xref: /openbmc/linux/drivers/infiniband/core/addr.c (revision 12eb4683)
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 void rdma_addr_register_client(struct rdma_addr_client *client)
90 {
91 	atomic_set(&client->refcount, 1);
92 	init_completion(&client->comp);
93 }
94 EXPORT_SYMBOL(rdma_addr_register_client);
95 
96 static inline void put_client(struct rdma_addr_client *client)
97 {
98 	if (atomic_dec_and_test(&client->refcount))
99 		complete(&client->comp);
100 }
101 
102 void rdma_addr_unregister_client(struct rdma_addr_client *client)
103 {
104 	put_client(client);
105 	wait_for_completion(&client->comp);
106 }
107 EXPORT_SYMBOL(rdma_addr_unregister_client);
108 
109 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
110 		     const unsigned char *dst_dev_addr)
111 {
112 	dev_addr->dev_type = dev->type;
113 	memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
114 	memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
115 	if (dst_dev_addr)
116 		memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
117 	dev_addr->bound_dev_if = dev->ifindex;
118 	return 0;
119 }
120 EXPORT_SYMBOL(rdma_copy_addr);
121 
122 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
123 {
124 	struct net_device *dev;
125 	int ret = -EADDRNOTAVAIL;
126 
127 	if (dev_addr->bound_dev_if) {
128 		dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
129 		if (!dev)
130 			return -ENODEV;
131 		ret = rdma_copy_addr(dev_addr, dev, NULL);
132 		dev_put(dev);
133 		return ret;
134 	}
135 
136 	switch (addr->sa_family) {
137 	case AF_INET:
138 		dev = ip_dev_find(&init_net,
139 			((struct sockaddr_in *) addr)->sin_addr.s_addr);
140 
141 		if (!dev)
142 			return ret;
143 
144 		ret = rdma_copy_addr(dev_addr, dev, NULL);
145 		dev_put(dev);
146 		break;
147 
148 #if IS_ENABLED(CONFIG_IPV6)
149 	case AF_INET6:
150 		rcu_read_lock();
151 		for_each_netdev_rcu(&init_net, dev) {
152 			if (ipv6_chk_addr(&init_net,
153 					  &((struct sockaddr_in6 *) addr)->sin6_addr,
154 					  dev, 1)) {
155 				ret = rdma_copy_addr(dev_addr, dev, NULL);
156 				break;
157 			}
158 		}
159 		rcu_read_unlock();
160 		break;
161 #endif
162 	}
163 	return ret;
164 }
165 EXPORT_SYMBOL(rdma_translate_ip);
166 
167 static void set_timeout(unsigned long time)
168 {
169 	unsigned long delay;
170 
171 	delay = time - jiffies;
172 	if ((long)delay <= 0)
173 		delay = 1;
174 
175 	mod_delayed_work(addr_wq, &work, delay);
176 }
177 
178 static void queue_req(struct addr_req *req)
179 {
180 	struct addr_req *temp_req;
181 
182 	mutex_lock(&lock);
183 	list_for_each_entry_reverse(temp_req, &req_list, list) {
184 		if (time_after_eq(req->timeout, temp_req->timeout))
185 			break;
186 	}
187 
188 	list_add(&req->list, &temp_req->list);
189 
190 	if (req_list.next == &req->list)
191 		set_timeout(req->timeout);
192 	mutex_unlock(&lock);
193 }
194 
195 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr, void *daddr)
196 {
197 	struct neighbour *n;
198 	int ret;
199 
200 	n = dst_neigh_lookup(dst, daddr);
201 
202 	rcu_read_lock();
203 	if (!n || !(n->nud_state & NUD_VALID)) {
204 		if (n)
205 			neigh_event_send(n, NULL);
206 		ret = -ENODATA;
207 	} else {
208 		ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
209 	}
210 	rcu_read_unlock();
211 
212 	if (n)
213 		neigh_release(n);
214 
215 	return ret;
216 }
217 
218 static int addr4_resolve(struct sockaddr_in *src_in,
219 			 struct sockaddr_in *dst_in,
220 			 struct rdma_dev_addr *addr)
221 {
222 	__be32 src_ip = src_in->sin_addr.s_addr;
223 	__be32 dst_ip = dst_in->sin_addr.s_addr;
224 	struct rtable *rt;
225 	struct flowi4 fl4;
226 	int ret;
227 
228 	memset(&fl4, 0, sizeof(fl4));
229 	fl4.daddr = dst_ip;
230 	fl4.saddr = src_ip;
231 	fl4.flowi4_oif = addr->bound_dev_if;
232 	rt = ip_route_output_key(&init_net, &fl4);
233 	if (IS_ERR(rt)) {
234 		ret = PTR_ERR(rt);
235 		goto out;
236 	}
237 	src_in->sin_family = AF_INET;
238 	src_in->sin_addr.s_addr = fl4.saddr;
239 
240 	if (rt->dst.dev->flags & IFF_LOOPBACK) {
241 		ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
242 		if (!ret)
243 			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
244 		goto put;
245 	}
246 
247 	/* If the device does ARP internally, return 'done' */
248 	if (rt->dst.dev->flags & IFF_NOARP) {
249 		ret = rdma_copy_addr(addr, rt->dst.dev, NULL);
250 		goto put;
251 	}
252 
253 	ret = dst_fetch_ha(&rt->dst, addr, &fl4.daddr);
254 put:
255 	ip_rt_put(rt);
256 out:
257 	return ret;
258 }
259 
260 #if IS_ENABLED(CONFIG_IPV6)
261 static int addr6_resolve(struct sockaddr_in6 *src_in,
262 			 struct sockaddr_in6 *dst_in,
263 			 struct rdma_dev_addr *addr)
264 {
265 	struct flowi6 fl6;
266 	struct dst_entry *dst;
267 	int ret;
268 
269 	memset(&fl6, 0, sizeof fl6);
270 	fl6.daddr = dst_in->sin6_addr;
271 	fl6.saddr = src_in->sin6_addr;
272 	fl6.flowi6_oif = addr->bound_dev_if;
273 
274 	dst = ip6_route_output(&init_net, NULL, &fl6);
275 	if ((ret = dst->error))
276 		goto put;
277 
278 	if (ipv6_addr_any(&fl6.saddr)) {
279 		ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
280 					 &fl6.daddr, 0, &fl6.saddr);
281 		if (ret)
282 			goto put;
283 
284 		src_in->sin6_family = AF_INET6;
285 		src_in->sin6_addr = fl6.saddr;
286 	}
287 
288 	if (dst->dev->flags & IFF_LOOPBACK) {
289 		ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
290 		if (!ret)
291 			memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
292 		goto put;
293 	}
294 
295 	/* If the device does ARP internally, return 'done' */
296 	if (dst->dev->flags & IFF_NOARP) {
297 		ret = rdma_copy_addr(addr, dst->dev, NULL);
298 		goto put;
299 	}
300 
301 	ret = dst_fetch_ha(dst, addr, &fl6.daddr);
302 put:
303 	dst_release(dst);
304 	return ret;
305 }
306 #else
307 static int addr6_resolve(struct sockaddr_in6 *src_in,
308 			 struct sockaddr_in6 *dst_in,
309 			 struct rdma_dev_addr *addr)
310 {
311 	return -EADDRNOTAVAIL;
312 }
313 #endif
314 
315 static int addr_resolve(struct sockaddr *src_in,
316 			struct sockaddr *dst_in,
317 			struct rdma_dev_addr *addr)
318 {
319 	if (src_in->sa_family == AF_INET) {
320 		return addr4_resolve((struct sockaddr_in *) src_in,
321 			(struct sockaddr_in *) dst_in, addr);
322 	} else
323 		return addr6_resolve((struct sockaddr_in6 *) src_in,
324 			(struct sockaddr_in6 *) dst_in, addr);
325 }
326 
327 static void process_req(struct work_struct *work)
328 {
329 	struct addr_req *req, *temp_req;
330 	struct sockaddr *src_in, *dst_in;
331 	struct list_head done_list;
332 
333 	INIT_LIST_HEAD(&done_list);
334 
335 	mutex_lock(&lock);
336 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
337 		if (req->status == -ENODATA) {
338 			src_in = (struct sockaddr *) &req->src_addr;
339 			dst_in = (struct sockaddr *) &req->dst_addr;
340 			req->status = addr_resolve(src_in, dst_in, req->addr);
341 			if (req->status && time_after_eq(jiffies, req->timeout))
342 				req->status = -ETIMEDOUT;
343 			else if (req->status == -ENODATA)
344 				continue;
345 		}
346 		list_move_tail(&req->list, &done_list);
347 	}
348 
349 	if (!list_empty(&req_list)) {
350 		req = list_entry(req_list.next, struct addr_req, list);
351 		set_timeout(req->timeout);
352 	}
353 	mutex_unlock(&lock);
354 
355 	list_for_each_entry_safe(req, temp_req, &done_list, list) {
356 		list_del(&req->list);
357 		req->callback(req->status, (struct sockaddr *) &req->src_addr,
358 			req->addr, req->context);
359 		put_client(req->client);
360 		kfree(req);
361 	}
362 }
363 
364 int rdma_resolve_ip(struct rdma_addr_client *client,
365 		    struct sockaddr *src_addr, struct sockaddr *dst_addr,
366 		    struct rdma_dev_addr *addr, int timeout_ms,
367 		    void (*callback)(int status, struct sockaddr *src_addr,
368 				     struct rdma_dev_addr *addr, void *context),
369 		    void *context)
370 {
371 	struct sockaddr *src_in, *dst_in;
372 	struct addr_req *req;
373 	int ret = 0;
374 
375 	req = kzalloc(sizeof *req, GFP_KERNEL);
376 	if (!req)
377 		return -ENOMEM;
378 
379 	src_in = (struct sockaddr *) &req->src_addr;
380 	dst_in = (struct sockaddr *) &req->dst_addr;
381 
382 	if (src_addr) {
383 		if (src_addr->sa_family != dst_addr->sa_family) {
384 			ret = -EINVAL;
385 			goto err;
386 		}
387 
388 		memcpy(src_in, src_addr, rdma_addr_size(src_addr));
389 	} else {
390 		src_in->sa_family = dst_addr->sa_family;
391 	}
392 
393 	memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
394 	req->addr = addr;
395 	req->callback = callback;
396 	req->context = context;
397 	req->client = client;
398 	atomic_inc(&client->refcount);
399 
400 	req->status = addr_resolve(src_in, dst_in, addr);
401 	switch (req->status) {
402 	case 0:
403 		req->timeout = jiffies;
404 		queue_req(req);
405 		break;
406 	case -ENODATA:
407 		req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
408 		queue_req(req);
409 		break;
410 	default:
411 		ret = req->status;
412 		atomic_dec(&client->refcount);
413 		goto err;
414 	}
415 	return ret;
416 err:
417 	kfree(req);
418 	return ret;
419 }
420 EXPORT_SYMBOL(rdma_resolve_ip);
421 
422 void rdma_addr_cancel(struct rdma_dev_addr *addr)
423 {
424 	struct addr_req *req, *temp_req;
425 
426 	mutex_lock(&lock);
427 	list_for_each_entry_safe(req, temp_req, &req_list, list) {
428 		if (req->addr == addr) {
429 			req->status = -ECANCELED;
430 			req->timeout = jiffies;
431 			list_move(&req->list, &req_list);
432 			set_timeout(req->timeout);
433 			break;
434 		}
435 	}
436 	mutex_unlock(&lock);
437 }
438 EXPORT_SYMBOL(rdma_addr_cancel);
439 
440 static int netevent_callback(struct notifier_block *self, unsigned long event,
441 	void *ctx)
442 {
443 	if (event == NETEVENT_NEIGH_UPDATE) {
444 		struct neighbour *neigh = ctx;
445 
446 		if (neigh->nud_state & NUD_VALID) {
447 			set_timeout(jiffies);
448 		}
449 	}
450 	return 0;
451 }
452 
453 static struct notifier_block nb = {
454 	.notifier_call = netevent_callback
455 };
456 
457 static int __init addr_init(void)
458 {
459 	addr_wq = create_singlethread_workqueue("ib_addr");
460 	if (!addr_wq)
461 		return -ENOMEM;
462 
463 	register_netevent_notifier(&nb);
464 	return 0;
465 }
466 
467 static void __exit addr_cleanup(void)
468 {
469 	unregister_netevent_notifier(&nb);
470 	destroy_workqueue(addr_wq);
471 }
472 
473 module_init(addr_init);
474 module_exit(addr_cleanup);
475