diff options
Diffstat (limited to 'bgpd/rfapi/vnc_zebra.c')
-rw-r--r-- | bgpd/rfapi/vnc_zebra.c | 887 |
1 files changed, 887 insertions, 0 deletions
diff --git a/bgpd/rfapi/vnc_zebra.c b/bgpd/rfapi/vnc_zebra.c new file mode 100644 index 0000000..c886bee --- /dev/null +++ b/bgpd/rfapi/vnc_zebra.c @@ -0,0 +1,887 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * + * Copyright 2009-2016, LabN Consulting, L.L.C. + * + */ + +/* + * File: vnc_zebra.c + * Purpose: Handle exchange of routes between VNC and Zebra + */ + +#include "lib/zebra.h" +#include "lib/prefix.h" +#include "lib/agg_table.h" +#include "lib/log.h" +#include "lib/command.h" +#include "lib/zclient.h" +#include "lib/stream.h" +#include "lib/ringbuf.h" +#include "lib/memory.h" +#include "lib/lib_errors.h" + +#include "bgpd/bgpd.h" +#include "bgpd/bgp_ecommunity.h" +#include "bgpd/bgp_route.h" +#include "bgpd/bgp_debug.h" +#include "bgpd/bgp_advertise.h" + +#include "bgpd/rfapi/bgp_rfapi_cfg.h" +#include "bgpd/rfapi/rfapi.h" +#include "bgpd/rfapi/rfapi_import.h" +#include "bgpd/rfapi/rfapi_private.h" +#include "bgpd/rfapi/vnc_zebra.h" +#include "bgpd/rfapi/rfapi_vty.h" +#include "bgpd/rfapi/rfapi_backend.h" +#include "bgpd/rfapi/vnc_debug.h" + +static struct rfapi_descriptor vncHD1VR; /* Single-VR export dummy nve descr */ +static struct zclient *zclient_vnc = NULL; + +/*********************************************************************** + * REDISTRIBUTE: Zebra sends updates/withdraws to BGPD + ***********************************************************************/ + +/* + * Routes coming from zebra get added to VNC here + */ +static void vnc_redistribute_add(struct prefix *p, uint32_t metric, + uint8_t type) +{ + struct bgp *bgp = bgp_get_default(); + struct prefix_rd prd; + struct rfapi_ip_addr vnaddr; + afi_t afi; + uint32_t local_pref = + rfp_cost_to_localpref(metric > 255 ? 255 : metric); + + if (!bgp) + return; + + if (!bgp->rfapi_cfg) { + vnc_zlog_debug_verbose("%s: bgp->rfapi_cfg is NULL, skipping", + __func__); + return; + } + + afi = family2afi(p->family); + if (!afi) { + vnc_zlog_debug_verbose("%s: unknown prefix address family %d", + __func__, p->family); + return; + } + + if (!bgp->rfapi_cfg->redist[afi][type]) { + vnc_zlog_debug_verbose( + "%s: bgp->rfapi_cfg->redist[afi=%d][type=%d] is 0, skipping", + __func__, afi, type); + return; + } + if (!bgp->rfapi_cfg->rfg_redist) { + vnc_zlog_debug_verbose("%s: no redist nve group, skipping", + __func__); + return; + } + + /* + * Assume nve group's configured VN address prefix is a host + * route which also happens to give the NVE VN address to use + * for redistributing into VNC. + */ + vnaddr.addr_family = bgp->rfapi_cfg->rfg_redist->vn_prefix.family; + switch (bgp->rfapi_cfg->rfg_redist->vn_prefix.family) { + case AF_INET: + if (bgp->rfapi_cfg->rfg_redist->vn_prefix.prefixlen + != IPV4_MAX_BITLEN) { + vnc_zlog_debug_verbose( + "%s: redist nve group VN prefix len (%d) != 32, skipping", + __func__, + bgp->rfapi_cfg->rfg_redist->vn_prefix + .prefixlen); + return; + } + vnaddr.addr.v4 = + bgp->rfapi_cfg->rfg_redist->vn_prefix.u.prefix4; + break; + case AF_INET6: + if (bgp->rfapi_cfg->rfg_redist->vn_prefix.prefixlen + != IPV6_MAX_BITLEN) { + vnc_zlog_debug_verbose( + "%s: redist nve group VN prefix len (%d) != 128, skipping", + __func__, + bgp->rfapi_cfg->rfg_redist->vn_prefix + .prefixlen); + return; + } + vnaddr.addr.v6 = + bgp->rfapi_cfg->rfg_redist->vn_prefix.u.prefix6; + break; + default: + vnc_zlog_debug_verbose( + "%s: no redist nve group VN host prefix configured, skipping", + __func__); + return; + } + + /* + * Assume nve group's configured UN address prefix is a host + * route which also happens to give the NVE UN address to use + * for redistributing into VNC. + */ + + /* + * Set UN address in dummy nve descriptor so add_vnc_route + * can use it in VNC tunnel SubTLV + */ + { + struct rfapi_ip_prefix pfx_un; + + rfapiQprefix2Rprefix(&bgp->rfapi_cfg->rfg_redist->un_prefix, + &pfx_un); + + switch (pfx_un.prefix.addr_family) { + case AF_INET: + if (pfx_un.length != IPV4_MAX_BITLEN) { + vnc_zlog_debug_verbose( + "%s: redist nve group UN prefix len (%d) != 32, skipping", + __func__, pfx_un.length); + return; + } + break; + case AF_INET6: + if (pfx_un.length != IPV6_MAX_BITLEN) { + vnc_zlog_debug_verbose( + "%s: redist nve group UN prefix len (%d) != 128, skipping", + __func__, pfx_un.length); + return; + } + break; + default: + vnc_zlog_debug_verbose( + "%s: no redist nve group UN host prefix configured, skipping", + __func__); + return; + } + + vncHD1VR.un_addr = pfx_un.prefix; + + if (!vncHD1VR.peer) { + /* + * Same setup as in rfapi_open() + */ + vncHD1VR.peer = peer_new(bgp); + vncHD1VR.peer->connection->status = + Established; /* keep bgp core happy */ + + bgp_peer_connection_buffers_free( + vncHD1VR.peer->connection); + + /* base code assumes have valid host pointer */ + vncHD1VR.peer->host = + XSTRDUP(MTYPE_BGP_PEER_HOST, ".zebra."); + + /* Mark peer as belonging to HD */ + SET_FLAG(vncHD1VR.peer->flags, PEER_FLAG_IS_RFAPI_HD); + } + } + + memset(&prd, 0, sizeof(prd)); + prd = bgp->rfapi_cfg->rfg_redist->rd; + prd.family = AF_UNSPEC; + prd.prefixlen = 64; + + add_vnc_route(&vncHD1VR, /* cookie + UN addr */ + bgp, SAFI_MPLS_VPN, p, &prd, &vnaddr, &local_pref, + &(bgp->rfapi_cfg->redist_lifetime), + NULL, /* RFP options */ + NULL, /* struct rfapi_un_option */ + NULL, /* struct rfapi_vn_option */ + bgp->rfapi_cfg->rfg_redist->rt_export_list, NULL, + NULL, /* label: default */ + type, BGP_ROUTE_REDISTRIBUTE, 0); /* flags */ +} + +/* + * Route deletions from zebra propagate to VNC here + */ +static void vnc_redistribute_delete(struct prefix *p, uint8_t type) +{ + struct bgp *bgp = bgp_get_default(); + struct prefix_rd prd; + afi_t afi; + + if (!bgp) + return; + + if (!bgp->rfapi_cfg) { + vnc_zlog_debug_verbose("%s: bgp->rfapi_cfg is NULL, skipping", + __func__); + return; + } + afi = family2afi(p->family); + if (!afi) { + vnc_zlog_debug_verbose("%s: unknown prefix address family %d", + __func__, p->family); + return; + } + if (!bgp->rfapi_cfg->redist[afi][type]) { + vnc_zlog_debug_verbose( + "%s: bgp->rfapi_cfg->redist[afi=%d][type=%d] is 0, skipping", + __func__, afi, type); + return; + } + if (!bgp->rfapi_cfg->rfg_redist) { + vnc_zlog_debug_verbose("%s: no redist nve group, skipping", + __func__); + return; + } + + memset(&prd, 0, sizeof(prd)); + prd = bgp->rfapi_cfg->rfg_redist->rd; + prd.family = AF_UNSPEC; + prd.prefixlen = 64; + + del_vnc_route(&vncHD1VR, /* use dummy ptr as cookie */ + vncHD1VR.peer, bgp, SAFI_MPLS_VPN, p, &prd, type, + BGP_ROUTE_REDISTRIBUTE, NULL, 0); +} + +/* + * Flush all redistributed routes of type <type> + */ +static void vnc_redistribute_withdraw(struct bgp *bgp, afi_t afi, uint8_t type) +{ + struct prefix_rd prd; + struct bgp_table *table; + struct bgp_dest *pdest; + struct bgp_dest *dest; + + vnc_zlog_debug_verbose("%s: entry", __func__); + + if (!bgp) + return; + if (!bgp->rfapi_cfg) { + vnc_zlog_debug_verbose("%s: bgp->rfapi_cfg is NULL, skipping", + __func__); + return; + } + + /* + * Loop over all the RDs + */ + for (pdest = bgp_table_top(bgp->rib[afi][SAFI_MPLS_VPN]); pdest; + pdest = bgp_route_next(pdest)) { + const struct prefix *pdest_p = bgp_dest_get_prefix(pdest); + + memset(&prd, 0, sizeof(prd)); + prd.family = AF_UNSPEC; + prd.prefixlen = 64; + memcpy(prd.val, pdest_p->u.val, 8); + + /* This is the per-RD table of prefixes */ + table = bgp_dest_get_bgp_table_info(pdest); + if (!table) + continue; + + for (dest = bgp_table_top(table); dest; + dest = bgp_route_next(dest)) { + + struct bgp_path_info *ri; + + for (ri = bgp_dest_get_bgp_path_info(dest); ri; + ri = ri->next) { + if (ri->type + == type) { /* has matching redist type */ + break; + } + } + if (ri) { + del_vnc_route( + &vncHD1VR, /* use dummy ptr as cookie */ + vncHD1VR.peer, bgp, SAFI_MPLS_VPN, + bgp_dest_get_prefix(dest), &prd, type, + BGP_ROUTE_REDISTRIBUTE, NULL, 0); + } + } + } + vnc_zlog_debug_verbose("%s: return", __func__); +} + +/* + * Zebra route add and delete treatment. + * + * Assumes 1 nexthop + */ +static int vnc_zebra_read_route(ZAPI_CALLBACK_ARGS) +{ + struct zapi_route api; + int add; + + if (zapi_route_decode(zclient->ibuf, &api) < 0) + return -1; + + /* we completely ignore srcdest routes for now. */ + if (CHECK_FLAG(api.message, ZAPI_MESSAGE_SRCPFX)) + return 0; + + add = (cmd == ZEBRA_REDISTRIBUTE_ROUTE_ADD); + if (add) + vnc_redistribute_add(&api.prefix, api.metric, api.type); + else + vnc_redistribute_delete(&api.prefix, api.type); + + if (BGP_DEBUG(zebra, ZEBRA)) + vnc_zlog_debug_verbose( + "%s: Zebra rcvd: route delete %s %pFX metric %u", + __func__, zebra_route_string(api.type), &api.prefix, + api.metric); + + return 0; +} + +/*********************************************************************** + * vnc_bgp_zebra_*: VNC sends updates/withdraws to Zebra + ***********************************************************************/ + +/* + * low-level message builder + */ +static void vnc_zebra_route_msg(const struct prefix *p, unsigned int nhp_count, + void *nhp_ary, int add) /* 1 = add, 0 = del */ +{ + struct zapi_route api; + struct zapi_nexthop *api_nh; + int i; + struct in_addr **nhp_ary4 = nhp_ary; + struct in6_addr **nhp_ary6 = nhp_ary; + + if (!nhp_count) { + vnc_zlog_debug_verbose("%s: empty nexthop list, skipping", + __func__); + return; + } + + memset(&api, 0, sizeof(api)); + api.vrf_id = VRF_DEFAULT; + api.type = ZEBRA_ROUTE_VNC; + api.safi = SAFI_UNICAST; + api.prefix = *p; + + /* Nexthops */ + SET_FLAG(api.message, ZAPI_MESSAGE_NEXTHOP); + api.nexthop_num = MIN(nhp_count, multipath_num); + for (i = 0; i < api.nexthop_num; i++) { + + api_nh = &api.nexthops[i]; + api_nh->vrf_id = VRF_DEFAULT; + switch (p->family) { + case AF_INET: + memcpy(&api_nh->gate.ipv4, nhp_ary4[i], + sizeof(api_nh->gate.ipv4)); + api_nh->type = NEXTHOP_TYPE_IPV4; + break; + case AF_INET6: + memcpy(&api_nh->gate.ipv6, nhp_ary6[i], + sizeof(api_nh->gate.ipv6)); + api_nh->type = NEXTHOP_TYPE_IPV6; + break; + } + } + + if (BGP_DEBUG(zebra, ZEBRA)) + vnc_zlog_debug_verbose( + "%s: Zebra send: route %s %pFX, nhp_count=%d", __func__, + (add ? "add" : "del"), &api.prefix, nhp_count); + + zclient_route_send((add ? ZEBRA_ROUTE_ADD : ZEBRA_ROUTE_DELETE), + zclient_vnc, &api); +} + + +static void +nve_list_to_nh_array(uint8_t family, struct list *nve_list, + unsigned int *nh_count_ret, + void **nh_ary_ret, /* returned address array */ + void **nhp_ary_ret) /* returned pointer array */ +{ + int nve_count = listcount(nve_list); + + *nh_count_ret = 0; + *nh_ary_ret = NULL; + *nhp_ary_ret = NULL; + + if (!nve_count) { + vnc_zlog_debug_verbose("%s: empty nve_list, skipping", + __func__); + return; + } + + if (family == AF_INET) { + struct listnode *ln; + struct in_addr *iap; + struct in_addr **v; + + /* + * Array of nexthop addresses + */ + *nh_ary_ret = + XCALLOC(MTYPE_TMP, nve_count * sizeof(struct in_addr)); + + /* + * Array of pointers to nexthop addresses + */ + *nhp_ary_ret = XCALLOC(MTYPE_TMP, + nve_count * sizeof(struct in_addr *)); + iap = *nh_ary_ret; + v = *nhp_ary_ret; + + for (ln = listhead(nve_list); ln; ln = listnextnode(ln)) { + + struct rfapi_descriptor *irfd; + struct prefix nhp; + + irfd = listgetdata(ln); + + if (rfapiRaddr2Qprefix(&irfd->vn_addr, &nhp)) + continue; + + *iap = nhp.u.prefix4; + *v = iap; + vnc_zlog_debug_verbose( + "%s: ipadr: (%p)<-0x%x, ptr: (%p)<-%p", + __func__, iap, nhp.u.prefix4.s_addr, v, iap); + + ++iap; + ++v; + ++*nh_count_ret; + } + + } else if (family == AF_INET6) { + + struct listnode *ln; + + *nh_ary_ret = + XCALLOC(MTYPE_TMP, nve_count * sizeof(struct in6_addr)); + + *nhp_ary_ret = XCALLOC(MTYPE_TMP, + nve_count * sizeof(struct in6_addr *)); + + for (ln = listhead(nve_list); ln; ln = listnextnode(ln)) { + + struct rfapi_descriptor *irfd; + struct in6_addr *iap = *nh_ary_ret; + struct in6_addr **v = *nhp_ary_ret; + struct prefix nhp; + + irfd = listgetdata(ln); + + if (rfapiRaddr2Qprefix(&irfd->vn_addr, &nhp)) + continue; + + *iap = nhp.u.prefix6; + *v = iap; + + ++iap; + ++v; + ++*nh_count_ret; + } + } +} + +static void import_table_to_nve_list_zebra(struct bgp *bgp, + struct rfapi_import_table *it, + struct list **nves, uint8_t family) +{ + struct listnode *node; + struct rfapi_rfg_name *rfgn; + + /* + * Loop over the list of NVE-Groups configured for + * exporting to direct-bgp. + * + * Build a list of NVEs that use this import table + */ + *nves = NULL; + for (ALL_LIST_ELEMENTS_RO(bgp->rfapi_cfg->rfg_export_zebra_l, node, + rfgn)) { + + /* + * If this NVE-Group's import table matches the current one + */ + if (rfgn->rfg && rfgn->rfg->nves + && rfgn->rfg->rfapi_import_table == it) { + + nve_group_to_nve_list(rfgn->rfg, nves, family); + } + } +} + +static void vnc_zebra_add_del_prefix(struct bgp *bgp, + struct rfapi_import_table *import_table, + struct agg_node *rn, + int add) /* !0 = add, 0 = del */ +{ + struct list *nves; + const struct prefix *p = agg_node_get_prefix(rn); + unsigned int nexthop_count = 0; + void *nh_ary = NULL; + void *nhp_ary = NULL; + + vnc_zlog_debug_verbose("%s: entry, add=%d", __func__, add); + + if (zclient_vnc->sock < 0) + return; + + if (p->family != AF_INET && p->family != AF_INET6) { + flog_err(EC_LIB_DEVELOPMENT, + "%s: invalid route node addr family", __func__); + return; + } + + if (!vrf_bitmap_check(&zclient_vnc->redist[family2afi(p->family)] + [ZEBRA_ROUTE_VNC], + VRF_DEFAULT)) + return; + + if (!bgp->rfapi_cfg) { + vnc_zlog_debug_verbose("%s: bgp->rfapi_cfg is NULL, skipping", + __func__); + return; + } + if (!listcount(bgp->rfapi_cfg->rfg_export_zebra_l)) { + vnc_zlog_debug_verbose( + "%s: no zebra export nve group, skipping", __func__); + return; + } + + import_table_to_nve_list_zebra(bgp, import_table, &nves, p->family); + + if (nves) { + nve_list_to_nh_array(p->family, nves, &nexthop_count, &nh_ary, + &nhp_ary); + + list_delete(&nves); + + if (nexthop_count) + vnc_zebra_route_msg(p, nexthop_count, nhp_ary, add); + } + + XFREE(MTYPE_TMP, nhp_ary); + XFREE(MTYPE_TMP, nh_ary); +} + +void vnc_zebra_add_prefix(struct bgp *bgp, + struct rfapi_import_table *import_table, + struct agg_node *rn) +{ + vnc_zebra_add_del_prefix(bgp, import_table, rn, 1); +} + +void vnc_zebra_del_prefix(struct bgp *bgp, + struct rfapi_import_table *import_table, + struct agg_node *rn) +{ + vnc_zebra_add_del_prefix(bgp, import_table, rn, 0); +} + + +static void vnc_zebra_add_del_nve(struct bgp *bgp, struct rfapi_descriptor *rfd, + int add) /* 0 = del, !0 = add */ +{ + struct listnode *node; + struct rfapi_rfg_name *rfgn; + struct rfapi_nve_group_cfg *rfg = rfd->rfg; + afi_t afi = family2afi(rfd->vn_addr.addr_family); + struct prefix nhp; + void *pAddr; + + vnc_zlog_debug_verbose("%s: entry, add=%d", __func__, add); + + if (zclient_vnc->sock < 0) + return; + + if (!vrf_bitmap_check(&zclient_vnc->redist[afi][ZEBRA_ROUTE_VNC], + VRF_DEFAULT)) + return; + + if (afi != AFI_IP && afi != AFI_IP6) { + flog_err(EC_LIB_DEVELOPMENT, "%s: invalid vn addr family", + __func__); + return; + } + + if (!bgp) + return; + if (!bgp->rfapi_cfg) { + vnc_zlog_debug_verbose("%s: bgp->rfapi_cfg is NULL, skipping", + __func__); + return; + } + + if (rfapiRaddr2Qprefix(&rfd->vn_addr, &nhp)) { + vnc_zlog_debug_verbose("%s: can't convert vn address, skipping", + __func__); + return; + } + + pAddr = &nhp.u.val; + + /* + * Loop over the list of NVE-Groups configured for + * exporting to zebra and see if this new NVE's + * group is among them. + */ + for (ALL_LIST_ELEMENTS_RO(bgp->rfapi_cfg->rfg_export_zebra_l, node, + rfgn)) { + + /* + * Yes, this NVE's group is configured for export to zebra + */ + if (rfgn->rfg == rfg) { + + struct agg_table *rt = NULL; + struct agg_node *rn; + struct rfapi_import_table *import_table; + import_table = rfg->rfapi_import_table; + + vnc_zlog_debug_verbose( + "%s: this nve's group is in zebra export list", + __func__); + + rt = import_table->imported_vpn[afi]; + + /* + * Walk the NVE-Group's VNC Import table + */ + for (rn = agg_route_top(rt); rn; + rn = agg_route_next(rn)) { + if (!rn->info) + continue; + + vnc_zlog_debug_verbose("%s: sending %s", + __func__, + (add ? "add" : "del")); + vnc_zebra_route_msg(agg_node_get_prefix(rn), 1, + &pAddr, add); + } + } + } +} + +void vnc_zebra_add_nve(struct bgp *bgp, struct rfapi_descriptor *rfd) +{ + vnc_zebra_add_del_nve(bgp, rfd, 1); +} + +void vnc_zebra_del_nve(struct bgp *bgp, struct rfapi_descriptor *rfd) +{ + vnc_zebra_add_del_nve(bgp, rfd, 0); +} + +static void vnc_zebra_add_del_group_afi(struct bgp *bgp, + struct rfapi_nve_group_cfg *rfg, + afi_t afi, int add) +{ + struct agg_table *rt = NULL; + struct agg_node *rn; + struct rfapi_import_table *import_table; + uint8_t family = afi2family(afi); + + struct list *nves = NULL; + unsigned int nexthop_count = 0; + void *nh_ary = NULL; + void *nhp_ary = NULL; + + vnc_zlog_debug_verbose("%s: entry", __func__); + import_table = rfg->rfapi_import_table; + if (!import_table) { + vnc_zlog_debug_verbose( + "%s: import table not defined, returning", __func__); + return; + } + + if (afi == AFI_IP || afi == AFI_IP6) { + rt = import_table->imported_vpn[afi]; + } else { + flog_err(EC_LIB_DEVELOPMENT, "%s: bad afi %d", __func__, afi); + return; + } + + if (!family) { + flog_err(EC_LIB_DEVELOPMENT, "%s: computed bad family: %d", + __func__, family); + return; + } + + if (!rfg->nves) { + /* avoid segfault below if list doesn't exist */ + vnc_zlog_debug_verbose("%s: no NVEs in this group", __func__); + return; + } + + nve_group_to_nve_list(rfg, &nves, family); + if (nves) { + vnc_zlog_debug_verbose("%s: have nves", __func__); + nve_list_to_nh_array(family, nves, &nexthop_count, &nh_ary, + &nhp_ary); + + vnc_zlog_debug_verbose("%s: family: %d, nve count: %d", + __func__, family, nexthop_count); + + list_delete(&nves); + + if (nexthop_count) { + /* + * Walk the NVE-Group's VNC Import table + */ + for (rn = agg_route_top(rt); rn; + rn = agg_route_next(rn)) { + if (rn->info) { + vnc_zebra_route_msg( + agg_node_get_prefix(rn), + nexthop_count, nhp_ary, add); + } + } + } + XFREE(MTYPE_TMP, nhp_ary); + XFREE(MTYPE_TMP, nh_ary); + } +} + +void vnc_zebra_add_group(struct bgp *bgp, struct rfapi_nve_group_cfg *rfg) +{ + vnc_zebra_add_del_group_afi(bgp, rfg, AFI_IP, 1); + vnc_zebra_add_del_group_afi(bgp, rfg, AFI_IP6, 1); +} + +void vnc_zebra_del_group(struct bgp *bgp, struct rfapi_nve_group_cfg *rfg) +{ + vnc_zlog_debug_verbose("%s: entry", __func__); + vnc_zebra_add_del_group_afi(bgp, rfg, AFI_IP, 0); + vnc_zebra_add_del_group_afi(bgp, rfg, AFI_IP6, 0); +} + +void vnc_zebra_reexport_group_afi(struct bgp *bgp, + struct rfapi_nve_group_cfg *rfg, afi_t afi) +{ + struct listnode *node; + struct rfapi_rfg_name *rfgn; + + for (ALL_LIST_ELEMENTS_RO(bgp->rfapi_cfg->rfg_export_zebra_l, node, + rfgn)) { + + if (rfgn->rfg == rfg) { + vnc_zebra_add_del_group_afi(bgp, rfg, afi, 0); + vnc_zebra_add_del_group_afi(bgp, rfg, afi, 1); + break; + } + } +} + + +/*********************************************************************** + * CONTROL INTERFACE + ***********************************************************************/ + + +/* Other routes redistribution into BGP. */ +int vnc_redistribute_set(struct bgp *bgp, afi_t afi, int type) +{ + if (!bgp->rfapi_cfg) { + return CMD_WARNING_CONFIG_FAILED; + } + + /* Set flag to BGP instance. */ + bgp->rfapi_cfg->redist[afi][type] = 1; + + // bgp->redist[afi][type] = 1; + + /* Return if already redistribute flag is set. */ + if (vrf_bitmap_check(&zclient_vnc->redist[afi][type], VRF_DEFAULT)) + return CMD_WARNING_CONFIG_FAILED; + + vrf_bitmap_set(&zclient_vnc->redist[afi][type], VRF_DEFAULT); + + // vrf_bitmap_set(&zclient_vnc->redist[afi][type], VRF_DEFAULT); + + /* Return if zebra connection is not established. */ + if (zclient_vnc->sock < 0) + return CMD_WARNING_CONFIG_FAILED; + + if (BGP_DEBUG(zebra, ZEBRA)) + vnc_zlog_debug_verbose("Zebra send: redistribute add %s", + zebra_route_string(type)); + + /* Send distribute add message to zebra. */ + zebra_redistribute_send(ZEBRA_REDISTRIBUTE_ADD, zclient_vnc, afi, type, + 0, VRF_DEFAULT); + + return CMD_SUCCESS; +} + +/* Unset redistribution. */ +int vnc_redistribute_unset(struct bgp *bgp, afi_t afi, int type) +{ + vnc_zlog_debug_verbose("%s: type=%d entry", __func__, type); + + if (!bgp->rfapi_cfg) { + vnc_zlog_debug_verbose("%s: return (no rfapi_cfg)", __func__); + return CMD_WARNING_CONFIG_FAILED; + } + + /* Unset flag from BGP instance. */ + bgp->rfapi_cfg->redist[afi][type] = 0; + + /* Return if zebra connection is disabled. */ + if (!vrf_bitmap_check(&zclient_vnc->redist[afi][type], VRF_DEFAULT)) + return CMD_WARNING_CONFIG_FAILED; + vrf_bitmap_unset(&zclient_vnc->redist[afi][type], VRF_DEFAULT); + + if (bgp->rfapi_cfg->redist[AFI_IP][type] == 0 + && bgp->rfapi_cfg->redist[AFI_IP6][type] == 0 + && zclient_vnc->sock >= 0) { + /* Send distribute delete message to zebra. */ + if (BGP_DEBUG(zebra, ZEBRA)) + vnc_zlog_debug_verbose( + "Zebra send: redistribute delete %s", + zebra_route_string(type)); + zebra_redistribute_send(ZEBRA_REDISTRIBUTE_DELETE, zclient_vnc, + afi, type, 0, VRF_DEFAULT); + } + + /* Withdraw redistributed routes from current BGP's routing table. */ + vnc_redistribute_withdraw(bgp, afi, type); + + vnc_zlog_debug_verbose("%s: return", __func__); + + return CMD_SUCCESS; +} + +extern struct zebra_privs_t bgpd_privs; + +static zclient_handler *const vnc_handlers[] = { + [ZEBRA_REDISTRIBUTE_ROUTE_ADD] = vnc_zebra_read_route, + [ZEBRA_REDISTRIBUTE_ROUTE_DEL] = vnc_zebra_read_route, +}; + +/* + * Modeled after bgp_zebra.c'bgp_zebra_init() + * Charriere asks, "Is it possible to carry two?" + */ +void vnc_zebra_init(struct event_loop *master) +{ + /* Set default values. */ + zclient_vnc = zclient_new(master, &zclient_options_default, + vnc_handlers, array_size(vnc_handlers)); + zclient_init(zclient_vnc, ZEBRA_ROUTE_VNC, 0, &bgpd_privs); +} + +void vnc_zebra_destroy(void) +{ + if (zclient_vnc == NULL) + return; + zclient_stop(zclient_vnc); + zclient_free(zclient_vnc); + zclient_vnc = NULL; +} |