1 // SPDX-License-Identifier: GPL-2.0
2 /* Generic nexthop implementation
4 * Copyright (c) 2017-19 Cumulus Networks
5 * Copyright (c) 2017-19 David Ahern <dsa@cumulusnetworks.com>
8 #include <linux/nexthop.h>
9 #include <linux/rtnetlink.h>
10 #include <linux/slab.h>
12 #include <net/ipv6_stubs.h>
13 #include <net/lwtunnel.h>
14 #include <net/ndisc.h>
15 #include <net/nexthop.h>
16 #include <net/route.h>
19 static void remove_nexthop(struct net *net, struct nexthop *nh,
20 struct nl_info *nlinfo);
22 #define NH_DEV_HASHBITS 8
23 #define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS)
25 static const struct nla_policy rtm_nh_policy[NHA_MAX + 1] = {
26 [NHA_ID] = { .type = NLA_U32 },
27 [NHA_GROUP] = { .type = NLA_BINARY },
28 [NHA_GROUP_TYPE] = { .type = NLA_U16 },
29 [NHA_BLACKHOLE] = { .type = NLA_FLAG },
30 [NHA_OIF] = { .type = NLA_U32 },
31 [NHA_GATEWAY] = { .type = NLA_BINARY },
32 [NHA_ENCAP_TYPE] = { .type = NLA_U16 },
33 [NHA_ENCAP] = { .type = NLA_NESTED },
34 [NHA_GROUPS] = { .type = NLA_FLAG },
35 [NHA_MASTER] = { .type = NLA_U32 },
38 static unsigned int nh_dev_hashfn(unsigned int val)
40 unsigned int mask = NH_DEV_HASHSIZE - 1;
43 (val >> NH_DEV_HASHBITS) ^
44 (val >> (NH_DEV_HASHBITS * 2))) & mask;
47 static void nexthop_devhash_add(struct net *net, struct nh_info *nhi)
49 struct net_device *dev = nhi->fib_nhc.nhc_dev;
50 struct hlist_head *head;
55 hash = nh_dev_hashfn(dev->ifindex);
56 head = &net->nexthop.devhash[hash];
57 hlist_add_head(&nhi->dev_hash, head);
60 static void nexthop_free_mpath(struct nexthop *nh)
65 nhg = rcu_dereference_raw(nh->nh_grp);
66 for (i = 0; i < nhg->num_nh; ++i)
67 WARN_ON(nhg->nh_entries[i].nh);
72 static void nexthop_free_single(struct nexthop *nh)
76 nhi = rcu_dereference_raw(nh->nh_info);
77 switch (nhi->family) {
79 fib_nh_release(nh->net, &nhi->fib_nh);
82 ipv6_stub->fib6_nh_release(&nhi->fib6_nh);
88 void nexthop_free_rcu(struct rcu_head *head)
90 struct nexthop *nh = container_of(head, struct nexthop, rcu);
93 nexthop_free_mpath(nh);
95 nexthop_free_single(nh);
99 EXPORT_SYMBOL_GPL(nexthop_free_rcu);
101 static struct nexthop *nexthop_alloc(void)
105 nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL);
107 INIT_LIST_HEAD(&nh->fi_list);
108 INIT_LIST_HEAD(&nh->f6i_list);
109 INIT_LIST_HEAD(&nh->grp_list);
114 static struct nh_group *nexthop_grp_alloc(u16 num_nh)
116 size_t sz = offsetof(struct nexthop, nh_grp)
117 + sizeof(struct nh_group)
118 + sizeof(struct nh_grp_entry) * num_nh;
119 struct nh_group *nhg;
121 nhg = kzalloc(sz, GFP_KERNEL);
123 nhg->num_nh = num_nh;
128 static void nh_base_seq_inc(struct net *net)
130 while (++net->nexthop.seq == 0)
134 /* no reference taken; rcu lock or rtnl must be held */
135 struct nexthop *nexthop_find_by_id(struct net *net, u32 id)
137 struct rb_node **pp, *parent = NULL, *next;
139 pp = &net->nexthop.rb_root.rb_node;
143 next = rcu_dereference_raw(*pp);
148 nh = rb_entry(parent, struct nexthop, rb_node);
151 else if (id > nh->id)
152 pp = &next->rb_right;
158 EXPORT_SYMBOL_GPL(nexthop_find_by_id);
160 /* used for auto id allocation; called with rtnl held */
161 static u32 nh_find_unused_id(struct net *net)
163 u32 id_start = net->nexthop.last_id_allocated;
166 net->nexthop.last_id_allocated++;
167 if (net->nexthop.last_id_allocated == id_start)
170 if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated))
171 return net->nexthop.last_id_allocated;
176 static int nla_put_nh_group(struct sk_buff *skb, struct nh_group *nhg)
178 struct nexthop_grp *p;
179 size_t len = nhg->num_nh * sizeof(*p);
185 group_type = NEXTHOP_GRP_TYPE_MPATH;
187 if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type))
188 goto nla_put_failure;
190 nla = nla_reserve(skb, NHA_GROUP, len);
192 goto nla_put_failure;
195 for (i = 0; i < nhg->num_nh; ++i) {
196 p->id = nhg->nh_entries[i].nh->id;
197 p->weight = nhg->nh_entries[i].weight - 1;
207 static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh,
208 int event, u32 portid, u32 seq, unsigned int nlflags)
210 struct fib6_nh *fib6_nh;
211 struct fib_nh *fib_nh;
212 struct nlmsghdr *nlh;
216 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags);
220 nhm = nlmsg_data(nlh);
221 nhm->nh_family = AF_UNSPEC;
222 nhm->nh_flags = nh->nh_flags;
223 nhm->nh_protocol = nh->protocol;
227 if (nla_put_u32(skb, NHA_ID, nh->id))
228 goto nla_put_failure;
231 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
233 if (nla_put_nh_group(skb, nhg))
234 goto nla_put_failure;
238 nhi = rtnl_dereference(nh->nh_info);
239 nhm->nh_family = nhi->family;
240 if (nhi->reject_nh) {
241 if (nla_put_flag(skb, NHA_BLACKHOLE))
242 goto nla_put_failure;
245 const struct net_device *dev;
247 dev = nhi->fib_nhc.nhc_dev;
248 if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex))
249 goto nla_put_failure;
252 nhm->nh_scope = nhi->fib_nhc.nhc_scope;
253 switch (nhi->family) {
255 fib_nh = &nhi->fib_nh;
256 if (fib_nh->fib_nh_gw_family &&
257 nla_put_u32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4))
258 goto nla_put_failure;
262 fib6_nh = &nhi->fib6_nh;
263 if (fib6_nh->fib_nh_gw_family &&
264 nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6))
265 goto nla_put_failure;
269 if (nhi->fib_nhc.nhc_lwtstate &&
270 lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate,
271 NHA_ENCAP, NHA_ENCAP_TYPE) < 0)
272 goto nla_put_failure;
282 static size_t nh_nlmsg_size_grp(struct nexthop *nh)
284 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
285 size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh;
287 return nla_total_size(sz) +
288 nla_total_size(2); /* NHA_GROUP_TYPE */
291 static size_t nh_nlmsg_size_single(struct nexthop *nh)
293 struct nh_info *nhi = rtnl_dereference(nh->nh_info);
296 /* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE
297 * are mutually exclusive
299 sz = nla_total_size(4); /* NHA_OIF */
301 switch (nhi->family) {
303 if (nhi->fib_nh.fib_nh_gw_family)
304 sz += nla_total_size(4); /* NHA_GATEWAY */
309 if (nhi->fib6_nh.fib_nh_gw_family)
310 sz += nla_total_size(sizeof(const struct in6_addr));
314 if (nhi->fib_nhc.nhc_lwtstate) {
315 sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate);
316 sz += nla_total_size(2); /* NHA_ENCAP_TYPE */
322 static size_t nh_nlmsg_size(struct nexthop *nh)
324 size_t sz = nla_total_size(4); /* NHA_ID */
327 sz += nh_nlmsg_size_grp(nh);
329 sz += nh_nlmsg_size_single(nh);
334 static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info)
336 unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0;
337 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
341 skb = nlmsg_new(nh_nlmsg_size(nh), gfp_any());
345 err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags);
347 /* -EMSGSIZE implies BUG in nh_nlmsg_size() */
348 WARN_ON(err == -EMSGSIZE);
353 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP,
354 info->nlh, gfp_any());
358 rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err);
361 static bool valid_group_nh(struct nexthop *nh, unsigned int npaths,
362 struct netlink_ext_ack *extack)
365 struct nh_group *nhg = rtnl_dereference(nh->nh_grp);
367 /* nested multipath (group within a group) is not
371 NL_SET_ERR_MSG(extack,
372 "Multipath group can not be a nexthop within a group");
376 struct nh_info *nhi = rtnl_dereference(nh->nh_info);
378 if (nhi->reject_nh && npaths > 1) {
379 NL_SET_ERR_MSG(extack,
380 "Blackhole nexthop can not be used in a group with more than 1 path");
388 static int nh_check_attr_group(struct net *net, struct nlattr *tb[],
389 struct netlink_ext_ack *extack)
391 unsigned int len = nla_len(tb[NHA_GROUP]);
392 struct nexthop_grp *nhg;
395 if (len & (sizeof(struct nexthop_grp) - 1)) {
396 NL_SET_ERR_MSG(extack,
397 "Invalid length for nexthop group attribute");
401 /* convert len to number of nexthop ids */
404 nhg = nla_data(tb[NHA_GROUP]);
405 for (i = 0; i < len; ++i) {
406 if (nhg[i].resvd1 || nhg[i].resvd2) {
407 NL_SET_ERR_MSG(extack, "Reserved fields in nexthop_grp must be 0");
410 if (nhg[i].weight > 254) {
411 NL_SET_ERR_MSG(extack, "Invalid value for weight");
414 for (j = i + 1; j < len; ++j) {
415 if (nhg[i].id == nhg[j].id) {
416 NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group");
422 nhg = nla_data(tb[NHA_GROUP]);
423 for (i = 0; i < len; ++i) {
426 nh = nexthop_find_by_id(net, nhg[i].id);
428 NL_SET_ERR_MSG(extack, "Invalid nexthop id");
431 if (!valid_group_nh(nh, len, extack))
434 for (i = NHA_GROUP + 1; i < __NHA_MAX; ++i) {
438 NL_SET_ERR_MSG(extack,
439 "No other attributes can be set in nexthop groups");
446 static bool ipv6_good_nh(const struct fib6_nh *nh)
448 int state = NUD_REACHABLE;
453 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6);
455 state = n->nud_state;
457 rcu_read_unlock_bh();
459 return !!(state & NUD_VALID);
462 static bool ipv4_good_nh(const struct fib_nh *nh)
464 int state = NUD_REACHABLE;
469 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
470 (__force u32)nh->fib_nh_gw4);
472 state = n->nud_state;
474 rcu_read_unlock_bh();
476 return !!(state & NUD_VALID);
479 struct nexthop *nexthop_select_path(struct nexthop *nh, int hash)
481 struct nexthop *rc = NULL;
482 struct nh_group *nhg;
488 nhg = rcu_dereference(nh->nh_grp);
489 for (i = 0; i < nhg->num_nh; ++i) {
490 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
493 if (hash > atomic_read(&nhge->upper_bound))
496 /* nexthops always check if it is good and does
497 * not rely on a sysctl for this behavior
499 nhi = rcu_dereference(nhge->nh->nh_info);
500 switch (nhi->family) {
502 if (ipv4_good_nh(&nhi->fib_nh))
506 if (ipv6_good_nh(&nhi->fib6_nh))
517 EXPORT_SYMBOL_GPL(nexthop_select_path);
519 int nexthop_for_each_fib6_nh(struct nexthop *nh,
520 int (*cb)(struct fib6_nh *nh, void *arg),
527 struct nh_group *nhg;
530 nhg = rcu_dereference_rtnl(nh->nh_grp);
531 for (i = 0; i < nhg->num_nh; i++) {
532 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
534 nhi = rcu_dereference_rtnl(nhge->nh->nh_info);
535 err = cb(&nhi->fib6_nh, arg);
540 nhi = rcu_dereference_rtnl(nh->nh_info);
541 err = cb(&nhi->fib6_nh, arg);
548 EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh);
550 static int check_src_addr(const struct in6_addr *saddr,
551 struct netlink_ext_ack *extack)
553 if (!ipv6_addr_any(saddr)) {
554 NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects");
560 int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg,
561 struct netlink_ext_ack *extack)
565 /* fib6_src is unique to a fib6_info and limits the ability to cache
566 * routes in fib6_nh within a nexthop that is potentially shared
567 * across multiple fib entries. If the config wants to use source
568 * routing it can not use nexthop objects. mlxsw also does not allow
569 * fib6_src on routes.
571 if (cfg && check_src_addr(&cfg->fc_src, extack) < 0)
575 struct nh_group *nhg;
577 nhg = rtnl_dereference(nh->nh_grp);
581 nhi = rtnl_dereference(nh->nh_info);
582 if (nhi->family == AF_INET)
588 NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop");
591 EXPORT_SYMBOL_GPL(fib6_check_nexthop);
593 /* if existing nexthop has ipv6 routes linked to it, need
594 * to verify this new spec works with ipv6
596 static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new,
597 struct netlink_ext_ack *extack)
599 struct fib6_info *f6i;
601 if (list_empty(&old->f6i_list))
604 list_for_each_entry(f6i, &old->f6i_list, nh_list) {
605 if (check_src_addr(&f6i->fib6_src.addr, extack) < 0)
609 return fib6_check_nexthop(new, NULL, extack);
612 static int nexthop_check_scope(struct nexthop *nh, u8 scope,
613 struct netlink_ext_ack *extack)
617 nhi = rtnl_dereference(nh->nh_info);
618 if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) {
619 NL_SET_ERR_MSG(extack,
620 "Route with host scope can not have a gateway");
624 if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) {
625 NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop");
632 /* Invoked by fib add code to verify nexthop by id is ok with
633 * config for prefix; parts of fib_check_nh not done when nexthop
636 int fib_check_nexthop(struct nexthop *nh, u8 scope,
637 struct netlink_ext_ack *extack)
642 struct nh_group *nhg;
644 if (scope == RT_SCOPE_HOST) {
645 NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops");
650 nhg = rtnl_dereference(nh->nh_grp);
651 /* all nexthops in a group have the same scope */
652 err = nexthop_check_scope(nhg->nh_entries[0].nh, scope, extack);
654 err = nexthop_check_scope(nh, scope, extack);
660 static int fib_check_nh_list(struct nexthop *old, struct nexthop *new,
661 struct netlink_ext_ack *extack)
665 list_for_each_entry(fi, &old->fi_list, nh_list) {
668 err = fib_check_nexthop(new, fi->fib_scope, extack);
675 static void nh_group_rebalance(struct nh_group *nhg)
681 for (i = 0; i < nhg->num_nh; ++i)
682 total += nhg->nh_entries[i].weight;
684 for (i = 0; i < nhg->num_nh; ++i) {
685 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
689 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1;
690 atomic_set(&nhge->upper_bound, upper_bound);
694 static void remove_nh_grp_entry(struct nh_grp_entry *nhge,
695 struct nh_group *nhg,
696 struct nl_info *nlinfo)
698 struct nexthop *nh = nhge->nh;
699 struct nh_grp_entry *nhges;
705 nhges = nhg->nh_entries;
706 for (i = 0; i < nhg->num_nh; ++i) {
708 nhges[i-1].nh = nhges[i].nh;
709 nhges[i-1].weight = nhges[i].weight;
710 list_del(&nhges[i].nh_list);
711 list_add(&nhges[i-1].nh_list, &nhges[i-1].nh->grp_list);
712 } else if (nhg->nh_entries[i].nh == nh) {
721 nhg->nh_entries[nhg->num_nh].nh = NULL;
723 nh_group_rebalance(nhg);
728 nexthop_notify(RTM_NEWNEXTHOP, nhge->nh_parent, nlinfo);
731 static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh,
732 struct nl_info *nlinfo)
734 struct nh_grp_entry *nhge, *tmp;
736 list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list) {
737 struct nh_group *nhg;
739 list_del(&nhge->nh_list);
740 nhg = rtnl_dereference(nhge->nh_parent->nh_grp);
741 remove_nh_grp_entry(nhge, nhg, nlinfo);
743 /* if this group has no more entries then remove it */
745 remove_nexthop(net, nhge->nh_parent, nlinfo);
749 static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo)
751 struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp);
752 int i, num_nh = nhg->num_nh;
754 for (i = 0; i < num_nh; ++i) {
755 struct nh_grp_entry *nhge = &nhg->nh_entries[i];
757 if (WARN_ON(!nhge->nh))
760 list_del(&nhge->nh_list);
761 nexthop_put(nhge->nh);
767 /* not called for nexthop replace */
768 static void __remove_nexthop_fib(struct net *net, struct nexthop *nh)
770 struct fib6_info *f6i, *tmp;
771 bool do_flush = false;
774 list_for_each_entry(fi, &nh->fi_list, nh_list) {
775 fi->fib_flags |= RTNH_F_DEAD;
781 /* ip6_del_rt removes the entry from this list hence the _safe */
782 list_for_each_entry_safe(f6i, tmp, &nh->f6i_list, nh_list) {
783 /* __ip6_del_rt does a release, so do a hold here */
785 ipv6_stub->ip6_del_rt(net, f6i);
789 static void __remove_nexthop(struct net *net, struct nexthop *nh,
790 struct nl_info *nlinfo)
792 __remove_nexthop_fib(net, nh);
795 remove_nexthop_group(nh, nlinfo);
799 nhi = rtnl_dereference(nh->nh_info);
800 if (nhi->fib_nhc.nhc_dev)
801 hlist_del(&nhi->dev_hash);
803 remove_nexthop_from_groups(net, nh, nlinfo);
807 static void remove_nexthop(struct net *net, struct nexthop *nh,
808 struct nl_info *nlinfo)
810 /* remove from the tree */
811 rb_erase(&nh->rb_node, &net->nexthop.rb_root);
814 nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo);
816 __remove_nexthop(net, nh, nlinfo);
817 nh_base_seq_inc(net);
822 /* if any FIB entries reference this nexthop, any dst entries
823 * need to be regenerated
825 static void nh_rt_cache_flush(struct net *net, struct nexthop *nh)
827 struct fib6_info *f6i;
829 if (!list_empty(&nh->fi_list))
832 list_for_each_entry(f6i, &nh->f6i_list, nh_list)
833 ipv6_stub->fib6_update_sernum(net, f6i);
836 static int replace_nexthop_grp(struct net *net, struct nexthop *old,
838 struct netlink_ext_ack *extack)
840 struct nh_group *oldg, *newg;
843 if (!new->is_group) {
844 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop.");
848 oldg = rtnl_dereference(old->nh_grp);
849 newg = rtnl_dereference(new->nh_grp);
851 /* update parents - used by nexthop code for cleanup */
852 for (i = 0; i < newg->num_nh; i++)
853 newg->nh_entries[i].nh_parent = old;
855 rcu_assign_pointer(old->nh_grp, newg);
857 for (i = 0; i < oldg->num_nh; i++)
858 oldg->nh_entries[i].nh_parent = new;
860 rcu_assign_pointer(new->nh_grp, oldg);
865 static int replace_nexthop_single(struct net *net, struct nexthop *old,
867 struct netlink_ext_ack *extack)
869 struct nh_info *oldi, *newi;
872 NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
876 oldi = rtnl_dereference(old->nh_info);
877 newi = rtnl_dereference(new->nh_info);
879 newi->nh_parent = old;
880 oldi->nh_parent = new;
882 old->protocol = new->protocol;
883 old->nh_flags = new->nh_flags;
885 rcu_assign_pointer(old->nh_info, newi);
886 rcu_assign_pointer(new->nh_info, oldi);
891 static void __nexthop_replace_notify(struct net *net, struct nexthop *nh,
892 struct nl_info *info)
894 struct fib6_info *f6i;
896 if (!list_empty(&nh->fi_list)) {
899 /* expectation is a few fib_info per nexthop and then
900 * a lot of routes per fib_info. So mark the fib_info
901 * and then walk the fib tables once
903 list_for_each_entry(fi, &nh->fi_list, nh_list)
904 fi->nh_updated = true;
906 fib_info_notify_update(net, info);
908 list_for_each_entry(fi, &nh->fi_list, nh_list)
909 fi->nh_updated = false;
912 list_for_each_entry(f6i, &nh->f6i_list, nh_list)
913 ipv6_stub->fib6_rt_update(net, f6i, info);
916 /* send RTM_NEWROUTE with REPLACE flag set for all FIB entries
917 * linked to this nexthop and for all groups that the nexthop
920 static void nexthop_replace_notify(struct net *net, struct nexthop *nh,
921 struct nl_info *info)
923 struct nh_grp_entry *nhge;
925 __nexthop_replace_notify(net, nh, info);
927 list_for_each_entry(nhge, &nh->grp_list, nh_list)
928 __nexthop_replace_notify(net, nhge->nh_parent, info);
931 static int replace_nexthop(struct net *net, struct nexthop *old,
932 struct nexthop *new, struct netlink_ext_ack *extack)
934 bool new_is_reject = false;
935 struct nh_grp_entry *nhge;
938 /* check that existing FIB entries are ok with the
939 * new nexthop definition
941 err = fib_check_nh_list(old, new, extack);
945 err = fib6_check_nh_list(old, new, extack);
949 if (!new->is_group) {
950 struct nh_info *nhi = rtnl_dereference(new->nh_info);
952 new_is_reject = nhi->reject_nh;
955 list_for_each_entry(nhge, &old->grp_list, nh_list) {
956 /* if new nexthop is a blackhole, any groups using this
957 * nexthop cannot have more than 1 path
960 nexthop_num_path(nhge->nh_parent) > 1) {
961 NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path");
965 err = fib_check_nh_list(nhge->nh_parent, new, extack);
969 err = fib6_check_nh_list(nhge->nh_parent, new, extack);
975 err = replace_nexthop_grp(net, old, new, extack);
977 err = replace_nexthop_single(net, old, new, extack);
980 nh_rt_cache_flush(net, old);
982 __remove_nexthop(net, new, NULL);
989 /* called with rtnl_lock held */
990 static int insert_nexthop(struct net *net, struct nexthop *new_nh,
991 struct nh_config *cfg, struct netlink_ext_ack *extack)
993 struct rb_node **pp, *parent = NULL, *next;
994 struct rb_root *root = &net->nexthop.rb_root;
995 bool replace = !!(cfg->nlflags & NLM_F_REPLACE);
996 bool create = !!(cfg->nlflags & NLM_F_CREATE);
997 u32 new_id = new_nh->id;
998 int replace_notify = 0;
1001 pp = &root->rb_node;
1005 next = rtnl_dereference(*pp);
1011 nh = rb_entry(parent, struct nexthop, rb_node);
1012 if (new_id < nh->id) {
1013 pp = &next->rb_left;
1014 } else if (new_id > nh->id) {
1015 pp = &next->rb_right;
1016 } else if (replace) {
1017 rc = replace_nexthop(net, nh, new_nh, extack);
1019 new_nh = nh; /* send notification with old nh */
1024 /* id already exists and not a replace */
1029 if (replace && !create) {
1030 NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists");
1035 rb_link_node_rcu(&new_nh->rb_node, parent, pp);
1036 rb_insert_color(&new_nh->rb_node, root);
1040 nh_base_seq_inc(net);
1041 nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo);
1043 nexthop_replace_notify(net, new_nh, &cfg->nlinfo);
1050 /* remove all nexthops tied to a device being deleted */
1051 static void nexthop_flush_dev(struct net_device *dev)
1053 unsigned int hash = nh_dev_hashfn(dev->ifindex);
1054 struct net *net = dev_net(dev);
1055 struct hlist_head *head = &net->nexthop.devhash[hash];
1056 struct hlist_node *n;
1057 struct nh_info *nhi;
1059 hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1060 if (nhi->fib_nhc.nhc_dev != dev)
1063 remove_nexthop(net, nhi->nh_parent, NULL);
1067 /* rtnl; called when net namespace is deleted */
1068 static void flush_all_nexthops(struct net *net)
1070 struct rb_root *root = &net->nexthop.rb_root;
1071 struct rb_node *node;
1074 while ((node = rb_first(root))) {
1075 nh = rb_entry(node, struct nexthop, rb_node);
1076 remove_nexthop(net, nh, NULL);
1081 static struct nexthop *nexthop_create_group(struct net *net,
1082 struct nh_config *cfg)
1084 struct nlattr *grps_attr = cfg->nh_grp;
1085 struct nexthop_grp *entry = nla_data(grps_attr);
1086 struct nh_group *nhg;
1090 nh = nexthop_alloc();
1092 return ERR_PTR(-ENOMEM);
1096 nhg = nexthop_grp_alloc(nla_len(grps_attr) / sizeof(*entry));
1099 return ERR_PTR(-ENOMEM);
1102 for (i = 0; i < nhg->num_nh; ++i) {
1103 struct nexthop *nhe;
1104 struct nh_info *nhi;
1106 nhe = nexthop_find_by_id(net, entry[i].id);
1107 if (!nexthop_get(nhe))
1110 nhi = rtnl_dereference(nhe->nh_info);
1111 if (nhi->family == AF_INET)
1114 nhg->nh_entries[i].nh = nhe;
1115 nhg->nh_entries[i].weight = entry[i].weight + 1;
1116 list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list);
1117 nhg->nh_entries[i].nh_parent = nh;
1120 if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) {
1122 nh_group_rebalance(nhg);
1125 rcu_assign_pointer(nh->nh_grp, nhg);
1131 nexthop_put(nhg->nh_entries[i].nh);
1136 return ERR_PTR(-ENOENT);
1139 static int nh_create_ipv4(struct net *net, struct nexthop *nh,
1140 struct nh_info *nhi, struct nh_config *cfg,
1141 struct netlink_ext_ack *extack)
1143 struct fib_nh *fib_nh = &nhi->fib_nh;
1144 struct fib_config fib_cfg = {
1145 .fc_oif = cfg->nh_ifindex,
1146 .fc_gw4 = cfg->gw.ipv4,
1147 .fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0,
1148 .fc_flags = cfg->nh_flags,
1149 .fc_encap = cfg->nh_encap,
1150 .fc_encap_type = cfg->nh_encap_type,
1152 u32 tb_id = l3mdev_fib_table(cfg->dev);
1155 err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack);
1157 fib_nh_release(net, fib_nh);
1161 /* sets nh_dev if successful */
1162 err = fib_check_nh(net, fib_nh, tb_id, 0, extack);
1164 nh->nh_flags = fib_nh->fib_nh_flags;
1165 fib_info_update_nhc_saddr(net, &fib_nh->nh_common,
1166 fib_nh->fib_nh_scope);
1168 fib_nh_release(net, fib_nh);
1174 static int nh_create_ipv6(struct net *net, struct nexthop *nh,
1175 struct nh_info *nhi, struct nh_config *cfg,
1176 struct netlink_ext_ack *extack)
1178 struct fib6_nh *fib6_nh = &nhi->fib6_nh;
1179 struct fib6_config fib6_cfg = {
1180 .fc_table = l3mdev_fib_table(cfg->dev),
1181 .fc_ifindex = cfg->nh_ifindex,
1182 .fc_gateway = cfg->gw.ipv6,
1183 .fc_flags = cfg->nh_flags,
1184 .fc_encap = cfg->nh_encap,
1185 .fc_encap_type = cfg->nh_encap_type,
1189 if (!ipv6_addr_any(&cfg->gw.ipv6))
1190 fib6_cfg.fc_flags |= RTF_GATEWAY;
1192 /* sets nh_dev if successful */
1193 err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL,
1196 ipv6_stub->fib6_nh_release(fib6_nh);
1198 nh->nh_flags = fib6_nh->fib_nh_flags;
1203 static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg,
1204 struct netlink_ext_ack *extack)
1206 struct nh_info *nhi;
1210 nh = nexthop_alloc();
1212 return ERR_PTR(-ENOMEM);
1214 nhi = kzalloc(sizeof(*nhi), GFP_KERNEL);
1217 return ERR_PTR(-ENOMEM);
1220 nh->nh_flags = cfg->nh_flags;
1223 nhi->nh_parent = nh;
1224 nhi->family = cfg->nh_family;
1225 nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK;
1227 if (cfg->nh_blackhole) {
1229 cfg->nh_ifindex = net->loopback_dev->ifindex;
1232 switch (cfg->nh_family) {
1234 err = nh_create_ipv4(net, nh, nhi, cfg, extack);
1237 err = nh_create_ipv6(net, nh, nhi, cfg, extack);
1244 return ERR_PTR(err);
1247 /* add the entry to the device based hash */
1248 nexthop_devhash_add(net, nhi);
1250 rcu_assign_pointer(nh->nh_info, nhi);
1255 /* called with rtnl lock held */
1256 static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg,
1257 struct netlink_ext_ack *extack)
1262 if (cfg->nlflags & NLM_F_REPLACE && !cfg->nh_id) {
1263 NL_SET_ERR_MSG(extack, "Replace requires nexthop id");
1264 return ERR_PTR(-EINVAL);
1268 cfg->nh_id = nh_find_unused_id(net);
1270 NL_SET_ERR_MSG(extack, "No unused id");
1271 return ERR_PTR(-EINVAL);
1276 nh = nexthop_create_group(net, cfg);
1278 nh = nexthop_create(net, cfg, extack);
1283 refcount_set(&nh->refcnt, 1);
1284 nh->id = cfg->nh_id;
1285 nh->protocol = cfg->nh_protocol;
1288 err = insert_nexthop(net, nh, cfg, extack);
1290 __remove_nexthop(net, nh, NULL);
1298 static int rtm_to_nh_config(struct net *net, struct sk_buff *skb,
1299 struct nlmsghdr *nlh, struct nh_config *cfg,
1300 struct netlink_ext_ack *extack)
1302 struct nhmsg *nhm = nlmsg_data(nlh);
1303 struct nlattr *tb[NHA_MAX + 1];
1306 err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1312 if (nhm->resvd || nhm->nh_scope) {
1313 NL_SET_ERR_MSG(extack, "Invalid values in ancillary header");
1316 if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) {
1317 NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header");
1321 switch (nhm->nh_family) {
1330 NL_SET_ERR_MSG(extack, "Invalid address family");
1334 if (tb[NHA_GROUPS] || tb[NHA_MASTER]) {
1335 NL_SET_ERR_MSG(extack, "Invalid attributes in request");
1339 memset(cfg, 0, sizeof(*cfg));
1340 cfg->nlflags = nlh->nlmsg_flags;
1341 cfg->nlinfo.portid = NETLINK_CB(skb).portid;
1342 cfg->nlinfo.nlh = nlh;
1343 cfg->nlinfo.nl_net = net;
1345 cfg->nh_family = nhm->nh_family;
1346 cfg->nh_protocol = nhm->nh_protocol;
1347 cfg->nh_flags = nhm->nh_flags;
1350 cfg->nh_id = nla_get_u32(tb[NHA_ID]);
1352 if (tb[NHA_GROUP]) {
1353 if (nhm->nh_family != AF_UNSPEC) {
1354 NL_SET_ERR_MSG(extack, "Invalid family for group");
1357 cfg->nh_grp = tb[NHA_GROUP];
1359 cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH;
1360 if (tb[NHA_GROUP_TYPE])
1361 cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]);
1363 if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) {
1364 NL_SET_ERR_MSG(extack, "Invalid group type");
1367 err = nh_check_attr_group(net, tb, extack);
1369 /* no other attributes should be set */
1373 if (tb[NHA_BLACKHOLE]) {
1374 if (tb[NHA_GATEWAY] || tb[NHA_OIF] ||
1375 tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE]) {
1376 NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway or oif");
1380 cfg->nh_blackhole = 1;
1386 NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole nexthops");
1390 cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]);
1391 if (cfg->nh_ifindex)
1392 cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex);
1395 NL_SET_ERR_MSG(extack, "Invalid device index");
1397 } else if (!(cfg->dev->flags & IFF_UP)) {
1398 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1401 } else if (!netif_carrier_ok(cfg->dev)) {
1402 NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down");
1408 if (tb[NHA_GATEWAY]) {
1409 struct nlattr *gwa = tb[NHA_GATEWAY];
1411 switch (cfg->nh_family) {
1413 if (nla_len(gwa) != sizeof(u32)) {
1414 NL_SET_ERR_MSG(extack, "Invalid gateway");
1417 cfg->gw.ipv4 = nla_get_be32(gwa);
1420 if (nla_len(gwa) != sizeof(struct in6_addr)) {
1421 NL_SET_ERR_MSG(extack, "Invalid gateway");
1424 cfg->gw.ipv6 = nla_get_in6_addr(gwa);
1427 NL_SET_ERR_MSG(extack,
1428 "Unknown address family for gateway");
1432 /* device only nexthop (no gateway) */
1433 if (cfg->nh_flags & RTNH_F_ONLINK) {
1434 NL_SET_ERR_MSG(extack,
1435 "ONLINK flag can not be set for nexthop without a gateway");
1440 if (tb[NHA_ENCAP]) {
1441 cfg->nh_encap = tb[NHA_ENCAP];
1443 if (!tb[NHA_ENCAP_TYPE]) {
1444 NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing");
1448 cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]);
1449 err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack);
1453 } else if (tb[NHA_ENCAP_TYPE]) {
1454 NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing");
1465 static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1466 struct netlink_ext_ack *extack)
1468 struct net *net = sock_net(skb->sk);
1469 struct nh_config cfg;
1473 err = rtm_to_nh_config(net, skb, nlh, &cfg, extack);
1475 nh = nexthop_add(net, &cfg, extack);
1483 static int nh_valid_get_del_req(struct nlmsghdr *nlh, u32 *id,
1484 struct netlink_ext_ack *extack)
1486 struct nhmsg *nhm = nlmsg_data(nlh);
1487 struct nlattr *tb[NHA_MAX + 1];
1490 err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1496 for (i = 0; i < __NHA_MAX; ++i) {
1504 NL_SET_ERR_MSG_ATTR(extack, tb[i],
1505 "Unexpected attribute in request");
1509 if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1510 NL_SET_ERR_MSG(extack, "Invalid values in header");
1515 NL_SET_ERR_MSG(extack, "Nexthop id is missing");
1519 *id = nla_get_u32(tb[NHA_ID]);
1521 NL_SET_ERR_MSG(extack, "Invalid nexthop id");
1529 static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh,
1530 struct netlink_ext_ack *extack)
1532 struct net *net = sock_net(skb->sk);
1533 struct nl_info nlinfo = {
1536 .portid = NETLINK_CB(skb).portid,
1542 err = nh_valid_get_del_req(nlh, &id, extack);
1546 nh = nexthop_find_by_id(net, id);
1550 remove_nexthop(net, nh, &nlinfo);
1556 static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh,
1557 struct netlink_ext_ack *extack)
1559 struct net *net = sock_net(in_skb->sk);
1560 struct sk_buff *skb = NULL;
1565 err = nh_valid_get_del_req(nlh, &id, extack);
1570 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
1575 nh = nexthop_find_by_id(net, id);
1579 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid,
1582 WARN_ON(err == -EMSGSIZE);
1586 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid);
1594 static bool nh_dump_filtered(struct nexthop *nh, int dev_idx, int master_idx,
1595 bool group_filter, u8 family)
1597 const struct net_device *dev;
1598 const struct nh_info *nhi;
1600 if (group_filter && !nh->is_group)
1603 if (!dev_idx && !master_idx && !family)
1609 nhi = rtnl_dereference(nh->nh_info);
1610 if (family && nhi->family != family)
1613 dev = nhi->fib_nhc.nhc_dev;
1614 if (dev_idx && (!dev || dev->ifindex != dev_idx))
1618 struct net_device *master;
1623 master = netdev_master_upper_dev_get((struct net_device *)dev);
1624 if (!master || master->ifindex != master_idx)
1631 static int nh_valid_dump_req(const struct nlmsghdr *nlh, int *dev_idx,
1632 int *master_idx, bool *group_filter,
1633 struct netlink_callback *cb)
1635 struct netlink_ext_ack *extack = cb->extack;
1636 struct nlattr *tb[NHA_MAX + 1];
1641 err = nlmsg_parse(nlh, sizeof(*nhm), tb, NHA_MAX, rtm_nh_policy,
1646 for (i = 0; i <= NHA_MAX; ++i) {
1652 idx = nla_get_u32(tb[i]);
1653 if (idx > INT_MAX) {
1654 NL_SET_ERR_MSG(extack, "Invalid device index");
1660 idx = nla_get_u32(tb[i]);
1661 if (idx > INT_MAX) {
1662 NL_SET_ERR_MSG(extack, "Invalid master device index");
1668 *group_filter = true;
1671 NL_SET_ERR_MSG(extack, "Unsupported attribute in dump request");
1676 nhm = nlmsg_data(nlh);
1677 if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) {
1678 NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request");
1686 static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb)
1688 struct nhmsg *nhm = nlmsg_data(cb->nlh);
1689 int dev_filter_idx = 0, master_idx = 0;
1690 struct net *net = sock_net(skb->sk);
1691 struct rb_root *root = &net->nexthop.rb_root;
1692 bool group_filter = false;
1693 struct rb_node *node;
1697 err = nh_valid_dump_req(cb->nlh, &dev_filter_idx, &master_idx,
1702 s_idx = cb->args[0];
1703 for (node = rb_first(root); node; node = rb_next(node)) {
1709 nh = rb_entry(node, struct nexthop, rb_node);
1710 if (nh_dump_filtered(nh, dev_filter_idx, master_idx,
1711 group_filter, nhm->nh_family))
1714 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP,
1715 NETLINK_CB(cb->skb).portid,
1716 cb->nlh->nlmsg_seq, NLM_F_MULTI);
1718 if (likely(skb->len))
1731 cb->seq = net->nexthop.seq;
1732 nl_dump_check_consistent(cb, nlmsg_hdr(skb));
1737 static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu)
1739 unsigned int hash = nh_dev_hashfn(dev->ifindex);
1740 struct net *net = dev_net(dev);
1741 struct hlist_head *head = &net->nexthop.devhash[hash];
1742 struct hlist_node *n;
1743 struct nh_info *nhi;
1745 hlist_for_each_entry_safe(nhi, n, head, dev_hash) {
1746 if (nhi->fib_nhc.nhc_dev == dev) {
1747 if (nhi->family == AF_INET)
1748 fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu,
1755 static int nh_netdev_event(struct notifier_block *this,
1756 unsigned long event, void *ptr)
1758 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1759 struct netdev_notifier_info_ext *info_ext;
1763 case NETDEV_UNREGISTER:
1764 nexthop_flush_dev(dev);
1767 if (!(dev_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP)))
1768 nexthop_flush_dev(dev);
1770 case NETDEV_CHANGEMTU:
1772 nexthop_sync_mtu(dev, info_ext->ext.mtu);
1773 rt_cache_flush(dev_net(dev));
1779 static struct notifier_block nh_netdev_notifier = {
1780 .notifier_call = nh_netdev_event,
1783 static void __net_exit nexthop_net_exit(struct net *net)
1786 flush_all_nexthops(net);
1788 kfree(net->nexthop.devhash);
1791 static int __net_init nexthop_net_init(struct net *net)
1793 size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE;
1795 net->nexthop.rb_root = RB_ROOT;
1796 net->nexthop.devhash = kzalloc(sz, GFP_KERNEL);
1797 if (!net->nexthop.devhash)
1803 static struct pernet_operations nexthop_net_ops = {
1804 .init = nexthop_net_init,
1805 .exit = nexthop_net_exit,
1808 static int __init nexthop_init(void)
1810 register_pernet_subsys(&nexthop_net_ops);
1812 register_netdevice_notifier(&nh_netdev_notifier);
1814 rtnl_register(PF_UNSPEC, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1815 rtnl_register(PF_UNSPEC, RTM_DELNEXTHOP, rtm_del_nexthop, NULL, 0);
1816 rtnl_register(PF_UNSPEC, RTM_GETNEXTHOP, rtm_get_nexthop,
1817 rtm_dump_nexthop, 0);
1819 rtnl_register(PF_INET, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1820 rtnl_register(PF_INET, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1822 rtnl_register(PF_INET6, RTM_NEWNEXTHOP, rtm_new_nexthop, NULL, 0);
1823 rtnl_register(PF_INET6, RTM_GETNEXTHOP, NULL, rtm_dump_nexthop, 0);
1827 subsys_initcall(nexthop_init);