xref: /openbmc/linux/drivers/infiniband/core/addr.c (revision 711aab1d)
1 /*
2  * Copyright (c) 2005 Voltaire Inc.  All rights reserved.
3  * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
4  * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
5  * Copyright (c) 2005 Intel Corporation.  All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  */
35 
36 #include <linux/mutex.h>
37 #include <linux/inetdevice.h>
38 #include <linux/slab.h>
39 #include <linux/workqueue.h>
40 #include <linux/module.h>
41 #include <net/arp.h>
42 #include <net/neighbour.h>
43 #include <net/route.h>
44 #include <net/netevent.h>
45 #include <net/addrconf.h>
46 #include <net/ip6_route.h>
47 #include <rdma/ib_addr.h>
48 #include <rdma/ib.h>
49 #include <rdma/rdma_netlink.h>
50 #include <net/netlink.h>
51 
52 #include "core_priv.h"
53 
54 struct addr_req {
55 	struct list_head list;
56 	struct sockaddr_storage src_addr;
57 	struct sockaddr_storage dst_addr;
58 	struct rdma_dev_addr *addr;
59 	struct rdma_addr_client *client;
60 	void *context;
61 	void (*callback)(int status, struct sockaddr *src_addr,
62 			 struct rdma_dev_addr *addr, void *context);
63 	unsigned long timeout;
64 	struct delayed_work work;
65 	int status;
66 	u32 seq;
67 };
68 
69 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
70 
71 static void process_req(struct work_struct *work);
72 
73 static DEFINE_MUTEX(lock);
74 static LIST_HEAD(req_list);
75 static DECLARE_DELAYED_WORK(work, process_req);
76 static struct workqueue_struct *addr_wq;
77 
78 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
79 	[LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
80 		.len = sizeof(struct rdma_nla_ls_gid)},
81 };
82 
83 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
84 {
85 	struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
86 	int ret;
87 
88 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
89 		return false;
90 
91 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
92 			nlmsg_len(nlh), ib_nl_addr_policy, NULL);
93 	if (ret)
94 		return false;
95 
96 	return true;
97 }
98 
99 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
100 {
101 	const struct nlattr *head, *curr;
102 	union ib_gid gid;
103 	struct addr_req *req;
104 	int len, rem;
105 	int found = 0;
106 
107 	head = (const struct nlattr *)nlmsg_data(nlh);
108 	len = nlmsg_len(nlh);
109 
110 	nla_for_each_attr(curr, head, len, rem) {
111 		if (curr->nla_type == LS_NLA_TYPE_DGID)
112 			memcpy(&gid, nla_data(curr), nla_len(curr));
113 	}
114 
115 	mutex_lock(&lock);
116 	list_for_each_entry(req, &req_list, list) {
117 		if (nlh->nlmsg_seq != req->seq)
118 			continue;
119 		/* We set the DGID part, the rest was set earlier */
120 		rdma_addr_set_dgid(req->addr, &gid);
121 		req->status = 0;
122 		found = 1;
123 		break;
124 	}
125 	mutex_unlock(&lock);
126 
127 	if (!found)
128 		pr_info("Couldn't find request waiting for DGID: %pI6\n",
129 			&gid);
130 }
131 
132 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
133 			     struct nlmsghdr *nlh,
134 			     struct netlink_ext_ack *extack)
135 {
136 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
137 	    !(NETLINK_CB(skb).sk))
138 		return -EPERM;
139 
140 	if (ib_nl_is_good_ip_resp(nlh))
141 		ib_nl_process_good_ip_rsep(nlh);
142 
143 	return skb->len;
144 }
145 
146 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
147 			     const void *daddr,
148 			     u32 seq, u16 family)
149 {
150 	struct sk_buff *skb = NULL;
151 	struct nlmsghdr *nlh;
152 	struct rdma_ls_ip_resolve_header *header;
153 	void *data;
154 	size_t size;
155 	int attrtype;
156 	int len;
157 
158 	if (family == AF_INET) {
159 		size = sizeof(struct in_addr);
160 		attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
161 	} else {
162 		size = sizeof(struct in6_addr);
163 		attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
164 	}
165 
166 	len = nla_total_size(sizeof(size));
167 	len += NLMSG_ALIGN(sizeof(*header));
168 
169 	skb = nlmsg_new(len, GFP_KERNEL);
170 	if (!skb)
171 		return -ENOMEM;
172 
173 	data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
174 			    RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
175 	if (!data) {
176 		nlmsg_free(skb);
177 		return -ENODATA;
178 	}
179 
180 	/* Construct the family header first */
181 	header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
182 	header->ifindex = dev_addr->bound_dev_if;
183 	nla_put(skb, attrtype, size, daddr);
184 
185 	/* Repair the nlmsg header length */
186 	nlmsg_end(skb, nlh);
187 	rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
188 
189 	/* Make the request retry, so when we get the response from userspace
190 	 * we will have something.
191 	 */
192 	return -ENODATA;
193 }
194 
195 int rdma_addr_size(struct sockaddr *addr)
196 {
197 	switch (addr->sa_family) {
198 	case AF_INET:
199 		return sizeof(struct sockaddr_in);
200 	case AF_INET6:
201 		return sizeof(struct sockaddr_in6);
202 	case AF_IB:
203 		return sizeof(struct sockaddr_ib);
204 	default:
205 		return 0;
206 	}
207 }
208 EXPORT_SYMBOL(rdma_addr_size);
209 
210 static struct rdma_addr_client self;
211 
212 void rdma_addr_register_client(struct rdma_addr_client *client)
213 {
214 	atomic_set(&client->refcount, 1);
215 	init_completion(&client->comp);
216 }
217 EXPORT_SYMBOL(rdma_addr_register_client);
218 
219 static inline void put_client(struct rdma_addr_client *client)
220 {
221 	if (atomic_dec_and_test(&client->refcount))
222 		complete(&client->comp);
223 }
224 
225 void rdma_addr_unregister_client(struct rdma_addr_client *client)
226 {
227 	put_client(client);
228 	wait_for_completion(&client->comp);
229 }
230 EXPORT_SYMBOL(rdma_addr_unregister_client);
231 
232 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
233 		     const unsigned char *dst_dev_addr)
234 {
235 	dev_addr->dev_type = dev->type;
236 	memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
237 	memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
238 	if (dst_dev_addr)
239 		memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
240 	dev_addr->bound_dev_if = dev->ifindex;
241 	return 0;
242 }
243 EXPORT_SYMBOL(rdma_copy_addr);
244 
245 int rdma_translate_ip(const struct sockaddr *addr,
246 		      struct rdma_dev_addr *dev_addr,
247 		      u16 *vlan_id)
248 {
249 	struct net_device *dev;
250 	int ret = -EADDRNOTAVAIL;
251 
252 	if (dev_addr->bound_dev_if) {
253 		dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
254 		if (!dev)
255 			return -ENODEV;
256 		ret = rdma_copy_addr(dev_addr, dev, NULL);
257 		dev_put(dev);
258 		return ret;
259 	}
260 
261 	switch (addr->sa_family) {
262 	case AF_INET:
263 		dev = ip_dev_find(dev_addr->net,
264 			((const struct sockaddr_in *)addr)->sin_addr.s_addr);
265 
266 		if (!dev)
267 			return ret;
268 
269 		ret = rdma_copy_addr(dev_addr, dev, NULL);
270 		dev_addr->bound_dev_if = dev->ifindex;
271 		if (vlan_id)
272 			*vlan_id = rdma_vlan_dev_vlan_id(dev);
273 		dev_put(dev);
274 		break;
275 #if IS_ENABLED(CONFIG_IPV6)
276 	case AF_INET6:
277 		rcu_read_lock();
278 		for_each_netdev_rcu(dev_addr->net, dev) {
279 			if (ipv6_chk_addr(dev_addr->net,
280 					  &((const struct sockaddr_in6 *)addr)->sin6_addr,
281 					  dev, 1)) {
282 				ret = rdma_copy_addr(dev_addr, dev, NULL);
283 				dev_addr->bound_dev_if = dev->ifindex;
284 				if (vlan_id)
285 					*vlan_id = rdma_vlan_dev_vlan_id(dev);
286 				break;
287 			}
288 		}
289 		rcu_read_unlock();
290 		break;
291 #endif
292 	}
293 	return ret;
294 }
295 EXPORT_SYMBOL(rdma_translate_ip);
296 
297 static void set_timeout(struct delayed_work *delayed_work, unsigned long time)
298 {
299 	unsigned long delay;
300 
301 	delay = time - jiffies;
302 	if ((long)delay < 0)
303 		delay = 0;
304 
305 	mod_delayed_work(addr_wq, delayed_work, delay);
306 }
307 
308 static void queue_req(struct addr_req *req)
309 {
310 	struct addr_req *temp_req;
311 
312 	mutex_lock(&lock);
313 	list_for_each_entry_reverse(temp_req, &req_list, list) {
314 		if (time_after_eq(req->timeout, temp_req->timeout))
315 			break;
316 	}
317 
318 	list_add(&req->list, &temp_req->list);
319 
320 	set_timeout(&req->work, req->timeout);
321 	mutex_unlock(&lock);
322 }
323 
324 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
325 			  const void *daddr, u32 seq, u16 family)
326 {
327 	if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
328 		return -EADDRNOTAVAIL;
329 
330 	/* We fill in what we can, the response will fill the rest */
331 	rdma_copy_addr(dev_addr, dst->dev, NULL);
332 	return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
333 }
334 
335 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
336 			const void *daddr)
337 {
338 	struct neighbour *n;
339 	int ret;
340 
341 	n = dst_neigh_lookup(dst, daddr);
342 
343 	rcu_read_lock();
344 	if (!n || !(n->nud_state & NUD_VALID)) {
345 		if (n)
346 			neigh_event_send(n, NULL);
347 		ret = -ENODATA;
348 	} else {
349 		ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
350 	}
351 	rcu_read_unlock();
352 
353 	if (n)
354 		neigh_release(n);
355 
356 	return ret;
357 }
358 
359 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
360 {
361 	struct rtable *rt;
362 	struct rt6_info *rt6;
363 
364 	if (family == AF_INET) {
365 		rt = container_of(dst, struct rtable, dst);
366 		return rt->rt_uses_gateway;
367 	}
368 
369 	rt6 = container_of(dst, struct rt6_info, dst);
370 	return rt6->rt6i_flags & RTF_GATEWAY;
371 }
372 
373 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
374 		    const struct sockaddr *dst_in, u32 seq)
375 {
376 	const struct sockaddr_in *dst_in4 =
377 		(const struct sockaddr_in *)dst_in;
378 	const struct sockaddr_in6 *dst_in6 =
379 		(const struct sockaddr_in6 *)dst_in;
380 	const void *daddr = (dst_in->sa_family == AF_INET) ?
381 		(const void *)&dst_in4->sin_addr.s_addr :
382 		(const void *)&dst_in6->sin6_addr;
383 	sa_family_t family = dst_in->sa_family;
384 
385 	/* Gateway + ARPHRD_INFINIBAND -> IB router */
386 	if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
387 		return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
388 	else
389 		return dst_fetch_ha(dst, dev_addr, daddr);
390 }
391 
392 static int addr4_resolve(struct sockaddr_in *src_in,
393 			 const struct sockaddr_in *dst_in,
394 			 struct rdma_dev_addr *addr,
395 			 struct rtable **prt)
396 {
397 	__be32 src_ip = src_in->sin_addr.s_addr;
398 	__be32 dst_ip = dst_in->sin_addr.s_addr;
399 	struct rtable *rt;
400 	struct flowi4 fl4;
401 	int ret;
402 
403 	memset(&fl4, 0, sizeof(fl4));
404 	fl4.daddr = dst_ip;
405 	fl4.saddr = src_ip;
406 	fl4.flowi4_oif = addr->bound_dev_if;
407 	rt = ip_route_output_key(addr->net, &fl4);
408 	ret = PTR_ERR_OR_ZERO(rt);
409 	if (ret)
410 		return ret;
411 
412 	src_in->sin_family = AF_INET;
413 	src_in->sin_addr.s_addr = fl4.saddr;
414 
415 	/* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
416 	 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
417 	 * type accordingly.
418 	 */
419 	if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
420 		addr->network = RDMA_NETWORK_IPV4;
421 
422 	addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
423 
424 	*prt = rt;
425 	return 0;
426 }
427 
428 #if IS_ENABLED(CONFIG_IPV6)
429 static int addr6_resolve(struct sockaddr_in6 *src_in,
430 			 const struct sockaddr_in6 *dst_in,
431 			 struct rdma_dev_addr *addr,
432 			 struct dst_entry **pdst)
433 {
434 	struct flowi6 fl6;
435 	struct dst_entry *dst;
436 	struct rt6_info *rt;
437 	int ret;
438 
439 	memset(&fl6, 0, sizeof fl6);
440 	fl6.daddr = dst_in->sin6_addr;
441 	fl6.saddr = src_in->sin6_addr;
442 	fl6.flowi6_oif = addr->bound_dev_if;
443 
444 	ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
445 	if (ret < 0)
446 		return ret;
447 
448 	rt = (struct rt6_info *)dst;
449 	if (ipv6_addr_any(&src_in->sin6_addr)) {
450 		src_in->sin6_family = AF_INET6;
451 		src_in->sin6_addr = fl6.saddr;
452 	}
453 
454 	/* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
455 	 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
456 	 * type accordingly.
457 	 */
458 	if (rt->rt6i_flags & RTF_GATEWAY &&
459 	    ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
460 		addr->network = RDMA_NETWORK_IPV6;
461 
462 	addr->hoplimit = ip6_dst_hoplimit(dst);
463 
464 	*pdst = dst;
465 	return 0;
466 }
467 #else
468 static int addr6_resolve(struct sockaddr_in6 *src_in,
469 			 const struct sockaddr_in6 *dst_in,
470 			 struct rdma_dev_addr *addr,
471 			 struct dst_entry **pdst)
472 {
473 	return -EADDRNOTAVAIL;
474 }
475 #endif
476 
477 static int addr_resolve_neigh(struct dst_entry *dst,
478 			      const struct sockaddr *dst_in,
479 			      struct rdma_dev_addr *addr,
480 			      u32 seq)
481 {
482 	if (dst->dev->flags & IFF_LOOPBACK) {
483 		int ret;
484 
485 		ret = rdma_translate_ip(dst_in, addr, NULL);
486 		if (!ret)
487 			memcpy(addr->dst_dev_addr, addr->src_dev_addr,
488 			       MAX_ADDR_LEN);
489 
490 		return ret;
491 	}
492 
493 	/* If the device doesn't do ARP internally */
494 	if (!(dst->dev->flags & IFF_NOARP))
495 		return fetch_ha(dst, addr, dst_in, seq);
496 
497 	return rdma_copy_addr(addr, dst->dev, NULL);
498 }
499 
500 static int addr_resolve(struct sockaddr *src_in,
501 			const struct sockaddr *dst_in,
502 			struct rdma_dev_addr *addr,
503 			bool resolve_neigh,
504 			u32 seq)
505 {
506 	struct net_device *ndev;
507 	struct dst_entry *dst;
508 	int ret;
509 
510 	if (!addr->net) {
511 		pr_warn_ratelimited("%s: missing namespace\n", __func__);
512 		return -EINVAL;
513 	}
514 
515 	if (src_in->sa_family == AF_INET) {
516 		struct rtable *rt = NULL;
517 		const struct sockaddr_in *dst_in4 =
518 			(const struct sockaddr_in *)dst_in;
519 
520 		ret = addr4_resolve((struct sockaddr_in *)src_in,
521 				    dst_in4, addr, &rt);
522 		if (ret)
523 			return ret;
524 
525 		if (resolve_neigh)
526 			ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
527 
528 		if (addr->bound_dev_if) {
529 			ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
530 		} else {
531 			ndev = rt->dst.dev;
532 			dev_hold(ndev);
533 		}
534 
535 		ip_rt_put(rt);
536 	} else {
537 		const struct sockaddr_in6 *dst_in6 =
538 			(const struct sockaddr_in6 *)dst_in;
539 
540 		ret = addr6_resolve((struct sockaddr_in6 *)src_in,
541 				    dst_in6, addr,
542 				    &dst);
543 		if (ret)
544 			return ret;
545 
546 		if (resolve_neigh)
547 			ret = addr_resolve_neigh(dst, dst_in, addr, seq);
548 
549 		if (addr->bound_dev_if) {
550 			ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
551 		} else {
552 			ndev = dst->dev;
553 			dev_hold(ndev);
554 		}
555 
556 		dst_release(dst);
557 	}
558 
559 	if (ndev->flags & IFF_LOOPBACK) {
560 		ret = rdma_translate_ip(dst_in, addr, NULL);
561 		/*
562 		 * Put the loopback device and get the translated
563 		 * device instead.
564 		 */
565 		dev_put(ndev);
566 		ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
567 	} else {
568 		addr->bound_dev_if = ndev->ifindex;
569 	}
570 	dev_put(ndev);
571 
572 	return ret;
573 }
574 
575 static void process_one_req(struct work_struct *_work)
576 {
577 	struct addr_req *req;
578 	struct sockaddr *src_in, *dst_in;
579 
580 	mutex_lock(&lock);
581 	req = container_of(_work, struct addr_req, work.work);
582 
583 	if (req->status == -ENODATA) {
584 		src_in = (struct sockaddr *)&req->src_addr;
585 		dst_in = (struct sockaddr *)&req->dst_addr;
586 		req->status = addr_resolve(src_in, dst_in, req->addr,
587 					   true, req->seq);
588 		if (req->status && time_after_eq(jiffies, req->timeout)) {
589 			req->status = -ETIMEDOUT;
590 		} else if (req->status == -ENODATA) {
591 			/* requeue the work for retrying again */
592 			set_timeout(&req->work, req->timeout);
593 			mutex_unlock(&lock);
594 			return;
595 		}
596 	}
597 	list_del(&req->list);
598 	mutex_unlock(&lock);
599 
600 	req->callback(req->status, (struct sockaddr *)&req->src_addr,
601 		req->addr, req->context);
602 	put_client(req->client);
603 	kfree(req);
604 }
605 
606 static void process_req(struct work_struct *work)
607 {
608 	struct addr_req *req, *temp_req;
609 	struct sockaddr *src_in, *dst_in;
610 	struct list_head done_list;
611 
612 	INIT_LIST_HEAD(&done_list);
613 
614 	mutex_lock(&lock);
615 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
616 		if (req->status == -ENODATA) {
617 			src_in = (struct sockaddr *) &req->src_addr;
618 			dst_in = (struct sockaddr *) &req->dst_addr;
619 			req->status = addr_resolve(src_in, dst_in, req->addr,
620 						   true, req->seq);
621 			if (req->status && time_after_eq(jiffies, req->timeout))
622 				req->status = -ETIMEDOUT;
623 			else if (req->status == -ENODATA) {
624 				set_timeout(&req->work, req->timeout);
625 				continue;
626 			}
627 		}
628 		list_move_tail(&req->list, &done_list);
629 	}
630 
631 	mutex_unlock(&lock);
632 
633 	list_for_each_entry_safe(req, temp_req, &done_list, list) {
634 		list_del(&req->list);
635 		/* It is safe to cancel other work items from this work item
636 		 * because at a time there can be only one work item running
637 		 * with this single threaded work queue.
638 		 */
639 		cancel_delayed_work(&req->work);
640 		req->callback(req->status, (struct sockaddr *) &req->src_addr,
641 			req->addr, req->context);
642 		put_client(req->client);
643 		kfree(req);
644 	}
645 }
646 
647 int rdma_resolve_ip(struct rdma_addr_client *client,
648 		    struct sockaddr *src_addr, struct sockaddr *dst_addr,
649 		    struct rdma_dev_addr *addr, int timeout_ms,
650 		    void (*callback)(int status, struct sockaddr *src_addr,
651 				     struct rdma_dev_addr *addr, void *context),
652 		    void *context)
653 {
654 	struct sockaddr *src_in, *dst_in;
655 	struct addr_req *req;
656 	int ret = 0;
657 
658 	req = kzalloc(sizeof *req, GFP_KERNEL);
659 	if (!req)
660 		return -ENOMEM;
661 
662 	src_in = (struct sockaddr *) &req->src_addr;
663 	dst_in = (struct sockaddr *) &req->dst_addr;
664 
665 	if (src_addr) {
666 		if (src_addr->sa_family != dst_addr->sa_family) {
667 			ret = -EINVAL;
668 			goto err;
669 		}
670 
671 		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
672 	} else {
673 		src_in->sa_family = dst_addr->sa_family;
674 	}
675 
676 	memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
677 	req->addr = addr;
678 	req->callback = callback;
679 	req->context = context;
680 	req->client = client;
681 	atomic_inc(&client->refcount);
682 	INIT_DELAYED_WORK(&req->work, process_one_req);
683 	req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
684 
685 	req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
686 	switch (req->status) {
687 	case 0:
688 		req->timeout = jiffies;
689 		queue_req(req);
690 		break;
691 	case -ENODATA:
692 		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
693 		queue_req(req);
694 		break;
695 	default:
696 		ret = req->status;
697 		atomic_dec(&client->refcount);
698 		goto err;
699 	}
700 	return ret;
701 err:
702 	kfree(req);
703 	return ret;
704 }
705 EXPORT_SYMBOL(rdma_resolve_ip);
706 
707 int rdma_resolve_ip_route(struct sockaddr *src_addr,
708 			  const struct sockaddr *dst_addr,
709 			  struct rdma_dev_addr *addr)
710 {
711 	struct sockaddr_storage ssrc_addr = {};
712 	struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
713 
714 	if (src_addr) {
715 		if (src_addr->sa_family != dst_addr->sa_family)
716 			return -EINVAL;
717 
718 		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
719 	} else {
720 		src_in->sa_family = dst_addr->sa_family;
721 	}
722 
723 	return addr_resolve(src_in, dst_addr, addr, false, 0);
724 }
725 EXPORT_SYMBOL(rdma_resolve_ip_route);
726 
727 void rdma_addr_cancel(struct rdma_dev_addr *addr)
728 {
729 	struct addr_req *req, *temp_req;
730 
731 	mutex_lock(&lock);
732 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
733 		if (req->addr == addr) {
734 			req->status = -ECANCELED;
735 			req->timeout = jiffies;
736 			list_move(&req->list, &req_list);
737 			set_timeout(&req->work, req->timeout);
738 			break;
739 		}
740 	}
741 	mutex_unlock(&lock);
742 }
743 EXPORT_SYMBOL(rdma_addr_cancel);
744 
745 struct resolve_cb_context {
746 	struct rdma_dev_addr *addr;
747 	struct completion comp;
748 	int status;
749 };
750 
751 static void resolve_cb(int status, struct sockaddr *src_addr,
752 	     struct rdma_dev_addr *addr, void *context)
753 {
754 	if (!status)
755 		memcpy(((struct resolve_cb_context *)context)->addr,
756 		       addr, sizeof(struct rdma_dev_addr));
757 	((struct resolve_cb_context *)context)->status = status;
758 	complete(&((struct resolve_cb_context *)context)->comp);
759 }
760 
761 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
762 				 const union ib_gid *dgid,
763 				 u8 *dmac, u16 *vlan_id, int *if_index,
764 				 int *hoplimit)
765 {
766 	int ret = 0;
767 	struct rdma_dev_addr dev_addr;
768 	struct resolve_cb_context ctx;
769 	struct net_device *dev;
770 
771 	union {
772 		struct sockaddr     _sockaddr;
773 		struct sockaddr_in  _sockaddr_in;
774 		struct sockaddr_in6 _sockaddr_in6;
775 	} sgid_addr, dgid_addr;
776 
777 
778 	rdma_gid2ip(&sgid_addr._sockaddr, sgid);
779 	rdma_gid2ip(&dgid_addr._sockaddr, dgid);
780 
781 	memset(&dev_addr, 0, sizeof(dev_addr));
782 	if (if_index)
783 		dev_addr.bound_dev_if = *if_index;
784 	dev_addr.net = &init_net;
785 
786 	ctx.addr = &dev_addr;
787 	init_completion(&ctx.comp);
788 	ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
789 			&dev_addr, 1000, resolve_cb, &ctx);
790 	if (ret)
791 		return ret;
792 
793 	wait_for_completion(&ctx.comp);
794 
795 	ret = ctx.status;
796 	if (ret)
797 		return ret;
798 
799 	memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
800 	dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
801 	if (!dev)
802 		return -ENODEV;
803 	if (if_index)
804 		*if_index = dev_addr.bound_dev_if;
805 	if (vlan_id)
806 		*vlan_id = rdma_vlan_dev_vlan_id(dev);
807 	if (hoplimit)
808 		*hoplimit = dev_addr.hoplimit;
809 	dev_put(dev);
810 	return ret;
811 }
812 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
813 
814 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
815 {
816 	int ret = 0;
817 	struct rdma_dev_addr dev_addr;
818 	union {
819 		struct sockaddr     _sockaddr;
820 		struct sockaddr_in  _sockaddr_in;
821 		struct sockaddr_in6 _sockaddr_in6;
822 	} gid_addr;
823 
824 	rdma_gid2ip(&gid_addr._sockaddr, sgid);
825 
826 	memset(&dev_addr, 0, sizeof(dev_addr));
827 	dev_addr.net = &init_net;
828 	ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
829 	if (ret)
830 		return ret;
831 
832 	memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
833 	return ret;
834 }
835 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
836 
837 static int netevent_callback(struct notifier_block *self, unsigned long event,
838 	void *ctx)
839 {
840 	if (event == NETEVENT_NEIGH_UPDATE) {
841 		struct neighbour *neigh = ctx;
842 
843 		if (neigh->nud_state & NUD_VALID)
844 			set_timeout(&work, jiffies);
845 	}
846 	return 0;
847 }
848 
849 static struct notifier_block nb = {
850 	.notifier_call = netevent_callback
851 };
852 
853 int addr_init(void)
854 {
855 	addr_wq = alloc_ordered_workqueue("ib_addr", WQ_MEM_RECLAIM);
856 	if (!addr_wq)
857 		return -ENOMEM;
858 
859 	register_netevent_notifier(&nb);
860 	rdma_addr_register_client(&self);
861 
862 	return 0;
863 }
864 
865 void addr_cleanup(void)
866 {
867 	rdma_addr_unregister_client(&self);
868 	unregister_netevent_notifier(&nb);
869 	destroy_workqueue(addr_wq);
870 }
871