RDMA/cma: Replace net_device pointer with index
[linux-2.6.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/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->bound_dev_if = dev->ifindex;
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         if (dev_addr->bound_dev_if) {
121                 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
122                 if (!dev)
123                         return -ENODEV;
124                 ret = rdma_copy_addr(dev_addr, dev, NULL);
125                 dev_put(dev);
126                 return ret;
127         }
128
129         switch (addr->sa_family) {
130         case AF_INET:
131                 dev = ip_dev_find(&init_net,
132                         ((struct sockaddr_in *) addr)->sin_addr.s_addr);
133
134                 if (!dev)
135                         return ret;
136
137                 ret = rdma_copy_addr(dev_addr, dev, NULL);
138                 dev_put(dev);
139                 break;
140
141 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
142         case AF_INET6:
143                 read_lock(&dev_base_lock);
144                 for_each_netdev(&init_net, dev) {
145                         if (ipv6_chk_addr(&init_net,
146                                           &((struct sockaddr_in6 *) addr)->sin6_addr,
147                                           dev, 1)) {
148                                 ret = rdma_copy_addr(dev_addr, dev, NULL);
149                                 break;
150                         }
151                 }
152                 read_unlock(&dev_base_lock);
153                 break;
154 #endif
155         }
156         return ret;
157 }
158 EXPORT_SYMBOL(rdma_translate_ip);
159
160 static void set_timeout(unsigned long time)
161 {
162         unsigned long delay;
163
164         cancel_delayed_work(&work);
165
166         delay = time - jiffies;
167         if ((long)delay <= 0)
168                 delay = 1;
169
170         queue_delayed_work(addr_wq, &work, delay);
171 }
172
173 static void queue_req(struct addr_req *req)
174 {
175         struct addr_req *temp_req;
176
177         mutex_lock(&lock);
178         list_for_each_entry_reverse(temp_req, &req_list, list) {
179                 if (time_after_eq(req->timeout, temp_req->timeout))
180                         break;
181         }
182
183         list_add(&req->list, &temp_req->list);
184
185         if (req_list.next == &req->list)
186                 set_timeout(req->timeout);
187         mutex_unlock(&lock);
188 }
189
190 static void addr_send_arp(struct sockaddr *dst_in)
191 {
192         struct rtable *rt;
193         struct flowi fl;
194
195         memset(&fl, 0, sizeof fl);
196
197         switch (dst_in->sa_family) {
198         case AF_INET:
199                 fl.nl_u.ip4_u.daddr =
200                         ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
201
202                 if (ip_route_output_key(&init_net, &rt, &fl))
203                         return;
204
205                 neigh_event_send(rt->u.dst.neighbour, NULL);
206                 ip_rt_put(rt);
207                 break;
208
209 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
210         case AF_INET6:
211         {
212                 struct dst_entry *dst;
213
214                 fl.nl_u.ip6_u.daddr =
215                         ((struct sockaddr_in6 *) dst_in)->sin6_addr;
216
217                 dst = ip6_route_output(&init_net, NULL, &fl);
218                 if (!dst)
219                         return;
220
221                 neigh_event_send(dst->neighbour, NULL);
222                 dst_release(dst);
223                 break;
224         }
225 #endif
226         }
227 }
228
229 static int addr4_resolve_remote(struct sockaddr_in *src_in,
230                                struct sockaddr_in *dst_in,
231                                struct rdma_dev_addr *addr)
232 {
233         __be32 src_ip = src_in->sin_addr.s_addr;
234         __be32 dst_ip = dst_in->sin_addr.s_addr;
235         struct flowi fl;
236         struct rtable *rt;
237         struct neighbour *neigh;
238         int ret;
239
240         memset(&fl, 0, sizeof fl);
241         fl.nl_u.ip4_u.daddr = dst_ip;
242         fl.nl_u.ip4_u.saddr = src_ip;
243         fl.oif = addr->bound_dev_if;
244
245         ret = ip_route_output_key(&init_net, &rt, &fl);
246         if (ret)
247                 goto out;
248
249         /* If the device does ARP internally, return 'done' */
250         if (rt->idev->dev->flags & IFF_NOARP) {
251                 rdma_copy_addr(addr, rt->idev->dev, NULL);
252                 goto put;
253         }
254
255         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
256         if (!neigh) {
257                 ret = -ENODATA;
258                 goto put;
259         }
260
261         if (!(neigh->nud_state & NUD_VALID)) {
262                 ret = -ENODATA;
263                 goto release;
264         }
265
266         if (!src_ip) {
267                 src_in->sin_family = dst_in->sin_family;
268                 src_in->sin_addr.s_addr = rt->rt_src;
269         }
270
271         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
272 release:
273         neigh_release(neigh);
274 put:
275         ip_rt_put(rt);
276 out:
277         return ret;
278 }
279
280 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
281 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
282                                struct sockaddr_in6 *dst_in,
283                                struct rdma_dev_addr *addr)
284 {
285         struct flowi fl;
286         struct neighbour *neigh;
287         struct dst_entry *dst;
288         int ret = -ENODATA;
289
290         memset(&fl, 0, sizeof fl);
291         fl.nl_u.ip6_u.daddr = dst_in->sin6_addr;
292         fl.nl_u.ip6_u.saddr = src_in->sin6_addr;
293         fl.oif = addr->bound_dev_if;
294
295         dst = ip6_route_output(&init_net, NULL, &fl);
296         if (!dst)
297                 return ret;
298
299         if (dst->dev->flags & IFF_NOARP) {
300                 ret = rdma_copy_addr(addr, dst->dev, NULL);
301         } else {
302                 neigh = dst->neighbour;
303                 if (neigh && (neigh->nud_state & NUD_VALID))
304                         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
305         }
306
307         dst_release(dst);
308         return ret;
309 }
310 #else
311 static int addr6_resolve_remote(struct sockaddr_in6 *src_in,
312                                struct sockaddr_in6 *dst_in,
313                                struct rdma_dev_addr *addr)
314 {
315         return -EADDRNOTAVAIL;
316 }
317 #endif
318
319 static int addr_resolve_remote(struct sockaddr *src_in,
320                                 struct sockaddr *dst_in,
321                                 struct rdma_dev_addr *addr)
322 {
323         if (src_in->sa_family == AF_INET) {
324                 return addr4_resolve_remote((struct sockaddr_in *) src_in,
325                         (struct sockaddr_in *) dst_in, addr);
326         } else
327                 return addr6_resolve_remote((struct sockaddr_in6 *) src_in,
328                         (struct sockaddr_in6 *) dst_in, addr);
329 }
330
331 static void process_req(struct work_struct *work)
332 {
333         struct addr_req *req, *temp_req;
334         struct sockaddr *src_in, *dst_in;
335         struct list_head done_list;
336
337         INIT_LIST_HEAD(&done_list);
338
339         mutex_lock(&lock);
340         list_for_each_entry_safe(req, temp_req, &req_list, list) {
341                 if (req->status == -ENODATA) {
342                         src_in = (struct sockaddr *) &req->src_addr;
343                         dst_in = (struct sockaddr *) &req->dst_addr;
344                         req->status = addr_resolve_remote(src_in, dst_in,
345                                                           req->addr);
346                         if (req->status && time_after_eq(jiffies, req->timeout))
347                                 req->status = -ETIMEDOUT;
348                         else if (req->status == -ENODATA)
349                                 continue;
350                 }
351                 list_move_tail(&req->list, &done_list);
352         }
353
354         if (!list_empty(&req_list)) {
355                 req = list_entry(req_list.next, struct addr_req, list);
356                 set_timeout(req->timeout);
357         }
358         mutex_unlock(&lock);
359
360         list_for_each_entry_safe(req, temp_req, &done_list, list) {
361                 list_del(&req->list);
362                 req->callback(req->status, (struct sockaddr *) &req->src_addr,
363                         req->addr, req->context);
364                 put_client(req->client);
365                 kfree(req);
366         }
367 }
368
369 static int addr_resolve_local(struct sockaddr *src_in,
370                               struct sockaddr *dst_in,
371                               struct rdma_dev_addr *addr)
372 {
373         struct net_device *dev;
374         int ret;
375
376         switch (dst_in->sa_family) {
377         case AF_INET:
378         {
379                 __be32 src_ip = ((struct sockaddr_in *) src_in)->sin_addr.s_addr;
380                 __be32 dst_ip = ((struct sockaddr_in *) dst_in)->sin_addr.s_addr;
381
382                 dev = ip_dev_find(&init_net, dst_ip);
383                 if (!dev)
384                         return -EADDRNOTAVAIL;
385
386                 if (ipv4_is_zeronet(src_ip)) {
387                         src_in->sa_family = dst_in->sa_family;
388                         ((struct sockaddr_in *) src_in)->sin_addr.s_addr = dst_ip;
389                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
390                 } else if (ipv4_is_loopback(src_ip)) {
391                         ret = rdma_translate_ip(dst_in, addr);
392                         if (!ret)
393                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
394                 } else {
395                         ret = rdma_translate_ip(src_in, addr);
396                         if (!ret)
397                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
398                 }
399                 dev_put(dev);
400                 break;
401         }
402
403 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
404         case AF_INET6:
405         {
406                 struct in6_addr *a;
407
408                 read_lock(&dev_base_lock);
409                 for_each_netdev(&init_net, dev)
410                         if (ipv6_chk_addr(&init_net,
411                                           &((struct sockaddr_in6 *) dst_in)->sin6_addr,
412                                           dev, 1))
413                                 break;
414
415                 if (!dev) {
416                         read_unlock(&dev_base_lock);
417                         return -EADDRNOTAVAIL;
418                 }
419
420                 a = &((struct sockaddr_in6 *) src_in)->sin6_addr;
421
422                 if (ipv6_addr_any(a)) {
423                         src_in->sa_family = dst_in->sa_family;
424                         ((struct sockaddr_in6 *) src_in)->sin6_addr =
425                                 ((struct sockaddr_in6 *) dst_in)->sin6_addr;
426                         ret = rdma_copy_addr(addr, dev, dev->dev_addr);
427                 } else if (ipv6_addr_loopback(a)) {
428                         ret = rdma_translate_ip(dst_in, addr);
429                         if (!ret)
430                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
431                 } else  {
432                         ret = rdma_translate_ip(src_in, addr);
433                         if (!ret)
434                                 memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
435                 }
436                 read_unlock(&dev_base_lock);
437                 break;
438         }
439 #endif
440
441         default:
442                 ret = -EADDRNOTAVAIL;
443                 break;
444         }
445
446         return ret;
447 }
448
449 int rdma_resolve_ip(struct rdma_addr_client *client,
450                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
451                     struct rdma_dev_addr *addr, int timeout_ms,
452                     void (*callback)(int status, struct sockaddr *src_addr,
453                                      struct rdma_dev_addr *addr, void *context),
454                     void *context)
455 {
456         struct sockaddr *src_in, *dst_in;
457         struct addr_req *req;
458         int ret = 0;
459
460         req = kzalloc(sizeof *req, GFP_KERNEL);
461         if (!req)
462                 return -ENOMEM;
463
464         if (src_addr)
465                 memcpy(&req->src_addr, src_addr, ip_addr_size(src_addr));
466         memcpy(&req->dst_addr, dst_addr, ip_addr_size(dst_addr));
467         req->addr = addr;
468         req->callback = callback;
469         req->context = context;
470         req->client = client;
471         atomic_inc(&client->refcount);
472
473         src_in = (struct sockaddr *) &req->src_addr;
474         dst_in = (struct sockaddr *) &req->dst_addr;
475
476         req->status = addr_resolve_local(src_in, dst_in, addr);
477         if (req->status == -EADDRNOTAVAIL)
478                 req->status = addr_resolve_remote(src_in, dst_in, addr);
479
480         switch (req->status) {
481         case 0:
482                 req->timeout = jiffies;
483                 queue_req(req);
484                 break;
485         case -ENODATA:
486                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
487                 queue_req(req);
488                 addr_send_arp(dst_in);
489                 break;
490         default:
491                 ret = req->status;
492                 atomic_dec(&client->refcount);
493                 kfree(req);
494                 break;
495         }
496         return ret;
497 }
498 EXPORT_SYMBOL(rdma_resolve_ip);
499
500 void rdma_addr_cancel(struct rdma_dev_addr *addr)
501 {
502         struct addr_req *req, *temp_req;
503
504         mutex_lock(&lock);
505         list_for_each_entry_safe(req, temp_req, &req_list, list) {
506                 if (req->addr == addr) {
507                         req->status = -ECANCELED;
508                         req->timeout = jiffies;
509                         list_move(&req->list, &req_list);
510                         set_timeout(req->timeout);
511                         break;
512                 }
513         }
514         mutex_unlock(&lock);
515 }
516 EXPORT_SYMBOL(rdma_addr_cancel);
517
518 static int netevent_callback(struct notifier_block *self, unsigned long event,
519         void *ctx)
520 {
521         if (event == NETEVENT_NEIGH_UPDATE) {
522                 struct neighbour *neigh = ctx;
523
524                 if (neigh->nud_state & NUD_VALID) {
525                         set_timeout(jiffies);
526                 }
527         }
528         return 0;
529 }
530
531 static struct notifier_block nb = {
532         .notifier_call = netevent_callback
533 };
534
535 static int __init addr_init(void)
536 {
537         addr_wq = create_singlethread_workqueue("ib_addr");
538         if (!addr_wq)
539                 return -ENOMEM;
540
541         register_netevent_notifier(&nb);
542         return 0;
543 }
544
545 static void __exit addr_cleanup(void)
546 {
547         unregister_netevent_notifier(&nb);
548         destroy_workqueue(addr_wq);
549 }
550
551 module_init(addr_init);
552 module_exit(addr_cleanup);