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) {
141 dev = ip_dev_find(NULL,
142 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
147 ret = rdma_copy_addr(dev_addr, dev, NULL);
155 read_lock(&dev_base_lock);
156 for_each_netdev(&init_net, dev) {
157 if (ipv6_chk_addr(&init_net,
158 &((struct sockaddr_in6 *) addr)->sin6_addr,
160 ret = rdma_copy_addr(dev_addr, dev, NULL);
164 read_unlock(&dev_base_lock);
167 struct sockaddr_in6 *sin6;
171 sin6 = (struct sockaddr_in6 *)addr;
172 port = sin6->sin6_port;
174 ifa = ifa_ifwithaddr(addr);
175 sin6->sin6_port = port;
180 ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
190 EXPORT_SYMBOL(rdma_translate_ip);
192 static void set_timeout(unsigned long time)
196 cancel_delayed_work(&work);
198 delay = time - jiffies;
199 if ((long)delay <= 0)
202 queue_delayed_work(addr_wq, &work, delay);
205 static void queue_req(struct addr_req *req)
207 struct addr_req *temp_req;
210 list_for_each_entry_reverse(temp_req, &req_list, list) {
211 if (time_after_eq(req->timeout, temp_req->timeout))
215 list_add(&req->list, &temp_req->list);
217 if (req_list.next == &req->list)
218 set_timeout(req->timeout);
223 static int addr4_resolve(struct sockaddr_in *src_in,
224 struct sockaddr_in *dst_in,
225 struct rdma_dev_addr *addr)
227 __be32 src_ip = src_in->sin_addr.s_addr;
228 __be32 dst_ip = dst_in->sin_addr.s_addr;
231 struct neighbour *neigh;
234 memset(&fl, 0, sizeof fl);
235 fl.nl_u.ip4_u.daddr = dst_ip;
236 fl.nl_u.ip4_u.saddr = src_ip;
237 fl.oif = addr->bound_dev_if;
239 ret = ip_route_output_key(&init_net, &rt, &fl);
243 src_in->sin_family = AF_INET;
244 src_in->sin_addr.s_addr = rt->rt_src;
246 if (rt->idev->dev->flags & IFF_LOOPBACK) {
247 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
249 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
253 /* If the device does ARP internally, return 'done' */
254 if (rt->idev->dev->flags & IFF_NOARP) {
255 rdma_copy_addr(addr, rt->idev->dev, NULL);
259 neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
260 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
261 neigh_event_send(rt->u.dst.neighbour, NULL);
268 ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
270 neigh_release(neigh);
278 static int addr6_resolve(struct sockaddr_in6 *src_in,
279 struct sockaddr_in6 *dst_in,
280 struct rdma_dev_addr *addr)
283 struct neighbour *neigh;
284 struct dst_entry *dst;
287 memset(&fl, 0, sizeof fl);
288 ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
289 ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
290 fl.oif = addr->bound_dev_if;
292 dst = ip6_route_output(&init_net, NULL, &fl);
293 if ((ret = dst->error))
296 if (ipv6_addr_any(&fl.fl6_src)) {
297 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
298 &fl.fl6_dst, 0, &fl.fl6_src);
302 src_in->sin6_family = AF_INET6;
303 ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
306 if (dst->dev->flags & IFF_LOOPBACK) {
307 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
309 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
313 /* If the device does ARP internally, return 'done' */
314 if (dst->dev->flags & IFF_NOARP) {
315 ret = rdma_copy_addr(addr, dst->dev, NULL);
319 neigh = dst->neighbour;
320 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
321 neigh_event_send(dst->neighbour, NULL);
326 ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
332 static int addr6_resolve(struct sockaddr_in6 *src_in,
333 struct sockaddr_in6 *dst_in,
334 struct rdma_dev_addr *addr)
336 return -EADDRNOTAVAIL;
341 #include <netinet/if_ether.h>
343 static int addr_resolve(struct sockaddr *src_in,
344 struct sockaddr *dst_in,
345 struct rdma_dev_addr *addr)
347 struct sockaddr_in *sin;
348 struct sockaddr_in6 *sin6;
351 #if defined(INET) || defined(INET6)
356 u_char edst[MAX_ADDR_LEN];
362 * Determine whether the address is unicast, multicast, or broadcast
363 * and whether the source interface is valid.
371 switch (dst_in->sa_family) {
374 sin = (struct sockaddr_in *)dst_in;
375 if (sin->sin_addr.s_addr == INADDR_BROADCAST)
377 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
379 sin = (struct sockaddr_in *)src_in;
380 if (sin->sin_addr.s_addr != INADDR_ANY) {
382 * Address comparison fails if the port is set
383 * cache it here to be restored later.
385 port = sin->sin_port;
387 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
394 sin6 = (struct sockaddr_in6 *)dst_in;
395 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
397 sin6 = (struct sockaddr_in6 *)src_in;
398 if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
399 port = sin6->sin6_port;
409 * If we have a source address to use look it up first and verify
410 * that it is a local interface.
413 ifa = ifa_ifwithaddr(src_in);
415 sin->sin_port = port;
417 sin6->sin6_port = port;
426 * Make sure the route exists and has a valid link.
428 rte = rtalloc1(dst_in, 1, 0);
429 if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
432 return -EHOSTUNREACH;
435 * If it's not multicast or broadcast and the route doesn't match the
436 * requested interface return unreachable. Otherwise fetch the
437 * correct interface pointer and unlock the route.
439 if (multi || bcast) {
443 } else if (ifp && ifp != rte->rt_ifp) {
453 return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
455 struct sockaddr *llsa;
457 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
460 error = rdma_copy_addr(addr, ifp,
461 LLADDR((struct sockaddr_dl *)llsa));
462 free(llsa, M_IFMADDR);
466 * Resolve the link local address.
468 switch (dst_in->sa_family) {
471 error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
476 error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
480 /* XXX: Shouldn't happen. */
485 return rdma_copy_addr(addr, ifp, edst);
486 if (error == EWOULDBLOCK)
493 static void process_req(struct work_struct *work)
495 struct addr_req *req, *temp_req;
496 struct sockaddr *src_in, *dst_in;
497 struct list_head done_list;
499 INIT_LIST_HEAD(&done_list);
502 list_for_each_entry_safe(req, temp_req, &req_list, list) {
503 if (req->status == -ENODATA) {
504 src_in = (struct sockaddr *) &req->src_addr;
505 dst_in = (struct sockaddr *) &req->dst_addr;
506 req->status = addr_resolve(src_in, dst_in, req->addr);
507 if (req->status && time_after_eq(jiffies, req->timeout))
508 req->status = -ETIMEDOUT;
509 else if (req->status == -ENODATA)
512 list_move_tail(&req->list, &done_list);
515 if (!list_empty(&req_list)) {
516 req = list_entry(req_list.next, struct addr_req, list);
517 set_timeout(req->timeout);
521 list_for_each_entry_safe(req, temp_req, &done_list, list) {
522 list_del(&req->list);
523 req->callback(req->status, (struct sockaddr *) &req->src_addr,
524 req->addr, req->context);
525 put_client(req->client);
530 int rdma_resolve_ip(struct rdma_addr_client *client,
531 struct sockaddr *src_addr, struct sockaddr *dst_addr,
532 struct rdma_dev_addr *addr, int timeout_ms,
533 void (*callback)(int status, struct sockaddr *src_addr,
534 struct rdma_dev_addr *addr, void *context),
537 struct sockaddr *src_in, *dst_in;
538 struct addr_req *req;
541 req = kzalloc(sizeof *req, GFP_KERNEL);
545 src_in = (struct sockaddr *) &req->src_addr;
546 dst_in = (struct sockaddr *) &req->dst_addr;
549 if (src_addr->sa_family != dst_addr->sa_family) {
554 memcpy(src_in, src_addr, ip_addr_size(src_addr));
556 src_in->sa_family = dst_addr->sa_family;
559 memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
561 req->callback = callback;
562 req->context = context;
563 req->client = client;
564 atomic_inc(&client->refcount);
566 req->status = addr_resolve(src_in, dst_in, addr);
567 switch (req->status) {
569 req->timeout = jiffies;
573 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
578 atomic_dec(&client->refcount);
586 EXPORT_SYMBOL(rdma_resolve_ip);
588 void rdma_addr_cancel(struct rdma_dev_addr *addr)
590 struct addr_req *req, *temp_req;
593 list_for_each_entry_safe(req, temp_req, &req_list, list) {
594 if (req->addr == addr) {
595 req->status = -ECANCELED;
596 req->timeout = jiffies;
597 list_move(&req->list, &req_list);
598 set_timeout(req->timeout);
604 EXPORT_SYMBOL(rdma_addr_cancel);
606 static int netevent_callback(struct notifier_block *self, unsigned long event,
609 if (event == NETEVENT_NEIGH_UPDATE) {
611 struct neighbour *neigh = ctx;
613 if (neigh->nud_state & NUD_VALID) {
614 set_timeout(jiffies);
617 set_timeout(jiffies);
623 static struct notifier_block nb = {
624 .notifier_call = netevent_callback
627 static int addr_init(void)
629 INIT_DELAYED_WORK(&work, process_req);
630 addr_wq = create_singlethread_workqueue("ib_addr");
634 register_netevent_notifier(&nb);
638 static void addr_cleanup(void)
640 unregister_netevent_notifier(&nb);
641 destroy_workqueue(addr_wq);
644 module_init(addr_init);
645 module_exit(addr_cleanup);