1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3 *
4 * Copyright (c) 2005 Voltaire Inc. All rights reserved.
5 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
6 * Copyright (c) 1999-2005, Mellanox Technologies, Inc. All rights reserved.
7 * Copyright (c) 2005 Intel Corporation. All rights reserved.
8 *
9 * This software is available to you under a choice of one of two
10 * licenses. You may choose to be licensed under the terms of the GNU
11 * General Public License (GPL) Version 2, available from the file
12 * COPYING in the main directory of this source tree, or the
13 * OpenIB.org BSD license below:
14 *
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
17 * conditions are met:
18 *
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer.
22 *
23 * - Redistributions in binary form must reproduce the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer in the documentation and/or other materials
26 * provided with the distribution.
27 *
28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 * SOFTWARE.
36 */
37
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40
41 #include <linux/mutex.h>
42 #include <linux/inetdevice.h>
43 #include <linux/slab.h>
44 #include <linux/workqueue.h>
45 #include <linux/module.h>
46 #include <net/route.h>
47 #include <net/netevent.h>
48 #include <rdma/ib_addr.h>
49 #include <rdma/ib.h>
50
51 #include <netinet/if_ether.h>
52 #include <netinet/ip_var.h>
53 #include <netinet6/scope6_var.h>
54 #include <netinet6/in6_pcb.h>
55
56 #include "core_priv.h"
57
58 struct addr_req {
59 struct list_head list;
60 struct sockaddr_storage src_addr;
61 struct sockaddr_storage dst_addr;
62 struct rdma_dev_addr *addr;
63 struct rdma_addr_client *client;
64 void *context;
65 void (*callback)(int status, struct sockaddr *src_addr,
66 struct rdma_dev_addr *addr, void *context);
67 int timeout;
68 int status;
69 };
70
71 static void process_req(struct work_struct *work);
72
73 static DEFINE_MUTEX(lock);
74 static LIST_HEAD(req_list);
75 static DECLARE_DELAYED_WORK(work, process_req);
76 static struct workqueue_struct *addr_wq;
77
rdma_addr_size(struct sockaddr * addr)78 int rdma_addr_size(struct sockaddr *addr)
79 {
80 switch (addr->sa_family) {
81 case AF_INET:
82 return sizeof(struct sockaddr_in);
83 case AF_INET6:
84 return sizeof(struct sockaddr_in6);
85 case AF_IB:
86 return sizeof(struct sockaddr_ib);
87 default:
88 return 0;
89 }
90 }
91 EXPORT_SYMBOL(rdma_addr_size);
92
rdma_addr_size_in6(struct sockaddr_in6 * addr)93 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
94 {
95 int ret = rdma_addr_size((struct sockaddr *) addr);
96
97 return ret <= sizeof(*addr) ? ret : 0;
98 }
99 EXPORT_SYMBOL(rdma_addr_size_in6);
100
rdma_addr_size_kss(struct sockaddr_storage * addr)101 int rdma_addr_size_kss(struct sockaddr_storage *addr)
102 {
103 int ret = rdma_addr_size((struct sockaddr *) addr);
104
105 return ret <= sizeof(*addr) ? ret : 0;
106 }
107 EXPORT_SYMBOL(rdma_addr_size_kss);
108
109 static struct rdma_addr_client self;
110
rdma_addr_register_client(struct rdma_addr_client * client)111 void rdma_addr_register_client(struct rdma_addr_client *client)
112 {
113 atomic_set(&client->refcount, 1);
114 init_completion(&client->comp);
115 }
116 EXPORT_SYMBOL(rdma_addr_register_client);
117
put_client(struct rdma_addr_client * client)118 static inline void put_client(struct rdma_addr_client *client)
119 {
120 if (atomic_dec_and_test(&client->refcount))
121 complete(&client->comp);
122 }
123
rdma_addr_unregister_client(struct rdma_addr_client * client)124 void rdma_addr_unregister_client(struct rdma_addr_client *client)
125 {
126 put_client(client);
127 wait_for_completion(&client->comp);
128 }
129 EXPORT_SYMBOL(rdma_addr_unregister_client);
130
131 static inline void
rdma_copy_addr_sub(u8 * dst,const u8 * src,unsigned min,unsigned max)132 rdma_copy_addr_sub(u8 *dst, const u8 *src, unsigned min, unsigned max)
133 {
134 if (min > max)
135 min = max;
136 memcpy(dst, src, min);
137 memset(dst + min, 0, max - min);
138 }
139
rdma_copy_addr(struct rdma_dev_addr * dev_addr,struct net_device * dev,const unsigned char * dst_dev_addr)140 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
141 const unsigned char *dst_dev_addr)
142 {
143 /* check for loopback device */
144 if (dev->if_flags & IFF_LOOPBACK) {
145 dev_addr->dev_type = ARPHRD_ETHER;
146 memset(dev_addr->src_dev_addr, 0, MAX_ADDR_LEN);
147 memset(dev_addr->broadcast, 0, MAX_ADDR_LEN);
148 memset(dev_addr->dst_dev_addr, 0, MAX_ADDR_LEN);
149 dev_addr->bound_dev_if = dev->if_index;
150 return (0);
151 } else if (dev->if_type == IFT_INFINIBAND)
152 dev_addr->dev_type = ARPHRD_INFINIBAND;
153 else if (dev->if_type == IFT_ETHER)
154 dev_addr->dev_type = ARPHRD_ETHER;
155 else
156 dev_addr->dev_type = 0;
157 rdma_copy_addr_sub(dev_addr->src_dev_addr, IF_LLADDR(dev),
158 dev->if_addrlen, MAX_ADDR_LEN);
159 rdma_copy_addr_sub(dev_addr->broadcast, dev->if_broadcastaddr,
160 dev->if_addrlen, MAX_ADDR_LEN);
161 if (dst_dev_addr != NULL) {
162 rdma_copy_addr_sub(dev_addr->dst_dev_addr, dst_dev_addr,
163 dev->if_addrlen, MAX_ADDR_LEN);
164 }
165 dev_addr->bound_dev_if = dev->if_index;
166 return 0;
167 }
168 EXPORT_SYMBOL(rdma_copy_addr);
169
rdma_translate_ip(const struct sockaddr * addr,struct rdma_dev_addr * dev_addr)170 int rdma_translate_ip(const struct sockaddr *addr,
171 struct rdma_dev_addr *dev_addr)
172 {
173 struct net_device *dev;
174 int ret;
175
176 if (dev_addr->bound_dev_if) {
177 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
178 } else switch (addr->sa_family) {
179 #ifdef INET
180 case AF_INET:
181 dev = ip_dev_find(dev_addr->net,
182 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
183 break;
184 #endif
185 #ifdef INET6
186 case AF_INET6:
187 dev = ip6_dev_find(dev_addr->net,
188 ((const struct sockaddr_in6 *)addr)->sin6_addr, 0);
189 break;
190 #endif
191 default:
192 dev = NULL;
193 break;
194 }
195
196 if (dev != NULL) {
197 /* disallow connections through 127.0.0.1 itself */
198 if (dev->if_flags & IFF_LOOPBACK)
199 ret = -EINVAL;
200 else
201 ret = rdma_copy_addr(dev_addr, dev, NULL);
202 dev_put(dev);
203 } else {
204 ret = -ENODEV;
205 }
206 return ret;
207 }
208 EXPORT_SYMBOL(rdma_translate_ip);
209
set_timeout(int time)210 static void set_timeout(int time)
211 {
212 int delay; /* under FreeBSD ticks are 32-bit */
213
214 delay = time - jiffies;
215 if (delay <= 0)
216 delay = 1;
217 else if (delay > hz)
218 delay = hz;
219
220 mod_delayed_work(addr_wq, &work, delay);
221 }
222
queue_req(struct addr_req * req)223 static void queue_req(struct addr_req *req)
224 {
225 struct addr_req *temp_req;
226
227 mutex_lock(&lock);
228 list_for_each_entry_reverse(temp_req, &req_list, list) {
229 if (time_after_eq(req->timeout, temp_req->timeout))
230 break;
231 }
232
233 list_add(&req->list, &temp_req->list);
234
235 if (req_list.next == &req->list)
236 set_timeout(req->timeout);
237 mutex_unlock(&lock);
238 }
239
240 #if defined(INET) || defined(INET6)
addr_resolve_multi(u8 * edst,struct ifnet * ifp,struct sockaddr * dst_in)241 static int addr_resolve_multi(u8 *edst, struct ifnet *ifp, struct sockaddr *dst_in)
242 {
243 struct sockaddr *llsa;
244 struct sockaddr_dl sdl;
245 int error;
246
247 sdl.sdl_len = sizeof(sdl);
248 llsa = (struct sockaddr *)&sdl;
249
250 if (ifp->if_resolvemulti == NULL) {
251 error = EOPNOTSUPP;
252 } else {
253 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
254 if (error == 0) {
255 rdma_copy_addr_sub(edst, LLADDR((struct sockaddr_dl *)llsa),
256 ifp->if_addrlen, MAX_ADDR_LEN);
257 }
258 }
259 return (error);
260 }
261 #endif
262
263 #ifdef INET
addr4_resolve(struct sockaddr_in * src_in,const struct sockaddr_in * dst_in,struct rdma_dev_addr * addr,u8 * edst,struct ifnet ** ifpp)264 static int addr4_resolve(struct sockaddr_in *src_in,
265 const struct sockaddr_in *dst_in,
266 struct rdma_dev_addr *addr,
267 u8 *edst,
268 struct ifnet **ifpp)
269 {
270 enum {
271 ADDR_VALID = 0,
272 ADDR_SRC_ANY = 1,
273 ADDR_DST_ANY = 2,
274 };
275 struct sockaddr_in dst_tmp = *dst_in;
276 in_port_t src_port;
277 struct sockaddr *saddr = NULL;
278 struct rtentry *rte;
279 struct ifnet *ifp;
280 int error;
281 int type;
282
283 /* set VNET, if any */
284 CURVNET_SET(addr->net);
285
286 /* set default TTL limit */
287 addr->hoplimit = V_ip_defttl;
288
289 type = ADDR_VALID;
290 if (src_in->sin_addr.s_addr == INADDR_ANY)
291 type |= ADDR_SRC_ANY;
292 if (dst_tmp.sin_addr.s_addr == INADDR_ANY)
293 type |= ADDR_DST_ANY;
294
295 /*
296 * Make sure the socket address length field
297 * is set, else rtalloc1() will fail.
298 */
299 dst_tmp.sin_len = sizeof(dst_tmp);
300
301 /* Step 1 - lookup destination route if any */
302 switch (type) {
303 case ADDR_VALID:
304 case ADDR_SRC_ANY:
305 /* regular destination route lookup */
306 rte = rtalloc1((struct sockaddr *)&dst_tmp, 1, 0);
307 if (rte == NULL) {
308 error = EHOSTUNREACH;
309 goto done;
310 } else if (rte->rt_ifp == NULL || RT_LINK_IS_UP(rte->rt_ifp) == 0) {
311 RTFREE_LOCKED(rte);
312 error = EHOSTUNREACH;
313 goto done;
314 }
315 RT_UNLOCK(rte);
316 break;
317 default:
318 error = ENETUNREACH;
319 goto done;
320 }
321
322 /* Step 2 - find outgoing network interface */
323 switch (type) {
324 case ADDR_VALID:
325 /* get source interface */
326 if (addr->bound_dev_if != 0) {
327 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
328 } else {
329 ifp = ip_dev_find(addr->net, src_in->sin_addr.s_addr);
330 }
331
332 /* check source interface */
333 if (ifp == NULL) {
334 error = ENETUNREACH;
335 goto error_rt_free;
336 } else if (ifp->if_flags & IFF_LOOPBACK) {
337 /*
338 * Source address cannot be a loopback device.
339 */
340 error = EHOSTUNREACH;
341 goto error_put_ifp;
342 } else if (rte->rt_ifp->if_flags & IFF_LOOPBACK) {
343 if (memcmp(&src_in->sin_addr, &dst_in->sin_addr,
344 sizeof(src_in->sin_addr))) {
345 /*
346 * Destination is loopback, but source
347 * and destination address is not the
348 * same.
349 */
350 error = EHOSTUNREACH;
351 goto error_put_ifp;
352 }
353 /* get destination network interface from route */
354 dev_put(ifp);
355 ifp = rte->rt_ifp;
356 dev_hold(ifp);
357 } else if (ifp != rte->rt_ifp) {
358 /*
359 * Source and destination interfaces are
360 * different.
361 */
362 error = ENETUNREACH;
363 goto error_put_ifp;
364 }
365 break;
366 case ADDR_SRC_ANY:
367 /* check for loopback device */
368 if (rte->rt_ifp->if_flags & IFF_LOOPBACK)
369 saddr = (struct sockaddr *)&dst_tmp;
370 else
371 saddr = rte->rt_ifa->ifa_addr;
372
373 /* get destination network interface from route */
374 ifp = rte->rt_ifp;
375 dev_hold(ifp);
376 break;
377 default:
378 break;
379 }
380
381 /*
382 * Step 3 - resolve destination MAC address
383 */
384 if (dst_tmp.sin_addr.s_addr == INADDR_BROADCAST) {
385 rdma_copy_addr_sub(edst, ifp->if_broadcastaddr,
386 ifp->if_addrlen, MAX_ADDR_LEN);
387 error = 0;
388 } else if (IN_MULTICAST(ntohl(dst_tmp.sin_addr.s_addr))) {
389 bool is_gw = (rte->rt_flags & RTF_GATEWAY) != 0;
390 error = addr_resolve_multi(edst, ifp, (struct sockaddr *)&dst_tmp);
391 if (error != 0)
392 goto error_put_ifp;
393 else if (is_gw)
394 addr->network = RDMA_NETWORK_IPV4;
395 } else if (ifp->if_flags & IFF_LOOPBACK) {
396 memset(edst, 0, MAX_ADDR_LEN);
397 error = 0;
398 } else {
399 bool is_gw = (rte->rt_flags & RTF_GATEWAY) != 0;
400 memset(edst, 0, MAX_ADDR_LEN);
401 error = arpresolve(ifp, is_gw, NULL, is_gw ?
402 rte->rt_gateway : (const struct sockaddr *)&dst_tmp,
403 edst, NULL, NULL);
404 if (error != 0)
405 goto error_put_ifp;
406 else if (is_gw)
407 addr->network = RDMA_NETWORK_IPV4;
408 }
409
410 /*
411 * Step 4 - update source address, if any
412 */
413 if (saddr != NULL) {
414 src_port = src_in->sin_port;
415 memcpy(src_in, saddr, rdma_addr_size(saddr));
416 src_in->sin_port = src_port; /* preserve port number */
417 }
418
419 if (rte != NULL)
420 RTFREE(rte);
421
422 *ifpp = ifp;
423
424 goto done;
425
426 error_put_ifp:
427 dev_put(ifp);
428 error_rt_free:
429 RTFREE(rte);
430 done:
431 CURVNET_RESTORE();
432
433 if (error == EWOULDBLOCK || error == EAGAIN)
434 error = ENODATA;
435 return (-error);
436 }
437 #else
addr4_resolve(struct sockaddr_in * src_in,const struct sockaddr_in * dst_in,struct rdma_dev_addr * addr,u8 * edst,struct ifnet ** ifpp)438 static int addr4_resolve(struct sockaddr_in *src_in,
439 const struct sockaddr_in *dst_in,
440 struct rdma_dev_addr *addr,
441 u8 *edst,
442 struct ifnet **ifpp)
443 {
444 return -EADDRNOTAVAIL;
445 }
446 #endif
447
448 #ifdef INET6
addr6_resolve(struct sockaddr_in6 * src_in,const struct sockaddr_in6 * dst_in,struct rdma_dev_addr * addr,u8 * edst,struct ifnet ** ifpp)449 static int addr6_resolve(struct sockaddr_in6 *src_in,
450 const struct sockaddr_in6 *dst_in,
451 struct rdma_dev_addr *addr,
452 u8 *edst,
453 struct ifnet **ifpp)
454 {
455 enum {
456 ADDR_VALID = 0,
457 ADDR_SRC_ANY = 1,
458 ADDR_DST_ANY = 2,
459 };
460 struct sockaddr_in6 dst_tmp = *dst_in;
461 in_port_t src_port;
462 struct sockaddr *saddr = NULL;
463 struct rtentry *rte;
464 struct ifnet *ifp;
465 int error;
466 int type;
467
468 /* set VNET, if any */
469 CURVNET_SET(addr->net);
470
471 /* set default TTL limit */
472 addr->hoplimit = V_ip_defttl;
473
474 type = ADDR_VALID;
475 if (ipv6_addr_any(&src_in->sin6_addr))
476 type |= ADDR_SRC_ANY;
477 if (ipv6_addr_any(&dst_tmp.sin6_addr))
478 type |= ADDR_DST_ANY;
479
480 /*
481 * Make sure the socket address length field
482 * is set, else rtalloc1() will fail.
483 */
484 dst_tmp.sin6_len = sizeof(dst_tmp);
485
486 /*
487 * Make sure the scope ID gets embedded, else rtalloc1() will
488 * resolve to the loopback interface.
489 */
490 dst_tmp.sin6_scope_id = addr->bound_dev_if;
491 sa6_embedscope(&dst_tmp, 0);
492
493 /* Step 1 - lookup destination route if any */
494 switch (type) {
495 case ADDR_VALID:
496 /* sanity check for IPv4 addresses */
497 if (ipv6_addr_v4mapped(&src_in->sin6_addr) !=
498 ipv6_addr_v4mapped(&dst_tmp.sin6_addr)) {
499 error = EAFNOSUPPORT;
500 goto done;
501 }
502 /* FALLTHROUGH */
503 case ADDR_SRC_ANY:
504 /* regular destination route lookup */
505 rte = rtalloc1((struct sockaddr *)&dst_tmp, 1, 0);
506 if (rte == NULL) {
507 error = EHOSTUNREACH;
508 goto done;
509 } else if (rte->rt_ifp == NULL || RT_LINK_IS_UP(rte->rt_ifp) == 0) {
510 RTFREE_LOCKED(rte);
511 error = EHOSTUNREACH;
512 goto done;
513 }
514 RT_UNLOCK(rte);
515 break;
516 default:
517 error = ENETUNREACH;
518 goto done;
519 }
520
521 /* Step 2 - find outgoing network interface */
522 switch (type) {
523 case ADDR_VALID:
524 /* get source interface */
525 if (addr->bound_dev_if != 0) {
526 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
527 } else {
528 ifp = ip6_dev_find(addr->net, src_in->sin6_addr, 0);
529 }
530
531 /* check source interface */
532 if (ifp == NULL) {
533 error = ENETUNREACH;
534 goto error_rt_free;
535 } else if (ifp->if_flags & IFF_LOOPBACK) {
536 /*
537 * Source address cannot be a loopback device.
538 */
539 error = EHOSTUNREACH;
540 goto error_put_ifp;
541 } else if (rte->rt_ifp->if_flags & IFF_LOOPBACK) {
542 if (memcmp(&src_in->sin6_addr, &dst_in->sin6_addr,
543 sizeof(src_in->sin6_addr))) {
544 /*
545 * Destination is loopback, but source
546 * and destination address is not the
547 * same.
548 */
549 error = EHOSTUNREACH;
550 goto error_put_ifp;
551 }
552 /* get destination network interface from route */
553 dev_put(ifp);
554 ifp = rte->rt_ifp;
555 dev_hold(ifp);
556 } else if (ifp != rte->rt_ifp) {
557 /*
558 * Source and destination interfaces are
559 * different.
560 */
561 error = ENETUNREACH;
562 goto error_put_ifp;
563 }
564 break;
565 case ADDR_SRC_ANY:
566 /* check for loopback device */
567 if (rte->rt_ifp->if_flags & IFF_LOOPBACK)
568 saddr = (struct sockaddr *)&dst_tmp;
569 else
570 saddr = rte->rt_ifa->ifa_addr;
571
572 /* get destination network interface from route */
573 ifp = rte->rt_ifp;
574 dev_hold(ifp);
575 break;
576 default:
577 break;
578 }
579
580 /*
581 * Step 3 - resolve destination MAC address
582 */
583 if (IN6_IS_ADDR_MULTICAST(&dst_tmp.sin6_addr)) {
584 bool is_gw = (rte->rt_flags & RTF_GATEWAY) != 0;
585 error = addr_resolve_multi(edst, ifp,
586 (struct sockaddr *)&dst_tmp);
587 if (error != 0)
588 goto error_put_ifp;
589 else if (is_gw)
590 addr->network = RDMA_NETWORK_IPV6;
591 } else if (rte->rt_ifp->if_flags & IFF_LOOPBACK) {
592 memset(edst, 0, MAX_ADDR_LEN);
593 error = 0;
594 } else {
595 bool is_gw = (rte->rt_flags & RTF_GATEWAY) != 0;
596 memset(edst, 0, MAX_ADDR_LEN);
597 error = nd6_resolve(ifp, is_gw, NULL, is_gw ?
598 rte->rt_gateway : (const struct sockaddr *)&dst_tmp,
599 edst, NULL, NULL);
600 if (error != 0)
601 goto error_put_ifp;
602 else if (is_gw)
603 addr->network = RDMA_NETWORK_IPV6;
604 }
605
606 /*
607 * Step 4 - update source address, if any
608 */
609 if (saddr != NULL) {
610 src_port = src_in->sin6_port;
611 memcpy(src_in, saddr, rdma_addr_size(saddr));
612 src_in->sin6_port = src_port; /* preserve port number */
613 }
614
615 if (rte != NULL)
616 RTFREE(rte);
617
618 *ifpp = ifp;
619
620 goto done;
621
622 error_put_ifp:
623 dev_put(ifp);
624 error_rt_free:
625 RTFREE(rte);
626 done:
627 CURVNET_RESTORE();
628
629 if (error == EWOULDBLOCK || error == EAGAIN)
630 error = ENODATA;
631 return (-error);
632 }
633 #else
addr6_resolve(struct sockaddr_in6 * src_in,const struct sockaddr_in6 * dst_in,struct rdma_dev_addr * addr,u8 * edst,struct ifnet ** ifpp)634 static int addr6_resolve(struct sockaddr_in6 *src_in,
635 const struct sockaddr_in6 *dst_in,
636 struct rdma_dev_addr *addr,
637 u8 *edst,
638 struct ifnet **ifpp)
639 {
640 return -EADDRNOTAVAIL;
641 }
642 #endif
643
addr_resolve_neigh(struct ifnet * dev,const struct sockaddr * dst_in,u8 * edst,struct rdma_dev_addr * addr)644 static int addr_resolve_neigh(struct ifnet *dev,
645 const struct sockaddr *dst_in,
646 u8 *edst,
647 struct rdma_dev_addr *addr)
648 {
649 if (dev->if_flags & IFF_LOOPBACK) {
650 int ret;
651
652 /*
653 * Binding to a loopback device is not allowed. Make
654 * sure the destination device address is global by
655 * clearing the bound device interface:
656 */
657 if (addr->bound_dev_if == dev->if_index)
658 addr->bound_dev_if = 0;
659
660 ret = rdma_translate_ip(dst_in, addr);
661 if (ret == 0) {
662 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
663 MAX_ADDR_LEN);
664 }
665 return ret;
666 }
667
668 /* If the device doesn't do ARP internally */
669 if (!(dev->if_flags & IFF_NOARP))
670 return rdma_copy_addr(addr, dev, edst);
671
672 return rdma_copy_addr(addr, dev, NULL);
673 }
674
addr_resolve(struct sockaddr * src_in,const struct sockaddr * dst_in,struct rdma_dev_addr * addr)675 static int addr_resolve(struct sockaddr *src_in,
676 const struct sockaddr *dst_in,
677 struct rdma_dev_addr *addr)
678 {
679 struct net_device *ndev = NULL;
680 u8 edst[MAX_ADDR_LEN];
681 int ret;
682
683 if (dst_in->sa_family != src_in->sa_family)
684 return -EINVAL;
685
686 switch (src_in->sa_family) {
687 case AF_INET:
688 ret = addr4_resolve((struct sockaddr_in *)src_in,
689 (const struct sockaddr_in *)dst_in,
690 addr, edst, &ndev);
691 break;
692 case AF_INET6:
693 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
694 (const struct sockaddr_in6 *)dst_in, addr,
695 edst, &ndev);
696 break;
697 default:
698 ret = -EADDRNOTAVAIL;
699 break;
700 }
701
702 /* check for error */
703 if (ret != 0)
704 return ret;
705
706 /* store MAC addresses and check for loopback */
707 ret = addr_resolve_neigh(ndev, dst_in, edst, addr);
708
709 /* set belonging VNET, if any */
710 addr->net = dev_net(ndev);
711 dev_put(ndev);
712
713 return ret;
714 }
715
process_req(struct work_struct * work)716 static void process_req(struct work_struct *work)
717 {
718 struct addr_req *req, *temp_req;
719 struct sockaddr *src_in, *dst_in;
720 struct list_head done_list;
721
722 INIT_LIST_HEAD(&done_list);
723
724 mutex_lock(&lock);
725 list_for_each_entry_safe(req, temp_req, &req_list, list) {
726 if (req->status == -ENODATA) {
727 src_in = (struct sockaddr *) &req->src_addr;
728 dst_in = (struct sockaddr *) &req->dst_addr;
729 req->status = addr_resolve(src_in, dst_in, req->addr);
730 if (req->status && time_after_eq(jiffies, req->timeout))
731 req->status = -ETIMEDOUT;
732 else if (req->status == -ENODATA)
733 continue;
734 }
735 list_move_tail(&req->list, &done_list);
736 }
737
738 if (!list_empty(&req_list)) {
739 req = list_entry(req_list.next, struct addr_req, list);
740 set_timeout(req->timeout);
741 }
742 mutex_unlock(&lock);
743
744 list_for_each_entry_safe(req, temp_req, &done_list, list) {
745 list_del(&req->list);
746 req->callback(req->status, (struct sockaddr *) &req->src_addr,
747 req->addr, req->context);
748 put_client(req->client);
749 kfree(req);
750 }
751 }
752
rdma_resolve_ip(struct rdma_addr_client * client,struct sockaddr * src_addr,struct sockaddr * dst_addr,struct rdma_dev_addr * addr,int timeout_ms,void (* callback)(int status,struct sockaddr * src_addr,struct rdma_dev_addr * addr,void * context),void * context)753 int rdma_resolve_ip(struct rdma_addr_client *client,
754 struct sockaddr *src_addr, struct sockaddr *dst_addr,
755 struct rdma_dev_addr *addr, int timeout_ms,
756 void (*callback)(int status, struct sockaddr *src_addr,
757 struct rdma_dev_addr *addr, void *context),
758 void *context)
759 {
760 struct sockaddr *src_in, *dst_in;
761 struct addr_req *req;
762 int ret = 0;
763
764 req = kzalloc(sizeof *req, GFP_KERNEL);
765 if (!req)
766 return -ENOMEM;
767
768 src_in = (struct sockaddr *) &req->src_addr;
769 dst_in = (struct sockaddr *) &req->dst_addr;
770
771 if (src_addr) {
772 if (src_addr->sa_family != dst_addr->sa_family) {
773 ret = -EINVAL;
774 goto err;
775 }
776
777 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
778 } else {
779 src_in->sa_family = dst_addr->sa_family;
780 }
781
782 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
783 req->addr = addr;
784 req->callback = callback;
785 req->context = context;
786 req->client = client;
787 atomic_inc(&client->refcount);
788
789 req->status = addr_resolve(src_in, dst_in, addr);
790 switch (req->status) {
791 case 0:
792 req->timeout = jiffies;
793 queue_req(req);
794 break;
795 case -ENODATA:
796 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
797 queue_req(req);
798 break;
799 default:
800 ret = req->status;
801 atomic_dec(&client->refcount);
802 goto err;
803 }
804 return ret;
805 err:
806 kfree(req);
807 return ret;
808 }
809 EXPORT_SYMBOL(rdma_resolve_ip);
810
rdma_resolve_ip_route(struct sockaddr * src_addr,const struct sockaddr * dst_addr,struct rdma_dev_addr * addr)811 int rdma_resolve_ip_route(struct sockaddr *src_addr,
812 const struct sockaddr *dst_addr,
813 struct rdma_dev_addr *addr)
814 {
815 struct sockaddr_storage ssrc_addr = {};
816 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
817
818 if (src_addr) {
819 if (src_addr->sa_family != dst_addr->sa_family)
820 return -EINVAL;
821
822 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
823 } else {
824 src_in->sa_family = dst_addr->sa_family;
825 }
826
827 return addr_resolve(src_in, dst_addr, addr);
828 }
829 EXPORT_SYMBOL(rdma_resolve_ip_route);
830
rdma_addr_cancel(struct rdma_dev_addr * addr)831 void rdma_addr_cancel(struct rdma_dev_addr *addr)
832 {
833 struct addr_req *req, *temp_req;
834
835 mutex_lock(&lock);
836 list_for_each_entry_safe(req, temp_req, &req_list, list) {
837 if (req->addr == addr) {
838 req->status = -ECANCELED;
839 req->timeout = jiffies;
840 list_move(&req->list, &req_list);
841 set_timeout(req->timeout);
842 break;
843 }
844 }
845 mutex_unlock(&lock);
846 }
847 EXPORT_SYMBOL(rdma_addr_cancel);
848
849 struct resolve_cb_context {
850 struct rdma_dev_addr *addr;
851 struct completion comp;
852 int status;
853 };
854
resolve_cb(int status,struct sockaddr * src_addr,struct rdma_dev_addr * addr,void * context)855 static void resolve_cb(int status, struct sockaddr *src_addr,
856 struct rdma_dev_addr *addr, void *context)
857 {
858 if (!status)
859 memcpy(((struct resolve_cb_context *)context)->addr,
860 addr, sizeof(struct rdma_dev_addr));
861 ((struct resolve_cb_context *)context)->status = status;
862 complete(&((struct resolve_cb_context *)context)->comp);
863 }
864
rdma_addr_find_l2_eth_by_grh(const union ib_gid * sgid,const union ib_gid * dgid,u8 * dmac,struct net_device * dev,int * hoplimit)865 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
866 const union ib_gid *dgid,
867 u8 *dmac, struct net_device *dev,
868 int *hoplimit)
869 {
870 int ret = 0;
871 struct rdma_dev_addr dev_addr;
872 struct resolve_cb_context ctx;
873
874 union {
875 struct sockaddr _sockaddr;
876 struct sockaddr_in _sockaddr_in;
877 struct sockaddr_in6 _sockaddr_in6;
878 } sgid_addr, dgid_addr;
879
880 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
881 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
882
883 memset(&dev_addr, 0, sizeof(dev_addr));
884
885 dev_addr.bound_dev_if = dev->if_index;
886 dev_addr.net = dev_net(dev);
887
888 ctx.addr = &dev_addr;
889 init_completion(&ctx.comp);
890 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
891 &dev_addr, 1000, resolve_cb, &ctx);
892 if (ret)
893 return ret;
894
895 wait_for_completion(&ctx.comp);
896
897 ret = ctx.status;
898 if (ret)
899 return ret;
900
901 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
902 if (hoplimit)
903 *hoplimit = dev_addr.hoplimit;
904 return ret;
905 }
906 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
907
addr_init(void)908 int addr_init(void)
909 {
910 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
911 if (!addr_wq)
912 return -ENOMEM;
913
914 rdma_addr_register_client(&self);
915
916 return 0;
917 }
918
addr_cleanup(void)919 void addr_cleanup(void)
920 {
921 rdma_addr_unregister_client(&self);
922 destroy_workqueue(addr_wq);
923 }
924