addr.c revision 324527 
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 <linux/notifier.h>
42#include <net/route.h>
43#include <net/netevent.h>
44#include <rdma/ib_addr.h>
45#include <netinet/if_ether.h>
46
47
48MODULE_AUTHOR("Sean Hefty");
49MODULE_DESCRIPTION("IB Address Translation");
50MODULE_LICENSE("Dual BSD/GPL");
51
52struct addr_req {
53	struct list_head list;
54	struct sockaddr_storage src_addr;
55	struct sockaddr_storage dst_addr;
56	struct rdma_dev_addr *addr;
57	struct rdma_addr_client *client;
58	void *context;
59	void (*callback)(int status, struct sockaddr *src_addr,
60			 struct rdma_dev_addr *addr, void *context);
61	unsigned long timeout;
62	int status;
63};
64
65static void process_req(struct work_struct *work);
66
67static DEFINE_MUTEX(lock);
68static LIST_HEAD(req_list);
69static struct delayed_work work;
70static struct workqueue_struct *addr_wq;
71
72void rdma_addr_register_client(struct rdma_addr_client *client)
73{
74	atomic_set(&client->refcount, 1);
75	init_completion(&client->comp);
76}
77EXPORT_SYMBOL(rdma_addr_register_client);
78
79static inline void put_client(struct rdma_addr_client *client)
80{
81	if (atomic_dec_and_test(&client->refcount))
82		complete(&client->comp);
83}
84
85void rdma_addr_unregister_client(struct rdma_addr_client *client)
86{
87	put_client(client);
88	wait_for_completion(&client->comp);
89}
90EXPORT_SYMBOL(rdma_addr_unregister_client);
91
92#ifdef __linux__
93int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
94		     const unsigned char *dst_dev_addr)
95{
96	dev_addr->dev_type = dev->type;
97	memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
98	memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
99	if (dst_dev_addr)
100		memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
101	dev_addr->bound_dev_if = dev->ifindex;
102	return 0;
103}
104#else
105int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct ifnet *dev,
106		     const unsigned char *dst_dev_addr)
107{
108	if (dev->if_type == IFT_INFINIBAND)
109		dev_addr->dev_type = ARPHRD_INFINIBAND;
110	else if (dev->if_type == IFT_ETHER)
111		dev_addr->dev_type = ARPHRD_ETHER;
112	else
113		dev_addr->dev_type = 0;
114	memcpy(dev_addr->src_dev_addr, IF_LLADDR(dev), dev->if_addrlen);
115	memcpy(dev_addr->broadcast, __DECONST(char *, dev->if_broadcastaddr),
116	    dev->if_addrlen);
117	if (dst_dev_addr)
118		memcpy(dev_addr->dst_dev_addr, dst_dev_addr, dev->if_addrlen);
119	dev_addr->bound_dev_if = dev->if_index;
120	return 0;
121}
122#endif
123EXPORT_SYMBOL(rdma_copy_addr);
124
125int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
126{
127	struct net_device *dev;
128	int ret = -EADDRNOTAVAIL;
129
130	if (dev_addr->bound_dev_if) {
131		dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
132		if (!dev)
133			return -ENODEV;
134		ret = rdma_copy_addr(dev_addr, dev, NULL);
135		dev_put(dev);
136		return ret;
137	}
138
139	switch (addr->sa_family) {
140#ifdef INET
141	case AF_INET:
142		dev = ip_dev_find(NULL,
143			((struct sockaddr_in *) addr)->sin_addr.s_addr);
144
145		if (!dev)
146			return ret;
147
148		ret = rdma_copy_addr(dev_addr, dev, NULL);
149		dev_put(dev);
150		break;
151#endif
152
153#if defined(INET6)
154	case AF_INET6:
155#ifdef __linux__
156		read_lock(&dev_base_lock);
157		for_each_netdev(&init_net, dev) {
158			if (ipv6_chk_addr(&init_net,
159					  &((struct sockaddr_in6 *) addr)->sin6_addr,
160					  dev, 1)) {
161				ret = rdma_copy_addr(dev_addr, dev, NULL);
162				break;
163			}
164		}
165		read_unlock(&dev_base_lock);
166#else
167		{
168			struct sockaddr_in6 *sin6;
169			struct ifaddr *ifa;
170			in_port_t port;
171
172			sin6 = (struct sockaddr_in6 *)addr;
173			port = sin6->sin6_port;
174			sin6->sin6_port = 0;
175			CURVNET_SET_QUIET(&init_net);
176			ifa = ifa_ifwithaddr(addr);
177			CURVNET_RESTORE();
178			sin6->sin6_port = port;
179			if (ifa == NULL) {
180				ret = -ENODEV;
181				break;
182			}
183			ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
184			ifa_free(ifa);
185			break;
186		}
187#endif
188		break;
189#endif
190	}
191	return ret;
192}
193EXPORT_SYMBOL(rdma_translate_ip);
194
195static void set_timeout(unsigned long time)
196{
197	int delay;	/* under FreeBSD ticks are 32-bit */
198
199	delay = time - jiffies;
200	if (delay <= 0)
201		delay = 1;
202
203	mod_delayed_work(addr_wq, &work, delay);
204}
205
206static void queue_req(struct addr_req *req)
207{
208	struct addr_req *temp_req;
209
210	mutex_lock(&lock);
211	list_for_each_entry_reverse(temp_req, &req_list, list) {
212		if (time_after_eq(req->timeout, temp_req->timeout))
213			break;
214	}
215
216	list_add(&req->list, &temp_req->list);
217
218	if (req_list.next == &req->list)
219		set_timeout(req->timeout);
220	mutex_unlock(&lock);
221}
222
223#ifdef __linux__
224static int addr4_resolve(struct sockaddr_in *src_in,
225			 struct sockaddr_in *dst_in,
226			 struct rdma_dev_addr *addr)
227{
228	__be32 src_ip = src_in->sin_addr.s_addr;
229	__be32 dst_ip = dst_in->sin_addr.s_addr;
230	struct flowi fl;
231	struct rtable *rt;
232	struct neighbour *neigh;
233	int ret;
234
235	memset(&fl, 0, sizeof fl);
236	fl.nl_u.ip4_u.daddr = dst_ip;
237	fl.nl_u.ip4_u.saddr = src_ip;
238	fl.oif = addr->bound_dev_if;
239
240	ret = ip_route_output_key(&init_net, &rt, &fl);
241	if (ret)
242		goto out;
243
244	src_in->sin_family = AF_INET;
245	src_in->sin_addr.s_addr = rt->rt_src;
246
247	if (rt->idev->dev->flags & IFF_LOOPBACK) {
248		ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
249		if (!ret)
250			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
251		goto put;
252	}
253
254	/* If the device does ARP internally, return 'done' */
255	if (rt->idev->dev->flags & IFF_NOARP) {
256		rdma_copy_addr(addr, rt->idev->dev, NULL);
257		goto put;
258	}
259
260	neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
261	if (!neigh || !(neigh->nud_state & NUD_VALID)) {
262		neigh_event_send(rt->u.dst.neighbour, NULL);
263		ret = -ENODATA;
264		if (neigh)
265			goto release;
266		goto put;
267	}
268
269	ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
270release:
271	neigh_release(neigh);
272put:
273	ip_rt_put(rt);
274out:
275	return ret;
276}
277
278#if defined(INET6)
279static int addr6_resolve(struct sockaddr_in6 *src_in,
280			 struct sockaddr_in6 *dst_in,
281			 struct rdma_dev_addr *addr)
282{
283	struct flowi fl;
284	struct neighbour *neigh;
285	struct dst_entry *dst;
286	int ret;
287
288	memset(&fl, 0, sizeof fl);
289	ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
290	ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
291	fl.oif = addr->bound_dev_if;
292
293	dst = ip6_route_output(&init_net, NULL, &fl);
294	if ((ret = dst->error))
295		goto put;
296
297	if (ipv6_addr_any(&fl.fl6_src)) {
298		ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
299					 &fl.fl6_dst, 0, &fl.fl6_src);
300		if (ret)
301			goto put;
302
303		src_in->sin6_family = AF_INET6;
304		ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
305	}
306
307	if (dst->dev->flags & IFF_LOOPBACK) {
308		ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
309		if (!ret)
310			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
311		goto put;
312	}
313
314	/* If the device does ARP internally, return 'done' */
315	if (dst->dev->flags & IFF_NOARP) {
316		ret = rdma_copy_addr(addr, dst->dev, NULL);
317		goto put;
318	}
319
320	neigh = dst->neighbour;
321	if (!neigh || !(neigh->nud_state & NUD_VALID)) {
322		neigh_event_send(dst->neighbour, NULL);
323		ret = -ENODATA;
324		goto put;
325	}
326
327	ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
328put:
329	dst_release(dst);
330	return ret;
331}
332#else
333static int addr6_resolve(struct sockaddr_in6 *src_in,
334			 struct sockaddr_in6 *dst_in,
335			 struct rdma_dev_addr *addr)
336{
337	return -EADDRNOTAVAIL;
338}
339#endif
340
341#else
342#include <netinet/if_ether.h>
343
344static int addr_resolve_sub(struct sockaddr *src_in,
345			struct sockaddr *dst_in,
346			struct rdma_dev_addr *addr)
347{
348	struct sockaddr_in *sin;
349	struct sockaddr_in6 *sin6;
350	struct ifaddr *ifa;
351	struct ifnet *ifp;
352#if defined(INET) || defined(INET6)
353	struct llentry *lle;
354#endif
355	struct rtentry *rte;
356	in_port_t port;
357	u_char edst[MAX_ADDR_LEN];
358	int multi;
359	int bcast;
360	int error = 0;
361
362	/*
363	 * Determine whether the address is unicast, multicast, or broadcast
364	 * and whether the source interface is valid.
365	 */
366	multi = 0;
367	bcast = 0;
368	sin = NULL;
369	sin6 = NULL;
370	ifp = NULL;
371	rte = NULL;
372	switch (dst_in->sa_family) {
373#ifdef INET
374	case AF_INET:
375		sin = (struct sockaddr_in *)dst_in;
376		if (sin->sin_addr.s_addr == INADDR_BROADCAST)
377			bcast = 1;
378		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
379			multi = 1;
380		sin = (struct sockaddr_in *)src_in;
381		if (sin->sin_addr.s_addr != INADDR_ANY) {
382			/*
383			 * Address comparison fails if the port is set
384			 * cache it here to be restored later.
385			 */
386			port = sin->sin_port;
387			sin->sin_port = 0;
388			memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
389		} else
390			src_in = NULL;
391		break;
392#endif
393#ifdef INET6
394	case AF_INET6:
395		sin6 = (struct sockaddr_in6 *)dst_in;
396		if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
397			multi = 1;
398		sin6 = (struct sockaddr_in6 *)src_in;
399		if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
400			port = sin6->sin6_port;
401			sin6->sin6_port = 0;
402		} else
403			src_in = NULL;
404		break;
405#endif
406	default:
407		return -EINVAL;
408	}
409	/*
410	 * If we have a source address to use look it up first and verify
411	 * that it is a local interface.
412	 */
413	if (src_in) {
414		ifa = ifa_ifwithaddr(src_in);
415		if (sin)
416			sin->sin_port = port;
417		if (sin6)
418			sin6->sin6_port = port;
419		if (ifa == NULL)
420			return -ENETUNREACH;
421		ifp = ifa->ifa_ifp;
422		ifa_free(ifa);
423		if (bcast || multi)
424			goto mcast;
425	}
426	/*
427	 * Make sure the route exists and has a valid link.
428	 */
429	rte = rtalloc1(dst_in, 1, 0);
430	if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
431		if (rte)
432			RTFREE_LOCKED(rte);
433		return -EHOSTUNREACH;
434	}
435	/*
436	 * If it's not multicast or broadcast and the route doesn't match the
437	 * requested interface return unreachable.  Otherwise fetch the
438	 * correct interface pointer and unlock the route.
439	 */
440	if (multi || bcast) {
441		if (ifp == NULL)
442			ifp = rte->rt_ifp;
443		RTFREE_LOCKED(rte);
444	} else if (ifp && ifp != rte->rt_ifp) {
445		RTFREE_LOCKED(rte);
446		return -ENETUNREACH;
447	} else {
448		if (ifp == NULL)
449			ifp = rte->rt_ifp;
450		RT_UNLOCK(rte);
451	}
452mcast:
453	if (bcast)
454		return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
455	if (multi) {
456		struct sockaddr *llsa;
457
458		error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
459		if (error)
460			return -error;
461		error = rdma_copy_addr(addr, ifp,
462		    LLADDR((struct sockaddr_dl *)llsa));
463		free(llsa, M_IFMADDR);
464		return error;
465	}
466	/*
467	 * Resolve the link local address.
468	 */
469	switch (dst_in->sa_family) {
470#ifdef INET
471	case AF_INET:
472		error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
473		break;
474#endif
475#ifdef INET6
476	case AF_INET6:
477		error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
478		break;
479#endif
480	default:
481		/* XXX: Shouldn't happen. */
482		error = -EINVAL;
483	}
484	RTFREE(rte);
485	if (error == 0)
486		return rdma_copy_addr(addr, ifp, edst);
487	if (error == EWOULDBLOCK)
488		return -ENODATA;
489	return -error;
490}
491
492static int addr_resolve(struct sockaddr *src_in,
493			struct sockaddr *dst_in,
494			struct rdma_dev_addr *addr)
495{
496	int error;
497
498	CURVNET_SET_QUIET(&init_net);
499	error = addr_resolve_sub(src_in, dst_in, addr);
500	CURVNET_RESTORE();
501
502	return (error);
503}
504#endif
505
506static void process_req(struct work_struct *work)
507{
508	struct addr_req *req, *temp_req;
509	struct sockaddr *src_in, *dst_in;
510	struct list_head done_list;
511
512	INIT_LIST_HEAD(&done_list);
513
514	mutex_lock(&lock);
515	list_for_each_entry_safe(req, temp_req, &req_list, list) {
516		if (req->status == -ENODATA) {
517			src_in = (struct sockaddr *) &req->src_addr;
518			dst_in = (struct sockaddr *) &req->dst_addr;
519			req->status = addr_resolve(src_in, dst_in, req->addr);
520			if (req->status && time_after_eq(jiffies, req->timeout))
521				req->status = -ETIMEDOUT;
522			else if (req->status == -ENODATA)
523				continue;
524		}
525		list_move_tail(&req->list, &done_list);
526	}
527
528	if (!list_empty(&req_list)) {
529		req = list_entry(req_list.next, struct addr_req, list);
530		set_timeout(req->timeout);
531	}
532	mutex_unlock(&lock);
533
534	list_for_each_entry_safe(req, temp_req, &done_list, list) {
535		list_del(&req->list);
536		req->callback(req->status, (struct sockaddr *) &req->src_addr,
537			req->addr, req->context);
538		put_client(req->client);
539		kfree(req);
540	}
541}
542
543int rdma_resolve_ip(struct rdma_addr_client *client,
544		    struct sockaddr *src_addr, struct sockaddr *dst_addr,
545		    struct rdma_dev_addr *addr, int timeout_ms,
546		    void (*callback)(int status, struct sockaddr *src_addr,
547				     struct rdma_dev_addr *addr, void *context),
548		    void *context)
549{
550	struct sockaddr *src_in, *dst_in;
551	struct addr_req *req;
552	int ret = 0;
553
554	req = kzalloc(sizeof *req, GFP_KERNEL);
555	if (!req)
556		return -ENOMEM;
557
558	src_in = (struct sockaddr *) &req->src_addr;
559	dst_in = (struct sockaddr *) &req->dst_addr;
560
561	if (src_addr) {
562		if (src_addr->sa_family != dst_addr->sa_family) {
563			ret = -EINVAL;
564			goto err;
565		}
566
567		memcpy(src_in, src_addr, ip_addr_size(src_addr));
568	} else {
569		src_in->sa_family = dst_addr->sa_family;
570	}
571
572	memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
573	req->addr = addr;
574	req->callback = callback;
575	req->context = context;
576	req->client = client;
577	atomic_inc(&client->refcount);
578
579	req->status = addr_resolve(src_in, dst_in, addr);
580	switch (req->status) {
581	case 0:
582		req->timeout = jiffies;
583		queue_req(req);
584		break;
585	case -ENODATA:
586		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
587		queue_req(req);
588		break;
589	default:
590		ret = req->status;
591		atomic_dec(&client->refcount);
592		goto err;
593	}
594	return ret;
595err:
596	kfree(req);
597	return ret;
598}
599EXPORT_SYMBOL(rdma_resolve_ip);
600
601void rdma_addr_cancel(struct rdma_dev_addr *addr)
602{
603	struct addr_req *req, *temp_req;
604
605	mutex_lock(&lock);
606	list_for_each_entry_safe(req, temp_req, &req_list, list) {
607		if (req->addr == addr) {
608			req->status = -ECANCELED;
609			req->timeout = jiffies;
610			list_move(&req->list, &req_list);
611			set_timeout(req->timeout);
612			break;
613		}
614	}
615	mutex_unlock(&lock);
616}
617EXPORT_SYMBOL(rdma_addr_cancel);
618
619static int netevent_callback(struct notifier_block *self, unsigned long event,
620	void *ctx)
621{
622	if (event == NETEVENT_NEIGH_UPDATE) {
623#ifdef __linux__
624		struct neighbour *neigh = ctx;
625
626		if (neigh->nud_state & NUD_VALID) {
627			set_timeout(jiffies);
628		}
629#else
630		set_timeout(jiffies);
631#endif
632	}
633	return 0;
634}
635
636static struct notifier_block nb = {
637	.notifier_call = netevent_callback
638};
639
640static int __init addr_init(void)
641{
642	INIT_DELAYED_WORK(&work, process_req);
643	addr_wq = create_singlethread_workqueue("ib_addr");
644	if (!addr_wq)
645		return -ENOMEM;
646
647	register_netevent_notifier(&nb);
648	return 0;
649}
650
651static void __exit addr_cleanup(void)
652{
653	unregister_netevent_notifier(&nb);
654	destroy_workqueue(addr_wq);
655}
656
657module_init(addr_init);
658module_exit(addr_cleanup);
659