/* * BIRD -- Routing Tables * * (c) 1998--2000 Martin Mares * * Can be freely distributed and used under the terms of the GNU GPL. */ /** * DOC: Routing tables * * Routing tables are probably the most important structures BIRD uses. They * hold all the information about known networks, the associated routes and * their attributes. * * There are multiple routing tables (a primary one together with any * number of secondary ones if requested by the configuration). Each table * is basically a FIB containing entries describing the individual * destination networks. For each network (represented by structure &net), * there is a one-way linked list of route entries (&rte), the first entry * on the list being the best one (i.e., the one we currently use * for routing), the order of the other ones is undetermined. * * The &rte contains information specific to the route (preference, protocol * metrics, time of last modification etc.) and a pointer to a &rta structure * (see the route attribute module for a precise explanation) holding the * remaining route attributes which are expected to be shared by multiple * routes in order to conserve memory. */ #undef LOCAL_DEBUG #include "nest/bird.h" #include "nest/route.h" #include "nest/protocol.h" #include "nest/iface.h" #include "lib/resource.h" #include "lib/event.h" #include "lib/timer.h" #include "lib/string.h" #include "conf/conf.h" #include "filter/filter.h" #include "filter/data.h" #include "lib/hash.h" #include "lib/string.h" #include "lib/alloca.h" pool *rt_table_pool; static linpool *rte_update_pool; list routing_tables; static void rt_free_hostcache(rtable *tab); static void rt_notify_hostcache(rtable *tab, net *net); static void rt_update_hostcache(rtable *tab); static void rt_next_hop_update(rtable *tab); static inline void rt_prune_table(rtable *tab); static inline void rt_schedule_notify(rtable *tab); /* Like fib_route(), but skips empty net entries */ static inline void * net_route_ip4(rtable *t, net_addr_ip4 *n) { net *r; while (r = net_find_valid(t, (net_addr *) n), (!r) && (n->pxlen > 0)) { n->pxlen--; ip4_clrbit(&n->prefix, n->pxlen); } return r; } static inline void * net_route_ip6(rtable *t, net_addr_ip6 *n) { net *r; while (r = net_find_valid(t, (net_addr *) n), (!r) && (n->pxlen > 0)) { n->pxlen--; ip6_clrbit(&n->prefix, n->pxlen); } return r; } static inline void * net_route_ip6_sadr(rtable *t, net_addr_ip6_sadr *n) { struct fib_node *fn; while (1) { net *best = NULL; int best_pxlen = 0; /* We need to do dst first matching. Since sadr addresses are hashed on dst prefix only, find the hash table chain and go through it to find the match with the smallest matching src prefix. */ for (fn = fib_get_chain(&t->fib, (net_addr *) n); fn; fn = fn->next) { net_addr_ip6_sadr *a = (void *) fn->addr; if (net_equal_dst_ip6_sadr(n, a) && net_in_net_src_ip6_sadr(n, a) && (a->src_pxlen >= best_pxlen)) { best = fib_node_to_user(&t->fib, fn); best_pxlen = a->src_pxlen; } } if (best) return best; if (!n->dst_pxlen) break; n->dst_pxlen--; ip6_clrbit(&n->dst_prefix, n->dst_pxlen); } return NULL; } void * net_route(rtable *tab, const net_addr *n) { ASSERT(tab->addr_type == n->type); net_addr *n0 = alloca(n->length); net_copy(n0, n); switch (n->type) { case NET_IP4: case NET_VPN4: case NET_ROA4: return net_route_ip4(tab, (net_addr_ip4 *) n0); case NET_IP6: case NET_VPN6: case NET_ROA6: return net_route_ip6(tab, (net_addr_ip6 *) n0); case NET_IP6_SADR: return net_route_ip6_sadr(tab, (net_addr_ip6_sadr *) n0); default: return NULL; } } static int net_roa_check_ip4(rtable *tab, const net_addr_ip4 *px, u32 asn) { struct net_addr_roa4 n = NET_ADDR_ROA4(px->prefix, px->pxlen, 0, 0); struct fib_node *fn; int anything = 0; while (1) { for (fn = fib_get_chain(&tab->fib, (net_addr *) &n); fn; fn = fn->next) { net_addr_roa4 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); if (net_equal_prefix_roa4(roa, &n) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) return ROA_VALID; } } if (n.pxlen == 0) break; n.pxlen--; ip4_clrbit(&n.prefix, n.pxlen); } return anything ? ROA_INVALID : ROA_UNKNOWN; } static int net_roa_check_ip6(rtable *tab, const net_addr_ip6 *px, u32 asn) { struct net_addr_roa6 n = NET_ADDR_ROA6(px->prefix, px->pxlen, 0, 0); struct fib_node *fn; int anything = 0; while (1) { for (fn = fib_get_chain(&tab->fib, (net_addr *) &n); fn; fn = fn->next) { net_addr_roa6 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); if (net_equal_prefix_roa6(roa, &n) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) return ROA_VALID; } } if (n.pxlen == 0) break; n.pxlen--; ip6_clrbit(&n.prefix, n.pxlen); } return anything ? ROA_INVALID : ROA_UNKNOWN; } /** * roa_check - check validity of route origination in a ROA table * @tab: ROA table * @n: network prefix to check * @asn: AS number of network prefix * * Implements RFC 6483 route validation for the given network prefix. The * procedure is to find all candidate ROAs - ROAs whose prefixes cover the given * network prefix. If there is no candidate ROA, return ROA_UNKNOWN. If there is * a candidate ROA with matching ASN and maxlen field greater than or equal to * the given prefix length, return ROA_VALID. Otherwise, return ROA_INVALID. If * caller cannot determine origin AS, 0 could be used (in that case ROA_VALID * cannot happen). Table @tab must have type NET_ROA4 or NET_ROA6, network @n * must have type NET_IP4 or NET_IP6, respectively. */ int net_roa_check(rtable *tab, const net_addr *n, u32 asn) { if ((tab->addr_type == NET_ROA4) && (n->type == NET_IP4)) return net_roa_check_ip4(tab, (const net_addr_ip4 *) n, asn); else if ((tab->addr_type == NET_ROA6) && (n->type == NET_IP6)) return net_roa_check_ip6(tab, (const net_addr_ip6 *) n, asn); else return ROA_UNKNOWN; /* Should not happen */ } /** * rte_find - find a route * @net: network node * @src: route source * * The rte_find() function returns a pointer to a route for destination @net * which is from route source @src. List end pointer is returned if no route is found. */ static struct rte_storage ** rte_find(net *net, struct rte_src *src) { struct rte_storage **e = &net->routes; while ((*e) && (*e)->rte.src != src) e = &(*e)->next; return e; } struct rte_storage * rte_store(const rte *r, net *net, rtable *tab) { struct rte_storage *e = sl_alloc(tab->rte_slab); e->rte = *r; e->rte.net = net->n.addr; rt_lock_source(e->rte.src); if (e->rte.attrs->cached) e->rte.attrs = rta_clone(e->rte.attrs); else e->rte.attrs = rta_lookup(e->rte.attrs); return e; } /** * rte_free - delete a &rte * @e: &struct rte_storage to be deleted * @tab: the table which the rte belongs to * * rte_free() deletes the given &rte from the routing table it's linked to. */ void rte_free(struct rte_storage *e, rtable *tab) { rt_unlock_source(e->rte.src); rta_free(e->rte.attrs); sl_free(tab->rte_slab, e); } static int /* Actually better or at least as good as */ rte_better(rte *new, rte *old) { int (*better)(rte *, rte *); if (!rte_is_valid(old)) return 1; if (!rte_is_valid(new)) return 0; if (new->attrs->pref > old->attrs->pref) return 1; if (new->attrs->pref < old->attrs->pref) return 0; if (new->src->proto->proto != old->src->proto->proto) { /* * If the user has configured protocol preferences, so that two different protocols * have the same preference, try to break the tie by comparing addresses. Not too * useful, but keeps the ordering of routes unambiguous. */ return new->src->proto->proto > old->src->proto->proto; } if (better = new->src->proto->rte_better) return better(new, old); return 0; } static int rte_mergable(rte *pri, rte *sec) { int (*mergable)(rte *, rte *); if (!rte_is_valid(pri) || !rte_is_valid(sec)) return 0; if (pri->attrs->pref != sec->attrs->pref) return 0; if (pri->src->proto->proto != sec->src->proto->proto) return 0; if (mergable = pri->src->proto->rte_mergable) return mergable(pri, sec); return 0; } static void rte_trace(struct channel *c, rte *e, int dir, const char *msg) { log(L_TRACE "%s.%s %c %s %N %uL %uG %s", c->proto->name, c->name ?: "?", dir, msg, e->net, e->src->private_id, e->src->global_id, rta_dest_name(e->attrs->dest)); } static inline void rte_trace_in(uint flag, struct channel *c, rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) rte_trace(c, e, '>', msg); } static inline void rte_trace_out(uint flag, struct channel *c, rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) rte_trace(c, e, '<', msg); } static rte * export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) { struct proto *p = c->proto; const struct filter *filter = c->out_filter; struct export_stats *stats = &c->export_stats; int v; v = p->preexport ? p->preexport(c, rt) : 0; if (v < 0) { if (silent) goto reject; stats->updates_rejected++; if (v == RIC_REJECT) rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); goto reject; } if (v > 0) { if (!silent) rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); goto accept; } v = filter && ((filter == FILTER_REJECT) || (f_run(filter, rt, pool, (silent ? FF_SILENT : 0)) > F_ACCEPT)); if (v) { if (silent) goto reject; stats->updates_filtered++; rte_trace_out(D_FILTERS, c, rt, "filtered out"); goto reject; } accept: return rt; reject: /* Discard temporary rte */ return NULL; } static inline rte * export_filter(struct channel *c, rte *rt, int silent) { return export_filter_(c, rt, rte_update_pool, silent); } static void do_rt_notify(struct channel *c, const net_addr *net, rte *new, rte *old, int refeed) { struct proto *p = c->proto; struct export_stats *stats = &c->export_stats; if (refeed && new) c->refeed_count++; /* Apply export limit */ struct channel_limit *l = &c->out_limit; if (l->action && !old && new) { if (stats->routes >= l->limit) channel_notify_limit(c, l, PLD_OUT, stats->routes); if (l->state == PLS_BLOCKED) { stats->updates_rejected++; rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); return; } } /* Apply export table */ struct rte_storage *old_exported = NULL; if (c->out_table) { if (!rte_update_out(c, net, new, old, &old_exported)) { rte_trace_out(D_ROUTES, c, new, "idempotent"); return; } } if (new) stats->updates_accepted++; else stats->withdraws_accepted++; if (old) { bmap_clear(&c->export_map, old->id); stats->routes--; } if (new) { bmap_set(&c->export_map, new->id); stats->routes++; } if (p->debug & D_ROUTES) { if (new && old) rte_trace_out(D_ROUTES, c, new, "replaced"); else if (new) rte_trace_out(D_ROUTES, c, new, "added"); else if (old) rte_trace_out(D_ROUTES, c, old, "removed"); } p->rt_notify(p, c, net, new, old_exported ? &old_exported->rte : old); if (c->out_table && old_exported) rte_free(old_exported, c->out_table); } static void rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old, int refeed) { if (new) c->export_stats.updates_received++; else c->export_stats.withdraws_received++; if (new) new = export_filter(c, new, 0); if (old && !bmap_test(&c->export_map, old->id)) old = NULL; if (!new && !old) return; do_rt_notify(c, net, new, old, refeed); } static void rt_notify_accepted(struct channel *c, net *net, rte *new_changed, rte *old_changed, int refeed) { // struct proto *p = c->proto; rte nb0; rte *new_best = NULL; rte *old_best = NULL; int new_first = 0; /* * We assume that there are no changes in net route order except (added) * new_changed and (removed) old_changed. Therefore, the function is not * compatible with deterministic_med (where nontrivial reordering can happen * as a result of a route change) and with recomputation of recursive routes * due to next hop update (where many routes can be changed in one step). * * Note that we need this assumption just for optimizations, we could just * run full new_best recomputation otherwise. * * There are three cases: * feed or old_best is old_changed -> we need to recompute new_best * old_best is before new_changed -> new_best is old_best, ignore * old_best is after new_changed -> try new_changed, otherwise old_best */ if (net->routes) c->export_stats.updates_received++; else c->export_stats.withdraws_received++; /* Find old_best - either old_changed, or route for net->routes */ if (old_changed && bmap_test(&c->export_map, old_changed->id)) old_best = old_changed; else { for (struct rte_storage *r = net->routes; r && rte_is_valid(&r->rte); r = r->next) { if (bmap_test(&c->export_map, r->rte.id)) { old_best = &r->rte; break; } /* Note if new_changed found before old_best */ if (&r->rte == new_changed) new_first = 1; } } /* Find new_best */ if ((new_changed == old_changed) || (old_best == old_changed)) { /* Feed or old_best changed -> find first accepted by filters */ for (struct rte_storage *r = net->routes; r && rte_is_valid(&r->rte); r = r->next) if (new_best = export_filter(c, ((nb0 = r->rte), &nb0), 0)) break; } else { /* Other cases -> either new_changed, or old_best (and nothing changed) */ if (new_first && (new_changed = export_filter(c, new_changed, 0))) new_best = new_changed; else return; } if (!new_best && !old_best) return; do_rt_notify(c, net->n.addr, new_best, old_best, refeed); } static struct nexthop * nexthop_merge_rta(struct nexthop *nhs, rta *a, linpool *pool, int max) { return nexthop_merge(nhs, &(a->nh), 1, 0, max, pool); } rte * rt_export_merged(struct channel *c, net *net, linpool *pool, int silent) { // struct proto *p = c->proto; struct nexthop *nhs = NULL; _Thread_local static rte rme; struct rte_storage *best0 = net->routes; rte *best; if (!best0 || !rte_is_valid(&best0->rte)) return NULL; best = export_filter_(c, ((rme = best0->rte), &rme), pool, silent); if (!best || !rte_is_reachable(best)) return best; for (struct rte_storage *rt0 = best0->next; rt0; rt0 = rt0->next) { if (!rte_mergable(best, &rt0->rte)) continue; rte rnh = rt0->rte; rte *rt = export_filter_(c, &rnh, pool, 1); if (!rt) continue; if (rte_is_reachable(rt)) nhs = nexthop_merge_rta(nhs, rt->attrs, pool, c->merge_limit); } if (nhs) { nhs = nexthop_merge_rta(nhs, best->attrs, pool, c->merge_limit); if (nhs->next) { best->attrs = rta_cow(best->attrs, pool); nexthop_link(best->attrs, nhs); } } return best; } static void rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed, rte *new_best, rte *old_best, int refeed) { /* We assume that all rte arguments are either NULL or rte_is_valid() */ /* This check should be done by the caller */ if (!new_best && !old_best) return; /* Check whether the change is relevant to the merged route */ if ((new_best == old_best) && (new_changed != old_changed) && !rte_mergable(new_best, new_changed) && !rte_mergable(old_best, old_changed)) return; if (new_best) c->export_stats.updates_received++; else c->export_stats.withdraws_received++; /* Prepare new merged route */ if (new_best) new_best = rt_export_merged(c, net, rte_update_pool, 0); /* Check old merged route */ if (old_best && !bmap_test(&c->export_map, old_best->id)) old_best = NULL; if (!new_best && !old_best) return; do_rt_notify(c, net->n.addr, new_best, old_best, refeed); } /** * rte_announce - announce a routing table change * @tab: table the route has been added to * @type: type of route announcement (RA_UNDEF or RA_ANY) * @net: network in question * @new: the new or changed route * @old: the previous route replaced by the new one * @new_best: the new best route for the same network * @old_best: the previous best route for the same network * * This function gets a routing table update and announces it to all protocols * that are connected to the same table by their channels. * * There are two ways of how routing table changes are announced. First, there * is a change of just one route in @net (which may caused a change of the best * route of the network). In this case @new and @old describes the changed route * and @new_best and @old_best describes best routes. Other routes are not * affected, but in sorted table the order of other routes might change. * * Second, There is a bulk change of multiple routes in @net, with shared best * route selection. In such case separate route changes are described using * @type of %RA_ANY, with @new and @old specifying the changed route, while * @new_best and @old_best are NULL. After that, another notification is done * where @new_best and @old_best are filled (may be the same), but @new and @old * are NULL. * * The function announces the change to all associated channels. For each * channel, an appropriate preprocessing is done according to channel &ra_mode. * For example, %RA_OPTIMAL channels receive just changes of best routes. * * In general, we first call preexport() hook of a protocol, which performs * basic checks on the route (each protocol has a right to veto or force accept * of the route before any filter is asked). Then we consult an export filter * of the channel and verify the old route in an export map of the channel. * Finally, the rt_notify() hook of the protocol gets called. * * Note that there are also calls of rt_notify() hooks due to feed, but that is * done outside of scope of rte_announce(). */ static void rte_announce(rtable *tab, uint type, net *net, struct rte_storage *new, struct rte_storage *old, struct rte_storage *new_best, struct rte_storage *old_best) { if (!new || !rte_is_valid(&new->rte)) new = NULL; if (!old || !rte_is_valid(&old->rte)) old = NULL; if (!new_best || !rte_is_valid(&new_best->rte)) new_best = NULL; if (!old_best || !rte_is_valid(&old_best->rte)) old_best = NULL; if (!new && !old && !new_best && !old_best) return; if (new_best != old_best) { if (new_best) new_best->rte.sender->import_stats.pref++; if (old_best) old_best->rte.sender->import_stats.pref--; if (tab->hostcache) rt_notify_hostcache(tab, net); } rt_schedule_notify(tab); struct channel *c; node *n; WALK_LIST2(c, n, tab->channels, table_node) { if (c->export_state == ES_DOWN) continue; if (type && (type != c->ra_mode)) continue; rte n0; switch (c->ra_mode) { case RA_OPTIMAL: if (new_best != old_best) rt_notify_basic(c, net->n.addr, RTE_COPY(new_best, &n0), RTE_OR_NULL(old_best), 0); break; case RA_ANY: if (new != old) rt_notify_basic(c, net->n.addr, RTE_COPY(new, &n0), RTE_OR_NULL(old), 0); break; case RA_ACCEPTED: rt_notify_accepted(c, net, RTE_OR_NULL(new), RTE_OR_NULL(old), 0); break; case RA_MERGED: rt_notify_merged(c, net, RTE_OR_NULL(new), RTE_OR_NULL(old), RTE_OR_NULL(new_best), RTE_OR_NULL(old_best), 0); break; } } } static inline int rte_validate(rte *e) { int c; const net_addr *n = e->net; if (!net_validate(n)) { log(L_WARN "Ignoring bogus prefix %N received via %s", n, e->sender->proto->name); return 0; } /* FIXME: better handling different nettypes */ c = !net_is_flow(n) ? net_classify(n): (IADDR_HOST | SCOPE_UNIVERSE); if ((c < 0) || !(c & IADDR_HOST) || ((c & IADDR_SCOPE_MASK) <= SCOPE_LINK)) { log(L_WARN "Ignoring bogus route %N received via %s", n, e->sender->proto->name); return 0; } if (net_type_match(n, NB_DEST) == !e->attrs->dest) { log(L_WARN "Ignoring route %N with invalid dest %d received via %s", n, e->attrs->dest, e->sender->proto->name); return 0; } if ((e->attrs->dest == RTD_UNICAST) && !nexthop_is_sorted(&(e->attrs->nh))) { log(L_WARN "Ignoring unsorted multipath route %N received via %s", n, e->sender->proto->name); return 0; } return 1; } static int rte_same(rte *x, rte *y) { /* rte.flags are not checked, as they are mostly internal to rtable */ return x->attrs == y->attrs && x->pflags == y->pflags && x->src == y->src && rte_is_filtered(x) == rte_is_filtered(y); } static inline int rte_is_ok(rte *e) { return e && !rte_is_filtered(e); } static void rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) { struct proto *p = c->proto; struct rtable *table = c->table; struct import_stats *stats = &c->import_stats; struct rte_storage *old_best_stored = net->routes, *old_stored = NULL; rte *old_best = old_best_stored ? &old_best_stored->rte : NULL; rte *old = NULL; /* Find and remove original route from the same protocol */ struct rte_storage **before_old = rte_find(net, src); if (*before_old) { old = &(old_stored = (*before_old))->rte; /* If there is the same route in the routing table but from * a different sender, then there are two paths from the * source protocol to this routing table through transparent * pipes, which is not allowed. * We log that and ignore the route. */ if (old->sender->proto != p) { if (!old->generation && !new->generation) bug("Two protocols claim to author a route with the same rte_src in table %s: %N %s/%u:%u", c->table->name, net->n.addr, old->src->proto->name, old->src->private_id, old->src->global_id); log_rl(&table->rl_pipe, L_ERR "Route source collision in table %s: %N %s/%u:%u", c->table->name, net->n.addr, old->src->proto->name, old->src->private_id, old->src->global_id); return; } if (new && rte_same(old, new)) { /* No changes, ignore the new route and refresh the old one */ old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); if (!rte_is_filtered(new)) { stats->updates_ignored++; rte_trace_in(D_ROUTES, c, new, "ignored"); } return; } *before_old = (*before_old)->next; table->rt_count--; } if (!old && !new) { stats->withdraws_ignored++; return; } int new_ok = rte_is_ok(new); int old_ok = rte_is_ok(old); struct channel_limit *l = &c->rx_limit; if (l->action && !old && new && !c->in_table) { u32 all_routes = stats->routes + stats->filtered; if (all_routes >= l->limit) channel_notify_limit(c, l, PLD_RX, all_routes); if (l->state == PLS_BLOCKED) { /* In receive limit the situation is simple, old is NULL so we just free new and exit like nothing happened */ stats->updates_ignored++; rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); return; } } l = &c->in_limit; if (l->action && !old_ok && new_ok) { if (stats->routes >= l->limit) channel_notify_limit(c, l, PLD_IN, stats->routes); if (l->state == PLS_BLOCKED) { /* In import limit the situation is more complicated. We shouldn't just drop the route, we should handle it like it was filtered. We also have to continue the route processing if old or new is non-NULL, but we should exit if both are NULL as this case is probably assumed to be already handled. */ stats->updates_ignored++; rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); if (c->in_keep_filtered) new->flags |= REF_FILTERED; else new = NULL; /* Note that old && !new could be possible when c->in_keep_filtered changed in the recent past. */ if (!old && !new) return; new_ok = 0; goto skip_stats1; } } if (new_ok) stats->updates_accepted++; else if (old_ok) stats->withdraws_accepted++; else stats->withdraws_ignored++; if (old_ok || new_ok) table->last_rt_change = current_time(); skip_stats1:; struct rte_storage *new_stored = new ? rte_store(new, net, table) : NULL; if (new) rte_is_filtered(new) ? stats->filtered++ : stats->routes++; if (old) rte_is_filtered(old) ? stats->filtered-- : stats->routes--; if (table->config->sorted) { /* If routes are sorted, just insert new route to appropriate position */ if (new_stored) { struct rte_storage **k; if ((before_old != &net->routes) && !rte_better(new, &SKIP_BACK(struct rte_storage, next, before_old)->rte)) k = before_old; else k = &net->routes; for (; *k; k=&(*k)->next) if (rte_better(new, &(*k)->rte)) break; new_stored->next = *k; *k = new_stored; table->rt_count++; } } else { /* If routes are not sorted, find the best route and move it on the first position. There are several optimized cases. */ if (src->proto->rte_recalculate && src->proto->rte_recalculate(table, net, new_stored ? &new_stored->rte : NULL, old, old_best)) goto do_recalculate; if (new_stored && rte_better(&new_stored->rte, old_best)) { /* The first case - the new route is cleary optimal, we link it at the first position */ new_stored->next = net->routes; net->routes = new_stored; table->rt_count++; } else if (old == old_best) { /* The second case - the old best route disappeared, we add the new route (if we have any) to the list (we don't care about position) and then we elect the new optimal route and relink that route at the first position and announce it. New optimal route might be NULL if there is no more routes */ do_recalculate: /* Add the new route to the list */ if (new_stored) { new_stored->next = *before_old; *before_old = new_stored; table->rt_count++; } /* Find a new optimal route (if there is any) */ if (net->routes) { struct rte_storage **bp = &net->routes; for (struct rte_storage **k=&(*bp)->next; *k; k=&(*k)->next) if (rte_better(&(*k)->rte, &(*bp)->rte)) bp = k; /* And relink it */ struct rte_storage *best = *bp; *bp = best->next; best->next = net->routes; net->routes = best; } } else if (new_stored) { /* The third case - the new route is not better than the old best route (therefore old_best != NULL) and the old best route was not removed (therefore old_best == net->routes). We just link the new route to the old/last position. */ new_stored->next = *before_old; *before_old = new_stored; table->rt_count++; } /* The fourth (empty) case - suboptimal route was removed, nothing to do */ } if (new_stored) { new_stored->rte.lastmod = current_time(); if (!old) { new_stored->rte.id = hmap_first_zero(&table->id_map); hmap_set(&table->id_map, new_stored->rte.id); } else new_stored->rte.id = old->id; } /* Log the route change */ if ((c->debug & D_ROUTES) || (p->debug & D_ROUTES)) { if (new_ok) rte_trace(c, &new_stored->rte, '>', new_stored == net->routes ? "added [best]" : "added"); else if (old_ok) { if (old != old_best) rte_trace(c, old, '>', "removed"); else if (net->routes && rte_is_ok(&net->routes->rte)) rte_trace(c, old, '>', "removed [replaced]"); else rte_trace(c, old, '>', "removed [sole]"); } } /* Propagate the route change */ rte_announce(table, RA_UNDEF, net, new_stored, old_stored, net->routes, old_best_stored); if (!net->routes && (table->gc_counter++ >= table->config->gc_max_ops) && (table->gc_time + table->config->gc_min_time <= current_time())) rt_schedule_prune(table); if (old_ok && p->rte_remove) p->rte_remove(net, old); if (new_ok && p->rte_insert) p->rte_insert(net, &new_stored->rte); if (old) { if (!new_stored) hmap_clear(&table->id_map, old->id); rte_free(old_stored, table); } } static int rte_update_nest_cnt; /* Nesting counter to allow recursive updates */ static inline void rte_update_lock(void) { rte_update_nest_cnt++; } static inline void rte_update_unlock(void) { if (!--rte_update_nest_cnt) lp_flush(rte_update_pool); } static int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); void rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { if (c->in_table && !rte_update_in(c, n, new, src)) return; struct import_stats *stats = &c->import_stats; const struct filter *filter = c->in_filter; net *nn; ASSERT(c->channel_state == CS_UP); rte_update_lock(); if (new) { new->net = n; new->sender = c; stats->updates_received++; if (!rte_validate(new)) { rte_trace_in(D_FILTERS, c, new, "invalid"); stats->updates_invalid++; goto drop; } if (filter == FILTER_REJECT) { stats->updates_filtered++; rte_trace_in(D_FILTERS, c, new, "filtered out"); if (! c->in_keep_filtered) goto drop; /* new is a private copy, i could modify it */ new->flags |= REF_FILTERED; } else if (filter) { int fr = f_run(filter, new, rte_update_pool, 0); if (fr > F_ACCEPT) { stats->updates_filtered++; rte_trace_in(D_FILTERS, c, new, "filtered out"); if (! c->in_keep_filtered) goto drop; new->flags |= REF_FILTERED; } } /* Use the actual struct network, not the dummy one */ nn = net_get(c->table, n); new->net = nn->n.addr; } else { stats->withdraws_received++; if (!(nn = net_find(c->table, n)) || !src) { stats->withdraws_ignored++; rte_update_unlock(); return; } } recalc: /* And recalculate the best route */ rte_recalculate(c, nn, new, src); rte_update_unlock(); return; drop: new = NULL; if (nn = net_find(c->table, n)) goto recalc; rte_update_unlock(); } /* Independent call to rte_announce(), used from next hop recalculation, outside of rte_update(). new must be non-NULL */ static inline void rte_announce_i(rtable *tab, uint type, net *net, struct rte_storage *new, struct rte_storage *old, struct rte_storage *new_best, struct rte_storage *old_best) { rte_update_lock(); rte_announce(tab, type, net, new, old, new_best, old_best); rte_update_unlock(); } static inline void rte_discard(net *net, rte *old) /* Non-filtered route deletion, used during garbage collection */ { rte_update_lock(); rte_recalculate(old->sender, net, NULL, old->src); rte_update_unlock(); } /* Modify existing route by protocol hook, used for long-lived graceful restart */ static inline void rte_modify(net *net, rte *old) { rte_update_lock(); rte *new = old->sender->proto->rte_modify(old, rte_update_pool); if (new != old) { if (new) new->flags = old->flags & ~REF_MODIFY; rte_recalculate(old->sender, net, new, old->src); } rte_update_unlock(); } /* Check rtable for best route to given net whether it would be exported do p */ int rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filter) { net *n = net_find(t, a); if (!n || !n->routes) return 0; rte rt = n->routes->rte; if (!rte_is_valid(&rt)) return 0; rte_update_lock(); /* Rest is stripped down export_filter() */ int v = c->proto->preexport ? c->proto->preexport(c, &rt) : 0; if (v == RIC_PROCESS) v = (f_run(filter, &rt, rte_update_pool, FF_SILENT) <= F_ACCEPT); rte_update_unlock(); return v > 0; } /** * rt_refresh_begin - start a refresh cycle * @t: related routing table * @c related channel * * This function starts a refresh cycle for given routing table and announce * hook. The refresh cycle is a sequence where the protocol sends all its valid * routes to the routing table (by rte_update()). After that, all protocol * routes (more precisely routes with @c as @sender) not sent during the * refresh cycle but still in the table from the past are pruned. This is * implemented by marking all related routes as stale by REF_STALE flag in * rt_refresh_begin(), then marking all related stale routes with REF_DISCARD * flag in rt_refresh_end() and then removing such routes in the prune loop. */ void rt_refresh_begin(rtable *t, struct channel *c) { FIB_WALK(&t->fib, net, n) { for (struct rte_storage *e = n->routes; e; e = e->next) if (e->rte.sender == c) e->rte.flags |= REF_STALE; } FIB_WALK_END; } /** * rt_refresh_end - end a refresh cycle * @t: related routing table * @c: related channel * * This function ends a refresh cycle for given routing table and announce * hook. See rt_refresh_begin() for description of refresh cycles. */ void rt_refresh_end(rtable *t, struct channel *c) { int prune = 0; FIB_WALK(&t->fib, net, n) { for (struct rte_storage *e = n->routes; e; e = e->next) if ((e->rte.sender == c) && (e->rte.flags & REF_STALE)) { e->rte.flags |= REF_DISCARD; prune = 1; } } FIB_WALK_END; if (prune) rt_schedule_prune(t); } void rt_modify_stale(rtable *t, struct channel *c) { int prune = 0; FIB_WALK(&t->fib, net, n) { for (struct rte_storage *e = n->routes; e; e = e->next) if ((e->rte.sender == c) && (e->rte.flags & REF_STALE) && !(e->rte.flags & REF_FILTERED)) { e->rte.flags |= REF_MODIFY; prune = 1; } } FIB_WALK_END; if (prune) rt_schedule_prune(t); } /** * rte_dump - dump a route * @e: &rte to be dumped * * This functions dumps contents of a &rte to debug output. */ void rte_dump(struct rte_storage *e) { debug("%-1N ", e->rte.net); debug("PF=%02x ", e->rte.pflags); rta_dump(e->rte.attrs); debug("\n"); } /** * rt_dump - dump a routing table * @t: routing table to be dumped * * This function dumps contents of a given routing table to debug output. */ void rt_dump(rtable *t) { debug("Dump of routing table <%s>\n", t->name); #ifdef DEBUGGING fib_check(&t->fib); #endif FIB_WALK(&t->fib, net, n) { for(struct rte_storage *e=n->routes; e; e=e->next) rte_dump(e); } FIB_WALK_END; debug("\n"); } /** * rt_dump_all - dump all routing tables * * This function dumps contents of all routing tables to debug output. */ void rt_dump_all(void) { rtable *t; node *n; WALK_LIST2(t, n, routing_tables, n) rt_dump(t); } static inline void rt_schedule_hcu(rtable *tab) { if (tab->hcu_scheduled) return; tab->hcu_scheduled = 1; ev_schedule(tab->rt_event); } static inline void rt_schedule_nhu(rtable *tab) { if (tab->nhu_state == NHU_CLEAN) ev_schedule(tab->rt_event); /* state change: * NHU_CLEAN -> NHU_SCHEDULED * NHU_RUNNING -> NHU_DIRTY */ tab->nhu_state |= NHU_SCHEDULED; } void rt_schedule_prune(rtable *tab) { if (tab->prune_state == 0) ev_schedule(tab->rt_event); /* state change 0->1, 2->3 */ tab->prune_state |= 1; } static void rt_event(void *ptr) { rtable *tab = ptr; rt_lock_table(tab); if (tab->hcu_scheduled) rt_update_hostcache(tab); if (tab->nhu_state) rt_next_hop_update(tab); if (tab->prune_state) rt_prune_table(tab); rt_unlock_table(tab); } static inline btime rt_settled_time(rtable *tab) { ASSUME(tab->base_settle_time != 0); return MIN(tab->last_rt_change + tab->config->min_settle_time, tab->base_settle_time + tab->config->max_settle_time); } static void rt_settle_timer(timer *t) { rtable *tab = t->data; if (!tab->base_settle_time) return; btime settled_time = rt_settled_time(tab); if (current_time() < settled_time) { tm_set(tab->settle_timer, settled_time); return; } /* Settled */ tab->base_settle_time = 0; struct rt_subscription *s; WALK_LIST(s, tab->subscribers) s->hook(s); } static void rt_kick_settle_timer(rtable *tab) { tab->base_settle_time = current_time(); if (!tab->settle_timer) tab->settle_timer = tm_new_init(tab->rp, rt_settle_timer, tab, 0, 0); if (!tm_active(tab->settle_timer)) tm_set(tab->settle_timer, rt_settled_time(tab)); } static inline void rt_schedule_notify(rtable *tab) { if (EMPTY_LIST(tab->subscribers)) return; if (tab->base_settle_time) return; rt_kick_settle_timer(tab); } void rt_subscribe(rtable *tab, struct rt_subscription *s) { s->tab = tab; rt_lock_table(tab); add_tail(&tab->subscribers, &s->n); } void rt_unsubscribe(struct rt_subscription *s) { rem_node(&s->n); rt_unlock_table(s->tab); } static void rt_free(resource *_r) { rtable *r = (rtable *) _r; DBG("Deleting routing table %s\n", r->name); ASSERT_DIE(r->use_count == 0); if (r->internal) return; r->config->table = NULL; rem_node(&r->n); if (r->hostcache) rt_free_hostcache(r); /* Freed automagically by the resource pool fib_free(&r->fib); hmap_free(&r->id_map); rfree(r->rt_event); rfree(r->settle_timer); mb_free(r); */ } static void rt_res_dump(resource *_r) { rtable *r = (rtable *) _r; debug("name \"%s\", addr_type=%s, rt_count=%u, use_count=%d\n", r->name, net_label[r->addr_type], r->rt_count, r->use_count); } static struct resclass rt_class = { .name = "Routing table", .size = sizeof(struct rtable), .free = rt_free, .dump = rt_res_dump, .lookup = NULL, .memsize = NULL, }; rtable * rt_setup(pool *pp, struct rtable_config *cf) { int ns = strlen("Routing table ") + strlen(cf->name) + 1; void *nb = mb_alloc(pp, ns); ASSERT_DIE(ns - 1 == bsnprintf(nb, ns, "Routing table %s", cf->name)); pool *p = rp_new(pp, nb); mb_move(nb, p); rtable *t = ralloc(p, &rt_class); t->rp = p; t->rte_slab = sl_new(p, sizeof(struct rte_storage)); t->name = cf->name; t->config = cf; t->addr_type = cf->addr_type; fib_init(&t->fib, p, t->addr_type, sizeof(net), OFFSETOF(net, n), 0, NULL); if (!(t->internal = cf->internal)) { init_list(&t->channels); hmap_init(&t->id_map, p, 1024); hmap_set(&t->id_map, 0); init_list(&t->subscribers); t->rt_event = ev_new_init(p, rt_event, t); t->last_rt_change = t->gc_time = current_time(); t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS; } return t; } /** * rt_init - initialize routing tables * * This function is called during BIRD startup. It initializes the * routing table module. */ void rt_init(void) { rta_init(); rt_table_pool = rp_new(&root_pool, "Routing tables"); rte_update_pool = lp_new_default(rt_table_pool); init_list(&routing_tables); } /** * rt_prune_table - prune a routing table * * The prune loop scans routing tables and removes routes belonging to flushing * protocols, discarded routes and also stale network entries. It is called from * rt_event(). The event is rescheduled if the current iteration do not finish * the table. The pruning is directed by the prune state (@prune_state), * specifying whether the prune cycle is scheduled or running, and there * is also a persistent pruning iterator (@prune_fit). * * The prune loop is used also for channel flushing. For this purpose, the * channels to flush are marked before the iteration and notified after the * iteration. */ static void rt_prune_table(rtable *tab) { struct fib_iterator *fit = &tab->prune_fit; int limit = 512; struct channel *c; node *n, *x; DBG("Pruning route table %s\n", tab->name); #ifdef DEBUGGING fib_check(&tab->fib); #endif if (tab->prune_state == 0) return; if (tab->prune_state == 1) { /* Mark channels to flush */ WALK_LIST2(c, n, tab->channels, table_node) if (c->channel_state == CS_FLUSHING) c->flush_active = 1; FIB_ITERATE_INIT(fit, &tab->fib); tab->prune_state = 2; } again: FIB_ITERATE_START(&tab->fib, fit, net, n) { rescan: for (struct rte_storage *e=n->routes; e; e=e->next) { if (e->rte.sender->flush_active || (e->rte.flags & REF_DISCARD)) { if (limit <= 0) { FIB_ITERATE_PUT(fit); ev_schedule(tab->rt_event); return; } rte_discard(n, &e->rte); limit--; goto rescan; } if (e->rte.flags & REF_MODIFY) { if (limit <= 0) { FIB_ITERATE_PUT(fit); ev_schedule(tab->rt_event); return; } rte_modify(n, &e->rte); limit--; goto rescan; } } if (!n->routes) /* Orphaned FIB entry */ { FIB_ITERATE_PUT(fit); fib_delete(&tab->fib, n); goto again; } } FIB_ITERATE_END; #ifdef DEBUGGING fib_check(&tab->fib); #endif tab->gc_counter = 0; tab->gc_time = current_time(); /* state change 2->0, 3->1 */ tab->prune_state &= 1; if (tab->prune_state > 0) ev_schedule(tab->rt_event); /* FIXME: This should be handled in a better way */ rt_prune_sources(); /* Close flushed channels */ WALK_LIST2_DELSAFE(c, n, x, tab->channels, table_node) if (c->flush_active) { c->flush_active = 0; channel_set_state(c, CS_DOWN); } return; } void rt_preconfig(struct config *c) { init_list(&c->tables); rt_new_table(cf_get_symbol("master4"), NET_IP4); rt_new_table(cf_get_symbol("master6"), NET_IP6); } /* * Some functions for handing internal next hop updates * triggered by rt_schedule_nhu(). */ static inline int rta_next_hop_outdated(rta *a) { struct hostentry *he = a->hostentry; if (!he) return 0; if (!he->src) return a->dest != RTD_UNREACHABLE; return (a->dest != he->dest) || (a->igp_metric != he->igp_metric) || (!he->nexthop_linkable) || !nexthop_same(&(a->nh), &(he->src->nh)); } void rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls) { a->hostentry = he; a->dest = he->dest; a->igp_metric = he->igp_metric; if (a->dest != RTD_UNICAST) { /* No nexthop */ no_nexthop: a->nh = (struct nexthop) {}; if (mls) { /* Store the label stack for later changes */ a->nh.labels_orig = a->nh.labels = mls->len; memcpy(a->nh.label, mls->stack, mls->len * sizeof(u32)); } return; } if (((!mls) || (!mls->len)) && he->nexthop_linkable) { /* Just link the nexthop chain, no label append happens. */ memcpy(&(a->nh), &(he->src->nh), nexthop_size(&(he->src->nh))); return; } struct nexthop *nhp = NULL, *nhr = NULL; int skip_nexthop = 0; for (struct nexthop *nh = &(he->src->nh); nh; nh = nh->next) { if (skip_nexthop) skip_nexthop--; else { nhr = nhp; nhp = (nhp ? (nhp->next = lp_alloc(rte_update_pool, NEXTHOP_MAX_SIZE)) : &(a->nh)); } memset(nhp, 0, NEXTHOP_MAX_SIZE); nhp->iface = nh->iface; nhp->weight = nh->weight; if (mls) { nhp->labels = nh->labels + mls->len; nhp->labels_orig = mls->len; if (nhp->labels <= MPLS_MAX_LABEL_STACK) { memcpy(nhp->label, nh->label, nh->labels * sizeof(u32)); /* First the hostentry labels */ memcpy(&(nhp->label[nh->labels]), mls->stack, mls->len * sizeof(u32)); /* Then the bottom labels */ } else { log(L_WARN "Sum of label stack sizes %d + %d = %d exceedes allowed maximum (%d)", nh->labels, mls->len, nhp->labels, MPLS_MAX_LABEL_STACK); skip_nexthop++; continue; } } else if (nh->labels) { nhp->labels = nh->labels; nhp->labels_orig = 0; memcpy(nhp->label, nh->label, nh->labels * sizeof(u32)); } if (ipa_nonzero(nh->gw)) { nhp->gw = nh->gw; /* Router nexthop */ nhp->flags |= (nh->flags & RNF_ONLINK); } else if (!(nh->iface->flags & IF_MULTIACCESS) || (nh->iface->flags & IF_LOOPBACK)) nhp->gw = IPA_NONE; /* PtP link - no need for nexthop */ else if (ipa_nonzero(he->link)) nhp->gw = he->link; /* Device nexthop with link-local address known */ else nhp->gw = he->addr; /* Device nexthop with link-local address unknown */ } if (skip_nexthop) if (nhr) nhr->next = NULL; else { a->dest = RTD_UNREACHABLE; log(L_WARN "No valid nexthop remaining, setting route unreachable"); goto no_nexthop; } } static inline struct rte_storage * rt_next_hop_update_rte(rtable *tab, net *n, rte *old) { rta *a = alloca(RTA_MAX_SIZE); memcpy(a, old->attrs, rta_size(old->attrs)); mpls_label_stack mls = { .len = a->nh.labels_orig }; memcpy(mls.stack, &a->nh.label[a->nh.labels - mls.len], mls.len * sizeof(u32)); rta_apply_hostentry(a, old->attrs->hostentry, &mls); a->cached = 0; rte e0 = *old; e0.attrs = a; return rte_store(&e0, n, tab); } static inline int rt_next_hop_update_net(rtable *tab, net *n) { struct rte_storage *new; int count = 0; struct rte_storage *old_best = n->routes; if (!old_best) return 0; for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) if (rta_next_hop_outdated(e->rte.attrs)) count++; if (!count) return 0; struct rte_multiupdate { struct rte_storage *old, *new; } *updates = alloca(sizeof(struct rte_multiupdate) * count); int pos = 0; for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) if (rta_next_hop_outdated(e->rte.attrs)) { struct rte_storage *new = rt_next_hop_update_rte(tab, n, &e->rte); /* Call a pre-comparison hook */ /* Not really an efficient way to compute this */ if (e->rte.src->proto->rte_recalculate) e->rte.src->proto->rte_recalculate(tab, n, &new->rte, &e->rte, &old_best->rte); updates[pos++] = (struct rte_multiupdate) { .old = e, .new = new, }; /* Replace the route in the list */ new->next = e->next; *k = e = new; } ASSERT_DIE(pos == count); /* Find the new best route */ struct rte_storage **new_best = NULL; for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) { if (!new_best || rte_better(&e->rte, &(*new_best)->rte)) new_best = k; } /* Relink the new best route to the first position */ new = *new_best; if (new != n->routes) { *new_best = new->next; new->next = n->routes; n->routes = new; } /* Announce the changes */ for (int i=0; irte.sender, &updates[i].new->rte, best_indicator[nb][ob]); rte_announce_i(tab, RA_UNDEF, n, updates[i].new, updates[i].old, new, old_best); } for (int i=0; inhu_fit; int max_feed = 32; if (tab->nhu_state == NHU_CLEAN) return; if (tab->nhu_state == NHU_SCHEDULED) { FIB_ITERATE_INIT(fit, &tab->fib); tab->nhu_state = NHU_RUNNING; } FIB_ITERATE_START(&tab->fib, fit, net, n) { if (max_feed <= 0) { FIB_ITERATE_PUT(fit); ev_schedule(tab->rt_event); return; } max_feed -= rt_next_hop_update_net(tab, n); } FIB_ITERATE_END; /* State change: * NHU_DIRTY -> NHU_SCHEDULED * NHU_RUNNING -> NHU_CLEAN */ tab->nhu_state &= 1; if (tab->nhu_state != NHU_CLEAN) ev_schedule(tab->rt_event); } struct rtable_config * rt_new_table(struct symbol *s, uint addr_type) { /* Hack that allows to 'redefine' the master table */ if ((s->class == SYM_TABLE) && (s->table == new_config->def_tables[addr_type]) && ((addr_type == NET_IP4) || (addr_type == NET_IP6))) return s->table; struct rtable_config *c = cfg_allocz(sizeof(struct rtable_config)); cf_define_symbol(s, SYM_TABLE, table, c); c->name = s->name; c->addr_type = addr_type; c->gc_max_ops = 1000; c->gc_min_time = 5; c->min_settle_time = 1 S; c->max_settle_time = 20 S; add_tail(&new_config->tables, &c->n); /* First table of each type is kept as default */ if (! new_config->def_tables[addr_type]) new_config->def_tables[addr_type] = c; return c; } /** * rt_lock_table - lock a routing table * @r: routing table to be locked * * Lock a routing table, because it's in use by a protocol, * preventing it from being freed when it gets undefined in a new * configuration. */ void rt_lock_table(rtable *r) { r->use_count++; } /** * rt_unlock_table - unlock a routing table * @r: routing table to be unlocked * * Unlock a routing table formerly locked by rt_lock_table(), * that is decrease its use count and delete it if it's scheduled * for deletion by configuration changes. */ void rt_unlock_table(rtable *r) { if (!--r->use_count && r->deleted) { struct config *conf = r->deleted; /* Delete the routing table by freeing its pool */ rt_shutdown(r); config_del_obstacle(conf); } } static struct rtable_config * rt_find_table_config(struct config *cf, char *name) { struct symbol *sym = cf_find_symbol(cf, name); return (sym && (sym->class == SYM_TABLE)) ? sym->table : NULL; } /** * rt_commit - commit new routing table configuration * @new: new configuration * @old: original configuration or %NULL if it's boot time config * * Scan differences between @old and @new configuration and modify * the routing tables according to these changes. If @new defines a * previously unknown table, create it, if it omits a table existing * in @old, schedule it for deletion (it gets deleted when all protocols * disconnect from it by calling rt_unlock_table()), if it exists * in both configurations, leave it unchanged. */ void rt_commit(struct config *new, struct config *old) { struct rtable_config *o, *r; DBG("rt_commit:\n"); if (old) { WALK_LIST(o, old->tables) { rtable *ot = o->table; if (!ot->deleted) { r = rt_find_table_config(new, o->name); if (r && (r->addr_type == o->addr_type) && !new->shutdown) { DBG("\t%s: same\n", o->name); r->table = ot; ot->name = r->name; ot->config = r; if (o->sorted != r->sorted) log(L_WARN "Reconfiguration of rtable sorted flag not implemented"); } else { DBG("\t%s: deleted\n", o->name); ot->deleted = old; config_add_obstacle(old); rt_lock_table(ot); rt_unlock_table(ot); } } } } WALK_LIST(r, new->tables) if (!r->table) { r->table = rt_setup(rt_table_pool, r); DBG("\t%s: created\n", r->name); add_tail(&routing_tables, &r->table->n); } DBG("\tdone\n"); } static inline void do_feed_channel(struct channel *c, net *n, rte *e) { rte_update_lock(); if (c->ra_mode == RA_ACCEPTED) rt_notify_accepted(c, n, NULL, NULL, c->refeeding); else if (c->ra_mode == RA_MERGED) rt_notify_merged(c, n, NULL, NULL, e, e, c->refeeding); else /* RA_BASIC */ { rte e0 = *e; rt_notify_basic(c, n->n.addr, &e0, &e0, c->refeeding); } rte_update_unlock(); } /** * rt_feed_channel - advertise all routes to a channel * @c: channel to be fed * * This function performs one pass of advertisement of routes to a channel that * is in the ES_FEEDING state. It is called by the protocol code as long as it * has something to do. (We avoid transferring all the routes in single pass in * order not to monopolize CPU time.) */ int rt_feed_channel(struct channel *c) { struct fib_iterator *fit = &c->feed_fit; int max_feed = 256; ASSERT(c->export_state == ES_FEEDING); if (!c->feed_active) { FIB_ITERATE_INIT(fit, &c->table->fib); c->feed_active = 1; } FIB_ITERATE_START(&c->table->fib, fit, net, n) { struct rte_storage *e = n->routes; if (max_feed <= 0) { FIB_ITERATE_PUT(fit); return 0; } if ((c->ra_mode == RA_OPTIMAL) || (c->ra_mode == RA_ACCEPTED) || (c->ra_mode == RA_MERGED)) if (e && rte_is_valid(&e->rte)) { /* In the meantime, the protocol may fell down */ if (c->export_state != ES_FEEDING) goto done; do_feed_channel(c, n, &e->rte); max_feed--; } if (c->ra_mode == RA_ANY) for(e = n->routes; e; e = e->next) { /* In the meantime, the protocol may fell down */ if (c->export_state != ES_FEEDING) goto done; if (!rte_is_valid(&e->rte)) continue; do_feed_channel(c, n, &e->rte); max_feed--; } } FIB_ITERATE_END; done: c->feed_active = 0; return 1; } /** * rt_feed_baby_abort - abort protocol feeding * @c: channel * * This function is called by the protocol code when the protocol stops or * ceases to exist during the feeding. */ void rt_feed_channel_abort(struct channel *c) { if (c->feed_active) { /* Unlink the iterator */ fit_get(&c->table->fib, &c->feed_fit); c->feed_active = 0; } } /* * Import table */ static int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { struct rtable *tab = c->in_table; net *net; if (new) net = net_get(tab, n); else { net = net_find(tab, n); if (!net) goto drop_withdraw; } /* Find the old rte */ struct rte_storage **pos = rte_find(net, src); if (*pos) { rte *old = &(*pos)->rte; if (new && rte_same(old, new)) { /* Refresh the old rte, continue with update to main rtable */ if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY)) { old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); return 1; } goto drop_update; } /* Move iterator if needed */ if (*pos == c->reload_next_rte) c->reload_next_rte = (*pos)->next; /* Remove the old rte */ struct rte_storage *del = *pos; *pos = (*pos)->next; rte_free(del, tab); tab->rt_count--; } else if (!new) goto drop_withdraw; if (!new) { if (!net->routes) fib_delete(&tab->fib, net); return 1; } struct channel_limit *l = &c->rx_limit; if (l->action && !*pos) { if (tab->rt_count >= l->limit) channel_notify_limit(c, l, PLD_RX, tab->rt_count); if (l->state == PLS_BLOCKED) { /* Required by rte_trace_in() */ new->net = n; rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); goto drop_update; } } /* Insert the new rte */ struct rte_storage *e = rte_store(new, net, tab); e->rte.sender = c; e->rte.lastmod = current_time(); e->next = *pos; *pos = e; tab->rt_count++; return 1; drop_update: c->import_stats.updates_received++; c->import_stats.updates_ignored++; if (!net->routes) fib_delete(&tab->fib, net); return 0; drop_withdraw: c->import_stats.withdraws_received++; c->import_stats.withdraws_ignored++; return 0; } int rt_reload_channel(struct channel *c) { struct rtable *tab = c->in_table; struct fib_iterator *fit = &c->reload_fit; int max_feed = 64; ASSERT(c->channel_state == CS_UP); if (!c->reload_active) { FIB_ITERATE_INIT(fit, &tab->fib); c->reload_active = 1; } do { for (struct rte_storage *e = c->reload_next_rte; e; e = e->next) { if (max_feed-- <= 0) { c->reload_next_rte = e; debug("%s channel reload burst split (max_feed=%d)", c->proto->name, max_feed); return 0; } rte r = e->rte; rte_update(c, r.net, &r, r.src); } c->reload_next_rte = NULL; FIB_ITERATE_START(&tab->fib, fit, net, n) { if (c->reload_next_rte = n->routes) { FIB_ITERATE_PUT_NEXT(fit, &tab->fib); break; } } FIB_ITERATE_END; } while (c->reload_next_rte); c->reload_active = 0; return 1; } void rt_reload_channel_abort(struct channel *c) { if (c->reload_active) { /* Unlink the iterator */ fit_get(&c->in_table->fib, &c->reload_fit); c->reload_next_rte = NULL; c->reload_active = 0; } } void rt_prune_sync(rtable *t, int all) { struct fib_iterator fit; FIB_ITERATE_INIT(&fit, &t->fib); again: FIB_ITERATE_START(&t->fib, &fit, net, n) { struct rte_storage *e, **ee = &n->routes; while (e = *ee) { if (all || (e->rte.flags & (REF_STALE | REF_DISCARD))) { *ee = e->next; rte_free(e, t); t->rt_count--; } else ee = &e->next; } if (all || !n->routes) { FIB_ITERATE_PUT(&fit); fib_delete(&t->fib, n); goto again; } } FIB_ITERATE_END; } /* * Export table */ int rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, struct rte_storage **old_exported) { struct rtable *tab = c->out_table; struct rte_src *src; net *net; if (new) { net = net_get(tab, n); src = new->src; } else { net = net_find(tab, n); src = old0->src; if (!net) goto drop; } /* Find the old rte */ struct rte_storage **pos = (c->ra_mode == RA_ANY) ? rte_find(net, src) : &net->routes; struct rte_storage *old = NULL; if (old = *pos) { if (new && rte_same(&(*pos)->rte, new)) goto drop; /* Remove the old rte */ *pos = old->next; *old_exported = old; tab->rt_count--; } if (!new) { if (!old) goto drop; if (!net->routes) fib_delete(&tab->fib, net); return 1; } /* Insert the new rte */ struct rte_storage *e = rte_store(new, net, tab); e->rte.lastmod = current_time(); e->next = *pos; *pos = e; tab->rt_count++; return 1; drop: return 0; } void rt_refeed_channel(struct channel *c) { if (!c->out_table) { channel_request_feeding(c); return; } ASSERT_DIE(c->ra_mode != RA_ANY); c->proto->feed_begin(c, 0); FIB_WALK(&c->out_table->fib, net, n) { if (!n->routes) continue; rte e = n->routes->rte; c->proto->rt_notify(c->proto, c, n->n.addr, &e, NULL); } FIB_WALK_END; c->proto->feed_end(c); } /* * Hostcache */ static inline u32 hc_hash(ip_addr a, rtable *dep) { return ipa_hash(a) ^ ptr_hash(dep); } static inline void hc_insert(struct hostcache *hc, struct hostentry *he) { uint k = he->hash_key >> hc->hash_shift; he->next = hc->hash_table[k]; hc->hash_table[k] = he; } static inline void hc_remove(struct hostcache *hc, struct hostentry *he) { struct hostentry **hep; uint k = he->hash_key >> hc->hash_shift; for (hep = &hc->hash_table[k]; *hep != he; hep = &(*hep)->next); *hep = he->next; } #define HC_DEF_ORDER 10 #define HC_HI_MARK *4 #define HC_HI_STEP 2 #define HC_HI_ORDER 16 /* Must be at most 16 */ #define HC_LO_MARK /5 #define HC_LO_STEP 2 #define HC_LO_ORDER 10 static void hc_alloc_table(struct hostcache *hc, pool *p, unsigned order) { uint hsize = 1 << order; hc->hash_order = order; hc->hash_shift = 32 - order; hc->hash_max = (order >= HC_HI_ORDER) ? ~0U : (hsize HC_HI_MARK); hc->hash_min = (order <= HC_LO_ORDER) ? 0U : (hsize HC_LO_MARK); hc->hash_table = mb_allocz(p, hsize * sizeof(struct hostentry *)); } static void hc_resize(struct hostcache *hc, pool *p, unsigned new_order) { struct hostentry **old_table = hc->hash_table; struct hostentry *he, *hen; uint old_size = 1 << hc->hash_order; uint i; hc_alloc_table(hc, p, new_order); for (i = 0; i < old_size; i++) for (he = old_table[i]; he != NULL; he=hen) { hen = he->next; hc_insert(hc, he); } mb_free(old_table); } static struct hostentry * hc_new_hostentry(struct hostcache *hc, pool *p, ip_addr a, ip_addr ll, rtable *dep, unsigned k) { struct hostentry *he = sl_alloc(hc->slab); *he = (struct hostentry) { .addr = a, .link = ll, .tab = dep, .hash_key = k, }; add_tail(&hc->hostentries, &he->ln); hc_insert(hc, he); hc->hash_items++; if (hc->hash_items > hc->hash_max) hc_resize(hc, p, hc->hash_order + HC_HI_STEP); return he; } static void hc_delete_hostentry(struct hostcache *hc, pool *p, struct hostentry *he) { rta_free(he->src); rem_node(&he->ln); hc_remove(hc, he); sl_free(hc->slab, he); hc->hash_items--; if (hc->hash_items < hc->hash_min) hc_resize(hc, p, hc->hash_order - HC_LO_STEP); } static void rt_init_hostcache(rtable *tab) { struct hostcache *hc = mb_allocz(tab->rp, sizeof(struct hostcache)); init_list(&hc->hostentries); hc->hash_items = 0; hc_alloc_table(hc, tab->rp, HC_DEF_ORDER); hc->slab = sl_new(tab->rp, sizeof(struct hostentry)); hc->lp = lp_new(tab->rp, LP_GOOD_SIZE(1024)); hc->trie = f_new_trie(hc->lp, 0); tab->hostcache = hc; } static void rt_free_hostcache(rtable *tab) { struct hostcache *hc = tab->hostcache; node *n; WALK_LIST(n, hc->hostentries) { struct hostentry *he = SKIP_BACK(struct hostentry, ln, n); rta_free(he->src); if (he->uc) log(L_ERR "Hostcache is not empty in table %s", tab->name); } /* Freed automagically by the resource pool rfree(hc->slab); rfree(hc->lp); mb_free(hc->hash_table); mb_free(hc); */ } static void rt_notify_hostcache(rtable *tab, net *net) { if (tab->hcu_scheduled) return; if (trie_match_net(tab->hostcache->trie, net->n.addr)) rt_schedule_hcu(tab); } static int if_local_addr(ip_addr a, struct iface *i) { struct ifa *b; WALK_LIST(b, i->addrs) if (ipa_equal(a, b->ip)) return 1; return 0; } u32 rt_get_igp_metric(rte *rt) { eattr *ea = ea_find(rt->attrs->eattrs, EA_GEN_IGP_METRIC); if (ea) return ea->u.data; if (rt->attrs->source == RTS_DEVICE) return 0; if (rt->src->proto->rte_igp_metric) return rt->src->proto->rte_igp_metric(rt); return IGP_METRIC_UNKNOWN; } static int rt_update_hostentry(rtable *tab, struct hostentry *he) { rta *old_src = he->src; int direct = 0; int pxlen = 0; /* Reset the hostentry */ he->src = NULL; he->dest = RTD_UNREACHABLE; he->nexthop_linkable = 0; he->igp_metric = 0; net_addr he_addr; net_fill_ip_host(&he_addr, he->addr); net *n = net_route(tab, &he_addr); if (n) { struct rte_storage *e = n->routes; rta *a = e->rte.attrs; pxlen = n->n.addr->pxlen; if (a->hostentry) { /* Recursive route should not depend on another recursive route */ log(L_WARN "Next hop address %I resolvable through recursive route for %N", he->addr, n->n.addr); goto done; } if (a->dest == RTD_UNICAST) { for (struct nexthop *nh = &(a->nh); nh; nh = nh->next) if (ipa_zero(nh->gw)) { if (if_local_addr(he->addr, nh->iface)) { /* The host address is a local address, this is not valid */ log(L_WARN "Next hop address %I is a local address of iface %s", he->addr, nh->iface->name); goto done; } direct++; } } he->src = rta_clone(a); he->dest = a->dest; he->nexthop_linkable = !direct; he->igp_metric = rt_get_igp_metric(&e->rte); } done: /* Add a prefix range to the trie */ trie_add_prefix(tab->hostcache->trie, &he_addr, pxlen, he_addr.pxlen); rta_free(old_src); return old_src != he->src; } static void rt_update_hostcache(rtable *tab) { struct hostcache *hc = tab->hostcache; struct hostentry *he; node *n, *x; /* Reset the trie */ lp_flush(hc->lp); hc->trie = f_new_trie(hc->lp, 0); WALK_LIST_DELSAFE(n, x, hc->hostentries) { he = SKIP_BACK(struct hostentry, ln, n); if (!he->uc) { hc_delete_hostentry(hc, tab->rp, he); continue; } if (rt_update_hostentry(tab, he)) rt_schedule_nhu(he->tab); } tab->hcu_scheduled = 0; } struct hostentry * rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep) { struct hostentry *he; if (!tab->hostcache) rt_init_hostcache(tab); u32 k = hc_hash(a, dep); struct hostcache *hc = tab->hostcache; for (he = hc->hash_table[k >> hc->hash_shift]; he != NULL; he = he->next) if (ipa_equal(he->addr, a) && (he->tab == dep)) return he; he = hc_new_hostentry(hc, tab->rp, a, ipa_zero(ll) ? a : ll, dep, k); rt_update_hostentry(tab, he); return he; } /* * Documentation for functions declared inline in route.h */ #if 0 /** * net_find - find a network entry * @tab: a routing table * @addr: address of the network * * net_find() looks up the given network in routing table @tab and * returns a pointer to its &net entry or %NULL if no such network * exists. */ static inline net *net_find(rtable *tab, net_addr *addr) { DUMMY; } /** * net_get - obtain a network entry * @tab: a routing table * @addr: address of the network * * net_get() looks up the given network in routing table @tab and * returns a pointer to its &net entry. If no such entry exists, it's * created. */ static inline net *net_get(rtable *tab, net_addr *addr) { DUMMY; } /** * rte_cow - copy a route for writing * @r: a route entry to be copied * * rte_cow() takes a &rte and prepares it for modification. The exact action * taken depends on the flags of the &rte -- if it's a temporary entry, it's * just returned unchanged, else a new temporary entry with the same contents * is created. * * The primary use of this function is inside the filter machinery -- when * a filter wants to modify &rte contents (to change the preference or to * attach another set of attributes), it must ensure that the &rte is not * shared with anyone else (and especially that it isn't stored in any routing * table). * * Result: a pointer to the new writable &rte. */ static inline rte * rte_cow(rte *r) { DUMMY; } #endif