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.
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:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
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.
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
36 #include <linux/mutex.h>
37 #include <linux/inetdevice.h>
38 #include <linux/workqueue.h>
39 #include <net/route.h>
40 #include <net/netevent.h>
41 #include <rdma/ib_addr.h>
43 MODULE_AUTHOR("Sean Hefty");
44 MODULE_DESCRIPTION("IB Address Translation");
45 MODULE_LICENSE("Dual BSD/GPL");
48 struct list_head list;
49 struct sockaddr_storage src_addr;
50 struct sockaddr_storage dst_addr;
51 struct rdma_dev_addr *addr;
52 struct rdma_addr_client *client;
54 void (*callback)(int status, struct sockaddr *src_addr,
55 struct rdma_dev_addr *addr, void *context);
56 unsigned long timeout;
60 static void process_req(struct work_struct *work);
62 static DEFINE_MUTEX(lock);
63 static LIST_HEAD(req_list);
64 static struct delayed_work work;
65 static struct workqueue_struct *addr_wq;
67 void rdma_addr_register_client(struct rdma_addr_client *client)
69 atomic_set(&client->refcount, 1);
70 init_completion(&client->comp);
72 EXPORT_SYMBOL(rdma_addr_register_client);
74 static inline void put_client(struct rdma_addr_client *client)
76 if (atomic_dec_and_test(&client->refcount))
77 complete(&client->comp);
80 void rdma_addr_unregister_client(struct rdma_addr_client *client)
83 wait_for_completion(&client->comp);
85 EXPORT_SYMBOL(rdma_addr_unregister_client);
88 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
89 const unsigned char *dst_dev_addr)
91 dev_addr->dev_type = dev->type;
92 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
93 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
95 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
96 dev_addr->bound_dev_if = dev->ifindex;
100 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct ifnet *dev,
101 const unsigned char *dst_dev_addr)
103 if (dev->if_type == IFT_INFINIBAND)
104 dev_addr->dev_type = ARPHRD_INFINIBAND;
105 else if (dev->if_type == IFT_ETHER)
106 dev_addr->dev_type = ARPHRD_ETHER;
108 dev_addr->dev_type = 0;
109 memcpy(dev_addr->src_dev_addr, IF_LLADDR(dev), dev->if_addrlen);
110 memcpy(dev_addr->broadcast, __DECONST(char *, dev->if_broadcastaddr),
113 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, dev->if_addrlen);
114 dev_addr->bound_dev_if = dev->if_index;
118 EXPORT_SYMBOL(rdma_copy_addr);
120 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
122 struct net_device *dev;
123 int ret = -EADDRNOTAVAIL;
125 if (dev_addr->bound_dev_if) {
126 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
129 ret = rdma_copy_addr(dev_addr, dev, NULL);
134 switch (addr->sa_family) {
137 dev = ip_dev_find(NULL,
138 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
143 ret = rdma_copy_addr(dev_addr, dev, NULL);
151 read_lock(&dev_base_lock);
152 for_each_netdev(&init_net, dev) {
153 if (ipv6_chk_addr(&init_net,
154 &((struct sockaddr_in6 *) addr)->sin6_addr,
156 ret = rdma_copy_addr(dev_addr, dev, NULL);
160 read_unlock(&dev_base_lock);
163 struct sockaddr_in6 *sin6;
167 sin6 = (struct sockaddr_in6 *)addr;
168 port = sin6->sin6_port;
170 ifa = ifa_ifwithaddr(addr);
171 sin6->sin6_port = port;
176 ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
186 EXPORT_SYMBOL(rdma_translate_ip);
188 static void set_timeout(unsigned long time)
192 cancel_delayed_work(&work);
194 delay = time - jiffies;
195 if ((long)delay <= 0)
198 queue_delayed_work(addr_wq, &work, delay);
201 static void queue_req(struct addr_req *req)
203 struct addr_req *temp_req;
206 list_for_each_entry_reverse(temp_req, &req_list, list) {
207 if (time_after_eq(req->timeout, temp_req->timeout))
211 list_add(&req->list, &temp_req->list);
213 if (req_list.next == &req->list)
214 set_timeout(req->timeout);
219 static int addr4_resolve(struct sockaddr_in *src_in,
220 struct sockaddr_in *dst_in,
221 struct rdma_dev_addr *addr)
223 __be32 src_ip = src_in->sin_addr.s_addr;
224 __be32 dst_ip = dst_in->sin_addr.s_addr;
227 struct neighbour *neigh;
230 memset(&fl, 0, sizeof fl);
231 fl.nl_u.ip4_u.daddr = dst_ip;
232 fl.nl_u.ip4_u.saddr = src_ip;
233 fl.oif = addr->bound_dev_if;
235 ret = ip_route_output_key(&init_net, &rt, &fl);
239 src_in->sin_family = AF_INET;
240 src_in->sin_addr.s_addr = rt->rt_src;
242 if (rt->idev->dev->flags & IFF_LOOPBACK) {
243 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
245 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
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);
255 neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
256 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
257 neigh_event_send(rt->u.dst.neighbour, NULL);
264 ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
266 neigh_release(neigh);
274 static int addr6_resolve(struct sockaddr_in6 *src_in,
275 struct sockaddr_in6 *dst_in,
276 struct rdma_dev_addr *addr)
279 struct neighbour *neigh;
280 struct dst_entry *dst;
283 memset(&fl, 0, sizeof fl);
284 ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
285 ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
286 fl.oif = addr->bound_dev_if;
288 dst = ip6_route_output(&init_net, NULL, &fl);
289 if ((ret = dst->error))
292 if (ipv6_addr_any(&fl.fl6_src)) {
293 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
294 &fl.fl6_dst, 0, &fl.fl6_src);
298 src_in->sin6_family = AF_INET6;
299 ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
302 if (dst->dev->flags & IFF_LOOPBACK) {
303 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
305 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
309 /* If the device does ARP internally, return 'done' */
310 if (dst->dev->flags & IFF_NOARP) {
311 ret = rdma_copy_addr(addr, dst->dev, NULL);
315 neigh = dst->neighbour;
316 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
317 neigh_event_send(dst->neighbour, NULL);
322 ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
328 static int addr6_resolve(struct sockaddr_in6 *src_in,
329 struct sockaddr_in6 *dst_in,
330 struct rdma_dev_addr *addr)
332 return -EADDRNOTAVAIL;
337 #include <netinet/if_ether.h>
339 static int addr_resolve(struct sockaddr *src_in,
340 struct sockaddr *dst_in,
341 struct rdma_dev_addr *addr)
343 struct sockaddr_in *sin;
344 struct sockaddr_in6 *sin6;
347 #if defined(INET) || defined(INET6)
352 u_char edst[MAX_ADDR_LEN];
358 * Determine whether the address is unicast, multicast, or broadcast
359 * and whether the source interface is valid.
367 switch (dst_in->sa_family) {
370 sin = (struct sockaddr_in *)dst_in;
371 if (sin->sin_addr.s_addr == INADDR_BROADCAST)
373 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
375 sin = (struct sockaddr_in *)src_in;
376 if (sin->sin_addr.s_addr != INADDR_ANY) {
378 * Address comparison fails if the port is set
379 * cache it here to be restored later.
381 port = sin->sin_port;
383 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
390 sin6 = (struct sockaddr_in6 *)dst_in;
391 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
393 sin6 = (struct sockaddr_in6 *)src_in;
394 if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
395 port = sin6->sin6_port;
405 * If we have a source address to use look it up first and verify
406 * that it is a local interface.
409 ifa = ifa_ifwithaddr(src_in);
411 sin->sin_port = port;
413 sin6->sin6_port = port;
422 * Make sure the route exists and has a valid link.
424 rte = rtalloc1(dst_in, 1, 0);
425 if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
428 return -EHOSTUNREACH;
431 * If it's not multicast or broadcast and the route doesn't match the
432 * requested interface return unreachable. Otherwise fetch the
433 * correct interface pointer and unlock the route.
435 if (multi || bcast) {
439 } else if (ifp && ifp != rte->rt_ifp) {
449 return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
451 struct sockaddr *llsa;
453 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
456 error = rdma_copy_addr(addr, ifp,
457 LLADDR((struct sockaddr_dl *)llsa));
458 free(llsa, M_IFMADDR);
462 * Resolve the link local address.
464 switch (dst_in->sa_family) {
467 error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
472 error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
476 /* XXX: Shouldn't happen. */
481 return rdma_copy_addr(addr, ifp, edst);
482 if (error == EWOULDBLOCK)
489 static void process_req(struct work_struct *work)
491 struct addr_req *req, *temp_req;
492 struct sockaddr *src_in, *dst_in;
493 struct list_head done_list;
495 INIT_LIST_HEAD(&done_list);
498 list_for_each_entry_safe(req, temp_req, &req_list, list) {
499 if (req->status == -ENODATA) {
500 src_in = (struct sockaddr *) &req->src_addr;
501 dst_in = (struct sockaddr *) &req->dst_addr;
502 req->status = addr_resolve(src_in, dst_in, req->addr);
503 if (req->status && time_after_eq(jiffies, req->timeout))
504 req->status = -ETIMEDOUT;
505 else if (req->status == -ENODATA)
508 list_move_tail(&req->list, &done_list);
511 if (!list_empty(&req_list)) {
512 req = list_entry(req_list.next, struct addr_req, list);
513 set_timeout(req->timeout);
517 list_for_each_entry_safe(req, temp_req, &done_list, list) {
518 list_del(&req->list);
519 req->callback(req->status, (struct sockaddr *) &req->src_addr,
520 req->addr, req->context);
521 put_client(req->client);
526 int rdma_resolve_ip(struct rdma_addr_client *client,
527 struct sockaddr *src_addr, struct sockaddr *dst_addr,
528 struct rdma_dev_addr *addr, int timeout_ms,
529 void (*callback)(int status, struct sockaddr *src_addr,
530 struct rdma_dev_addr *addr, void *context),
533 struct sockaddr *src_in, *dst_in;
534 struct addr_req *req;
537 req = kzalloc(sizeof *req, GFP_KERNEL);
541 src_in = (struct sockaddr *) &req->src_addr;
542 dst_in = (struct sockaddr *) &req->dst_addr;
545 if (src_addr->sa_family != dst_addr->sa_family) {
550 memcpy(src_in, src_addr, ip_addr_size(src_addr));
552 src_in->sa_family = dst_addr->sa_family;
555 memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
557 req->callback = callback;
558 req->context = context;
559 req->client = client;
560 atomic_inc(&client->refcount);
562 req->status = addr_resolve(src_in, dst_in, addr);
563 switch (req->status) {
565 req->timeout = jiffies;
569 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
574 atomic_dec(&client->refcount);
582 EXPORT_SYMBOL(rdma_resolve_ip);
584 void rdma_addr_cancel(struct rdma_dev_addr *addr)
586 struct addr_req *req, *temp_req;
589 list_for_each_entry_safe(req, temp_req, &req_list, list) {
590 if (req->addr == addr) {
591 req->status = -ECANCELED;
592 req->timeout = jiffies;
593 list_move(&req->list, &req_list);
594 set_timeout(req->timeout);
600 EXPORT_SYMBOL(rdma_addr_cancel);
602 static int netevent_callback(struct notifier_block *self, unsigned long event,
605 if (event == NETEVENT_NEIGH_UPDATE) {
607 struct neighbour *neigh = ctx;
609 if (neigh->nud_state & NUD_VALID) {
610 set_timeout(jiffies);
613 set_timeout(jiffies);
619 static struct notifier_block nb = {
620 .notifier_call = netevent_callback
623 static int addr_init(void)
625 INIT_DELAYED_WORK(&work, process_req);
626 addr_wq = create_singlethread_workqueue("ib_addr");
630 register_netevent_notifier(&nb);
634 static void addr_cleanup(void)
636 unregister_netevent_notifier(&nb);
637 destroy_workqueue(addr_wq);
640 module_init(addr_init);
641 module_exit(addr_cleanup);