1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36#include <linux/mutex.h>
37#include <linux/inetdevice.h>
38#include <linux/slab.h>
39#include <linux/workqueue.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
48MODULE_AUTHOR("Sean Hefty");
49MODULE_DESCRIPTION("IB Address Translation");
50MODULE_LICENSE("Dual BSD/GPL");
51
52struct 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
65static void process_req(struct work_struct *work);
66
67static DEFINE_MUTEX(lock);
68static LIST_HEAD(req_list);
69static DECLARE_DELAYED_WORK(work, process_req);
70static struct workqueue_struct *addr_wq;
71
72void rdma_addr_register_client(struct rdma_addr_client *client)
73{
74 atomic_set(&client->refcount, 1);
75 init_completion(&client->comp);
76}
77EXPORT_SYMBOL(rdma_addr_register_client);
78
79static inline void put_client(struct rdma_addr_client *client)
80{
81 if (atomic_dec_and_test(&client->refcount))
82 complete(&client->comp);
83}
84
85void rdma_addr_unregister_client(struct rdma_addr_client *client)
86{
87 put_client(client);
88 wait_for_completion(&client->comp);
89}
90EXPORT_SYMBOL(rdma_addr_unregister_client);
91
92int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
93 const unsigned char *dst_dev_addr)
94{
95 dev_addr->dev_type = dev->type;
96 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
97 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
98 if (dst_dev_addr)
99 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
100 dev_addr->bound_dev_if = dev->ifindex;
101 return 0;
102}
103EXPORT_SYMBOL(rdma_copy_addr);
104
105int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
106{
107 struct net_device *dev;
108 int ret = -EADDRNOTAVAIL;
109
110 if (dev_addr->bound_dev_if) {
111 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
112 if (!dev)
113 return -ENODEV;
114 ret = rdma_copy_addr(dev_addr, dev, NULL);
115 dev_put(dev);
116 return ret;
117 }
118
119 switch (addr->sa_family) {
120 case AF_INET:
121 dev = ip_dev_find(&init_net,
122 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
123
124 if (!dev)
125 return ret;
126
127 ret = rdma_copy_addr(dev_addr, dev, NULL);
128 dev_put(dev);
129 break;
130
131#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
132 case AF_INET6:
133 rcu_read_lock();
134 for_each_netdev_rcu(&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 rcu_read_unlock();
143 break;
144#endif
145 }
146 return ret;
147}
148EXPORT_SYMBOL(rdma_translate_ip);
149
150static void set_timeout(unsigned long time)
151{
152 unsigned long delay;
153
154 cancel_delayed_work(&work);
155
156 delay = time - jiffies;
157 if ((long)delay <= 0)
158 delay = 1;
159
160 queue_delayed_work(addr_wq, &work, delay);
161}
162
163static void queue_req(struct addr_req *req)
164{
165 struct addr_req *temp_req;
166
167 mutex_lock(&lock);
168 list_for_each_entry_reverse(temp_req, &req_list, list) {
169 if (time_after_eq(req->timeout, temp_req->timeout))
170 break;
171 }
172
173 list_add(&req->list, &temp_req->list);
174
175 if (req_list.next == &req->list)
176 set_timeout(req->timeout);
177 mutex_unlock(&lock);
178}
179
180static int addr4_resolve(struct sockaddr_in *src_in,
181 struct sockaddr_in *dst_in,
182 struct rdma_dev_addr *addr)
183{
184 __be32 src_ip = src_in->sin_addr.s_addr;
185 __be32 dst_ip = dst_in->sin_addr.s_addr;
186 struct flowi fl;
187 struct rtable *rt;
188 struct neighbour *neigh;
189 int ret;
190
191 memset(&fl, 0, sizeof fl);
192 fl.nl_u.ip4_u.daddr = dst_ip;
193 fl.nl_u.ip4_u.saddr = src_ip;
194 fl.oif = addr->bound_dev_if;
195
196 ret = ip_route_output_key(&init_net, &rt, &fl);
197 if (ret)
198 goto out;
199
200 src_in->sin_family = AF_INET;
201 src_in->sin_addr.s_addr = rt->rt_src;
202
203 if (rt->dst.dev->flags & IFF_LOOPBACK) {
204 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
205 if (!ret)
206 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
207 goto put;
208 }
209
210
211 if (rt->dst.dev->flags & IFF_NOARP) {
212 rdma_copy_addr(addr, rt->dst.dev, NULL);
213 goto put;
214 }
215
216 neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->dst.dev);
217 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
218 neigh_event_send(rt->dst.neighbour, NULL);
219 ret = -ENODATA;
220 if (neigh)
221 goto release;
222 goto put;
223 }
224
225 ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
226release:
227 neigh_release(neigh);
228put:
229 ip_rt_put(rt);
230out:
231 return ret;
232}
233
234#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
235static int addr6_resolve(struct sockaddr_in6 *src_in,
236 struct sockaddr_in6 *dst_in,
237 struct rdma_dev_addr *addr)
238{
239 struct flowi fl;
240 struct neighbour *neigh;
241 struct dst_entry *dst;
242 int ret;
243
244 memset(&fl, 0, sizeof fl);
245 ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
246 ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
247 fl.oif = addr->bound_dev_if;
248
249 dst = ip6_route_output(&init_net, NULL, &fl);
250 if ((ret = dst->error))
251 goto put;
252
253 if (ipv6_addr_any(&fl.fl6_src)) {
254 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
255 &fl.fl6_dst, 0, &fl.fl6_src);
256 if (ret)
257 goto put;
258
259 src_in->sin6_family = AF_INET6;
260 ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
261 }
262
263 if (dst->dev->flags & IFF_LOOPBACK) {
264 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
265 if (!ret)
266 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
267 goto put;
268 }
269
270
271 if (dst->dev->flags & IFF_NOARP) {
272 ret = rdma_copy_addr(addr, dst->dev, NULL);
273 goto put;
274 }
275
276 neigh = dst->neighbour;
277 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
278 neigh_event_send(dst->neighbour, NULL);
279 ret = -ENODATA;
280 goto put;
281 }
282
283 ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
284put:
285 dst_release(dst);
286 return ret;
287}
288#else
289static 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
297static 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
309static 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
346int 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;
398err:
399 kfree(req);
400 return ret;
401}
402EXPORT_SYMBOL(rdma_resolve_ip);
403
404void 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}
420EXPORT_SYMBOL(rdma_addr_cancel);
421
422static 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
435static struct notifier_block nb = {
436 .notifier_call = netevent_callback
437};
438
439static 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
449static void __exit addr_cleanup(void)
450{
451 unregister_netevent_notifier(&nb);
452 destroy_workqueue(addr_wq);
453}
454
455module_init(addr_init);
456module_exit(addr_cleanup);
457
