if (FIELD_GET(NFP_FLOWER_CMSG_PORT_TYPE, port) ==
528 	    NFP_FLOWER_CMSG_PORT_TYPE_PHYS_PORT)
529 		return true;
530 
531 	return false;
532 }
533 
534 static u16 nfp_tunnel_get_mac_idx_from_phy_port_id(int port)
535 {
536 	return port << 8 | NFP_FLOWER_CMSG_PORT_TYPE_PHYS_PORT;
537 }
538 
539 static u16 nfp_tunnel_get_global_mac_idx_from_ida(int id)
540 {
541 	return id << 8 | NFP_FLOWER_CMSG_PORT_TYPE_OTHER_PORT;
542 }
543 
544 static int nfp_tunnel_get_ida_from_global_mac_idx(u16 nfp_mac_idx)
545 {
546 	return nfp_mac_idx >> 8;
547 }
548 
549 static bool nfp_tunnel_is_mac_idx_global(u16 nfp_mac_idx)
550 {
551 	return (nfp_mac_idx & 0xff) == NFP_FLOWER_CMSG_PORT_TYPE_OTHER_PORT;
552 }
553 
554 static struct nfp_tun_offloaded_mac *
555 nfp_tunnel_lookup_offloaded_macs(struct nfp_app *app, u8 *mac)
556 {
557 	struct nfp_flower_priv *priv = app->priv;
558 
559 	return rhashtable_lookup_fast(&priv->tun.offloaded_macs, mac,
560 				      offloaded_macs_params);
561 }
562 
563 static void
564 nfp_tunnel_offloaded_macs_inc_ref_and_link(struct nfp_tun_offloaded_mac *entry,
565 					   struct net_device *netdev, bool mod)
566 {
567 	if (nfp_netdev_is_nfp_repr(netdev)) {
568 		struct nfp_flower_repr_priv *repr_priv;
569 		struct nfp_repr *repr;
570 
571 		repr = netdev_priv(netdev);
572 		repr_priv = repr->app_priv;
573 
574 		/* If modifing MAC, remove repr from old list first. */
575 		if (mod)
576 			list_del(&repr_priv->mac_list);
577 
578 		list_add_tail(&repr_priv->mac_list, &entry->repr_list);
579 	} else if (nfp_flower_is_supported_bridge(netdev)) {
580 		entry->bridge_count++;
581 	}
582 
583 	entry->ref_count++;
584 }
585 
586 static int
587 nfp_tunnel_add_shared_mac(struct nfp_app *app, struct net_device *netdev,
588 			  int port, bool mod)
589 {
590 	struct nfp_flower_priv *priv = app->priv;
591 	int ida_idx = NFP_MAX_MAC_INDEX, err;
592 	struct nfp_tun_offloaded_mac *entry;
593 	u16 nfp_mac_idx = 0;
594 
595 	entry = nfp_tunnel_lookup_offloaded_macs(app, netdev->dev_addr);
596 	if (entry && nfp_tunnel_is_mac_idx_global(entry->index)) {
597 		if (entry->bridge_count ||
598 		    !nfp_flower_is_supported_bridge(netdev)) {
599 			nfp_tunnel_offloaded_macs_inc_ref_and_link(entry,
600 								   netdev, mod);
601 			return 0;
602 		}
603 
604 		/* MAC is global but matches need to go to pre_tun table. */
605 		nfp_mac_idx = entry->index | NFP_TUN_PRE_TUN_IDX_BIT;
606 	}
607 
608 	if (!nfp_mac_idx) {
609 		/* Assign a global index if non-repr or MAC is now shared. */
610 		if (entry || !port) {
611 			ida_idx = ida_simple_get(&priv->tun.mac_off_ids, 0,
612 						 NFP_MAX_MAC_INDEX, GFP_KERNEL);
613 			if (ida_idx < 0)
614 				return ida_idx;
615 
616 			nfp_mac_idx =
617 				nfp_tunnel_get_global_mac_idx_from_ida(ida_idx);
618 
619 			if (nfp_flower_is_supported_bridge(netdev))
620 				nfp_mac_idx |= NFP_TUN_PRE_TUN_IDX_BIT;
621 
622 		} else {
623 			nfp_mac_idx =
624 				nfp_tunnel_get_mac_idx_from_phy_port_id(port);
625 		}
626 	}
627 
628 	if (!entry) {
629 		entry = kzalloc(sizeof(*entry), GFP_KERNEL);
630 		if (!entry) {
631 			err = -ENOMEM;
632 			goto err_free_ida;
633 		}
634 
635 		ether_addr_copy(entry->addr, netdev->dev_addr);
636 		INIT_LIST_HEAD(&entry->repr_list);
637 
638 		if (rhashtable_insert_fast(&priv->tun.offloaded_macs,
639 					   &entry->ht_node,
640 					   offloaded_macs_params)) {
641 			err = -ENOMEM;
642 			goto err_free_entry;
643 		}
644 	}
645 
646 	err = __nfp_tunnel_offload_mac(app, netdev->dev_addr,
647 				       nfp_mac_idx, false);
648 	if (err) {
649 		/* If not shared then free. */
650 		if (!entry->ref_count)
651 			goto err_remove_hash;
652 		goto err_free_ida;
653 	}
654 
655 	entry->index = nfp_mac_idx;
656 	nfp_tunnel_offloaded_macs_inc_ref_and_link(entry, netdev, mod);
657 
658 	return 0;
659 
660 err_remove_hash:
661 	rhashtable_remove_fast(&priv->tun.offloaded_macs, &entry->ht_node,
662 			       offloaded_macs_params);
663 err_free_entry:
664 	kfree(entry);
665 err_free_ida:
666 	if (ida_idx != NFP_MAX_MAC_INDEX)
667 		ida_simple_remove(&priv->tun.mac_off_ids, ida_idx);
668 
669 	return err;
670 }
671 
672 static int
673 nfp_tunnel_del_shared_mac(struct nfp_app *app, struct net_device *netdev,
674 			  u8 *mac, bool mod)
675 {
676 	struct nfp_flower_priv *priv = app->priv;
677 	struct nfp_flower_repr_priv *repr_priv;
678 	struct nfp_tun_offloaded_mac *entry;
679 	struct nfp_repr *repr;
680 	int ida_idx;
681 
682 	entry = nfp_tunnel_lookup_offloaded_macs(app, mac);
683 	if (!entry)
684 		return 0;
685 
686 	entry->ref_count--;
687 	/* If del is part of a mod then mac_list is still in use elsewheree. */
688 	if (nfp_netdev_is_nfp_repr(netdev) && !mod) {
689 		repr = netdev_priv(netdev);
690 		repr_priv = repr->app_priv;
691 		list_del(&repr_priv->mac_list);
692 	}
693 
694 	if (nfp_flower_is_supported_bridge(netdev)) {
695 		entry->bridge_count--;
696 
697 		if (!entry->bridge_count && entry->ref_count) {
698 			u16 nfp_mac_idx;
699 
700 			nfp_mac_idx = entry->index & ~NFP_TUN_PRE_TUN_IDX_BIT;
701 			if (__nfp_tunnel_offload_mac(app, mac, nfp_mac_idx,
702 						     false)) {
703 				nfp_flower_cmsg_warn(app, "MAC offload index revert failed on %s.\n",
704 						     netdev_name(netdev));
705 				return 0;
706 			}
707 
708 			entry->index = nfp_mac_idx;
709 			return 0;
710 		}
711 	}
712 
713 	/* If MAC is now used by 1 repr set the offloaded MAC index to port. */
714 	if (entry->ref_count == 1 && list_is_singular(&entry->repr_list)) {
715 		u16 nfp_mac_idx;
716 		int port, err;
717 
718 		repr_priv = list_first_entry(&entry->repr_list,
719 					     struct nfp_flower_repr_priv,
720 					     mac_list);
721 		repr = repr_priv->nfp_repr;
722 		port = nfp_repr_get_port_id(repr->netdev);
723 		nfp_mac_idx = nfp_tunnel_get_mac_idx_from_phy_port_id(port);
724 		err = __nfp_tunnel_offload_mac(app, mac, nfp_mac_idx, false);
725 		if (err) {
726 			nfp_flower_cmsg_warn(app, "MAC offload index revert failed on %s.\n",
727 					     netdev_name(netdev));
728 			return 0;
729 		}
730 
731 		ida_idx = nfp_tunnel_get_ida_from_global_mac_idx(entry->index);
732 		ida_simple_remove(&priv->tun.mac_off_ids, ida_idx);
733 		entry->index = nfp_mac_idx;
734 		return 0;
735 	}
736 
737 	if (entry->ref_count)
738 		return 0;
739 
740 	WARN_ON_ONCE(rhashtable_remove_fast(&priv->tun.offloaded_macs,
741 					    &entry->ht_node,
742 					    offloaded_macs_params));
743 	/* If MAC has global ID then extract and free the ida entry. */
744 	if (nfp_tunnel_is_mac_idx_global(entry->index)) {
745 		ida_idx = nfp_tunnel_get_ida_from_global_mac_idx(entry->index);
746 		ida_simple_remove(&priv->tun.mac_off_ids, ida_idx);
747 	}
748 
749 	kfree(entry);
750 
751 	return __nfp_tunnel_offload_mac(app, mac, 0, true);
752 }
753 
754 static int
755 nfp_tunnel_offload_mac(struct nfp_app *app, struct net_device *netdev,
756 		       enum nfp_flower_mac_offload_cmd cmd)
757 {
758 	struct nfp_flower_non_repr_priv *nr_priv = NULL;
759 	bool non_repr = false, *mac_offloaded;
760 	u8 *off_mac = NULL;
761 	int err, port = 0;
762 
763 	if (nfp_netdev_is_nfp_repr(netdev)) {
764 		struct nfp_flower_repr_priv *repr_priv;
765 		struct nfp_repr *repr;
766 
767 		repr = netdev_priv(netdev);
768 		if (repr->app != app)
769 			return 0;
770 
771 		repr_priv = repr->app_priv;
772 		if (repr_priv->on_bridge)
773 			return 0;
774 
775 		mac_offloaded = &repr_priv->mac_offloaded;
776 		off_mac = &repr_priv->offloaded_mac_addr[0];
777 		port = nfp_repr_get_port_id(netdev);
778 		if (!nfp_tunnel_port_is_phy_repr(port))
779 			return 0;
780 	} else if (nfp_fl_is_netdev_to_offload(netdev)) {
781 		nr_priv = nfp_flower_non_repr_priv_get(app, netdev);
782 		if (!nr_priv)
783 			return -ENOMEM;
784 
785 		mac_offloaded = &nr_priv->mac_offloaded;
786 		off_mac = &nr_priv->offloaded_mac_addr[0];
787 		non_repr = true;
788 	} else {
789 		return 0;
790 	}
791 
792 	if (!is_valid_ether_addr(netdev->dev_addr)) {
793 		err = -EINVAL;
794 		goto err_put_non_repr_priv;
795 	}
796 
797 	if (cmd == NFP_TUNNEL_MAC_OFFLOAD_MOD && !*mac_offloaded)
798 		cmd = NFP_TUNNEL_MAC_OFFLOAD_ADD;
799 
800 	switch (cmd) {
801 	case NFP_TUNNEL_MAC_OFFLOAD_ADD:
802 		err = nfp_tunnel_add_shared_mac(app, netdev, port, false);
803 		if (err)
804 			goto err_put_non_repr_priv;
805 
806 		if (non_repr)
807 			__nfp_flower_non_repr_priv_get(nr_priv);
808 
809 		*mac_offloaded = true;
810 		ether_addr_copy(off_mac, netdev->dev_addr);
811 		break;
812 	case NFP_TUNNEL_MAC_OFFLOAD_DEL:
813 		/* Only attempt delete if add was successful. */
814 		if (!*mac_offloaded)
815 			break;
816 
817 		if (non_repr)
818 			__nfp_flower_non_repr_priv_put(nr_priv);
819 
820 		*mac_offloaded = false;
821 
822 		err = nfp_tunnel_del_shared_mac(app, netdev, netdev->dev_addr,
823 						false);
824 		if (err)
825 			goto err_put_non_repr_priv;
826 
827 		break;
828 	case NFP_TUNNEL_MAC_OFFLOAD_MOD:
829 		/* Ignore if changing to the same address. */
830 		if (ether_addr_equal(netdev->dev_addr, off_mac))
831 			break;
832 
833 		err = nfp_tunnel_add_shared_mac(app, netdev, port, true);
834 		if (err)
835 			goto err_put_non_repr_priv;
836 
837 		/* Delete the previous MAC address. */
838 		err = nfp_tunnel_del_shared_mac(app, netdev, off_mac, true);
839 		if (err)
840 			nfp_flower_cmsg_warn(app, "Failed to remove offload of replaced MAC addr on %s.\n",
841 					     netdev_name(netdev));
842 
843 		ether_addr_copy(off_mac, netdev->dev_addr);
844 		break;
845 	default:
846 		err = -EINVAL;
847 		goto err_put_non_repr_priv;
848 	}
849 
850 	if (non_repr)
851 		__nfp_flower_non_repr_priv_put(nr_priv);
852 
853 	return 0;
854 
855 err_put_non_repr_priv:
856 	if (non_repr)
857 		__nfp_flower_non_repr_priv_put(nr_priv);
858 
859 	return err;
860 }
861 
862 int nfp_tunnel_mac_event_handler(struct nfp_app *app,
863 				 struct net_device *netdev,
864 				 unsigned long event, void *ptr)
865 {
866 	int err;
867 
868 	if (event == NETDEV_DOWN) {
869 		err = nfp_tunnel_offload_mac(app, netdev,
870 					     NFP_TUNNEL_MAC_OFFLOAD_DEL);
871 		if (err)
872 			nfp_flower_cmsg_warn(app, "Failed to delete offload MAC on %s.\n",
873 					     netdev_name(netdev));
874 	} else if (event == NETDEV_UP) {
875 		err = nfp_tunnel_offload_mac(app, netdev,
876 					     NFP_TUNNEL_MAC_OFFLOAD_ADD);
877 		if (err)
878 			nfp_flower_cmsg_warn(app, "Failed to offload MAC on %s.\n",
879 					     netdev_name(netdev));
880 	} else if (event == NETDEV_CHANGEADDR) {
881 		/* Only offload addr change if netdev is already up. */
882 		if (!(netdev->flags & IFF_UP))
883 			return NOTIFY_OK;
884 
885 		err = nfp_tunnel_offload_mac(app, netdev,
886 					     NFP_TUNNEL_MAC_OFFLOAD_MOD);
887 		if (err)
888 			nfp_flower_cmsg_warn(app, "Failed to offload MAC change on %s.\n",
889 					     netdev_name(netdev));
890 	} else if (event == NETDEV_CHANGEUPPER) {
891 		/* If a repr is attached to a bridge then tunnel packets
892 		 * entering the physical port are directed through the bridge
893 		 * datapath and cannot be directly detunneled. Therefore,
894 		 * associated offloaded MACs and indexes should not be used
895 		 * by fw for detunneling.
896 		 */
897 		struct netdev_notifier_changeupper_info *info = ptr;
898 		struct net_device *upper = info->upper_dev;
899 		struct nfp_flower_repr_priv *repr_priv;
900 		struct nfp_repr *repr;
901 
902 		if (!nfp_netdev_is_nfp_repr(netdev) ||
903 		    !nfp_flower_is_supported_bridge(upper))
904 			return NOTIFY_OK;
905 
906 		repr = netdev_priv(netdev);
907 		if (repr->app != app)
908 			return NOTIFY_OK;
909 
910 		repr_priv = repr->app_priv;
911 
912 		if (info->linking) {
913 			if (nfp_tunnel_offload_mac(app, netdev,
914 						   NFP_TUNNEL_MAC_OFFLOAD_DEL))
915 				nfp_flower_cmsg_warn(app, "Failed to delete offloaded MAC on %s.\n",
916 						     netdev_name(netdev));
917 			repr_priv->on_bridge = true;
918 		} else {
919 			repr_priv->on_bridge = false;
920 
921 			if (!(netdev->flags & IFF_UP))
922 				return NOTIFY_OK;
923 
924 			if (nfp_tunnel_offload_mac(app, netdev,
925 						   NFP_TUNNEL_MAC_OFFLOAD_ADD))
926 				nfp_flower_cmsg_warn(app, "Failed to offload MAC on %s.\n",
927 						     netdev_name(netdev));
928 		}
929 	}
930 	return NOTIFY_OK;
931 }
932 
933 int nfp_flower_xmit_pre_tun_flow(struct nfp_app *app,
934 				 struct nfp_fl_payload *flow)
935 {
936 	struct nfp_flower_priv *app_priv = app->priv;
937 	struct nfp_tun_offloaded_mac *mac_entry;
938 	struct nfp_tun_pre_tun_rule payload;
939 	struct net_device *internal_dev;
940 	int err;
941 
942 	if (app_priv->pre_tun_rule_cnt == NFP_TUN_PRE_TUN_RULE_LIMIT)
943 		return -ENOSPC;
944 
945 	memset(&payload, 0, sizeof(struct nfp_tun_pre_tun_rule));
946 
947 	internal_dev = flow->pre_tun_rule.dev;
948 	payload.vlan_tci = flow->pre_tun_rule.vlan_tci;
949 	payload.host_ctx_id = flow->meta.host_ctx_id;
950 
951 	/* Lookup MAC index for the pre-tunnel rule egress device.
952 	 * Note that because the device is always an internal port, it will
953 	 * have a constant global index so does not need to be tracked.
954 	 */
955 	mac_entry = nfp_tunnel_lookup_offloaded_macs(app,
956 						     internal_dev->dev_addr);
957 	if (!mac_entry)
958 		return -ENOENT;
959 
960 	payload.port_idx = cpu_to_be16(mac_entry->index);
961 
962 	/* Copy mac id and vlan to flow - dev may not exist at delete time. */
963 	flow->pre_tun_rule.vlan_tci = payload.vlan_tci;
964 	flow->pre_tun_rule.port_idx = payload.port_idx;
965 
966 	err = nfp_flower_xmit_tun_conf(app, NFP_FLOWER_CMSG_TYPE_PRE_TUN_RULE,
967 				       sizeof(struct nfp_tun_pre_tun_rule),
968 				       (unsigned char *)&payload, GFP_KERNEL);
969 	if (err)
970 		return err;
971 
972 	app_priv->pre_tun_rule_cnt++;
973 
974 	return 0;
975 }
976 
977 int nfp_flower_xmit_pre_tun_del_flow(struct nfp_app *app,
978 				     struct nfp_fl_payload *flow)
979 {
980 	struct nfp_flower_priv *app_priv = app->priv;
981 	struct nfp_tun_pre_tun_rule payload;
982 	u32 tmp_flags = 0;
983 	int err;
984 
985 	memset(&payload, 0, sizeof(struct nfp_tun_pre_tun_rule));
986 
987 	tmp_flags |= NFP_TUN_PRE_TUN_RULE_DEL;
988 	payload.flags = cpu_to_be32(tmp_flags);
989 	payload.vlan_tci = flow->pre_tun_rule.vlan_tci;
990 	payload.port_idx = flow->pre_tun_rule.port_idx;
991 
992 	err = nfp_flower_xmit_tun_conf(app, NFP_FLOWER_CMSG_TYPE_PRE_TUN_RULE,
993 				       sizeof(struct nfp_tun_pre_tun_rule),
994 				       (unsigned char *)&payload, GFP_KERNEL);
995 	if (err)
996 		return err;
997 
998 	app_priv->pre_tun_rule_cnt--;
999 
1000 	return 0;
1001 }
1002 
1003 int nfp_tunnel_config_start(struct nfp_app *app)
1004 {
1005 	struct nfp_flower_priv *priv = app->priv;
1006 	int err;
1007 
1008 	/* Initialise rhash for MAC offload tracking. */
1009 	err = rhashtable_init(&priv->tun.offloaded_macs,
1010 			      &offloaded_macs_params);
1011 	if (err)
1012 		return err;
1013 
1014 	ida_init(&priv->tun.mac_off_ids);
1015 
1016 	/* Initialise priv data for IPv4 offloading. */
1017 	mutex_init(&priv->tun.ipv4_off_lock);
1018 	INIT_LIST_HEAD(&priv->tun.ipv4_off_list);
1019 
1020 	/* Initialise priv data for neighbour offloading. */
1021 	spin_lock_init(&priv->tun.neigh_off_lock);
1022 	INIT_LIST_HEAD(&priv->tun.neigh_off_list);
1023 	priv->tun.neigh_nb.notifier_call = nfp_tun_neigh_event_handler;
1024 
1025 	err = register_netevent_notifier(&priv->tun.neigh_nb);
1026 	if (err) {
1027 		rhashtable_free_and_destroy(&priv->tun.offloaded_macs,
1028 					    nfp_check_rhashtable_empty, NULL);
1029 		return err;
1030 	}
1031 
1032 	return 0;
1033 }
1034 
1035 void nfp_tunnel_config_stop(struct nfp_app *app)
1036 {
1037 	struct nfp_flower_priv *priv = app->priv;
1038 	struct nfp_ipv4_route_entry *route_entry;
1039 	struct nfp_ipv4_addr_entry *ip_entry;
1040 	struct list_head *ptr, *storage;
1041 
1042 	unregister_netevent_notifier(&priv->tun.neigh_nb);
1043 
1044 	ida_destroy(&priv->tun.mac_off_ids);
1045 
1046 	/* Free any memory that may be occupied by ipv4 list. */
1047 	list_for_each_safe(ptr, storage, &priv->tun.ipv4_off_list) {
1048 		ip_entry = list_entry(ptr, struct nfp_ipv4_addr_entry, list);
1049 		list_del(&ip_entry->list);
1050 		kfree(ip_entry);
1051 	}
1052 
1053 	/* Free any memory that may be occupied by the route list. */
1054 	list_for_each_safe(ptr, storage, &priv->tun.neigh_off_list) {
1055 		route_entry = list_entry(ptr, struct nfp_ipv4_route_entry,
1056 					 list);
1057 		list_del(&route_entry->list);
1058 		kfree(route_entry);
1059 	}
1060 
1061 	/* Destroy rhash. Entries should be cleaned on netdev notifier unreg. */
1062 	rhashtable_free_and_destroy(&priv->tun.offloaded_macs,
1063 				    nfp_check_rhashtable_empty, NULL);
1064 }
1065