// SPDX-License-Identifier: GPL-2.0-or-later /* Kernel routing table updates using netlink over GNU/Linux system. * Copyright (C) 1997, 98, 99 Kunihiro Ishiguro */ #include #ifdef HAVE_NETLINK /* The following definition is to workaround an issue in the Linux kernel * header files with redefinition of 'struct in6_addr' in both * netinet/in.h and linux/in6.h. * Reference - https://sourceware.org/ml/libc-alpha/2013-01/msg00599.html */ #define _LINUX_IN6_H #include #include #include #include #include #include #include #include /* Hack for GNU libc version 2. */ #ifndef MSG_TRUNC #define MSG_TRUNC 0x20 #endif /* MSG_TRUNC */ #include "linklist.h" #include "if.h" #include "log.h" #include "prefix.h" #include "plist.h" #include "plist_int.h" #include "connected.h" #include "table.h" #include "memory.h" #include "rib.h" #include "thread.h" #include "privs.h" #include "nexthop.h" #include "vrf.h" #include "vty.h" #include "mpls.h" #include "vxlan.h" #include "printfrr.h" #include "zebra/zapi_msg.h" #include "zebra/zebra_ns.h" #include "zebra/zebra_vrf.h" #include "zebra/rt.h" #include "zebra/redistribute.h" #include "zebra/interface.h" #include "zebra/debug.h" #include "zebra/rtadv.h" #include "zebra/zebra_ptm.h" #include "zebra/zebra_mpls.h" #include "zebra/kernel_netlink.h" #include "zebra/rt_netlink.h" #include "zebra/zebra_nhg.h" #include "zebra/zebra_mroute.h" #include "zebra/zebra_vxlan.h" #include "zebra/zebra_errors.h" #include "zebra/zebra_evpn_mh.h" #include "zebra/zebra_trace.h" #include "zebra/zebra_neigh.h" #ifndef AF_MPLS #define AF_MPLS 28 #endif /* Re-defining as I am unable to include which has the * UAPI for MAC sync. */ #ifndef _UAPI_LINUX_IF_BRIDGE_H #define BR_SPH_LIST_SIZE 10 #endif static vlanid_t filter_vlan = 0; /* We capture whether the current kernel supports nexthop ids; by * default, we'll use them if possible. There's also a configuration * available to _disable_ use of kernel nexthops. */ static bool supports_nh; struct gw_family_t { uint16_t filler; uint16_t family; union g_addr gate; }; static const char ipv4_ll_buf[16] = "169.254.0.1"; static struct in_addr ipv4_ll; /* Is this a ipv4 over ipv6 route? */ static bool is_route_v4_over_v6(unsigned char rtm_family, enum nexthop_types_t nexthop_type) { if (rtm_family == AF_INET && (nexthop_type == NEXTHOP_TYPE_IPV6 || nexthop_type == NEXTHOP_TYPE_IPV6_IFINDEX)) return true; return false; } /* Helper to control use of kernel-level nexthop ids */ static bool kernel_nexthops_supported(void) { return (supports_nh && !vrf_is_backend_netns() && zebra_nhg_kernel_nexthops_enabled()); } /* * Some people may only want to use NHGs created by protos and not * implicitly created by Zebra. This check accounts for that. */ static bool proto_nexthops_only(void) { return zebra_nhg_proto_nexthops_only(); } /* Is this a proto created NHG? */ static bool is_proto_nhg(uint32_t id, int type) { /* If type is available, use it as the source of truth */ if (type) { if (type != ZEBRA_ROUTE_NHG) return true; return false; } if (id >= ZEBRA_NHG_PROTO_LOWER) return true; return false; } /* Is vni mcast group */ static bool is_mac_vni_mcast_group(struct ethaddr *mac, vni_t vni, struct in_addr grp_addr) { if (!vni) return false; if (!is_zero_mac(mac)) return false; if (!IN_MULTICAST(ntohl(grp_addr.s_addr))) return false; return true; } /* * The ipv4_ll data structure is used for all 5549 * additions to the kernel. Let's figure out the * correct value one time instead for every * install/remove of a 5549 type route */ void rt_netlink_init(void) { inet_pton(AF_INET, ipv4_ll_buf, &ipv4_ll); } /* * Mapping from dataplane neighbor flags to netlink flags */ static uint8_t neigh_flags_to_netlink(uint8_t dplane_flags) { uint8_t flags = 0; if (dplane_flags & DPLANE_NTF_EXT_LEARNED) flags |= NTF_EXT_LEARNED; if (dplane_flags & DPLANE_NTF_ROUTER) flags |= NTF_ROUTER; if (dplane_flags & DPLANE_NTF_USE) flags |= NTF_USE; return flags; } /* * Mapping from dataplane neighbor state to netlink state */ static uint16_t neigh_state_to_netlink(uint16_t dplane_state) { uint16_t state = 0; if (dplane_state & DPLANE_NUD_REACHABLE) state |= NUD_REACHABLE; if (dplane_state & DPLANE_NUD_STALE) state |= NUD_STALE; if (dplane_state & DPLANE_NUD_NOARP) state |= NUD_NOARP; if (dplane_state & DPLANE_NUD_PROBE) state |= NUD_PROBE; if (dplane_state & DPLANE_NUD_INCOMPLETE) state |= NUD_INCOMPLETE; if (dplane_state & DPLANE_NUD_PERMANENT) state |= NUD_PERMANENT; if (dplane_state & DPLANE_NUD_FAILED) state |= NUD_FAILED; return state; } static inline bool is_selfroute(int proto) { if ((proto == RTPROT_BGP) || (proto == RTPROT_OSPF) || (proto == RTPROT_ZSTATIC) || (proto == RTPROT_ZEBRA) || (proto == RTPROT_ISIS) || (proto == RTPROT_RIPNG) || (proto == RTPROT_NHRP) || (proto == RTPROT_EIGRP) || (proto == RTPROT_LDP) || (proto == RTPROT_BABEL) || (proto == RTPROT_RIP) || (proto == RTPROT_SHARP) || (proto == RTPROT_PBR) || (proto == RTPROT_OPENFABRIC) || (proto == RTPROT_SRTE)) { return true; } return false; } int zebra2proto(int proto) { switch (proto) { case ZEBRA_ROUTE_BABEL: proto = RTPROT_BABEL; break; case ZEBRA_ROUTE_BGP: proto = RTPROT_BGP; break; case ZEBRA_ROUTE_OSPF: case ZEBRA_ROUTE_OSPF6: proto = RTPROT_OSPF; break; case ZEBRA_ROUTE_STATIC: proto = RTPROT_ZSTATIC; break; case ZEBRA_ROUTE_ISIS: proto = RTPROT_ISIS; break; case ZEBRA_ROUTE_RIP: proto = RTPROT_RIP; break; case ZEBRA_ROUTE_RIPNG: proto = RTPROT_RIPNG; break; case ZEBRA_ROUTE_NHRP: proto = RTPROT_NHRP; break; case ZEBRA_ROUTE_EIGRP: proto = RTPROT_EIGRP; break; case ZEBRA_ROUTE_LDP: proto = RTPROT_LDP; break; case ZEBRA_ROUTE_SHARP: proto = RTPROT_SHARP; break; case ZEBRA_ROUTE_PBR: proto = RTPROT_PBR; break; case ZEBRA_ROUTE_OPENFABRIC: proto = RTPROT_OPENFABRIC; break; case ZEBRA_ROUTE_SRTE: proto = RTPROT_SRTE; break; case ZEBRA_ROUTE_TABLE: case ZEBRA_ROUTE_NHG: proto = RTPROT_ZEBRA; break; case ZEBRA_ROUTE_CONNECT: case ZEBRA_ROUTE_KERNEL: proto = RTPROT_KERNEL; break; default: /* * When a user adds a new protocol this will show up * to let them know to do something about it. This * is intentionally a warn because we should see * this as part of development of a new protocol */ zlog_debug( "%s: Please add this protocol(%d) to proper rt_netlink.c handling", __func__, proto); proto = RTPROT_ZEBRA; break; } return proto; } static inline int proto2zebra(int proto, int family, bool is_nexthop) { switch (proto) { case RTPROT_BABEL: proto = ZEBRA_ROUTE_BABEL; break; case RTPROT_BGP: proto = ZEBRA_ROUTE_BGP; break; case RTPROT_OSPF: proto = (family == AF_INET) ? ZEBRA_ROUTE_OSPF : ZEBRA_ROUTE_OSPF6; break; case RTPROT_ISIS: proto = ZEBRA_ROUTE_ISIS; break; case RTPROT_RIP: proto = ZEBRA_ROUTE_RIP; break; case RTPROT_RIPNG: proto = ZEBRA_ROUTE_RIPNG; break; case RTPROT_NHRP: proto = ZEBRA_ROUTE_NHRP; break; case RTPROT_EIGRP: proto = ZEBRA_ROUTE_EIGRP; break; case RTPROT_LDP: proto = ZEBRA_ROUTE_LDP; break; case RTPROT_STATIC: case RTPROT_ZSTATIC: proto = ZEBRA_ROUTE_STATIC; break; case RTPROT_SHARP: proto = ZEBRA_ROUTE_SHARP; break; case RTPROT_PBR: proto = ZEBRA_ROUTE_PBR; break; case RTPROT_OPENFABRIC: proto = ZEBRA_ROUTE_OPENFABRIC; break; case RTPROT_SRTE: proto = ZEBRA_ROUTE_SRTE; break; case RTPROT_UNSPEC: case RTPROT_REDIRECT: case RTPROT_KERNEL: case RTPROT_BOOT: case RTPROT_GATED: case RTPROT_RA: case RTPROT_MRT: case RTPROT_BIRD: case RTPROT_DNROUTED: case RTPROT_XORP: case RTPROT_NTK: case RTPROT_MROUTED: case RTPROT_KEEPALIVED: case RTPROT_OPENR: proto = ZEBRA_ROUTE_KERNEL; break; case RTPROT_ZEBRA: if (is_nexthop) { proto = ZEBRA_ROUTE_NHG; break; } /* Intentional fall thru */ default: /* * When a user adds a new protocol this will show up * to let them know to do something about it. This * is intentionally a warn because we should see * this as part of development of a new protocol */ zlog_debug( "%s: Please add this protocol(%d) to proper rt_netlink.c handling", __func__, proto); proto = ZEBRA_ROUTE_KERNEL; break; } return proto; } /* Pending: create an efficient table_id (in a tree/hash) based lookup) */ vrf_id_t vrf_lookup_by_table(uint32_t table_id, ns_id_t ns_id) { struct vrf *vrf; struct zebra_vrf *zvrf; RB_FOREACH (vrf, vrf_id_head, &vrfs_by_id) { zvrf = vrf->info; if (zvrf == NULL) continue; /* case vrf with netns : match the netnsid */ if (vrf_is_backend_netns()) { if (ns_id == zvrf_id(zvrf)) return zvrf_id(zvrf); } else { /* VRF is VRF_BACKEND_VRF_LITE */ if (zvrf->table_id != table_id) continue; return zvrf_id(zvrf); } } return VRF_DEFAULT; } /** * @parse_encap_mpls() - Parses encapsulated mpls attributes * @tb: Pointer to rtattr to look for nested items in. * @labels: Pointer to store labels in. * * Return: Number of mpls labels found. */ static int parse_encap_mpls(struct rtattr *tb, mpls_label_t *labels) { struct rtattr *tb_encap[MPLS_IPTUNNEL_MAX + 1] = {0}; mpls_lse_t *lses = NULL; int num_labels = 0; uint32_t ttl = 0; uint32_t bos = 0; uint32_t exp = 0; mpls_label_t label = 0; netlink_parse_rtattr_nested(tb_encap, MPLS_IPTUNNEL_MAX, tb); lses = (mpls_lse_t *)RTA_DATA(tb_encap[MPLS_IPTUNNEL_DST]); while (!bos && num_labels < MPLS_MAX_LABELS) { mpls_lse_decode(lses[num_labels], &label, &ttl, &exp, &bos); labels[num_labels++] = label; } return num_labels; } static enum seg6local_action_t parse_encap_seg6local(struct rtattr *tb, struct seg6local_context *ctx) { struct rtattr *tb_encap[SEG6_LOCAL_MAX + 1] = {}; enum seg6local_action_t act = ZEBRA_SEG6_LOCAL_ACTION_UNSPEC; netlink_parse_rtattr_nested(tb_encap, SEG6_LOCAL_MAX, tb); if (tb_encap[SEG6_LOCAL_ACTION]) act = *(uint32_t *)RTA_DATA(tb_encap[SEG6_LOCAL_ACTION]); if (tb_encap[SEG6_LOCAL_NH4]) ctx->nh4 = *(struct in_addr *)RTA_DATA( tb_encap[SEG6_LOCAL_NH4]); if (tb_encap[SEG6_LOCAL_NH6]) ctx->nh6 = *(struct in6_addr *)RTA_DATA( tb_encap[SEG6_LOCAL_NH6]); if (tb_encap[SEG6_LOCAL_TABLE]) ctx->table = *(uint32_t *)RTA_DATA(tb_encap[SEG6_LOCAL_TABLE]); if (tb_encap[SEG6_LOCAL_VRFTABLE]) ctx->table = *(uint32_t *)RTA_DATA(tb_encap[SEG6_LOCAL_VRFTABLE]); return act; } static int parse_encap_seg6(struct rtattr *tb, struct in6_addr *segs) { struct rtattr *tb_encap[SEG6_IPTUNNEL_MAX + 1] = {}; struct seg6_iptunnel_encap *ipt = NULL; struct in6_addr *segments = NULL; netlink_parse_rtattr_nested(tb_encap, SEG6_IPTUNNEL_MAX, tb); /* * TODO: It's not support multiple SID list. */ if (tb_encap[SEG6_IPTUNNEL_SRH]) { ipt = (struct seg6_iptunnel_encap *) RTA_DATA(tb_encap[SEG6_IPTUNNEL_SRH]); segments = ipt->srh[0].segments; *segs = segments[0]; return 1; } return 0; } static struct nexthop parse_nexthop_unicast(ns_id_t ns_id, struct rtmsg *rtm, struct rtattr **tb, enum blackhole_type bh_type, int index, void *prefsrc, void *gate, afi_t afi, vrf_id_t vrf_id) { struct interface *ifp = NULL; struct nexthop nh = {0}; mpls_label_t labels[MPLS_MAX_LABELS] = {0}; int num_labels = 0; enum seg6local_action_t seg6l_act = ZEBRA_SEG6_LOCAL_ACTION_UNSPEC; struct seg6local_context seg6l_ctx = {}; struct in6_addr seg6_segs = {}; int num_segs = 0; vrf_id_t nh_vrf_id = vrf_id; size_t sz = (afi == AFI_IP) ? 4 : 16; if (bh_type == BLACKHOLE_UNSPEC) { if (index && !gate) nh.type = NEXTHOP_TYPE_IFINDEX; else if (index && gate) nh.type = (afi == AFI_IP) ? NEXTHOP_TYPE_IPV4_IFINDEX : NEXTHOP_TYPE_IPV6_IFINDEX; else if (!index && gate) nh.type = (afi == AFI_IP) ? NEXTHOP_TYPE_IPV4 : NEXTHOP_TYPE_IPV6; else { nh.type = NEXTHOP_TYPE_BLACKHOLE; nh.bh_type = bh_type; } } else { nh.type = NEXTHOP_TYPE_BLACKHOLE; nh.bh_type = bh_type; } nh.ifindex = index; if (prefsrc) memcpy(&nh.src, prefsrc, sz); if (gate) memcpy(&nh.gate, gate, sz); if (index) { ifp = if_lookup_by_index_per_ns(zebra_ns_lookup(ns_id), index); if (ifp) nh_vrf_id = ifp->vrf->vrf_id; } nh.vrf_id = nh_vrf_id; if (tb[RTA_ENCAP] && tb[RTA_ENCAP_TYPE] && *(uint16_t *)RTA_DATA(tb[RTA_ENCAP_TYPE]) == LWTUNNEL_ENCAP_MPLS) { num_labels = parse_encap_mpls(tb[RTA_ENCAP], labels); } if (tb[RTA_ENCAP] && tb[RTA_ENCAP_TYPE] && *(uint16_t *)RTA_DATA(tb[RTA_ENCAP_TYPE]) == LWTUNNEL_ENCAP_SEG6_LOCAL) { seg6l_act = parse_encap_seg6local(tb[RTA_ENCAP], &seg6l_ctx); } if (tb[RTA_ENCAP] && tb[RTA_ENCAP_TYPE] && *(uint16_t *)RTA_DATA(tb[RTA_ENCAP_TYPE]) == LWTUNNEL_ENCAP_SEG6) { num_segs = parse_encap_seg6(tb[RTA_ENCAP], &seg6_segs); } if (rtm->rtm_flags & RTNH_F_ONLINK) SET_FLAG(nh.flags, NEXTHOP_FLAG_ONLINK); if (rtm->rtm_flags & RTNH_F_LINKDOWN) SET_FLAG(nh.flags, NEXTHOP_FLAG_LINKDOWN); if (num_labels) nexthop_add_labels(&nh, ZEBRA_LSP_STATIC, num_labels, labels); if (seg6l_act != ZEBRA_SEG6_LOCAL_ACTION_UNSPEC) nexthop_add_srv6_seg6local(&nh, seg6l_act, &seg6l_ctx); if (num_segs) nexthop_add_srv6_seg6(&nh, &seg6_segs); return nh; } static uint8_t parse_multipath_nexthops_unicast(ns_id_t ns_id, struct nexthop_group *ng, struct rtmsg *rtm, struct rtnexthop *rtnh, struct rtattr **tb, void *prefsrc, vrf_id_t vrf_id) { void *gate = NULL; struct interface *ifp = NULL; int index = 0; /* MPLS labels */ mpls_label_t labels[MPLS_MAX_LABELS] = {0}; int num_labels = 0; enum seg6local_action_t seg6l_act = ZEBRA_SEG6_LOCAL_ACTION_UNSPEC; struct seg6local_context seg6l_ctx = {}; struct in6_addr seg6_segs = {}; int num_segs = 0; struct rtattr *rtnh_tb[RTA_MAX + 1] = {}; int len = RTA_PAYLOAD(tb[RTA_MULTIPATH]); vrf_id_t nh_vrf_id = vrf_id; for (;;) { struct nexthop *nh = NULL; if (len < (int)sizeof(*rtnh) || rtnh->rtnh_len > len) break; index = rtnh->rtnh_ifindex; if (index) { /* * Yes we are looking this up * for every nexthop and just * using the last one looked * up right now */ ifp = if_lookup_by_index_per_ns(zebra_ns_lookup(ns_id), index); if (ifp) nh_vrf_id = ifp->vrf->vrf_id; else { flog_warn( EC_ZEBRA_UNKNOWN_INTERFACE, "%s: Unknown interface %u specified, defaulting to VRF_DEFAULT", __func__, index); nh_vrf_id = VRF_DEFAULT; } } else nh_vrf_id = vrf_id; if (rtnh->rtnh_len > sizeof(*rtnh)) { netlink_parse_rtattr(rtnh_tb, RTA_MAX, RTNH_DATA(rtnh), rtnh->rtnh_len - sizeof(*rtnh)); if (rtnh_tb[RTA_GATEWAY]) gate = RTA_DATA(rtnh_tb[RTA_GATEWAY]); if (rtnh_tb[RTA_ENCAP] && rtnh_tb[RTA_ENCAP_TYPE] && *(uint16_t *)RTA_DATA(rtnh_tb[RTA_ENCAP_TYPE]) == LWTUNNEL_ENCAP_MPLS) { num_labels = parse_encap_mpls( rtnh_tb[RTA_ENCAP], labels); } if (rtnh_tb[RTA_ENCAP] && rtnh_tb[RTA_ENCAP_TYPE] && *(uint16_t *)RTA_DATA(rtnh_tb[RTA_ENCAP_TYPE]) == LWTUNNEL_ENCAP_SEG6_LOCAL) { seg6l_act = parse_encap_seg6local( rtnh_tb[RTA_ENCAP], &seg6l_ctx); } if (rtnh_tb[RTA_ENCAP] && rtnh_tb[RTA_ENCAP_TYPE] && *(uint16_t *)RTA_DATA(rtnh_tb[RTA_ENCAP_TYPE]) == LWTUNNEL_ENCAP_SEG6) { num_segs = parse_encap_seg6(rtnh_tb[RTA_ENCAP], &seg6_segs); } } if (gate && rtm->rtm_family == AF_INET) { if (index) nh = nexthop_from_ipv4_ifindex( gate, prefsrc, index, nh_vrf_id); else nh = nexthop_from_ipv4(gate, prefsrc, nh_vrf_id); } else if (gate && rtm->rtm_family == AF_INET6) { if (index) nh = nexthop_from_ipv6_ifindex( gate, index, nh_vrf_id); else nh = nexthop_from_ipv6(gate, nh_vrf_id); } else nh = nexthop_from_ifindex(index, nh_vrf_id); if (nh) { nh->weight = rtnh->rtnh_hops + 1; if (num_labels) nexthop_add_labels(nh, ZEBRA_LSP_STATIC, num_labels, labels); if (seg6l_act != ZEBRA_SEG6_LOCAL_ACTION_UNSPEC) nexthop_add_srv6_seg6local(nh, seg6l_act, &seg6l_ctx); if (num_segs) nexthop_add_srv6_seg6(nh, &seg6_segs); if (rtnh->rtnh_flags & RTNH_F_ONLINK) SET_FLAG(nh->flags, NEXTHOP_FLAG_ONLINK); /* Add to temporary list */ nexthop_group_add_sorted(ng, nh); } if (rtnh->rtnh_len == 0) break; len -= NLMSG_ALIGN(rtnh->rtnh_len); rtnh = RTNH_NEXT(rtnh); } uint8_t nhop_num = nexthop_group_nexthop_num(ng); return nhop_num; } /* Looking up routing table by netlink interface. */ int netlink_route_change_read_unicast_internal(struct nlmsghdr *h, ns_id_t ns_id, int startup, struct zebra_dplane_ctx *ctx) { int len; struct rtmsg *rtm; struct rtattr *tb[RTA_MAX + 1]; uint32_t flags = 0; struct prefix p; struct prefix_ipv6 src_p = {}; vrf_id_t vrf_id; bool selfroute; char anyaddr[16] = {0}; int proto = ZEBRA_ROUTE_KERNEL; int index = 0; int table; int metric = 0; uint32_t mtu = 0; uint8_t distance = 0; route_tag_t tag = 0; uint32_t nhe_id = 0; void *dest = NULL; void *gate = NULL; void *prefsrc = NULL; /* IPv4 preferred source host address */ void *src = NULL; /* IPv6 srcdest source prefix */ enum blackhole_type bh_type = BLACKHOLE_UNSPEC; frrtrace(3, frr_zebra, netlink_route_change_read_unicast, h, ns_id, startup); rtm = NLMSG_DATA(h); if (startup && h->nlmsg_type != RTM_NEWROUTE) return 0; switch (rtm->rtm_type) { case RTN_UNICAST: break; case RTN_BLACKHOLE: bh_type = BLACKHOLE_NULL; break; case RTN_UNREACHABLE: bh_type = BLACKHOLE_REJECT; break; case RTN_PROHIBIT: bh_type = BLACKHOLE_ADMINPROHIB; break; default: if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("Route rtm_type: %s(%d) intentionally ignoring", nl_rttype_to_str(rtm->rtm_type), rtm->rtm_type); return 0; } len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct rtmsg)); if (len < 0) { zlog_err( "%s: Message received from netlink is of a broken size %d %zu", __func__, h->nlmsg_len, (size_t)NLMSG_LENGTH(sizeof(struct rtmsg))); return -1; } netlink_parse_rtattr(tb, RTA_MAX, RTM_RTA(rtm), len); if (rtm->rtm_flags & RTM_F_CLONED) return 0; if (rtm->rtm_protocol == RTPROT_REDIRECT) return 0; if (rtm->rtm_protocol == RTPROT_KERNEL) return 0; selfroute = is_selfroute(rtm->rtm_protocol); if (!startup && selfroute && h->nlmsg_type == RTM_NEWROUTE && !zrouter.asic_offloaded && !ctx) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("Route type: %d Received that we think we have originated, ignoring", rtm->rtm_protocol); return 0; } /* We don't care about change notifications for the MPLS table. */ /* TODO: Revisit this. */ if (rtm->rtm_family == AF_MPLS) return 0; /* Table corresponding to route. */ if (tb[RTA_TABLE]) table = *(int *)RTA_DATA(tb[RTA_TABLE]); else table = rtm->rtm_table; /* Map to VRF */ vrf_id = vrf_lookup_by_table(table, ns_id); if (vrf_id == VRF_DEFAULT) { if (!is_zebra_valid_kernel_table(table) && !is_zebra_main_routing_table(table)) return 0; } if (rtm->rtm_flags & RTM_F_TRAP) flags |= ZEBRA_FLAG_TRAPPED; if (rtm->rtm_flags & RTM_F_OFFLOAD) flags |= ZEBRA_FLAG_OFFLOADED; if (rtm->rtm_flags & RTM_F_OFFLOAD_FAILED) flags |= ZEBRA_FLAG_OFFLOAD_FAILED; if (h->nlmsg_flags & NLM_F_APPEND) flags |= ZEBRA_FLAG_OUTOFSYNC; /* Route which inserted by Zebra. */ if (selfroute) { flags |= ZEBRA_FLAG_SELFROUTE; proto = proto2zebra(rtm->rtm_protocol, rtm->rtm_family, false); } if (tb[RTA_OIF]) index = *(int *)RTA_DATA(tb[RTA_OIF]); if (tb[RTA_DST]) dest = RTA_DATA(tb[RTA_DST]); else dest = anyaddr; if (tb[RTA_SRC]) src = RTA_DATA(tb[RTA_SRC]); else src = anyaddr; if (tb[RTA_PREFSRC]) prefsrc = RTA_DATA(tb[RTA_PREFSRC]); if (tb[RTA_GATEWAY]) gate = RTA_DATA(tb[RTA_GATEWAY]); if (tb[RTA_NH_ID]) nhe_id = *(uint32_t *)RTA_DATA(tb[RTA_NH_ID]); if (tb[RTA_PRIORITY]) metric = *(int *)RTA_DATA(tb[RTA_PRIORITY]); #if defined(SUPPORT_REALMS) if (tb[RTA_FLOW]) tag = *(uint32_t *)RTA_DATA(tb[RTA_FLOW]); #endif if (tb[RTA_METRICS]) { struct rtattr *mxrta[RTAX_MAX + 1]; netlink_parse_rtattr(mxrta, RTAX_MAX, RTA_DATA(tb[RTA_METRICS]), RTA_PAYLOAD(tb[RTA_METRICS])); if (mxrta[RTAX_MTU]) mtu = *(uint32_t *)RTA_DATA(mxrta[RTAX_MTU]); } if (rtm->rtm_family == AF_INET) { p.family = AF_INET; if (rtm->rtm_dst_len > IPV4_MAX_BITLEN) { zlog_err( "Invalid destination prefix length: %u received from kernel route change", rtm->rtm_dst_len); return -1; } memcpy(&p.u.prefix4, dest, 4); p.prefixlen = rtm->rtm_dst_len; if (rtm->rtm_src_len != 0) { flog_warn( EC_ZEBRA_UNSUPPORTED_V4_SRCDEST, "unsupported IPv4 sourcedest route (dest %pFX vrf %u)", &p, vrf_id); return 0; } /* Force debug below to not display anything for source */ src_p.prefixlen = 0; } else if (rtm->rtm_family == AF_INET6) { p.family = AF_INET6; if (rtm->rtm_dst_len > IPV6_MAX_BITLEN) { zlog_err( "Invalid destination prefix length: %u received from kernel route change", rtm->rtm_dst_len); return -1; } memcpy(&p.u.prefix6, dest, 16); p.prefixlen = rtm->rtm_dst_len; src_p.family = AF_INET6; if (rtm->rtm_src_len > IPV6_MAX_BITLEN) { zlog_err( "Invalid source prefix length: %u received from kernel route change", rtm->rtm_src_len); return -1; } memcpy(&src_p.prefix, src, 16); src_p.prefixlen = rtm->rtm_src_len; } else { /* We only handle the AFs we handle... */ if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: unknown address-family %u", __func__, rtm->rtm_family); return 0; } /* * For ZEBRA_ROUTE_KERNEL types: * * The metric/priority of the route received from the kernel * is a 32 bit number. We are going to interpret the high * order byte as the Admin Distance and the low order 3 bytes * as the metric. * * This will allow us to do two things: * 1) Allow the creation of kernel routes that can be * overridden by zebra. * 2) Allow the old behavior for 'most' kernel route types * if a user enters 'ip route ...' v4 routes get a metric * of 0 and v6 routes get a metric of 1024. Both of these * values will end up with a admin distance of 0, which * will cause them to win for the purposes of zebra. */ if (proto == ZEBRA_ROUTE_KERNEL) { distance = (metric >> 24) & 0xFF; metric = (metric & 0x00FFFFFF); } if (IS_ZEBRA_DEBUG_KERNEL) { char buf2[PREFIX_STRLEN]; zlog_debug( "%s %pFX%s%s vrf %s(%u) table_id: %u metric: %d Admin Distance: %d", nl_msg_type_to_str(h->nlmsg_type), &p, src_p.prefixlen ? " from " : "", src_p.prefixlen ? prefix2str(&src_p, buf2, sizeof(buf2)) : "", vrf_id_to_name(vrf_id), vrf_id, table, metric, distance); } afi_t afi = AFI_IP; if (rtm->rtm_family == AF_INET6) afi = AFI_IP6; if (h->nlmsg_type == RTM_NEWROUTE) { struct route_entry *re; struct nexthop_group *ng = NULL; re = zebra_rib_route_entry_new(vrf_id, proto, 0, flags, nhe_id, table, metric, mtu, distance, tag); if (!nhe_id) ng = nexthop_group_new(); if (!tb[RTA_MULTIPATH]) { struct nexthop *nexthop, nh; if (!nhe_id) { nh = parse_nexthop_unicast( ns_id, rtm, tb, bh_type, index, prefsrc, gate, afi, vrf_id); nexthop = nexthop_new(); *nexthop = nh; nexthop_group_add_sorted(ng, nexthop); } } else { /* This is a multipath route */ struct rtnexthop *rtnh = (struct rtnexthop *)RTA_DATA(tb[RTA_MULTIPATH]); if (!nhe_id) { uint8_t nhop_num; /* Use temporary list of nexthops; parse * message payload's nexthops. */ nhop_num = parse_multipath_nexthops_unicast( ns_id, ng, rtm, rtnh, tb, prefsrc, vrf_id); zserv_nexthop_num_warn( __func__, (const struct prefix *)&p, nhop_num); if (nhop_num == 0) { nexthop_group_delete(&ng); ng = NULL; } } } if (nhe_id || ng) { dplane_rib_add_multipath(afi, SAFI_UNICAST, &p, &src_p, re, ng, startup, ctx); if (ng) nexthop_group_delete(&ng); } else { /* * I really don't see how this is possible * but since we are testing for it let's * let the end user know why the route * that was just received was swallowed * up and forgotten */ zlog_err( "%s: %pFX multipath RTM_NEWROUTE has a invalid nexthop group from the kernel", __func__, &p); XFREE(MTYPE_RE, re); } } else { if (ctx) { zlog_err( "%s: %pFX RTM_DELROUTE received but received a context as well", __func__, &p); return 0; } if (nhe_id) { rib_delete(afi, SAFI_UNICAST, vrf_id, proto, 0, flags, &p, &src_p, NULL, nhe_id, table, metric, distance, true); } else { if (!tb[RTA_MULTIPATH]) { struct nexthop nh; nh = parse_nexthop_unicast( ns_id, rtm, tb, bh_type, index, prefsrc, gate, afi, vrf_id); rib_delete(afi, SAFI_UNICAST, vrf_id, proto, 0, flags, &p, &src_p, &nh, 0, table, metric, distance, true); } else { /* XXX: need to compare the entire list of * nexthops here for NLM_F_APPEND stupidity */ rib_delete(afi, SAFI_UNICAST, vrf_id, proto, 0, flags, &p, &src_p, NULL, 0, table, metric, distance, true); } } } return 1; } static int netlink_route_change_read_unicast(struct nlmsghdr *h, ns_id_t ns_id, int startup) { return netlink_route_change_read_unicast_internal(h, ns_id, startup, NULL); } static struct mcast_route_data *mroute = NULL; static int netlink_route_change_read_multicast(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct rtmsg *rtm; struct rtattr *tb[RTA_MAX + 1]; struct mcast_route_data *m; int iif = 0; int count; int oif[256]; int oif_count = 0; char oif_list[256] = "\0"; vrf_id_t vrf; int table; assert(mroute); m = mroute; rtm = NLMSG_DATA(h); len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct rtmsg)); netlink_parse_rtattr(tb, RTA_MAX, RTM_RTA(rtm), len); if (tb[RTA_TABLE]) table = *(int *)RTA_DATA(tb[RTA_TABLE]); else table = rtm->rtm_table; vrf = vrf_lookup_by_table(table, ns_id); if (tb[RTA_IIF]) iif = *(int *)RTA_DATA(tb[RTA_IIF]); if (tb[RTA_SRC]) { if (rtm->rtm_family == RTNL_FAMILY_IPMR) m->src.ipaddr_v4 = *(struct in_addr *)RTA_DATA(tb[RTA_SRC]); else m->src.ipaddr_v6 = *(struct in6_addr *)RTA_DATA(tb[RTA_SRC]); } if (tb[RTA_DST]) { if (rtm->rtm_family == RTNL_FAMILY_IPMR) m->grp.ipaddr_v4 = *(struct in_addr *)RTA_DATA(tb[RTA_DST]); else m->grp.ipaddr_v6 = *(struct in6_addr *)RTA_DATA(tb[RTA_DST]); } if (tb[RTA_EXPIRES]) m->lastused = *(unsigned long long *)RTA_DATA(tb[RTA_EXPIRES]); if (tb[RTA_MULTIPATH]) { struct rtnexthop *rtnh = (struct rtnexthop *)RTA_DATA(tb[RTA_MULTIPATH]); len = RTA_PAYLOAD(tb[RTA_MULTIPATH]); for (;;) { if (len < (int)sizeof(*rtnh) || rtnh->rtnh_len > len) break; oif[oif_count] = rtnh->rtnh_ifindex; oif_count++; if (rtnh->rtnh_len == 0) break; len -= NLMSG_ALIGN(rtnh->rtnh_len); rtnh = RTNH_NEXT(rtnh); } } if (rtm->rtm_family == RTNL_FAMILY_IPMR) { SET_IPADDR_V4(&m->src); SET_IPADDR_V4(&m->grp); } else if (rtm->rtm_family == RTNL_FAMILY_IP6MR) { SET_IPADDR_V6(&m->src); SET_IPADDR_V6(&m->grp); } else { zlog_warn("%s: Invalid rtm_family received", __func__); return 0; } if (IS_ZEBRA_DEBUG_KERNEL) { struct interface *ifp = NULL; struct zebra_vrf *zvrf = NULL; for (count = 0; count < oif_count; count++) { ifp = if_lookup_by_index(oif[count], vrf); char temp[256]; snprintf(temp, sizeof(temp), "%s(%d) ", ifp ? ifp->name : "Unknown", oif[count]); strlcat(oif_list, temp, sizeof(oif_list)); } zvrf = zebra_vrf_lookup_by_id(vrf); ifp = if_lookup_by_index(iif, vrf); zlog_debug( "MCAST VRF: %s(%d) %s (%pIA,%pIA) IIF: %s(%d) OIF: %s jiffies: %lld", zvrf_name(zvrf), vrf, nl_msg_type_to_str(h->nlmsg_type), &m->src, &m->grp, ifp ? ifp->name : "Unknown", iif, oif_list, m->lastused); } return 0; } int netlink_route_change(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct rtmsg *rtm; rtm = NLMSG_DATA(h); if (!(h->nlmsg_type == RTM_NEWROUTE || h->nlmsg_type == RTM_DELROUTE)) { /* If this is not route add/delete message print warning. */ zlog_debug("Kernel message: %s NS %u", nl_msg_type_to_str(h->nlmsg_type), ns_id); return 0; } switch (rtm->rtm_family) { case AF_INET: case AF_INET6: break; case RTNL_FAMILY_IPMR: case RTNL_FAMILY_IP6MR: /* notifications on IPMR are irrelevant to zebra, we only care * about responses to RTM_GETROUTE requests we sent. */ return 0; default: flog_warn( EC_ZEBRA_UNKNOWN_FAMILY, "Invalid address family: %u received from kernel route change: %s", rtm->rtm_family, nl_msg_type_to_str(h->nlmsg_type)); return 0; } /* Connected route. */ if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s %s %s proto %s NS %u", nl_msg_type_to_str(h->nlmsg_type), nl_family_to_str(rtm->rtm_family), nl_rttype_to_str(rtm->rtm_type), nl_rtproto_to_str(rtm->rtm_protocol), ns_id); len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct rtmsg)); if (len < 0) { zlog_err( "%s: Message received from netlink is of a broken size: %d %zu", __func__, h->nlmsg_len, (size_t)NLMSG_LENGTH(sizeof(struct rtmsg))); return -1; } /* these are "magic" kernel-managed *unicast* routes used for * outputting locally generated multicast traffic (which uses unicast * handling on Linux because ~reasons~. */ if (rtm->rtm_type == RTN_MULTICAST) return 0; netlink_route_change_read_unicast(h, ns_id, startup); return 0; } /* Request for specific route information from the kernel */ static int netlink_request_route(struct zebra_ns *zns, int family, int type) { struct { struct nlmsghdr n; struct rtmsg rtm; } req; /* Form the request, specifying filter (rtattr) if needed. */ memset(&req, 0, sizeof(req)); req.n.nlmsg_type = type; req.n.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST; req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg)); req.rtm.rtm_family = family; return netlink_request(&zns->netlink_cmd, &req); } /* Routing table read function using netlink interface. Only called bootstrap time. */ int netlink_route_read(struct zebra_ns *zns) { int ret; struct zebra_dplane_info dp_info; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Get IPv4 routing table. */ ret = netlink_request_route(zns, AF_INET, RTM_GETROUTE); if (ret < 0) return ret; ret = netlink_parse_info(netlink_route_change_read_unicast, &zns->netlink_cmd, &dp_info, 0, true); if (ret < 0) return ret; /* Get IPv6 routing table. */ ret = netlink_request_route(zns, AF_INET6, RTM_GETROUTE); if (ret < 0) return ret; ret = netlink_parse_info(netlink_route_change_read_unicast, &zns->netlink_cmd, &dp_info, 0, true); if (ret < 0) return ret; return 0; } /* * The function returns true if the gateway info could be added * to the message, otherwise false is returned. */ static bool _netlink_route_add_gateway_info(uint8_t route_family, uint8_t gw_family, struct nlmsghdr *nlmsg, size_t req_size, int bytelen, const struct nexthop *nexthop) { if (route_family == AF_MPLS) { struct gw_family_t gw_fam; gw_fam.family = gw_family; if (gw_family == AF_INET) memcpy(&gw_fam.gate.ipv4, &nexthop->gate.ipv4, bytelen); else memcpy(&gw_fam.gate.ipv6, &nexthop->gate.ipv6, bytelen); if (!nl_attr_put(nlmsg, req_size, RTA_VIA, &gw_fam.family, bytelen + 2)) return false; } else { if (!(nexthop->rparent && IS_MAPPED_IPV6(&nexthop->rparent->gate.ipv6))) { if (gw_family == AF_INET) { if (!nl_attr_put(nlmsg, req_size, RTA_GATEWAY, &nexthop->gate.ipv4, bytelen)) return false; } else { if (!nl_attr_put(nlmsg, req_size, RTA_GATEWAY, &nexthop->gate.ipv6, bytelen)) return false; } } } return true; } static int build_label_stack(struct mpls_label_stack *nh_label, enum lsp_types_t nh_label_type, mpls_lse_t *out_lse, char *label_buf, size_t label_buf_size) { char label_buf1[20]; int num_labels = 0; for (int i = 0; nh_label && i < nh_label->num_labels; i++) { if (nh_label_type != ZEBRA_LSP_EVPN && nh_label->label[i] == MPLS_LABEL_IMPLICIT_NULL) continue; if (IS_ZEBRA_DEBUG_KERNEL) { if (!num_labels) snprintf(label_buf, label_buf_size, "label %u", nh_label->label[i]); else { snprintf(label_buf1, sizeof(label_buf1), "/%u", nh_label->label[i]); strlcat(label_buf, label_buf1, label_buf_size); } } if (nh_label_type == ZEBRA_LSP_EVPN) out_lse[num_labels] = label2vni(&nh_label->label[i]); else out_lse[num_labels] = mpls_lse_encode(nh_label->label[i], 0, 0, 0); num_labels++; } return num_labels; } static bool _netlink_nexthop_encode_dvni_label(const struct nexthop *nexthop, struct nlmsghdr *nlmsg, mpls_lse_t *out_lse, size_t buflen, char *label_buf) { struct in_addr ipv4; if (!nl_attr_put64(nlmsg, buflen, LWTUNNEL_IP_ID, htonll((uint64_t)out_lse[0]))) return false; if (nexthop->type == NEXTHOP_TYPE_IPV4_IFINDEX) { if (!nl_attr_put(nlmsg, buflen, LWTUNNEL_IP_DST, &nexthop->gate.ipv4, 4)) return false; } else if (nexthop->type == NEXTHOP_TYPE_IPV6_IFINDEX) { if (IS_MAPPED_IPV6(&nexthop->gate.ipv6)) { ipv4_mapped_ipv6_to_ipv4(&nexthop->gate.ipv6, &ipv4); if (!nl_attr_put(nlmsg, buflen, LWTUNNEL_IP_DST, &ipv4, 4)) return false; } else { if (!nl_attr_put(nlmsg, buflen, LWTUNNEL_IP_DST, &nexthop->gate.ipv6, 16)) return false; } } else { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "%s: nexthop %pNHv %s must NEXTHOP_TYPE_IPV*_IFINDEX to be vxlan encapped", __func__, nexthop, label_buf); return false; } return true; } static bool _netlink_route_encode_label_info(const struct nexthop *nexthop, struct nlmsghdr *nlmsg, size_t buflen, struct rtmsg *rtmsg, char *label_buf, size_t label_buf_size) { mpls_lse_t out_lse[MPLS_MAX_LABELS]; int num_labels; struct rtattr *nest; struct mpls_label_stack *nh_label; enum lsp_types_t nh_label_type; nh_label = nexthop->nh_label; nh_label_type = nexthop->nh_label_type; /* * label_buf is *only* currently used within debugging. * As such when we assign it we are guarding it inside * a debug test. If you want to change this make sure * you fix this assumption */ label_buf[0] = '\0'; num_labels = build_label_stack(nh_label, nh_label_type, out_lse, label_buf, label_buf_size); if (num_labels && nh_label_type == ZEBRA_LSP_EVPN) { if (!nl_attr_put16(nlmsg, buflen, RTA_ENCAP_TYPE, LWTUNNEL_ENCAP_IP)) return false; nest = nl_attr_nest(nlmsg, buflen, RTA_ENCAP); if (!nest) return false; if (_netlink_nexthop_encode_dvni_label(nexthop, nlmsg, out_lse, buflen, label_buf) == false) return false; nl_attr_nest_end(nlmsg, nest); } else if (num_labels) { /* Set the BoS bit */ out_lse[num_labels - 1] |= htonl(1 << MPLS_LS_S_SHIFT); if (rtmsg->rtm_family == AF_MPLS) { if (!nl_attr_put(nlmsg, buflen, RTA_NEWDST, &out_lse, num_labels * sizeof(mpls_lse_t))) return false; } else { if (!nl_attr_put16(nlmsg, buflen, RTA_ENCAP_TYPE, LWTUNNEL_ENCAP_MPLS)) return false; nest = nl_attr_nest(nlmsg, buflen, RTA_ENCAP); if (!nest) return false; if (!nl_attr_put(nlmsg, buflen, MPLS_IPTUNNEL_DST, &out_lse, num_labels * sizeof(mpls_lse_t))) return false; nl_attr_nest_end(nlmsg, nest); } } return true; } static bool _netlink_route_encode_nexthop_src(const struct nexthop *nexthop, int family, struct nlmsghdr *nlmsg, size_t buflen, int bytelen) { if (family == AF_INET) { if (nexthop->rmap_src.ipv4.s_addr != INADDR_ANY) { if (!nl_attr_put(nlmsg, buflen, RTA_PREFSRC, &nexthop->rmap_src.ipv4, bytelen)) return false; } else if (nexthop->src.ipv4.s_addr != INADDR_ANY) { if (!nl_attr_put(nlmsg, buflen, RTA_PREFSRC, &nexthop->src.ipv4, bytelen)) return false; } } else if (family == AF_INET6) { if (!IN6_IS_ADDR_UNSPECIFIED(&nexthop->rmap_src.ipv6)) { if (!nl_attr_put(nlmsg, buflen, RTA_PREFSRC, &nexthop->rmap_src.ipv6, bytelen)) return false; } else if (!IN6_IS_ADDR_UNSPECIFIED(&nexthop->src.ipv6)) { if (!nl_attr_put(nlmsg, buflen, RTA_PREFSRC, &nexthop->src.ipv6, bytelen)) return false; } } return true; } static ssize_t fill_seg6ipt_encap(char *buffer, size_t buflen, const struct in6_addr *seg) { struct seg6_iptunnel_encap *ipt; struct ipv6_sr_hdr *srh; const size_t srhlen = 24; /* * Caution: Support only SINGLE-SID, not MULTI-SID * This function only supports the case where segs represents * a single SID. If you want to extend the SRv6 functionality, * you should improve the Boundary Check. * Ex. In case of set a SID-List include multiple-SIDs as an * argument of the Transit Behavior, we must support variable * boundary check for buflen. */ if (buflen < (sizeof(struct seg6_iptunnel_encap) + sizeof(struct ipv6_sr_hdr) + 16)) return -1; memset(buffer, 0, buflen); ipt = (struct seg6_iptunnel_encap *)buffer; ipt->mode = SEG6_IPTUN_MODE_ENCAP; srh = ipt->srh; srh->hdrlen = (srhlen >> 3) - 1; srh->type = 4; srh->segments_left = 0; srh->first_segment = 0; memcpy(&srh->segments[0], seg, sizeof(struct in6_addr)); return srhlen + 4; } /* This function takes a nexthop as argument and adds * the appropriate netlink attributes to an existing * netlink message. * * @param routedesc: Human readable description of route type * (direct/recursive, single-/multipath) * @param bytelen: Length of addresses in bytes. * @param nexthop: Nexthop information * @param nlmsg: nlmsghdr structure to fill in. * @param req_size: The size allocated for the message. * * The function returns true if the nexthop could be added * to the message, otherwise false is returned. */ static bool _netlink_route_build_singlepath(const struct prefix *p, const char *routedesc, int bytelen, const struct nexthop *nexthop, struct nlmsghdr *nlmsg, struct rtmsg *rtmsg, size_t req_size, int cmd) { char label_buf[256]; struct vrf *vrf; char addrstr[INET6_ADDRSTRLEN]; assert(nexthop); vrf = vrf_lookup_by_id(nexthop->vrf_id); if (!_netlink_route_encode_label_info(nexthop, nlmsg, req_size, rtmsg, label_buf, sizeof(label_buf))) return false; if (nexthop->nh_srv6) { if (nexthop->nh_srv6->seg6local_action != ZEBRA_SEG6_LOCAL_ACTION_UNSPEC) { struct rtattr *nest; const struct seg6local_context *ctx; ctx = &nexthop->nh_srv6->seg6local_ctx; if (!nl_attr_put16(nlmsg, req_size, RTA_ENCAP_TYPE, LWTUNNEL_ENCAP_SEG6_LOCAL)) return false; nest = nl_attr_nest(nlmsg, req_size, RTA_ENCAP); if (!nest) return false; switch (nexthop->nh_srv6->seg6local_action) { case ZEBRA_SEG6_LOCAL_ACTION_END: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END)) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_X: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_X)) return false; if (!nl_attr_put(nlmsg, req_size, SEG6_LOCAL_NH6, &ctx->nh6, sizeof(struct in6_addr))) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_T: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_T)) return false; if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_TABLE, ctx->table)) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_DX4: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DX4)) return false; if (!nl_attr_put(nlmsg, req_size, SEG6_LOCAL_NH4, &ctx->nh4, sizeof(struct in_addr))) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_DT6: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DT6)) return false; if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_TABLE, ctx->table)) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_DT4: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DT4)) return false; if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_VRFTABLE, ctx->table)) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_DT46: if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DT46)) return false; if (!nl_attr_put32(nlmsg, req_size, SEG6_LOCAL_VRFTABLE, ctx->table)) return false; break; case ZEBRA_SEG6_LOCAL_ACTION_END_DX2: case ZEBRA_SEG6_LOCAL_ACTION_END_DX6: case ZEBRA_SEG6_LOCAL_ACTION_END_B6: case ZEBRA_SEG6_LOCAL_ACTION_END_B6_ENCAP: case ZEBRA_SEG6_LOCAL_ACTION_END_BM: case ZEBRA_SEG6_LOCAL_ACTION_END_S: case ZEBRA_SEG6_LOCAL_ACTION_END_AS: case ZEBRA_SEG6_LOCAL_ACTION_END_AM: case ZEBRA_SEG6_LOCAL_ACTION_END_BPF: case ZEBRA_SEG6_LOCAL_ACTION_UNSPEC: zlog_err("%s: unsupport seg6local behaviour action=%u", __func__, nexthop->nh_srv6->seg6local_action); return false; } nl_attr_nest_end(nlmsg, nest); } if (!sid_zero(&nexthop->nh_srv6->seg6_segs)) { char tun_buf[4096]; ssize_t tun_len; struct rtattr *nest; if (!nl_attr_put16(nlmsg, req_size, RTA_ENCAP_TYPE, LWTUNNEL_ENCAP_SEG6)) return false; nest = nl_attr_nest(nlmsg, req_size, RTA_ENCAP); if (!nest) return false; tun_len = fill_seg6ipt_encap(tun_buf, sizeof(tun_buf), &nexthop->nh_srv6->seg6_segs); if (tun_len < 0) return false; if (!nl_attr_put(nlmsg, req_size, SEG6_IPTUNNEL_SRH, tun_buf, tun_len)) return false; nl_attr_nest_end(nlmsg, nest); } } if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ONLINK)) rtmsg->rtm_flags |= RTNH_F_ONLINK; if (is_route_v4_over_v6(rtmsg->rtm_family, nexthop->type)) { rtmsg->rtm_flags |= RTNH_F_ONLINK; if (!nl_attr_put(nlmsg, req_size, RTA_GATEWAY, &ipv4_ll, 4)) return false; if (!nl_attr_put32(nlmsg, req_size, RTA_OIF, nexthop->ifindex)) return false; if (cmd == RTM_NEWROUTE) { if (!_netlink_route_encode_nexthop_src( nexthop, AF_INET, nlmsg, req_size, bytelen)) return false; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: 5549 (%s): %pFX nexthop via %s %s if %u vrf %s(%u)", __func__, routedesc, p, ipv4_ll_buf, label_buf, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); return true; } if (nexthop->type == NEXTHOP_TYPE_IPV4 || nexthop->type == NEXTHOP_TYPE_IPV4_IFINDEX) { /* Send deletes to the kernel without specifying the next-hop */ if (cmd != RTM_DELROUTE) { if (!_netlink_route_add_gateway_info( rtmsg->rtm_family, AF_INET, nlmsg, req_size, bytelen, nexthop)) return false; } if (cmd == RTM_NEWROUTE) { if (!_netlink_route_encode_nexthop_src( nexthop, AF_INET, nlmsg, req_size, bytelen)) return false; } if (IS_ZEBRA_DEBUG_KERNEL) { inet_ntop(AF_INET, &nexthop->gate.ipv4, addrstr, sizeof(addrstr)); zlog_debug("%s: (%s): %pFX nexthop via %s %s if %u vrf %s(%u)", __func__, routedesc, p, addrstr, label_buf, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); } } if (nexthop->type == NEXTHOP_TYPE_IPV6 || nexthop->type == NEXTHOP_TYPE_IPV6_IFINDEX) { if (!_netlink_route_add_gateway_info(rtmsg->rtm_family, AF_INET6, nlmsg, req_size, bytelen, nexthop)) return false; if (cmd == RTM_NEWROUTE) { if (!_netlink_route_encode_nexthop_src( nexthop, AF_INET6, nlmsg, req_size, bytelen)) return false; } if (IS_ZEBRA_DEBUG_KERNEL) { inet_ntop(AF_INET6, &nexthop->gate.ipv6, addrstr, sizeof(addrstr)); zlog_debug("%s: (%s): %pFX nexthop via %s %s if %u vrf %s(%u)", __func__, routedesc, p, addrstr, label_buf, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); } } /* * We have the ifindex so we should always send it * This is especially useful if we are doing route * leaking. */ if (nexthop->type != NEXTHOP_TYPE_BLACKHOLE) { if (!nl_attr_put32(nlmsg, req_size, RTA_OIF, nexthop->ifindex)) return false; } if (nexthop->type == NEXTHOP_TYPE_IFINDEX) { if (cmd == RTM_NEWROUTE) { if (!_netlink_route_encode_nexthop_src( nexthop, AF_INET, nlmsg, req_size, bytelen)) return false; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: (%s): %pFX nexthop via if %u vrf %s(%u)", __func__, routedesc, p, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); } return true; } /* This function appends tag value as rtnl flow attribute * to the given netlink msg only if value is less than 256. * Used only if SUPPORT_REALMS enabled. * * @param nlmsg: nlmsghdr structure to fill in. * @param maxlen: The size allocated for the message. * @param tag: The route tag. * * The function returns true if the flow attribute could * be added to the message, otherwise false is returned. */ static inline bool _netlink_set_tag(struct nlmsghdr *n, unsigned int maxlen, route_tag_t tag) { if (tag > 0 && tag <= 255) { if (!nl_attr_put32(n, maxlen, RTA_FLOW, tag)) return false; } return true; } /* This function takes a nexthop as argument and * appends to the given netlink msg. If the nexthop * defines a preferred source, the src parameter * will be modified to point to that src, otherwise * it will be kept unmodified. * * @param routedesc: Human readable description of route type * (direct/recursive, single-/multipath) * @param bytelen: Length of addresses in bytes. * @param nexthop: Nexthop information * @param nlmsg: nlmsghdr structure to fill in. * @param req_size: The size allocated for the message. * @param src: pointer pointing to a location where * the prefsrc should be stored. * * The function returns true if the nexthop could be added * to the message, otherwise false is returned. */ static bool _netlink_route_build_multipath( const struct prefix *p, const char *routedesc, int bytelen, const struct nexthop *nexthop, struct nlmsghdr *nlmsg, size_t req_size, struct rtmsg *rtmsg, const union g_addr **src, route_tag_t tag) { char label_buf[256]; struct vrf *vrf; struct rtnexthop *rtnh; rtnh = nl_attr_rtnh(nlmsg, req_size); if (rtnh == NULL) return false; assert(nexthop); vrf = vrf_lookup_by_id(nexthop->vrf_id); if (!_netlink_route_encode_label_info(nexthop, nlmsg, req_size, rtmsg, label_buf, sizeof(label_buf))) return false; if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ONLINK)) rtnh->rtnh_flags |= RTNH_F_ONLINK; if (is_route_v4_over_v6(rtmsg->rtm_family, nexthop->type)) { rtnh->rtnh_flags |= RTNH_F_ONLINK; if (!nl_attr_put(nlmsg, req_size, RTA_GATEWAY, &ipv4_ll, 4)) return false; rtnh->rtnh_ifindex = nexthop->ifindex; if (nexthop->weight) rtnh->rtnh_hops = nexthop->weight - 1; if (nexthop->rmap_src.ipv4.s_addr != INADDR_ANY) *src = &nexthop->rmap_src; else if (nexthop->src.ipv4.s_addr != INADDR_ANY) *src = &nexthop->src; if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "%s: 5549 (%s): %pFX nexthop via %s %s if %u vrf %s(%u)", __func__, routedesc, p, ipv4_ll_buf, label_buf, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); nl_attr_rtnh_end(nlmsg, rtnh); return true; } if (nexthop->type == NEXTHOP_TYPE_IPV4 || nexthop->type == NEXTHOP_TYPE_IPV4_IFINDEX) { if (!_netlink_route_add_gateway_info(rtmsg->rtm_family, AF_INET, nlmsg, req_size, bytelen, nexthop)) return false; if (nexthop->rmap_src.ipv4.s_addr != INADDR_ANY) *src = &nexthop->rmap_src; else if (nexthop->src.ipv4.s_addr != INADDR_ANY) *src = &nexthop->src; if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: (%s): %pFX nexthop via %pI4 %s if %u vrf %s(%u)", __func__, routedesc, p, &nexthop->gate.ipv4, label_buf, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); } if (nexthop->type == NEXTHOP_TYPE_IPV6 || nexthop->type == NEXTHOP_TYPE_IPV6_IFINDEX) { if (!_netlink_route_add_gateway_info(rtmsg->rtm_family, AF_INET6, nlmsg, req_size, bytelen, nexthop)) return false; if (!IN6_IS_ADDR_UNSPECIFIED(&nexthop->rmap_src.ipv6)) *src = &nexthop->rmap_src; else if (!IN6_IS_ADDR_UNSPECIFIED(&nexthop->src.ipv6)) *src = &nexthop->src; if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: (%s): %pFX nexthop via %pI6 %s if %u vrf %s(%u)", __func__, routedesc, p, &nexthop->gate.ipv6, label_buf, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); } /* * We have figured out the ifindex so we should always send it * This is especially useful if we are doing route * leaking. */ if (nexthop->type != NEXTHOP_TYPE_BLACKHOLE) rtnh->rtnh_ifindex = nexthop->ifindex; /* ifindex */ if (nexthop->type == NEXTHOP_TYPE_IFINDEX) { if (nexthop->rmap_src.ipv4.s_addr != INADDR_ANY) *src = &nexthop->rmap_src; else if (nexthop->src.ipv4.s_addr != INADDR_ANY) *src = &nexthop->src; if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: (%s): %pFX nexthop via if %u vrf %s(%u)", __func__, routedesc, p, nexthop->ifindex, VRF_LOGNAME(vrf), nexthop->vrf_id); } if (nexthop->weight) rtnh->rtnh_hops = nexthop->weight - 1; if (!_netlink_set_tag(nlmsg, req_size, tag)) return false; nl_attr_rtnh_end(nlmsg, rtnh); return true; } static inline bool _netlink_mpls_build_singlepath(const struct prefix *p, const char *routedesc, const struct zebra_nhlfe *nhlfe, struct nlmsghdr *nlmsg, struct rtmsg *rtmsg, size_t req_size, int cmd) { int bytelen; uint8_t family; family = NHLFE_FAMILY(nhlfe); bytelen = (family == AF_INET ? 4 : 16); return _netlink_route_build_singlepath(p, routedesc, bytelen, nhlfe->nexthop, nlmsg, rtmsg, req_size, cmd); } static inline bool _netlink_mpls_build_multipath(const struct prefix *p, const char *routedesc, const struct zebra_nhlfe *nhlfe, struct nlmsghdr *nlmsg, size_t req_size, struct rtmsg *rtmsg, const union g_addr **src) { int bytelen; uint8_t family; family = NHLFE_FAMILY(nhlfe); bytelen = (family == AF_INET ? 4 : 16); return _netlink_route_build_multipath(p, routedesc, bytelen, nhlfe->nexthop, nlmsg, req_size, rtmsg, src, 0); } static void _netlink_mpls_debug(int cmd, uint32_t label, const char *routedesc) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("netlink_mpls_multipath_msg_encode() (%s): %s %u/20", routedesc, nl_msg_type_to_str(cmd), label); } static int netlink_neigh_update(int cmd, int ifindex, void *addr, char *lla, int llalen, ns_id_t ns_id, uint8_t family, bool permanent, uint8_t protocol) { struct { struct nlmsghdr n; struct ndmsg ndm; char buf[256]; } req; struct zebra_ns *zns = zebra_ns_lookup(ns_id); memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ndmsg)); req.n.nlmsg_flags = NLM_F_CREATE | NLM_F_REQUEST; req.n.nlmsg_type = cmd; // RTM_NEWNEIGH or RTM_DELNEIGH req.n.nlmsg_pid = zns->netlink_cmd.snl.nl_pid; req.ndm.ndm_family = family; req.ndm.ndm_ifindex = ifindex; req.ndm.ndm_type = RTN_UNICAST; if (cmd == RTM_NEWNEIGH) { if (!permanent) req.ndm.ndm_state = NUD_REACHABLE; else req.ndm.ndm_state = NUD_PERMANENT; } else req.ndm.ndm_state = NUD_FAILED; nl_attr_put(&req.n, sizeof(req), NDA_PROTOCOL, &protocol, sizeof(protocol)); req.ndm.ndm_type = RTN_UNICAST; nl_attr_put(&req.n, sizeof(req), NDA_DST, addr, family2addrsize(family)); if (lla) nl_attr_put(&req.n, sizeof(req), NDA_LLADDR, lla, llalen); return netlink_talk(netlink_talk_filter, &req.n, &zns->netlink_cmd, zns, false); } static bool nexthop_set_src(const struct nexthop *nexthop, int family, union g_addr *src) { if (family == AF_INET) { if (nexthop->rmap_src.ipv4.s_addr != INADDR_ANY) { src->ipv4 = nexthop->rmap_src.ipv4; return true; } else if (nexthop->src.ipv4.s_addr != INADDR_ANY) { src->ipv4 = nexthop->src.ipv4; return true; } } else if (family == AF_INET6) { if (!IN6_IS_ADDR_UNSPECIFIED(&nexthop->rmap_src.ipv6)) { src->ipv6 = nexthop->rmap_src.ipv6; return true; } else if (!IN6_IS_ADDR_UNSPECIFIED(&nexthop->src.ipv6)) { src->ipv6 = nexthop->src.ipv6; return true; } } return false; } /* * The function returns true if the attribute could be added * to the message, otherwise false is returned. */ static int netlink_route_nexthop_encap(struct nlmsghdr *n, size_t nlen, struct nexthop *nh) { struct rtattr *nest; switch (nh->nh_encap_type) { case NET_VXLAN: if (!nl_attr_put16(n, nlen, RTA_ENCAP_TYPE, nh->nh_encap_type)) return false; nest = nl_attr_nest(n, nlen, RTA_ENCAP); if (!nest) return false; if (!nl_attr_put32(n, nlen, 0 /* VXLAN_VNI */, nh->nh_encap.vni)) return false; nl_attr_nest_end(n, nest); break; } return true; } /* * Routing table change via netlink interface, using a dataplane context object * * Returns -1 on failure, 0 when the msg doesn't fit entirely in the buffer * otherwise the number of bytes written to buf. */ ssize_t netlink_route_multipath_msg_encode(int cmd, struct zebra_dplane_ctx *ctx, uint8_t *data, size_t datalen, bool fpm, bool force_nhg) { int bytelen; struct nexthop *nexthop = NULL; unsigned int nexthop_num; const char *routedesc; bool setsrc = false; union g_addr src; const struct prefix *p, *src_p; uint32_t table_id; struct nlsock *nl; route_tag_t tag = 0; struct { struct nlmsghdr n; struct rtmsg r; char buf[]; } *req = (void *)data; p = dplane_ctx_get_dest(ctx); src_p = dplane_ctx_get_src(ctx); if (datalen < sizeof(*req)) return 0; nl = kernel_netlink_nlsock_lookup(dplane_ctx_get_ns_sock(ctx)); memset(req, 0, sizeof(*req)); bytelen = (p->family == AF_INET ? 4 : 16); req->n.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg)); req->n.nlmsg_flags = NLM_F_CREATE | NLM_F_REQUEST; if ((cmd == RTM_NEWROUTE) && ((p->family == AF_INET) || v6_rr_semantics)) req->n.nlmsg_flags |= NLM_F_REPLACE; req->n.nlmsg_type = cmd; req->n.nlmsg_pid = nl->snl.nl_pid; req->r.rtm_family = p->family; req->r.rtm_dst_len = p->prefixlen; req->r.rtm_src_len = src_p ? src_p->prefixlen : 0; req->r.rtm_scope = RT_SCOPE_UNIVERSE; if (cmd == RTM_DELROUTE) req->r.rtm_protocol = zebra2proto(dplane_ctx_get_old_type(ctx)); else req->r.rtm_protocol = zebra2proto(dplane_ctx_get_type(ctx)); /* * blackhole routes are not RTN_UNICAST, they are * RTN_ BLACKHOLE|UNREACHABLE|PROHIBIT * so setting this value as a RTN_UNICAST would * cause the route lookup of just the prefix * to fail. So no need to specify this for * the RTM_DELROUTE case */ if (cmd != RTM_DELROUTE) req->r.rtm_type = RTN_UNICAST; if (!nl_attr_put(&req->n, datalen, RTA_DST, &p->u.prefix, bytelen)) return 0; if (src_p) { if (!nl_attr_put(&req->n, datalen, RTA_SRC, &src_p->u.prefix, bytelen)) return 0; } /* Metric. */ /* Hardcode the metric for all routes coming from zebra. Metric isn't * used * either by the kernel or by zebra. Its purely for calculating best * path(s) * by the routing protocol and for communicating with protocol peers. */ if (!nl_attr_put32(&req->n, datalen, RTA_PRIORITY, ROUTE_INSTALLATION_METRIC)) return 0; #if defined(SUPPORT_REALMS) if (cmd == RTM_DELROUTE) tag = dplane_ctx_get_old_tag(ctx); else tag = dplane_ctx_get_tag(ctx); #endif /* Table corresponding to this route. */ table_id = dplane_ctx_get_table(ctx); if (table_id < 256) req->r.rtm_table = table_id; else { req->r.rtm_table = RT_TABLE_UNSPEC; if (!nl_attr_put32(&req->n, datalen, RTA_TABLE, table_id)) return 0; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "%s: %s %pFX vrf %u(%u)", __func__, nl_msg_type_to_str(cmd), p, dplane_ctx_get_vrf(ctx), table_id); /* * If we are not updating the route and we have received * a route delete, then all we need to fill in is the * prefix information to tell the kernel to schwack * it. */ if (cmd == RTM_DELROUTE) { if (!_netlink_set_tag(&req->n, datalen, tag)) return 0; return NLMSG_ALIGN(req->n.nlmsg_len); } if (dplane_ctx_get_mtu(ctx) || dplane_ctx_get_nh_mtu(ctx)) { struct rtattr *nest; uint32_t mtu = dplane_ctx_get_mtu(ctx); uint32_t nexthop_mtu = dplane_ctx_get_nh_mtu(ctx); if (!mtu || (nexthop_mtu && nexthop_mtu < mtu)) mtu = nexthop_mtu; nest = nl_attr_nest(&req->n, datalen, RTA_METRICS); if (nest == NULL) return 0; if (!nl_attr_put(&req->n, datalen, RTAX_MTU, &mtu, sizeof(mtu))) return 0; nl_attr_nest_end(&req->n, nest); } /* * Always install blackhole routes without using nexthops, because of * the following kernel problems: * 1. Kernel nexthops don't suport unreachable/prohibit route types. * 2. Blackhole kernel nexthops are deleted when loopback is down. */ nexthop = dplane_ctx_get_ng(ctx)->nexthop; if (nexthop) { if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE)) nexthop = nexthop->resolved; if (nexthop->type == NEXTHOP_TYPE_BLACKHOLE) { switch (nexthop->bh_type) { case BLACKHOLE_ADMINPROHIB: req->r.rtm_type = RTN_PROHIBIT; break; case BLACKHOLE_REJECT: req->r.rtm_type = RTN_UNREACHABLE; break; case BLACKHOLE_UNSPEC: case BLACKHOLE_NULL: req->r.rtm_type = RTN_BLACKHOLE; break; } return NLMSG_ALIGN(req->n.nlmsg_len); } } if ((!fpm && kernel_nexthops_supported() && (!proto_nexthops_only() || is_proto_nhg(dplane_ctx_get_nhe_id(ctx), 0))) || (fpm && force_nhg)) { /* Kernel supports nexthop objects */ if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: %pFX nhg_id is %u", __func__, p, dplane_ctx_get_nhe_id(ctx)); if (!nl_attr_put32(&req->n, datalen, RTA_NH_ID, dplane_ctx_get_nhe_id(ctx))) return 0; /* Have to determine src still */ for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx), nexthop)) { if (setsrc) break; setsrc = nexthop_set_src(nexthop, p->family, &src); } if (setsrc) { if (p->family == AF_INET) { if (!nl_attr_put(&req->n, datalen, RTA_PREFSRC, &src.ipv4, bytelen)) return 0; } else if (p->family == AF_INET6) { if (!nl_attr_put(&req->n, datalen, RTA_PREFSRC, &src.ipv6, bytelen)) return 0; } } return NLMSG_ALIGN(req->n.nlmsg_len); } /* Count overall nexthops so we can decide whether to use singlepath * or multipath case. */ nexthop_num = 0; for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx), nexthop)) { if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE)) continue; if (!NEXTHOP_IS_ACTIVE(nexthop->flags)) continue; nexthop_num++; } /* Singlepath case. */ if (nexthop_num == 1) { nexthop_num = 0; for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx), nexthop)) { if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE)) { if (setsrc) continue; setsrc = nexthop_set_src(nexthop, p->family, &src); continue; } if (NEXTHOP_IS_ACTIVE(nexthop->flags)) { routedesc = nexthop->rparent ? "recursive, single-path" : "single-path"; if (!_netlink_set_tag(&req->n, datalen, tag)) return 0; if (!_netlink_route_build_singlepath( p, routedesc, bytelen, nexthop, &req->n, &req->r, datalen, cmd)) return 0; nexthop_num++; break; } /* * Add encapsulation information when installing via * FPM. */ if (fpm) { if (!netlink_route_nexthop_encap( &req->n, datalen, nexthop)) return 0; } } if (setsrc) { if (p->family == AF_INET) { if (!nl_attr_put(&req->n, datalen, RTA_PREFSRC, &src.ipv4, bytelen)) return 0; } else if (p->family == AF_INET6) { if (!nl_attr_put(&req->n, datalen, RTA_PREFSRC, &src.ipv6, bytelen)) return 0; } } } else { /* Multipath case */ struct rtattr *nest; const union g_addr *src1 = NULL; nest = nl_attr_nest(&req->n, datalen, RTA_MULTIPATH); if (nest == NULL) return 0; nexthop_num = 0; for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx), nexthop)) { if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE)) { /* This only works for IPv4 now */ if (setsrc) continue; setsrc = nexthop_set_src(nexthop, p->family, &src); continue; } if (NEXTHOP_IS_ACTIVE(nexthop->flags)) { routedesc = nexthop->rparent ? "recursive, multipath" : "multipath"; nexthop_num++; if (!_netlink_route_build_multipath( p, routedesc, bytelen, nexthop, &req->n, datalen, &req->r, &src1, tag)) return 0; if (!setsrc && src1) { if (p->family == AF_INET) src.ipv4 = src1->ipv4; else if (p->family == AF_INET6) src.ipv6 = src1->ipv6; setsrc = 1; } } } nl_attr_nest_end(&req->n, nest); /* * Add encapsulation information when installing via * FPM. */ if (fpm) { for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx), nexthop)) { if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE)) continue; if (!netlink_route_nexthop_encap( &req->n, datalen, nexthop)) return 0; } } if (setsrc) { if (p->family == AF_INET) { if (!nl_attr_put(&req->n, datalen, RTA_PREFSRC, &src.ipv4, bytelen)) return 0; } else if (p->family == AF_INET6) { if (!nl_attr_put(&req->n, datalen, RTA_PREFSRC, &src.ipv6, bytelen)) return 0; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("Setting source"); } } /* If there is no useful nexthop then return. */ if (nexthop_num == 0) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: No useful nexthop.", __func__); } return NLMSG_ALIGN(req->n.nlmsg_len); } int kernel_get_ipmr_sg_stats(struct zebra_vrf *zvrf, void *in) { uint32_t actual_table; int suc = 0; struct mcast_route_data *mr = (struct mcast_route_data *)in; struct { struct nlmsghdr n; struct rtmsg rtm; char buf[256]; } req; mroute = mr; struct zebra_ns *zns; zns = zvrf->zns; memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg)); req.n.nlmsg_flags = NLM_F_REQUEST; req.n.nlmsg_pid = zns->netlink_cmd.snl.nl_pid; req.n.nlmsg_type = RTM_GETROUTE; if (mroute->family == AF_INET) { req.rtm.rtm_family = RTNL_FAMILY_IPMR; req.rtm.rtm_dst_len = IPV4_MAX_BITLEN; req.rtm.rtm_src_len = IPV4_MAX_BITLEN; nl_attr_put(&req.n, sizeof(req), RTA_SRC, &mroute->src.ipaddr_v4, sizeof(mroute->src.ipaddr_v4)); nl_attr_put(&req.n, sizeof(req), RTA_DST, &mroute->grp.ipaddr_v4, sizeof(mroute->grp.ipaddr_v4)); } else { req.rtm.rtm_family = RTNL_FAMILY_IP6MR; req.rtm.rtm_dst_len = IPV6_MAX_BITLEN; req.rtm.rtm_src_len = IPV6_MAX_BITLEN; nl_attr_put(&req.n, sizeof(req), RTA_SRC, &mroute->src.ipaddr_v6, sizeof(mroute->src.ipaddr_v6)); nl_attr_put(&req.n, sizeof(req), RTA_DST, &mroute->grp.ipaddr_v6, sizeof(mroute->grp.ipaddr_v6)); } /* * What? * * So during the namespace cleanup we started storing * the zvrf table_id for the default table as RT_TABLE_MAIN * which is what the normal routing table for ip routing is. * This change caused this to break our lookups of sg data * because prior to this change the zvrf->table_id was 0 * and when the pim multicast kernel code saw a 0, * it was auto-translated to RT_TABLE_DEFAULT. But since * we are now passing in RT_TABLE_MAIN there is no auto-translation * and the kernel goes screw you and the delicious cookies you * are trying to give me. So now we have this little hack. */ if (mroute->family == AF_INET) actual_table = (zvrf->table_id == RT_TABLE_MAIN) ? RT_TABLE_DEFAULT : zvrf->table_id; else actual_table = zvrf->table_id; nl_attr_put32(&req.n, sizeof(req), RTA_TABLE, actual_table); suc = netlink_talk(netlink_route_change_read_multicast, &req.n, &zns->netlink_cmd, zns, false); mroute = NULL; return suc; } /* Char length to debug ID with */ #define ID_LENGTH 10 static bool _netlink_nexthop_build_group(struct nlmsghdr *n, size_t req_size, uint32_t id, const struct nh_grp *z_grp, const uint8_t count, bool resilient, const struct nhg_resilience *nhgr) { struct nexthop_grp grp[count]; /* Need space for max group size, "/", and null term */ char buf[(MULTIPATH_NUM * (ID_LENGTH + 1)) + 1]; char buf1[ID_LENGTH + 2]; buf[0] = '\0'; memset(grp, 0, sizeof(grp)); if (count) { for (int i = 0; i < count; i++) { grp[i].id = z_grp[i].id; grp[i].weight = z_grp[i].weight - 1; if (IS_ZEBRA_DEBUG_KERNEL) { if (i == 0) snprintf(buf, sizeof(buf1), "group %u", grp[i].id); else { snprintf(buf1, sizeof(buf1), "/%u", grp[i].id); strlcat(buf, buf1, sizeof(buf)); } } } if (!nl_attr_put(n, req_size, NHA_GROUP, grp, count * sizeof(*grp))) return false; if (resilient) { struct rtattr *nest; nest = nl_attr_nest(n, req_size, NHA_RES_GROUP); nl_attr_put16(n, req_size, NHA_RES_GROUP_BUCKETS, nhgr->buckets); nl_attr_put32(n, req_size, NHA_RES_GROUP_IDLE_TIMER, nhgr->idle_timer * 1000); nl_attr_put32(n, req_size, NHA_RES_GROUP_UNBALANCED_TIMER, nhgr->unbalanced_timer * 1000); nl_attr_nest_end(n, nest); nl_attr_put16(n, req_size, NHA_GROUP_TYPE, NEXTHOP_GRP_TYPE_RES); } } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: ID (%u): %s", __func__, id, buf); return true; } /** * Next hop packet encoding helper function. * * \param[in] cmd netlink command. * \param[in] ctx dataplane context (information snapshot). * \param[out] buf buffer to hold the packet. * \param[in] buflen amount of buffer bytes. * * \returns -1 on failure, 0 when the msg doesn't fit entirely in the buffer * otherwise the number of bytes written to buf. */ ssize_t netlink_nexthop_msg_encode(uint16_t cmd, const struct zebra_dplane_ctx *ctx, void *buf, size_t buflen, bool fpm) { struct { struct nlmsghdr n; struct nhmsg nhm; char buf[]; } *req = buf; mpls_lse_t out_lse[MPLS_MAX_LABELS]; char label_buf[256]; int num_labels = 0; uint32_t id = dplane_ctx_get_nhe_id(ctx); int type = dplane_ctx_get_nhe_type(ctx); struct rtattr *nest; uint16_t encap; struct nlsock *nl = kernel_netlink_nlsock_lookup(dplane_ctx_get_ns_sock(ctx)); if (!id) { flog_err( EC_ZEBRA_NHG_FIB_UPDATE, "Failed trying to update a nexthop group in the kernel that does not have an ID"); return -1; } /* * Nothing to do if the kernel doesn't support nexthop objects or * we dont want to install this type of NHG, but FPM may possible to * handle this. */ if (!fpm && !kernel_nexthops_supported()) { if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_NHG) zlog_debug( "%s: nhg_id %u (%s): kernel nexthops not supported, ignoring", __func__, id, zebra_route_string(type)); return 0; } if (proto_nexthops_only() && !is_proto_nhg(id, type)) { if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_NHG) zlog_debug( "%s: nhg_id %u (%s): proto-based nexthops only, ignoring", __func__, id, zebra_route_string(type)); return 0; } label_buf[0] = '\0'; if (buflen < sizeof(*req)) return 0; memset(req, 0, sizeof(*req)); req->n.nlmsg_len = NLMSG_LENGTH(sizeof(struct nhmsg)); req->n.nlmsg_flags = NLM_F_CREATE | NLM_F_REQUEST; if (cmd == RTM_NEWNEXTHOP) req->n.nlmsg_flags |= NLM_F_REPLACE; req->n.nlmsg_type = cmd; req->n.nlmsg_pid = nl->snl.nl_pid; req->nhm.nh_family = AF_UNSPEC; /* TODO: Scope? */ if (!nl_attr_put32(&req->n, buflen, NHA_ID, id)) return 0; if (cmd == RTM_NEWNEXTHOP) { /* * We distinguish between a "group", which is a collection * of ids, and a singleton nexthop with an id. The * group is installed as an id that just refers to a list of * other ids. */ if (dplane_ctx_get_nhe_nh_grp_count(ctx)) { const struct nexthop_group *nhg; const struct nhg_resilience *nhgr; nhg = dplane_ctx_get_nhe_ng(ctx); nhgr = &nhg->nhgr; if (!_netlink_nexthop_build_group( &req->n, buflen, id, dplane_ctx_get_nhe_nh_grp(ctx), dplane_ctx_get_nhe_nh_grp_count(ctx), !!nhgr->buckets, nhgr)) return 0; } else { const struct nexthop *nh = dplane_ctx_get_nhe_ng(ctx)->nexthop; afi_t afi = dplane_ctx_get_nhe_afi(ctx); if (afi == AFI_IP) req->nhm.nh_family = AF_INET; else if (afi == AFI_IP6) req->nhm.nh_family = AF_INET6; switch (nh->type) { case NEXTHOP_TYPE_IPV4: case NEXTHOP_TYPE_IPV4_IFINDEX: if (!nl_attr_put(&req->n, buflen, NHA_GATEWAY, &nh->gate.ipv4, IPV4_MAX_BYTELEN)) return 0; break; case NEXTHOP_TYPE_IPV6: case NEXTHOP_TYPE_IPV6_IFINDEX: if (!nl_attr_put(&req->n, buflen, NHA_GATEWAY, &nh->gate.ipv6, IPV6_MAX_BYTELEN)) return 0; break; case NEXTHOP_TYPE_BLACKHOLE: if (!nl_attr_put(&req->n, buflen, NHA_BLACKHOLE, NULL, 0)) return 0; /* Blackhole shouldn't have anymore attributes */ goto nexthop_done; case NEXTHOP_TYPE_IFINDEX: /* Don't need anymore info for this */ break; } if (!nh->ifindex) { flog_err( EC_ZEBRA_NHG_FIB_UPDATE, "Context received for kernel nexthop update without an interface"); return -1; } if (!nl_attr_put32(&req->n, buflen, NHA_OIF, nh->ifindex)) return 0; if (CHECK_FLAG(nh->flags, NEXTHOP_FLAG_ONLINK)) req->nhm.nh_flags |= RTNH_F_ONLINK; num_labels = build_label_stack( nh->nh_label, nh->nh_label_type, out_lse, label_buf, sizeof(label_buf)); if (num_labels && nh->nh_label_type == ZEBRA_LSP_EVPN) { if (!nl_attr_put16(&req->n, buflen, NHA_ENCAP_TYPE, LWTUNNEL_ENCAP_IP)) return 0; nest = nl_attr_nest(&req->n, buflen, NHA_ENCAP); if (!nest) return 0; if (_netlink_nexthop_encode_dvni_label( nh, &req->n, out_lse, buflen, label_buf) == false) return 0; nl_attr_nest_end(&req->n, nest); } else if (num_labels) { /* Set the BoS bit */ out_lse[num_labels - 1] |= htonl(1 << MPLS_LS_S_SHIFT); /* * TODO: MPLS unsupported for now in kernel. */ if (req->nhm.nh_family == AF_MPLS) goto nexthop_done; encap = LWTUNNEL_ENCAP_MPLS; if (!nl_attr_put16(&req->n, buflen, NHA_ENCAP_TYPE, encap)) return 0; nest = nl_attr_nest(&req->n, buflen, NHA_ENCAP); if (!nest) return 0; if (!nl_attr_put( &req->n, buflen, MPLS_IPTUNNEL_DST, &out_lse, num_labels * sizeof(mpls_lse_t))) return 0; nl_attr_nest_end(&req->n, nest); } if (nh->nh_srv6) { if (nh->nh_srv6->seg6local_action != ZEBRA_SEG6_LOCAL_ACTION_UNSPEC) { uint32_t action; uint16_t encap; struct rtattr *nest; const struct seg6local_context *ctx; req->nhm.nh_family = AF_INET6; action = nh->nh_srv6->seg6local_action; ctx = &nh->nh_srv6->seg6local_ctx; encap = LWTUNNEL_ENCAP_SEG6_LOCAL; if (!nl_attr_put(&req->n, buflen, NHA_ENCAP_TYPE, &encap, sizeof(uint16_t))) return 0; nest = nl_attr_nest(&req->n, buflen, NHA_ENCAP | NLA_F_NESTED); if (!nest) return 0; switch (action) { case SEG6_LOCAL_ACTION_END: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END)) return 0; break; case SEG6_LOCAL_ACTION_END_X: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_X)) return 0; if (!nl_attr_put( &req->n, buflen, SEG6_LOCAL_NH6, &ctx->nh6, sizeof(struct in6_addr))) return 0; break; case SEG6_LOCAL_ACTION_END_T: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_T)) return 0; if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_TABLE, ctx->table)) return 0; break; case SEG6_LOCAL_ACTION_END_DX4: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DX4)) return 0; if (!nl_attr_put( &req->n, buflen, SEG6_LOCAL_NH4, &ctx->nh4, sizeof(struct in_addr))) return 0; break; case SEG6_LOCAL_ACTION_END_DT6: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DT6)) return 0; if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_TABLE, ctx->table)) return 0; break; case SEG6_LOCAL_ACTION_END_DT4: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DT4)) return 0; if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_VRFTABLE, ctx->table)) return 0; break; case SEG6_LOCAL_ACTION_END_DT46: if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_ACTION, SEG6_LOCAL_ACTION_END_DT46)) return 0; if (!nl_attr_put32( &req->n, buflen, SEG6_LOCAL_VRFTABLE, ctx->table)) return 0; break; default: zlog_err("%s: unsupport seg6local behaviour action=%u", __func__, action); return 0; } nl_attr_nest_end(&req->n, nest); } if (!sid_zero(&nh->nh_srv6->seg6_segs)) { char tun_buf[4096]; ssize_t tun_len; struct rtattr *nest; if (!nl_attr_put16(&req->n, buflen, NHA_ENCAP_TYPE, LWTUNNEL_ENCAP_SEG6)) return 0; nest = nl_attr_nest(&req->n, buflen, NHA_ENCAP | NLA_F_NESTED); if (!nest) return 0; tun_len = fill_seg6ipt_encap(tun_buf, sizeof(tun_buf), &nh->nh_srv6->seg6_segs); if (tun_len < 0) return 0; if (!nl_attr_put(&req->n, buflen, SEG6_IPTUNNEL_SRH, tun_buf, tun_len)) return 0; nl_attr_nest_end(&req->n, nest); } } nexthop_done: if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: ID (%u): %pNHv(%d) vrf %s(%u) %s ", __func__, id, nh, nh->ifindex, vrf_id_to_name(nh->vrf_id), nh->vrf_id, label_buf); } req->nhm.nh_protocol = zebra2proto(type); } else if (cmd != RTM_DELNEXTHOP) { flog_err( EC_ZEBRA_NHG_FIB_UPDATE, "Nexthop group kernel update command (%d) does not exist", cmd); return -1; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: %s, id=%u", __func__, nl_msg_type_to_str(cmd), id); return NLMSG_ALIGN(req->n.nlmsg_len); } static ssize_t netlink_nexthop_msg_encoder(struct zebra_dplane_ctx *ctx, void *buf, size_t buflen) { enum dplane_op_e op; int cmd = 0; op = dplane_ctx_get_op(ctx); if (op == DPLANE_OP_NH_INSTALL || op == DPLANE_OP_NH_UPDATE) cmd = RTM_NEWNEXTHOP; else if (op == DPLANE_OP_NH_DELETE) cmd = RTM_DELNEXTHOP; else { flog_err(EC_ZEBRA_NHG_FIB_UPDATE, "Context received for kernel nexthop update with incorrect OP code (%u)", op); return -1; } return netlink_nexthop_msg_encode(cmd, ctx, buf, buflen, false); } enum netlink_msg_status netlink_put_nexthop_update_msg(struct nl_batch *bth, struct zebra_dplane_ctx *ctx) { /* Nothing to do if the kernel doesn't support nexthop objects */ if (!kernel_nexthops_supported()) return FRR_NETLINK_SUCCESS; return netlink_batch_add_msg(bth, ctx, netlink_nexthop_msg_encoder, false); } static ssize_t netlink_newroute_msg_encoder(struct zebra_dplane_ctx *ctx, void *buf, size_t buflen) { return netlink_route_multipath_msg_encode(RTM_NEWROUTE, ctx, buf, buflen, false, false); } static ssize_t netlink_delroute_msg_encoder(struct zebra_dplane_ctx *ctx, void *buf, size_t buflen) { return netlink_route_multipath_msg_encode(RTM_DELROUTE, ctx, buf, buflen, false, false); } enum netlink_msg_status netlink_put_route_update_msg(struct nl_batch *bth, struct zebra_dplane_ctx *ctx) { int cmd; const struct prefix *p = dplane_ctx_get_dest(ctx); if (dplane_ctx_get_op(ctx) == DPLANE_OP_ROUTE_DELETE) { cmd = RTM_DELROUTE; } else if (dplane_ctx_get_op(ctx) == DPLANE_OP_ROUTE_INSTALL) { cmd = RTM_NEWROUTE; } else if (dplane_ctx_get_op(ctx) == DPLANE_OP_ROUTE_UPDATE) { if (p->family == AF_INET || v6_rr_semantics) { /* Single 'replace' operation */ /* * With route replace semantics in place * for v4 routes and the new route is a system * route we do not install anything. * The problem here is that the new system * route should cause us to withdraw from * the kernel the old non-system route */ if (RSYSTEM_ROUTE(dplane_ctx_get_type(ctx)) && !RSYSTEM_ROUTE(dplane_ctx_get_old_type(ctx))) return netlink_batch_add_msg( bth, ctx, netlink_delroute_msg_encoder, true); } else { /* * So v6 route replace semantics are not in * the kernel at this point as I understand it. * so let's do a delete then an add. * In the future once v6 route replace semantics * are in we can figure out what to do here to * allow working with old and new kernels. * * I'm also intentionally ignoring the failure case * of the route delete. If that happens yeah we're * screwed. */ if (!RSYSTEM_ROUTE(dplane_ctx_get_old_type(ctx))) netlink_batch_add_msg( bth, ctx, netlink_delroute_msg_encoder, true); } cmd = RTM_NEWROUTE; } else return FRR_NETLINK_ERROR; if (RSYSTEM_ROUTE(dplane_ctx_get_type(ctx))) return FRR_NETLINK_SUCCESS; return netlink_batch_add_msg(bth, ctx, cmd == RTM_NEWROUTE ? netlink_newroute_msg_encoder : netlink_delroute_msg_encoder, false); } /** * netlink_nexthop_process_nh() - Parse the gatway/if info from a new nexthop * * @tb: Netlink RTA data * @family: Address family in the nhmsg * @ifp: Interface connected - this should be NULL, we fill it in * @ns_id: Namspace id * * Return: New nexthop */ static struct nexthop netlink_nexthop_process_nh(struct rtattr **tb, unsigned char family, struct interface **ifp, ns_id_t ns_id) { struct nexthop nh = {}; void *gate = NULL; enum nexthop_types_t type = 0; int if_index = 0; size_t sz = 0; struct interface *ifp_lookup; if_index = *(int *)RTA_DATA(tb[NHA_OIF]); if (tb[NHA_GATEWAY]) { switch (family) { case AF_INET: type = NEXTHOP_TYPE_IPV4_IFINDEX; sz = 4; break; case AF_INET6: type = NEXTHOP_TYPE_IPV6_IFINDEX; sz = 16; break; default: flog_warn( EC_ZEBRA_BAD_NHG_MESSAGE, "Nexthop gateway with bad address family (%d) received from kernel", family); return nh; } gate = RTA_DATA(tb[NHA_GATEWAY]); } else type = NEXTHOP_TYPE_IFINDEX; if (type) nh.type = type; if (gate) memcpy(&(nh.gate), gate, sz); if (if_index) nh.ifindex = if_index; ifp_lookup = if_lookup_by_index_per_ns(zebra_ns_lookup(ns_id), nh.ifindex); if (ifp) *ifp = ifp_lookup; if (ifp_lookup) nh.vrf_id = ifp_lookup->vrf->vrf_id; else { flog_warn( EC_ZEBRA_UNKNOWN_INTERFACE, "%s: Unknown nexthop interface %u received, defaulting to VRF_DEFAULT", __func__, nh.ifindex); nh.vrf_id = VRF_DEFAULT; } if (tb[NHA_ENCAP] && tb[NHA_ENCAP_TYPE]) { uint16_t encap_type = *(uint16_t *)RTA_DATA(tb[NHA_ENCAP_TYPE]); int num_labels = 0; mpls_label_t labels[MPLS_MAX_LABELS] = {0}; if (encap_type == LWTUNNEL_ENCAP_MPLS) num_labels = parse_encap_mpls(tb[NHA_ENCAP], labels); if (num_labels) nexthop_add_labels(&nh, ZEBRA_LSP_STATIC, num_labels, labels); } return nh; } static int netlink_nexthop_process_group(struct rtattr **tb, struct nh_grp *z_grp, int z_grp_size, struct nhg_resilience *nhgr) { uint8_t count = 0; /* linux/nexthop.h group struct */ struct nexthop_grp *n_grp = NULL; n_grp = (struct nexthop_grp *)RTA_DATA(tb[NHA_GROUP]); count = (RTA_PAYLOAD(tb[NHA_GROUP]) / sizeof(*n_grp)); if (!count || (count * sizeof(*n_grp)) != RTA_PAYLOAD(tb[NHA_GROUP])) { flog_warn(EC_ZEBRA_BAD_NHG_MESSAGE, "Invalid nexthop group received from the kernel"); return count; } for (int i = 0; ((i < count) && (i < z_grp_size)); i++) { z_grp[i].id = n_grp[i].id; z_grp[i].weight = n_grp[i].weight + 1; } memset(nhgr, 0, sizeof(*nhgr)); if (tb[NHA_RES_GROUP]) { struct rtattr *tbn[NHA_RES_GROUP_MAX + 1]; struct rtattr *rta; struct rtattr *res_group = tb[NHA_RES_GROUP]; netlink_parse_rtattr_nested(tbn, NHA_RES_GROUP_MAX, res_group); if (tbn[NHA_RES_GROUP_BUCKETS]) { rta = tbn[NHA_RES_GROUP_BUCKETS]; nhgr->buckets = *(uint16_t *)RTA_DATA(rta); } if (tbn[NHA_RES_GROUP_IDLE_TIMER]) { rta = tbn[NHA_RES_GROUP_IDLE_TIMER]; nhgr->idle_timer = *(uint32_t *)RTA_DATA(rta); } if (tbn[NHA_RES_GROUP_UNBALANCED_TIMER]) { rta = tbn[NHA_RES_GROUP_UNBALANCED_TIMER]; nhgr->unbalanced_timer = *(uint32_t *)RTA_DATA(rta); } if (tbn[NHA_RES_GROUP_UNBALANCED_TIME]) { rta = tbn[NHA_RES_GROUP_UNBALANCED_TIME]; nhgr->unbalanced_time = *(uint64_t *)RTA_DATA(rta); } } return count; } /** * netlink_nexthop_change() - Read in change about nexthops from the kernel * * @h: Netlink message header * @ns_id: Namspace id * @startup: Are we reading under startup conditions? * * Return: Result status */ int netlink_nexthop_change(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; /* nexthop group id */ uint32_t id; unsigned char family; int type; afi_t afi = AFI_UNSPEC; vrf_id_t vrf_id = VRF_DEFAULT; struct interface *ifp = NULL; struct nhmsg *nhm = NULL; struct nexthop nh = {}; struct nh_grp grp[MULTIPATH_NUM] = {}; /* Count of nexthops in group array */ uint8_t grp_count = 0; struct rtattr *tb[NHA_MAX + 1] = {}; frrtrace(3, frr_zebra, netlink_nexthop_change, h, ns_id, startup); nhm = NLMSG_DATA(h); if (ns_id) vrf_id = ns_id; if (startup && h->nlmsg_type != RTM_NEWNEXTHOP) return 0; len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct nhmsg)); if (len < 0) { zlog_warn( "%s: Message received from netlink is of a broken size %d %zu", __func__, h->nlmsg_len, (size_t)NLMSG_LENGTH(sizeof(struct nhmsg))); return -1; } netlink_parse_rtattr_flags(tb, NHA_MAX, RTM_NHA(nhm), len, NLA_F_NESTED); if (!tb[NHA_ID]) { flog_warn( EC_ZEBRA_BAD_NHG_MESSAGE, "Nexthop group without an ID received from the kernel"); return -1; } /* We use the ID key'd nhg table for kernel updates */ id = *((uint32_t *)RTA_DATA(tb[NHA_ID])); if (zebra_evpn_mh_is_fdb_nh(id)) { /* If this is a L2 NH just ignore it */ if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_EVPN_MH_NH) { zlog_debug("Ignore kernel update (%u) for fdb-nh 0x%x", h->nlmsg_type, id); } return 0; } family = nhm->nh_family; afi = family2afi(family); type = proto2zebra(nhm->nh_protocol, 0, true); if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s ID (%u) %s NS %u", nl_msg_type_to_str(h->nlmsg_type), id, nl_family_to_str(family), ns_id); if (h->nlmsg_type == RTM_NEWNEXTHOP) { struct nhg_resilience nhgr = {}; if (tb[NHA_GROUP]) { /** * If this is a group message its only going to have * an array of nexthop IDs associated with it */ grp_count = netlink_nexthop_process_group( tb, grp, array_size(grp), &nhgr); } else { if (tb[NHA_BLACKHOLE]) { /** * This nexthop is just for blackhole-ing * traffic, it should not have an OIF, GATEWAY, * or ENCAP */ nh.type = NEXTHOP_TYPE_BLACKHOLE; nh.bh_type = BLACKHOLE_UNSPEC; } else if (tb[NHA_OIF]) /** * This is a true new nexthop, so we need * to parse the gateway and device info */ nh = netlink_nexthop_process_nh(tb, family, &ifp, ns_id); else { flog_warn( EC_ZEBRA_BAD_NHG_MESSAGE, "Invalid Nexthop message received from the kernel with ID (%u)", id); return -1; } SET_FLAG(nh.flags, NEXTHOP_FLAG_ACTIVE); if (nhm->nh_flags & RTNH_F_ONLINK) SET_FLAG(nh.flags, NEXTHOP_FLAG_ONLINK); vrf_id = nh.vrf_id; } if (zebra_nhg_kernel_find(id, &nh, grp, grp_count, vrf_id, afi, type, startup, &nhgr)) return -1; } else if (h->nlmsg_type == RTM_DELNEXTHOP) zebra_nhg_kernel_del(id, vrf_id); return 0; } /** * netlink_request_nexthop() - Request nextop information from the kernel * @zns: Zebra namespace * @family: AF_* netlink family * @type: RTM_* route type * * Return: Result status */ static int netlink_request_nexthop(struct zebra_ns *zns, int family, int type) { struct { struct nlmsghdr n; struct nhmsg nhm; } req; /* Form the request, specifying filter (rtattr) if needed. */ memset(&req, 0, sizeof(req)); req.n.nlmsg_type = type; req.n.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST; req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct nhmsg)); req.nhm.nh_family = family; return netlink_request(&zns->netlink_cmd, &req); } /** * netlink_nexthop_read() - Nexthop read function using netlink interface * * @zns: Zebra name space * * Return: Result status * Only called at bootstrap time. */ int netlink_nexthop_read(struct zebra_ns *zns) { int ret; struct zebra_dplane_info dp_info; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Get nexthop objects */ ret = netlink_request_nexthop(zns, AF_UNSPEC, RTM_GETNEXTHOP); if (ret < 0) return ret; ret = netlink_parse_info(netlink_nexthop_change, &zns->netlink_cmd, &dp_info, 0, true); if (!ret) /* If we succesfully read in nexthop objects, * this kernel must support them. */ supports_nh = true; if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_NHG) zlog_debug("Nexthop objects %ssupported on this kernel", supports_nh ? "" : "not "); zebra_router_set_supports_nhgs(supports_nh); return ret; } int kernel_neigh_update(int add, int ifindex, void *addr, char *lla, int llalen, ns_id_t ns_id, uint8_t family, bool permanent) { return netlink_neigh_update(add ? RTM_NEWNEIGH : RTM_DELNEIGH, ifindex, addr, lla, llalen, ns_id, family, permanent, RTPROT_ZEBRA); } /** * netlink_neigh_update_msg_encode() - Common helper api for encoding * evpn neighbor update as netlink messages using dataplane context object. * Here, a neighbor refers to a bridge forwarding database entry for * either unicast forwarding or head-end replication or an IP neighbor * entry. * @ctx: Dataplane context * @cmd: Netlink command (RTM_NEWNEIGH or RTM_DELNEIGH) * @lla: A pointer to neighbor cache link layer address * @llalen: Length of the pointer to neighbor cache link layer * address * @ip: A neighbor cache n/w layer destination address * In the case of bridge FDB, this represnts the remote * VTEP IP. * @replace_obj: Whether NEW request should replace existing object or * add to the end of the list * @family: AF_* netlink family * @type: RTN_* route type * @flags: NTF_* flags * @state: NUD_* states * @data: data buffer pointer * @datalen: total amount of data buffer space * @protocol: protocol information * * Return: 0 when the msg doesn't fit entirely in the buffer * otherwise the number of bytes written to buf. */ static ssize_t netlink_neigh_update_msg_encode( const struct zebra_dplane_ctx *ctx, int cmd, const void *lla, int llalen, const struct ipaddr *ip, bool replace_obj, uint8_t family, uint8_t type, uint8_t flags, uint16_t state, uint32_t nhg_id, bool nfy, uint8_t nfy_flags, bool ext, uint32_t ext_flags, void *data, size_t datalen, uint8_t protocol) { struct { struct nlmsghdr n; struct ndmsg ndm; char buf[]; } *req = data; int ipa_len; enum dplane_op_e op; if (datalen < sizeof(*req)) return 0; memset(req, 0, sizeof(*req)); op = dplane_ctx_get_op(ctx); req->n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ndmsg)); req->n.nlmsg_flags = NLM_F_REQUEST; if (cmd == RTM_NEWNEIGH) req->n.nlmsg_flags |= NLM_F_CREATE | (replace_obj ? NLM_F_REPLACE : NLM_F_APPEND); req->n.nlmsg_type = cmd; req->ndm.ndm_family = family; req->ndm.ndm_type = type; req->ndm.ndm_state = state; req->ndm.ndm_flags = flags; req->ndm.ndm_ifindex = dplane_ctx_get_ifindex(ctx); if (!nl_attr_put(&req->n, datalen, NDA_PROTOCOL, &protocol, sizeof(protocol))) return 0; if (lla) { if (!nl_attr_put(&req->n, datalen, NDA_LLADDR, lla, llalen)) return 0; } if (nfy) { struct rtattr *nest; nest = nl_attr_nest(&req->n, datalen, NDA_FDB_EXT_ATTRS | NLA_F_NESTED); if (!nest) return 0; if (!nl_attr_put(&req->n, datalen, NFEA_ACTIVITY_NOTIFY, &nfy_flags, sizeof(nfy_flags))) return 0; if (!nl_attr_put(&req->n, datalen, NFEA_DONT_REFRESH, NULL, 0)) return 0; nl_attr_nest_end(&req->n, nest); } if (ext) { if (!nl_attr_put(&req->n, datalen, NDA_EXT_FLAGS, &ext_flags, sizeof(ext_flags))) return 0; } if (nhg_id) { if (!nl_attr_put32(&req->n, datalen, NDA_NH_ID, nhg_id)) return 0; } else { ipa_len = IS_IPADDR_V4(ip) ? IPV4_MAX_BYTELEN : IPV6_MAX_BYTELEN; if (!nl_attr_put(&req->n, datalen, NDA_DST, &ip->ip.addr, ipa_len)) return 0; } if (op == DPLANE_OP_MAC_INSTALL || op == DPLANE_OP_MAC_DELETE) { vlanid_t vid = dplane_ctx_mac_get_vlan(ctx); vni_t vni = dplane_ctx_mac_get_vni(ctx); if (vid > 0) { if (!nl_attr_put16(&req->n, datalen, NDA_VLAN, vid)) return 0; } if (vni > 0) { if (!nl_attr_put32(&req->n, datalen, NDA_SRC_VNI, vni)) return 0; } if (!nl_attr_put32(&req->n, datalen, NDA_MASTER, dplane_ctx_mac_get_br_ifindex(ctx))) return 0; } if (op == DPLANE_OP_VTEP_ADD || op == DPLANE_OP_VTEP_DELETE) { vni_t vni = dplane_ctx_neigh_get_vni(ctx); if (vni > 0) { if (!nl_attr_put32(&req->n, datalen, NDA_SRC_VNI, vni)) return 0; } } return NLMSG_ALIGN(req->n.nlmsg_len); } /* * Add remote VTEP to the flood list for this VxLAN interface (VNI). This * is done by adding an FDB entry with a MAC of 00:00:00:00:00:00. */ static ssize_t netlink_vxlan_flood_update_ctx(const struct zebra_dplane_ctx *ctx, int cmd, void *buf, size_t buflen) { struct ethaddr dst_mac = {.octet = {0}}; int proto = RTPROT_ZEBRA; if (dplane_ctx_get_type(ctx) != 0) proto = zebra2proto(dplane_ctx_get_type(ctx)); return netlink_neigh_update_msg_encode( ctx, cmd, (const void *)&dst_mac, ETH_ALEN, dplane_ctx_neigh_get_ipaddr(ctx), false, PF_BRIDGE, 0, NTF_SELF, (NUD_NOARP | NUD_PERMANENT), 0 /*nhg*/, false /*nfy*/, 0 /*nfy_flags*/, false /*ext*/, 0 /*ext_flags*/, buf, buflen, proto); } #ifndef NDA_RTA #define NDA_RTA(r) \ ((struct rtattr *)(((char *)(r)) + NLMSG_ALIGN(sizeof(struct ndmsg)))) #endif static int netlink_macfdb_change(struct nlmsghdr *h, int len, ns_id_t ns_id) { struct ndmsg *ndm; struct interface *ifp; struct zebra_if *zif; struct rtattr *tb[NDA_MAX + 1]; struct interface *br_if; struct ethaddr mac; vlanid_t vid = 0; struct in_addr vtep_ip; int vid_present = 0, dst_present = 0; char vid_buf[20]; char dst_buf[30]; bool sticky; bool local_inactive = false; bool dp_static = false; vni_t vni = 0; uint32_t nhg_id = 0; bool vni_mcast_grp = false; ndm = NLMSG_DATA(h); /* We only process macfdb notifications if EVPN is enabled */ if (!is_evpn_enabled()) return 0; /* Parse attributes and extract fields of interest. Do basic * validation of the fields. */ netlink_parse_rtattr_flags(tb, NDA_MAX, NDA_RTA(ndm), len, NLA_F_NESTED); if (!tb[NDA_LLADDR]) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s AF_BRIDGE IF %u - no LLADDR", nl_msg_type_to_str(h->nlmsg_type), ndm->ndm_ifindex); return 0; } if (RTA_PAYLOAD(tb[NDA_LLADDR]) != ETH_ALEN) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "%s AF_BRIDGE IF %u - LLADDR is not MAC, len %lu", nl_msg_type_to_str(h->nlmsg_type), ndm->ndm_ifindex, (unsigned long)RTA_PAYLOAD(tb[NDA_LLADDR])); return 0; } memcpy(&mac, RTA_DATA(tb[NDA_LLADDR]), ETH_ALEN); if (tb[NDA_VLAN]) { vid_present = 1; vid = *(uint16_t *)RTA_DATA(tb[NDA_VLAN]); snprintf(vid_buf, sizeof(vid_buf), " VLAN %u", vid); } if (tb[NDA_DST]) { /* TODO: Only IPv4 supported now. */ dst_present = 1; memcpy(&vtep_ip.s_addr, RTA_DATA(tb[NDA_DST]), IPV4_MAX_BYTELEN); snprintfrr(dst_buf, sizeof(dst_buf), " dst %pI4", &vtep_ip); } else memset(&vtep_ip, 0, sizeof(vtep_ip)); if (tb[NDA_NH_ID]) nhg_id = *(uint32_t *)RTA_DATA(tb[NDA_NH_ID]); if (ndm->ndm_state & NUD_STALE) local_inactive = true; if (tb[NDA_FDB_EXT_ATTRS]) { struct rtattr *attr = tb[NDA_FDB_EXT_ATTRS]; struct rtattr *nfea_tb[NFEA_MAX + 1] = {0}; netlink_parse_rtattr_nested(nfea_tb, NFEA_MAX, attr); if (nfea_tb[NFEA_ACTIVITY_NOTIFY]) { uint8_t nfy_flags; nfy_flags = *(uint8_t *)RTA_DATA( nfea_tb[NFEA_ACTIVITY_NOTIFY]); if (nfy_flags & FDB_NOTIFY_BIT) dp_static = true; if (nfy_flags & FDB_NOTIFY_INACTIVE_BIT) local_inactive = true; } } if (tb[NDA_SRC_VNI]) vni = *(vni_t *)RTA_DATA(tb[NDA_SRC_VNI]); if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "Rx %s AF_BRIDGE IF %u%s st 0x%x fl 0x%x MAC %pEA%s nhg %d vni %d", nl_msg_type_to_str(h->nlmsg_type), ndm->ndm_ifindex, vid_present ? vid_buf : "", ndm->ndm_state, ndm->ndm_flags, &mac, dst_present ? dst_buf : "", nhg_id, vni); /* The interface should exist. */ ifp = if_lookup_by_index_per_ns(zebra_ns_lookup(ns_id), ndm->ndm_ifindex); if (!ifp || !ifp->info) return 0; /* The interface should be something we're interested in. */ if (!IS_ZEBRA_IF_BRIDGE_SLAVE(ifp)) return 0; zif = (struct zebra_if *)ifp->info; if ((br_if = zif->brslave_info.br_if) == NULL) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "%s AF_BRIDGE IF %s(%u) brIF %u - no bridge master", nl_msg_type_to_str(h->nlmsg_type), ifp->name, ndm->ndm_ifindex, zif->brslave_info.bridge_ifindex); return 0; } /* For per vni device, vni comes from device itself */ if (IS_ZEBRA_IF_VXLAN(ifp) && IS_ZEBRA_VXLAN_IF_VNI(zif)) { struct zebra_vxlan_vni *vnip; vnip = zebra_vxlan_if_vni_find(zif, 0); vni = vnip->vni; } sticky = !!(ndm->ndm_flags & NTF_STICKY); if (filter_vlan && vid != filter_vlan) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug(" Filtered due to filter vlan: %d", filter_vlan); return 0; } /* * Check if this is a mcast group update (svd case) */ vni_mcast_grp = is_mac_vni_mcast_group(&mac, vni, vtep_ip); /* If add or update, do accordingly if learnt on a "local" interface; if * the notification is over VxLAN, this has to be related to * multi-homing, * so perform an implicit delete of any local entry (if it exists). */ if (h->nlmsg_type == RTM_NEWNEIGH) { /* Drop "permanent" entries. */ if (!vni_mcast_grp && (ndm->ndm_state & NUD_PERMANENT)) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( " Dropping entry because of NUD_PERMANENT"); return 0; } if (IS_ZEBRA_IF_VXLAN(ifp)) { if (!dst_present) return 0; if (vni_mcast_grp) return zebra_vxlan_if_vni_mcast_group_add_update( ifp, vni, &vtep_ip); return zebra_vxlan_dp_network_mac_add( ifp, br_if, &mac, vid, vni, nhg_id, sticky, !!(ndm->ndm_flags & NTF_EXT_LEARNED)); } return zebra_vxlan_local_mac_add_update(ifp, br_if, &mac, vid, sticky, local_inactive, dp_static); } /* This is a delete notification. * Ignore the notification with IP dest as it may just signify that the * MAC has moved from remote to local. The exception is the special * all-zeros MAC that represents the BUM flooding entry; we may have * to readd it. Otherwise, * 1. For a MAC over VxLan, check if it needs to be refreshed(readded) * 2. For a MAC over "local" interface, delete the mac * Note: We will get notifications from both bridge driver and VxLAN * driver. */ if (nhg_id) return 0; if (dst_present) { if (vni_mcast_grp) return zebra_vxlan_if_vni_mcast_group_del(ifp, vni, &vtep_ip); if (is_zero_mac(&mac) && vni) return zebra_vxlan_check_readd_vtep(ifp, vni, vtep_ip); return 0; } if (IS_ZEBRA_IF_VXLAN(ifp)) return 0; return zebra_vxlan_local_mac_del(ifp, br_if, &mac, vid); } static int netlink_macfdb_table(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ndmsg *ndm; if (h->nlmsg_type != RTM_NEWNEIGH) return 0; /* Length validity. */ len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct ndmsg)); if (len < 0) return -1; /* We are interested only in AF_BRIDGE notifications. */ ndm = NLMSG_DATA(h); if (ndm->ndm_family != AF_BRIDGE) return 0; return netlink_macfdb_change(h, len, ns_id); } /* Request for MAC FDB information from the kernel */ static int netlink_request_macs(struct nlsock *netlink_cmd, int family, int type, ifindex_t master_ifindex) { struct { struct nlmsghdr n; struct ifinfomsg ifm; char buf[256]; } req; /* Form the request, specifying filter (rtattr) if needed. */ memset(&req, 0, sizeof(req)); req.n.nlmsg_type = type; req.n.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST; req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)); req.ifm.ifi_family = family; if (master_ifindex) nl_attr_put32(&req.n, sizeof(req), IFLA_MASTER, master_ifindex); return netlink_request(netlink_cmd, &req); } /* * MAC forwarding database read using netlink interface. This is invoked * at startup. */ int netlink_macfdb_read(struct zebra_ns *zns) { int ret; struct zebra_dplane_info dp_info; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Get bridge FDB table. */ ret = netlink_request_macs(&zns->netlink_cmd, AF_BRIDGE, RTM_GETNEIGH, 0); if (ret < 0) return ret; /* We are reading entire table. */ filter_vlan = 0; ret = netlink_parse_info(netlink_macfdb_table, &zns->netlink_cmd, &dp_info, 0, true); return ret; } /* * MAC forwarding database read using netlink interface. This is for a * specific bridge and matching specific access VLAN (if VLAN-aware bridge). */ int netlink_macfdb_read_for_bridge(struct zebra_ns *zns, struct interface *ifp, struct interface *br_if, vlanid_t vid) { struct zebra_if *br_zif; struct zebra_dplane_info dp_info; int ret = 0; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Save VLAN we're filtering on, if needed. */ br_zif = (struct zebra_if *)br_if->info; if (IS_ZEBRA_IF_BRIDGE_VLAN_AWARE(br_zif)) filter_vlan = vid; /* Get bridge FDB table for specific bridge - we do the VLAN filtering. */ ret = netlink_request_macs(&zns->netlink_cmd, AF_BRIDGE, RTM_GETNEIGH, br_if->ifindex); if (ret < 0) return ret; ret = netlink_parse_info(netlink_macfdb_table, &zns->netlink_cmd, &dp_info, 0, false); /* Reset VLAN filter. */ filter_vlan = 0; return ret; } /* Request for MAC FDB for a specific MAC address in VLAN from the kernel */ static int netlink_request_specific_mac(struct zebra_ns *zns, int family, int type, struct interface *ifp, const struct ethaddr *mac, vlanid_t vid, vni_t vni, uint8_t flags) { struct { struct nlmsghdr n; struct ndmsg ndm; char buf[256]; } req; struct zebra_if *zif; memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ndmsg)); req.n.nlmsg_type = type; /* RTM_GETNEIGH */ req.n.nlmsg_flags = NLM_F_REQUEST; req.ndm.ndm_family = family; /* AF_BRIDGE */ req.ndm.ndm_flags = flags; /* req.ndm.ndm_state = NUD_REACHABLE; */ nl_attr_put(&req.n, sizeof(req), NDA_LLADDR, mac, 6); zif = (struct zebra_if *)ifp->info; /* Is this a read on a VXLAN interface? */ if (IS_ZEBRA_IF_VXLAN(ifp)) { nl_attr_put32(&req.n, sizeof(req), NDA_VNI, vni); /* TBD: Why is ifindex not filled in the non-vxlan case? */ req.ndm.ndm_ifindex = ifp->ifindex; } else { if (IS_ZEBRA_IF_BRIDGE_VLAN_AWARE(zif) && vid > 0) nl_attr_put16(&req.n, sizeof(req), NDA_VLAN, vid); nl_attr_put32(&req.n, sizeof(req), NDA_MASTER, ifp->ifindex); } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("Tx %s %s IF %s(%u) MAC %pEA vid %u vni %u", nl_msg_type_to_str(type), nl_family_to_str(req.ndm.ndm_family), ifp->name, ifp->ifindex, mac, vid, vni); return netlink_request(&zns->netlink_cmd, &req); } int netlink_macfdb_read_specific_mac(struct zebra_ns *zns, struct interface *br_if, const struct ethaddr *mac, vlanid_t vid) { int ret = 0; struct zebra_dplane_info dp_info; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Get bridge FDB table for specific bridge - we do the VLAN filtering. */ ret = netlink_request_specific_mac(zns, AF_BRIDGE, RTM_GETNEIGH, br_if, mac, vid, 0, 0); if (ret < 0) return ret; ret = netlink_parse_info(netlink_macfdb_table, &zns->netlink_cmd, &dp_info, 1, 0); return ret; } int netlink_macfdb_read_mcast_for_vni(struct zebra_ns *zns, struct interface *ifp, vni_t vni) { struct zebra_if *zif; struct ethaddr mac = {.octet = {0}}; struct zebra_dplane_info dp_info; int ret = 0; zif = ifp->info; if (IS_ZEBRA_VXLAN_IF_VNI(zif)) return 0; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Get specific FDB entry for BUM handling, if any */ ret = netlink_request_specific_mac(zns, AF_BRIDGE, RTM_GETNEIGH, ifp, &mac, 0, vni, NTF_SELF); if (ret < 0) return ret; ret = netlink_parse_info(netlink_macfdb_table, &zns->netlink_cmd, &dp_info, 1, false); return ret; } /* * Netlink-specific handler for MAC updates using dataplane context object. */ ssize_t netlink_macfdb_update_ctx(struct zebra_dplane_ctx *ctx, void *data, size_t datalen) { struct ipaddr vtep_ip; vlanid_t vid; ssize_t total; int cmd; uint8_t flags; uint16_t state; uint32_t nhg_id; uint32_t update_flags; bool nfy = false; uint8_t nfy_flags = 0; int proto = RTPROT_ZEBRA; if (dplane_ctx_get_type(ctx) != 0) proto = zebra2proto(dplane_ctx_get_type(ctx)); cmd = dplane_ctx_get_op(ctx) == DPLANE_OP_MAC_INSTALL ? RTM_NEWNEIGH : RTM_DELNEIGH; flags = NTF_MASTER; state = NUD_REACHABLE; update_flags = dplane_ctx_mac_get_update_flags(ctx); if (update_flags & DPLANE_MAC_REMOTE) { flags |= NTF_SELF; if (dplane_ctx_mac_is_sticky(ctx)) { /* NUD_NOARP prevents the entry from expiring */ state |= NUD_NOARP; /* sticky the entry from moving */ flags |= NTF_STICKY; } else { flags |= NTF_EXT_LEARNED; } /* if it was static-local previously we need to clear the * notify flags on replace with remote */ if (update_flags & DPLANE_MAC_WAS_STATIC) nfy = true; } else { /* local mac */ if (update_flags & DPLANE_MAC_SET_STATIC) { nfy_flags |= FDB_NOTIFY_BIT; state |= NUD_NOARP; } if (update_flags & DPLANE_MAC_SET_INACTIVE) nfy_flags |= FDB_NOTIFY_INACTIVE_BIT; nfy = true; } nhg_id = dplane_ctx_mac_get_nhg_id(ctx); vtep_ip.ipaddr_v4 = *(dplane_ctx_mac_get_vtep_ip(ctx)); SET_IPADDR_V4(&vtep_ip); if (IS_ZEBRA_DEBUG_KERNEL) { char vid_buf[20]; const struct ethaddr *mac = dplane_ctx_mac_get_addr(ctx); vid = dplane_ctx_mac_get_vlan(ctx); if (vid > 0) snprintf(vid_buf, sizeof(vid_buf), " VLAN %u", vid); else vid_buf[0] = '\0'; zlog_debug( "Tx %s family %s IF %s(%u)%s %sMAC %pEA dst %pIA nhg %u%s%s%s%s%s", nl_msg_type_to_str(cmd), nl_family_to_str(AF_BRIDGE), dplane_ctx_get_ifname(ctx), dplane_ctx_get_ifindex(ctx), vid_buf, dplane_ctx_mac_is_sticky(ctx) ? "sticky " : "", mac, &vtep_ip, nhg_id, (update_flags & DPLANE_MAC_REMOTE) ? " rem" : "", (update_flags & DPLANE_MAC_WAS_STATIC) ? " clr_sync" : "", (update_flags & DPLANE_MAC_SET_STATIC) ? " static" : "", (update_flags & DPLANE_MAC_SET_INACTIVE) ? " inactive" : "", nfy ? " nfy" : ""); } total = netlink_neigh_update_msg_encode( ctx, cmd, (const void *)dplane_ctx_mac_get_addr(ctx), ETH_ALEN, &vtep_ip, true, AF_BRIDGE, 0, flags, state, nhg_id, nfy, nfy_flags, false /*ext*/, 0 /*ext_flags*/, data, datalen, proto); return total; } /* * In the event the kernel deletes ipv4 link-local neighbor entries created for * 5549 support, re-install them. */ static void netlink_handle_5549(struct ndmsg *ndm, struct zebra_if *zif, struct interface *ifp, struct ipaddr *ip, bool handle_failed) { if (ndm->ndm_family != AF_INET) return; if (!zif->v6_2_v4_ll_neigh_entry) return; if (ipv4_ll.s_addr != ip->ip._v4_addr.s_addr) return; if (handle_failed && ndm->ndm_state & NUD_FAILED) { zlog_info("Neighbor Entry for %s has entered a failed state, not reinstalling", ifp->name); return; } if_nbr_ipv6ll_to_ipv4ll_neigh_update(ifp, &zif->v6_2_v4_ll_addr6, true); } #define NUD_VALID \ (NUD_PERMANENT | NUD_NOARP | NUD_REACHABLE | NUD_PROBE | NUD_STALE \ | NUD_DELAY) #define NUD_LOCAL_ACTIVE \ (NUD_PERMANENT | NUD_NOARP | NUD_REACHABLE) static int netlink_nbr_entry_state_to_zclient(int nbr_state) { /* an exact match is done between * - netlink neighbor state values: NDM_XXX (see in linux/neighbour.h) * - zclient neighbor state values: ZEBRA_NEIGH_STATE_XXX * (see in lib/zclient.h) */ return nbr_state; } static int netlink_ipneigh_change(struct nlmsghdr *h, int len, ns_id_t ns_id) { struct ndmsg *ndm; struct interface *ifp; struct zebra_if *zif; struct rtattr *tb[NDA_MAX + 1]; struct interface *link_if; struct ethaddr mac; struct ipaddr ip; char buf[ETHER_ADDR_STRLEN]; int mac_present = 0; bool is_ext; bool is_router; bool local_inactive; uint32_t ext_flags = 0; bool dp_static = false; int l2_len = 0; int cmd; ndm = NLMSG_DATA(h); /* The interface should exist. */ ifp = if_lookup_by_index_per_ns(zebra_ns_lookup(ns_id), ndm->ndm_ifindex); if (!ifp || !ifp->info) return 0; zif = (struct zebra_if *)ifp->info; /* Parse attributes and extract fields of interest. */ netlink_parse_rtattr(tb, NDA_MAX, NDA_RTA(ndm), len); if (!tb[NDA_DST]) { zlog_debug("%s family %s IF %s(%u) vrf %s(%u) - no DST", nl_msg_type_to_str(h->nlmsg_type), nl_family_to_str(ndm->ndm_family), ifp->name, ndm->ndm_ifindex, ifp->vrf->name, ifp->vrf->vrf_id); return 0; } memset(&ip, 0, sizeof(ip)); ip.ipa_type = (ndm->ndm_family == AF_INET) ? IPADDR_V4 : IPADDR_V6; memcpy(&ip.ip.addr, RTA_DATA(tb[NDA_DST]), RTA_PAYLOAD(tb[NDA_DST])); /* if kernel deletes our rfc5549 neighbor entry, re-install it */ if (h->nlmsg_type == RTM_DELNEIGH && (ndm->ndm_state & NUD_PERMANENT)) { netlink_handle_5549(ndm, zif, ifp, &ip, false); if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( " Neighbor Entry Received is a 5549 entry, finished"); return 0; } /* if kernel marks our rfc5549 neighbor entry invalid, re-install it */ if (h->nlmsg_type == RTM_NEWNEIGH && !(ndm->ndm_state & NUD_VALID)) netlink_handle_5549(ndm, zif, ifp, &ip, true); /* we send link layer information to client: * - nlmsg_type = RTM_DELNEIGH|NEWNEIGH|GETNEIGH * - struct ipaddr ( for DEL and GET) * - struct ethaddr mac; (for NEW) */ if (h->nlmsg_type == RTM_NEWNEIGH) cmd = ZEBRA_NHRP_NEIGH_ADDED; else if (h->nlmsg_type == RTM_GETNEIGH) cmd = ZEBRA_NHRP_NEIGH_GET; else if (h->nlmsg_type == RTM_DELNEIGH) cmd = ZEBRA_NHRP_NEIGH_REMOVED; else { zlog_debug("%s(): unknown nlmsg type %u", __func__, h->nlmsg_type); return 0; } if (tb[NDA_LLADDR]) { /* copy LLADDR information */ l2_len = RTA_PAYLOAD(tb[NDA_LLADDR]); } if (l2_len == IPV4_MAX_BYTELEN || l2_len == 0) { union sockunion link_layer_ipv4; if (l2_len) { sockunion_family(&link_layer_ipv4) = AF_INET; memcpy((void *)sockunion_get_addr(&link_layer_ipv4), RTA_DATA(tb[NDA_LLADDR]), l2_len); } else sockunion_family(&link_layer_ipv4) = AF_UNSPEC; zsend_nhrp_neighbor_notify( cmd, ifp, &ip, netlink_nbr_entry_state_to_zclient(ndm->ndm_state), &link_layer_ipv4); } if (h->nlmsg_type == RTM_GETNEIGH) return 0; /* The neighbor is present on an SVI. From this, we locate the * underlying * bridge because we're only interested in neighbors on a VxLAN bridge. * The bridge is located based on the nature of the SVI: * (a) In the case of a VLAN-aware bridge, the SVI is a L3 VLAN * interface * and is linked to the bridge * (b) In the case of a VLAN-unaware bridge, the SVI is the bridge * interface * itself */ if (IS_ZEBRA_IF_VLAN(ifp)) { link_if = if_lookup_by_index_per_ns(zebra_ns_lookup(ns_id), zif->link_ifindex); if (!link_if) return 0; } else if (IS_ZEBRA_IF_BRIDGE(ifp)) link_if = ifp; else { link_if = NULL; if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( " Neighbor Entry received is not on a VLAN or a BRIDGE, ignoring"); } memset(&mac, 0, sizeof(mac)); if (h->nlmsg_type == RTM_NEWNEIGH) { if (tb[NDA_LLADDR]) { if (RTA_PAYLOAD(tb[NDA_LLADDR]) != ETH_ALEN) { if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "%s family %s IF %s(%u) vrf %s(%u) - LLADDR is not MAC, len %lu", nl_msg_type_to_str( h->nlmsg_type), nl_family_to_str( ndm->ndm_family), ifp->name, ndm->ndm_ifindex, ifp->vrf->name, ifp->vrf->vrf_id, (unsigned long)RTA_PAYLOAD( tb[NDA_LLADDR])); return 0; } mac_present = 1; memcpy(&mac, RTA_DATA(tb[NDA_LLADDR]), ETH_ALEN); } is_ext = !!(ndm->ndm_flags & NTF_EXT_LEARNED); is_router = !!(ndm->ndm_flags & NTF_ROUTER); if (tb[NDA_EXT_FLAGS]) { ext_flags = *(uint32_t *)RTA_DATA(tb[NDA_EXT_FLAGS]); if (ext_flags & NTF_E_MH_PEER_SYNC) dp_static = true; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "Rx %s family %s IF %s(%u) vrf %s(%u) IP %pIA MAC %s state 0x%x flags 0x%x ext_flags 0x%x", nl_msg_type_to_str(h->nlmsg_type), nl_family_to_str(ndm->ndm_family), ifp->name, ndm->ndm_ifindex, ifp->vrf->name, ifp->vrf->vrf_id, &ip, mac_present ? prefix_mac2str(&mac, buf, sizeof(buf)) : "", ndm->ndm_state, ndm->ndm_flags, ext_flags); /* If the neighbor state is valid for use, process as an add or * update * else process as a delete. Note that the delete handling may * result * in re-adding the neighbor if it is a valid "remote" neighbor. */ if (ndm->ndm_state & NUD_VALID) { if (zebra_evpn_mh_do_adv_reachable_neigh_only()) local_inactive = !(ndm->ndm_state & NUD_LOCAL_ACTIVE); else /* If EVPN-MH is not enabled we treat STALE * neighbors as locally-active and advertise * them */ local_inactive = false; /* Add local neighbors to the l3 interface database */ if (is_ext) zebra_neigh_del(ifp, &ip); else zebra_neigh_add(ifp, &ip, &mac); if (link_if) zebra_vxlan_handle_kernel_neigh_update( ifp, link_if, &ip, &mac, ndm->ndm_state, is_ext, is_router, local_inactive, dp_static); return 0; } zebra_neigh_del(ifp, &ip); if (link_if) zebra_vxlan_handle_kernel_neigh_del(ifp, link_if, &ip); return 0; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("Rx %s family %s IF %s(%u) vrf %s(%u) IP %pIA", nl_msg_type_to_str(h->nlmsg_type), nl_family_to_str(ndm->ndm_family), ifp->name, ndm->ndm_ifindex, ifp->vrf->name, ifp->vrf->vrf_id, &ip); /* Process the delete - it may result in re-adding the neighbor if it is * a valid "remote" neighbor. */ zebra_neigh_del(ifp, &ip); if (link_if) zebra_vxlan_handle_kernel_neigh_del(ifp, link_if, &ip); return 0; } static int netlink_neigh_table(struct nlmsghdr *h, ns_id_t ns_id, int startup) { int len; struct ndmsg *ndm; if (h->nlmsg_type != RTM_NEWNEIGH) return 0; /* Length validity. */ len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct ndmsg)); if (len < 0) return -1; /* We are interested only in AF_INET or AF_INET6 notifications. */ ndm = NLMSG_DATA(h); if (ndm->ndm_family != AF_INET && ndm->ndm_family != AF_INET6) return 0; return netlink_neigh_change(h, len); } /* Request for IP neighbor information from the kernel */ static int netlink_request_neigh(struct nlsock *netlink_cmd, int family, int type, ifindex_t ifindex) { struct { struct nlmsghdr n; struct ndmsg ndm; char buf[256]; } req; /* Form the request, specifying filter (rtattr) if needed. */ memset(&req, 0, sizeof(req)); req.n.nlmsg_type = type; req.n.nlmsg_flags = NLM_F_ROOT | NLM_F_MATCH | NLM_F_REQUEST; req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ndmsg)); req.ndm.ndm_family = family; if (ifindex) nl_attr_put32(&req.n, sizeof(req), NDA_IFINDEX, ifindex); return netlink_request(netlink_cmd, &req); } /* * IP Neighbor table read using netlink interface. This is invoked * at startup. */ int netlink_neigh_read(struct zebra_ns *zns) { int ret; struct zebra_dplane_info dp_info; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); /* Get IP neighbor table. */ ret = netlink_request_neigh(&zns->netlink_cmd, AF_UNSPEC, RTM_GETNEIGH, 0); if (ret < 0) return ret; ret = netlink_parse_info(netlink_neigh_table, &zns->netlink_cmd, &dp_info, 0, true); return ret; } /* * IP Neighbor table read using netlink interface. This is for a specific * VLAN device. */ int netlink_neigh_read_for_vlan(struct zebra_ns *zns, struct interface *vlan_if) { int ret = 0; struct zebra_dplane_info dp_info; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); ret = netlink_request_neigh(&zns->netlink_cmd, AF_UNSPEC, RTM_GETNEIGH, vlan_if->ifindex); if (ret < 0) return ret; ret = netlink_parse_info(netlink_neigh_table, &zns->netlink_cmd, &dp_info, 0, false); return ret; } /* * Request for a specific IP in VLAN (SVI) device from IP Neighbor table, * read using netlink interface. */ static int netlink_request_specific_neigh_in_vlan(struct zebra_ns *zns, int type, const struct ipaddr *ip, ifindex_t ifindex) { struct { struct nlmsghdr n; struct ndmsg ndm; char buf[256]; } req; int ipa_len; /* Form the request, specifying filter (rtattr) if needed. */ memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ndmsg)); req.n.nlmsg_flags = NLM_F_REQUEST; req.n.nlmsg_type = type; /* RTM_GETNEIGH */ req.ndm.ndm_ifindex = ifindex; if (IS_IPADDR_V4(ip)) { ipa_len = IPV4_MAX_BYTELEN; req.ndm.ndm_family = AF_INET; } else { ipa_len = IPV6_MAX_BYTELEN; req.ndm.ndm_family = AF_INET6; } nl_attr_put(&req.n, sizeof(req), NDA_DST, &ip->ip.addr, ipa_len); if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: Tx %s family %s IF %u IP %pIA flags 0x%x", __func__, nl_msg_type_to_str(type), nl_family_to_str(req.ndm.ndm_family), ifindex, ip, req.n.nlmsg_flags); return netlink_request(&zns->netlink_cmd, &req); } int netlink_neigh_read_specific_ip(const struct ipaddr *ip, struct interface *vlan_if) { int ret = 0; struct zebra_ns *zns; struct zebra_vrf *zvrf = vlan_if->vrf->info; struct zebra_dplane_info dp_info; zns = zvrf->zns; zebra_dplane_info_from_zns(&dp_info, zns, true /*is_cmd*/); if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug("%s: neigh request IF %s(%u) IP %pIA vrf %s(%u)", __func__, vlan_if->name, vlan_if->ifindex, ip, vlan_if->vrf->name, vlan_if->vrf->vrf_id); ret = netlink_request_specific_neigh_in_vlan(zns, RTM_GETNEIGH, ip, vlan_if->ifindex); if (ret < 0) return ret; ret = netlink_parse_info(netlink_neigh_table, &zns->netlink_cmd, &dp_info, 1, false); return ret; } int netlink_neigh_change(struct nlmsghdr *h, ns_id_t ns_id) { int len; struct ndmsg *ndm; if (!(h->nlmsg_type == RTM_NEWNEIGH || h->nlmsg_type == RTM_DELNEIGH || h->nlmsg_type == RTM_GETNEIGH)) return 0; /* Length validity. */ len = h->nlmsg_len - NLMSG_LENGTH(sizeof(struct ndmsg)); if (len < 0) { zlog_err( "%s: Message received from netlink is of a broken size %d %zu", __func__, h->nlmsg_len, (size_t)NLMSG_LENGTH(sizeof(struct ndmsg))); return -1; } /* Is this a notification for the MAC FDB or IP neighbor table? */ ndm = NLMSG_DATA(h); if (ndm->ndm_family == AF_BRIDGE) return netlink_macfdb_change(h, len, ns_id); if (ndm->ndm_type != RTN_UNICAST) return 0; if (ndm->ndm_family == AF_INET || ndm->ndm_family == AF_INET6) return netlink_ipneigh_change(h, len, ns_id); else { flog_warn( EC_ZEBRA_UNKNOWN_FAMILY, "Invalid address family: %u received from kernel neighbor change: %s", ndm->ndm_family, nl_msg_type_to_str(h->nlmsg_type)); return 0; } return 0; } /* * Utility neighbor-update function, using info from dplane context. */ static ssize_t netlink_neigh_update_ctx(const struct zebra_dplane_ctx *ctx, int cmd, void *buf, size_t buflen) { const struct ipaddr *ip; const struct ethaddr *mac = NULL; const struct ipaddr *link_ip = NULL; const void *link_ptr = NULL; char buf2[ETHER_ADDR_STRLEN]; int llalen; uint8_t flags; uint16_t state; uint8_t family; uint32_t update_flags; uint32_t ext_flags = 0; bool ext = false; int proto = RTPROT_ZEBRA; if (dplane_ctx_get_type(ctx) != 0) proto = zebra2proto(dplane_ctx_get_type(ctx)); ip = dplane_ctx_neigh_get_ipaddr(ctx); if (dplane_ctx_get_op(ctx) == DPLANE_OP_NEIGH_IP_INSTALL || dplane_ctx_get_op(ctx) == DPLANE_OP_NEIGH_IP_DELETE) { link_ip = dplane_ctx_neigh_get_link_ip(ctx); llalen = IPADDRSZ(link_ip); link_ptr = (const void *)&(link_ip->ip.addr); ipaddr2str(link_ip, buf2, sizeof(buf2)); } else { mac = dplane_ctx_neigh_get_mac(ctx); llalen = ETH_ALEN; link_ptr = (const void *)mac; if (is_zero_mac(mac)) mac = NULL; if (mac) prefix_mac2str(mac, buf2, sizeof(buf2)); else snprintf(buf2, sizeof(buf2), "null"); } update_flags = dplane_ctx_neigh_get_update_flags(ctx); flags = neigh_flags_to_netlink(dplane_ctx_neigh_get_flags(ctx)); state = neigh_state_to_netlink(dplane_ctx_neigh_get_state(ctx)); family = IS_IPADDR_V4(ip) ? AF_INET : AF_INET6; if (update_flags & DPLANE_NEIGH_REMOTE) { flags |= NTF_EXT_LEARNED; /* if it was static-local previously we need to clear the * ext flags on replace with remote */ if (update_flags & DPLANE_NEIGH_WAS_STATIC) ext = true; } else if (!(update_flags & DPLANE_NEIGH_NO_EXTENSION)) { ext = true; /* local neigh */ if (update_flags & DPLANE_NEIGH_SET_STATIC) ext_flags |= NTF_E_MH_PEER_SYNC; } if (IS_ZEBRA_DEBUG_KERNEL) zlog_debug( "Tx %s family %s IF %s(%u) Neigh %pIA %s %s flags 0x%x state 0x%x %sext_flags 0x%x", nl_msg_type_to_str(cmd), nl_family_to_str(family), dplane_ctx_get_ifname(ctx), dplane_ctx_get_ifindex(ctx), ip, link_ip ? "Link" : "MAC", buf2, flags, state, ext ? "ext " : "", ext_flags); return netlink_neigh_update_msg_encode( ctx, cmd, link_ptr, llalen, ip, true, family, RTN_UNICAST, flags, state, 0 /*nhg*/, false /*nfy*/, 0 /*nfy_flags*/, ext, ext_flags, buf, buflen, proto); } static int netlink_neigh_table_update_ctx(const struct zebra_dplane_ctx *ctx, void *data, size_t datalen) { struct { struct nlmsghdr n; struct ndtmsg ndtm; char buf[]; } *req = data; struct rtattr *nest; uint8_t family; ifindex_t idx; uint32_t val; if (datalen < sizeof(*req)) return 0; memset(req, 0, sizeof(*req)); family = dplane_ctx_neightable_get_family(ctx); idx = dplane_ctx_get_ifindex(ctx); req->n.nlmsg_len = NLMSG_LENGTH(sizeof(struct ndtmsg)); req->n.nlmsg_flags = NLM_F_REQUEST | NLM_F_REPLACE; req->n.nlmsg_type = RTM_SETNEIGHTBL; req->ndtm.ndtm_family = family; nl_attr_put(&req->n, datalen, NDTA_NAME, family == AF_INET ? "arp_cache" : "ndisc_cache", 10); nest = nl_attr_nest(&req->n, datalen, NDTA_PARMS); if (nest == NULL) return 0; if (!nl_attr_put(&req->n, datalen, NDTPA_IFINDEX, &idx, sizeof(idx))) return 0; val = dplane_ctx_neightable_get_app_probes(ctx); if (!nl_attr_put(&req->n, datalen, NDTPA_APP_PROBES, &val, sizeof(val))) return 0; val = dplane_ctx_neightable_get_mcast_probes(ctx); if (!nl_attr_put(&req->n, datalen, NDTPA_MCAST_PROBES, &val, sizeof(val))) return 0; val = dplane_ctx_neightable_get_ucast_probes(ctx); if (!nl_attr_put(&req->n, datalen, NDTPA_UCAST_PROBES, &val, sizeof(val))) return 0; nl_attr_nest_end(&req->n, nest); return NLMSG_ALIGN(req->n.nlmsg_len); } static ssize_t netlink_neigh_msg_encoder(struct zebra_dplane_ctx *ctx, void *buf, size_t buflen) { ssize_t ret = 0; switch (dplane_ctx_get_op(ctx)) { case DPLANE_OP_NEIGH_INSTALL: case DPLANE_OP_NEIGH_UPDATE: case DPLANE_OP_NEIGH_DISCOVER: case DPLANE_OP_NEIGH_IP_INSTALL: ret = netlink_neigh_update_ctx(ctx, RTM_NEWNEIGH, buf, buflen); break; case DPLANE_OP_NEIGH_DELETE: case DPLANE_OP_NEIGH_IP_DELETE: ret = netlink_neigh_update_ctx(ctx, RTM_DELNEIGH, buf, buflen); break; case DPLANE_OP_VTEP_ADD: ret = netlink_vxlan_flood_update_ctx(ctx, RTM_NEWNEIGH, buf, buflen); break; case DPLANE_OP_VTEP_DELETE: ret = netlink_vxlan_flood_update_ctx(ctx, RTM_DELNEIGH, buf, buflen); break; case DPLANE_OP_NEIGH_TABLE_UPDATE: ret = netlink_neigh_table_update_ctx(ctx, buf, buflen); break; case DPLANE_OP_ROUTE_INSTALL: case DPLANE_OP_ROUTE_UPDATE: case DPLANE_OP_ROUTE_DELETE: case DPLANE_OP_ROUTE_NOTIFY: case DPLANE_OP_NH_INSTALL: case DPLANE_OP_NH_UPDATE: case DPLANE_OP_NH_DELETE: case DPLANE_OP_LSP_INSTALL: case DPLANE_OP_LSP_UPDATE: case DPLANE_OP_LSP_DELETE: case DPLANE_OP_LSP_NOTIFY: case DPLANE_OP_PW_INSTALL: case DPLANE_OP_PW_UNINSTALL: case DPLANE_OP_SYS_ROUTE_ADD: case DPLANE_OP_SYS_ROUTE_DELETE: case DPLANE_OP_ADDR_INSTALL: case DPLANE_OP_ADDR_UNINSTALL: case DPLANE_OP_MAC_INSTALL: case DPLANE_OP_MAC_DELETE: case DPLANE_OP_RULE_ADD: case DPLANE_OP_RULE_DELETE: case DPLANE_OP_RULE_UPDATE: case DPLANE_OP_BR_PORT_UPDATE: case DPLANE_OP_IPTABLE_ADD: case DPLANE_OP_IPTABLE_DELETE: case DPLANE_OP_IPSET_ADD: case DPLANE_OP_IPSET_DELETE: case DPLANE_OP_IPSET_ENTRY_ADD: case DPLANE_OP_IPSET_ENTRY_DELETE: case DPLANE_OP_GRE_SET: case DPLANE_OP_INTF_ADDR_ADD: case DPLANE_OP_INTF_ADDR_DEL: case DPLANE_OP_INTF_NETCONFIG: case DPLANE_OP_INTF_INSTALL: case DPLANE_OP_INTF_UPDATE: case DPLANE_OP_INTF_DELETE: case DPLANE_OP_TC_QDISC_INSTALL: case DPLANE_OP_TC_QDISC_UNINSTALL: case DPLANE_OP_TC_CLASS_ADD: case DPLANE_OP_TC_CLASS_DELETE: case DPLANE_OP_TC_CLASS_UPDATE: case DPLANE_OP_TC_FILTER_ADD: case DPLANE_OP_TC_FILTER_DELETE: case DPLANE_OP_TC_FILTER_UPDATE: case DPLANE_OP_NONE: ret = -1; } return ret; } /* * Update MAC, using dataplane context object. */ enum netlink_msg_status netlink_put_mac_update_msg(struct nl_batch *bth, struct zebra_dplane_ctx *ctx) { return netlink_batch_add_msg(bth, ctx, netlink_macfdb_update_ctx, false); } enum netlink_msg_status netlink_put_neigh_update_msg(struct nl_batch *bth, struct zebra_dplane_ctx *ctx) { return netlink_batch_add_msg(bth, ctx, netlink_neigh_msg_encoder, false); } /* * MPLS label forwarding table change via netlink interface, using dataplane * context information. */ ssize_t netlink_mpls_multipath_msg_encode(int cmd, struct zebra_dplane_ctx *ctx, void *buf, size_t buflen) { mpls_lse_t lse; const struct nhlfe_list_head *head; const struct zebra_nhlfe *nhlfe; struct nexthop *nexthop = NULL; unsigned int nexthop_num; const char *routedesc; int route_type; struct prefix p = {0}; struct nlsock *nl = kernel_netlink_nlsock_lookup(dplane_ctx_get_ns_sock(ctx)); struct { struct nlmsghdr n; struct rtmsg r; char buf[0]; } *req = buf; if (buflen < sizeof(*req)) return 0; memset(req, 0, sizeof(*req)); /* * Count # nexthops so we can decide whether to use singlepath * or multipath case. */ nexthop_num = 0; head = dplane_ctx_get_nhlfe_list(ctx); frr_each(nhlfe_list_const, head, nhlfe) { nexthop = nhlfe->nexthop; if (!nexthop) continue; if (cmd == RTM_NEWROUTE) { /* Count all selected NHLFEs */ if (CHECK_FLAG(nhlfe->flags, NHLFE_FLAG_SELECTED) && CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ACTIVE)) nexthop_num++; } else { /* DEL */ /* Count all installed NHLFEs */ if (CHECK_FLAG(nhlfe->flags, NHLFE_FLAG_INSTALLED) && CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB)) nexthop_num++; } } if ((nexthop_num == 0) || (!dplane_ctx_get_best_nhlfe(ctx) && (cmd != RTM_DELROUTE))) return 0; req->n.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg)); req->n.nlmsg_flags = NLM_F_CREATE | NLM_F_REQUEST; req->n.nlmsg_type = cmd; req->n.nlmsg_pid = nl->snl.nl_pid; req->r.rtm_family = AF_MPLS; req->r.rtm_table = RT_TABLE_MAIN; req->r.rtm_dst_len = MPLS_LABEL_LEN_BITS; req->r.rtm_scope = RT_SCOPE_UNIVERSE; req->r.rtm_type = RTN_UNICAST; if (cmd == RTM_NEWROUTE) { /* We do a replace to handle update. */ req->n.nlmsg_flags |= NLM_F_REPLACE; /* set the protocol value if installing */ route_type = re_type_from_lsp_type( dplane_ctx_get_best_nhlfe(ctx)->type); req->r.rtm_protocol = zebra2proto(route_type); } /* Fill destination */ lse = mpls_lse_encode(dplane_ctx_get_in_label(ctx), 0, 0, 1); if (!nl_attr_put(&req->n, buflen, RTA_DST, &lse, sizeof(mpls_lse_t))) return 0; /* Fill nexthops (paths) based on single-path or multipath. The paths * chosen depend on the operation. */ if (nexthop_num == 1) { routedesc = "single-path"; _netlink_mpls_debug(cmd, dplane_ctx_get_in_label(ctx), routedesc); nexthop_num = 0; frr_each(nhlfe_list_const, head, nhlfe) { nexthop = nhlfe->nexthop; if (!nexthop) continue; if ((cmd == RTM_NEWROUTE && (CHECK_FLAG(nhlfe->flags, NHLFE_FLAG_SELECTED) && CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ACTIVE))) || (cmd == RTM_DELROUTE && (CHECK_FLAG(nhlfe->flags, NHLFE_FLAG_INSTALLED) && CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB)))) { /* Add the gateway */ if (!_netlink_mpls_build_singlepath( &p, routedesc, nhlfe, &req->n, &req->r, buflen, cmd)) return false; nexthop_num++; break; } } } else { /* Multipath case */ struct rtattr *nest; const union g_addr *src1 = NULL; nest = nl_attr_nest(&req->n, buflen, RTA_MULTIPATH); if (!nest) return 0; routedesc = "multipath"; _netlink_mpls_debug(cmd, dplane_ctx_get_in_label(ctx), routedesc); nexthop_num = 0; frr_each(nhlfe_list_const, head, nhlfe) { nexthop = nhlfe->nexthop; if (!nexthop) continue; if ((cmd == RTM_NEWROUTE && (CHECK_FLAG(nhlfe->flags, NHLFE_FLAG_SELECTED) && CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ACTIVE))) || (cmd == RTM_DELROUTE && (CHECK_FLAG(nhlfe->flags, NHLFE_FLAG_INSTALLED) && CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB)))) { nexthop_num++; /* Build the multipath */ if (!_netlink_mpls_build_multipath( &p, routedesc, nhlfe, &req->n, buflen, &req->r, &src1)) return 0; } } /* Add the multipath */ nl_attr_nest_end(&req->n, nest); } return NLMSG_ALIGN(req->n.nlmsg_len); } /**************************************************************************** * This code was developed in a branch that didn't have dplane APIs for * MAC updates. Hence the use of the legacy style. It will be moved to * the new dplane style pre-merge to master. XXX */ static int netlink_fdb_nh_update(uint32_t nh_id, struct in_addr vtep_ip) { struct { struct nlmsghdr n; struct nhmsg nhm; char buf[256]; } req; int cmd = RTM_NEWNEXTHOP; struct zebra_vrf *zvrf; struct zebra_ns *zns; zvrf = zebra_vrf_get_evpn(); zns = zvrf->zns; memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct nhmsg)); req.n.nlmsg_flags = NLM_F_REQUEST; req.n.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE); req.n.nlmsg_type = cmd; req.nhm.nh_family = AF_INET; if (!nl_attr_put32(&req.n, sizeof(req), NHA_ID, nh_id)) return -1; if (!nl_attr_put(&req.n, sizeof(req), NHA_FDB, NULL, 0)) return -1; if (!nl_attr_put(&req.n, sizeof(req), NHA_GATEWAY, &vtep_ip, IPV4_MAX_BYTELEN)) return -1; if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_EVPN_MH_NH) { zlog_debug("Tx %s fdb-nh 0x%x %pI4", nl_msg_type_to_str(cmd), nh_id, &vtep_ip); } return netlink_talk(netlink_talk_filter, &req.n, &zns->netlink_cmd, zns, false); } static int netlink_fdb_nh_del(uint32_t nh_id) { struct { struct nlmsghdr n; struct nhmsg nhm; char buf[256]; } req; int cmd = RTM_DELNEXTHOP; struct zebra_vrf *zvrf; struct zebra_ns *zns; zvrf = zebra_vrf_get_evpn(); zns = zvrf->zns; memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct nhmsg)); req.n.nlmsg_flags = NLM_F_REQUEST; req.n.nlmsg_type = cmd; req.nhm.nh_family = AF_UNSPEC; if (!nl_attr_put32(&req.n, sizeof(req), NHA_ID, nh_id)) return -1; if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_EVPN_MH_NH) { zlog_debug("Tx %s fdb-nh 0x%x", nl_msg_type_to_str(cmd), nh_id); } return netlink_talk(netlink_talk_filter, &req.n, &zns->netlink_cmd, zns, false); } static int netlink_fdb_nhg_update(uint32_t nhg_id, uint32_t nh_cnt, struct nh_grp *nh_ids) { struct { struct nlmsghdr n; struct nhmsg nhm; char buf[256]; } req; int cmd = RTM_NEWNEXTHOP; struct zebra_vrf *zvrf; struct zebra_ns *zns; struct nexthop_grp grp[nh_cnt]; uint32_t i; zvrf = zebra_vrf_get_evpn(); zns = zvrf->zns; memset(&req, 0, sizeof(req)); req.n.nlmsg_len = NLMSG_LENGTH(sizeof(struct nhmsg)); req.n.nlmsg_flags = NLM_F_REQUEST; req.n.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE); req.n.nlmsg_type = cmd; req.nhm.nh_family = AF_UNSPEC; if (!nl_attr_put32(&req.n, sizeof(req), NHA_ID, nhg_id)) return -1; if (!nl_attr_put(&req.n, sizeof(req), NHA_FDB, NULL, 0)) return -1; memset(&grp, 0, sizeof(grp)); for (i = 0; i < nh_cnt; ++i) { grp[i].id = nh_ids[i].id; grp[i].weight = nh_ids[i].weight; } if (!nl_attr_put(&req.n, sizeof(req), NHA_GROUP, grp, nh_cnt * sizeof(struct nexthop_grp))) return -1; if (IS_ZEBRA_DEBUG_KERNEL || IS_ZEBRA_DEBUG_EVPN_MH_NH) { char vtep_str[ES_VTEP_LIST_STR_SZ]; char nh_buf[16]; vtep_str[0] = '\0'; for (i = 0; i < nh_cnt; ++i) { snprintf(nh_buf, sizeof(nh_buf), "%u ", grp[i].id); strlcat(vtep_str, nh_buf, sizeof(vtep_str)); } zlog_debug("Tx %s fdb-nhg 0x%x %s", nl_msg_type_to_str(cmd), nhg_id, vtep_str); } return netlink_talk(netlink_talk_filter, &req.n, &zns->netlink_cmd, zns, false); } static int netlink_fdb_nhg_del(uint32_t nhg_id) { return netlink_fdb_nh_del(nhg_id); } int kernel_upd_mac_nh(uint32_t nh_id, struct in_addr vtep_ip) { return netlink_fdb_nh_update(nh_id, vtep_ip); } int kernel_del_mac_nh(uint32_t nh_id) { return netlink_fdb_nh_del(nh_id); } int kernel_upd_mac_nhg(uint32_t nhg_id, uint32_t nh_cnt, struct nh_grp *nh_ids) { return netlink_fdb_nhg_update(nhg_id, nh_cnt, nh_ids); } int kernel_del_mac_nhg(uint32_t nhg_id) { return netlink_fdb_nhg_del(nhg_id); } #endif /* HAVE_NETLINK */