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 struct delayed_work work;
65 int status;
66 u32 seq;
67};
68
69static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
70
71static void process_req(struct work_struct *work);
72
73static DEFINE_MUTEX(lock);
74static LIST_HEAD(req_list);
75static DECLARE_DELAYED_WORK(work, process_req);
76static struct workqueue_struct *addr_wq;
77
78static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
79 [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
80 .len = sizeof(struct rdma_nla_ls_gid)},
81};
82
83static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
84{
85 struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
86 int ret;
87
88 if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
89 return false;
90
91 ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
92 nlmsg_len(nlh), ib_nl_addr_policy, NULL);
93 if (ret)
94 return false;
95
96 return true;
97}
98
99static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
100{
101 const struct nlattr *head, *curr;
102 union ib_gid gid;
103 struct addr_req *req;
104 int len, rem;
105 int found = 0;
106
107 head = (const struct nlattr *)nlmsg_data(nlh);
108 len = nlmsg_len(nlh);
109
110 nla_for_each_attr(curr, head, len, rem) {
111 if (curr->nla_type == LS_NLA_TYPE_DGID)
112 memcpy(&gid, nla_data(curr), nla_len(curr));
113 }
114
115 mutex_lock(&lock);
116 list_for_each_entry(req, &req_list, list) {
117 if (nlh->nlmsg_seq != req->seq)
118 continue;
119 /* We set the DGID part, the rest was set earlier */
120 rdma_addr_set_dgid(req->addr, &gid);
121 req->status = 0;
122 found = 1;
123 break;
124 }
125 mutex_unlock(&lock);
126
127 if (!found)
128 pr_info("Couldn't find request waiting for DGID: %pI6\n",
129 &gid);
130}
131
132int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
133 struct nlmsghdr *nlh,
134 struct netlink_ext_ack *extack)
135{
136 if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
137 !(NETLINK_CB(skb).sk))
138 return -EPERM;
139
140 if (ib_nl_is_good_ip_resp(nlh))
141 ib_nl_process_good_ip_rsep(nlh);
142
143 return skb->len;
144}
145
146static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
147 const void *daddr,
148 u32 seq, u16 family)
149{
150 struct sk_buff *skb = NULL;
151 struct nlmsghdr *nlh;
152 struct rdma_ls_ip_resolve_header *header;
153 void *data;
154 size_t size;
155 int attrtype;
156 int len;
157
158 if (family == AF_INET) {
159 size = sizeof(struct in_addr);
160 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
161 } else {
162 size = sizeof(struct in6_addr);
163 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
164 }
165
166 len = nla_total_size(sizeof(size));
167 len += NLMSG_ALIGN(sizeof(*header));
168
169 skb = nlmsg_new(len, GFP_KERNEL);
170 if (!skb)
171 return -ENOMEM;
172
173 data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
174 RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
175 if (!data) {
176 nlmsg_free(skb);
177 return -ENODATA;
178 }
179
180 /* Construct the family header first */
181 header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
182 header->ifindex = dev_addr->bound_dev_if;
183 nla_put(skb, attrtype, size, daddr);
184
185 /* Repair the nlmsg header length */
186 nlmsg_end(skb, nlh);
187 rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
188
189 /* Make the request retry, so when we get the response from userspace
190 * we will have something.
191 */
192 return -ENODATA;
193}
194
195int rdma_addr_size(struct sockaddr *addr)
196{
197 switch (addr->sa_family) {
198 case AF_INET:
199 return sizeof(struct sockaddr_in);
200 case AF_INET6:
201 return sizeof(struct sockaddr_in6);
202 case AF_IB:
203 return sizeof(struct sockaddr_ib);
204 default:
205 return 0;
206 }
207}
208EXPORT_SYMBOL(rdma_addr_size);
209
210int rdma_addr_size_in6(struct sockaddr_in6 *addr)
211{
212 int ret = rdma_addr_size((struct sockaddr *) addr);
213
214 return ret <= sizeof(*addr) ? ret : 0;
215}
216EXPORT_SYMBOL(rdma_addr_size_in6);
217
218int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
219{
220 int ret = rdma_addr_size((struct sockaddr *) addr);
221
222 return ret <= sizeof(*addr) ? ret : 0;
223}
224EXPORT_SYMBOL(rdma_addr_size_kss);
225
226static struct rdma_addr_client self;
227
228void rdma_addr_register_client(struct rdma_addr_client *client)
229{
230 atomic_set(&client->refcount, 1);
231 init_completion(&client->comp);
232}
233EXPORT_SYMBOL(rdma_addr_register_client);
234
235static inline void put_client(struct rdma_addr_client *client)
236{
237 if (atomic_dec_and_test(&client->refcount))
238 complete(&client->comp);
239}
240
241void rdma_addr_unregister_client(struct rdma_addr_client *client)
242{
243 put_client(client);
244 wait_for_completion(&client->comp);
245}
246EXPORT_SYMBOL(rdma_addr_unregister_client);
247
248void rdma_copy_addr(struct rdma_dev_addr *dev_addr,
249 const struct net_device *dev,
250 const unsigned char *dst_dev_addr)
251{
252 dev_addr->dev_type = dev->type;
253 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
254 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
255 if (dst_dev_addr)
256 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
257 dev_addr->bound_dev_if = dev->ifindex;
258}
259EXPORT_SYMBOL(rdma_copy_addr);
260
261int rdma_translate_ip(const struct sockaddr *addr,
262 struct rdma_dev_addr *dev_addr)
263{
264 struct net_device *dev;
265
266 if (dev_addr->bound_dev_if) {
267 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
268 if (!dev)
269 return -ENODEV;
270 rdma_copy_addr(dev_addr, dev, NULL);
271 dev_put(dev);
272 return 0;
273 }
274
275 switch (addr->sa_family) {
276 case AF_INET:
277 dev = ip_dev_find(dev_addr->net,
278 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
279
280 if (!dev)
281 return -EADDRNOTAVAIL;
282
283 rdma_copy_addr(dev_addr, dev, NULL);
284 dev_put(dev);
285 break;
286#if IS_ENABLED(CONFIG_IPV6)
287 case AF_INET6:
288 rcu_read_lock();
289 for_each_netdev_rcu(dev_addr->net, dev) {
290 if (ipv6_chk_addr(dev_addr->net,
291 &((const struct sockaddr_in6 *)addr)->sin6_addr,
292 dev, 1)) {
293 rdma_copy_addr(dev_addr, dev, NULL);
294 break;
295 }
296 }
297 rcu_read_unlock();
298 break;
299#endif
300 }
301 return 0;
302}
303EXPORT_SYMBOL(rdma_translate_ip);
304
305static void set_timeout(struct delayed_work *delayed_work, unsigned long time)
306{
307 unsigned long delay;
308
309 delay = time - jiffies;
310 if ((long)delay < 0)
311 delay = 0;
312
313 mod_delayed_work(addr_wq, delayed_work, delay);
314}
315
316static void queue_req(struct addr_req *req)
317{
318 struct addr_req *temp_req;
319
320 mutex_lock(&lock);
321 list_for_each_entry_reverse(temp_req, &req_list, list) {
322 if (time_after_eq(req->timeout, temp_req->timeout))
323 break;
324 }
325
326 list_add(&req->list, &temp_req->list);
327
328 set_timeout(&req->work, req->timeout);
329 mutex_unlock(&lock);
330}
331
332static int ib_nl_fetch_ha(const struct dst_entry *dst,
333 struct rdma_dev_addr *dev_addr,
334 const void *daddr, u32 seq, u16 family)
335{
336 if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
337 return -EADDRNOTAVAIL;
338
339 /* We fill in what we can, the response will fill the rest */
340 rdma_copy_addr(dev_addr, dst->dev, NULL);
341 return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
342}
343
344static int dst_fetch_ha(const struct dst_entry *dst,
345 struct rdma_dev_addr *dev_addr,
346 const void *daddr)
347{
348 struct neighbour *n;
349 int ret = 0;
350
351 n = dst_neigh_lookup(dst, daddr);
352
353 rcu_read_lock();
354 if (!n || !(n->nud_state & NUD_VALID)) {
355 if (n)
356 neigh_event_send(n, NULL);
357 ret = -ENODATA;
358 } else {
359 rdma_copy_addr(dev_addr, dst->dev, n->ha);
360 }
361 rcu_read_unlock();
362
363 if (n)
364 neigh_release(n);
365
366 return ret;
367}
368
369static bool has_gateway(const struct dst_entry *dst, sa_family_t family)
370{
371 struct rtable *rt;
372 struct rt6_info *rt6;
373
374 if (family == AF_INET) {
375 rt = container_of(dst, struct rtable, dst);
376 return rt->rt_uses_gateway;
377 }
378
379 rt6 = container_of(dst, struct rt6_info, dst);
380 return rt6->rt6i_flags & RTF_GATEWAY;
381}
382
383static int fetch_ha(const struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
384 const struct sockaddr *dst_in, u32 seq)
385{
386 const struct sockaddr_in *dst_in4 =
387 (const struct sockaddr_in *)dst_in;
388 const struct sockaddr_in6 *dst_in6 =
389 (const struct sockaddr_in6 *)dst_in;
390 const void *daddr = (dst_in->sa_family == AF_INET) ?
391 (const void *)&dst_in4->sin_addr.s_addr :
392 (const void *)&dst_in6->sin6_addr;
393 sa_family_t family = dst_in->sa_family;
394
395 /* Gateway + ARPHRD_INFINIBAND -> IB router */
396 if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
397 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
398 else
399 return dst_fetch_ha(dst, dev_addr, daddr);
400}
401
402static int addr4_resolve(struct sockaddr_in *src_in,
403 const struct sockaddr_in *dst_in,
404 struct rdma_dev_addr *addr,
405 struct rtable **prt)
406{
407 __be32 src_ip = src_in->sin_addr.s_addr;
408 __be32 dst_ip = dst_in->sin_addr.s_addr;
409 struct rtable *rt;
410 struct flowi4 fl4;
411 int ret;
412
413 memset(&fl4, 0, sizeof(fl4));
414 fl4.daddr = dst_ip;
415 fl4.saddr = src_ip;
416 fl4.flowi4_oif = addr->bound_dev_if;
417 rt = ip_route_output_key(addr->net, &fl4);
418 ret = PTR_ERR_OR_ZERO(rt);
419 if (ret)
420 return ret;
421
422 src_in->sin_family = AF_INET;
423 src_in->sin_addr.s_addr = fl4.saddr;
424
425 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
426 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
427 * type accordingly.
428 */
429 if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
430 addr->network = RDMA_NETWORK_IPV4;
431
432 addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
433
434 *prt = rt;
435 return 0;
436}
437
438#if IS_ENABLED(CONFIG_IPV6)
439static int addr6_resolve(struct sockaddr_in6 *src_in,
440 const struct sockaddr_in6 *dst_in,
441 struct rdma_dev_addr *addr,
442 struct dst_entry **pdst)
443{
444 struct flowi6 fl6;
445 struct dst_entry *dst;
446 struct rt6_info *rt;
447 int ret;
448
449 memset(&fl6, 0, sizeof fl6);
450 fl6.daddr = dst_in->sin6_addr;
451 fl6.saddr = src_in->sin6_addr;
452 fl6.flowi6_oif = addr->bound_dev_if;
453
454 ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
455 if (ret < 0)
456 return ret;
457
458 rt = (struct rt6_info *)dst;
459 if (ipv6_addr_any(&src_in->sin6_addr)) {
460 src_in->sin6_family = AF_INET6;
461 src_in->sin6_addr = fl6.saddr;
462 }
463
464 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
465 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
466 * type accordingly.
467 */
468 if (rt->rt6i_flags & RTF_GATEWAY &&
469 ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
470 addr->network = RDMA_NETWORK_IPV6;
471
472 addr->hoplimit = ip6_dst_hoplimit(dst);
473
474 *pdst = dst;
475 return 0;
476}
477#else
478static int addr6_resolve(struct sockaddr_in6 *src_in,
479 const struct sockaddr_in6 *dst_in,
480 struct rdma_dev_addr *addr,
481 struct dst_entry **pdst)
482{
483 return -EADDRNOTAVAIL;
484}
485#endif
486
487static int addr_resolve_neigh(const struct dst_entry *dst,
488 const struct sockaddr *dst_in,
489 struct rdma_dev_addr *addr,
490 u32 seq)
491{
492 if (dst->dev->flags & IFF_LOOPBACK) {
493 int ret;
494
495 ret = rdma_translate_ip(dst_in, addr);
496 if (!ret)
497 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
498 MAX_ADDR_LEN);
499
500 return ret;
501 }
502
503 /* If the device doesn't do ARP internally */
504 if (!(dst->dev->flags & IFF_NOARP))
505 return fetch_ha(dst, addr, dst_in, seq);
506
507 rdma_copy_addr(addr, dst->dev, NULL);
508
509 return 0;
510}
511
512static int addr_resolve(struct sockaddr *src_in,
513 const struct sockaddr *dst_in,
514 struct rdma_dev_addr *addr,
515 bool resolve_neigh,
516 u32 seq)
517{
518 struct net_device *ndev;
519 struct dst_entry *dst;
520 int ret;
521
522 if (!addr->net) {
523 pr_warn_ratelimited("%s: missing namespace\n", __func__);
524 return -EINVAL;
525 }
526
527 if (src_in->sa_family == AF_INET) {
528 struct rtable *rt = NULL;
529 const struct sockaddr_in *dst_in4 =
530 (const struct sockaddr_in *)dst_in;
531
532 ret = addr4_resolve((struct sockaddr_in *)src_in,
533 dst_in4, addr, &rt);
534 if (ret)
535 return ret;
536
537 if (resolve_neigh)
538 ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
539
540 if (addr->bound_dev_if) {
541 ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
542 } else {
543 ndev = rt->dst.dev;
544 dev_hold(ndev);
545 }
546
547 ip_rt_put(rt);
548 } else {
549 const struct sockaddr_in6 *dst_in6 =
550 (const struct sockaddr_in6 *)dst_in;
551
552 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
553 dst_in6, addr,
554 &dst);
555 if (ret)
556 return ret;
557
558 if (resolve_neigh)
559 ret = addr_resolve_neigh(dst, dst_in, addr, seq);
560
561 if (addr->bound_dev_if) {
562 ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
563 } else {
564 ndev = dst->dev;
565 dev_hold(ndev);
566 }
567
568 dst_release(dst);
569 }
570
571 if (ndev) {
572 if (ndev->flags & IFF_LOOPBACK)
573 ret = rdma_translate_ip(dst_in, addr);
574 else
575 addr->bound_dev_if = ndev->ifindex;
576 dev_put(ndev);
577 }
578
579 return ret;
580}
581
582static void process_one_req(struct work_struct *_work)
583{
584 struct addr_req *req;
585 struct sockaddr *src_in, *dst_in;
586
587 mutex_lock(&lock);
588 req = container_of(_work, struct addr_req, work.work);
589
590 if (req->status == -ENODATA) {
591 src_in = (struct sockaddr *)&req->src_addr;
592 dst_in = (struct sockaddr *)&req->dst_addr;
593 req->status = addr_resolve(src_in, dst_in, req->addr,
594 true, req->seq);
595 if (req->status && time_after_eq(jiffies, req->timeout)) {
596 req->status = -ETIMEDOUT;
597 } else if (req->status == -ENODATA) {
598 /* requeue the work for retrying again */
599 set_timeout(&req->work, req->timeout);
600 mutex_unlock(&lock);
601 return;
602 }
603 }
604 list_del(&req->list);
605 mutex_unlock(&lock);
606
607 /*
608 * Although the work will normally have been canceled by the
609 * workqueue, it can still be requeued as long as it is on the
610 * req_list, so it could have been requeued before we grabbed &lock.
611 * We need to cancel it after it is removed from req_list to really be
612 * sure it is safe to free.
613 */
614 cancel_delayed_work(&req->work);
615
616 req->callback(req->status, (struct sockaddr *)&req->src_addr,
617 req->addr, req->context);
618 put_client(req->client);
619 kfree(req);
620}
621
622static void process_req(struct work_struct *work)
623{
624 struct addr_req *req, *temp_req;
625 struct sockaddr *src_in, *dst_in;
626 struct list_head done_list;
627
628 INIT_LIST_HEAD(&done_list);
629
630 mutex_lock(&lock);
631 list_for_each_entry_safe(req, temp_req, &req_list, list) {
632 if (req->status == -ENODATA) {
633 src_in = (struct sockaddr *) &req->src_addr;
634 dst_in = (struct sockaddr *) &req->dst_addr;
635 req->status = addr_resolve(src_in, dst_in, req->addr,
636 true, req->seq);
637 if (req->status && time_after_eq(jiffies, req->timeout))
638 req->status = -ETIMEDOUT;
639 else if (req->status == -ENODATA) {
640 set_timeout(&req->work, req->timeout);
641 continue;
642 }
643 }
644 list_move_tail(&req->list, &done_list);
645 }
646
647 mutex_unlock(&lock);
648
649 list_for_each_entry_safe(req, temp_req, &done_list, list) {
650 list_del(&req->list);
651 /* It is safe to cancel other work items from this work item
652 * because at a time there can be only one work item running
653 * with this single threaded work queue.
654 */
655 cancel_delayed_work(&req->work);
656 req->callback(req->status, (struct sockaddr *) &req->src_addr,
657 req->addr, req->context);
658 put_client(req->client);
659 kfree(req);
660 }
661}
662
663int rdma_resolve_ip(struct rdma_addr_client *client,
664 struct sockaddr *src_addr, struct sockaddr *dst_addr,
665 struct rdma_dev_addr *addr, int timeout_ms,
666 void (*callback)(int status, struct sockaddr *src_addr,
667 struct rdma_dev_addr *addr, void *context),
668 void *context)
669{
670 struct sockaddr *src_in, *dst_in;
671 struct addr_req *req;
672 int ret = 0;
673
674 req = kzalloc(sizeof *req, GFP_KERNEL);
675 if (!req)
676 return -ENOMEM;
677
678 src_in = (struct sockaddr *) &req->src_addr;
679 dst_in = (struct sockaddr *) &req->dst_addr;
680
681 if (src_addr) {
682 if (src_addr->sa_family != dst_addr->sa_family) {
683 ret = -EINVAL;
684 goto err;
685 }
686
687 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
688 } else {
689 src_in->sa_family = dst_addr->sa_family;
690 }
691
692 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
693 req->addr = addr;
694 req->callback = callback;
695 req->context = context;
696 req->client = client;
697 atomic_inc(&client->refcount);
698 INIT_DELAYED_WORK(&req->work, process_one_req);
699 req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
700
701 req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
702 switch (req->status) {
703 case 0:
704 req->timeout = jiffies;
705 queue_req(req);
706 break;
707 case -ENODATA:
708 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
709 queue_req(req);
710 break;
711 default:
712 ret = req->status;
713 atomic_dec(&client->refcount);
714 goto err;
715 }
716 return ret;
717err:
718 kfree(req);
719 return ret;
720}
721EXPORT_SYMBOL(rdma_resolve_ip);
722
723int rdma_resolve_ip_route(struct sockaddr *src_addr,
724 const struct sockaddr *dst_addr,
725 struct rdma_dev_addr *addr)
726{
727 struct sockaddr_storage ssrc_addr = {};
728 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
729
730 if (src_addr) {
731 if (src_addr->sa_family != dst_addr->sa_family)
732 return -EINVAL;
733
734 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
735 } else {
736 src_in->sa_family = dst_addr->sa_family;
737 }
738
739 return addr_resolve(src_in, dst_addr, addr, false, 0);
740}
741
742void rdma_addr_cancel(struct rdma_dev_addr *addr)
743{
744 struct addr_req *req, *temp_req;
745
746 mutex_lock(&lock);
747 list_for_each_entry_safe(req, temp_req, &req_list, list) {
748 if (req->addr == addr) {
749 req->status = -ECANCELED;
750 req->timeout = jiffies;
751 list_move(&req->list, &req_list);
752 set_timeout(&req->work, req->timeout);
753 break;
754 }
755 }
756 mutex_unlock(&lock);
757}
758EXPORT_SYMBOL(rdma_addr_cancel);
759
760struct resolve_cb_context {
761 struct completion comp;
762 int status;
763};
764
765static void resolve_cb(int status, struct sockaddr *src_addr,
766 struct rdma_dev_addr *addr, void *context)
767{
768 ((struct resolve_cb_context *)context)->status = status;
769 complete(&((struct resolve_cb_context *)context)->comp);
770}
771
772int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
773 const union ib_gid *dgid,
774 u8 *dmac, const struct net_device *ndev,
775 int *hoplimit)
776{
777 struct rdma_dev_addr dev_addr;
778 struct resolve_cb_context ctx;
779 union {
780 struct sockaddr _sockaddr;
781 struct sockaddr_in _sockaddr_in;
782 struct sockaddr_in6 _sockaddr_in6;
783 } sgid_addr, dgid_addr;
784 int ret;
785
786 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
787 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
788
789 memset(&dev_addr, 0, sizeof(dev_addr));
790 dev_addr.bound_dev_if = ndev->ifindex;
791 dev_addr.net = &init_net;
792
793 init_completion(&ctx.comp);
794 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
795 &dev_addr, 1000, resolve_cb, &ctx);
796 if (ret)
797 return ret;
798
799 wait_for_completion(&ctx.comp);
800
801 ret = ctx.status;
802 if (ret)
803 return ret;
804
805 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
806 *hoplimit = dev_addr.hoplimit;
807 return 0;
808}
809
810static int netevent_callback(struct notifier_block *self, unsigned long event,
811 void *ctx)
812{
813 if (event == NETEVENT_NEIGH_UPDATE) {
814 struct neighbour *neigh = ctx;
815
816 if (neigh->nud_state & NUD_VALID)
817 set_timeout(&work, jiffies);
818 }
819 return 0;
820}
821
822static struct notifier_block nb = {
823 .notifier_call = netevent_callback
824};
825
826int addr_init(void)
827{
828 addr_wq = alloc_ordered_workqueue("ib_addr", 0);
829 if (!addr_wq)
830 return -ENOMEM;
831
832 register_netevent_notifier(&nb);
833 rdma_addr_register_client(&self);
834
835 return 0;
836}
837
838void addr_cleanup(void)
839{
840 rdma_addr_unregister_client(&self);
841 unregister_netevent_notifier(&nb);
842 destroy_workqueue(addr_wq);
843}