infiniband: addr: Consolidate code to fetch neighbour hardware address from dst.
[~shefty/rdma-dev.git] / drivers / infiniband / core / 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 <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
49 MODULE_AUTHOR("Sean Hefty");
50 MODULE_DESCRIPTION("IB Address Translation");
51 MODULE_LICENSE("Dual BSD/GPL");
52
53 struct addr_req {
54         struct list_head list;
55         struct sockaddr_storage src_addr;
56         struct sockaddr_storage dst_addr;
57         struct rdma_dev_addr *addr;
58         struct rdma_addr_client *client;
59         void *context;
60         void (*callback)(int status, struct sockaddr *src_addr,
61                          struct rdma_dev_addr *addr, void *context);
62         unsigned long timeout;
63         int status;
64 };
65
66 static void process_req(struct work_struct *work);
67
68 static DEFINE_MUTEX(lock);
69 static LIST_HEAD(req_list);
70 static DECLARE_DELAYED_WORK(work, process_req);
71 static struct workqueue_struct *addr_wq;
72
73 void rdma_addr_register_client(struct rdma_addr_client *client)
74 {
75         atomic_set(&client->refcount, 1);
76         init_completion(&client->comp);
77 }
78 EXPORT_SYMBOL(rdma_addr_register_client);
79
80 static inline void put_client(struct rdma_addr_client *client)
81 {
82         if (atomic_dec_and_test(&client->refcount))
83                 complete(&client->comp);
84 }
85
86 void rdma_addr_unregister_client(struct rdma_addr_client *client)
87 {
88         put_client(client);
89         wait_for_completion(&client->comp);
90 }
91 EXPORT_SYMBOL(rdma_addr_unregister_client);
92
93 int 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 EXPORT_SYMBOL(rdma_copy_addr);
105
106 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
107 {
108         struct net_device *dev;
109         int ret = -EADDRNOTAVAIL;
110
111         if (dev_addr->bound_dev_if) {
112                 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
113                 if (!dev)
114                         return -ENODEV;
115                 ret = rdma_copy_addr(dev_addr, dev, NULL);
116                 dev_put(dev);
117                 return ret;
118         }
119
120         switch (addr->sa_family) {
121         case AF_INET:
122                 dev = ip_dev_find(&init_net,
123                         ((struct sockaddr_in *) addr)->sin_addr.s_addr);
124
125                 if (!dev)
126                         return ret;
127
128                 ret = rdma_copy_addr(dev_addr, dev, NULL);
129                 dev_put(dev);
130                 break;
131
132 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
133         case AF_INET6:
134                 rcu_read_lock();
135                 for_each_netdev_rcu(&init_net, dev) {
136                         if (ipv6_chk_addr(&init_net,
137                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
138                                           dev, 1)) {
139                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
140                                 break;
141                         }
142                 }
143                 rcu_read_unlock();
144                 break;
145 #endif
146         }
147         return ret;
148 }
149 EXPORT_SYMBOL(rdma_translate_ip);
150
151 static void set_timeout(unsigned long time)
152 {
153         unsigned long delay;
154
155         cancel_delayed_work(&work);
156
157         delay = time - jiffies;
158         if ((long)delay <= 0)
159                 delay = 1;
160
161         queue_delayed_work(addr_wq, &work, delay);
162 }
163
164 static void queue_req(struct addr_req *req)
165 {
166         struct addr_req *temp_req;
167
168         mutex_lock(&lock);
169         list_for_each_entry_reverse(temp_req, &req_list, list) {
170                 if (time_after_eq(req->timeout, temp_req->timeout))
171                         break;
172         }
173
174         list_add(&req->list, &temp_req->list);
175
176         if (req_list.next == &req->list)
177                 set_timeout(req->timeout);
178         mutex_unlock(&lock);
179 }
180
181 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *addr)
182 {
183         struct neighbour *n;
184         int ret;
185
186         rcu_read_lock();
187         n = dst_get_neighbour_noref(dst);
188         if (!n || !(n->nud_state & NUD_VALID)) {
189                 if (n)
190                         neigh_event_send(n, NULL);
191                 ret = -ENODATA;
192         } else {
193                 ret = rdma_copy_addr(addr, dst->dev, n->ha);
194         }
195         rcu_read_unlock();
196
197         return ret;
198 }
199
200 static int addr4_resolve(struct sockaddr_in *src_in,
201                          struct sockaddr_in *dst_in,
202                          struct rdma_dev_addr *addr)
203 {
204         __be32 src_ip = src_in->sin_addr.s_addr;
205         __be32 dst_ip = dst_in->sin_addr.s_addr;
206         struct rtable *rt;
207         struct flowi4 fl4;
208         int ret;
209
210         memset(&fl4, 0, sizeof(fl4));
211         fl4.daddr = dst_ip;
212         fl4.saddr = src_ip;
213         fl4.flowi4_oif = addr->bound_dev_if;
214         rt = ip_route_output_key(&init_net, &fl4);
215         if (IS_ERR(rt)) {
216                 ret = PTR_ERR(rt);
217                 goto out;
218         }
219         src_in->sin_family = AF_INET;
220         src_in->sin_addr.s_addr = fl4.saddr;
221
222         if (rt->dst.dev->flags & IFF_LOOPBACK) {
223                 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
224                 if (!ret)
225                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
226                 goto put;
227         }
228
229         /* If the device does ARP internally, return 'done' */
230         if (rt->dst.dev->flags & IFF_NOARP) {
231                 ret = rdma_copy_addr(addr, rt->dst.dev, NULL);
232                 goto put;
233         }
234
235         ret = dst_fetch_ha(&rt->dst, addr);
236 put:
237         ip_rt_put(rt);
238 out:
239         return ret;
240 }
241
242 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
243 static int addr6_resolve(struct sockaddr_in6 *src_in,
244                          struct sockaddr_in6 *dst_in,
245                          struct rdma_dev_addr *addr)
246 {
247         struct flowi6 fl6;
248         struct dst_entry *dst;
249         int ret;
250
251         memset(&fl6, 0, sizeof fl6);
252         fl6.daddr = dst_in->sin6_addr;
253         fl6.saddr = src_in->sin6_addr;
254         fl6.flowi6_oif = addr->bound_dev_if;
255
256         dst = ip6_route_output(&init_net, NULL, &fl6);
257         if ((ret = dst->error))
258                 goto put;
259
260         if (ipv6_addr_any(&fl6.saddr)) {
261                 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
262                                          &fl6.daddr, 0, &fl6.saddr);
263                 if (ret)
264                         goto put;
265
266                 src_in->sin6_family = AF_INET6;
267                 src_in->sin6_addr = fl6.saddr;
268         }
269
270         if (dst->dev->flags & IFF_LOOPBACK) {
271                 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
272                 if (!ret)
273                         memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
274                 goto put;
275         }
276
277         /* If the device does ARP internally, return 'done' */
278         if (dst->dev->flags & IFF_NOARP) {
279                 ret = rdma_copy_addr(addr, dst->dev, NULL);
280                 goto put;
281         }
282
283         ret = dst_fetch_ha(dst, addr);
284 put:
285         dst_release(dst);
286         return ret;
287 }
288 #else
289 static int addr6_resolve(struct sockaddr_in6 *src_in,
290                          struct sockaddr_in6 *dst_in,
291                          struct rdma_dev_addr *addr)
292 {
293         return -EADDRNOTAVAIL;
294 }
295 #endif
296
297 static int addr_resolve(struct sockaddr *src_in,
298                         struct sockaddr *dst_in,
299                         struct rdma_dev_addr *addr)
300 {
301         if (src_in->sa_family == AF_INET) {
302                 return addr4_resolve((struct sockaddr_in *) src_in,
303                         (struct sockaddr_in *) dst_in, addr);
304         } else
305                 return addr6_resolve((struct sockaddr_in6 *) src_in,
306                         (struct sockaddr_in6 *) dst_in, addr);
307 }
308
309 static void process_req(struct work_struct *work)
310 {
311         struct addr_req *req, *temp_req;
312         struct sockaddr *src_in, *dst_in;
313         struct list_head done_list;
314
315         INIT_LIST_HEAD(&done_list);
316
317         mutex_lock(&lock);
318         list_for_each_entry_safe(req, temp_req, &req_list, list) {
319                 if (req->status == -ENODATA) {
320                         src_in = (struct sockaddr *) &req->src_addr;
321                         dst_in = (struct sockaddr *) &req->dst_addr;
322                         req->status = addr_resolve(src_in, dst_in, req->addr);
323                         if (req->status && time_after_eq(jiffies, req->timeout))
324                                 req->status = -ETIMEDOUT;
325                         else if (req->status == -ENODATA)
326                                 continue;
327                 }
328                 list_move_tail(&req->list, &done_list);
329         }
330
331         if (!list_empty(&req_list)) {
332                 req = list_entry(req_list.next, struct addr_req, list);
333                 set_timeout(req->timeout);
334         }
335         mutex_unlock(&lock);
336
337         list_for_each_entry_safe(req, temp_req, &done_list, list) {
338                 list_del(&req->list);
339                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
340                         req->addr, req->context);
341                 put_client(req->client);
342                 kfree(req);
343         }
344 }
345
346 int rdma_resolve_ip(struct rdma_addr_client *client,
347                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
348                     struct rdma_dev_addr *addr, int timeout_ms,
349                     void (*callback)(int status, struct sockaddr *src_addr,
350                                      struct rdma_dev_addr *addr, void *context),
351                     void *context)
352 {
353         struct sockaddr *src_in, *dst_in;
354         struct addr_req *req;
355         int ret = 0;
356
357         req = kzalloc(sizeof *req, GFP_KERNEL);
358         if (!req)
359                 return -ENOMEM;
360
361         src_in = (struct sockaddr *) &req->src_addr;
362         dst_in = (struct sockaddr *) &req->dst_addr;
363
364         if (src_addr) {
365                 if (src_addr->sa_family != dst_addr->sa_family) {
366                         ret = -EINVAL;
367                         goto err;
368                 }
369
370                 memcpy(src_in, src_addr, ip_addr_size(src_addr));
371         } else {
372                 src_in->sa_family = dst_addr->sa_family;
373         }
374
375         memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
376         req->addr = addr;
377         req->callback = callback;
378         req->context = context;
379         req->client = client;
380         atomic_inc(&client->refcount);
381
382         req->status = addr_resolve(src_in, dst_in, addr);
383         switch (req->status) {
384         case 0:
385                 req->timeout = jiffies;
386                 queue_req(req);
387                 break;
388         case -ENODATA:
389                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
390                 queue_req(req);
391                 break;
392         default:
393                 ret = req->status;
394                 atomic_dec(&client->refcount);
395                 goto err;
396         }
397         return ret;
398 err:
399         kfree(req);
400         return ret;
401 }
402 EXPORT_SYMBOL(rdma_resolve_ip);
403
404 void rdma_addr_cancel(struct rdma_dev_addr *addr)
405 {
406         struct addr_req *req, *temp_req;
407
408         mutex_lock(&lock);
409         list_for_each_entry_safe(req, temp_req, &req_list, list) {
410                 if (req->addr == addr) {
411                         req->status = -ECANCELED;
412                         req->timeout = jiffies;
413                         list_move(&req->list, &req_list);
414                         set_timeout(req->timeout);
415                         break;
416                 }
417         }
418         mutex_unlock(&lock);
419 }
420 EXPORT_SYMBOL(rdma_addr_cancel);
421
422 static int netevent_callback(struct notifier_block *self, unsigned long event,
423         void *ctx)
424 {
425         if (event == NETEVENT_NEIGH_UPDATE) {
426                 struct neighbour *neigh = ctx;
427
428                 if (neigh->nud_state & NUD_VALID) {
429                         set_timeout(jiffies);
430                 }
431         }
432         return 0;
433 }
434
435 static struct notifier_block nb = {
436         .notifier_call = netevent_callback
437 };
438
439 static int __init addr_init(void)
440 {
441         addr_wq = create_singlethread_workqueue("ib_addr");
442         if (!addr_wq)
443                 return -ENOMEM;
444
445         register_netevent_notifier(&nb);
446         return 0;
447 }
448
449 static void __exit addr_cleanup(void)
450 {
451         unregister_netevent_notifier(&nb);
452         destroy_workqueue(addr_wq);
453 }
454
455 module_init(addr_init);
456 module_exit(addr_cleanup);