Loading...
1/*
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.
6 *
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:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
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.
25 *
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
33 * SOFTWARE.
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 rtable *rt;
187 struct neighbour *neigh;
188 struct flowi4 fl4;
189 int ret;
190
191 memset(&fl4, 0, sizeof(fl4));
192 fl4.daddr = dst_ip;
193 fl4.saddr = src_ip;
194 fl4.flowi4_oif = addr->bound_dev_if;
195 rt = ip_route_output_key(&init_net, &fl4);
196 if (IS_ERR(rt)) {
197 ret = PTR_ERR(rt);
198 goto out;
199 }
200 src_in->sin_family = AF_INET;
201 src_in->sin_addr.s_addr = fl4.saddr;
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 /* If the device does ARP internally, return 'done' */
211 if (rt->dst.dev->flags & IFF_NOARP) {
212 ret = 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(dst_get_neighbour(&rt->dst), 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 flowi6 fl6;
240 struct neighbour *neigh;
241 struct dst_entry *dst;
242 int ret;
243
244 memset(&fl6, 0, sizeof fl6);
245 ipv6_addr_copy(&fl6.daddr, &dst_in->sin6_addr);
246 ipv6_addr_copy(&fl6.saddr, &src_in->sin6_addr);
247 fl6.flowi6_oif = addr->bound_dev_if;
248
249 dst = ip6_route_output(&init_net, NULL, &fl6);
250 if ((ret = dst->error))
251 goto put;
252
253 if (ipv6_addr_any(&fl6.saddr)) {
254 ret = ipv6_dev_get_saddr(&init_net, ip6_dst_idev(dst)->dev,
255 &fl6.daddr, 0, &fl6.saddr);
256 if (ret)
257 goto put;
258
259 src_in->sin6_family = AF_INET6;
260 ipv6_addr_copy(&src_in->sin6_addr, &fl6.saddr);
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 /* If the device does ARP internally, return 'done' */
271 if (dst->dev->flags & IFF_NOARP) {
272 ret = rdma_copy_addr(addr, dst->dev, NULL);
273 goto put;
274 }
275
276 neigh = dst_get_neighbour(dst);
277 if (!neigh || !(neigh->nud_state & NUD_VALID)) {
278 if (neigh)
279 neigh_event_send(neigh, NULL);
280 ret = -ENODATA;
281 goto put;
282 }
283
284 ret = rdma_copy_addr(addr, dst->dev, neigh->ha);
285put:
286 dst_release(dst);
287 return ret;
288}
289#else
290static int addr6_resolve(struct sockaddr_in6 *src_in,
291 struct sockaddr_in6 *dst_in,
292 struct rdma_dev_addr *addr)
293{
294 return -EADDRNOTAVAIL;
295}
296#endif
297
298static int addr_resolve(struct sockaddr *src_in,
299 struct sockaddr *dst_in,
300 struct rdma_dev_addr *addr)
301{
302 if (src_in->sa_family == AF_INET) {
303 return addr4_resolve((struct sockaddr_in *) src_in,
304 (struct sockaddr_in *) dst_in, addr);
305 } else
306 return addr6_resolve((struct sockaddr_in6 *) src_in,
307 (struct sockaddr_in6 *) dst_in, addr);
308}
309
310static void process_req(struct work_struct *work)
311{
312 struct addr_req *req, *temp_req;
313 struct sockaddr *src_in, *dst_in;
314 struct list_head done_list;
315
316 INIT_LIST_HEAD(&done_list);
317
318 mutex_lock(&lock);
319 list_for_each_entry_safe(req, temp_req, &req_list, list) {
320 if (req->status == -ENODATA) {
321 src_in = (struct sockaddr *) &req->src_addr;
322 dst_in = (struct sockaddr *) &req->dst_addr;
323 req->status = addr_resolve(src_in, dst_in, req->addr);
324 if (req->status && time_after_eq(jiffies, req->timeout))
325 req->status = -ETIMEDOUT;
326 else if (req->status == -ENODATA)
327 continue;
328 }
329 list_move_tail(&req->list, &done_list);
330 }
331
332 if (!list_empty(&req_list)) {
333 req = list_entry(req_list.next, struct addr_req, list);
334 set_timeout(req->timeout);
335 }
336 mutex_unlock(&lock);
337
338 list_for_each_entry_safe(req, temp_req, &done_list, list) {
339 list_del(&req->list);
340 req->callback(req->status, (struct sockaddr *) &req->src_addr,
341 req->addr, req->context);
342 put_client(req->client);
343 kfree(req);
344 }
345}
346
347int rdma_resolve_ip(struct rdma_addr_client *client,
348 struct sockaddr *src_addr, struct sockaddr *dst_addr,
349 struct rdma_dev_addr *addr, int timeout_ms,
350 void (*callback)(int status, struct sockaddr *src_addr,
351 struct rdma_dev_addr *addr, void *context),
352 void *context)
353{
354 struct sockaddr *src_in, *dst_in;
355 struct addr_req *req;
356 int ret = 0;
357
358 req = kzalloc(sizeof *req, GFP_KERNEL);
359 if (!req)
360 return -ENOMEM;
361
362 src_in = (struct sockaddr *) &req->src_addr;
363 dst_in = (struct sockaddr *) &req->dst_addr;
364
365 if (src_addr) {
366 if (src_addr->sa_family != dst_addr->sa_family) {
367 ret = -EINVAL;
368 goto err;
369 }
370
371 memcpy(src_in, src_addr, ip_addr_size(src_addr));
372 } else {
373 src_in->sa_family = dst_addr->sa_family;
374 }
375
376 memcpy(dst_in, dst_addr, ip_addr_size(dst_addr));
377 req->addr = addr;
378 req->callback = callback;
379 req->context = context;
380 req->client = client;
381 atomic_inc(&client->refcount);
382
383 req->status = addr_resolve(src_in, dst_in, addr);
384 switch (req->status) {
385 case 0:
386 req->timeout = jiffies;
387 queue_req(req);
388 break;
389 case -ENODATA:
390 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
391 queue_req(req);
392 break;
393 default:
394 ret = req->status;
395 atomic_dec(&client->refcount);
396 goto err;
397 }
398 return ret;
399err:
400 kfree(req);
401 return ret;
402}
403EXPORT_SYMBOL(rdma_resolve_ip);
404
405void rdma_addr_cancel(struct rdma_dev_addr *addr)
406{
407 struct addr_req *req, *temp_req;
408
409 mutex_lock(&lock);
410 list_for_each_entry_safe(req, temp_req, &req_list, list) {
411 if (req->addr == addr) {
412 req->status = -ECANCELED;
413 req->timeout = jiffies;
414 list_move(&req->list, &req_list);
415 set_timeout(req->timeout);
416 break;
417 }
418 }
419 mutex_unlock(&lock);
420}
421EXPORT_SYMBOL(rdma_addr_cancel);
422
423static int netevent_callback(struct notifier_block *self, unsigned long event,
424 void *ctx)
425{
426 if (event == NETEVENT_NEIGH_UPDATE) {
427 struct neighbour *neigh = ctx;
428
429 if (neigh->nud_state & NUD_VALID) {
430 set_timeout(jiffies);
431 }
432 }
433 return 0;
434}
435
436static struct notifier_block nb = {
437 .notifier_call = netevent_callback
438};
439
440static int __init addr_init(void)
441{
442 addr_wq = create_singlethread_workqueue("ib_addr");
443 if (!addr_wq)
444 return -ENOMEM;
445
446 register_netevent_notifier(&nb);
447 return 0;
448}
449
450static void __exit addr_cleanup(void)
451{
452 unregister_netevent_notifier(&nb);
453 destroy_workqueue(addr_wq);
454}
455
456module_init(addr_init);
457module_exit(addr_cleanup);
1/*
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.
6 *
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:
12 *
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
15 * conditions are met:
16 *
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer.
20 *
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.
25 *
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
33 * SOFTWARE.
34 */
35
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 <net/arp.h>
42#include <net/neighbour.h>
43#include <net/route.h>
44#include <net/netevent.h>
45#include <net/addrconf.h>
46#include <net/ip6_route.h>
47#include <rdma/ib_addr.h>
48#include <rdma/ib.h>
49#include <rdma/rdma_netlink.h>
50#include <net/netlink.h>
51
52#include "core_priv.h"
53
54struct addr_req {
55 struct list_head list;
56 struct sockaddr_storage src_addr;
57 struct sockaddr_storage dst_addr;
58 struct rdma_dev_addr *addr;
59 struct rdma_addr_client *client;
60 void *context;
61 void (*callback)(int status, struct sockaddr *src_addr,
62 struct rdma_dev_addr *addr, void *context);
63 unsigned long timeout;
64 int status;
65 u32 seq;
66};
67
68static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
69
70static void process_req(struct work_struct *work);
71
72static DEFINE_MUTEX(lock);
73static LIST_HEAD(req_list);
74static DECLARE_DELAYED_WORK(work, process_req);
75static struct workqueue_struct *addr_wq;
76
77static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
78 [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
79 .len = sizeof(struct rdma_nla_ls_gid)},
80};
81
82static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
83{
84 struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
85 int ret;
86
87 if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
88 return false;
89
90 ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
91 nlmsg_len(nlh), ib_nl_addr_policy);
92 if (ret)
93 return false;
94
95 return true;
96}
97
98static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
99{
100 const struct nlattr *head, *curr;
101 union ib_gid gid;
102 struct addr_req *req;
103 int len, rem;
104 int found = 0;
105
106 head = (const struct nlattr *)nlmsg_data(nlh);
107 len = nlmsg_len(nlh);
108
109 nla_for_each_attr(curr, head, len, rem) {
110 if (curr->nla_type == LS_NLA_TYPE_DGID)
111 memcpy(&gid, nla_data(curr), nla_len(curr));
112 }
113
114 mutex_lock(&lock);
115 list_for_each_entry(req, &req_list, list) {
116 if (nlh->nlmsg_seq != req->seq)
117 continue;
118 /* We set the DGID part, the rest was set earlier */
119 rdma_addr_set_dgid(req->addr, &gid);
120 req->status = 0;
121 found = 1;
122 break;
123 }
124 mutex_unlock(&lock);
125
126 if (!found)
127 pr_info("Couldn't find request waiting for DGID: %pI6\n",
128 &gid);
129}
130
131int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
132 struct netlink_callback *cb)
133{
134 const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
135
136 if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
137 !(NETLINK_CB(skb).sk) ||
138 !netlink_capable(skb, CAP_NET_ADMIN))
139 return -EPERM;
140
141 if (ib_nl_is_good_ip_resp(nlh))
142 ib_nl_process_good_ip_rsep(nlh);
143
144 return skb->len;
145}
146
147static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
148 const void *daddr,
149 u32 seq, u16 family)
150{
151 struct sk_buff *skb = NULL;
152 struct nlmsghdr *nlh;
153 struct rdma_ls_ip_resolve_header *header;
154 void *data;
155 size_t size;
156 int attrtype;
157 int len;
158
159 if (family == AF_INET) {
160 size = sizeof(struct in_addr);
161 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
162 } else {
163 size = sizeof(struct in6_addr);
164 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
165 }
166
167 len = nla_total_size(sizeof(size));
168 len += NLMSG_ALIGN(sizeof(*header));
169
170 skb = nlmsg_new(len, GFP_KERNEL);
171 if (!skb)
172 return -ENOMEM;
173
174 data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
175 RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
176 if (!data) {
177 nlmsg_free(skb);
178 return -ENODATA;
179 }
180
181 /* Construct the family header first */
182 header = (struct rdma_ls_ip_resolve_header *)
183 skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
184 header->ifindex = dev_addr->bound_dev_if;
185 nla_put(skb, attrtype, size, daddr);
186
187 /* Repair the nlmsg header length */
188 nlmsg_end(skb, nlh);
189 ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, GFP_KERNEL);
190
191 /* Make the request retry, so when we get the response from userspace
192 * we will have something.
193 */
194 return -ENODATA;
195}
196
197int rdma_addr_size(struct sockaddr *addr)
198{
199 switch (addr->sa_family) {
200 case AF_INET:
201 return sizeof(struct sockaddr_in);
202 case AF_INET6:
203 return sizeof(struct sockaddr_in6);
204 case AF_IB:
205 return sizeof(struct sockaddr_ib);
206 default:
207 return 0;
208 }
209}
210EXPORT_SYMBOL(rdma_addr_size);
211
212static struct rdma_addr_client self;
213
214void rdma_addr_register_client(struct rdma_addr_client *client)
215{
216 atomic_set(&client->refcount, 1);
217 init_completion(&client->comp);
218}
219EXPORT_SYMBOL(rdma_addr_register_client);
220
221static inline void put_client(struct rdma_addr_client *client)
222{
223 if (atomic_dec_and_test(&client->refcount))
224 complete(&client->comp);
225}
226
227void rdma_addr_unregister_client(struct rdma_addr_client *client)
228{
229 put_client(client);
230 wait_for_completion(&client->comp);
231}
232EXPORT_SYMBOL(rdma_addr_unregister_client);
233
234int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
235 const unsigned char *dst_dev_addr)
236{
237 dev_addr->dev_type = dev->type;
238 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
239 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
240 if (dst_dev_addr)
241 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
242 dev_addr->bound_dev_if = dev->ifindex;
243 return 0;
244}
245EXPORT_SYMBOL(rdma_copy_addr);
246
247int rdma_translate_ip(const struct sockaddr *addr,
248 struct rdma_dev_addr *dev_addr,
249 u16 *vlan_id)
250{
251 struct net_device *dev;
252 int ret = -EADDRNOTAVAIL;
253
254 if (dev_addr->bound_dev_if) {
255 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
256 if (!dev)
257 return -ENODEV;
258 ret = rdma_copy_addr(dev_addr, dev, NULL);
259 dev_put(dev);
260 return ret;
261 }
262
263 switch (addr->sa_family) {
264 case AF_INET:
265 dev = ip_dev_find(dev_addr->net,
266 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
267
268 if (!dev)
269 return ret;
270
271 ret = rdma_copy_addr(dev_addr, dev, NULL);
272 if (vlan_id)
273 *vlan_id = rdma_vlan_dev_vlan_id(dev);
274 dev_put(dev);
275 break;
276#if IS_ENABLED(CONFIG_IPV6)
277 case AF_INET6:
278 rcu_read_lock();
279 for_each_netdev_rcu(dev_addr->net, dev) {
280 if (ipv6_chk_addr(dev_addr->net,
281 &((const struct sockaddr_in6 *)addr)->sin6_addr,
282 dev, 1)) {
283 ret = rdma_copy_addr(dev_addr, dev, NULL);
284 if (vlan_id)
285 *vlan_id = rdma_vlan_dev_vlan_id(dev);
286 break;
287 }
288 }
289 rcu_read_unlock();
290 break;
291#endif
292 }
293 return ret;
294}
295EXPORT_SYMBOL(rdma_translate_ip);
296
297static void set_timeout(unsigned long time)
298{
299 unsigned long delay;
300
301 delay = time - jiffies;
302 if ((long)delay < 0)
303 delay = 0;
304
305 mod_delayed_work(addr_wq, &work, delay);
306}
307
308static void queue_req(struct addr_req *req)
309{
310 struct addr_req *temp_req;
311
312 mutex_lock(&lock);
313 list_for_each_entry_reverse(temp_req, &req_list, list) {
314 if (time_after_eq(req->timeout, temp_req->timeout))
315 break;
316 }
317
318 list_add(&req->list, &temp_req->list);
319
320 if (req_list.next == &req->list)
321 set_timeout(req->timeout);
322 mutex_unlock(&lock);
323}
324
325static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
326 const void *daddr, u32 seq, u16 family)
327{
328 if (ibnl_chk_listeners(RDMA_NL_GROUP_LS))
329 return -EADDRNOTAVAIL;
330
331 /* We fill in what we can, the response will fill the rest */
332 rdma_copy_addr(dev_addr, dst->dev, NULL);
333 return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
334}
335
336static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
337 const void *daddr)
338{
339 struct neighbour *n;
340 int ret;
341
342 n = dst_neigh_lookup(dst, daddr);
343
344 rcu_read_lock();
345 if (!n || !(n->nud_state & NUD_VALID)) {
346 if (n)
347 neigh_event_send(n, NULL);
348 ret = -ENODATA;
349 } else {
350 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
351 }
352 rcu_read_unlock();
353
354 if (n)
355 neigh_release(n);
356
357 return ret;
358}
359
360static bool has_gateway(struct dst_entry *dst, sa_family_t family)
361{
362 struct rtable *rt;
363 struct rt6_info *rt6;
364
365 if (family == AF_INET) {
366 rt = container_of(dst, struct rtable, dst);
367 return rt->rt_uses_gateway;
368 }
369
370 rt6 = container_of(dst, struct rt6_info, dst);
371 return rt6->rt6i_flags & RTF_GATEWAY;
372}
373
374static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
375 const struct sockaddr *dst_in, u32 seq)
376{
377 const struct sockaddr_in *dst_in4 =
378 (const struct sockaddr_in *)dst_in;
379 const struct sockaddr_in6 *dst_in6 =
380 (const struct sockaddr_in6 *)dst_in;
381 const void *daddr = (dst_in->sa_family == AF_INET) ?
382 (const void *)&dst_in4->sin_addr.s_addr :
383 (const void *)&dst_in6->sin6_addr;
384 sa_family_t family = dst_in->sa_family;
385
386 /* Gateway + ARPHRD_INFINIBAND -> IB router */
387 if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
388 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
389 else
390 return dst_fetch_ha(dst, dev_addr, daddr);
391}
392
393static int addr4_resolve(struct sockaddr_in *src_in,
394 const struct sockaddr_in *dst_in,
395 struct rdma_dev_addr *addr,
396 struct rtable **prt)
397{
398 __be32 src_ip = src_in->sin_addr.s_addr;
399 __be32 dst_ip = dst_in->sin_addr.s_addr;
400 struct rtable *rt;
401 struct flowi4 fl4;
402 int ret;
403
404 memset(&fl4, 0, sizeof(fl4));
405 fl4.daddr = dst_ip;
406 fl4.saddr = src_ip;
407 fl4.flowi4_oif = addr->bound_dev_if;
408 rt = ip_route_output_key(addr->net, &fl4);
409 if (IS_ERR(rt)) {
410 ret = PTR_ERR(rt);
411 goto out;
412 }
413 src_in->sin_family = AF_INET;
414 src_in->sin_addr.s_addr = fl4.saddr;
415
416 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
417 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
418 * type accordingly.
419 */
420 if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
421 addr->network = RDMA_NETWORK_IPV4;
422
423 addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
424
425 *prt = rt;
426 return 0;
427out:
428 return ret;
429}
430
431#if IS_ENABLED(CONFIG_IPV6)
432static int addr6_resolve(struct sockaddr_in6 *src_in,
433 const struct sockaddr_in6 *dst_in,
434 struct rdma_dev_addr *addr,
435 struct dst_entry **pdst)
436{
437 struct flowi6 fl6;
438 struct dst_entry *dst;
439 struct rt6_info *rt;
440 int ret;
441
442 memset(&fl6, 0, sizeof fl6);
443 fl6.daddr = dst_in->sin6_addr;
444 fl6.saddr = src_in->sin6_addr;
445 fl6.flowi6_oif = addr->bound_dev_if;
446
447 dst = ip6_route_output(addr->net, NULL, &fl6);
448 if ((ret = dst->error))
449 goto put;
450
451 rt = (struct rt6_info *)dst;
452 if (ipv6_addr_any(&fl6.saddr)) {
453 ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev,
454 &fl6.daddr, 0, &fl6.saddr);
455 if (ret)
456 goto put;
457
458 src_in->sin6_family = AF_INET6;
459 src_in->sin6_addr = fl6.saddr;
460 }
461
462 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
463 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
464 * type accordingly.
465 */
466 if (rt->rt6i_flags & RTF_GATEWAY &&
467 ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
468 addr->network = RDMA_NETWORK_IPV6;
469
470 addr->hoplimit = ip6_dst_hoplimit(dst);
471
472 *pdst = dst;
473 return 0;
474put:
475 dst_release(dst);
476 return ret;
477}
478#else
479static int addr6_resolve(struct sockaddr_in6 *src_in,
480 const struct sockaddr_in6 *dst_in,
481 struct rdma_dev_addr *addr,
482 struct dst_entry **pdst)
483{
484 return -EADDRNOTAVAIL;
485}
486#endif
487
488static int addr_resolve_neigh(struct dst_entry *dst,
489 const struct sockaddr *dst_in,
490 struct rdma_dev_addr *addr,
491 u32 seq)
492{
493 if (dst->dev->flags & IFF_LOOPBACK) {
494 int ret;
495
496 ret = rdma_translate_ip(dst_in, addr, NULL);
497 if (!ret)
498 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
499 MAX_ADDR_LEN);
500
501 return ret;
502 }
503
504 /* If the device doesn't do ARP internally */
505 if (!(dst->dev->flags & IFF_NOARP))
506 return fetch_ha(dst, addr, dst_in, seq);
507
508 return rdma_copy_addr(addr, dst->dev, NULL);
509}
510
511static int addr_resolve(struct sockaddr *src_in,
512 const struct sockaddr *dst_in,
513 struct rdma_dev_addr *addr,
514 bool resolve_neigh,
515 u32 seq)
516{
517 struct net_device *ndev;
518 struct dst_entry *dst;
519 int ret;
520
521 if (src_in->sa_family == AF_INET) {
522 struct rtable *rt = NULL;
523 const struct sockaddr_in *dst_in4 =
524 (const struct sockaddr_in *)dst_in;
525
526 ret = addr4_resolve((struct sockaddr_in *)src_in,
527 dst_in4, addr, &rt);
528 if (ret)
529 return ret;
530
531 if (resolve_neigh)
532 ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
533
534 ndev = rt->dst.dev;
535 dev_hold(ndev);
536
537 ip_rt_put(rt);
538 } else {
539 const struct sockaddr_in6 *dst_in6 =
540 (const struct sockaddr_in6 *)dst_in;
541
542 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
543 dst_in6, addr,
544 &dst);
545 if (ret)
546 return ret;
547
548 if (resolve_neigh)
549 ret = addr_resolve_neigh(dst, dst_in, addr, seq);
550
551 ndev = dst->dev;
552 dev_hold(ndev);
553
554 dst_release(dst);
555 }
556
557 addr->bound_dev_if = ndev->ifindex;
558 addr->net = dev_net(ndev);
559 dev_put(ndev);
560
561 return ret;
562}
563
564static void process_req(struct work_struct *work)
565{
566 struct addr_req *req, *temp_req;
567 struct sockaddr *src_in, *dst_in;
568 struct list_head done_list;
569
570 INIT_LIST_HEAD(&done_list);
571
572 mutex_lock(&lock);
573 list_for_each_entry_safe(req, temp_req, &req_list, list) {
574 if (req->status == -ENODATA) {
575 src_in = (struct sockaddr *) &req->src_addr;
576 dst_in = (struct sockaddr *) &req->dst_addr;
577 req->status = addr_resolve(src_in, dst_in, req->addr,
578 true, req->seq);
579 if (req->status && time_after_eq(jiffies, req->timeout))
580 req->status = -ETIMEDOUT;
581 else if (req->status == -ENODATA)
582 continue;
583 }
584 list_move_tail(&req->list, &done_list);
585 }
586
587 if (!list_empty(&req_list)) {
588 req = list_entry(req_list.next, struct addr_req, list);
589 set_timeout(req->timeout);
590 }
591 mutex_unlock(&lock);
592
593 list_for_each_entry_safe(req, temp_req, &done_list, list) {
594 list_del(&req->list);
595 req->callback(req->status, (struct sockaddr *) &req->src_addr,
596 req->addr, req->context);
597 put_client(req->client);
598 kfree(req);
599 }
600}
601
602int rdma_resolve_ip(struct rdma_addr_client *client,
603 struct sockaddr *src_addr, struct sockaddr *dst_addr,
604 struct rdma_dev_addr *addr, int timeout_ms,
605 void (*callback)(int status, struct sockaddr *src_addr,
606 struct rdma_dev_addr *addr, void *context),
607 void *context)
608{
609 struct sockaddr *src_in, *dst_in;
610 struct addr_req *req;
611 int ret = 0;
612
613 req = kzalloc(sizeof *req, GFP_KERNEL);
614 if (!req)
615 return -ENOMEM;
616
617 src_in = (struct sockaddr *) &req->src_addr;
618 dst_in = (struct sockaddr *) &req->dst_addr;
619
620 if (src_addr) {
621 if (src_addr->sa_family != dst_addr->sa_family) {
622 ret = -EINVAL;
623 goto err;
624 }
625
626 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
627 } else {
628 src_in->sa_family = dst_addr->sa_family;
629 }
630
631 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
632 req->addr = addr;
633 req->callback = callback;
634 req->context = context;
635 req->client = client;
636 atomic_inc(&client->refcount);
637 req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
638
639 req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
640 switch (req->status) {
641 case 0:
642 req->timeout = jiffies;
643 queue_req(req);
644 break;
645 case -ENODATA:
646 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
647 queue_req(req);
648 break;
649 default:
650 ret = req->status;
651 atomic_dec(&client->refcount);
652 goto err;
653 }
654 return ret;
655err:
656 kfree(req);
657 return ret;
658}
659EXPORT_SYMBOL(rdma_resolve_ip);
660
661int rdma_resolve_ip_route(struct sockaddr *src_addr,
662 const struct sockaddr *dst_addr,
663 struct rdma_dev_addr *addr)
664{
665 struct sockaddr_storage ssrc_addr = {};
666 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
667
668 if (src_addr) {
669 if (src_addr->sa_family != dst_addr->sa_family)
670 return -EINVAL;
671
672 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
673 } else {
674 src_in->sa_family = dst_addr->sa_family;
675 }
676
677 return addr_resolve(src_in, dst_addr, addr, false, 0);
678}
679EXPORT_SYMBOL(rdma_resolve_ip_route);
680
681void rdma_addr_cancel(struct rdma_dev_addr *addr)
682{
683 struct addr_req *req, *temp_req;
684
685 mutex_lock(&lock);
686 list_for_each_entry_safe(req, temp_req, &req_list, list) {
687 if (req->addr == addr) {
688 req->status = -ECANCELED;
689 req->timeout = jiffies;
690 list_move(&req->list, &req_list);
691 set_timeout(req->timeout);
692 break;
693 }
694 }
695 mutex_unlock(&lock);
696}
697EXPORT_SYMBOL(rdma_addr_cancel);
698
699struct resolve_cb_context {
700 struct rdma_dev_addr *addr;
701 struct completion comp;
702 int status;
703};
704
705static void resolve_cb(int status, struct sockaddr *src_addr,
706 struct rdma_dev_addr *addr, void *context)
707{
708 if (!status)
709 memcpy(((struct resolve_cb_context *)context)->addr,
710 addr, sizeof(struct rdma_dev_addr));
711 ((struct resolve_cb_context *)context)->status = status;
712 complete(&((struct resolve_cb_context *)context)->comp);
713}
714
715int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
716 const union ib_gid *dgid,
717 u8 *dmac, u16 *vlan_id, int *if_index,
718 int *hoplimit)
719{
720 int ret = 0;
721 struct rdma_dev_addr dev_addr;
722 struct resolve_cb_context ctx;
723 struct net_device *dev;
724
725 union {
726 struct sockaddr _sockaddr;
727 struct sockaddr_in _sockaddr_in;
728 struct sockaddr_in6 _sockaddr_in6;
729 } sgid_addr, dgid_addr;
730
731
732 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
733 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
734
735 memset(&dev_addr, 0, sizeof(dev_addr));
736 if (if_index)
737 dev_addr.bound_dev_if = *if_index;
738 dev_addr.net = &init_net;
739
740 ctx.addr = &dev_addr;
741 init_completion(&ctx.comp);
742 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
743 &dev_addr, 1000, resolve_cb, &ctx);
744 if (ret)
745 return ret;
746
747 wait_for_completion(&ctx.comp);
748
749 ret = ctx.status;
750 if (ret)
751 return ret;
752
753 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
754 dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
755 if (!dev)
756 return -ENODEV;
757 if (if_index)
758 *if_index = dev_addr.bound_dev_if;
759 if (vlan_id)
760 *vlan_id = rdma_vlan_dev_vlan_id(dev);
761 if (hoplimit)
762 *hoplimit = dev_addr.hoplimit;
763 dev_put(dev);
764 return ret;
765}
766EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
767
768int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
769{
770 int ret = 0;
771 struct rdma_dev_addr dev_addr;
772 union {
773 struct sockaddr _sockaddr;
774 struct sockaddr_in _sockaddr_in;
775 struct sockaddr_in6 _sockaddr_in6;
776 } gid_addr;
777
778 rdma_gid2ip(&gid_addr._sockaddr, sgid);
779
780 memset(&dev_addr, 0, sizeof(dev_addr));
781 dev_addr.net = &init_net;
782 ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
783 if (ret)
784 return ret;
785
786 memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
787 return ret;
788}
789EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
790
791static int netevent_callback(struct notifier_block *self, unsigned long event,
792 void *ctx)
793{
794 if (event == NETEVENT_NEIGH_UPDATE) {
795 struct neighbour *neigh = ctx;
796
797 if (neigh->nud_state & NUD_VALID) {
798 set_timeout(jiffies);
799 }
800 }
801 return 0;
802}
803
804static struct notifier_block nb = {
805 .notifier_call = netevent_callback
806};
807
808int addr_init(void)
809{
810 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
811 if (!addr_wq)
812 return -ENOMEM;
813
814 register_netevent_notifier(&nb);
815 rdma_addr_register_client(&self);
816
817 return 0;
818}
819
820void addr_cleanup(void)
821{
822 rdma_addr_unregister_client(&self);
823 unregister_netevent_notifier(&nb);
824 destroy_workqueue(addr_wq);
825}