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;
60 void (*callback)(int status, struct sockaddr *src_addr,
61 struct rdma_dev_addr *addr, void *context);
62 unsigned long timeout;
63 struct delayed_work work;
68 static atomic_t ib_nl_addr_request_seq = ATOMIC_INIT(0);
70 static DEFINE_SPINLOCK(lock);
71 static LIST_HEAD(req_list);
72 static struct workqueue_struct *addr_wq;
74 static const struct nla_policy ib_nl_addr_policy[LS_NLA_TYPE_MAX] = {
75 [LS_NLA_TYPE_DGID] = {.type = NLA_BINARY,
76 .len = sizeof(struct rdma_nla_ls_gid)},
79 static inline bool ib_nl_is_good_ip_resp(const struct nlmsghdr *nlh)
81 struct nlattr *tb[LS_NLA_TYPE_MAX] = {};
84 if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
87 ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
88 nlmsg_len(nlh), ib_nl_addr_policy, NULL);
95 static void ib_nl_process_good_ip_rsep(const struct nlmsghdr *nlh)
97 const struct nlattr *head, *curr;
103 head = (const struct nlattr *)nlmsg_data(nlh);
104 len = nlmsg_len(nlh);
106 nla_for_each_attr(curr, head, len, rem) {
107 if (curr->nla_type == LS_NLA_TYPE_DGID)
108 memcpy(&gid, nla_data(curr), nla_len(curr));
112 list_for_each_entry(req, &req_list, list) {
113 if (nlh->nlmsg_seq != req->seq)
115 /* We set the DGID part, the rest was set earlier */
116 rdma_addr_set_dgid(req->addr, &gid);
121 spin_unlock_bh(&lock);
124 pr_info("Couldn't find request waiting for DGID: %pI6\n",
128 int ib_nl_handle_ip_res_resp(struct sk_buff *skb,
129 struct nlmsghdr *nlh,
130 struct netlink_ext_ack *extack)
132 if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
133 !(NETLINK_CB(skb).sk))
136 if (ib_nl_is_good_ip_resp(nlh))
137 ib_nl_process_good_ip_rsep(nlh);
142 static int ib_nl_ip_send_msg(struct rdma_dev_addr *dev_addr,
146 struct sk_buff *skb = NULL;
147 struct nlmsghdr *nlh;
148 struct rdma_ls_ip_resolve_header *header;
154 if (family == AF_INET) {
155 size = sizeof(struct in_addr);
156 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV4;
158 size = sizeof(struct in6_addr);
159 attrtype = RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_IPV6;
162 len = nla_total_size(sizeof(size));
163 len += NLMSG_ALIGN(sizeof(*header));
165 skb = nlmsg_new(len, GFP_KERNEL);
169 data = ibnl_put_msg(skb, &nlh, seq, 0, RDMA_NL_LS,
170 RDMA_NL_LS_OP_IP_RESOLVE, NLM_F_REQUEST);
176 /* Construct the family header first */
177 header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
178 header->ifindex = dev_addr->bound_dev_if;
179 nla_put(skb, attrtype, size, daddr);
181 /* Repair the nlmsg header length */
183 rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, GFP_KERNEL);
185 /* Make the request retry, so when we get the response from userspace
186 * we will have something.
191 int rdma_addr_size(const struct sockaddr *addr)
193 switch (addr->sa_family) {
195 return sizeof(struct sockaddr_in);
197 return sizeof(struct sockaddr_in6);
199 return sizeof(struct sockaddr_ib);
204 EXPORT_SYMBOL(rdma_addr_size);
206 int rdma_addr_size_in6(struct sockaddr_in6 *addr)
208 int ret = rdma_addr_size((struct sockaddr *) addr);
210 return ret <= sizeof(*addr) ? ret : 0;
212 EXPORT_SYMBOL(rdma_addr_size_in6);
214 int rdma_addr_size_kss(struct __kernel_sockaddr_storage *addr)
216 int ret = rdma_addr_size((struct sockaddr *) addr);
218 return ret <= sizeof(*addr) ? ret : 0;
220 EXPORT_SYMBOL(rdma_addr_size_kss);
222 void rdma_copy_addr(struct rdma_dev_addr *dev_addr,
223 const struct net_device *dev,
224 const unsigned char *dst_dev_addr)
226 dev_addr->dev_type = dev->type;
227 memcpy(dev_addr->src_dev_addr, dev->dev_addr, MAX_ADDR_LEN);
228 memcpy(dev_addr->broadcast, dev->broadcast, MAX_ADDR_LEN);
230 memcpy(dev_addr->dst_dev_addr, dst_dev_addr, MAX_ADDR_LEN);
231 dev_addr->bound_dev_if = dev->ifindex;
233 EXPORT_SYMBOL(rdma_copy_addr);
235 int rdma_translate_ip(const struct sockaddr *addr,
236 struct rdma_dev_addr *dev_addr)
238 struct net_device *dev;
240 if (dev_addr->bound_dev_if) {
241 dev = dev_get_by_index(dev_addr->net, dev_addr->bound_dev_if);
244 rdma_copy_addr(dev_addr, dev, NULL);
249 switch (addr->sa_family) {
251 dev = ip_dev_find(dev_addr->net,
252 ((const struct sockaddr_in *)addr)->sin_addr.s_addr);
255 return -EADDRNOTAVAIL;
257 rdma_copy_addr(dev_addr, dev, NULL);
260 #if IS_ENABLED(CONFIG_IPV6)
263 for_each_netdev_rcu(dev_addr->net, dev) {
264 if (ipv6_chk_addr(dev_addr->net,
265 &((const struct sockaddr_in6 *)addr)->sin6_addr,
267 rdma_copy_addr(dev_addr, dev, NULL);
277 EXPORT_SYMBOL(rdma_translate_ip);
279 static void set_timeout(struct addr_req *req, unsigned long time)
283 delay = time - jiffies;
287 mod_delayed_work(addr_wq, &req->work, delay);
290 static void queue_req(struct addr_req *req)
293 list_add_tail(&req->list, &req_list);
294 set_timeout(req, req->timeout);
295 spin_unlock_bh(&lock);
298 static int ib_nl_fetch_ha(const struct dst_entry *dst,
299 struct rdma_dev_addr *dev_addr,
300 const void *daddr, u32 seq, u16 family)
302 if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS))
303 return -EADDRNOTAVAIL;
305 /* We fill in what we can, the response will fill the rest */
306 rdma_copy_addr(dev_addr, dst->dev, NULL);
307 return ib_nl_ip_send_msg(dev_addr, daddr, seq, family);
310 static int dst_fetch_ha(const struct dst_entry *dst,
311 struct rdma_dev_addr *dev_addr,
317 n = dst_neigh_lookup(dst, daddr);
321 if (!(n->nud_state & NUD_VALID)) {
322 neigh_event_send(n, NULL);
325 rdma_copy_addr(dev_addr, dst->dev, n->ha);
333 static bool has_gateway(const struct dst_entry *dst, sa_family_t family)
336 struct rt6_info *rt6;
338 if (family == AF_INET) {
339 rt = container_of(dst, struct rtable, dst);
340 return rt->rt_uses_gateway;
343 rt6 = container_of(dst, struct rt6_info, dst);
344 return rt6->rt6i_flags & RTF_GATEWAY;
347 static int fetch_ha(const struct dst_entry *dst, struct rdma_dev_addr *dev_addr,
348 const struct sockaddr *dst_in, u32 seq)
350 const struct sockaddr_in *dst_in4 =
351 (const struct sockaddr_in *)dst_in;
352 const struct sockaddr_in6 *dst_in6 =
353 (const struct sockaddr_in6 *)dst_in;
354 const void *daddr = (dst_in->sa_family == AF_INET) ?
355 (const void *)&dst_in4->sin_addr.s_addr :
356 (const void *)&dst_in6->sin6_addr;
357 sa_family_t family = dst_in->sa_family;
359 /* Gateway + ARPHRD_INFINIBAND -> IB router */
360 if (has_gateway(dst, family) && dst->dev->type == ARPHRD_INFINIBAND)
361 return ib_nl_fetch_ha(dst, dev_addr, daddr, seq, family);
363 return dst_fetch_ha(dst, dev_addr, daddr);
366 static int addr4_resolve(struct sockaddr_in *src_in,
367 const struct sockaddr_in *dst_in,
368 struct rdma_dev_addr *addr,
371 __be32 src_ip = src_in->sin_addr.s_addr;
372 __be32 dst_ip = dst_in->sin_addr.s_addr;
377 memset(&fl4, 0, sizeof(fl4));
380 fl4.flowi4_oif = addr->bound_dev_if;
381 rt = ip_route_output_key(addr->net, &fl4);
382 ret = PTR_ERR_OR_ZERO(rt);
386 src_in->sin_family = AF_INET;
387 src_in->sin_addr.s_addr = fl4.saddr;
389 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
390 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
393 if (rt->rt_uses_gateway && rt->dst.dev->type != ARPHRD_INFINIBAND)
394 addr->network = RDMA_NETWORK_IPV4;
396 addr->hoplimit = ip4_dst_hoplimit(&rt->dst);
402 #if IS_ENABLED(CONFIG_IPV6)
403 static int addr6_resolve(struct sockaddr_in6 *src_in,
404 const struct sockaddr_in6 *dst_in,
405 struct rdma_dev_addr *addr,
406 struct dst_entry **pdst)
409 struct dst_entry *dst;
413 memset(&fl6, 0, sizeof fl6);
414 fl6.daddr = dst_in->sin6_addr;
415 fl6.saddr = src_in->sin6_addr;
416 fl6.flowi6_oif = addr->bound_dev_if;
418 ret = ipv6_stub->ipv6_dst_lookup(addr->net, NULL, &dst, &fl6);
422 rt = (struct rt6_info *)dst;
423 if (ipv6_addr_any(&src_in->sin6_addr)) {
424 src_in->sin6_family = AF_INET6;
425 src_in->sin6_addr = fl6.saddr;
428 /* If there's a gateway and type of device not ARPHRD_INFINIBAND, we're
429 * definitely in RoCE v2 (as RoCE v1 isn't routable) set the network
432 if (rt->rt6i_flags & RTF_GATEWAY &&
433 ip6_dst_idev(dst)->dev->type != ARPHRD_INFINIBAND)
434 addr->network = RDMA_NETWORK_IPV6;
436 addr->hoplimit = ip6_dst_hoplimit(dst);
442 static int addr6_resolve(struct sockaddr_in6 *src_in,
443 const struct sockaddr_in6 *dst_in,
444 struct rdma_dev_addr *addr,
445 struct dst_entry **pdst)
447 return -EADDRNOTAVAIL;
451 static int addr_resolve_neigh(const struct dst_entry *dst,
452 const struct sockaddr *dst_in,
453 struct rdma_dev_addr *addr,
456 if (dst->dev->flags & IFF_LOOPBACK) {
459 ret = rdma_translate_ip(dst_in, addr);
461 memcpy(addr->dst_dev_addr, addr->src_dev_addr,
467 /* If the device doesn't do ARP internally */
468 if (!(dst->dev->flags & IFF_NOARP))
469 return fetch_ha(dst, addr, dst_in, seq);
471 rdma_copy_addr(addr, dst->dev, NULL);
476 static int addr_resolve(struct sockaddr *src_in,
477 const struct sockaddr *dst_in,
478 struct rdma_dev_addr *addr,
482 struct net_device *ndev;
483 struct dst_entry *dst;
487 pr_warn_ratelimited("%s: missing namespace\n", __func__);
491 if (src_in->sa_family == AF_INET) {
492 struct rtable *rt = NULL;
493 const struct sockaddr_in *dst_in4 =
494 (const struct sockaddr_in *)dst_in;
496 ret = addr4_resolve((struct sockaddr_in *)src_in,
502 ret = addr_resolve_neigh(&rt->dst, dst_in, addr, seq);
504 if (addr->bound_dev_if) {
505 ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
513 const struct sockaddr_in6 *dst_in6 =
514 (const struct sockaddr_in6 *)dst_in;
516 ret = addr6_resolve((struct sockaddr_in6 *)src_in,
523 ret = addr_resolve_neigh(dst, dst_in, addr, seq);
525 if (addr->bound_dev_if) {
526 ndev = dev_get_by_index(addr->net, addr->bound_dev_if);
536 if (ndev->flags & IFF_LOOPBACK)
537 ret = rdma_translate_ip(dst_in, addr);
539 addr->bound_dev_if = ndev->ifindex;
546 static void process_one_req(struct work_struct *_work)
548 struct addr_req *req;
549 struct sockaddr *src_in, *dst_in;
551 req = container_of(_work, struct addr_req, work.work);
553 if (req->status == -ENODATA) {
554 src_in = (struct sockaddr *)&req->src_addr;
555 dst_in = (struct sockaddr *)&req->dst_addr;
556 req->status = addr_resolve(src_in, dst_in, req->addr,
558 if (req->status && time_after_eq(jiffies, req->timeout)) {
559 req->status = -ETIMEDOUT;
560 } else if (req->status == -ENODATA) {
561 /* requeue the work for retrying again */
563 if (!list_empty(&req->list))
564 set_timeout(req, req->timeout);
565 spin_unlock_bh(&lock);
570 req->callback(req->status, (struct sockaddr *)&req->src_addr,
571 req->addr, req->context);
572 req->callback = NULL;
575 if (!list_empty(&req->list)) {
577 * Although the work will normally have been canceled by the
578 * workqueue, it can still be requeued as long as it is on the
581 cancel_delayed_work(&req->work);
582 list_del_init(&req->list);
585 spin_unlock_bh(&lock);
588 int rdma_resolve_ip(struct sockaddr *src_addr, const struct sockaddr *dst_addr,
589 struct rdma_dev_addr *addr, int timeout_ms,
590 void (*callback)(int status, struct sockaddr *src_addr,
591 struct rdma_dev_addr *addr, void *context),
594 struct sockaddr *src_in, *dst_in;
595 struct addr_req *req;
598 req = kzalloc(sizeof *req, GFP_KERNEL);
602 src_in = (struct sockaddr *) &req->src_addr;
603 dst_in = (struct sockaddr *) &req->dst_addr;
606 if (src_addr->sa_family != dst_addr->sa_family) {
611 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
613 src_in->sa_family = dst_addr->sa_family;
616 memcpy(dst_in, dst_addr, rdma_addr_size(dst_addr));
618 req->callback = callback;
619 req->context = context;
620 INIT_DELAYED_WORK(&req->work, process_one_req);
621 req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
623 req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
624 switch (req->status) {
626 req->timeout = jiffies;
630 req->timeout = msecs_to_jiffies(timeout_ms) + jiffies;
642 EXPORT_SYMBOL(rdma_resolve_ip);
644 int rdma_resolve_ip_route(struct sockaddr *src_addr,
645 const struct sockaddr *dst_addr,
646 struct rdma_dev_addr *addr)
648 struct sockaddr_storage ssrc_addr = {};
649 struct sockaddr *src_in = (struct sockaddr *)&ssrc_addr;
652 if (src_addr->sa_family != dst_addr->sa_family)
655 memcpy(src_in, src_addr, rdma_addr_size(src_addr));
657 src_in->sa_family = dst_addr->sa_family;
660 return addr_resolve(src_in, dst_addr, addr, false, 0);
663 void rdma_addr_cancel(struct rdma_dev_addr *addr)
665 struct addr_req *req, *temp_req;
666 struct addr_req *found = NULL;
669 list_for_each_entry_safe(req, temp_req, &req_list, list) {
670 if (req->addr == addr) {
672 * Removing from the list means we take ownership of
675 list_del_init(&req->list);
680 spin_unlock_bh(&lock);
686 * sync canceling the work after removing it from the req_list
687 * guarentees no work is running and none will be started.
689 cancel_delayed_work_sync(&found->work);
692 found->callback(-ECANCELED, (struct sockaddr *)&found->src_addr,
693 found->addr, found->context);
697 EXPORT_SYMBOL(rdma_addr_cancel);
699 struct resolve_cb_context {
700 struct completion comp;
704 static void resolve_cb(int status, struct sockaddr *src_addr,
705 struct rdma_dev_addr *addr, void *context)
707 ((struct resolve_cb_context *)context)->status = status;
708 complete(&((struct resolve_cb_context *)context)->comp);
711 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
712 const union ib_gid *dgid,
713 u8 *dmac, const struct net_device *ndev,
716 struct rdma_dev_addr dev_addr;
717 struct resolve_cb_context ctx;
719 struct sockaddr _sockaddr;
720 struct sockaddr_in _sockaddr_in;
721 struct sockaddr_in6 _sockaddr_in6;
722 } sgid_addr, dgid_addr;
725 rdma_gid2ip(&sgid_addr._sockaddr, sgid);
726 rdma_gid2ip(&dgid_addr._sockaddr, dgid);
728 memset(&dev_addr, 0, sizeof(dev_addr));
729 dev_addr.bound_dev_if = ndev->ifindex;
730 dev_addr.net = &init_net;
732 init_completion(&ctx.comp);
733 ret = rdma_resolve_ip(&sgid_addr._sockaddr, &dgid_addr._sockaddr,
734 &dev_addr, 1000, resolve_cb, &ctx);
738 wait_for_completion(&ctx.comp);
744 memcpy(dmac, dev_addr.dst_dev_addr, ETH_ALEN);
745 *hoplimit = dev_addr.hoplimit;
749 static int netevent_callback(struct notifier_block *self, unsigned long event,
752 struct addr_req *req;
754 if (event == NETEVENT_NEIGH_UPDATE) {
755 struct neighbour *neigh = ctx;
757 if (neigh->nud_state & NUD_VALID) {
759 list_for_each_entry(req, &req_list, list)
760 set_timeout(req, jiffies);
761 spin_unlock_bh(&lock);
767 static struct notifier_block nb = {
768 .notifier_call = netevent_callback
773 addr_wq = alloc_ordered_workqueue("ib_addr", 0);
777 register_netevent_notifier(&nb);
782 void addr_cleanup(void)
784 unregister_netevent_notifier(&nb);
785 destroy_workqueue(addr_wq);
786 WARN_ON(!list_empty(&req_list));