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.
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:
13 * Redistribution and use in source and binary forms, with or
14 * without modification, are permitted provided that the following
17 * - Redistributions of source code must retain the above
18 * copyright notice, this list of conditions and the following
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.
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
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>
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>
49 #include <rdma/rdma_netlink.h>
50 #include <net/netlink.h>
52 #include "core_priv.h"
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;
61 void (*callback)(int status, struct sockaddr *src_addr,
62 struct rdma_dev_addr *addr, void *context);
63 unsigned long timeout;
68 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
70 static void process_req(struct work_struct *work);
72 static DEFINE_MUTEX(lock);
73 static LIST_HEAD(req_list);
74 static DECLARE_DELAYED_WORK(work, process_req);
75 static struct workqueue_struct *addr_wq;
77 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
78 [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
79 .len = sizeof(struct rdma_nla_ls_gid)},
82 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
84 struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
87 if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
90 ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
91 nlmsg_len(nlh), ib_nl_addr_policy);
98 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
100 const struct nlattr *head, *curr;
102 struct addr_req *req;
106 head = (const struct nlattr *)nlmsg_data(nlh);
107 len = nlmsg_len(nlh);
109 nla_for_each_attr(curr, head, len, rem) {
110 if (curr->nla_type == LS_NLA_TYPE_DGID)
111 memcpy(&gid, nla_data(curr), nla_len(curr));
115 list_for_each_entry(req, &req_list, list) {
116 if (nlh->nlmsg_seq != req->seq)
118 /* We set the DGID part, the rest was set earlier */
119 rdma_addr_set_dgid(req->addr, &gid);
127 pr_info("Couldn't find request waiting for DGID: %pI6\n",
131 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
132 struct netlink_callback *cb)
134 const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
136 if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
137 !(NETLINK_CB(skb).sk) ||
138 !netlink_capable(skb, CAP_NET_ADMIN))
141 if (ib_nl_is_good_ip_resp(nlh))
142 ib_nl_process_good_ip_rsep(nlh);
147 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
151 struct sk_buff *skb = NULL;
152 struct nlmsghdr *nlh;
153 struct rdma_ls_ip_resolve_header *header;
159 if (family == AF_INET) {
160 size = sizeof(struct in_addr);
161 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
163 size = sizeof(struct in6_addr);
164 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
167 len = nla_total_size(sizeof(size));
168 len += NLMSG_ALIGN(sizeof(*header));
170 skb = nlmsg_new(len, GFP_KERNEL);
174 data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
175 RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
181 /* Construct the family header first */
182 header = (struct rdma_ls_ip_resolve_header *)
183 skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
184 header->ifindex = dev_addr->bound_dev_if;
185 nla_put(skb, attrtype, size, daddr);
187 /* Repair the nlmsg header length */
189 ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, GFP_KERNEL);
191 /* Make the request retry, so when we get the response from userspace
192 * we will have something.
197 int rdma_addr_size(struct sockaddr *addr)
199 switch (addr->sa_family) {
201 return sizeof(struct sockaddr_in);
203 return sizeof(struct sockaddr_in6);
205 return sizeof(struct sockaddr_ib);
210 EXPORT_SYMBOL(rdma_addr_size);
212 static struct rdma_addr_client self;
214 void rdma_addr_register_client(struct rdma_addr_client *client)
216 atomic_set(&client->refcount, 1);
217 init_completion(&client->comp);
219 EXPORT_SYMBOL(rdma_addr_register_client);
221 static inline void put_client(struct rdma_addr_client *client)
223 if (atomic_dec_and_test(&client->refcount))
224 complete(&client->comp);
227 void rdma_addr_unregister_client(struct rdma_addr_client *client)
230 wait_for_completion(&client->comp);
232 EXPORT_SYMBOL(rdma_addr_unregister_client);
234 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
235 const unsigned char *dst_dev_addr)
237 dev_addr->dev_type = dev->type;
238 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
239 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
241 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
242 dev_addr->bound_dev_if = dev->ifindex;
245 EXPORT_SYMBOL(rdma_copy_addr);
247 int rdma_translate_ip(const struct sockaddr *addr,
248 struct rdma_dev_addr *dev_addr,
251 struct net_device *dev;
252 int ret = -EADDRNOTAVAIL;
254 if (dev_addr->bound_dev_if) {
255 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
258 ret = rdma_copy_addr(dev_addr, dev, NULL);
263 switch (addr->sa_family) {
265 dev = ip_dev_find(dev_addr->net,
266 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
271 ret = rdma_copy_addr(dev_addr, dev, NULL);
273 *vlan_id = rdma_vlan_dev_vlan_id(dev);
276 #if IS_ENABLED(CONFIG_IPV6)
279 for_each_netdev_rcu(dev_addr->net, dev) {
280 if (ipv6_chk_addr(dev_addr->net,
281 &((const struct sockaddr_in6 *)addr)->sin6_addr,
283 ret = rdma_copy_addr(dev_addr, dev, NULL);
285 *vlan_id = rdma_vlan_dev_vlan_id(dev);
295 EXPORT_SYMBOL(rdma_translate_ip);
297 static void set_timeout(unsigned long time)
301 delay = time - jiffies;
305 mod_delayed_work(addr_wq, &work, delay);
308 static void queue_req(struct addr_req *req)
310 struct addr_req *temp_req;
313 list_for_each_entry_reverse(temp_req, &req_list, list) {
314 if (time_after_eq(req->timeout, temp_req->timeout))
318 list_add(&req->list, &temp_req->list);
320 if (req_list.next == &req->list)
321 set_timeout(req->timeout);
325 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
326 const void *daddr, u32 seq, u16 family)
328 if (ibnl_chk_listeners(RDMA_NL_GROUP_LS))
329 return -EADDRNOTAVAIL;
331 /* We fill in what we can, the response will fill the rest */
332 rdma_copy_addr(dev_addr, dst->dev, NULL);
333 return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
336 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
342 n = dst_neigh_lookup(dst, daddr);
345 if (!n || !(n->nud_state & NUD_VALID)) {
347 neigh_event_send(n, NULL);
350 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
360 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
363 struct rt6_info *rt6;
365 if (family == AF_INET) {
366 rt = container_of(dst, struct rtable, dst);
367 return rt->rt_uses_gateway;
370 rt6 = container_of(dst, struct rt6_info, dst);
371 return rt6->rt6i_flags & RTF_GATEWAY;
374 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
375 const struct sockaddr *dst_in, u32 seq)
377 const struct sockaddr_in *dst_in4 =
378 (const struct sockaddr_in *)dst_in;
379 const struct sockaddr_in6 *dst_in6 =
380 (const struct sockaddr_in6 *)dst_in;
381 const void *daddr = (dst_in->sa_family == AF_INET) ?
382 (const void *)&dst_in4->sin_addr.s_addr :
383 (const void *)&dst_in6->sin6_addr;
384 sa_family_t family = dst_in->sa_family;
386 /* Gateway + ARPHRD_INFINIBAND -> IB router */
387 if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
388 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
390 return dst_fetch_ha(dst, dev_addr, daddr);
393 static int addr4_resolve(struct sockaddr_in *src_in,
394 const struct sockaddr_in *dst_in,
395 struct rdma_dev_addr *addr,
398 __be32 src_ip = src_in->sin_addr.s_addr;
399 __be32 dst_ip = dst_in->sin_addr.s_addr;
404 memset(&fl4, 0, sizeof(fl4));
407 fl4.flowi4_oif = addr->bound_dev_if;
408 rt = ip_route_output_key(addr->net, &fl4);
413 src_in->sin_family = AF_INET;
414 src_in->sin_addr.s_addr = fl4.saddr;
416 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
417 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
420 if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
421 addr->network = RDMA_NETWORK_IPV4;
423 addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
431 #if IS_ENABLED(CONFIG_IPV6)
432 static int addr6_resolve(struct sockaddr_in6 *src_in,
433 const struct sockaddr_in6 *dst_in,
434 struct rdma_dev_addr *addr,
435 struct dst_entry **pdst)
438 struct dst_entry *dst;
442 memset(&fl6, 0, sizeof fl6);
443 fl6.daddr = dst_in->sin6_addr;
444 fl6.saddr = src_in->sin6_addr;
445 fl6.flowi6_oif = addr->bound_dev_if;
447 dst = ip6_route_output(addr->net, NULL, &fl6);
448 if ((ret = dst->error))
451 rt = (struct rt6_info *)dst;
452 if (ipv6_addr_any(&fl6.saddr)) {
453 ret = ipv6_dev_get_saddr(addr->net, ip6_dst_idev(dst)->dev,
454 &fl6.daddr, 0, &fl6.saddr);
458 src_in->sin6_family = AF_INET6;
459 src_in->sin6_addr = fl6.saddr;
462 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
463 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
466 if (rt->rt6i_flags & RTF_GATEWAY &&
467 ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
468 addr->network = RDMA_NETWORK_IPV6;
470 addr->hoplimit = ip6_dst_hoplimit(dst);
479 static int addr6_resolve(struct sockaddr_in6 *src_in,
480 const struct sockaddr_in6 *dst_in,
481 struct rdma_dev_addr *addr,
482 struct dst_entry **pdst)
484 return -EADDRNOTAVAIL;
488 static int addr_resolve_neigh(struct dst_entry *dst,
489 const struct sockaddr *dst_in,
490 struct rdma_dev_addr *addr,
493 if (dst->dev->flags & IFF_LOOPBACK) {
496 ret = rdma_translate_ip(dst_in, addr, NULL);
498 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
504 /* If the device doesn't do ARP internally */
505 if (!(dst->dev->flags & IFF_NOARP))
506 return fetch_ha(dst, addr, dst_in, seq);
508 return rdma_copy_addr(addr, dst->dev, NULL);
511 static int addr_resolve(struct sockaddr *src_in,
512 const struct sockaddr *dst_in,
513 struct rdma_dev_addr *addr,
517 struct net_device *ndev;
518 struct dst_entry *dst;
521 if (src_in->sa_family == AF_INET) {
522 struct rtable *rt = NULL;
523 const struct sockaddr_in *dst_in4 =
524 (const struct sockaddr_in *)dst_in;
526 ret = addr4_resolve((struct sockaddr_in *)src_in,
532 ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
539 const struct sockaddr_in6 *dst_in6 =
540 (const struct sockaddr_in6 *)dst_in;
542 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
549 ret = addr_resolve_neigh(dst, dst_in, addr, seq);
557 addr->bound_dev_if = ndev->ifindex;
558 addr->net = dev_net(ndev);
564 static void process_req(struct work_struct *work)
566 struct addr_req *req, *temp_req;
567 struct sockaddr *src_in, *dst_in;
568 struct list_head done_list;
570 INIT_LIST_HEAD(&done_list);
573 list_for_each_entry_safe(req, temp_req, &req_list, list) {
574 if (req->status == -ENODATA) {
575 src_in = (struct sockaddr *) &req->src_addr;
576 dst_in = (struct sockaddr *) &req->dst_addr;
577 req->status = addr_resolve(src_in, dst_in, req->addr,
579 if (req->status && time_after_eq(jiffies, req->timeout))
580 req->status = -ETIMEDOUT;
581 else if (req->status == -ENODATA)
584 list_move_tail(&req->list, &done_list);
587 if (!list_empty(&req_list)) {
588 req = list_entry(req_list.next, struct addr_req, list);
589 set_timeout(req->timeout);
593 list_for_each_entry_safe(req, temp_req, &done_list, list) {
594 list_del(&req->list);
595 req->callback(req->status, (struct sockaddr *) &req->src_addr,
596 req->addr, req->context);
597 put_client(req->client);
602 int rdma_resolve_ip(struct rdma_addr_client *client,
603 struct sockaddr *src_addr, struct sockaddr *dst_addr,
604 struct rdma_dev_addr *addr, int timeout_ms,
605 void (*callback)(int status, struct sockaddr *src_addr,
606 struct rdma_dev_addr *addr, void *context),
609 struct sockaddr *src_in, *dst_in;
610 struct addr_req *req;
613 req = kzalloc(sizeof *req, GFP_KERNEL);
617 src_in = (struct sockaddr *) &req->src_addr;
618 dst_in = (struct sockaddr *) &req->dst_addr;
621 if (src_addr->sa_family != dst_addr->sa_family) {
626 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
628 src_in->sa_family = dst_addr->sa_family;
631 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
633 req->callback = callback;
634 req->context = context;
635 req->client = client;
636 atomic_inc(&client->refcount);
637 req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
639 req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
640 switch (req->status) {
642 req->timeout = jiffies;
646 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
651 atomic_dec(&client->refcount);
659 EXPORT_SYMBOL(rdma_resolve_ip);
661 int rdma_resolve_ip_route(struct sockaddr *src_addr,
662 const struct sockaddr *dst_addr,
663 struct rdma_dev_addr *addr)
665 struct sockaddr_storage ssrc_addr = {};
666 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
669 if (src_addr->sa_family != dst_addr->sa_family)
672 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
674 src_in->sa_family = dst_addr->sa_family;
677 return addr_resolve(src_in, dst_addr, addr, false, 0);
679 EXPORT_SYMBOL(rdma_resolve_ip_route);
681 void rdma_addr_cancel(struct rdma_dev_addr *addr)
683 struct addr_req *req, *temp_req;
686 list_for_each_entry_safe(req, temp_req, &req_list, list) {
687 if (req->addr == addr) {
688 req->status = -ECANCELED;
689 req->timeout = jiffies;
690 list_move(&req->list, &req_list);
691 set_timeout(req->timeout);
697 EXPORT_SYMBOL(rdma_addr_cancel);
699 struct resolve_cb_context {
700 struct rdma_dev_addr *addr;
701 struct completion comp;
704 static void resolve_cb(int status, struct sockaddr *src_addr,
705 struct rdma_dev_addr *addr, void *context)
707 memcpy(((struct resolve_cb_context *)context)->addr, addr, sizeof(struct
709 complete(&((struct resolve_cb_context *)context)->comp);
712 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
713 const union ib_gid *dgid,
714 u8 *dmac, u16 *vlan_id, int *if_index,
718 struct rdma_dev_addr dev_addr;
719 struct resolve_cb_context ctx;
720 struct net_device *dev;
723 struct sockaddr _sockaddr;
724 struct sockaddr_in _sockaddr_in;
725 struct sockaddr_in6 _sockaddr_in6;
726 } sgid_addr, dgid_addr;
729 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
730 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
732 memset(&dev_addr, 0, sizeof(dev_addr));
734 dev_addr.bound_dev_if = *if_index;
735 dev_addr.net = &init_net;
737 ctx.addr = &dev_addr;
738 init_completion(&ctx.comp);
739 ret = rdma_resolve_ip(&self, &sgid_addr._sockaddr, &dgid_addr._sockaddr,
740 &dev_addr, 1000, resolve_cb, &ctx);
744 wait_for_completion(&ctx.comp);
746 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
747 dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
751 *if_index = dev_addr.bound_dev_if;
753 *vlan_id = rdma_vlan_dev_vlan_id(dev);
755 *hoplimit = dev_addr.hoplimit;
759 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
761 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
764 struct rdma_dev_addr dev_addr;
766 struct sockaddr _sockaddr;
767 struct sockaddr_in _sockaddr_in;
768 struct sockaddr_in6 _sockaddr_in6;
771 rdma_gid2ip(&gid_addr._sockaddr, sgid);
773 memset(&dev_addr, 0, sizeof(dev_addr));
774 dev_addr.net = &init_net;
775 ret = rdma_translate_ip(&gid_addr._sockaddr, &dev_addr, vlan_id);
779 memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
782 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
784 static int netevent_callback(struct notifier_block *self, unsigned long event,
787 if (event == NETEVENT_NEIGH_UPDATE) {
788 struct neighbour *neigh = ctx;
790 if (neigh->nud_state & NUD_VALID) {
791 set_timeout(jiffies);
797 static struct notifier_block nb = {
798 .notifier_call = netevent_callback
803 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
807 register_netevent_notifier(&nb);
808 rdma_addr_register_client(&self);
813 void addr_cleanup(void)
815 rdma_addr_unregister_client(&self);
816 unregister_netevent_notifier(&nb);
817 destroy_workqueue(addr_wq);