xref: /openbmc/linux/drivers/infiniband/core/addr.c (revision a17922de)
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 	void *context;
60 	void (*callback)(int status, struct sockaddr *src_addr,
61 			 struct rdma_dev_addr *addr, void *context);
62 	unsigned long timeout;
63 	struct delayed_work work;
64 	int status;
65 	u32 seq;
66 };
67 
68 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
69 
70 static DEFINE_SPINLOCK(lock);
71 static LIST_HEAD(req_list);
72 static struct workqueue_struct *addr_wq;
73 
74 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
75 	[LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
76 		.len = sizeof(struct rdma_nla_ls_gid)},
77 };
78 
79 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
80 {
81 	struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
82 	int ret;
83 
84 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
85 		return false;
86 
87 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
88 			nlmsg_len(nlh), ib_nl_addr_policy, NULL);
89 	if (ret)
90 		return false;
91 
92 	return true;
93 }
94 
95 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
96 {
97 	const struct nlattr *head, *curr;
98 	union ib_gid gid;
99 	struct addr_req *req;
100 	int len, rem;
101 	int found = 0;
102 
103 	head = (const struct nlattr *)nlmsg_data(nlh);
104 	len = nlmsg_len(nlh);
105 
106 	nla_for_each_attr(curr, head, len, rem) {
107 		if (curr->nla_type == LS_NLA_TYPE_DGID)
108 			memcpy(&gid, nla_data(curr), nla_len(curr));
109 	}
110 
111 	spin_lock_bh(&lock);
112 	list_for_each_entry(req, &req_list, list) {
113 		if (nlh->nlmsg_seq != req->seq)
114 			continue;
115 		/* We set the DGID part, the rest was set earlier */
116 		rdma_addr_set_dgid(req->addr, &gid);
117 		req->status = 0;
118 		found = 1;
119 		break;
120 	}
121 	spin_unlock_bh(&lock);
122 
123 	if (!found)
124 		pr_info("Couldn't find request waiting for DGID: %pI6\n",
125 			&gid);
126 }
127 
128 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
129 			     struct nlmsghdr *nlh,
130 			     struct netlink_ext_ack *extack)
131 {
132 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
133 	    !(NETLINK_CB(skb).sk))
134 		return -EPERM;
135 
136 	if (ib_nl_is_good_ip_resp(nlh))
137 		ib_nl_process_good_ip_rsep(nlh);
138 
139 	return skb->len;
140 }
141 
142 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
143 			     const void *daddr,
144 			     u32 seq, u16 family)
145 {
146 	struct sk_buff *skb = NULL;
147 	struct nlmsghdr *nlh;
148 	struct rdma_ls_ip_resolve_header *header;
149 	void *data;
150 	size_t size;
151 	int attrtype;
152 	int len;
153 
154 	if (family == AF_INET) {
155 		size = sizeof(struct in_addr);
156 		attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
157 	} else {
158 		size = sizeof(struct in6_addr);
159 		attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
160 	}
161 
162 	len = nla_total_size(sizeof(size));
163 	len += NLMSG_ALIGN(sizeof(*header));
164 
165 	skb = nlmsg_new(len, GFP_KERNEL);
166 	if (!skb)
167 		return -ENOMEM;
168 
169 	data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
170 			    RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
171 	if (!data) {
172 		nlmsg_free(skb);
173 		return -ENODATA;
174 	}
175 
176 	/* Construct the family header first */
177 	header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
178 	header->ifindex = dev_addr->bound_dev_if;
179 	nla_put(skb, attrtype, size, daddr);
180 
181 	/* Repair the nlmsg header length */
182 	nlmsg_end(skb, nlh);
183 	rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
184 
185 	/* Make the request retry, so when we get the response from userspace
186 	 * we will have something.
187 	 */
188 	return -ENODATA;
189 }
190 
191 int rdma_addr_size(struct sockaddr *addr)
192 {
193 	switch (addr->sa_family) {
194 	case AF_INET:
195 		return sizeof(struct sockaddr_in);
196 	case AF_INET6:
197 		return sizeof(struct sockaddr_in6);
198 	case AF_IB:
199 		return sizeof(struct sockaddr_ib);
200 	default:
201 		return 0;
202 	}
203 }
204 EXPORT_SYMBOL(rdma_addr_size);
205 
206 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
207 {
208 	int ret = rdma_addr_size((struct sockaddr *) addr);
209 
210 	return ret <= sizeof(*addr) ? ret : 0;
211 }
212 EXPORT_SYMBOL(rdma_addr_size_in6);
213 
214 int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
215 {
216 	int ret = rdma_addr_size((struct sockaddr *) addr);
217 
218 	return ret <= sizeof(*addr) ? ret : 0;
219 }
220 EXPORT_SYMBOL(rdma_addr_size_kss);
221 
222 void rdma_copy_addr(struct rdma_dev_addr *dev_addr,
223 		    const struct net_device *dev,
224 		    const unsigned char *dst_dev_addr)
225 {
226 	dev_addr->dev_type = dev->type;
227 	memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
228 	memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
229 	if (dst_dev_addr)
230 		memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
231 	dev_addr->bound_dev_if = dev->ifindex;
232 }
233 EXPORT_SYMBOL(rdma_copy_addr);
234 
235 int rdma_translate_ip(const struct sockaddr *addr,
236 		      struct rdma_dev_addr *dev_addr)
237 {
238 	struct net_device *dev;
239 
240 	if (dev_addr->bound_dev_if) {
241 		dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
242 		if (!dev)
243 			return -ENODEV;
244 		rdma_copy_addr(dev_addr, dev, NULL);
245 		dev_put(dev);
246 		return 0;
247 	}
248 
249 	switch (addr->sa_family) {
250 	case AF_INET:
251 		dev = ip_dev_find(dev_addr->net,
252 			((const struct sockaddr_in *)addr)->sin_addr.s_addr);
253 
254 		if (!dev)
255 			return -EADDRNOTAVAIL;
256 
257 		rdma_copy_addr(dev_addr, dev, NULL);
258 		dev_put(dev);
259 		break;
260 #if IS_ENABLED(CONFIG_IPV6)
261 	case AF_INET6:
262 		rcu_read_lock();
263 		for_each_netdev_rcu(dev_addr->net, dev) {
264 			if (ipv6_chk_addr(dev_addr->net,
265 					  &((const struct sockaddr_in6 *)addr)->sin6_addr,
266 					  dev, 1)) {
267 				rdma_copy_addr(dev_addr, dev, NULL);
268 				break;
269 			}
270 		}
271 		rcu_read_unlock();
272 		break;
273 #endif
274 	}
275 	return 0;
276 }
277 EXPORT_SYMBOL(rdma_translate_ip);
278 
279 static void set_timeout(struct addr_req *req, unsigned long time)
280 {
281 	unsigned long delay;
282 
283 	delay = time - jiffies;
284 	if ((long)delay < 0)
285 		delay = 0;
286 
287 	mod_delayed_work(addr_wq, &req->work, delay);
288 }
289 
290 static void queue_req(struct addr_req *req)
291 {
292 	spin_lock_bh(&lock);
293 	list_add_tail(&req->list, &req_list);
294 	set_timeout(req, req->timeout);
295 	spin_unlock_bh(&lock);
296 }
297 
298 static int ib_nl_fetch_ha(const struct dst_entry *dst,
299 			  struct rdma_dev_addr *dev_addr,
300 			  const void *daddr, u32 seq, u16 family)
301 {
302 	if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
303 		return -EADDRNOTAVAIL;
304 
305 	/* We fill in what we can, the response will fill the rest */
306 	rdma_copy_addr(dev_addr, dst->dev, NULL);
307 	return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
308 }
309 
310 static int dst_fetch_ha(const struct dst_entry *dst,
311 			struct rdma_dev_addr *dev_addr,
312 			const void *daddr)
313 {
314 	struct neighbour *n;
315 	int ret = 0;
316 
317 	n = dst_neigh_lookup(dst, daddr);
318 
319 	rcu_read_lock();
320 	if (!n || !(n->nud_state & NUD_VALID)) {
321 		if (n)
322 			neigh_event_send(n, NULL);
323 		ret = -ENODATA;
324 	} else {
325 		rdma_copy_addr(dev_addr, dst->dev, n->ha);
326 	}
327 	rcu_read_unlock();
328 
329 	if (n)
330 		neigh_release(n);
331 
332 	return ret;
333 }
334 
335 static bool has_gateway(const struct dst_entry *dst, sa_family_t family)
336 {
337 	struct rtable *rt;
338 	struct rt6_info *rt6;
339 
340 	if (family == AF_INET) {
341 		rt = container_of(dst, struct rtable, dst);
342 		return rt->rt_uses_gateway;
343 	}
344 
345 	rt6 = container_of(dst, struct rt6_info, dst);
346 	return rt6->rt6i_flags & RTF_GATEWAY;
347 }
348 
349 static int fetch_ha(const struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
350 		    const struct sockaddr *dst_in, u32 seq)
351 {
352 	const struct sockaddr_in *dst_in4 =
353 		(const struct sockaddr_in *)dst_in;
354 	const struct sockaddr_in6 *dst_in6 =
355 		(const struct sockaddr_in6 *)dst_in;
356 	const void *daddr = (dst_in->sa_family == AF_INET) ?
357 		(const void *)&dst_in4->sin_addr.s_addr :
358 		(const void *)&dst_in6->sin6_addr;
359 	sa_family_t family = dst_in->sa_family;
360 
361 	/* Gateway + ARPHRD_INFINIBAND -> IB router */
362 	if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
363 		return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
364 	else
365 		return dst_fetch_ha(dst, dev_addr, daddr);
366 }
367 
368 static int addr4_resolve(struct sockaddr_in *src_in,
369 			 const struct sockaddr_in *dst_in,
370 			 struct rdma_dev_addr *addr,
371 			 struct rtable **prt)
372 {
373 	__be32 src_ip = src_in->sin_addr.s_addr;
374 	__be32 dst_ip = dst_in->sin_addr.s_addr;
375 	struct rtable *rt;
376 	struct flowi4 fl4;
377 	int ret;
378 
379 	memset(&fl4, 0, sizeof(fl4));
380 	fl4.daddr = dst_ip;
381 	fl4.saddr = src_ip;
382 	fl4.flowi4_oif = addr->bound_dev_if;
383 	rt = ip_route_output_key(addr->net, &fl4);
384 	ret = PTR_ERR_OR_ZERO(rt);
385 	if (ret)
386 		return ret;
387 
388 	src_in->sin_family = AF_INET;
389 	src_in->sin_addr.s_addr = fl4.saddr;
390 
391 	/* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
392 	 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
393 	 * type accordingly.
394 	 */
395 	if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
396 		addr->network = RDMA_NETWORK_IPV4;
397 
398 	addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
399 
400 	*prt = rt;
401 	return 0;
402 }
403 
404 #if IS_ENABLED(CONFIG_IPV6)
405 static int addr6_resolve(struct sockaddr_in6 *src_in,
406 			 const struct sockaddr_in6 *dst_in,
407 			 struct rdma_dev_addr *addr,
408 			 struct dst_entry **pdst)
409 {
410 	struct flowi6 fl6;
411 	struct dst_entry *dst;
412 	struct rt6_info *rt;
413 	int ret;
414 
415 	memset(&fl6, 0, sizeof fl6);
416 	fl6.daddr = dst_in->sin6_addr;
417 	fl6.saddr = src_in->sin6_addr;
418 	fl6.flowi6_oif = addr->bound_dev_if;
419 
420 	ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
421 	if (ret < 0)
422 		return ret;
423 
424 	rt = (struct rt6_info *)dst;
425 	if (ipv6_addr_any(&src_in->sin6_addr)) {
426 		src_in->sin6_family = AF_INET6;
427 		src_in->sin6_addr = fl6.saddr;
428 	}
429 
430 	/* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
431 	 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
432 	 * type accordingly.
433 	 */
434 	if (rt->rt6i_flags & RTF_GATEWAY &&
435 	    ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
436 		addr->network = RDMA_NETWORK_IPV6;
437 
438 	addr->hoplimit = ip6_dst_hoplimit(dst);
439 
440 	*pdst = dst;
441 	return 0;
442 }
443 #else
444 static int addr6_resolve(struct sockaddr_in6 *src_in,
445 			 const struct sockaddr_in6 *dst_in,
446 			 struct rdma_dev_addr *addr,
447 			 struct dst_entry **pdst)
448 {
449 	return -EADDRNOTAVAIL;
450 }
451 #endif
452 
453 static int addr_resolve_neigh(const struct dst_entry *dst,
454 			      const struct sockaddr *dst_in,
455 			      struct rdma_dev_addr *addr,
456 			      u32 seq)
457 {
458 	if (dst->dev->flags & IFF_LOOPBACK) {
459 		int ret;
460 
461 		ret = rdma_translate_ip(dst_in, addr);
462 		if (!ret)
463 			memcpy(addr->dst_dev_addr, addr->src_dev_addr,
464 			       MAX_ADDR_LEN);
465 
466 		return ret;
467 	}
468 
469 	/* If the device doesn't do ARP internally */
470 	if (!(dst->dev->flags & IFF_NOARP))
471 		return fetch_ha(dst, addr, dst_in, seq);
472 
473 	rdma_copy_addr(addr, dst->dev, NULL);
474 
475 	return 0;
476 }
477 
478 static int addr_resolve(struct sockaddr *src_in,
479 			const struct sockaddr *dst_in,
480 			struct rdma_dev_addr *addr,
481 			bool resolve_neigh,
482 			u32 seq)
483 {
484 	struct net_device *ndev;
485 	struct dst_entry *dst;
486 	int ret;
487 
488 	if (!addr->net) {
489 		pr_warn_ratelimited("%s: missing namespace\n", __func__);
490 		return -EINVAL;
491 	}
492 
493 	if (src_in->sa_family == AF_INET) {
494 		struct rtable *rt = NULL;
495 		const struct sockaddr_in *dst_in4 =
496 			(const struct sockaddr_in *)dst_in;
497 
498 		ret = addr4_resolve((struct sockaddr_in *)src_in,
499 				    dst_in4, addr, &rt);
500 		if (ret)
501 			return ret;
502 
503 		if (resolve_neigh)
504 			ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
505 
506 		if (addr->bound_dev_if) {
507 			ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
508 		} else {
509 			ndev = rt->dst.dev;
510 			dev_hold(ndev);
511 		}
512 
513 		ip_rt_put(rt);
514 	} else {
515 		const struct sockaddr_in6 *dst_in6 =
516 			(const struct sockaddr_in6 *)dst_in;
517 
518 		ret = addr6_resolve((struct sockaddr_in6 *)src_in,
519 				    dst_in6, addr,
520 				    &dst);
521 		if (ret)
522 			return ret;
523 
524 		if (resolve_neigh)
525 			ret = addr_resolve_neigh(dst, dst_in, addr, seq);
526 
527 		if (addr->bound_dev_if) {
528 			ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
529 		} else {
530 			ndev = dst->dev;
531 			dev_hold(ndev);
532 		}
533 
534 		dst_release(dst);
535 	}
536 
537 	if (ndev) {
538 		if (ndev->flags & IFF_LOOPBACK)
539 			ret = rdma_translate_ip(dst_in, addr);
540 		else
541 			addr->bound_dev_if = ndev->ifindex;
542 		dev_put(ndev);
543 	}
544 
545 	return ret;
546 }
547 
548 static void process_one_req(struct work_struct *_work)
549 {
550 	struct addr_req *req;
551 	struct sockaddr *src_in, *dst_in;
552 
553 	req = container_of(_work, struct addr_req, work.work);
554 
555 	if (req->status == -ENODATA) {
556 		src_in = (struct sockaddr *)&req->src_addr;
557 		dst_in = (struct sockaddr *)&req->dst_addr;
558 		req->status = addr_resolve(src_in, dst_in, req->addr,
559 					   true, req->seq);
560 		if (req->status && time_after_eq(jiffies, req->timeout)) {
561 			req->status = -ETIMEDOUT;
562 		} else if (req->status == -ENODATA) {
563 			/* requeue the work for retrying again */
564 			spin_lock_bh(&lock);
565 			if (!list_empty(&req->list))
566 				set_timeout(req, req->timeout);
567 			spin_unlock_bh(&lock);
568 			return;
569 		}
570 	}
571 
572 	req->callback(req->status, (struct sockaddr *)&req->src_addr,
573 		req->addr, req->context);
574 	req->callback = NULL;
575 
576 	spin_lock_bh(&lock);
577 	if (!list_empty(&req->list)) {
578 		/*
579 		 * Although the work will normally have been canceled by the
580 		 * workqueue, it can still be requeued as long as it is on the
581 		 * req_list.
582 		 */
583 		cancel_delayed_work(&req->work);
584 		list_del_init(&req->list);
585 		kfree(req);
586 	}
587 	spin_unlock_bh(&lock);
588 }
589 
590 int rdma_resolve_ip(struct sockaddr *src_addr, struct sockaddr *dst_addr,
591 		    struct rdma_dev_addr *addr, int timeout_ms,
592 		    void (*callback)(int status, struct sockaddr *src_addr,
593 				     struct rdma_dev_addr *addr, void *context),
594 		    void *context)
595 {
596 	struct sockaddr *src_in, *dst_in;
597 	struct addr_req *req;
598 	int ret = 0;
599 
600 	req = kzalloc(sizeof *req, GFP_KERNEL);
601 	if (!req)
602 		return -ENOMEM;
603 
604 	src_in = (struct sockaddr *) &req->src_addr;
605 	dst_in = (struct sockaddr *) &req->dst_addr;
606 
607 	if (src_addr) {
608 		if (src_addr->sa_family != dst_addr->sa_family) {
609 			ret = -EINVAL;
610 			goto err;
611 		}
612 
613 		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
614 	} else {
615 		src_in->sa_family = dst_addr->sa_family;
616 	}
617 
618 	memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
619 	req->addr = addr;
620 	req->callback = callback;
621 	req->context = context;
622 	INIT_DELAYED_WORK(&req->work, process_one_req);
623 	req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
624 
625 	req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
626 	switch (req->status) {
627 	case 0:
628 		req->timeout = jiffies;
629 		queue_req(req);
630 		break;
631 	case -ENODATA:
632 		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
633 		queue_req(req);
634 		break;
635 	default:
636 		ret = req->status;
637 		goto err;
638 	}
639 	return ret;
640 err:
641 	kfree(req);
642 	return ret;
643 }
644 EXPORT_SYMBOL(rdma_resolve_ip);
645 
646 int rdma_resolve_ip_route(struct sockaddr *src_addr,
647 			  const struct sockaddr *dst_addr,
648 			  struct rdma_dev_addr *addr)
649 {
650 	struct sockaddr_storage ssrc_addr = {};
651 	struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
652 
653 	if (src_addr) {
654 		if (src_addr->sa_family != dst_addr->sa_family)
655 			return -EINVAL;
656 
657 		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
658 	} else {
659 		src_in->sa_family = dst_addr->sa_family;
660 	}
661 
662 	return addr_resolve(src_in, dst_addr, addr, false, 0);
663 }
664 
665 void rdma_addr_cancel(struct rdma_dev_addr *addr)
666 {
667 	struct addr_req *req, *temp_req;
668 	struct addr_req *found = NULL;
669 
670 	spin_lock_bh(&lock);
671 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
672 		if (req->addr == addr) {
673 			/*
674 			 * Removing from the list means we take ownership of
675 			 * the req
676 			 */
677 			list_del_init(&req->list);
678 			found = req;
679 			break;
680 		}
681 	}
682 	spin_unlock_bh(&lock);
683 
684 	if (!found)
685 		return;
686 
687 	/*
688 	 * sync canceling the work after removing it from the req_list
689 	 * guarentees no work is running and none will be started.
690 	 */
691 	cancel_delayed_work_sync(&found->work);
692 
693 	if (found->callback)
694 		found->callback(-ECANCELED, (struct sockaddr *)&found->src_addr,
695 			      found->addr, found->context);
696 
697 	kfree(found);
698 }
699 EXPORT_SYMBOL(rdma_addr_cancel);
700 
701 struct resolve_cb_context {
702 	struct completion comp;
703 	int status;
704 };
705 
706 static void resolve_cb(int status, struct sockaddr *src_addr,
707 	     struct rdma_dev_addr *addr, void *context)
708 {
709 	((struct resolve_cb_context *)context)->status = status;
710 	complete(&((struct resolve_cb_context *)context)->comp);
711 }
712 
713 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
714 				 const union ib_gid *dgid,
715 				 u8 *dmac, const struct net_device *ndev,
716 				 int *hoplimit)
717 {
718 	struct rdma_dev_addr dev_addr;
719 	struct resolve_cb_context ctx;
720 	union {
721 		struct sockaddr     _sockaddr;
722 		struct sockaddr_in  _sockaddr_in;
723 		struct sockaddr_in6 _sockaddr_in6;
724 	} sgid_addr, dgid_addr;
725 	int ret;
726 
727 	rdma_gid2ip(&sgid_addr._sockaddr, sgid);
728 	rdma_gid2ip(&dgid_addr._sockaddr, dgid);
729 
730 	memset(&dev_addr, 0, sizeof(dev_addr));
731 	dev_addr.bound_dev_if = ndev->ifindex;
732 	dev_addr.net = &init_net;
733 
734 	init_completion(&ctx.comp);
735 	ret = rdma_resolve_ip(&sgid_addr._sockaddr, &dgid_addr._sockaddr,
736 			      &dev_addr, 1000, resolve_cb, &ctx);
737 	if (ret)
738 		return ret;
739 
740 	wait_for_completion(&ctx.comp);
741 
742 	ret = ctx.status;
743 	if (ret)
744 		return ret;
745 
746 	memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
747 	*hoplimit = dev_addr.hoplimit;
748 	return 0;
749 }
750 
751 static int netevent_callback(struct notifier_block *self, unsigned long event,
752 	void *ctx)
753 {
754 	struct addr_req *req;
755 
756 	if (event == NETEVENT_NEIGH_UPDATE) {
757 		struct neighbour *neigh = ctx;
758 
759 		if (neigh->nud_state & NUD_VALID) {
760 			spin_lock_bh(&lock);
761 			list_for_each_entry(req, &req_list, list)
762 				set_timeout(req, jiffies);
763 			spin_unlock_bh(&lock);
764 		}
765 	}
766 	return 0;
767 }
768 
769 static struct notifier_block nb = {
770 	.notifier_call = netevent_callback
771 };
772 
773 int addr_init(void)
774 {
775 	addr_wq = alloc_ordered_workqueue("ib_addr", 0);
776 	if (!addr_wq)
777 		return -ENOMEM;
778 
779 	register_netevent_notifier(&nb);
780 
781 	return 0;
782 }
783 
784 void addr_cleanup(void)
785 {
786 	unregister_netevent_notifier(&nb);
787 	destroy_workqueue(addr_wq);
788 	WARN_ON(!list_empty(&req_list));
789 }
790