Merge commit 'jk/jk-merge'
[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 <rdma/ib_addr.h>
45
46 MODULE_AUTHOR("Sean Hefty");
47 MODULE_DESCRIPTION("IB Address Translation");
48 MODULE_LICENSE("Dual BSD/GPL");
49
50 struct addr_req {
51         struct list_head list;
52         struct sockaddr src_addr;
53         struct sockaddr dst_addr;
54         struct rdma_dev_addr *addr;
55         struct rdma_addr_client *client;
56         void *context;
57         void (*callback)(int status, struct sockaddr *src_addr,
58                          struct rdma_dev_addr *addr, void *context);
59         unsigned long timeout;
60         int status;
61 };
62
63 static void process_req(struct work_struct *work);
64
65 static DEFINE_MUTEX(lock);
66 static LIST_HEAD(req_list);
67 static DECLARE_DELAYED_WORK(work, process_req);
68 static struct workqueue_struct *addr_wq;
69
70 void rdma_addr_register_client(struct rdma_addr_client *client)
71 {
72         atomic_set(&client->refcount, 1);
73         init_completion(&client->comp);
74 }
75 EXPORT_SYMBOL(rdma_addr_register_client);
76
77 static inline void put_client(struct rdma_addr_client *client)
78 {
79         if (atomic_dec_and_test(&client->refcount))
80                 complete(&client->comp);
81 }
82
83 void rdma_addr_unregister_client(struct rdma_addr_client *client)
84 {
85         put_client(client);
86         wait_for_completion(&client->comp);
87 }
88 EXPORT_SYMBOL(rdma_addr_unregister_client);
89
90 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
91                      const unsigned char *dst_dev_addr)
92 {
93         switch (dev->type) {
94         case ARPHRD_INFINIBAND:
95                 dev_addr->dev_type = RDMA_NODE_IB_CA;
96                 break;
97         case ARPHRD_ETHER:
98                 dev_addr->dev_type = RDMA_NODE_RNIC;
99                 break;
100         default:
101                 return -EADDRNOTAVAIL;
102         }
103
104         memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
105         memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
106         if (dst_dev_addr)
107                 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
108         dev_addr->src_dev = dev;
109         return 0;
110 }
111 EXPORT_SYMBOL(rdma_copy_addr);
112
113 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
114 {
115         struct net_device *dev;
116         __be32 ip = ((struct sockaddr_in *) addr)->sin_addr.s_addr;
117         int ret;
118
119         dev = ip_dev_find(&init_net, ip);
120         if (!dev)
121                 return -EADDRNOTAVAIL;
122
123         ret = rdma_copy_addr(dev_addr, dev, NULL);
124         dev_put(dev);
125         return ret;
126 }
127 EXPORT_SYMBOL(rdma_translate_ip);
128
129 static void set_timeout(unsigned long time)
130 {
131         unsigned long delay;
132
133         cancel_delayed_work(&work);
134
135         delay = time - jiffies;
136         if ((long)delay <= 0)
137                 delay = 1;
138
139         queue_delayed_work(addr_wq, &work, delay);
140 }
141
142 static void queue_req(struct addr_req *req)
143 {
144         struct addr_req *temp_req;
145
146         mutex_lock(&lock);
147         list_for_each_entry_reverse(temp_req, &req_list, list) {
148                 if (time_after_eq(req->timeout, temp_req->timeout))
149                         break;
150         }
151
152         list_add(&req->list, &temp_req->list);
153
154         if (req_list.next == &req->list)
155                 set_timeout(req->timeout);
156         mutex_unlock(&lock);
157 }
158
159 static void addr_send_arp(struct sockaddr_in *dst_in)
160 {
161         struct rtable *rt;
162         struct flowi fl;
163         __be32 dst_ip = dst_in->sin_addr.s_addr;
164
165         memset(&fl, 0, sizeof fl);
166         fl.nl_u.ip4_u.daddr = dst_ip;
167         if (ip_route_output_key(&init_net, &rt, &fl))
168                 return;
169
170         neigh_event_send(rt->u.dst.neighbour, NULL);
171         ip_rt_put(rt);
172 }
173
174 static int addr_resolve_remote(struct sockaddr_in *src_in,
175                                struct sockaddr_in *dst_in,
176                                struct rdma_dev_addr *addr)
177 {
178         __be32 src_ip = src_in->sin_addr.s_addr;
179         __be32 dst_ip = dst_in->sin_addr.s_addr;
180         struct flowi fl;
181         struct rtable *rt;
182         struct neighbour *neigh;
183         int ret;
184
185         memset(&fl, 0, sizeof fl);
186         fl.nl_u.ip4_u.daddr = dst_ip;
187         fl.nl_u.ip4_u.saddr = src_ip;
188         ret = ip_route_output_key(&init_net, &rt, &fl);
189         if (ret)
190                 goto out;
191
192         /* If the device does ARP internally, return 'done' */
193         if (rt->idev->dev->flags & IFF_NOARP) {
194                 rdma_copy_addr(addr, rt->idev->dev, NULL);
195                 goto put;
196         }
197
198         neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
199         if (!neigh) {
200                 ret = -ENODATA;
201                 goto put;
202         }
203
204         if (!(neigh->nud_state & NUD_VALID)) {
205                 ret = -ENODATA;
206                 goto release;
207         }
208
209         if (!src_ip) {
210                 src_in->sin_family = dst_in->sin_family;
211                 src_in->sin_addr.s_addr = rt->rt_src;
212         }
213
214         ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
215 release:
216         neigh_release(neigh);
217 put:
218         ip_rt_put(rt);
219 out:
220         return ret;
221 }
222
223 static void process_req(struct work_struct *work)
224 {
225         struct addr_req *req, *temp_req;
226         struct sockaddr_in *src_in, *dst_in;
227         struct list_head done_list;
228
229         INIT_LIST_HEAD(&done_list);
230
231         mutex_lock(&lock);
232         list_for_each_entry_safe(req, temp_req, &req_list, list) {
233                 if (req->status == -ENODATA) {
234                         src_in = (struct sockaddr_in *) &req->src_addr;
235                         dst_in = (struct sockaddr_in *) &req->dst_addr;
236                         req->status = addr_resolve_remote(src_in, dst_in,
237                                                           req->addr);
238                         if (req->status && time_after_eq(jiffies, req->timeout))
239                                 req->status = -ETIMEDOUT;
240                         else if (req->status == -ENODATA)
241                                 continue;
242                 }
243                 list_move_tail(&req->list, &done_list);
244         }
245
246         if (!list_empty(&req_list)) {
247                 req = list_entry(req_list.next, struct addr_req, list);
248                 set_timeout(req->timeout);
249         }
250         mutex_unlock(&lock);
251
252         list_for_each_entry_safe(req, temp_req, &done_list, list) {
253                 list_del(&req->list);
254                 req->callback(req->status, &req->src_addr, req->addr,
255                               req->context);
256                 put_client(req->client);
257                 kfree(req);
258         }
259 }
260
261 static int addr_resolve_local(struct sockaddr_in *src_in,
262                               struct sockaddr_in *dst_in,
263                               struct rdma_dev_addr *addr)
264 {
265         struct net_device *dev;
266         __be32 src_ip = src_in->sin_addr.s_addr;
267         __be32 dst_ip = dst_in->sin_addr.s_addr;
268         int ret;
269
270         dev = ip_dev_find(&init_net, dst_ip);
271         if (!dev)
272                 return -EADDRNOTAVAIL;
273
274         if (ipv4_is_zeronet(src_ip)) {
275                 src_in->sin_family = dst_in->sin_family;
276                 src_in->sin_addr.s_addr = dst_ip;
277                 ret = rdma_copy_addr(addr, dev, dev->dev_addr);
278         } else if (ipv4_is_loopback(src_ip)) {
279                 ret = rdma_translate_ip((struct sockaddr *)dst_in, addr);
280                 if (!ret)
281                         memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
282         } else {
283                 ret = rdma_translate_ip((struct sockaddr *)src_in, addr);
284                 if (!ret)
285                         memcpy(addr->dst_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
286         }
287
288         dev_put(dev);
289         return ret;
290 }
291
292 int rdma_resolve_ip(struct rdma_addr_client *client,
293                     struct sockaddr *src_addr, struct sockaddr *dst_addr,
294                     struct rdma_dev_addr *addr, int timeout_ms,
295                     void (*callback)(int status, struct sockaddr *src_addr,
296                                      struct rdma_dev_addr *addr, void *context),
297                     void *context)
298 {
299         struct sockaddr_in *src_in, *dst_in;
300         struct addr_req *req;
301         int ret = 0;
302
303         req = kzalloc(sizeof *req, GFP_KERNEL);
304         if (!req)
305                 return -ENOMEM;
306
307         if (src_addr)
308                 memcpy(&req->src_addr, src_addr, ip_addr_size(src_addr));
309         memcpy(&req->dst_addr, dst_addr, ip_addr_size(dst_addr));
310         req->addr = addr;
311         req->callback = callback;
312         req->context = context;
313         req->client = client;
314         atomic_inc(&client->refcount);
315
316         src_in = (struct sockaddr_in *) &req->src_addr;
317         dst_in = (struct sockaddr_in *) &req->dst_addr;
318
319         req->status = addr_resolve_local(src_in, dst_in, addr);
320         if (req->status == -EADDRNOTAVAIL)
321                 req->status = addr_resolve_remote(src_in, dst_in, addr);
322
323         switch (req->status) {
324         case 0:
325                 req->timeout = jiffies;
326                 queue_req(req);
327                 break;
328         case -ENODATA:
329                 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
330                 queue_req(req);
331                 addr_send_arp(dst_in);
332                 break;
333         default:
334                 ret = req->status;
335                 atomic_dec(&client->refcount);
336                 kfree(req);
337                 break;
338         }
339         return ret;
340 }
341 EXPORT_SYMBOL(rdma_resolve_ip);
342
343 void rdma_addr_cancel(struct rdma_dev_addr *addr)
344 {
345         struct addr_req *req, *temp_req;
346
347         mutex_lock(&lock);
348         list_for_each_entry_safe(req, temp_req, &req_list, list) {
349                 if (req->addr == addr) {
350                         req->status = -ECANCELED;
351                         req->timeout = jiffies;
352                         list_move(&req->list, &req_list);
353                         set_timeout(req->timeout);
354                         break;
355                 }
356         }
357         mutex_unlock(&lock);
358 }
359 EXPORT_SYMBOL(rdma_addr_cancel);
360
361 static int netevent_callback(struct notifier_block *self, unsigned long event,
362         void *ctx)
363 {
364         if (event == NETEVENT_NEIGH_UPDATE) {
365                 struct neighbour *neigh = ctx;
366
367                 if (neigh->nud_state & NUD_VALID) {
368                         set_timeout(jiffies);
369                 }
370         }
371         return 0;
372 }
373
374 static struct notifier_block nb = {
375         .notifier_call = netevent_callback
376 };
377
378 static int addr_init(void)
379 {
380         addr_wq = create_singlethread_workqueue("ib_addr");
381         if (!addr_wq)
382                 return -ENOMEM;
383
384         register_netevent_notifier(&nb);
385         return 0;
386 }
387
388 static void addr_cleanup(void)
389 {
390         unregister_netevent_notifier(&nb);
391         destroy_workqueue(addr_wq);
392 }
393
394 module_init(addr_init);
395 module_exit(addr_cleanup);