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			ifa = ifa_ifwithaddr(addr);
176			sin6->sin6_port = port;
177			if (ifa == NULL) {
178				ret = -ENODEV;
179				break;
180			}
181			ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
182			ifa_free(ifa);
183			break;
184		}
185#endif
186		break;
187#endif
188	}
189	return ret;
190}
191EXPORT_SYMBOL(rdma_translate_ip);
192
193static void set_timeout(unsigned long time)
194{
195	unsigned long delay;
196
197	delay = time - jiffies;
198	if ((long)delay <= 0)
199		delay = 1;
200
201	mod_delayed_work(addr_wq, &work, delay);
202}
203
204static void queue_req(struct addr_req *req)
205{
206	struct addr_req *temp_req;
207
208	mutex_lock(&lock);
209	list_for_each_entry_reverse(temp_req, &req_list, list) {
210		if (time_after_eq(req->timeout, temp_req->timeout))
211			break;
212	}
213
214	list_add(&req->list, &temp_req->list);
215
216	if (req_list.next == &req->list)
217		set_timeout(req->timeout);
218	mutex_unlock(&lock);
219}
220
221#ifdef __linux__
222static int addr4_resolve(struct sockaddr_in *src_in,
223			 struct sockaddr_in *dst_in,
224			 struct rdma_dev_addr *addr)
225{
226	__be32 src_ip = src_in->sin_addr.s_addr;
227	__be32 dst_ip = dst_in->sin_addr.s_addr;
228	struct flowi fl;
229	struct rtable *rt;
230	struct neighbour *neigh;
231	int ret;
232
233	memset(&fl, 0, sizeof fl);
234	fl.nl_u.ip4_u.daddr = dst_ip;
235	fl.nl_u.ip4_u.saddr = src_ip;
236	fl.oif = addr->bound_dev_if;
237
238	ret = ip_route_output_key(&init_net, &rt, &fl);
239	if (ret)
240		goto out;
241
242	src_in->sin_family = AF_INET;
243	src_in->sin_addr.s_addr = rt->rt_src;
244
245	if (rt->idev->dev->flags & IFF_LOOPBACK) {
246		ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
247		if (!ret)
248			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
249		goto put;
250	}
251
252	/* If the device does ARP internally, return 'done' */
253	if (rt->idev->dev->flags & IFF_NOARP) {
254		rdma_copy_addr(addr, rt->idev->dev, NULL);
255		goto put;
256	}
257
258	neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
259	if (!neigh || !(neigh->nud_state & NUD_VALID)) {
260		neigh_event_send(rt->u.dst.neighbour, NULL);
261		ret = -ENODATA;
262		if (neigh)
263			goto release;
264		goto put;
265	}
266
267	ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
268release:
269	neigh_release(neigh);
270put:
271	ip_rt_put(rt);
272out:
273	return ret;
274}
275
276#if defined(INET6)
277static int addr6_resolve(struct sockaddr_in6 *src_in,
278			 struct sockaddr_in6 *dst_in,
279			 struct rdma_dev_addr *addr)
280{
281	struct flowi fl;
282	struct neighbour *neigh;
283	struct dst_entry *dst;
284	int ret;
285
286	memset(&fl, 0, sizeof fl);
287	ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
288	ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
289	fl.oif = addr->bound_dev_if;
290
291	dst = ip6_route_output(&init_net, NULL, &fl);
292	if ((ret = dst->error))
293		goto put;
294
295	if (ipv6_addr_any(&fl.fl6_src)) {
296		ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
297					 &fl.fl6_dst, 0, &fl.fl6_src);
298		if (ret)
299			goto put;
300
301		src_in->sin6_family = AF_INET6;
302		ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
303	}
304
305	if (dst->dev->flags & IFF_LOOPBACK) {
306		ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
307		if (!ret)
308			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
309		goto put;
310	}
311
312	/* If the device does ARP internally, return 'done' */
313	if (dst->dev->flags & IFF_NOARP) {
314		ret = rdma_copy_addr(addr, dst->dev, NULL);
315		goto put;
316	}
317
318	neigh = dst->neighbour;
319	if (!neigh || !(neigh->nud_state & NUD_VALID)) {
320		neigh_event_send(dst->neighbour, NULL);
321		ret = -ENODATA;
322		goto put;
323	}
324
325	ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
326put:
327	dst_release(dst);
328	return ret;
329}
330#else
331static int addr6_resolve(struct sockaddr_in6 *src_in,
332			 struct sockaddr_in6 *dst_in,
333			 struct rdma_dev_addr *addr)
334{
335	return -EADDRNOTAVAIL;
336}
337#endif
338
339#else
340#include <netinet/if_ether.h>
341
342static int addr_resolve(struct sockaddr *src_in,
343			struct sockaddr *dst_in,
344			struct rdma_dev_addr *addr)
345{
346	struct sockaddr_in *sin;
347	struct sockaddr_in6 *sin6;
348	struct ifaddr *ifa;
349	struct ifnet *ifp;
350#if defined(INET) || defined(INET6)
351	struct llentry *lle;
352#endif
353	struct rtentry *rte;
354	in_port_t port;
355	u_char edst[MAX_ADDR_LEN];
356	int multi;
357	int bcast;
358	int error = 0;
359
360	/*
361	 * Determine whether the address is unicast, multicast, or broadcast
362	 * and whether the source interface is valid.
363	 */
364	multi = 0;
365	bcast = 0;
366	sin = NULL;
367	sin6 = NULL;
368	ifp = NULL;
369	rte = NULL;
370	switch (dst_in->sa_family) {
371#ifdef INET
372	case AF_INET:
373		sin = (struct sockaddr_in *)dst_in;
374		if (sin->sin_addr.s_addr == INADDR_BROADCAST)
375			bcast = 1;
376		if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
377			multi = 1;
378		sin = (struct sockaddr_in *)src_in;
379		if (sin->sin_addr.s_addr != INADDR_ANY) {
380			/*
381			 * Address comparison fails if the port is set
382			 * cache it here to be restored later.
383			 */
384			port = sin->sin_port;
385			sin->sin_port = 0;
386			memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
387		} else
388			src_in = NULL;
389		break;
390#endif
391#ifdef INET6
392	case AF_INET6:
393		sin6 = (struct sockaddr_in6 *)dst_in;
394		if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
395			multi = 1;
396		sin6 = (struct sockaddr_in6 *)src_in;
397		if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
398			port = sin6->sin6_port;
399			sin6->sin6_port = 0;
400		} else
401			src_in = NULL;
402		break;
403#endif
404	default:
405		return -EINVAL;
406	}
407	/*
408	 * If we have a source address to use look it up first and verify
409	 * that it is a local interface.
410	 */
411	if (src_in) {
412		ifa = ifa_ifwithaddr(src_in);
413		if (sin)
414			sin->sin_port = port;
415		if (sin6)
416			sin6->sin6_port = port;
417		if (ifa == NULL)
418			return -ENETUNREACH;
419		ifp = ifa->ifa_ifp;
420		ifa_free(ifa);
421		if (bcast || multi)
422			goto mcast;
423	}
424	/*
425	 * Make sure the route exists and has a valid link.
426	 */
427	rte = rtalloc1(dst_in, 1, 0);
428	if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
429		if (rte)
430			RTFREE_LOCKED(rte);
431		return -EHOSTUNREACH;
432	}
433	/*
434	 * If it's not multicast or broadcast and the route doesn't match the
435	 * requested interface return unreachable.  Otherwise fetch the
436	 * correct interface pointer and unlock the route.
437	 */
438	if (multi || bcast) {
439		if (ifp == NULL)
440			ifp = rte->rt_ifp;
441		RTFREE_LOCKED(rte);
442	} else if (ifp && ifp != rte->rt_ifp) {
443		RTFREE_LOCKED(rte);
444		return -ENETUNREACH;
445	} else {
446		if (ifp == NULL)
447			ifp = rte->rt_ifp;
448		RT_UNLOCK(rte);
449	}
450mcast:
451	if (bcast)
452		return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
453	if (multi) {
454		struct sockaddr *llsa;
455
456		error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
457		if (error)
458			return -error;
459		error = rdma_copy_addr(addr, ifp,
460		    LLADDR((struct sockaddr_dl *)llsa));
461		free(llsa, M_IFMADDR);
462		return error;
463	}
464	/*
465	 * Resolve the link local address.
466	 */
467	switch (dst_in->sa_family) {
468#ifdef INET
469	case AF_INET:
470		error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
471		break;
472#endif
473#ifdef INET6
474	case AF_INET6:
475		error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
476		break;
477#endif
478	default:
479		/* XXX: Shouldn't happen. */
480		error = -EINVAL;
481	}
482	RTFREE(rte);
483	if (error == 0)
484		return rdma_copy_addr(addr, ifp, edst);
485	if (error == EWOULDBLOCK)
486		return -ENODATA;
487	return -error;
488}
489
490#endif
491
492static void process_req(struct work_struct *work)
493{
494	struct addr_req *req, *temp_req;
495	struct sockaddr *src_in, *dst_in;
496	struct list_head done_list;
497
498	INIT_LIST_HEAD(&done_list);
499
500	mutex_lock(&lock);
501	list_for_each_entry_safe(req, temp_req, &req_list, list) {
502		if (req->status == -ENODATA) {
503			src_in = (struct sockaddr *) &req->src_addr;
504			dst_in = (struct sockaddr *) &req->dst_addr;
505			req->status = addr_resolve(src_in, dst_in, req->addr);
506			if (req->status && time_after_eq(jiffies, req->timeout))
507				req->status = -ETIMEDOUT;
508			else if (req->status == -ENODATA)
509				continue;
510		}
511		list_move_tail(&req->list, &done_list);
512	}
513
514	if (!list_empty(&req_list)) {
515		req = list_entry(req_list.next, struct addr_req, list);
516		set_timeout(req->timeout);
517	}
518	mutex_unlock(&lock);
519
520	list_for_each_entry_safe(req, temp_req, &done_list, list) {
521		list_del(&req->list);
522		req->callback(req->status, (struct sockaddr *) &req->src_addr,
523			req->addr, req->context);
524		put_client(req->client);
525		kfree(req);
526	}
527}
528
529int rdma_resolve_ip(struct rdma_addr_client *client,
530		    struct sockaddr *src_addr, struct sockaddr *dst_addr,
531		    struct rdma_dev_addr *addr, int timeout_ms,
532		    void (*callback)(int status, struct sockaddr *src_addr,
533				     struct rdma_dev_addr *addr, void *context),
534		    void *context)
535{
536	struct sockaddr *src_in, *dst_in;
537	struct addr_req *req;
538	int ret = 0;
539
540	req = kzalloc(sizeof *req, GFP_KERNEL);
541	if (!req)
542		return -ENOMEM;
543
544	src_in = (struct sockaddr *) &req->src_addr;
545	dst_in = (struct sockaddr *) &req->dst_addr;
546
547	if (src_addr) {
548		if (src_addr->sa_family != dst_addr->sa_family) {
549			ret = -EINVAL;
550			goto err;
551		}
552
553		memcpy(src_in, src_addr, ip_addr_size(src_addr));
554	} else {
555		src_in->sa_family = dst_addr->sa_family;
556	}
557
558	memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
559	req->addr = addr;
560	req->callback = callback;
561	req->context = context;
562	req->client = client;
563	atomic_inc(&client->refcount);
564
565	req->status = addr_resolve(src_in, dst_in, addr);
566	switch (req->status) {
567	case 0:
568		req->timeout = jiffies;
569		queue_req(req);
570		break;
571	case -ENODATA:
572		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
573		queue_req(req);
574		break;
575	default:
576		ret = req->status;
577		atomic_dec(&client->refcount);
578		goto err;
579	}
580	return ret;
581err:
582	kfree(req);
583	return ret;
584}
585EXPORT_SYMBOL(rdma_resolve_ip);
586
587void rdma_addr_cancel(struct rdma_dev_addr *addr)
588{
589	struct addr_req *req, *temp_req;
590
591	mutex_lock(&lock);
592	list_for_each_entry_safe(req, temp_req, &req_list, list) {
593		if (req->addr == addr) {
594			req->status = -ECANCELED;
595			req->timeout = jiffies;
596			list_move(&req->list, &req_list);
597			set_timeout(req->timeout);
598			break;
599		}
600	}
601	mutex_unlock(&lock);
602}
603EXPORT_SYMBOL(rdma_addr_cancel);
604
605static int netevent_callback(struct notifier_block *self, unsigned long event,
606	void *ctx)
607{
608	if (event == NETEVENT_NEIGH_UPDATE) {
609#ifdef __linux__
610		struct neighbour *neigh = ctx;
611
612		if (neigh->nud_state & NUD_VALID) {
613			set_timeout(jiffies);
614		}
615#else
616		set_timeout(jiffies);
617#endif
618	}
619	return 0;
620}
621
622static struct notifier_block nb = {
623	.notifier_call = netevent_callback
624};
625
626static int __init addr_init(void)
627{
628	INIT_DELAYED_WORK(&work, process_req);
629	addr_wq = create_singlethread_workqueue("ib_addr");
630	if (!addr_wq)
631		return -ENOMEM;
632
633	register_netevent_notifier(&nb);
634	return 0;
635}
636
637static void __exit addr_cleanup(void)
638{
639	unregister_netevent_notifier(&nb);
640	destroy_workqueue(addr_wq);
641}
642
643module_init(addr_init);
644module_exit(addr_cleanup);
645