diff options
Diffstat (limited to 'ldpd/interface.c')
-rw-r--r-- | ldpd/interface.c | 983 |
1 files changed, 983 insertions, 0 deletions
diff --git a/ldpd/interface.c b/ldpd/interface.c new file mode 100644 index 0000000..392e254 --- /dev/null +++ b/ldpd/interface.c @@ -0,0 +1,983 @@ +/* $OpenBSD$ */ + +/* + * Copyright (c) 2013, 2016 Renato Westphal <renato@openbsd.org> + * Copyright (c) 2005 Claudio Jeker <claudio@openbsd.org> + * Copyright (c) 2004, 2005, 2008 Esben Norby <norby@openbsd.org> + * + * Permission to use, copy, modify, and distribute this software for any + * purpose with or without fee is hereby granted, provided that the above + * copyright notice and this permission notice appear in all copies. + * + * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES + * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR + * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES + * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN + * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF + * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. + */ + +#include <zebra.h> + +#include "ldpd.h" +#include "ldpe.h" +#include "log.h" +#include "ldp_debug.h" + +#include "sockopt.h" + +static __inline int iface_compare(const struct iface *, const struct iface *); +static struct if_addr *if_addr_new(struct kaddr *); +static struct if_addr *if_addr_lookup(struct if_addr_head *, struct kaddr *); +static int if_start(struct iface *, int); +static int if_reset(struct iface *, int); +static void if_update_af(struct iface_af *); +static void if_hello_timer(struct thread *thread); +static void if_start_hello_timer(struct iface_af *); +static void if_stop_hello_timer(struct iface_af *); +static int if_join_ipv4_group(struct iface *, struct in_addr *); +static int if_leave_ipv4_group(struct iface *, struct in_addr *); +static int if_join_ipv6_group(struct iface *, struct in6_addr *); +static int if_leave_ipv6_group(struct iface *, struct in6_addr *); + +static int ldp_sync_fsm_init(struct iface *iface, int state); +static int ldp_sync_act_iface_start_sync(struct iface *iface); +static void iface_wait_for_ldp_sync_timer(struct thread *thread); +static void start_wait_for_ldp_sync_timer(struct iface *iface); +static void stop_wait_for_ldp_sync_timer(struct iface *iface); +static int ldp_sync_act_ldp_start_sync(struct iface *iface); +static int ldp_sync_act_ldp_complete_sync(struct iface *iface); +static int iface_to_oper_nbr_count(struct iface *iface, unsigned int type); +static void ldp_sync_get_peer_ldp_id(struct iface *iface, + struct in_addr *peer_ldp_id); + +RB_GENERATE(iface_head, iface, entry, iface_compare) + +static __inline int +iface_compare(const struct iface *a, const struct iface *b) +{ + return if_cmp_name_func(a->name, b->name); +} + +struct iface * +if_new(const char *name) +{ + struct iface *iface; + + if ((iface = calloc(1, sizeof(*iface))) == NULL) + fatal("if_new: calloc"); + + strlcpy(iface->name, name, sizeof(iface->name)); + + /* ipv4 */ + iface->ipv4.af = AF_INET; + iface->ipv4.iface = iface; + iface->ipv4.enabled = 0; + + /* ipv6 */ + iface->ipv6.af = AF_INET6; + iface->ipv6.iface = iface; + iface->ipv6.enabled = 0; + + return (iface); +} + +void +ldpe_if_init(struct iface *iface) +{ + log_debug("%s: interface %s", __func__, iface->name); + + LIST_INIT(&iface->addr_list); + + /* ipv4 */ + iface->ipv4.iface = iface; + iface->ipv4.state = IF_STA_DOWN; + RB_INIT(ia_adj_head, &iface->ipv4.adj_tree); + + /* ipv6 */ + iface->ipv6.iface = iface; + iface->ipv6.state = IF_STA_DOWN; + RB_INIT(ia_adj_head, &iface->ipv6.adj_tree); + + /* LGP IGP Sync */ + ldp_sync_fsm_init(iface, LDP_SYNC_STA_NOT_ACH); +} + +void +ldpe_if_exit(struct iface *iface) +{ + struct if_addr *if_addr; + + log_debug("%s: interface %s", __func__, iface->name); + + ldp_sync_fsm(iface, LDP_SYNC_EVT_CONFIG_LDP_OFF); + + if (iface->ipv4.state == IF_STA_ACTIVE) + if_reset(iface, AF_INET); + if (iface->ipv6.state == IF_STA_ACTIVE) + if_reset(iface, AF_INET6); + + while ((if_addr = LIST_FIRST(&iface->addr_list)) != NULL) { + LIST_REMOVE(if_addr, entry); + assert(if_addr != LIST_FIRST(&iface->addr_list)); + free(if_addr); + } +} + +struct iface * +if_lookup(struct ldpd_conf *xconf, ifindex_t ifindex) +{ + struct iface *iface; + + RB_FOREACH(iface, iface_head, &xconf->iface_tree) + if (iface->ifindex == ifindex) + return (iface); + + return (NULL); +} + +struct iface * +if_lookup_name(struct ldpd_conf *xconf, const char *ifname) +{ + struct iface iface; + strlcpy(iface.name, ifname, sizeof(iface.name)); + return (RB_FIND(iface_head, &xconf->iface_tree, &iface)); +} + +void +if_update_info(struct iface *iface, struct kif *kif) +{ + /* get type */ + if (kif->flags & IFF_POINTOPOINT) + iface->type = IF_TYPE_POINTOPOINT; + if (kif->flags & IFF_BROADCAST && + kif->flags & IFF_MULTICAST) + iface->type = IF_TYPE_BROADCAST; + + if (ldpd_process == PROC_LDP_ENGINE && iface->operative && + !kif->operative) + ldp_sync_fsm(iface, LDP_SYNC_EVT_IFACE_SHUTDOWN); + + /* get index and flags */ + iface->ifindex = kif->ifindex; + iface->operative = kif->operative; +} + +struct iface_af * +iface_af_get(struct iface *iface, int af) +{ + switch (af) { + case AF_INET: + return (&iface->ipv4); + case AF_INET6: + return (&iface->ipv6); + default: + fatalx("iface_af_get: unknown af"); + } +} + +static struct if_addr * +if_addr_new(struct kaddr *ka) +{ + struct if_addr *if_addr; + + if ((if_addr = calloc(1, sizeof(*if_addr))) == NULL) + fatal(__func__); + + if_addr->af = ka->af; + if_addr->addr = ka->addr; + if_addr->prefixlen = ka->prefixlen; + if_addr->dstbrd = ka->dstbrd; + + return (if_addr); +} + +static struct if_addr * +if_addr_lookup(struct if_addr_head *addr_list, struct kaddr *ka) +{ + struct if_addr *if_addr; + int af = ka->af; + + LIST_FOREACH(if_addr, addr_list, entry) + if (!ldp_addrcmp(af, &if_addr->addr, &ka->addr) && + if_addr->prefixlen == ka->prefixlen && + !ldp_addrcmp(af, &if_addr->dstbrd, &ka->dstbrd)) + return (if_addr); + + return (NULL); +} + +void +if_addr_add(struct kaddr *ka) +{ + struct iface *iface; + struct if_addr *if_addr; + struct nbr *nbr; + + if (if_addr_lookup(&global.addr_list, ka) == NULL) { + if_addr = if_addr_new(ka); + + LIST_INSERT_HEAD(&global.addr_list, if_addr, entry); + RB_FOREACH(nbr, nbr_id_head, &nbrs_by_id) { + if (nbr->state != NBR_STA_OPER) + continue; + if (if_addr->af == AF_INET && !nbr->v4_enabled) + continue; + if (if_addr->af == AF_INET6 && !nbr->v6_enabled) + continue; + + send_address_single(nbr, if_addr, 0); + } + } + + iface = if_lookup_name(leconf, ka->ifname); + if (iface) { + if (ka->af == AF_INET6 && IN6_IS_ADDR_LINKLOCAL(&ka->addr.v6)) + iface->linklocal = ka->addr.v6; + + if (if_addr_lookup(&iface->addr_list, ka) == NULL) { + if_addr = if_addr_new(ka); + LIST_INSERT_HEAD(&iface->addr_list, if_addr, entry); + ldp_if_update(iface, if_addr->af); + } + } +} + +void +if_addr_del(struct kaddr *ka) +{ + struct iface *iface; + struct if_addr *if_addr; + struct nbr *nbr; + + iface = if_lookup_name(leconf, ka->ifname); + if (iface) { + if (ka->af == AF_INET6 && + IN6_ARE_ADDR_EQUAL(&iface->linklocal, &ka->addr.v6)) + memset(&iface->linklocal, 0, sizeof(iface->linklocal)); + + if_addr = if_addr_lookup(&iface->addr_list, ka); + if (if_addr) { + LIST_REMOVE(if_addr, entry); + ldp_if_update(iface, if_addr->af); + free(if_addr); + } + } + + if_addr = if_addr_lookup(&global.addr_list, ka); + if (if_addr) { + RB_FOREACH(nbr, nbr_id_head, &nbrs_by_id) { + if (nbr->state != NBR_STA_OPER) + continue; + if (if_addr->af == AF_INET && !nbr->v4_enabled) + continue; + if (if_addr->af == AF_INET6 && !nbr->v6_enabled) + continue; + send_address_single(nbr, if_addr, 1); + } + LIST_REMOVE(if_addr, entry); + free(if_addr); + } +} + +static int +if_start(struct iface *iface, int af) +{ + struct iface_af *ia; + struct timeval now; + + log_debug("%s: %s address-family %s", __func__, iface->name, + af_name(af)); + + ia = iface_af_get(iface, af); + + gettimeofday(&now, NULL); + ia->uptime = now.tv_sec; + + switch (af) { + case AF_INET: + if (if_join_ipv4_group(iface, &global.mcast_addr_v4)) + return (-1); + break; + case AF_INET6: + if (if_join_ipv6_group(iface, &global.mcast_addr_v6)) + return (-1); + break; + default: + fatalx("if_start: unknown af"); + } + + send_hello(HELLO_LINK, ia, NULL); + if_start_hello_timer(ia); + ia->state = IF_STA_ACTIVE; + + return (0); +} + +static int +if_reset(struct iface *iface, int af) +{ + struct iface_af *ia; + struct adj *adj; + + log_debug("%s: %s address-family %s", __func__, iface->name, + af_name(af)); + + ia = iface_af_get(iface, af); + if_stop_hello_timer(ia); + + while (!RB_EMPTY(ia_adj_head, &ia->adj_tree)) { + adj = RB_ROOT(ia_adj_head, &ia->adj_tree); + + adj_del(adj, S_SHUTDOWN); + } + + /* try to cleanup */ + switch (af) { + case AF_INET: + if (global.ipv4.ldp_disc_socket != -1) + if_leave_ipv4_group(iface, &global.mcast_addr_v4); + break; + case AF_INET6: + if (global.ipv6.ldp_disc_socket != -1) + if_leave_ipv6_group(iface, &global.mcast_addr_v6); + break; + default: + fatalx("if_reset: unknown af"); + } + + ia->state = IF_STA_DOWN; + + return (0); +} + +static void +if_update_af(struct iface_af *ia) +{ + int addr_ok = 0, socket_ok, rtr_id_ok; + struct if_addr *if_addr; + + switch (ia->af) { + case AF_INET: + /* + * NOTE: for LDPv4, each interface should have at least one + * valid IP address otherwise they can not be enabled. + */ + LIST_FOREACH(if_addr, &ia->iface->addr_list, entry) { + if (if_addr->af == AF_INET) { + addr_ok = 1; + break; + } + } + break; + case AF_INET6: + /* for IPv6 the link-local address is enough. */ + if (IN6_IS_ADDR_LINKLOCAL(&ia->iface->linklocal)) + addr_ok = 1; + break; + default: + fatalx("if_update_af: unknown af"); + } + + if ((ldp_af_global_get(&global, ia->af))->ldp_disc_socket != -1) + socket_ok = 1; + else + socket_ok = 0; + + if (ldp_rtr_id_get(leconf) != INADDR_ANY) + rtr_id_ok = 1; + else + rtr_id_ok = 0; + + if (ia->state == IF_STA_DOWN) { + if (!ia->enabled || !ia->iface->operative || !addr_ok || + !socket_ok || !rtr_id_ok) + return; + + if_start(ia->iface, ia->af); + } else if (ia->state == IF_STA_ACTIVE) { + if (ia->enabled && ia->iface->operative && addr_ok && + socket_ok && rtr_id_ok) + return; + + if_reset(ia->iface, ia->af); + } +} + +void +ldp_if_update(struct iface *iface, int af) +{ + if (af == AF_INET || af == AF_UNSPEC) + if_update_af(&iface->ipv4); + if (af == AF_INET6 || af == AF_UNSPEC) + if_update_af(&iface->ipv6); +} + +void +if_update_all(int af) +{ + struct iface *iface; + + RB_FOREACH(iface, iface_head, &leconf->iface_tree) + ldp_if_update(iface, af); +} + +uint16_t +if_get_hello_holdtime(struct iface_af *ia) +{ + if (ia->hello_holdtime != 0) + return (ia->hello_holdtime); + + if ((ldp_af_conf_get(leconf, ia->af))->lhello_holdtime != 0) + return ((ldp_af_conf_get(leconf, ia->af))->lhello_holdtime); + + return (leconf->lhello_holdtime); +} + +uint16_t +if_get_hello_interval(struct iface_af *ia) +{ + if (ia->hello_interval != 0) + return (ia->hello_interval); + + if ((ldp_af_conf_get(leconf, ia->af))->lhello_interval != 0) + return ((ldp_af_conf_get(leconf, ia->af))->lhello_interval); + + return (leconf->lhello_interval); +} + +uint16_t +if_get_wait_for_sync_interval(void) +{ + return (leconf->wait_for_sync_interval); +} + +/* timers */ +/* ARGSUSED */ +static void if_hello_timer(struct thread *thread) +{ + struct iface_af *ia = THREAD_ARG(thread); + + ia->hello_timer = NULL; + send_hello(HELLO_LINK, ia, NULL); + if_start_hello_timer(ia); +} + +static void +if_start_hello_timer(struct iface_af *ia) +{ + THREAD_OFF(ia->hello_timer); + thread_add_timer(master, if_hello_timer, ia, if_get_hello_interval(ia), + &ia->hello_timer); +} + +static void +if_stop_hello_timer(struct iface_af *ia) +{ + THREAD_OFF(ia->hello_timer); +} + +struct ctl_iface * +if_to_ctl(struct iface_af *ia) +{ + static struct ctl_iface ictl; + struct timeval now; + struct adj *adj; + + ictl.af = ia->af; + memcpy(ictl.name, ia->iface->name, sizeof(ictl.name)); + ictl.ifindex = ia->iface->ifindex; + ictl.state = ia->state; + ictl.type = ia->iface->type; + ictl.hello_holdtime = if_get_hello_holdtime(ia); + ictl.hello_interval = if_get_hello_interval(ia); + + gettimeofday(&now, NULL); + if (ia->state != IF_STA_DOWN && + ia->uptime != 0) { + ictl.uptime = now.tv_sec - ia->uptime; + } else + ictl.uptime = 0; + + ictl.adj_cnt = 0; + RB_FOREACH(adj, ia_adj_head, &ia->adj_tree) + ictl.adj_cnt++; + + return (&ictl); +} + +static void +ldp_sync_get_peer_ldp_id(struct iface *iface, struct in_addr *peer_ldp_id) +{ + struct iface_af *ia; + struct adj *adj; + + if (iface->ipv4.state == IF_STA_ACTIVE) { + ia = iface_af_get(iface, AF_INET); + RB_FOREACH(adj, ia_adj_head, &ia->adj_tree) + if (adj->nbr && adj->nbr->state == NBR_STA_OPER) { + *peer_ldp_id = adj->nbr->id; + return; + } + } + + if (iface->ipv6.state == IF_STA_ACTIVE) { + ia = iface_af_get(iface, AF_INET6); + RB_FOREACH(adj, ia_adj_head, &ia->adj_tree) + if (adj->nbr && adj->nbr->state == NBR_STA_OPER) { + *peer_ldp_id = adj->nbr->id; + return; + } + } +} + +struct ctl_ldp_sync * +ldp_sync_to_ctl(struct iface *iface) +{ + static struct ctl_ldp_sync ictl; + + memcpy(ictl.name, iface->name, sizeof(ictl.name)); + ictl.ifindex = iface->ifindex; + ictl.in_sync = (iface->ldp_sync.state == LDP_SYNC_STA_ACH); + ictl.wait_time = if_get_wait_for_sync_interval(); + ictl.timer_running = iface->ldp_sync.wait_for_sync_timer ? true : false; + + ictl.wait_time_remaining = + thread_timer_remain_second(iface->ldp_sync.wait_for_sync_timer); + + memset(&ictl.peer_ldp_id, 0, sizeof(ictl.peer_ldp_id)); + + ldp_sync_get_peer_ldp_id(iface, &ictl.peer_ldp_id); + + return (&ictl); +} + +/* multicast membership sockopts */ +in_addr_t +if_get_ipv4_addr(struct iface *iface) +{ + struct if_addr *if_addr; + + LIST_FOREACH(if_addr, &iface->addr_list, entry) + if (if_addr->af == AF_INET) + return (if_addr->addr.v4.s_addr); + + return (INADDR_ANY); +} + +static int +if_join_ipv4_group(struct iface *iface, struct in_addr *addr) +{ + struct in_addr if_addr; + + log_debug("%s: interface %s addr %pI4", __func__, iface->name, + addr); + + if_addr.s_addr = if_get_ipv4_addr(iface); + + if (setsockopt_ipv4_multicast(global.ipv4.ldp_disc_socket, + IP_ADD_MEMBERSHIP, if_addr, addr->s_addr, iface->ifindex) < 0) { + log_warn("%s: error IP_ADD_MEMBERSHIP, interface %s address %pI4", + __func__, iface->name, addr); + return (-1); + } + return (0); +} + +static int +if_leave_ipv4_group(struct iface *iface, struct in_addr *addr) +{ + struct in_addr if_addr; + + log_debug("%s: interface %s addr %pI4", __func__, iface->name, + addr); + + if_addr.s_addr = if_get_ipv4_addr(iface); + + if (setsockopt_ipv4_multicast(global.ipv4.ldp_disc_socket, + IP_DROP_MEMBERSHIP, if_addr, addr->s_addr, iface->ifindex) < 0) { + log_warn("%s: error IP_DROP_MEMBERSHIP, interface %s address %pI4", __func__, iface->name, addr); + return (-1); + } + + return (0); +} + +static int +if_join_ipv6_group(struct iface *iface, struct in6_addr *addr) +{ + struct ipv6_mreq mreq; + + log_debug("%s: interface %s addr %s", __func__, iface->name, + log_in6addr(addr)); + + mreq.ipv6mr_multiaddr = *addr; + mreq.ipv6mr_interface = iface->ifindex; + + if (setsockopt(global.ipv6.ldp_disc_socket, IPPROTO_IPV6, + IPV6_JOIN_GROUP, &mreq, sizeof(mreq)) < 0) { + log_warn("%s: error IPV6_JOIN_GROUP, interface %s address %s", + __func__, iface->name, log_in6addr(addr)); + return (-1); + } + + return (0); +} + +static int +if_leave_ipv6_group(struct iface *iface, struct in6_addr *addr) +{ + struct ipv6_mreq mreq; + + log_debug("%s: interface %s addr %s", __func__, iface->name, + log_in6addr(addr)); + + mreq.ipv6mr_multiaddr = *addr; + mreq.ipv6mr_interface = iface->ifindex; + + if (setsockopt(global.ipv6.ldp_disc_socket, IPPROTO_IPV6, + IPV6_LEAVE_GROUP, (void *)&mreq, sizeof(mreq)) < 0) { + log_warn("%s: error IPV6_LEAVE_GROUP, interface %s address %s", + __func__, iface->name, log_in6addr(addr)); + return (-1); + } + + return (0); +} + +const struct { + int state; + enum ldp_sync_event event; + enum ldp_sync_action action; + int new_state; +} ldp_sync_fsm_tbl[] = { + /* current state event that happened action to take resulting state */ +/* LDP IGP Sync not achieved */ + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_LDP_SYNC_START, LDP_SYNC_ACT_LDP_START_SYNC, 0}, + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_LDP_SYNC_COMPLETE, LDP_SYNC_ACT_LDP_COMPLETE_SYNC, LDP_SYNC_STA_ACH}, + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_CONFIG_LDP_OFF, LDP_SYNC_ACT_CONFIG_LDP_OFF, 0}, + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_IFACE_SHUTDOWN, LDP_SYNC_ACT_IFACE_SHUTDOWN, 0}, + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_SESSION_CLOSE, LDP_SYNC_ACT_NOTHING, 0}, + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_ADJ_DEL, LDP_SYNC_ACT_NOTHING, 0}, + {LDP_SYNC_STA_NOT_ACH, LDP_SYNC_EVT_ADJ_NEW, LDP_SYNC_ACT_NOTHING, 0}, +/* LDP IGP Sync achieved */ + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_CONFIG_LDP_OFF, LDP_SYNC_ACT_CONFIG_LDP_OFF, LDP_SYNC_STA_NOT_ACH}, + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_LDP_SYNC_COMPLETE, LDP_SYNC_ACT_NOTHING, 0}, + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_LDP_SYNC_START, LDP_SYNC_ACT_NOTHING, 0}, + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_IFACE_SHUTDOWN, LDP_SYNC_ACT_IFACE_SHUTDOWN, LDP_SYNC_STA_NOT_ACH}, + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_SESSION_CLOSE, LDP_SYNC_ACT_IFACE_START_SYNC, LDP_SYNC_STA_NOT_ACH}, + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_ADJ_DEL, LDP_SYNC_ACT_IFACE_START_SYNC, LDP_SYNC_STA_NOT_ACH}, + {LDP_SYNC_STA_ACH, LDP_SYNC_EVT_ADJ_NEW, LDP_SYNC_ACT_NOTHING, 0}, + {-1, LDP_SYNC_EVT_NOTHING, LDP_SYNC_ACT_NOTHING, 0}, +}; + +const char * const ldp_sync_event_names[] = { + "NOTHING", + "LDP SYNC START", + "LDP SYNC COMPLETE", + "CONFIG LDP OFF", + "IFACE SYNC START (ADJ DEL)", + "IFACE SYNC START (ADJ NEW)", + "IFACE SYNC START (SESSION CLOSE)", + "IFACE SYNC START (CONFIG LDP ON)", + "IFACE SHUTDOWN", + "N/A" +}; + +const char * const ldp_sync_action_names[] = { + "NOTHING", + "IFACE SYNC START", + "LDP START SYNC", + "LDP COMPLETE SYNC", + "CONFIG LDP OFF", + "IFACE SHUTDOWN", + "N/A" +}; + +const char * +ldp_sync_state_name(int state) +{ + switch (state) { + case LDP_SYNC_STA_NOT_ACH: + return ("NOT ACHIEVED"); + case LDP_SYNC_STA_ACH: + return ("ACHIEVED"); + default: + return ("UNKNOWN"); + } +} + +static int +send_ldp_sync_state_update(char *name, int ifindex, int sync_start) +{ + debug_evt_ldp_sync("%s: interface %s (%d), sync_start=%d", + __func__, name, ifindex, sync_start); + + struct ldp_igp_sync_if_state state; + + state.ifindex = ifindex; + state.sync_start = sync_start; + + return ldpe_imsg_compose_parent(IMSG_LDP_SYNC_IF_STATE_UPDATE, + getpid(), &state, sizeof(state)); +} + +static int +ldp_sync_act_iface_start_sync(struct iface *iface) +{ + send_ldp_sync_state_update(iface->name, iface->ifindex, true); + + return (0); +} + +static void iface_wait_for_ldp_sync_timer(struct thread *thread) +{ + struct iface *iface = THREAD_ARG(thread); + + ldp_sync_fsm(iface, LDP_SYNC_EVT_LDP_SYNC_COMPLETE); +} + +static void start_wait_for_ldp_sync_timer(struct iface *iface) +{ + if (iface->ldp_sync.wait_for_sync_timer) + return; + + THREAD_OFF(iface->ldp_sync.wait_for_sync_timer); + thread_add_timer(master, iface_wait_for_ldp_sync_timer, iface, + if_get_wait_for_sync_interval(), + &iface->ldp_sync.wait_for_sync_timer); +} + +static void stop_wait_for_ldp_sync_timer(struct iface *iface) +{ + THREAD_OFF(iface->ldp_sync.wait_for_sync_timer); +} + +static int +ldp_sync_act_ldp_start_sync(struct iface *iface) +{ + start_wait_for_ldp_sync_timer(iface); + + return 0; +} + +static int +ldp_sync_act_ldp_complete_sync(struct iface *iface) +{ + send_ldp_sync_state_update(iface->name, iface->ifindex, false); + + return 0; +} + +static int +iface_to_oper_nbr_count(struct iface *iface, unsigned int type) +{ + int oper_nbr_count = 0; + struct adj *adj; + + RB_FOREACH(adj, ia_adj_head, &iface->ipv4.adj_tree) { + if (type == adj->source.type && adj->nbr && + adj->nbr->state == NBR_STA_OPER) + oper_nbr_count++; + } + + RB_FOREACH(adj, ia_adj_head, &iface->ipv6.adj_tree) { + if (type == adj->source.type && adj->nbr && + adj->nbr->state == NBR_STA_OPER) + oper_nbr_count++; + } + + return oper_nbr_count; +} + +int +ldp_sync_fsm_adj_event(struct adj *adj, enum ldp_sync_event event) +{ + if (adj->source.type != HELLO_LINK) + return -1; + + struct iface *iface = adj->source.link.ia->iface; + + if (!iface->operative) + return 0; + + if (event == LDP_SYNC_EVT_ADJ_NEW) { + struct nbr *nbr = adj->nbr; + if (nbr && nbr->state == NBR_STA_OPER) { + event = LDP_SYNC_EVT_LDP_SYNC_START; + } + } else if (event == LDP_SYNC_EVT_ADJ_DEL) { + /* Ignore if an operational neighbor exists. + */ + int oper_nbr_count = iface_to_oper_nbr_count(iface, HELLO_LINK); + if (oper_nbr_count > 0) + return 0; + } + + debug_evt_ldp_sync("%s: event %s, " + "adj iface %s (%d) lsr-id %pI4 " + "source address %s transport address %s", + __func__, ldp_sync_event_names[event], + adj->source.link.ia->iface->name, + adj->source.link.ia->iface->ifindex, + &adj->lsr_id, + log_addr(adj_get_af(adj), &adj->source.link.src_addr), + log_addr(adj_get_af(adj), &adj->trans_addr)); + + return ldp_sync_fsm(iface, event); +} + +int +ldp_sync_fsm_nbr_event(struct nbr *nbr, enum ldp_sync_event event) +{ + struct adj *adj; + struct iface *iface = NULL; + RB_FOREACH(adj, nbr_adj_head, &nbr->adj_tree) { + if (HELLO_LINK != adj->source.type) + continue; + + iface = adj->source.link.ia->iface; + + if (!iface || !iface->operative) + continue; + + int oper_nbr_count = iface_to_oper_nbr_count(iface, HELLO_LINK); + + if (event == LDP_SYNC_EVT_SESSION_CLOSE && oper_nbr_count > 0) + /* Ignore if an operational neighbor exists. + */ + continue; + + debug_evt_ldp_sync("%s: event %s, iface %s, lsr-id %pI4", + __func__, ldp_sync_event_names[event], + iface->name, &nbr->id); + + ldp_sync_fsm(iface, event); + } + + return 0; +} + +int +ldp_sync_fsm_state_req(struct ldp_igp_sync_if_state_req *state_req) +{ + debug_evt_ldp_sync("%s: interface %s (%d) proto %s", + __func__, state_req->name, state_req->ifindex, + zebra_route_string(state_req->proto)); + + struct iface *iface = if_lookup_name(leconf, state_req->name); + + if (!iface) { + debug_evt_ldp_sync("%s: Warning: Ignoring LDP IGP SYNC " + "interface state request for interface %s (%d). " + "Interface does not exist in LDP.", + __func__, state_req->name, state_req->ifindex); + + return 0; + } + + return send_ldp_sync_state_update(state_req->name, + state_req->ifindex, + (iface->ldp_sync.state != LDP_SYNC_STA_ACH)); +} + +static int +ldp_sync_fsm_init(struct iface *iface, int state) +{ + int old_state = iface->ldp_sync.state; + + iface->ldp_sync.state = state; + stop_wait_for_ldp_sync_timer(iface); + + send_ldp_sync_state_update(iface->name, iface->ifindex, + (iface->ldp_sync.state != LDP_SYNC_STA_ACH)); + + if (old_state != iface->ldp_sync.state) { + debug_evt_ldp_sync("%s: resulted in " + "changing state for interface %s (%d) from %s to %s", + __func__, + iface->name, iface->ifindex, + ldp_sync_state_name(old_state), + ldp_sync_state_name(iface->ldp_sync.state)); + } + + return 0; +} + +int +ldp_sync_fsm(struct iface *iface, enum ldp_sync_event event) +{ + int old_state = iface->ldp_sync.state; + int new_state = 0; + int i; + + for (i = 0; ldp_sync_fsm_tbl[i].state != -1; i++) + if ((ldp_sync_fsm_tbl[i].state & old_state) && + (ldp_sync_fsm_tbl[i].event == event)) { + new_state = ldp_sync_fsm_tbl[i].new_state; + break; + } + + if (ldp_sync_fsm_tbl[i].state == -1) { + /* event outside of the defined fsm, ignore it. */ + log_warnx("%s: interface %s, event %s not expected in " + "state %s ", __func__, iface->name, + ldp_sync_event_names[event], + ldp_sync_state_name(old_state)); + return (0); + } + + if (new_state != 0) + iface->ldp_sync.state = new_state; + + switch (ldp_sync_fsm_tbl[i].action) { + case LDP_SYNC_ACT_IFACE_START_SYNC: + ldp_sync_act_iface_start_sync(iface); + break; + case LDP_SYNC_ACT_LDP_START_SYNC: + ldp_sync_act_ldp_start_sync(iface); + break; + case LDP_SYNC_ACT_LDP_COMPLETE_SYNC: + ldp_sync_act_ldp_complete_sync(iface); + break; + case LDP_SYNC_ACT_CONFIG_LDP_OFF: + ldp_sync_fsm_init(iface, LDP_SYNC_STA_NOT_ACH); + break; + case LDP_SYNC_ACT_IFACE_SHUTDOWN: + ldp_sync_fsm_init(iface, iface->ldp_sync.state); + break; + case LDP_SYNC_ACT_NOTHING: + /* do nothing */ + break; + } + + if (old_state != iface->ldp_sync.state) { + + debug_evt_ldp_sync("%s: event %s resulted in action %s " + "for interface %s, changing state from %s to %s", + __func__, ldp_sync_event_names[event], + ldp_sync_action_names[ldp_sync_fsm_tbl[i].action], + iface->name, ldp_sync_state_name(old_state), + ldp_sync_state_name(iface->ldp_sync.state)); + + } else { + debug_evt_ldp_sync("%s: event %s resulted in action %s " + "for interface %s, remaining in state %s", + __func__, ldp_sync_event_names[event], + ldp_sync_action_names[ldp_sync_fsm_tbl[i].action], + iface->name, + ldp_sync_state_name(iface->ldp_sync.state)); + } + + return (0); +} + +void +ldp_sync_fsm_reset_all(void) +{ + struct iface *iface; + + RB_FOREACH(iface, iface_head, &leconf->iface_tree) + ldp_sync_fsm(iface, LDP_SYNC_EVT_CONFIG_LDP_OFF); +} |