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/slab.h>
39 #include <linux/workqueue.h>
40 #include <linux/module.h>
41 #include <linux/notifier.h>
42 #include <net/route.h>
43 #include <net/netevent.h>
44 #include <rdma/ib_addr.h>
45 #include <netinet/if_ether.h>
48 MODULE_AUTHOR("Sean Hefty");
49 MODULE_DESCRIPTION("IB Address Translation");
50 MODULE_LICENSE("Dual BSD/GPL");
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;
59 void (*callback)(int status, struct sockaddr *src_addr,
60 struct rdma_dev_addr *addr, void *context);
61 unsigned long timeout;
65 static void process_req(struct work_struct *work);
67 static DEFINE_MUTEX(lock);
68 static LIST_HEAD(req_list);
69 static struct delayed_work work;
70 static struct workqueue_struct *addr_wq;
72 void rdma_addr_register_client(struct rdma_addr_client *client)
74 atomic_set(&client->refcount, 1);
75 init_completion(&client->comp);
77 EXPORT_SYMBOL(rdma_addr_register_client);
79 static inline void put_client(struct rdma_addr_client *client)
81 if (atomic_dec_and_test(&client->refcount))
82 complete(&client->comp);
85 void rdma_addr_unregister_client(struct rdma_addr_client *client)
88 wait_for_completion(&client->comp);
90 EXPORT_SYMBOL(rdma_addr_unregister_client);
93 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
94 const unsigned char *dst_dev_addr)
96 dev_addr->dev_type = dev->type;
97 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
98 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
100 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
101 dev_addr->bound_dev_if = dev->ifindex;
105 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct ifnet *dev,
106 const unsigned char *dst_dev_addr)
108 if (dev->if_type == IFT_INFINIBAND)
109 dev_addr->dev_type = ARPHRD_INFINIBAND;
110 else if (dev->if_type == IFT_ETHER)
111 dev_addr->dev_type = ARPHRD_ETHER;
113 dev_addr->dev_type = 0;
114 memcpy(dev_addr->src_dev_addr, IF_LLADDR(dev), dev->if_addrlen);
115 memcpy(dev_addr->broadcast, __DECONST(char *, dev->if_broadcastaddr),
118 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, dev->if_addrlen);
119 dev_addr->bound_dev_if = dev->if_index;
123 EXPORT_SYMBOL(rdma_copy_addr);
125 int rdma_translate_ip(struct sockaddr *addr, struct rdma_dev_addr *dev_addr)
127 struct net_device *dev;
128 int ret = -EADDRNOTAVAIL;
130 if (dev_addr->bound_dev_if) {
131 dev = dev_get_by_index(&init_net, dev_addr->bound_dev_if);
134 ret = rdma_copy_addr(dev_addr, dev, NULL);
139 switch (addr->sa_family) {
142 dev = ip_dev_find(NULL,
143 ((struct sockaddr_in *) addr)->sin_addr.s_addr);
148 ret = rdma_copy_addr(dev_addr, dev, NULL);
156 read_lock(&dev_base_lock);
157 for_each_netdev(&init_net, dev) {
158 if (ipv6_chk_addr(&init_net,
159 &((struct sockaddr_in6 *) addr)->sin6_addr,
161 ret = rdma_copy_addr(dev_addr, dev, NULL);
165 read_unlock(&dev_base_lock);
168 struct sockaddr_in6 *sin6;
172 sin6 = (struct sockaddr_in6 *)addr;
173 port = sin6->sin6_port;
175 ifa = ifa_ifwithaddr(addr);
176 sin6->sin6_port = port;
181 ret = rdma_copy_addr(dev_addr, ifa->ifa_ifp, NULL);
191 EXPORT_SYMBOL(rdma_translate_ip);
193 static void set_timeout(unsigned long time)
197 delay = time - jiffies;
198 if ((long)delay <= 0)
201 mod_delayed_work(addr_wq, &work, delay);
204 static void queue_req(struct addr_req *req)
206 struct addr_req *temp_req;
209 list_for_each_entry_reverse(temp_req, &req_list, list) {
210 if (time_after_eq(req->timeout, temp_req->timeout))
214 list_add(&req->list, &temp_req->list);
216 if (req_list.next == &req->list)
217 set_timeout(req->timeout);
222 static int addr4_resolve(struct sockaddr_in *src_in,
223 struct sockaddr_in *dst_in,
224 struct rdma_dev_addr *addr)
226 __be32 src_ip = src_in->sin_addr.s_addr;
227 __be32 dst_ip = dst_in->sin_addr.s_addr;
230 struct neighbour *neigh;
233 memset(&fl, 0, sizeof fl);
234 fl.nl_u.ip4_u.daddr = dst_ip;
235 fl.nl_u.ip4_u.saddr = src_ip;
236 fl.oif = addr->bound_dev_if;
238 ret = ip_route_output_key(&init_net, &rt, &fl);
242 src_in->sin_family = AF_INET;
243 src_in->sin_addr.s_addr = rt->rt_src;
245 if (rt->idev->dev->flags & IFF_LOOPBACK) {
246 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
248 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
252 /* If the device does ARP internally, return 'done' */
253 if (rt->idev->dev->flags & IFF_NOARP) {
254 rdma_copy_addr(addr, rt->idev->dev, NULL);
258 neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, rt->idev->dev);
259 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
260 neigh_event_send(rt->u.dst.neighbour, NULL);
267 ret = rdma_copy_addr(addr, neigh->dev, neigh->ha);
269 neigh_release(neigh);
277 static int addr6_resolve(struct sockaddr_in6 *src_in,
278 struct sockaddr_in6 *dst_in,
279 struct rdma_dev_addr *addr)
282 struct neighbour *neigh;
283 struct dst_entry *dst;
286 memset(&fl, 0, sizeof fl);
287 ipv6_addr_copy(&fl.fl6_dst, &dst_in->sin6_addr);
288 ipv6_addr_copy(&fl.fl6_src, &src_in->sin6_addr);
289 fl.oif = addr->bound_dev_if;
291 dst = ip6_route_output(&init_net, NULL, &fl);
292 if ((ret = dst->error))
295 if (ipv6_addr_any(&fl.fl6_src)) {
296 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
297 &fl.fl6_dst, 0, &fl.fl6_src);
301 src_in->sin6_family = AF_INET6;
302 ipv6_addr_copy(&src_in->sin6_addr, &fl.fl6_src);
305 if (dst->dev->flags & IFF_LOOPBACK) {
306 ret = rdma_translate_ip((struct sockaddr *) dst_in, addr);
308 memcpy(addr->dst_dev_addr, addr->src_dev_addr, MAX_ADDR_LEN);
312 /* If the device does ARP internally, return 'done' */
313 if (dst->dev->flags & IFF_NOARP) {
314 ret = rdma_copy_addr(addr, dst->dev, NULL);
318 neigh = dst->neighbour;
319 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
320 neigh_event_send(dst->neighbour, NULL);
325 ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
331 static int addr6_resolve(struct sockaddr_in6 *src_in,
332 struct sockaddr_in6 *dst_in,
333 struct rdma_dev_addr *addr)
335 return -EADDRNOTAVAIL;
340 #include <netinet/if_ether.h>
342 static int addr_resolve(struct sockaddr *src_in,
343 struct sockaddr *dst_in,
344 struct rdma_dev_addr *addr)
346 struct sockaddr_in *sin;
347 struct sockaddr_in6 *sin6;
350 #if defined(INET) || defined(INET6)
355 u_char edst[MAX_ADDR_LEN];
361 * Determine whether the address is unicast, multicast, or broadcast
362 * and whether the source interface is valid.
370 switch (dst_in->sa_family) {
373 sin = (struct sockaddr_in *)dst_in;
374 if (sin->sin_addr.s_addr == INADDR_BROADCAST)
376 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
378 sin = (struct sockaddr_in *)src_in;
379 if (sin->sin_addr.s_addr != INADDR_ANY) {
381 * Address comparison fails if the port is set
382 * cache it here to be restored later.
384 port = sin->sin_port;
386 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
393 sin6 = (struct sockaddr_in6 *)dst_in;
394 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
396 sin6 = (struct sockaddr_in6 *)src_in;
397 if (!IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
398 port = sin6->sin6_port;
408 * If we have a source address to use look it up first and verify
409 * that it is a local interface.
412 ifa = ifa_ifwithaddr(src_in);
414 sin->sin_port = port;
416 sin6->sin6_port = port;
425 * Make sure the route exists and has a valid link.
427 rte = rtalloc1(dst_in, 1, 0);
428 if (rte == NULL || rte->rt_ifp == NULL || !RT_LINK_IS_UP(rte->rt_ifp)) {
431 return -EHOSTUNREACH;
434 * If it's not multicast or broadcast and the route doesn't match the
435 * requested interface return unreachable. Otherwise fetch the
436 * correct interface pointer and unlock the route.
438 if (multi || bcast) {
442 } else if (ifp && ifp != rte->rt_ifp) {
452 return rdma_copy_addr(addr, ifp, ifp->if_broadcastaddr);
454 struct sockaddr *llsa;
456 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
459 error = rdma_copy_addr(addr, ifp,
460 LLADDR((struct sockaddr_dl *)llsa));
461 free(llsa, M_IFMADDR);
465 * Resolve the link local address.
467 switch (dst_in->sa_family) {
470 error = arpresolve(ifp, rte, NULL, dst_in, edst, &lle);
475 error = nd6_storelladdr(ifp, NULL, dst_in, (u_char *)edst, &lle);
479 /* XXX: Shouldn't happen. */
484 return rdma_copy_addr(addr, ifp, edst);
485 if (error == EWOULDBLOCK)
492 static void process_req(struct work_struct *work)
494 struct addr_req *req, *temp_req;
495 struct sockaddr *src_in, *dst_in;
496 struct list_head done_list;
498 INIT_LIST_HEAD(&done_list);
501 list_for_each_entry_safe(req, temp_req, &req_list, list) {
502 if (req->status == -ENODATA) {
503 src_in = (struct sockaddr *) &req->src_addr;
504 dst_in = (struct sockaddr *) &req->dst_addr;
505 req->status = addr_resolve(src_in, dst_in, req->addr);
506 if (req->status && time_after_eq(jiffies, req->timeout))
507 req->status = -ETIMEDOUT;
508 else if (req->status == -ENODATA)
511 list_move_tail(&req->list, &done_list);
514 if (!list_empty(&req_list)) {
515 req = list_entry(req_list.next, struct addr_req, list);
516 set_timeout(req->timeout);
520 list_for_each_entry_safe(req, temp_req, &done_list, list) {
521 list_del(&req->list);
522 req->callback(req->status, (struct sockaddr *) &req->src_addr,
523 req->addr, req->context);
524 put_client(req->client);
529 int rdma_resolve_ip(struct rdma_addr_client *client,
530 struct sockaddr *src_addr, struct sockaddr *dst_addr,
531 struct rdma_dev_addr *addr, int timeout_ms,
532 void (*callback)(int status, struct sockaddr *src_addr,
533 struct rdma_dev_addr *addr, void *context),
536 struct sockaddr *src_in, *dst_in;
537 struct addr_req *req;
540 req = kzalloc(sizeof *req, GFP_KERNEL);
544 src_in = (struct sockaddr *) &req->src_addr;
545 dst_in = (struct sockaddr *) &req->dst_addr;
548 if (src_addr->sa_family != dst_addr->sa_family) {
553 memcpy(src_in, src_addr, ip_addr_size(src_addr));
555 src_in->sa_family = dst_addr->sa_family;
558 memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
560 req->callback = callback;
561 req->context = context;
562 req->client = client;
563 atomic_inc(&client->refcount);
565 req->status = addr_resolve(src_in, dst_in, addr);
566 switch (req->status) {
568 req->timeout = jiffies;
572 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
577 atomic_dec(&client->refcount);
585 EXPORT_SYMBOL(rdma_resolve_ip);
587 void rdma_addr_cancel(struct rdma_dev_addr *addr)
589 struct addr_req *req, *temp_req;
592 list_for_each_entry_safe(req, temp_req, &req_list, list) {
593 if (req->addr == addr) {
594 req->status = -ECANCELED;
595 req->timeout = jiffies;
596 list_move(&req->list, &req_list);
597 set_timeout(req->timeout);
603 EXPORT_SYMBOL(rdma_addr_cancel);
605 static int netevent_callback(struct notifier_block *self, unsigned long event,
608 if (event == NETEVENT_NEIGH_UPDATE) {
610 struct neighbour *neigh = ctx;
612 if (neigh->nud_state & NUD_VALID) {
613 set_timeout(jiffies);
616 set_timeout(jiffies);
622 static struct notifier_block nb = {
623 .notifier_call = netevent_callback
626 static int __init addr_init(void)
628 INIT_DELAYED_WORK(&work, process_req);
629 addr_wq = create_singlethread_workqueue("ib_addr");
633 register_netevent_notifier(&nb);
637 static void __exit addr_cleanup(void)
639 unregister_netevent_notifier(&nb);
640 destroy_workqueue(addr_wq);
643 module_init(addr_init);
644 module_exit(addr_cleanup);