xref: /linux/drivers/infiniband/core/addr.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 
50 MODULE_AUTHOR("Sean Hefty");
51 MODULE_DESCRIPTION("IB Address Translation");
52 MODULE_LICENSE("Dual BSD/GPL");
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 	int status;
65 };
66 
67 static void process_req(struct work_struct *work);
68 
69 static DEFINE_MUTEX(lock);
70 static LIST_HEAD(req_list);
71 static DECLARE_DELAYED_WORK(work, process_req);
72 static struct workqueue_struct *addr_wq;
73 
74 int rdma_addr_size(struct sockaddr *addr)
75 {
76 	switch (addr->sa_family) {
77 	case AF_INET:
78 		return sizeof(struct sockaddr_in);
79 	case AF_INET6:
80 		return sizeof(struct sockaddr_in6);
81 	case AF_IB:
82 		return sizeof(struct sockaddr_ib);
83 	default:
84 		return 0;
85 	}
86 }
87 EXPORT_SYMBOL(rdma_addr_size);
88 
89 static struct rdma_addr_client self;
90 
91 void rdma_addr_register_client(struct rdma_addr_client *client)
92 {
93 	atomic_set(&client->refcount, 1);
94 	init_completion(&client->comp);
95 }
96 EXPORT_SYMBOL(rdma_addr_register_client);
97 
98 static inline void put_client(struct rdma_addr_client *client)
99 {
100 	if (atomic_dec_and_test(&client->refcount))
101 		complete(&client->comp);
102 }
103 
104 void rdma_addr_unregister_client(struct rdma_addr_client *client)
105 {
106 	put_client(client);
107 	wait_for_completion(&client->comp);
108 }
109 EXPORT_SYMBOL(rdma_addr_unregister_client);
110 
111 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
112 		     const unsigned char *dst_dev_addr)
113 {
114 	dev_addr->dev_type = dev->type;
115 	memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
116 	memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
117 	if (dst_dev_addr)
118 		memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
119 	dev_addr->bound_dev_if = dev->ifindex;
120 	return 0;
121 }
122 EXPORT_SYMBOL(rdma_copy_addr);
123 
124 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr,
125 		      u16 *vlan_id)
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 	case AF_INET:
141 		dev = ip_dev_find(&init_net,
142 			((struct sockaddr_in *) addr)->sin_addr.s_addr);
143 
144 		if (!dev)
145 			return ret;
146 
147 		ret = rdma_copy_addr(dev_addr, dev, NULL);
148 		if (vlan_id)
149 			*vlan_id = rdma_vlan_dev_vlan_id(dev);
150 		dev_put(dev);
151 		break;
152 
153 #if IS_ENABLED(CONFIG_IPV6)
154 	case AF_INET6:
155 		rcu_read_lock();
156 		for_each_netdev_rcu(&init_net, dev) {
157 			if (ipv6_chk_addr(&init_net,
158 					  &((struct sockaddr_in6 *) addr)->sin6_addr,
159 					  dev, 1)) {
160 				ret = rdma_copy_addr(dev_addr, dev, NULL);
161 				if (vlan_id)
162 					*vlan_id = rdma_vlan_dev_vlan_id(dev);
163 				break;
164 			}
165 		}
166 		rcu_read_unlock();
167 		break;
168 #endif
169 	}
170 	return ret;
171 }
172 EXPORT_SYMBOL(rdma_translate_ip);
173 
174 static void set_timeout(unsigned long time)
175 {
176 	unsigned long delay;
177 
178 	delay = time - jiffies;
179 	if ((long)delay < 0)
180 		delay = 0;
181 
182 	mod_delayed_work(addr_wq, &work, delay);
183 }
184 
185 static void queue_req(struct addr_req *req)
186 {
187 	struct addr_req *temp_req;
188 
189 	mutex_lock(&lock);
190 	list_for_each_entry_reverse(temp_req, &req_list, list) {
191 		if (time_after_eq(req->timeout, temp_req->timeout))
192 			break;
193 	}
194 
195 	list_add(&req->list, &temp_req->list);
196 
197 	if (req_list.next == &req->list)
198 		set_timeout(req->timeout);
199 	mutex_unlock(&lock);
200 }
201 
202 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr, void *daddr)
203 {
204 	struct neighbour *n;
205 	int ret;
206 
207 	n = dst_neigh_lookup(dst, daddr);
208 
209 	rcu_read_lock();
210 	if (!n || !(n->nud_state & NUD_VALID)) {
211 		if (n)
212 			neigh_event_send(n, NULL);
213 		ret = -ENODATA;
214 	} else {
215 		ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
216 	}
217 	rcu_read_unlock();
218 
219 	if (n)
220 		neigh_release(n);
221 
222 	return ret;
223 }
224 
225 static int addr4_resolve(struct sockaddr_in *src_in,
226 			 struct sockaddr_in *dst_in,
227 			 struct rdma_dev_addr *addr)
228 {
229 	__be32 src_ip = src_in->sin_addr.s_addr;
230 	__be32 dst_ip = dst_in->sin_addr.s_addr;
231 	struct rtable *rt;
232 	struct flowi4 fl4;
233 	int ret;
234 
235 	memset(&fl4, 0, sizeof(fl4));
236 	fl4.daddr = dst_ip;
237 	fl4.saddr = src_ip;
238 	fl4.flowi4_oif = addr->bound_dev_if;
239 	rt = ip_route_output_key(&init_net, &fl4);
240 	if (IS_ERR(rt)) {
241 		ret = PTR_ERR(rt);
242 		goto out;
243 	}
244 	src_in->sin_family = AF_INET;
245 	src_in->sin_addr.s_addr = fl4.saddr;
246 
247 	if (rt->dst.dev->flags & IFF_LOOPBACK) {
248 		ret = rdma_translate_ip((struct sockaddr *)dst_in, addr, NULL);
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->dst.dev->flags & IFF_NOARP) {
256 		ret = rdma_copy_addr(addr, rt->dst.dev, NULL);
257 		goto put;
258 	}
259 
260 	ret = dst_fetch_ha(&rt->dst, addr, &fl4.daddr);
261 put:
262 	ip_rt_put(rt);
263 out:
264 	return ret;
265 }
266 
267 #if IS_ENABLED(CONFIG_IPV6)
268 static int addr6_resolve(struct sockaddr_in6 *src_in,
269 			 struct sockaddr_in6 *dst_in,
270 			 struct rdma_dev_addr *addr)
271 {
272 	struct flowi6 fl6;
273 	struct dst_entry *dst;
274 	int ret;
275 
276 	memset(&fl6, 0, sizeof fl6);
277 	fl6.daddr = dst_in->sin6_addr;
278 	fl6.saddr = src_in->sin6_addr;
279 	fl6.flowi6_oif = addr->bound_dev_if;
280 
281 	dst = ip6_route_output(&init_net, NULL, &fl6);
282 	if ((ret = dst->error))
283 		goto put;
284 
285 	if (ipv6_addr_any(&fl6.saddr)) {
286 		ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
287 					 &fl6.daddr, 0, &fl6.saddr);
288 		if (ret)
289 			goto put;
290 
291 		src_in->sin6_family = AF_INET6;
292 		src_in->sin6_addr = fl6.saddr;
293 	}
294 
295 	if (dst->dev->flags & IFF_LOOPBACK) {
296 		ret = rdma_translate_ip((struct sockaddr *)dst_in, addr, NULL);
297 		if (!ret)
298 			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
299 		goto put;
300 	}
301 
302 	/* If the device does ARP internally, return 'done' */
303 	if (dst->dev->flags & IFF_NOARP) {
304 		ret = rdma_copy_addr(addr, dst->dev, NULL);
305 		goto put;
306 	}
307 
308 	ret = dst_fetch_ha(dst, addr, &fl6.daddr);
309 put:
310 	dst_release(dst);
311 	return ret;
312 }
313 #else
314 static int addr6_resolve(struct sockaddr_in6 *src_in,
315 			 struct sockaddr_in6 *dst_in,
316 			 struct rdma_dev_addr *addr)
317 {
318 	return -EADDRNOTAVAIL;
319 }
320 #endif
321 
322 static int addr_resolve(struct sockaddr *src_in,
323 			struct sockaddr *dst_in,
324 			struct rdma_dev_addr *addr)
325 {
326 	if (src_in->sa_family == AF_INET) {
327 		return addr4_resolve((struct sockaddr_in *) src_in,
328 			(struct sockaddr_in *) dst_in, addr);
329 	} else
330 		return addr6_resolve((struct sockaddr_in6 *) src_in,
331 			(struct sockaddr_in6 *) dst_in, addr);
332 }
333 
334 static void process_req(struct work_struct *work)
335 {
336 	struct addr_req *req, *temp_req;
337 	struct sockaddr *src_in, *dst_in;
338 	struct list_head done_list;
339 
340 	INIT_LIST_HEAD(&done_list);
341 
342 	mutex_lock(&lock);
343 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
344 		if (req->status == -ENODATA) {
345 			src_in = (struct sockaddr *) &req->src_addr;
346 			dst_in = (struct sockaddr *) &req->dst_addr;
347 			req->status = addr_resolve(src_in, dst_in, req->addr);
348 			if (req->status && time_after_eq(jiffies, req->timeout))
349 				req->status = -ETIMEDOUT;
350 			else if (req->status == -ENODATA)
351 				continue;
352 		}
353 		list_move_tail(&req->list, &done_list);
354 	}
355 
356 	if (!list_empty(&req_list)) {
357 		req = list_entry(req_list.next, struct addr_req, list);
358 		set_timeout(req->timeout);
359 	}
360 	mutex_unlock(&lock);
361 
362 	list_for_each_entry_safe(req, temp_req, &done_list, list) {
363 		list_del(&req->list);
364 		req->callback(req->status, (struct sockaddr *) &req->src_addr,
365 			req->addr, req->context);
366 		put_client(req->client);
367 		kfree(req);
368 	}
369 }
370 
371 int rdma_resolve_ip(struct rdma_addr_client *client,
372 		    struct sockaddr *src_addr, struct sockaddr *dst_addr,
373 		    struct rdma_dev_addr *addr, int timeout_ms,
374 		    void (*callback)(int status, struct sockaddr *src_addr,
375 				     struct rdma_dev_addr *addr, void *context),
376 		    void *context)
377 {
378 	struct sockaddr *src_in, *dst_in;
379 	struct addr_req *req;
380 	int ret = 0;
381 
382 	req = kzalloc(sizeof *req, GFP_KERNEL);
383 	if (!req)
384 		return -ENOMEM;
385 
386 	src_in = (struct sockaddr *) &req->src_addr;
387 	dst_in = (struct sockaddr *) &req->dst_addr;
388 
389 	if (src_addr) {
390 		if (src_addr->sa_family != dst_addr->sa_family) {
391 			ret = -EINVAL;
392 			goto err;
393 		}
394 
395 		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
396 	} else {
397 		src_in->sa_family = dst_addr->sa_family;
398 	}
399 
400 	memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
401 	req->addr = addr;
402 	req->callback = callback;
403 	req->context = context;
404 	req->client = client;
405 	atomic_inc(&client->refcount);
406 
407 	req->status = addr_resolve(src_in, dst_in, addr);
408 	switch (req->status) {
409 	case 0:
410 		req->timeout = jiffies;
411 		queue_req(req);
412 		break;
413 	case -ENODATA:
414 		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
415 		queue_req(req);
416 		break;
417 	default:
418 		ret = req->status;
419 		atomic_dec(&client->refcount);
420 		goto err;
421 	}
422 	return ret;
423 err:
424 	kfree(req);
425 	return ret;
426 }
427 EXPORT_SYMBOL(rdma_resolve_ip);
428 
429 void rdma_addr_cancel(struct rdma_dev_addr *addr)
430 {
431 	struct addr_req *req, *temp_req;
432 
433 	mutex_lock(&lock);
434 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
435 		if (req->addr == addr) {
436 			req->status = -ECANCELED;
437 			req->timeout = jiffies;
438 			list_move(&req->list, &req_list);
439 			set_timeout(req->timeout);
440 			break;
441 		}
442 	}
443 	mutex_unlock(&lock);
444 }
445 EXPORT_SYMBOL(rdma_addr_cancel);
446 
447 struct resolve_cb_context {
448 	struct rdma_dev_addr *addr;
449 	struct completion comp;
450 };
451 
452 static void resolve_cb(int status, struct sockaddr *src_addr,
453 	     struct rdma_dev_addr *addr, void *context)
454 {
455 	memcpy(((struct resolve_cb_context *)context)->addr, addr, sizeof(struct
456 				rdma_dev_addr));
457 	complete(&((struct resolve_cb_context *)context)->comp);
458 }
459 
460 int rdma_addr_find_dmac_by_grh(const union ib_gid *sgid, const union ib_gid *dgid,
461 			       u8 *dmac, u16 *vlan_id)
462 {
463 	int ret = 0;
464 	struct rdma_dev_addr dev_addr;
465 	struct resolve_cb_context ctx;
466 	struct net_device *dev;
467 
468 	union {
469 		struct sockaddr     _sockaddr;
470 		struct sockaddr_in  _sockaddr_in;
471 		struct sockaddr_in6 _sockaddr_in6;
472 	} sgid_addr, dgid_addr;
473 
474 
475 	rdma_gid2ip(&sgid_addr._sockaddr, sgid);
476 	rdma_gid2ip(&dgid_addr._sockaddr, dgid);
477 
478 	memset(&dev_addr, 0, sizeof(dev_addr));
479 
480 	ctx.addr = &dev_addr;
481 	init_completion(&ctx.comp);
482 	ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
483 			&dev_addr, 1000, resolve_cb, &ctx);
484 	if (ret)
485 		return ret;
486 
487 	wait_for_completion(&ctx.comp);
488 
489 	memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
490 	dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
491 	if (!dev)
492 		return -ENODEV;
493 	if (vlan_id)
494 		*vlan_id = rdma_vlan_dev_vlan_id(dev);
495 	dev_put(dev);
496 	return ret;
497 }
498 EXPORT_SYMBOL(rdma_addr_find_dmac_by_grh);
499 
500 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
501 {
502 	int ret = 0;
503 	struct rdma_dev_addr dev_addr;
504 	union {
505 		struct sockaddr     _sockaddr;
506 		struct sockaddr_in  _sockaddr_in;
507 		struct sockaddr_in6 _sockaddr_in6;
508 	} gid_addr;
509 
510 	rdma_gid2ip(&gid_addr._sockaddr, sgid);
511 
512 	memset(&dev_addr, 0, sizeof(dev_addr));
513 	ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
514 	if (ret)
515 		return ret;
516 
517 	memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
518 	return ret;
519 }
520 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
521 
522 static int netevent_callback(struct notifier_block *self, unsigned long event,
523 	void *ctx)
524 {
525 	if (event == NETEVENT_NEIGH_UPDATE) {
526 		struct neighbour *neigh = ctx;
527 
528 		if (neigh->nud_state & NUD_VALID) {
529 			set_timeout(jiffies);
530 		}
531 	}
532 	return 0;
533 }
534 
535 static struct notifier_block nb = {
536 	.notifier_call = netevent_callback
537 };
538 
539 static int __init addr_init(void)
540 {
541 	addr_wq = create_singlethread_workqueue("ib_addr");
542 	if (!addr_wq)
543 		return -ENOMEM;
544 
545 	register_netevent_notifier(&nb);
546 	rdma_addr_register_client(&self);
547 	return 0;
548 }
549 
550 static void __exit addr_cleanup(void)
551 {
552 	rdma_addr_unregister_client(&self);
553 	unregister_netevent_notifier(&nb);
554 	destroy_workqueue(addr_wq);
555 }
556 
557 module_init(addr_init);
558 module_exit(addr_cleanup);
559