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 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
214 int ret = rdma_addr_size((struct sockaddr *) addr);
216 return ret <= sizeof(*addr) ? ret : 0;
218 EXPORT_SYMBOL(rdma_addr_size_in6);
220 int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
222 int ret = rdma_addr_size((struct sockaddr *) addr);
224 return ret <= sizeof(*addr) ? ret : 0;
226 EXPORT_SYMBOL(rdma_addr_size_kss);
228 static struct rdma_addr_client self;
230 void rdma_addr_register_client(struct rdma_addr_client *client)
232 atomic_set(&client->refcount, 1);
233 init_completion(&client->comp);
235 EXPORT_SYMBOL(rdma_addr_register_client);
237 static inline void put_client(struct rdma_addr_client *client)
239 if (atomic_dec_and_test(&client->refcount))
240 complete(&client->comp);
243 void rdma_addr_unregister_client(struct rdma_addr_client *client)
246 wait_for_completion(&client->comp);
248 EXPORT_SYMBOL(rdma_addr_unregister_client);
250 int rdma_copy_addr(struct rdma_dev_addr *dev_addr, struct net_device *dev,
251 const unsigned char *dst_dev_addr)
253 dev_addr->dev_type = dev->type;
254 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
255 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
257 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
258 dev_addr->bound_dev_if = dev->ifindex;
261 EXPORT_SYMBOL(rdma_copy_addr);
263 int rdma_translate_ip(const struct sockaddr *addr,
264 struct rdma_dev_addr *dev_addr,
267 struct net_device *dev;
268 int ret = -EADDRNOTAVAIL;
270 if (dev_addr->bound_dev_if) {
271 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
274 ret = rdma_copy_addr(dev_addr, dev, NULL);
279 switch (addr->sa_family) {
281 dev = ip_dev_find(dev_addr->net,
282 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
287 ret = rdma_copy_addr(dev_addr, dev, NULL);
289 *vlan_id = rdma_vlan_dev_vlan_id(dev);
292 #if IS_ENABLED(CONFIG_IPV6)
295 for_each_netdev_rcu(dev_addr->net, dev) {
296 if (ipv6_chk_addr(dev_addr->net,
297 &((const struct sockaddr_in6 *)addr)->sin6_addr,
299 ret = rdma_copy_addr(dev_addr, dev, NULL);
301 *vlan_id = rdma_vlan_dev_vlan_id(dev);
311 EXPORT_SYMBOL(rdma_translate_ip);
313 static void set_timeout(unsigned long time)
317 delay = time - jiffies;
321 mod_delayed_work(addr_wq, &work, delay);
324 static void queue_req(struct addr_req *req)
326 struct addr_req *temp_req;
329 list_for_each_entry_reverse(temp_req, &req_list, list) {
330 if (time_after_eq(req->timeout, temp_req->timeout))
334 list_add(&req->list, &temp_req->list);
336 if (req_list.next == &req->list)
337 set_timeout(req->timeout);
341 static int ib_nl_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
342 const void *daddr, u32 seq, u16 family)
344 if (ibnl_chk_listeners(RDMA_NL_GROUP_LS))
345 return -EADDRNOTAVAIL;
347 /* We fill in what we can, the response will fill the rest */
348 rdma_copy_addr(dev_addr, dst->dev, NULL);
349 return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
352 static int dst_fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
358 n = dst_neigh_lookup(dst, daddr);
361 if (!n || !(n->nud_state & NUD_VALID)) {
363 neigh_event_send(n, NULL);
366 ret = rdma_copy_addr(dev_addr, dst->dev, n->ha);
376 static bool has_gateway(struct dst_entry *dst, sa_family_t family)
379 struct rt6_info *rt6;
381 if (family == AF_INET) {
382 rt = container_of(dst, struct rtable, dst);
383 return rt->rt_uses_gateway;
386 rt6 = container_of(dst, struct rt6_info, dst);
387 return rt6->rt6i_flags & RTF_GATEWAY;
390 static int fetch_ha(struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
391 const struct sockaddr *dst_in, u32 seq)
393 const struct sockaddr_in *dst_in4 =
394 (const struct sockaddr_in *)dst_in;
395 const struct sockaddr_in6 *dst_in6 =
396 (const struct sockaddr_in6 *)dst_in;
397 const void *daddr = (dst_in->sa_family == AF_INET) ?
398 (const void *)&dst_in4->sin_addr.s_addr :
399 (const void *)&dst_in6->sin6_addr;
400 sa_family_t family = dst_in->sa_family;
402 /* Gateway + ARPHRD_INFINIBAND -> IB router */
403 if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
404 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
406 return dst_fetch_ha(dst, dev_addr, daddr);
409 static int addr4_resolve(struct sockaddr_in *src_in,
410 const struct sockaddr_in *dst_in,
411 struct rdma_dev_addr *addr,
414 __be32 src_ip = src_in->sin_addr.s_addr;
415 __be32 dst_ip = dst_in->sin_addr.s_addr;
420 memset(&fl4, 0, sizeof(fl4));
423 fl4.flowi4_oif = addr->bound_dev_if;
424 rt = ip_route_output_key(addr->net, &fl4);
429 src_in->sin_family = AF_INET;
430 src_in->sin_addr.s_addr = fl4.saddr;
432 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
433 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
436 if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
437 addr->network = RDMA_NETWORK_IPV4;
439 addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
447 #if IS_ENABLED(CONFIG_IPV6)
448 static int addr6_resolve(struct sockaddr_in6 *src_in,
449 const struct sockaddr_in6 *dst_in,
450 struct rdma_dev_addr *addr,
451 struct dst_entry **pdst)
454 struct dst_entry *dst;
457 memset(&fl6, 0, sizeof fl6);
458 fl6.daddr = dst_in->sin6_addr;
459 fl6.saddr = src_in->sin6_addr;
460 fl6.flowi6_oif = addr->bound_dev_if;
462 dst = ipv6_stub->ipv6_dst_lookup_flow(addr->net, NULL, &fl6, NULL);
466 rt = (struct rt6_info *)dst;
467 if (ipv6_addr_any(&src_in->sin6_addr)) {
468 src_in->sin6_family = AF_INET6;
469 src_in->sin6_addr = fl6.saddr;
472 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
473 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
476 if (rt->rt6i_flags & RTF_GATEWAY &&
477 ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
478 addr->network = RDMA_NETWORK_IPV6;
480 addr->hoplimit = ip6_dst_hoplimit(dst);
486 static int addr6_resolve(struct sockaddr_in6 *src_in,
487 const struct sockaddr_in6 *dst_in,
488 struct rdma_dev_addr *addr,
489 struct dst_entry **pdst)
491 return -EADDRNOTAVAIL;
495 static int addr_resolve_neigh(struct dst_entry *dst,
496 const struct sockaddr *dst_in,
497 struct rdma_dev_addr *addr,
500 if (dst->dev->flags & IFF_LOOPBACK) {
503 ret = rdma_translate_ip(dst_in, addr, NULL);
505 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
511 /* If the device doesn't do ARP internally */
512 if (!(dst->dev->flags & IFF_NOARP))
513 return fetch_ha(dst, addr, dst_in, seq);
515 return rdma_copy_addr(addr, dst->dev, NULL);
518 static int addr_resolve(struct sockaddr *src_in,
519 const struct sockaddr *dst_in,
520 struct rdma_dev_addr *addr,
524 struct net_device *ndev;
525 struct dst_entry *dst;
529 pr_warn_ratelimited("%s: missing namespace\n", __func__);
533 if (src_in->sa_family == AF_INET) {
534 struct rtable *rt = NULL;
535 const struct sockaddr_in *dst_in4 =
536 (const struct sockaddr_in *)dst_in;
538 ret = addr4_resolve((struct sockaddr_in *)src_in,
544 ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
551 const struct sockaddr_in6 *dst_in6 =
552 (const struct sockaddr_in6 *)dst_in;
555 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
562 ret = addr_resolve_neigh(dst, dst_in, addr, seq);
570 addr->bound_dev_if = ndev->ifindex;
576 static void process_req(struct work_struct *work)
578 struct addr_req *req, *temp_req;
579 struct sockaddr *src_in, *dst_in;
580 struct list_head done_list;
582 INIT_LIST_HEAD(&done_list);
585 list_for_each_entry_safe(req, temp_req, &req_list, list) {
586 if (req->status == -ENODATA) {
587 src_in = (struct sockaddr *) &req->src_addr;
588 dst_in = (struct sockaddr *) &req->dst_addr;
589 req->status = addr_resolve(src_in, dst_in, req->addr,
591 if (req->status && time_after_eq(jiffies, req->timeout))
592 req->status = -ETIMEDOUT;
593 else if (req->status == -ENODATA)
596 list_move_tail(&req->list, &done_list);
599 if (!list_empty(&req_list)) {
600 req = list_entry(req_list.next, struct addr_req, list);
601 set_timeout(req->timeout);
605 list_for_each_entry_safe(req, temp_req, &done_list, list) {
606 list_del(&req->list);
607 req->callback(req->status, (struct sockaddr *) &req->src_addr,
608 req->addr, req->context);
609 put_client(req->client);
614 int rdma_resolve_ip(struct rdma_addr_client *client,
615 struct sockaddr *src_addr, struct sockaddr *dst_addr,
616 struct rdma_dev_addr *addr, int timeout_ms,
617 void (*callback)(int status, struct sockaddr *src_addr,
618 struct rdma_dev_addr *addr, void *context),
621 struct sockaddr *src_in, *dst_in;
622 struct addr_req *req;
625 req = kzalloc(sizeof *req, GFP_KERNEL);
629 src_in = (struct sockaddr *) &req->src_addr;
630 dst_in = (struct sockaddr *) &req->dst_addr;
633 if (src_addr->sa_family != dst_addr->sa_family) {
638 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
640 src_in->sa_family = dst_addr->sa_family;
643 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
645 req->callback = callback;
646 req->context = context;
647 req->client = client;
648 atomic_inc(&client->refcount);
649 req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
651 req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
652 switch (req->status) {
654 req->timeout = jiffies;
658 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
663 atomic_dec(&client->refcount);
671 EXPORT_SYMBOL(rdma_resolve_ip);
673 int rdma_resolve_ip_route(struct sockaddr *src_addr,
674 const struct sockaddr *dst_addr,
675 struct rdma_dev_addr *addr)
677 struct sockaddr_storage ssrc_addr = {};
678 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
681 if (src_addr->sa_family != dst_addr->sa_family)
684 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
686 src_in->sa_family = dst_addr->sa_family;
689 return addr_resolve(src_in, dst_addr, addr, false, 0);
691 EXPORT_SYMBOL(rdma_resolve_ip_route);
693 void rdma_addr_cancel(struct rdma_dev_addr *addr)
695 struct addr_req *req, *temp_req;
698 list_for_each_entry_safe(req, temp_req, &req_list, list) {
699 if (req->addr == addr) {
700 req->status = -ECANCELED;
701 req->timeout = jiffies;
702 list_move(&req->list, &req_list);
703 set_timeout(req->timeout);
709 EXPORT_SYMBOL(rdma_addr_cancel);
711 struct resolve_cb_context {
712 struct rdma_dev_addr *addr;
713 struct completion comp;
717 static void resolve_cb(int status, struct sockaddr *src_addr,
718 struct rdma_dev_addr *addr, void *context)
721 memcpy(((struct resolve_cb_context *)context)->addr,
722 addr, sizeof(struct rdma_dev_addr));
723 ((struct resolve_cb_context *)context)->status = status;
724 complete(&((struct resolve_cb_context *)context)->comp);
727 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
728 const union ib_gid *dgid,
729 u8 *dmac, u16 *vlan_id, int *if_index,
733 struct rdma_dev_addr dev_addr;
734 struct resolve_cb_context ctx;
735 struct net_device *dev;
738 struct sockaddr_in _sockaddr_in;
739 struct sockaddr_in6 _sockaddr_in6;
740 } sgid_addr, dgid_addr;
743 rdma_gid2ip((struct sockaddr *)&sgid_addr, sgid);
744 rdma_gid2ip((struct sockaddr *)&dgid_addr, dgid);
746 memset(&dev_addr, 0, sizeof(dev_addr));
748 dev_addr.bound_dev_if = *if_index;
749 dev_addr.net = &init_net;
751 ctx.addr = &dev_addr;
752 init_completion(&ctx.comp);
753 ret = rdma_resolve_ip(&self, (struct sockaddr *)&sgid_addr,
754 (struct sockaddr *)&dgid_addr, &dev_addr, 1000,
759 wait_for_completion(&ctx.comp);
765 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
766 dev = dev_get_by_index(&init_net, dev_addr.bound_dev_if);
770 *if_index = dev_addr.bound_dev_if;
772 *vlan_id = rdma_vlan_dev_vlan_id(dev);
774 *hoplimit = dev_addr.hoplimit;
778 EXPORT_SYMBOL(rdma_addr_find_l2_eth_by_grh);
780 int rdma_addr_find_smac_by_sgid(union ib_gid *sgid, u8 *smac, u16 *vlan_id)
783 struct rdma_dev_addr dev_addr;
785 struct sockaddr_in _sockaddr_in;
786 struct sockaddr_in6 _sockaddr_in6;
789 rdma_gid2ip((struct sockaddr *)&gid_addr, sgid);
791 memset(&dev_addr, 0, sizeof(dev_addr));
792 dev_addr.net = &init_net;
793 ret = rdma_translate_ip((struct sockaddr *)&gid_addr, &dev_addr, vlan_id);
797 memcpy(smac, dev_addr.src_dev_addr, ETH_ALEN);
800 EXPORT_SYMBOL(rdma_addr_find_smac_by_sgid);
802 static int netevent_callback(struct notifier_block *self, unsigned long event,
805 if (event == NETEVENT_NEIGH_UPDATE) {
806 struct neighbour *neigh = ctx;
808 if (neigh->nud_state & NUD_VALID) {
809 set_timeout(jiffies);
815 static struct notifier_block nb = {
816 .notifier_call = netevent_callback
821 addr_wq = alloc_workqueue("ib_addr", WQ_MEM_RECLAIM, 0);
825 register_netevent_notifier(&nb);
826 rdma_addr_register_client(&self);
831 void addr_cleanup(void)
833 rdma_addr_unregister_client(&self);
834 unregister_netevent_notifier(&nb);
835 destroy_workqueue(addr_wq);