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>
40 #include <net/neighbour.h>
41 #include <net/route.h>
42 #include <net/netevent.h>
43 #include <net/addrconf.h>
44 #include <net/ip6_route.h>
45 #include <rdma/ib_addr.h>
47 MODULE_AUTHOR("Sean Hefty");
48 MODULE_DESCRIPTION("IB Address Translation");
49 MODULE_LICENSE("Dual BSD/GPL");
52 struct list_head list;
53 struct sockaddr_storage src_addr;
54 struct sockaddr_storage dst_addr;
55 struct rdma_dev_addr *addr;
56 struct rdma_addr_client *client;
58 void (*callback)(int status, struct sockaddr *src_addr,
59 struct rdma_dev_addr *addr, void *context);
60 unsigned long timeout;
64 static void process_req(struct work_struct *work);
66 static DEFINE_MUTEX(lock);
67 static LIST_HEAD(req_list);
68 static struct delayed_work work;
69 static struct workqueue_struct *addr_wq;
71 void rdma_addr_register_client(struct rdma_addr_client *client)
73 atomic_set(&client->refcount, 1);
74 init_completion(&client->comp);
76 EXPORT_SYMBOL(rdma_addr_register_client);
78 static inline void put_client(struct rdma_addr_client *client)
80 if (atomic_dec_and_test(&client->refcount))
81 complete(&client->comp);
84 void rdma_addr_unregister_client(struct rdma_addr_client *client)
87 wait_for_completion(&client->comp);
89 EXPORT_SYMBOL(rdma_addr_unregister_client);
92 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
93 const unsigned char *dst_dev_addr)
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);
99 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
100 dev_addr->bound_dev_if = dev->ifindex;
104 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct ifnet *dev,
105 const unsigned char *dst_dev_addr)
107 if (dev->if_type == IFT_INFINIBAND)
108 dev_addr->dev_type = ARPHRD_INFINIBAND;
109 else if (dev->if_type == IFT_ETHER)
110 dev_addr->dev_type = ARPHRD_ETHER;
112 dev_addr->dev_type = 0;
113 memcpy(dev_addr->src_dev_addr, IF_LLADDR(dev), dev->if_addrlen);
114 memcpy(dev_addr->broadcast, __DECONST(char *, dev->if_broadcastaddr),
117 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, dev->if_addrlen);
118 dev_addr->bound_dev_if = dev->if_index;
122 EXPORT_SYMBOL(rdma_copy_addr);
124 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
126 struct net_device *dev;
127 int ret = -EADDRNOTAVAIL;
129 if (dev_addr->bound_dev_if) {
130 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
133 ret = rdma_copy_addr(dev_addr, dev, NULL);
138 switch (addr->sa_family) {
140 dev = ip_dev_find(NULL,
141 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
146 ret = rdma_copy_addr(dev_addr, dev, NULL);
153 read_lock(&dev_base_lock);
154 for_each_netdev(&init_net, dev) {
155 if (ipv6_chk_addr(&init_net,
156 &((struct sockaddr_in6 *) addr)->sin6_addr,
158 ret = rdma_copy_addr(dev_addr, dev, NULL);
162 read_unlock(&dev_base_lock);
165 struct sockaddr_in6 *sin6;
169 sin6 = (struct sockaddr_in6 *)addr;
170 port = sin6->sin6_port;
172 ifa = ifa_ifwithaddr(addr);
173 sin6->sin6_port = port;
178 ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
188 EXPORT_SYMBOL(rdma_translate_ip);
190 static void set_timeout(unsigned long time)
194 cancel_delayed_work(&work);
196 delay = time - jiffies;
197 if ((long)delay <= 0)
200 queue_delayed_work(addr_wq, &work, delay);
203 static void queue_req(struct addr_req *req)
205 struct addr_req *temp_req;
208 list_for_each_entry_reverse(temp_req, &req_list, list) {
209 if (time_after_eq(req->timeout, temp_req->timeout))
213 list_add(&req->list, &temp_req->list);
215 if (req_list.next == &req->list)
216 set_timeout(req->timeout);
221 static int addr4_resolve(struct sockaddr_in *src_in,
222 struct sockaddr_in *dst_in,
223 struct rdma_dev_addr *addr)
225 __be32 src_ip = src_in->sin_addr.s_addr;
226 __be32 dst_ip = dst_in->sin_addr.s_addr;
229 struct neighbour *neigh;
232 memset(&fl, 0, sizeof fl);
233 fl.nl_u.ip4_u.daddr = dst_ip;
234 fl.nl_u.ip4_u.saddr = src_ip;
235 fl.oif = addr->bound_dev_if;
237 ret = ip_route_output_key(&init_net, &rt, &fl);
241 src_in->sin_family = AF_INET;
242 src_in->sin_addr.s_addr = rt->rt_src;
244 if (rt->idev->dev->flags & IFF_LOOPBACK) {
245 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
247 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
251 /* If the device does ARP internally, return 'done' */
252 if (rt->idev->dev->flags & IFF_NOARP) {
253 rdma_copy_addr(addr, rt->idev->dev, NULL);
257 neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
258 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
259 neigh_event_send(rt->u.dst.neighbour, NULL);
266 ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
268 neigh_release(neigh);
276 static int addr6_resolve(struct sockaddr_in6 *src_in,
277 struct sockaddr_in6 *dst_in,
278 struct rdma_dev_addr *addr)
281 struct neighbour *neigh;
282 struct dst_entry *dst;
285 memset(&fl, 0, sizeof fl);
286 ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
287 ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
288 fl.oif = addr->bound_dev_if;
290 dst = ip6_route_output(&init_net, NULL, &fl);
291 if ((ret = dst->error))
294 if (ipv6_addr_any(&fl.fl6_src)) {
295 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
296 &fl.fl6_dst, 0, &fl.fl6_src);
300 src_in->sin6_family = AF_INET6;
301 ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
304 if (dst->dev->flags & IFF_LOOPBACK) {
305 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
307 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
311 /* If the device does ARP internally, return 'done' */
312 if (dst->dev->flags & IFF_NOARP) {
313 ret = rdma_copy_addr(addr, dst->dev, NULL);
317 neigh = dst->neighbour;
318 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
319 neigh_event_send(dst->neighbour, NULL);
324 ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
330 static int addr6_resolve(struct sockaddr_in6 *src_in,
331 struct sockaddr_in6 *dst_in,
332 struct rdma_dev_addr *addr)
334 return -EADDRNOTAVAIL;
339 #include <netinet/if_ether.h>
341 static int addr_resolve(struct sockaddr *src_in,
342 struct sockaddr *dst_in,
343 struct rdma_dev_addr *addr)
345 struct sockaddr_in *sin;
346 struct sockaddr_in6 *sin6;
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) {
369 sin = (struct sockaddr_in *)dst_in;
370 if (sin->sin_addr.s_addr == INADDR_BROADCAST)
372 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
374 sin = (struct sockaddr_in *)src_in;
375 if (sin->sin_addr.s_addr != INADDR_ANY) {
377 * Address comparison fails if the port is set
378 * cache it here to be restored later.
380 port = sin->sin_port;
382 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
388 sin6 = (struct sockaddr_in6 *)dst_in;
389 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
391 sin6 = (struct sockaddr_in6 *)src_in;
392 if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
393 port = sin6->sin6_port;
403 * If we have a source address to use look it up first and verify
404 * that it is a local interface.
407 ifa = ifa_ifwithaddr(src_in);
409 sin->sin_port = port;
411 sin6->sin6_port = port;
420 * Make sure the route exists and has a valid link.
422 rte = rtalloc1(dst_in, 1, 0);
423 if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
426 return -EHOSTUNREACH;
429 * If it's not multicast or broadcast and the route doesn't match the
430 * requested interface return unreachable. Otherwise fetch the
431 * correct interface pointer and unlock the route.
433 if (multi || bcast) {
437 } else if (ifp && ifp != rte->rt_ifp) {
447 return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
449 struct sockaddr *llsa;
451 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
454 error = rdma_copy_addr(addr, ifp,
455 LLADDR((struct sockaddr_dl *)llsa));
456 free(llsa, M_IFMADDR);
460 * Resolve the link local address.
462 if (dst_in->sa_family == AF_INET)
463 error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
466 error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
470 return rdma_copy_addr(addr, ifp, edst);
471 if (error == EWOULDBLOCK)
478 static void process_req(struct work_struct *work)
480 struct addr_req *req, *temp_req;
481 struct sockaddr *src_in, *dst_in;
482 struct list_head done_list;
484 INIT_LIST_HEAD(&done_list);
487 list_for_each_entry_safe(req, temp_req, &req_list, list) {
488 if (req->status == -ENODATA) {
489 src_in = (struct sockaddr *) &req->src_addr;
490 dst_in = (struct sockaddr *) &req->dst_addr;
491 req->status = addr_resolve(src_in, dst_in, req->addr);
492 if (req->status && time_after_eq(jiffies, req->timeout))
493 req->status = -ETIMEDOUT;
494 else if (req->status == -ENODATA)
497 list_move_tail(&req->list, &done_list);
500 if (!list_empty(&req_list)) {
501 req = list_entry(req_list.next, struct addr_req, list);
502 set_timeout(req->timeout);
506 list_for_each_entry_safe(req, temp_req, &done_list, list) {
507 list_del(&req->list);
508 req->callback(req->status, (struct sockaddr *) &req->src_addr,
509 req->addr, req->context);
510 put_client(req->client);
515 int rdma_resolve_ip(struct rdma_addr_client *client,
516 struct sockaddr *src_addr, struct sockaddr *dst_addr,
517 struct rdma_dev_addr *addr, int timeout_ms,
518 void (*callback)(int status, struct sockaddr *src_addr,
519 struct rdma_dev_addr *addr, void *context),
522 struct sockaddr *src_in, *dst_in;
523 struct addr_req *req;
526 req = kzalloc(sizeof *req, GFP_KERNEL);
530 src_in = (struct sockaddr *) &req->src_addr;
531 dst_in = (struct sockaddr *) &req->dst_addr;
534 if (src_addr->sa_family != dst_addr->sa_family) {
539 memcpy(src_in, src_addr, ip_addr_size(src_addr));
541 src_in->sa_family = dst_addr->sa_family;
544 memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
546 req->callback = callback;
547 req->context = context;
548 req->client = client;
549 atomic_inc(&client->refcount);
551 req->status = addr_resolve(src_in, dst_in, addr);
552 switch (req->status) {
554 req->timeout = jiffies;
558 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
563 atomic_dec(&client->refcount);
571 EXPORT_SYMBOL(rdma_resolve_ip);
573 void rdma_addr_cancel(struct rdma_dev_addr *addr)
575 struct addr_req *req, *temp_req;
578 list_for_each_entry_safe(req, temp_req, &req_list, list) {
579 if (req->addr == addr) {
580 req->status = -ECANCELED;
581 req->timeout = jiffies;
582 list_move(&req->list, &req_list);
583 set_timeout(req->timeout);
589 EXPORT_SYMBOL(rdma_addr_cancel);
591 static int netevent_callback(struct notifier_block *self, unsigned long event,
594 if (event == NETEVENT_NEIGH_UPDATE) {
596 struct neighbour *neigh = ctx;
598 if (neigh->nud_state & NUD_VALID) {
599 set_timeout(jiffies);
602 set_timeout(jiffies);
608 static struct notifier_block nb = {
609 .notifier_call = netevent_callback
612 static int addr_init(void)
614 INIT_DELAYED_WORK(&work, process_req);
615 addr_wq = create_singlethread_workqueue("ib_addr");
619 register_netevent_notifier(&nb);
623 static void addr_cleanup(void)
625 unregister_netevent_notifier(&nb);
626 destroy_workqueue(addr_wq);
629 module_init(addr_init);
630 module_exit(addr_cleanup);