addr.c

  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 <net/arp.h>
 41#include <net/neighbour.h>
 42#include <net/route.h>
 43#include <net/netevent.h>
 44#include <net/ipv6_stubs.h>
 45#include <net/ip6_route.h>
 46#include <rdma/ib_addr.h>
 47#include <rdma/ib_cache.h>
 48#include <rdma/ib_sa.h>
 49#include <rdma/ib.h>
 50#include <rdma/rdma_netlink.h>
 51#include <net/netlink.h>
 52
 53#include "core_priv.h"
 54
 55struct addr_req {
 56	struct list_head list;
 57	struct sockaddr_storage src_addr;
 58	struct sockaddr_storage dst_addr;
 59	struct rdma_dev_addr *addr;
 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	bool resolve_by_gid_attr;	/* Consider gid attr in resolve phase */
 66	int status;
 67	u32 seq;
 68};
 69
 70static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
 71
 72static DEFINE_SPINLOCK(lock);
 73static LIST_HEAD(req_list);
 74static struct workqueue_struct *addr_wq;
 75
 76static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
 77	[LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
 78		.len = sizeof(struct rdma_nla_ls_gid),
 79		.validation_type = NLA_VALIDATE_MIN,
 80		.min = sizeof(struct rdma_nla_ls_gid)},
 81};
 82
 83static void ib_nl_process_ip_rsep(const struct nlmsghdr *nlh)
 84{
 85	struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
 86	union ib_gid gid;
 87	struct addr_req *req;
 88	int found = 0;
 89	int ret;
 90
 91	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
 92		return;
 93
 94	ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
 95				   nlmsg_len(nlh), ib_nl_addr_policy, NULL);
 96	if (ret)
 97		return;
 98
 99	if (!tb[LS_NLA_TYPE_DGID])
100		return;
101	memcpy(&gid, nla_data(tb[LS_NLA_TYPE_DGID]), sizeof(gid));
102
103	spin_lock_bh(&lock);
104	list_for_each_entry(req, &req_list, list) {
105		if (nlh->nlmsg_seq != req->seq)
106			continue;
107		/* We set the DGID part, the rest was set earlier */
108		rdma_addr_set_dgid(req->addr, &gid);
109		req->status = 0;
110		found = 1;
111		break;
112	}
113	spin_unlock_bh(&lock);
114
115	if (!found)
116		pr_info("Couldn't find request waiting for DGID: %pI6\n",
117			&gid);
118}
119
120int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
121			     struct nlmsghdr *nlh,
122			     struct netlink_ext_ack *extack)
123{
124	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
125	    !(NETLINK_CB(skb).sk))
126		return -EPERM;
127
128	ib_nl_process_ip_rsep(nlh);
129
130	return 0;
131}
132
133static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
134			     const void *daddr,
135			     u32 seq, u16 family)
136{
137	struct sk_buff *skb = NULL;
138	struct nlmsghdr *nlh;
139	struct rdma_ls_ip_resolve_header *header;
140	void *data;
141	size_t size;
142	int attrtype;
143	int len;
144
145	if (family == AF_INET) {
146		size = sizeof(struct in_addr);
147		attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
148	} else {
149		size = sizeof(struct in6_addr);
150		attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
151	}
152
153	len = nla_total_size(sizeof(size));
154	len += NLMSG_ALIGN(sizeof(*header));
155
156	skb = nlmsg_new(len, GFP_KERNEL);
157	if (!skb)
158		return -ENOMEM;
159
160	data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
161			    RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
162	if (!data) {
163		nlmsg_free(skb);
164		return -ENODATA;
165	}
166
167	/* Construct the family header first */
168	header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
169	header->ifindex = dev_addr->bound_dev_if;
170	nla_put(skb, attrtype, size, daddr);
171
172	/* Repair the nlmsg header length */
173	nlmsg_end(skb, nlh);
174	rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
175
176	/* Make the request retry, so when we get the response from userspace
177	 * we will have something.
178	 */
179	return -ENODATA;
180}
181
182int rdma_addr_size(const struct sockaddr *addr)
183{
184	switch (addr->sa_family) {
185	case AF_INET:
186		return sizeof(struct sockaddr_in);
187	case AF_INET6:
188		return sizeof(struct sockaddr_in6);
189	case AF_IB:
190		return sizeof(struct sockaddr_ib);
191	default:
192		return 0;
193	}
194}
195EXPORT_SYMBOL(rdma_addr_size);
196
197int rdma_addr_size_in6(struct sockaddr_in6 *addr)
198{
199	int ret = rdma_addr_size((struct sockaddr *) addr);
200
201	return ret <= sizeof(*addr) ? ret : 0;
202}
203EXPORT_SYMBOL(rdma_addr_size_in6);
204
205int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
206{
207	int ret = rdma_addr_size((struct sockaddr *) addr);
208
209	return ret <= sizeof(*addr) ? ret : 0;
210}
211EXPORT_SYMBOL(rdma_addr_size_kss);
212
213/**
214 * rdma_copy_src_l2_addr - Copy netdevice source addresses
215 * @dev_addr:	Destination address pointer where to copy the addresses
216 * @dev:	Netdevice whose source addresses to copy
217 *
218 * rdma_copy_src_l2_addr() copies source addresses from the specified netdevice.
219 * This includes unicast address, broadcast address, device type and
220 * interface index.
221 */
222void rdma_copy_src_l2_addr(struct rdma_dev_addr *dev_addr,
223			   const struct net_device *dev)
224{
225	dev_addr->dev_type = dev->type;
226	memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
227	memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
228	dev_addr->bound_dev_if = dev->ifindex;
229}
230EXPORT_SYMBOL(rdma_copy_src_l2_addr);
231
232static struct net_device *
233rdma_find_ndev_for_src_ip_rcu(struct net *net, const struct sockaddr *src_in)
234{
235	struct net_device *dev = NULL;
236	int ret = -EADDRNOTAVAIL;
237
238	switch (src_in->sa_family) {
239	case AF_INET:
240		dev = __ip_dev_find(net,
241				    ((const struct sockaddr_in *)src_in)->sin_addr.s_addr,
242				    false);
243		if (dev)
244			ret = 0;
245		break;
246#if IS_ENABLED(CONFIG_IPV6)
247	case AF_INET6:
248		for_each_netdev_rcu(net, dev) {
249			if (ipv6_chk_addr(net,
250					  &((const struct sockaddr_in6 *)src_in)->sin6_addr,
251					  dev, 1)) {
252				ret = 0;
253				break;
254			}
255		}
256		break;
257#endif
258	}
259	return ret ? ERR_PTR(ret) : dev;
260}
261
262int rdma_translate_ip(const struct sockaddr *addr,
263		      struct rdma_dev_addr *dev_addr)
264{
265	struct net_device *dev;
266
267	if (dev_addr->bound_dev_if) {
268		dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
269		if (!dev)
270			return -ENODEV;
271		rdma_copy_src_l2_addr(dev_addr, dev);
272		dev_put(dev);
273		return 0;
274	}
275
276	rcu_read_lock();
277	dev = rdma_find_ndev_for_src_ip_rcu(dev_addr->net, addr);
278	if (!IS_ERR(dev))
279		rdma_copy_src_l2_addr(dev_addr, dev);
280	rcu_read_unlock();
281	return PTR_ERR_OR_ZERO(dev);
282}
283EXPORT_SYMBOL(rdma_translate_ip);
284
285static void set_timeout(struct addr_req *req, unsigned long time)
286{
287	unsigned long delay;
288
289	delay = time - jiffies;
290	if ((long)delay < 0)
291		delay = 0;
292
293	mod_delayed_work(addr_wq, &req->work, delay);
294}
295
296static void queue_req(struct addr_req *req)
297{
298	spin_lock_bh(&lock);
299	list_add_tail(&req->list, &req_list);
300	set_timeout(req, req->timeout);
301	spin_unlock_bh(&lock);
302}
303
304static int ib_nl_fetch_ha(struct rdma_dev_addr *dev_addr,
305			  const void *daddr, u32 seq, u16 family)
306{
307	if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
308		return -EADDRNOTAVAIL;
309
310	return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
311}
312
313static int dst_fetch_ha(const struct dst_entry *dst,
314			struct rdma_dev_addr *dev_addr,
315			const void *daddr)
316{
317	struct neighbour *n;
318	int ret = 0;
319
320	n = dst_neigh_lookup(dst, daddr);
321	if (!n)
322		return -ENODATA;
323
324	if (!(n->nud_state & NUD_VALID)) {
325		neigh_event_send(n, NULL);
326		ret = -ENODATA;
327	} else {
328		neigh_ha_snapshot(dev_addr->dst_dev_addr, n, dst->dev);
329	}
330
331	neigh_release(n);
332
333	return ret;
334}
335
336static bool has_gateway(const struct dst_entry *dst, sa_family_t family)
337{
338	if (family == AF_INET)
339		return dst_rtable(dst)->rt_uses_gateway;
340
341	return dst_rt6_info(dst)->rt6i_flags & RTF_GATEWAY;
342}
343
344static int fetch_ha(const struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
345		    const struct sockaddr *dst_in, u32 seq)
346{
347	const struct sockaddr_in *dst_in4 =
348		(const struct sockaddr_in *)dst_in;
349	const struct sockaddr_in6 *dst_in6 =
350		(const struct sockaddr_in6 *)dst_in;
351	const void *daddr = (dst_in->sa_family == AF_INET) ?
352		(const void *)&dst_in4->sin_addr.s_addr :
353		(const void *)&dst_in6->sin6_addr;
354	sa_family_t family = dst_in->sa_family;
355
356	might_sleep();
357
358	/* If we have a gateway in IB mode then it must be an IB network */
359	if (has_gateway(dst, family) && dev_addr->network == RDMA_NETWORK_IB)
360		return ib_nl_fetch_ha(dev_addr, daddr, seq, family);
361	else
362		return dst_fetch_ha(dst, dev_addr, daddr);
363}
364
365static int addr4_resolve(struct sockaddr *src_sock,
366			 const struct sockaddr *dst_sock,
367			 struct rdma_dev_addr *addr,
368			 struct rtable **prt)
369{
370	struct sockaddr_in *src_in = (struct sockaddr_in *)src_sock;
371	const struct sockaddr_in *dst_in =
372			(const struct sockaddr_in *)dst_sock;
373
374	__be32 src_ip = src_in->sin_addr.s_addr;
375	__be32 dst_ip = dst_in->sin_addr.s_addr;
376	struct rtable *rt;
377	struct flowi4 fl4;
378	int ret;
379
380	memset(&fl4, 0, sizeof(fl4));
381	fl4.daddr = dst_ip;
382	fl4.saddr = src_ip;
383	fl4.flowi4_oif = addr->bound_dev_if;
384	rt = ip_route_output_key(addr->net, &fl4);
385	ret = PTR_ERR_OR_ZERO(rt);
386	if (ret)
387		return ret;
388
389	src_in->sin_addr.s_addr = fl4.saddr;
390
391	addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
392
393	*prt = rt;
394	return 0;
395}
396
397#if IS_ENABLED(CONFIG_IPV6)
398static int addr6_resolve(struct sockaddr *src_sock,
399			 const struct sockaddr *dst_sock,
400			 struct rdma_dev_addr *addr,
401			 struct dst_entry **pdst)
402{
403	struct sockaddr_in6 *src_in = (struct sockaddr_in6 *)src_sock;
404	const struct sockaddr_in6 *dst_in =
405				(const struct sockaddr_in6 *)dst_sock;
406	struct flowi6 fl6;
407	struct dst_entry *dst;
408
409	memset(&fl6, 0, sizeof fl6);
410	fl6.daddr = dst_in->sin6_addr;
411	fl6.saddr = src_in->sin6_addr;
412	fl6.flowi6_oif = addr->bound_dev_if;
413
414	dst = ipv6_stub->ipv6_dst_lookup_flow(addr->net, NULL, &fl6, NULL);
415	if (IS_ERR(dst))
416		return PTR_ERR(dst);
417
418	if (ipv6_addr_any(&src_in->sin6_addr))
419		src_in->sin6_addr = fl6.saddr;
420
421	addr->hoplimit = ip6_dst_hoplimit(dst);
422
423	*pdst = dst;
424	return 0;
425}
426#else
427static int addr6_resolve(struct sockaddr *src_sock,
428			 const struct sockaddr *dst_sock,
429			 struct rdma_dev_addr *addr,
430			 struct dst_entry **pdst)
431{
432	return -EADDRNOTAVAIL;
433}
434#endif
435
436static bool is_dst_local(const struct dst_entry *dst)
437{
438	if (dst->ops->family == AF_INET)
439		return !!(dst_rtable(dst)->rt_type & RTN_LOCAL);
440	else if (dst->ops->family == AF_INET6)
441		return !!(dst_rt6_info(dst)->rt6i_flags & RTF_LOCAL);
442	else
443		return false;
444}
445
446static int addr_resolve_neigh(const struct dst_entry *dst,
447			      const struct sockaddr *dst_in,
448			      struct rdma_dev_addr *addr,
449			      u32 seq)
450{
451	if (is_dst_local(dst)) {
452		/* When the destination is local entry, source and destination
453		 * are same. Skip the neighbour lookup.
454		 */
455		memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
456		return 0;
457	}
458
459	return fetch_ha(dst, addr, dst_in, seq);
460}
461
462static int rdma_set_src_addr_rcu(struct rdma_dev_addr *dev_addr,
463				 const struct sockaddr *dst_in,
464				 const struct dst_entry *dst)
465{
466	struct net_device *ndev = READ_ONCE(dst->dev);
467
468	/* A physical device must be the RDMA device to use */
469	if (is_dst_local(dst)) {
470		int ret;
471		/*
472		 * RDMA (IB/RoCE, iWarp) doesn't run on lo interface or
473		 * loopback IP address. So if route is resolved to loopback
474		 * interface, translate that to a real ndev based on non
475		 * loopback IP address.
476		 */
477		ndev = rdma_find_ndev_for_src_ip_rcu(dev_net(ndev), dst_in);
478		if (IS_ERR(ndev))
479			return -ENODEV;
480		ret = rdma_translate_ip(dst_in, dev_addr);
481		if (ret)
482			return ret;
483	} else {
484		rdma_copy_src_l2_addr(dev_addr, dst->dev);
485	}
486
487	/*
488	 * If there's a gateway and type of device not ARPHRD_INFINIBAND,
489	 * we're definitely in RoCE v2 (as RoCE v1 isn't routable) set the
490	 * network type accordingly.
491	 */
492	if (has_gateway(dst, dst_in->sa_family) &&
493	    ndev->type != ARPHRD_INFINIBAND)
494		dev_addr->network = dst_in->sa_family == AF_INET ?
495						RDMA_NETWORK_IPV4 :
496						RDMA_NETWORK_IPV6;
497	else
498		dev_addr->network = RDMA_NETWORK_IB;
499
500	return 0;
501}
502
503static int set_addr_netns_by_gid_rcu(struct rdma_dev_addr *addr)
504{
505	struct net_device *ndev;
506
507	ndev = rdma_read_gid_attr_ndev_rcu(addr->sgid_attr);
508	if (IS_ERR(ndev))
509		return PTR_ERR(ndev);
510
511	/*
512	 * Since we are holding the rcu, reading net and ifindex
513	 * are safe without any additional reference; because
514	 * change_net_namespace() in net/core/dev.c does rcu sync
515	 * after it changes the state to IFF_DOWN and before
516	 * updating netdev fields {net, ifindex}.
517	 */
518	addr->net = dev_net(ndev);
519	addr->bound_dev_if = ndev->ifindex;
520	return 0;
521}
522
523static void rdma_addr_set_net_defaults(struct rdma_dev_addr *addr)
524{
525	addr->net = &init_net;
526	addr->bound_dev_if = 0;
527}
528
529static int addr_resolve(struct sockaddr *src_in,
530			const struct sockaddr *dst_in,
531			struct rdma_dev_addr *addr,
532			bool resolve_neigh,
533			bool resolve_by_gid_attr,
534			u32 seq)
535{
536	struct dst_entry *dst = NULL;
537	struct rtable *rt = NULL;
538	int ret;
539
540	if (!addr->net) {
541		pr_warn_ratelimited("%s: missing namespace\n", __func__);
542		return -EINVAL;
543	}
544
545	rcu_read_lock();
546	if (resolve_by_gid_attr) {
547		if (!addr->sgid_attr) {
548			rcu_read_unlock();
549			pr_warn_ratelimited("%s: missing gid_attr\n", __func__);
550			return -EINVAL;
551		}
552		/*
553		 * If the request is for a specific gid attribute of the
554		 * rdma_dev_addr, derive net from the netdevice of the
555		 * GID attribute.
556		 */
557		ret = set_addr_netns_by_gid_rcu(addr);
558		if (ret) {
559			rcu_read_unlock();
560			return ret;
561		}
562	}
563	if (src_in->sa_family == AF_INET) {
564		ret = addr4_resolve(src_in, dst_in, addr, &rt);
565		dst = &rt->dst;
566	} else {
567		ret = addr6_resolve(src_in, dst_in, addr, &dst);
568	}
569	if (ret) {
570		rcu_read_unlock();
571		goto done;
572	}
573	ret = rdma_set_src_addr_rcu(addr, dst_in, dst);
574	rcu_read_unlock();
575
576	/*
577	 * Resolve neighbor destination address if requested and
578	 * only if src addr translation didn't fail.
579	 */
580	if (!ret && resolve_neigh)
581		ret = addr_resolve_neigh(dst, dst_in, addr, seq);
582
583	if (src_in->sa_family == AF_INET)
584		ip_rt_put(rt);
585	else
586		dst_release(dst);
587done:
588	/*
589	 * Clear the addr net to go back to its original state, only if it was
590	 * derived from GID attribute in this context.
591	 */
592	if (resolve_by_gid_attr)
593		rdma_addr_set_net_defaults(addr);
594	return ret;
595}
596
597static void process_one_req(struct work_struct *_work)
598{
599	struct addr_req *req;
600	struct sockaddr *src_in, *dst_in;
601
602	req = container_of(_work, struct addr_req, work.work);
603
604	if (req->status == -ENODATA) {
605		src_in = (struct sockaddr *)&req->src_addr;
606		dst_in = (struct sockaddr *)&req->dst_addr;
607		req->status = addr_resolve(src_in, dst_in, req->addr,
608					   true, req->resolve_by_gid_attr,
609					   req->seq);
610		if (req->status && time_after_eq(jiffies, req->timeout)) {
611			req->status = -ETIMEDOUT;
612		} else if (req->status == -ENODATA) {
613			/* requeue the work for retrying again */
614			spin_lock_bh(&lock);
615			if (!list_empty(&req->list))
616				set_timeout(req, req->timeout);
617			spin_unlock_bh(&lock);
618			return;
619		}
620	}
621
622	req->callback(req->status, (struct sockaddr *)&req->src_addr,
623		req->addr, req->context);
624	req->callback = NULL;
625
626	spin_lock_bh(&lock);
627	/*
628	 * Although the work will normally have been canceled by the workqueue,
629	 * it can still be requeued as long as it is on the req_list.
630	 */
631	cancel_delayed_work(&req->work);
632	if (!list_empty(&req->list)) {
633		list_del_init(&req->list);
634		kfree(req);
635	}
636	spin_unlock_bh(&lock);
637}
638
639int rdma_resolve_ip(struct sockaddr *src_addr, const struct sockaddr *dst_addr,
640		    struct rdma_dev_addr *addr, unsigned long timeout_ms,
641		    void (*callback)(int status, struct sockaddr *src_addr,
642				     struct rdma_dev_addr *addr, void *context),
643		    bool resolve_by_gid_attr, void *context)
644{
645	struct sockaddr *src_in, *dst_in;
646	struct addr_req *req;
647	int ret = 0;
648
649	req = kzalloc(sizeof *req, GFP_KERNEL);
650	if (!req)
651		return -ENOMEM;
652
653	src_in = (struct sockaddr *) &req->src_addr;
654	dst_in = (struct sockaddr *) &req->dst_addr;
655
656	if (src_addr) {
657		if (src_addr->sa_family != dst_addr->sa_family) {
658			ret = -EINVAL;
659			goto err;
660		}
661
662		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
663	} else {
664		src_in->sa_family = dst_addr->sa_family;
665	}
666
667	memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
668	req->addr = addr;
669	req->callback = callback;
670	req->context = context;
671	req->resolve_by_gid_attr = resolve_by_gid_attr;
672	INIT_DELAYED_WORK(&req->work, process_one_req);
673	req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
674
675	req->status = addr_resolve(src_in, dst_in, addr, true,
676				   req->resolve_by_gid_attr, req->seq);
677	switch (req->status) {
678	case 0:
679		req->timeout = jiffies;
680		queue_req(req);
681		break;
682	case -ENODATA:
683		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
684		queue_req(req);
685		break;
686	default:
687		ret = req->status;
688		goto err;
689	}
690	return ret;
691err:
692	kfree(req);
693	return ret;
694}
695EXPORT_SYMBOL(rdma_resolve_ip);
696
697int roce_resolve_route_from_path(struct sa_path_rec *rec,
698				 const struct ib_gid_attr *attr)
699{
700	union {
701		struct sockaddr     _sockaddr;
702		struct sockaddr_in  _sockaddr_in;
703		struct sockaddr_in6 _sockaddr_in6;
704	} sgid, dgid;
705	struct rdma_dev_addr dev_addr = {};
706	int ret;
707
708	might_sleep();
709
710	if (rec->roce.route_resolved)
711		return 0;
712
713	rdma_gid2ip((struct sockaddr *)&sgid, &rec->sgid);
714	rdma_gid2ip((struct sockaddr *)&dgid, &rec->dgid);
715
716	if (sgid._sockaddr.sa_family != dgid._sockaddr.sa_family)
717		return -EINVAL;
718
719	if (!attr || !attr->ndev)
720		return -EINVAL;
721
722	dev_addr.net = &init_net;
723	dev_addr.sgid_attr = attr;
724
725	ret = addr_resolve((struct sockaddr *)&sgid, (struct sockaddr *)&dgid,
726			   &dev_addr, false, true, 0);
727	if (ret)
728		return ret;
729
730	if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
731	     dev_addr.network == RDMA_NETWORK_IPV6) &&
732	    rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
733		return -EINVAL;
734
735	rec->roce.route_resolved = true;
736	return 0;
737}
738
739/**
740 * rdma_addr_cancel - Cancel resolve ip request
741 * @addr:	Pointer to address structure given previously
742 *		during rdma_resolve_ip().
743 * rdma_addr_cancel() is synchronous function which cancels any pending
744 * request if there is any.
745 */
746void rdma_addr_cancel(struct rdma_dev_addr *addr)
747{
748	struct addr_req *req, *temp_req;
749	struct addr_req *found = NULL;
750
751	spin_lock_bh(&lock);
752	list_for_each_entry_safe(req, temp_req, &req_list, list) {
753		if (req->addr == addr) {
754			/*
755			 * Removing from the list means we take ownership of
756			 * the req
757			 */
758			list_del_init(&req->list);
759			found = req;
760			break;
761		}
762	}
763	spin_unlock_bh(&lock);
764
765	if (!found)
766		return;
767
768	/*
769	 * sync canceling the work after removing it from the req_list
770	 * guarentees no work is running and none will be started.
771	 */
772	cancel_delayed_work_sync(&found->work);
773	kfree(found);
774}
775EXPORT_SYMBOL(rdma_addr_cancel);
776
777struct resolve_cb_context {
778	struct completion comp;
779	int status;
780};
781
782static void resolve_cb(int status, struct sockaddr *src_addr,
783	     struct rdma_dev_addr *addr, void *context)
784{
785	((struct resolve_cb_context *)context)->status = status;
786	complete(&((struct resolve_cb_context *)context)->comp);
787}
788
789int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
790				 const union ib_gid *dgid,
791				 u8 *dmac, const struct ib_gid_attr *sgid_attr,
792				 int *hoplimit)
793{
794	struct rdma_dev_addr dev_addr;
795	struct resolve_cb_context ctx;
796	union {
797		struct sockaddr_in  _sockaddr_in;
798		struct sockaddr_in6 _sockaddr_in6;
799	} sgid_addr, dgid_addr;
800	int ret;
801
802	rdma_gid2ip((struct sockaddr *)&sgid_addr, sgid);
803	rdma_gid2ip((struct sockaddr *)&dgid_addr, dgid);
804
805	memset(&dev_addr, 0, sizeof(dev_addr));
806	dev_addr.net = &init_net;
807	dev_addr.sgid_attr = sgid_attr;
808
809	init_completion(&ctx.comp);
810	ret = rdma_resolve_ip((struct sockaddr *)&sgid_addr,
811			      (struct sockaddr *)&dgid_addr, &dev_addr, 1000,
812			      resolve_cb, true, &ctx);
813	if (ret)
814		return ret;
815
816	wait_for_completion(&ctx.comp);
817
818	ret = ctx.status;
819	if (ret)
820		return ret;
821
822	memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
823	*hoplimit = dev_addr.hoplimit;
824	return 0;
825}
826
827static int netevent_callback(struct notifier_block *self, unsigned long event,
828	void *ctx)
829{
830	struct addr_req *req;
831
832	if (event == NETEVENT_NEIGH_UPDATE) {
833		struct neighbour *neigh = ctx;
834
835		if (neigh->nud_state & NUD_VALID) {
836			spin_lock_bh(&lock);
837			list_for_each_entry(req, &req_list, list)
838				set_timeout(req, jiffies);
839			spin_unlock_bh(&lock);
840		}
841	}
842	return 0;
843}
844
845static struct notifier_block nb = {
846	.notifier_call = netevent_callback
847};
848
849int addr_init(void)
850{
851	addr_wq = alloc_ordered_workqueue("ib_addr", 0);
852	if (!addr_wq)
853		return -ENOMEM;
854
855	register_netevent_notifier(&nb);
856
857	return 0;
858}
859
860void addr_cleanup(void)
861{
862	unregister_netevent_notifier(&nb);
863	destroy_workqueue(addr_wq);
864	WARN_ON(!list_empty(&req_list));
865}