Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-next-2.6
[linux-2.6] / 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/workqueue.h>
39 #include <linux/if_arp.h>
40 #include <net/arp.h>
41 #include <net/neighbour.h>
42 #include <net/route.h>
43 #include <net/netevent.h>
44 #include <net/addrconf.h>
45 #include <net/ip6_route.h>
46 #include <rdma/ib_addr.h>
47
48 MODULE_AUTHOR("Sean Hefty");
49 MODULE_DESCRIPTION("IB Address Translation");
50 MODULE_LICENSE("Dual BSD/GPL");
51
52 struct 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
65 static void process_req(struct work_struct *work);
66
67 static DEFINE_MUTEX(lock);
68 static LIST_HEAD(req_list);
69 static DECLARE_DELAYED_WORK(work, process_req);
70 static struct workqueue_struct *addr_wq;
71
72 void rdma_addr_register_client(struct rdma_addr_client *client)
73 {
74         atomic_set(&client->refcount, 1);
75         init_completion(&client->comp);
76 }
77 EXPORT_SYMBOL(rdma_addr_register_client);
78
79 static inline void put_client(struct rdma_addr_client *client)
80 {
81         if (atomic_dec_and_test(&client->refcount))
82                 complete(&client->comp);
83 }
84
85 void rdma_addr_unregister_client(struct rdma_addr_client *client)
86 {
87         put_client(client);
88         wait_for_completion(&client->comp);
89 }
90 EXPORT_SYMBOL(rdma_addr_unregister_client);
91
92 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
93                      const unsigned char *dst_dev_addr)
94 {
95         switch (dev->type) {
96         case ARPHRD_INFINIBAND:
97                 dev_addr->dev_type = RDMA_NODE_IB_CA;
98                 break;
99         case ARPHRD_ETHER:
100                 dev_addr->dev_type = RDMA_NODE_RNIC;
101                 break;
102         default:
103                 return -EADDRNOTAVAIL;
104         }
105
106         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
107         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
108         if (dst_dev_addr)
109                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
110         dev_addr->src_dev = dev;
111         return 0;
112 }
113 EXPORT_SYMBOL(rdma_copy_addr);
114
115 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
116 {
117         struct net_device *dev;
118         int ret = -EADDRNOTAVAIL;
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                 for_each_netdev(&init_net, dev) {
135                         if (ipv6_chk_addr(&init_net,
136                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
137                                           dev, 1)) {
138                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
139                                 break;
140                         }
141                 }
142                 break;
143 #endif
144         }
145         return ret;
146 }
147 EXPORT_SYMBOL(rdma_translate_ip);
148
149 static void set_timeout(unsigned long time)
150 {
151         unsigned long delay;
152
153         cancel_delayed_work(&work);
154
155         delay = time - jiffies;
156         if ((long)delay <= 0)
157                 delay = 1;
158
159         queue_delayed_work(addr_wq, &work, delay);
160 }
161
162 static void queue_req(struct addr_req *req)
163 {
164         struct addr_req *temp_req;
165
166         mutex_lock(&lock);
167         list_for_each_entry_reverse(temp_req, &req_list, list) {
168                 if (time_after_eq(req->timeout, temp_req->timeout))
169                         break;
170         }
171
172         list_add(&req->list, &temp_req->list);
173
174         if (req_list.next == &req->list)
175                 set_timeout(req->timeout);
176         mutex_unlock(&lock);
177 }
178
179 static void addr_send_arp(struct sockaddr *dst_in)
180 {
181         struct rtable *rt;
182         struct flowi fl;
183
184         memset(&fl, 0, sizeof fl);
185
186         switch (dst_in->sa_family) {
187         case AF_INET:
188                 fl.nl_u.ip4_u.daddr =
189                         ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
190
191                 if (ip_route_output_key(&init_net, &rt, &fl))
192                         return;
193
194                 neigh_event_send(rt->u.dst.neighbour, NULL);
195                 ip_rt_put(rt);
196                 break;
197
198 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
199         case AF_INET6:
200         {
201                 struct dst_entry *dst;
202
203                 fl.nl_u.ip6_u.daddr =
204                         ((struct sockaddr_in6 *) dst_in)->sin6_addr;
205
206                 dst = ip6_route_output(&init_net, NULL, &fl);
207                 if (!dst)
208                         return;
209
210                 neigh_event_send(dst->neighbour, NULL);
211                 dst_release(dst);
212                 break;
213         }
214 #endif
215         }
216 }
217
218 static int addr4_resolve_remote(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 flowi fl;
225         struct rtable *rt;
226         struct neighbour *neigh;
227         int ret;
228
229         memset(&fl, 0, sizeof fl);
230         fl.nl_u.ip4_u.daddr = dst_ip;
231         fl.nl_u.ip4_u.saddr = src_ip;
232         ret = ip_route_output_key(&init_net, &rt, &fl);
233         if (ret)
234                 goto out;
235
236         /* If the device does ARP internally, return 'done' */
237         if (rt->idev->dev->flags & IFF_NOARP) {
238                 rdma_copy_addr(addr, rt->idev->dev, NULL);
239                 goto put;
240         }
241
242         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
243         if (!neigh) {
244                 ret = -ENODATA;
245                 goto put;
246         }
247
248         if (!(neigh->nud_state & NUD_VALID)) {
249                 ret = -ENODATA;
250                 goto release;
251         }
252
253         if (!src_ip) {
254                 src_in->sin_family = dst_in->sin_family;
255                 src_in->sin_addr.s_addr = rt->rt_src;
256         }
257
258         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
259 release:
260         neigh_release(neigh);
261 put:
262         ip_rt_put(rt);
263 out:
264         return ret;
265 }
266
267 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
268 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
269                                struct sockaddr_in6 *dst_in,
270                                struct rdma_dev_addr *addr)
271 {
272         struct flowi fl;
273         struct neighbour *neigh;
274         struct dst_entry *dst;
275         int ret = -ENODATA;
276
277         memset(&fl, 0, sizeof fl);
278         fl.nl_u.ip6_u.daddr = dst_in->sin6_addr;
279         fl.nl_u.ip6_u.saddr = src_in->sin6_addr;
280
281         dst = ip6_route_output(&init_net, NULL, &fl);
282         if (!dst)
283                 return ret;
284
285         if (dst->dev->flags & IFF_NOARP) {
286                 ret = rdma_copy_addr(addr, dst->dev, NULL);
287         } else {
288                 neigh = dst->neighbour;
289                 if (neigh && (neigh->nud_state & NUD_VALID))
290                         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
291         }
292
293         dst_release(dst);
294         return ret;
295 }
296 #else
297 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
298                                struct sockaddr_in6 *dst_in,
299                                struct rdma_dev_addr *addr)
300 {
301         return -EADDRNOTAVAIL;
302 }
303 #endif
304
305 static int addr_resolve_remote(struct sockaddr *src_in,
306                                 struct sockaddr *dst_in,
307                                 struct rdma_dev_addr *addr)
308 {
309         if (src_in->sa_family == AF_INET) {
310                 return addr4_resolve_remote((struct sockaddr_in *) src_in,
311                         (struct sockaddr_in *) dst_in, addr);
312         } else
313                 return addr6_resolve_remote((struct sockaddr_in6 *) src_in,
314                         (struct sockaddr_in6 *) dst_in, addr);
315 }
316
317 static void process_req(struct work_struct *work)
318 {
319         struct addr_req *req, *temp_req;
320         struct sockaddr *src_in, *dst_in;
321         struct list_head done_list;
322
323         INIT_LIST_HEAD(&done_list);
324
325         mutex_lock(&lock);
326         list_for_each_entry_safe(req, temp_req, &req_list, list) {
327                 if (req->status == -ENODATA) {
328                         src_in = (struct sockaddr *) &req->src_addr;
329                         dst_in = (struct sockaddr *) &req->dst_addr;
330                         req->status = addr_resolve_remote(src_in, dst_in,
331                                                           req->addr);
332                         if (req->status && time_after_eq(jiffies, req->timeout))
333                                 req->status = -ETIMEDOUT;
334                         else if (req->status == -ENODATA)
335                                 continue;
336                 }
337                 list_move_tail(&req->list, &done_list);
338         }
339
340         if (!list_empty(&req_list)) {
341                 req = list_entry(req_list.next, struct addr_req, list);
342                 set_timeout(req->timeout);
343         }
344         mutex_unlock(&lock);
345
346         list_for_each_entry_safe(req, temp_req, &done_list, list) {
347                 list_del(&req->list);
348                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
349                         req->addr, req->context);
350                 put_client(req->client);
351                 kfree(req);
352         }
353 }
354
355 static int addr_resolve_local(struct sockaddr *src_in,
356                               struct sockaddr *dst_in,
357                               struct rdma_dev_addr *addr)
358 {
359         struct net_device *dev;
360         int ret;
361
362         switch (dst_in->sa_family) {
363         case AF_INET:
364         {
365                 __be32 src_ip = ((struct sockaddr_in *) src_in)->sin_addr.s_addr;
366                 __be32 dst_ip = ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
367
368                 dev = ip_dev_find(&init_net, dst_ip);
369                 if (!dev)
370                         return -EADDRNOTAVAIL;
371
372                 if (ipv4_is_zeronet(src_ip)) {
373                         src_in->sa_family = dst_in->sa_family;
374                         ((struct sockaddr_in *) src_in)->sin_addr.s_addr = dst_ip;
375                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
376                 } else if (ipv4_is_loopback(src_ip)) {
377                         ret = rdma_translate_ip(dst_in, addr);
378                         if (!ret)
379                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
380                 } else {
381                         ret = rdma_translate_ip(src_in, addr);
382                         if (!ret)
383                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
384                 }
385                 dev_put(dev);
386                 break;
387         }
388
389 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
390         case AF_INET6:
391         {
392                 struct in6_addr *a;
393
394                 for_each_netdev(&init_net, dev)
395                         if (ipv6_chk_addr(&init_net,
396                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
397                                           dev, 1))
398                                 break;
399
400                 if (!dev)
401                         return -EADDRNOTAVAIL;
402
403                 a = &((struct sockaddr_in6 *) src_in)->sin6_addr;
404
405                 if (ipv6_addr_any(a)) {
406                         src_in->sa_family = dst_in->sa_family;
407                         ((struct sockaddr_in6 *) src_in)->sin6_addr =
408                                 ((struct sockaddr_in6 *) dst_in)->sin6_addr;
409                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
410                 } else if (ipv6_addr_loopback(a)) {
411                         ret = rdma_translate_ip(dst_in, addr);
412                         if (!ret)
413                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
414                 } else  {
415                         ret = rdma_translate_ip(src_in, addr);
416                         if (!ret)
417                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
418                 }
419                 break;
420         }
421 #endif
422
423         default:
424                 ret = -EADDRNOTAVAIL;
425                 break;
426         }
427
428         return ret;
429 }
430
431 int rdma_resolve_ip(struct rdma_addr_client *client,
432                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
433                     struct rdma_dev_addr *addr, int timeout_ms,
434                     void (*callback)(int status, struct sockaddr *src_addr,
435                                      struct rdma_dev_addr *addr, void *context),
436                     void *context)
437 {
438         struct sockaddr *src_in, *dst_in;
439         struct addr_req *req;
440         int ret = 0;
441
442         req = kzalloc(sizeof *req, GFP_KERNEL);
443         if (!req)
444                 return -ENOMEM;
445
446         if (src_addr)
447                 memcpy(&req->src_addr, src_addr, ip_addr_size(src_addr));
448         memcpy(&req->dst_addr, dst_addr, ip_addr_size(dst_addr));
449         req->addr = addr;
450         req->callback = callback;
451         req->context = context;
452         req->client = client;
453         atomic_inc(&client->refcount);
454
455         src_in = (struct sockaddr *) &req->src_addr;
456         dst_in = (struct sockaddr *) &req->dst_addr;
457
458         req->status = addr_resolve_local(src_in, dst_in, addr);
459         if (req->status == -EADDRNOTAVAIL)
460                 req->status = addr_resolve_remote(src_in, dst_in, addr);
461
462         switch (req->status) {
463         case 0:
464                 req->timeout = jiffies;
465                 queue_req(req);
466                 break;
467         case -ENODATA:
468                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
469                 queue_req(req);
470                 addr_send_arp(dst_in);
471                 break;
472         default:
473                 ret = req->status;
474                 atomic_dec(&client->refcount);
475                 kfree(req);
476                 break;
477         }
478         return ret;
479 }
480 EXPORT_SYMBOL(rdma_resolve_ip);
481
482 void rdma_addr_cancel(struct rdma_dev_addr *addr)
483 {
484         struct addr_req *req, *temp_req;
485
486         mutex_lock(&lock);
487         list_for_each_entry_safe(req, temp_req, &req_list, list) {
488                 if (req->addr == addr) {
489                         req->status = -ECANCELED;
490                         req->timeout = jiffies;
491                         list_move(&req->list, &req_list);
492                         set_timeout(req->timeout);
493                         break;
494                 }
495         }
496         mutex_unlock(&lock);
497 }
498 EXPORT_SYMBOL(rdma_addr_cancel);
499
500 static int netevent_callback(struct notifier_block *self, unsigned long event,
501         void *ctx)
502 {
503         if (event == NETEVENT_NEIGH_UPDATE) {
504                 struct neighbour *neigh = ctx;
505
506                 if (neigh->nud_state & NUD_VALID) {
507                         set_timeout(jiffies);
508                 }
509         }
510         return 0;
511 }
512
513 static struct notifier_block nb = {
514         .notifier_call = netevent_callback
515 };
516
517 static int addr_init(void)
518 {
519         addr_wq = create_singlethread_workqueue("ib_addr");
520         if (!addr_wq)
521                 return -ENOMEM;
522
523         register_netevent_notifier(&nb);
524         return 0;
525 }
526
527 static void addr_cleanup(void)
528 {
529         unregister_netevent_notifier(&nb);
530         destroy_workqueue(addr_wq);
531 }
532
533 module_init(addr_init);
534 module_exit(addr_cleanup);