2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
4 * Copyright (c) 2005 Voltaire Inc. All rights reserved.
5 * Copyright (c) 2002-2005, Network Appliance, Inc. All rights reserved.
6 * Copyright (c) 1999-2019, Mellanox Technologies, Inc. All rights reserved.
7 * Copyright (c) 2005 Intel Corporation. All rights reserved.
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:
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
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.
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
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
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>
51 #include <netinet/if_ether.h>
52 #include <netinet/ip_var.h>
53 #include <netinet6/scope6_var.h>
54 #include <netinet6/in6_pcb.h>
56 #include "core_priv.h"
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;
65 void (*callback)(int status, struct sockaddr *src_addr,
66 struct rdma_dev_addr *addr, void *context);
71 static void process_req(struct work_struct *work);
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;
78 int rdma_addr_size(struct sockaddr *addr)
80 switch (addr->sa_family) {
82 return sizeof(struct sockaddr_in);
84 return sizeof(struct sockaddr_in6);
86 return sizeof(struct sockaddr_ib);
91 EXPORT_SYMBOL(rdma_addr_size);
93 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
95 int ret = rdma_addr_size((struct sockaddr *) addr);
97 return ret <= sizeof(*addr) ? ret : 0;
99 EXPORT_SYMBOL(rdma_addr_size_in6);
101 int rdma_addr_size_kss(struct sockaddr_storage *addr)
103 int ret = rdma_addr_size((struct sockaddr *) addr);
105 return ret <= sizeof(*addr) ? ret : 0;
107 EXPORT_SYMBOL(rdma_addr_size_kss);
109 static struct rdma_addr_client self;
111 void rdma_addr_register_client(struct rdma_addr_client *client)
113 atomic_set(&client->refcount, 1);
114 init_completion(&client->comp);
116 EXPORT_SYMBOL(rdma_addr_register_client);
118 static inline void put_client(struct rdma_addr_client *client)
120 if (atomic_dec_and_test(&client->refcount))
121 complete(&client->comp);
124 void rdma_addr_unregister_client(struct rdma_addr_client *client)
127 wait_for_completion(&client->comp);
129 EXPORT_SYMBOL(rdma_addr_unregister_client);
132 rdma_copy_addr_sub(u8 *dst, const u8 *src, unsigned min, unsigned max)
136 memcpy(dst, src, min);
137 memset(dst + min, 0, max - min);
140 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
141 const unsigned char *dst_dev_addr)
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;
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;
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);
165 dev_addr->bound_dev_if = dev->if_index;
168 EXPORT_SYMBOL(rdma_copy_addr);
170 int rdma_translate_ip(const struct sockaddr *addr,
171 struct rdma_dev_addr *dev_addr)
173 struct net_device *dev;
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) {
181 dev = ip_dev_find(dev_addr->net,
182 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
187 dev = ip6_dev_find(dev_addr->net,
188 ((const struct sockaddr_in6 *)addr)->sin6_addr, 0);
197 /* disallow connections through 127.0.0.1 itself */
198 if (dev->if_flags & IFF_LOOPBACK)
201 ret = rdma_copy_addr(dev_addr, dev, NULL);
208 EXPORT_SYMBOL(rdma_translate_ip);
210 static void set_timeout(int time)
212 int delay; /* under FreeBSD ticks are 32-bit */
214 delay = time - jiffies;
220 mod_delayed_work(addr_wq, &work, delay);
223 static void queue_req(struct addr_req *req)
225 struct addr_req *temp_req;
228 list_for_each_entry_reverse(temp_req, &req_list, list) {
229 if (time_after_eq(req->timeout, temp_req->timeout))
233 list_add(&req->list, &temp_req->list);
235 if (req_list.next == &req->list)
236 set_timeout(req->timeout);
240 #if defined(INET) || defined(INET6)
241 static int addr_resolve_multi(u8 *edst, struct ifnet *ifp, struct sockaddr *dst_in)
243 struct sockaddr *llsa;
244 struct sockaddr_dl sdl;
247 sdl.sdl_len = sizeof(sdl);
248 llsa = (struct sockaddr *)&sdl;
250 if (ifp->if_resolvemulti == NULL) {
253 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
255 rdma_copy_addr_sub(edst, LLADDR((struct sockaddr_dl *)llsa),
256 ifp->if_addrlen, MAX_ADDR_LEN);
264 static int addr4_resolve(struct sockaddr_in *src_in,
265 const struct sockaddr_in *dst_in,
266 struct rdma_dev_addr *addr,
275 struct sockaddr_in dst_tmp = *dst_in;
277 struct sockaddr *saddr = NULL;
283 /* set VNET, if any */
284 CURVNET_SET(addr->net);
286 /* set default TTL limit */
287 addr->hoplimit = V_ip_defttl;
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;
296 * Make sure the socket address length field
297 * is set, else rtalloc1() will fail.
299 dst_tmp.sin_len = sizeof(dst_tmp);
301 /* Step 1 - lookup destination route if any */
305 /* regular destination route lookup */
306 rte = rtalloc1((struct sockaddr *)&dst_tmp, 1, 0);
308 error = EHOSTUNREACH;
310 } else if (rte->rt_ifp == NULL || RT_LINK_IS_UP(rte->rt_ifp) == 0) {
312 error = EHOSTUNREACH;
322 /* Step 2 - find outgoing network interface */
325 /* get source interface */
326 if (addr->bound_dev_if != 0) {
327 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
329 ifp = ip_dev_find(addr->net, src_in->sin_addr.s_addr);
332 /* check source interface */
336 } else if (ifp->if_flags & IFF_LOOPBACK) {
338 * Source address cannot be a loopback device.
340 error = EHOSTUNREACH;
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))) {
346 * Destination is loopback, but source
347 * and destination address is not the
350 error = EHOSTUNREACH;
353 /* get destination network interface from route */
357 } else if (ifp != rte->rt_ifp) {
359 * Source and destination interfaces are
367 /* check for loopback device */
368 if (rte->rt_ifp->if_flags & IFF_LOOPBACK)
369 saddr = (struct sockaddr *)&dst_tmp;
371 saddr = rte->rt_ifa->ifa_addr;
373 /* get destination network interface from route */
382 * Step 3 - resolve destination MAC address
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);
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);
394 addr->network = RDMA_NETWORK_IPV4;
395 } else if (ifp->if_flags & IFF_LOOPBACK) {
396 memset(edst, 0, MAX_ADDR_LEN);
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,
407 addr->network = RDMA_NETWORK_IPV4;
411 * Step 4 - update source address, if any
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 */
433 if (error == EWOULDBLOCK || error == EAGAIN)
438 static int addr4_resolve(struct sockaddr_in *src_in,
439 const struct sockaddr_in *dst_in,
440 struct rdma_dev_addr *addr,
444 return -EADDRNOTAVAIL;
449 static int addr6_resolve(struct sockaddr_in6 *src_in,
450 const struct sockaddr_in6 *dst_in,
451 struct rdma_dev_addr *addr,
460 struct sockaddr_in6 dst_tmp = *dst_in;
462 struct sockaddr *saddr = NULL;
468 /* set VNET, if any */
469 CURVNET_SET(addr->net);
471 /* set default TTL limit */
472 addr->hoplimit = V_ip_defttl;
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;
481 * Make sure the socket address length field
482 * is set, else rtalloc1() will fail.
484 dst_tmp.sin6_len = sizeof(dst_tmp);
487 * Make sure the scope ID gets embedded, else rtalloc1() will
488 * resolve to the loopback interface.
490 dst_tmp.sin6_scope_id = addr->bound_dev_if;
491 sa6_embedscope(&dst_tmp, 0);
493 /* Step 1 - lookup destination route if any */
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;
504 /* regular destination route lookup */
505 rte = rtalloc1((struct sockaddr *)&dst_tmp, 1, 0);
507 error = EHOSTUNREACH;
509 } else if (rte->rt_ifp == NULL || RT_LINK_IS_UP(rte->rt_ifp) == 0) {
511 error = EHOSTUNREACH;
521 /* Step 2 - find outgoing network interface */
524 /* get source interface */
525 if (addr->bound_dev_if != 0) {
526 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
528 ifp = ip6_dev_find(addr->net, src_in->sin6_addr, 0);
531 /* check source interface */
535 } else if (ifp->if_flags & IFF_LOOPBACK) {
537 * Source address cannot be a loopback device.
539 error = EHOSTUNREACH;
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))) {
545 * Destination is loopback, but source
546 * and destination address is not the
549 error = EHOSTUNREACH;
552 /* get destination network interface from route */
556 } else if (ifp != rte->rt_ifp) {
558 * Source and destination interfaces are
566 /* check for loopback device */
567 if (rte->rt_ifp->if_flags & IFF_LOOPBACK)
568 saddr = (struct sockaddr *)&dst_tmp;
570 saddr = rte->rt_ifa->ifa_addr;
572 /* get destination network interface from route */
581 * Step 3 - resolve destination MAC address
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);
590 addr->network = RDMA_NETWORK_IPV6;
591 } else if (rte->rt_ifp->if_flags & IFF_LOOPBACK) {
592 memset(edst, 0, MAX_ADDR_LEN);
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,
603 addr->network = RDMA_NETWORK_IPV6;
607 * Step 4 - update source address, if any
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 */
629 if (error == EWOULDBLOCK || error == EAGAIN)
634 static int addr6_resolve(struct sockaddr_in6 *src_in,
635 const struct sockaddr_in6 *dst_in,
636 struct rdma_dev_addr *addr,
640 return -EADDRNOTAVAIL;
644 static int addr_resolve_neigh(struct ifnet *dev,
645 const struct sockaddr *dst_in,
647 struct rdma_dev_addr *addr)
649 if (dev->if_flags & IFF_LOOPBACK) {
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:
657 if (addr->bound_dev_if == dev->if_index)
658 addr->bound_dev_if = 0;
660 ret = rdma_translate_ip(dst_in, addr);
662 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
668 /* If the device doesn't do ARP internally */
669 if (!(dev->if_flags & IFF_NOARP))
670 return rdma_copy_addr(addr, dev, edst);
672 return rdma_copy_addr(addr, dev, NULL);
675 static int addr_resolve(struct sockaddr *src_in,
676 const struct sockaddr *dst_in,
677 struct rdma_dev_addr *addr)
679 struct epoch_tracker et;
680 struct net_device *ndev = NULL;
681 u8 edst[MAX_ADDR_LEN];
684 if (dst_in->sa_family != src_in->sa_family)
688 switch (src_in->sa_family) {
690 ret = addr4_resolve((struct sockaddr_in *)src_in,
691 (const struct sockaddr_in *)dst_in,
695 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
696 (const struct sockaddr_in6 *)dst_in, addr,
700 ret = -EADDRNOTAVAIL;
705 /* check for error */
709 /* store MAC addresses and check for loopback */
710 ret = addr_resolve_neigh(ndev, dst_in, edst, addr);
712 /* set belonging VNET, if any */
713 addr->net = dev_net(ndev);
719 static void process_req(struct work_struct *work)
721 struct addr_req *req, *temp_req;
722 struct sockaddr *src_in, *dst_in;
723 struct list_head done_list;
725 INIT_LIST_HEAD(&done_list);
728 list_for_each_entry_safe(req, temp_req, &req_list, list) {
729 if (req->status == -ENODATA) {
730 src_in = (struct sockaddr *) &req->src_addr;
731 dst_in = (struct sockaddr *) &req->dst_addr;
732 req->status = addr_resolve(src_in, dst_in, req->addr);
733 if (req->status && time_after_eq(jiffies, req->timeout))
734 req->status = -ETIMEDOUT;
735 else if (req->status == -ENODATA)
738 list_move_tail(&req->list, &done_list);
741 if (!list_empty(&req_list)) {
742 req = list_entry(req_list.next, struct addr_req, list);
743 set_timeout(req->timeout);
747 list_for_each_entry_safe(req, temp_req, &done_list, list) {
748 list_del(&req->list);
749 req->callback(req->status, (struct sockaddr *) &req->src_addr,
750 req->addr, req->context);
751 put_client(req->client);
756 int rdma_resolve_ip(struct rdma_addr_client *client,
757 struct sockaddr *src_addr, struct sockaddr *dst_addr,
758 struct rdma_dev_addr *addr, int timeout_ms,
759 void (*callback)(int status, struct sockaddr *src_addr,
760 struct rdma_dev_addr *addr, void *context),
763 struct sockaddr *src_in, *dst_in;
764 struct addr_req *req;
767 req = kzalloc(sizeof *req, GFP_KERNEL);
771 src_in = (struct sockaddr *) &req->src_addr;
772 dst_in = (struct sockaddr *) &req->dst_addr;
775 if (src_addr->sa_family != dst_addr->sa_family) {
780 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
782 src_in->sa_family = dst_addr->sa_family;
785 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
787 req->callback = callback;
788 req->context = context;
789 req->client = client;
790 atomic_inc(&client->refcount);
792 req->status = addr_resolve(src_in, dst_in, addr);
793 switch (req->status) {
795 req->timeout = jiffies;
799 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
804 atomic_dec(&client->refcount);
812 EXPORT_SYMBOL(rdma_resolve_ip);
814 int rdma_resolve_ip_route(struct sockaddr *src_addr,
815 const struct sockaddr *dst_addr,
816 struct rdma_dev_addr *addr)
818 struct sockaddr_storage ssrc_addr = {};
819 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
822 if (src_addr->sa_family != dst_addr->sa_family)
825 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
827 src_in->sa_family = dst_addr->sa_family;
830 return addr_resolve(src_in, dst_addr, addr);
832 EXPORT_SYMBOL(rdma_resolve_ip_route);
834 void rdma_addr_cancel(struct rdma_dev_addr *addr)
836 struct addr_req *req, *temp_req;
839 list_for_each_entry_safe(req, temp_req, &req_list, list) {
840 if (req->addr == addr) {
841 req->status = -ECANCELED;
842 req->timeout = jiffies;
843 list_move(&req->list, &req_list);
844 set_timeout(req->timeout);
850 EXPORT_SYMBOL(rdma_addr_cancel);
852 struct resolve_cb_context {
853 struct rdma_dev_addr *addr;
854 struct completion comp;
858 static void resolve_cb(int status, struct sockaddr *src_addr,
859 struct rdma_dev_addr *addr, void *context)
862 memcpy(((struct resolve_cb_context *)context)->addr,
863 addr, sizeof(struct rdma_dev_addr));
864 ((struct resolve_cb_context *)context)->status = status;
865 complete(&((struct resolve_cb_context *)context)->comp);
868 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
869 const union ib_gid *dgid,
870 u8 *dmac, struct net_device *dev,
874 struct rdma_dev_addr dev_addr;
875 struct resolve_cb_context ctx;
878 struct sockaddr _sockaddr;
879 struct sockaddr_in _sockaddr_in;
880 struct sockaddr_in6 _sockaddr_in6;
881 } sgid_addr, dgid_addr;
883 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
884 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
886 memset(&dev_addr, 0, sizeof(dev_addr));
888 dev_addr.bound_dev_if = dev->if_index;
889 dev_addr.net = dev_net(dev);
891 ctx.addr = &dev_addr;
892 init_completion(&ctx.comp);
893 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
894 &dev_addr, 1000, resolve_cb, &ctx);
898 wait_for_completion(&ctx.comp);
904 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
906 *hoplimit = dev_addr.hoplimit;
909 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
913 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
917 rdma_addr_register_client(&self);
922 void addr_cleanup(void)
924 rdma_addr_unregister_client(&self);
925 destroy_workqueue(addr_wq);