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/slab.h>
43 #include <linux/workqueue.h>
44 #include <linux/module.h>
45 #include <net/if_llatbl.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 ifnet *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)
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_ifp_find(dev_addr->net,
185 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
190 dev = ip6_ifp_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_ifp_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);
401 if (is_gw && nh->gw_sa.sa_family == AF_INET6)
402 error = nd6_resolve(ifp, LLE_SF(AF_INET, is_gw), NULL,
403 &nh->gw_sa, edst, NULL, NULL);
406 error = arpresolve(ifp, is_gw, NULL, is_gw ?
407 &nh->gw_sa : (const struct sockaddr *)&dst_tmp,
413 addr->network = RDMA_NETWORK_IPV4;
417 * Step 4 - update source address, if any
420 src_port = src_in->sin_port;
421 memcpy(src_in, saddr, rdma_addr_size(saddr));
422 src_in->sin_port = src_port; /* preserve port number */
434 if (error == EWOULDBLOCK || error == EAGAIN)
439 static int addr4_resolve(struct sockaddr_in *src_in,
440 const struct sockaddr_in *dst_in,
441 struct rdma_dev_addr *addr,
445 return -EADDRNOTAVAIL;
450 static int addr6_resolve(struct sockaddr_in6 *src_in,
451 const struct sockaddr_in6 *dst_in,
452 struct rdma_dev_addr *addr,
461 struct sockaddr_in6 dst_tmp = *dst_in;
463 struct sockaddr *saddr = NULL;
464 struct nhop_object *nh;
471 /* set VNET, if any */
472 CURVNET_SET(addr->net);
474 /* set default TTL limit */
475 addr->hoplimit = V_ip_defttl;
478 if (ipv6_addr_any(&src_in->sin6_addr))
479 type |= ADDR_SRC_ANY;
480 if (ipv6_addr_any(&dst_tmp.sin6_addr))
481 type |= ADDR_DST_ANY;
484 * Make sure the socket address length field is set.
486 dst_tmp.sin6_len = sizeof(dst_tmp);
489 * Make sure the scope ID gets embedded, else nd6_resolve() will
490 * not find the record.
492 dst_tmp.sin6_scope_id = addr->bound_dev_if;
493 sa6_embedscope(&dst_tmp, 0);
495 /* Step 1 - lookup destination route if any */
498 /* sanity check for IPv4 addresses */
499 if (ipv6_addr_v4mapped(&src_in->sin6_addr) !=
500 ipv6_addr_v4mapped(&dst_tmp.sin6_addr)) {
501 error = EAFNOSUPPORT;
506 /* regular destination route lookup */
507 nh = fib6_lookup(RT_DEFAULT_FIB, &dst_in->sin6_addr,
508 addr->bound_dev_if, NHR_NONE, 0);
510 error = EHOSTUNREACH;
519 /* Step 2 - find outgoing network interface */
522 /* get source interface */
523 if (addr->bound_dev_if != 0) {
524 ifp = dev_get_by_index(addr->net, addr->bound_dev_if);
526 ifp = ip6_ifp_find(addr->net, src_in->sin6_addr, 0);
529 /* check source interface */
533 } else if (ifp->if_flags & IFF_LOOPBACK) {
535 * Source address cannot be a loopback device.
537 error = EHOSTUNREACH;
539 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) {
540 if (memcmp(&src_in->sin6_addr, &dst_in->sin6_addr,
541 sizeof(src_in->sin6_addr))) {
543 * Destination is loopback, but source
544 * and destination address is not the
547 error = EHOSTUNREACH;
550 /* get destination network interface from route */
554 } else if (ifp != nh->nh_ifp) {
556 * Source and destination interfaces are
564 /* check for loopback device */
565 if (nh->nh_ifp->if_flags & IFF_LOOPBACK)
566 saddr = (struct sockaddr *)&dst_tmp;
568 saddr = nh->nh_ifa->ifa_addr;
570 /* get destination network interface from route */
579 * Step 3 - resolve destination MAC address
581 if (IN6_IS_ADDR_MULTICAST(&dst_tmp.sin6_addr)) {
582 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0;
583 error = addr_resolve_multi(edst, ifp,
584 (struct sockaddr *)&dst_tmp);
588 addr->network = RDMA_NETWORK_IPV6;
589 } else if (nh->nh_ifp->if_flags & IFF_LOOPBACK) {
590 memset(edst, 0, MAX_ADDR_LEN);
593 bool is_gw = (nh->nh_flags & NHF_GATEWAY) != 0;
594 memset(edst, 0, MAX_ADDR_LEN);
595 error = nd6_resolve(ifp, LLE_SF(AF_INET6, is_gw), NULL,
596 is_gw ? &nh->gw_sa : (const struct sockaddr *)&dst_tmp,
601 addr->network = RDMA_NETWORK_IPV6;
605 * Step 4 - update source address, if any
608 src_port = src_in->sin6_port;
609 memcpy(src_in, saddr, rdma_addr_size(saddr));
610 src_in->sin6_port = src_port; /* preserve port number */
622 if (error == EWOULDBLOCK || error == EAGAIN)
627 static int addr6_resolve(struct sockaddr_in6 *src_in,
628 const struct sockaddr_in6 *dst_in,
629 struct rdma_dev_addr *addr,
633 return -EADDRNOTAVAIL;
637 static int addr_resolve_neigh(struct ifnet *dev,
638 const struct sockaddr *dst_in,
640 struct rdma_dev_addr *addr)
642 if (dev->if_flags & IFF_LOOPBACK) {
646 * Binding to a loopback device is not allowed. Make
647 * sure the destination device address is global by
648 * clearing the bound device interface:
650 if (addr->bound_dev_if == dev->if_index)
651 addr->bound_dev_if = 0;
653 ret = rdma_translate_ip(dst_in, addr);
655 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
661 /* If the device doesn't do ARP internally */
662 if (!(dev->if_flags & IFF_NOARP))
663 return rdma_copy_addr(addr, dev, edst);
665 return rdma_copy_addr(addr, dev, NULL);
668 static int addr_resolve(struct sockaddr *src_in,
669 const struct sockaddr *dst_in,
670 struct rdma_dev_addr *addr)
672 struct epoch_tracker et;
673 struct ifnet *ndev = NULL;
674 u8 edst[MAX_ADDR_LEN];
677 if (dst_in->sa_family != src_in->sa_family)
681 switch (src_in->sa_family) {
683 ret = addr4_resolve((struct sockaddr_in *)src_in,
684 (const struct sockaddr_in *)dst_in,
688 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
689 (const struct sockaddr_in6 *)dst_in, addr,
693 ret = -EADDRNOTAVAIL;
698 /* check for error */
702 /* store MAC addresses and check for loopback */
703 ret = addr_resolve_neigh(ndev, dst_in, edst, addr);
705 /* set belonging VNET, if any */
706 addr->net = dev_net(ndev);
712 static void process_req(struct work_struct *work)
714 struct addr_req *req, *temp_req;
715 struct sockaddr *src_in, *dst_in;
716 struct list_head done_list;
718 INIT_LIST_HEAD(&done_list);
721 list_for_each_entry_safe(req, temp_req, &req_list, list) {
722 if (req->status == -ENODATA) {
723 src_in = (struct sockaddr *) &req->src_addr;
724 dst_in = (struct sockaddr *) &req->dst_addr;
725 req->status = addr_resolve(src_in, dst_in, req->addr);
726 if (req->status && time_after_eq(jiffies, req->timeout))
727 req->status = -ETIMEDOUT;
728 else if (req->status == -ENODATA)
731 list_move_tail(&req->list, &done_list);
734 if (!list_empty(&req_list)) {
735 req = list_entry(req_list.next, struct addr_req, list);
736 set_timeout(req->timeout);
740 list_for_each_entry_safe(req, temp_req, &done_list, list) {
741 list_del(&req->list);
742 req->callback(req->status, (struct sockaddr *) &req->src_addr,
743 req->addr, req->context);
744 put_client(req->client);
749 int rdma_resolve_ip(struct rdma_addr_client *client,
750 struct sockaddr *src_addr, struct sockaddr *dst_addr,
751 struct rdma_dev_addr *addr, int timeout_ms,
752 void (*callback)(int status, struct sockaddr *src_addr,
753 struct rdma_dev_addr *addr, void *context),
756 struct sockaddr *src_in, *dst_in;
757 struct addr_req *req;
760 req = kzalloc(sizeof *req, GFP_KERNEL);
764 src_in = (struct sockaddr *) &req->src_addr;
765 dst_in = (struct sockaddr *) &req->dst_addr;
768 if (src_addr->sa_family != dst_addr->sa_family) {
773 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
775 src_in->sa_family = dst_addr->sa_family;
778 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
780 req->callback = callback;
781 req->context = context;
782 req->client = client;
783 atomic_inc(&client->refcount);
785 req->status = addr_resolve(src_in, dst_in, addr);
786 switch (req->status) {
788 req->timeout = jiffies;
792 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
797 atomic_dec(&client->refcount);
805 EXPORT_SYMBOL(rdma_resolve_ip);
807 int rdma_resolve_ip_route(struct sockaddr *src_addr,
808 const struct sockaddr *dst_addr,
809 struct rdma_dev_addr *addr)
811 struct sockaddr_storage ssrc_addr = {};
812 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
815 if (src_addr->sa_family != dst_addr->sa_family)
818 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
820 src_in->sa_family = dst_addr->sa_family;
823 return addr_resolve(src_in, dst_addr, addr);
825 EXPORT_SYMBOL(rdma_resolve_ip_route);
827 void rdma_addr_cancel(struct rdma_dev_addr *addr)
829 struct addr_req *req, *temp_req;
832 list_for_each_entry_safe(req, temp_req, &req_list, list) {
833 if (req->addr == addr) {
834 req->status = -ECANCELED;
835 req->timeout = jiffies;
836 list_move(&req->list, &req_list);
837 set_timeout(req->timeout);
843 EXPORT_SYMBOL(rdma_addr_cancel);
845 struct resolve_cb_context {
846 struct rdma_dev_addr *addr;
847 struct completion comp;
851 static void resolve_cb(int status, struct sockaddr *src_addr,
852 struct rdma_dev_addr *addr, void *context)
855 memcpy(((struct resolve_cb_context *)context)->addr,
856 addr, sizeof(struct rdma_dev_addr));
857 ((struct resolve_cb_context *)context)->status = status;
858 complete(&((struct resolve_cb_context *)context)->comp);
861 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
862 const union ib_gid *dgid,
863 u8 *dmac, struct ifnet *dev,
867 struct rdma_dev_addr dev_addr;
868 struct resolve_cb_context ctx;
870 union rdma_sockaddr sgid_addr, dgid_addr;
872 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
873 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
875 memset(&dev_addr, 0, sizeof(dev_addr));
877 dev_addr.bound_dev_if = dev->if_index;
878 dev_addr.net = dev_net(dev);
880 ctx.addr = &dev_addr;
881 init_completion(&ctx.comp);
882 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
883 &dev_addr, 1000, resolve_cb, &ctx);
887 wait_for_completion(&ctx.comp);
893 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
895 *hoplimit = dev_addr.hoplimit;
898 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
902 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
906 rdma_addr_register_client(&self);
911 void addr_cleanup(void)
913 rdma_addr_unregister_client(&self);
914 destroy_workqueue(addr_wq);