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/route/nhop.h>
48 #include <net/netevent.h>
49 #include <rdma/ib_addr.h>
52 #include <netinet/in_fib.h>
53 #include <netinet/if_ether.h>
54 #include <netinet/ip_var.h>
55 #include <netinet6/scope6_var.h>
56 #include <netinet6/in6_pcb.h>
57 #include <netinet6/in6_fib.h>
59 #include "core_priv.h"
62 struct list_head list;
63 struct sockaddr_storage src_addr;
64 struct sockaddr_storage dst_addr;
65 struct rdma_dev_addr *addr;
66 struct rdma_addr_client *client;
68 void (*callback)(int status, struct sockaddr *src_addr,
69 struct rdma_dev_addr *addr, void *context);
74 static void process_req(struct work_struct *work);
76 static DEFINE_MUTEX(lock);
77 static LIST_HEAD(req_list);
78 static DECLARE_DELAYED_WORK(work, process_req);
79 static struct workqueue_struct *addr_wq;
81 int rdma_addr_size(struct sockaddr *addr)
83 switch (addr->sa_family) {
85 return sizeof(struct sockaddr_in);
87 return sizeof(struct sockaddr_in6);
89 return sizeof(struct sockaddr_ib);
94 EXPORT_SYMBOL(rdma_addr_size);
96 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
98 int ret = rdma_addr_size((struct sockaddr *) addr);
100 return ret <= sizeof(*addr) ? ret : 0;
102 EXPORT_SYMBOL(rdma_addr_size_in6);
104 int rdma_addr_size_kss(struct sockaddr_storage *addr)
106 int ret = rdma_addr_size((struct sockaddr *) addr);
108 return ret <= sizeof(*addr) ? ret : 0;
110 EXPORT_SYMBOL(rdma_addr_size_kss);
112 static struct rdma_addr_client self;
114 void rdma_addr_register_client(struct rdma_addr_client *client)
116 atomic_set(&client->refcount, 1);
117 init_completion(&client->comp);
119 EXPORT_SYMBOL(rdma_addr_register_client);
121 static inline void put_client(struct rdma_addr_client *client)
123 if (atomic_dec_and_test(&client->refcount))
124 complete(&client->comp);
127 void rdma_addr_unregister_client(struct rdma_addr_client *client)
130 wait_for_completion(&client->comp);
132 EXPORT_SYMBOL(rdma_addr_unregister_client);
135 rdma_copy_addr_sub(u8 *dst, const u8 *src, unsigned min, unsigned max)
139 memcpy(dst, src, min);
140 memset(dst + min, 0, max - min);
143 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
144 const unsigned char *dst_dev_addr)
146 /* check for loopback device */
147 if (dev->if_flags & IFF_LOOPBACK) {
148 dev_addr->dev_type = ARPHRD_ETHER;
149 memset(dev_addr->src_dev_addr, 0, MAX_ADDR_LEN);
150 memset(dev_addr->broadcast, 0, MAX_ADDR_LEN);
151 memset(dev_addr->dst_dev_addr, 0, MAX_ADDR_LEN);
152 dev_addr->bound_dev_if = dev->if_index;
154 } else if (dev->if_type == IFT_INFINIBAND)
155 dev_addr->dev_type = ARPHRD_INFINIBAND;
156 else if (dev->if_type == IFT_ETHER)
157 dev_addr->dev_type = ARPHRD_ETHER;
159 dev_addr->dev_type = 0;
160 rdma_copy_addr_sub(dev_addr->src_dev_addr, IF_LLADDR(dev),
161 dev->if_addrlen, MAX_ADDR_LEN);
162 rdma_copy_addr_sub(dev_addr->broadcast, dev->if_broadcastaddr,
163 dev->if_addrlen, MAX_ADDR_LEN);
164 if (dst_dev_addr != NULL) {
165 rdma_copy_addr_sub(dev_addr->dst_dev_addr, dst_dev_addr,
166 dev->if_addrlen, MAX_ADDR_LEN);
168 dev_addr->bound_dev_if = dev->if_index;
171 EXPORT_SYMBOL(rdma_copy_addr);
173 int rdma_translate_ip(const struct sockaddr *addr,
174 struct rdma_dev_addr *dev_addr)
176 struct net_device *dev;
179 if (dev_addr->bound_dev_if) {
180 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
181 } else switch (addr->sa_family) {
184 dev = ip_dev_find(dev_addr->net,
185 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
190 dev = ip6_dev_find(dev_addr->net,
191 ((const struct sockaddr_in6 *)addr)->sin6_addr, 0);
200 /* disallow connections through 127.0.0.1 itself */
201 if (dev->if_flags & IFF_LOOPBACK)
204 ret = rdma_copy_addr(dev_addr, dev, NULL);
211 EXPORT_SYMBOL(rdma_translate_ip);
213 static void set_timeout(int time)
215 int delay; /* under FreeBSD ticks are 32-bit */
217 delay = time - jiffies;
223 mod_delayed_work(addr_wq, &work, delay);
226 static void queue_req(struct addr_req *req)
228 struct addr_req *temp_req;
231 list_for_each_entry_reverse(temp_req, &req_list, list) {
232 if (time_after_eq(req->timeout, temp_req->timeout))
236 list_add(&req->list, &temp_req->list);
238 if (req_list.next == &req->list)
239 set_timeout(req->timeout);
243 #if defined(INET) || defined(INET6)
244 static int addr_resolve_multi(u8 *edst, struct ifnet *ifp, struct sockaddr *dst_in)
246 struct sockaddr *llsa;
247 struct sockaddr_dl sdl;
250 sdl.sdl_len = sizeof(sdl);
251 llsa = (struct sockaddr *)&sdl;
253 if (ifp->if_resolvemulti == NULL) {
256 error = ifp->if_resolvemulti(ifp, &llsa, dst_in);
258 rdma_copy_addr_sub(edst, LLADDR((struct sockaddr_dl *)llsa),
259 ifp->if_addrlen, MAX_ADDR_LEN);
267 static int addr4_resolve(struct sockaddr_in *src_in,
268 const struct sockaddr_in *dst_in,
269 struct rdma_dev_addr *addr,
278 struct sockaddr_in dst_tmp = *dst_in;
280 struct sockaddr *saddr = NULL;
281 struct nhop_object *nh;
288 /* set VNET, if any */
289 CURVNET_SET(addr->net);
291 /* set default TTL limit */
292 addr->hoplimit = V_ip_defttl;
295 if (src_in->sin_addr.s_addr == INADDR_ANY)
296 type |= ADDR_SRC_ANY;
297 if (dst_tmp.sin_addr.s_addr == INADDR_ANY)
298 type |= ADDR_DST_ANY;
301 * Make sure the socket address length field is set.
303 dst_tmp.sin_len = sizeof(dst_tmp);
305 /* Step 1 - lookup destination route if any */
309 /* regular destination route lookup */
310 nh = fib4_lookup(RT_DEFAULT_FIB, dst_tmp.sin_addr,0,NHR_NONE,0);
312 error = EHOSTUNREACH;
321 /* Step 2 - find outgoing network interface */
324 /* get source interface */
325 if (addr->bound_dev_if != 0) {
326 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
328 ifp = ip_dev_find(addr->net, src_in->sin_addr.s_addr);
331 /* check source interface */
335 } else if (ifp->if_flags & IFF_LOOPBACK) {
337 * Source address cannot be a loopback device.
339 error = EHOSTUNREACH;
341 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) {
342 if (memcmp(&src_in->sin_addr, &dst_in->sin_addr,
343 sizeof(src_in->sin_addr))) {
345 * Destination is loopback, but source
346 * and destination address is not the
349 error = EHOSTUNREACH;
352 /* get destination network interface from route */
356 } else if (ifp != nh->nh_ifp) {
358 * Source and destination interfaces are
366 /* check for loopback device */
367 if (nh->nh_ifp->if_flags & IFF_LOOPBACK)
368 saddr = (struct sockaddr *)&dst_tmp;
370 saddr = nh->nh_ifa->ifa_addr;
372 /* get destination network interface from route */
381 * Step 3 - resolve destination MAC address
383 if (dst_tmp.sin_addr.s_addr == INADDR_BROADCAST) {
384 rdma_copy_addr_sub(edst, ifp->if_broadcastaddr,
385 ifp->if_addrlen, MAX_ADDR_LEN);
387 } else if (IN_MULTICAST(ntohl(dst_tmp.sin_addr.s_addr))) {
388 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0;
389 error = addr_resolve_multi(edst, ifp, (struct sockaddr *)&dst_tmp);
393 addr->network = RDMA_NETWORK_IPV4;
394 } else if (ifp->if_flags & IFF_LOOPBACK) {
395 memset(edst, 0, MAX_ADDR_LEN);
398 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0;
399 memset(edst, 0, MAX_ADDR_LEN);
400 error = arpresolve(ifp, is_gw, NULL, is_gw ?
401 &nh->gw_sa : (const struct sockaddr *)&dst_tmp,
406 addr->network = RDMA_NETWORK_IPV4;
410 * Step 4 - update source address, if any
413 src_port = src_in->sin_port;
414 memcpy(src_in, saddr, rdma_addr_size(saddr));
415 src_in->sin_port = src_port; /* preserve port number */
427 if (error == EWOULDBLOCK || error == EAGAIN)
432 static int addr4_resolve(struct sockaddr_in *src_in,
433 const struct sockaddr_in *dst_in,
434 struct rdma_dev_addr *addr,
438 return -EADDRNOTAVAIL;
443 static int addr6_resolve(struct sockaddr_in6 *src_in,
444 const struct sockaddr_in6 *dst_in,
445 struct rdma_dev_addr *addr,
454 struct sockaddr_in6 dst_tmp = *dst_in;
456 struct sockaddr *saddr = NULL;
457 struct nhop_object *nh;
464 /* set VNET, if any */
465 CURVNET_SET(addr->net);
467 /* set default TTL limit */
468 addr->hoplimit = V_ip_defttl;
471 if (ipv6_addr_any(&src_in->sin6_addr))
472 type |= ADDR_SRC_ANY;
473 if (ipv6_addr_any(&dst_tmp.sin6_addr))
474 type |= ADDR_DST_ANY;
477 * Make sure the socket address length field is set.
479 dst_tmp.sin6_len = sizeof(dst_tmp);
482 * Make sure the scope ID gets embedded, else nd6_resolve() will
483 * not find the record.
485 dst_tmp.sin6_scope_id = addr->bound_dev_if;
486 sa6_embedscope(&dst_tmp, 0);
488 /* Step 1 - lookup destination route if any */
491 /* sanity check for IPv4 addresses */
492 if (ipv6_addr_v4mapped(&src_in->sin6_addr) !=
493 ipv6_addr_v4mapped(&dst_tmp.sin6_addr)) {
494 error = EAFNOSUPPORT;
499 /* regular destination route lookup */
500 nh = fib6_lookup(RT_DEFAULT_FIB, &dst_in->sin6_addr,
501 addr->bound_dev_if, NHR_NONE, 0);
503 error = EHOSTUNREACH;
512 /* Step 2 - find outgoing network interface */
515 /* get source interface */
516 if (addr->bound_dev_if != 0) {
517 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
519 ifp = ip6_dev_find(addr->net, src_in->sin6_addr, 0);
522 /* check source interface */
526 } else if (ifp->if_flags & IFF_LOOPBACK) {
528 * Source address cannot be a loopback device.
530 error = EHOSTUNREACH;
532 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) {
533 if (memcmp(&src_in->sin6_addr, &dst_in->sin6_addr,
534 sizeof(src_in->sin6_addr))) {
536 * Destination is loopback, but source
537 * and destination address is not the
540 error = EHOSTUNREACH;
543 /* get destination network interface from route */
547 } else if (ifp != nh->nh_ifp) {
549 * Source and destination interfaces are
557 /* check for loopback device */
558 if (nh->nh_ifp->if_flags & IFF_LOOPBACK)
559 saddr = (struct sockaddr *)&dst_tmp;
561 saddr = nh->nh_ifa->ifa_addr;
563 /* get destination network interface from route */
572 * Step 3 - resolve destination MAC address
574 if (IN6_IS_ADDR_MULTICAST(&dst_tmp.sin6_addr)) {
575 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0;
576 error = addr_resolve_multi(edst, ifp,
577 (struct sockaddr *)&dst_tmp);
581 addr->network = RDMA_NETWORK_IPV6;
582 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) {
583 memset(edst, 0, MAX_ADDR_LEN);
586 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0;
587 memset(edst, 0, MAX_ADDR_LEN);
588 error = nd6_resolve(ifp, is_gw, NULL, is_gw ?
589 &nh->gw_sa : (const struct sockaddr *)&dst_tmp,
594 addr->network = RDMA_NETWORK_IPV6;
598 * Step 4 - update source address, if any
601 src_port = src_in->sin6_port;
602 memcpy(src_in, saddr, rdma_addr_size(saddr));
603 src_in->sin6_port = src_port; /* preserve port number */
615 if (error == EWOULDBLOCK || error == EAGAIN)
620 static int addr6_resolve(struct sockaddr_in6 *src_in,
621 const struct sockaddr_in6 *dst_in,
622 struct rdma_dev_addr *addr,
626 return -EADDRNOTAVAIL;
630 static int addr_resolve_neigh(struct ifnet *dev,
631 const struct sockaddr *dst_in,
633 struct rdma_dev_addr *addr)
635 if (dev->if_flags & IFF_LOOPBACK) {
639 * Binding to a loopback device is not allowed. Make
640 * sure the destination device address is global by
641 * clearing the bound device interface:
643 if (addr->bound_dev_if == dev->if_index)
644 addr->bound_dev_if = 0;
646 ret = rdma_translate_ip(dst_in, addr);
648 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
654 /* If the device doesn't do ARP internally */
655 if (!(dev->if_flags & IFF_NOARP))
656 return rdma_copy_addr(addr, dev, edst);
658 return rdma_copy_addr(addr, dev, NULL);
661 static int addr_resolve(struct sockaddr *src_in,
662 const struct sockaddr *dst_in,
663 struct rdma_dev_addr *addr)
665 struct epoch_tracker et;
666 struct net_device *ndev = NULL;
667 u8 edst[MAX_ADDR_LEN];
670 if (dst_in->sa_family != src_in->sa_family)
674 switch (src_in->sa_family) {
676 ret = addr4_resolve((struct sockaddr_in *)src_in,
677 (const struct sockaddr_in *)dst_in,
681 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
682 (const struct sockaddr_in6 *)dst_in, addr,
686 ret = -EADDRNOTAVAIL;
691 /* check for error */
695 /* store MAC addresses and check for loopback */
696 ret = addr_resolve_neigh(ndev, dst_in, edst, addr);
698 /* set belonging VNET, if any */
699 addr->net = dev_net(ndev);
705 static void process_req(struct work_struct *work)
707 struct addr_req *req, *temp_req;
708 struct sockaddr *src_in, *dst_in;
709 struct list_head done_list;
711 INIT_LIST_HEAD(&done_list);
714 list_for_each_entry_safe(req, temp_req, &req_list, list) {
715 if (req->status == -ENODATA) {
716 src_in = (struct sockaddr *) &req->src_addr;
717 dst_in = (struct sockaddr *) &req->dst_addr;
718 req->status = addr_resolve(src_in, dst_in, req->addr);
719 if (req->status && time_after_eq(jiffies, req->timeout))
720 req->status = -ETIMEDOUT;
721 else if (req->status == -ENODATA)
724 list_move_tail(&req->list, &done_list);
727 if (!list_empty(&req_list)) {
728 req = list_entry(req_list.next, struct addr_req, list);
729 set_timeout(req->timeout);
733 list_for_each_entry_safe(req, temp_req, &done_list, list) {
734 list_del(&req->list);
735 req->callback(req->status, (struct sockaddr *) &req->src_addr,
736 req->addr, req->context);
737 put_client(req->client);
742 int rdma_resolve_ip(struct rdma_addr_client *client,
743 struct sockaddr *src_addr, struct sockaddr *dst_addr,
744 struct rdma_dev_addr *addr, int timeout_ms,
745 void (*callback)(int status, struct sockaddr *src_addr,
746 struct rdma_dev_addr *addr, void *context),
749 struct sockaddr *src_in, *dst_in;
750 struct addr_req *req;
753 req = kzalloc(sizeof *req, GFP_KERNEL);
757 src_in = (struct sockaddr *) &req->src_addr;
758 dst_in = (struct sockaddr *) &req->dst_addr;
761 if (src_addr->sa_family != dst_addr->sa_family) {
766 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
768 src_in->sa_family = dst_addr->sa_family;
771 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
773 req->callback = callback;
774 req->context = context;
775 req->client = client;
776 atomic_inc(&client->refcount);
778 req->status = addr_resolve(src_in, dst_in, addr);
779 switch (req->status) {
781 req->timeout = jiffies;
785 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
790 atomic_dec(&client->refcount);
798 EXPORT_SYMBOL(rdma_resolve_ip);
800 int rdma_resolve_ip_route(struct sockaddr *src_addr,
801 const struct sockaddr *dst_addr,
802 struct rdma_dev_addr *addr)
804 struct sockaddr_storage ssrc_addr = {};
805 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
808 if (src_addr->sa_family != dst_addr->sa_family)
811 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
813 src_in->sa_family = dst_addr->sa_family;
816 return addr_resolve(src_in, dst_addr, addr);
818 EXPORT_SYMBOL(rdma_resolve_ip_route);
820 void rdma_addr_cancel(struct rdma_dev_addr *addr)
822 struct addr_req *req, *temp_req;
825 list_for_each_entry_safe(req, temp_req, &req_list, list) {
826 if (req->addr == addr) {
827 req->status = -ECANCELED;
828 req->timeout = jiffies;
829 list_move(&req->list, &req_list);
830 set_timeout(req->timeout);
836 EXPORT_SYMBOL(rdma_addr_cancel);
838 struct resolve_cb_context {
839 struct rdma_dev_addr *addr;
840 struct completion comp;
844 static void resolve_cb(int status, struct sockaddr *src_addr,
845 struct rdma_dev_addr *addr, void *context)
848 memcpy(((struct resolve_cb_context *)context)->addr,
849 addr, sizeof(struct rdma_dev_addr));
850 ((struct resolve_cb_context *)context)->status = status;
851 complete(&((struct resolve_cb_context *)context)->comp);
854 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
855 const union ib_gid *dgid,
856 u8 *dmac, struct net_device *dev,
860 struct rdma_dev_addr dev_addr;
861 struct resolve_cb_context ctx;
864 struct sockaddr _sockaddr;
865 struct sockaddr_in _sockaddr_in;
866 struct sockaddr_in6 _sockaddr_in6;
867 } sgid_addr, dgid_addr;
869 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
870 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
872 memset(&dev_addr, 0, sizeof(dev_addr));
874 dev_addr.bound_dev_if = dev->if_index;
875 dev_addr.net = dev_net(dev);
877 ctx.addr = &dev_addr;
878 init_completion(&ctx.comp);
879 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
880 &dev_addr, 1000, resolve_cb, &ctx);
884 wait_for_completion(&ctx.comp);
890 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
892 *hoplimit = dev_addr.hoplimit;
895 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
899 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
903 rdma_addr_register_client(&self);
908 void addr_cleanup(void)
910 rdma_addr_unregister_client(&self);
911 destroy_workqueue(addr_wq);