diff options
Diffstat (limited to 'nest/rt-table.c')
| -rw-r--r-- | nest/rt-table.c | 1956 |
1 files changed, 1305 insertions, 651 deletions
diff --git a/nest/rt-table.c b/nest/rt-table.c index e7ff2816..50ddc141 100644 --- a/nest/rt-table.c +++ b/nest/rt-table.c @@ -26,12 +26,72 @@ * (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. + * + * There are several mechanisms that allow automatic update of routes in one + * routing table (dst) as a result of changes in another routing table (src). + * They handle issues of recursive next hop resolving, flowspec validation and + * RPKI validation. + * + * The first such mechanism is handling of recursive next hops. A route in the + * dst table has an indirect next hop address, which is resolved through a route + * in the src table (which may also be the same table) to get an immediate next + * hop. This is implemented using structure &hostcache attached to the src + * table, which contains &hostentry structures for each tracked next hop + * address. These structures are linked from recursive routes in dst tables, + * possibly multiple routes sharing one hostentry (as many routes may have the + * same indirect next hop). There is also a trie in the hostcache, which matches + * all prefixes that may influence resolving of tracked next hops. + * + * When a best route changes in the src table, the hostcache is notified using + * rt_notify_hostcache(), which immediately checks using the trie whether the + * change is relevant and if it is, then it schedules asynchronous hostcache + * recomputation. The recomputation is done by rt_update_hostcache() (called + * from rt_event() of src table), it walks through all hostentries and resolves + * them (by rt_update_hostentry()). It also updates the trie. If a change in + * hostentry resolution was found, then it schedules asynchronous nexthop + * recomputation of associated dst table. That is done by rt_next_hop_update() + * (called from rt_event() of dst table), it iterates over all routes in the dst + * table and re-examines their hostentries for changes. Note that in contrast to + * hostcache update, next hop update can be interrupted by main loop. These two + * full-table walks (over hostcache and dst table) are necessary due to absence + * of direct lookups (route -> affected nexthop, nexthop -> its route). + * + * The second mechanism is for flowspec validation, where validity of flowspec + * routes depends of resolving their network prefixes in IP routing tables. This + * is similar to the recursive next hop mechanism, but simpler as there are no + * intermediate hostcache and hostentries (because flows are less likely to + * share common net prefix than routes sharing a common next hop). In src table, + * there is a list of dst tables (list flowspec_links), this list is updated by + * flowpsec channels (by rt_flowspec_link() and rt_flowspec_unlink() during + * channel start/stop). Each dst table has its own trie of prefixes that may + * influence validation of flowspec routes in it (flowspec_trie). + * + * When a best route changes in the src table, rt_flowspec_notify() immediately + * checks all dst tables from the list using their tries to see whether the + * change is relevant for them. If it is, then an asynchronous re-validation of + * flowspec routes in the dst table is scheduled. That is also done by function + * rt_next_hop_update(), like nexthop recomputation above. It iterates over all + * flowspec routes and re-validates them. It also recalculates the trie. + * + * Note that in contrast to the hostcache update, here the trie is recalculated + * during the rt_next_hop_update(), which may be interleaved with IP route + * updates. The trie is flushed at the beginning of recalculation, which means + * that such updates may use partial trie to see if they are relevant. But it + * works anyway! Either affected flowspec was already re-validated and added to + * the trie, then IP route change would match the trie and trigger a next round + * of re-validation, or it was not yet re-validated and added to the trie, but + * will be re-validated later in this round anyway. + * + * The third mechanism is used for RPKI re-validation of IP routes and it is the + * simplest. It is just a list of subscribers in src table, who are notified + * when any change happened, but only after a settle time. Also, in RPKI case + * the dst is not a table, but a channel, who refeeds routes through a filter. */ #undef LOCAL_DEBUG #include "nest/bird.h" -#include "nest/route.h" +#include "nest/rt.h" #include "nest/protocol.h" #include "nest/iface.h" #include "lib/resource.h" @@ -44,6 +104,11 @@ #include "lib/hash.h" #include "lib/string.h" #include "lib/alloca.h" +#include "lib/flowspec.h" + +#ifdef CONFIG_BGP +#include "proto/bgp/bgp.h" +#endif pool *rt_table_pool; @@ -56,9 +121,16 @@ 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_next_hop_resolve_rte(rte *r); +static inline void rt_flowspec_resolve_rte(rte *r, struct channel *c); static inline void rt_prune_table(rtable *tab); static inline void rt_schedule_notify(rtable *tab); -static void rt_feed_channel(void *); +static void rt_flowspec_notify(rtable *tab, net *net); +static void rt_feed_by_fib(void *); +static void rt_feed_by_trie(void *); +static void rt_feed_equal(void *); +static void rt_feed_for(void *); +static uint rt_feed_net(struct rt_export_hook *c, net *n); const char *rt_import_state_name_array[TIS_MAX] = { [TIS_DOWN] = "DOWN", @@ -71,7 +143,6 @@ const char *rt_import_state_name_array[TIS_MAX] = { const char *rt_export_state_name_array[TES_MAX] = { [TES_DOWN] = "DOWN", - [TES_HUNGRY] = "HUNGRY", [TES_FEEDING] = "FEEDING", [TES_READY] = "READY", [TES_STOP] = "STOP" @@ -93,40 +164,184 @@ const char *rt_export_state_name(u8 state) return rt_export_state_name_array[state]; } +static inline struct rte_storage *rt_next_hop_update_rte(rtable *tab, net *n, rte *old); +static struct hostentry *rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep); + +static void +net_init_with_trie(struct fib *f, void *N) +{ + rtable *tab = SKIP_BACK(rtable, fib, f); + net *n = N; + + if (tab->trie) + trie_add_prefix(tab->trie, n->n.addr, n->n.addr->pxlen, n->n.addr->pxlen); + + if (tab->trie_new) + trie_add_prefix(tab->trie_new, n->n.addr, n->n.addr->pxlen, n->n.addr->pxlen); +} + +static inline net * +net_route_ip4_trie(rtable *t, const net_addr_ip4 *n0) +{ + TRIE_WALK_TO_ROOT_IP4(t->trie, n0, n) + { + net *r; + if (r = net_find_valid(t, (net_addr *) &n)) + return r; + } + TRIE_WALK_TO_ROOT_END; + + return NULL; +} + +static inline net * +net_route_vpn4_trie(rtable *t, const net_addr_vpn4 *n0) +{ + TRIE_WALK_TO_ROOT_IP4(t->trie, (const net_addr_ip4 *) n0, px) + { + net_addr_vpn4 n = NET_ADDR_VPN4(px.prefix, px.pxlen, n0->rd); + + net *r; + if (r = net_find_valid(t, (net_addr *) &n)) + return r; + } + TRIE_WALK_TO_ROOT_END; + + return NULL; +} + +static inline net * +net_route_ip6_trie(rtable *t, const net_addr_ip6 *n0) +{ + TRIE_WALK_TO_ROOT_IP6(t->trie, n0, n) + { + net *r; + if (r = net_find_valid(t, (net_addr *) &n)) + return r; + } + TRIE_WALK_TO_ROOT_END; + + return NULL; +} + +static inline net * +net_route_vpn6_trie(rtable *t, const net_addr_vpn6 *n0) +{ + TRIE_WALK_TO_ROOT_IP6(t->trie, (const net_addr_ip6 *) n0, px) + { + net_addr_vpn6 n = NET_ADDR_VPN6(px.prefix, px.pxlen, n0->rd); + + net *r; + if (r = net_find_valid(t, (net_addr *) &n)) + return r; + } + TRIE_WALK_TO_ROOT_END; + + return NULL; +} -/* Like fib_route(), but skips empty net entries */ static inline void * -net_route_ip4(rtable *t, net_addr_ip4 *n) +net_route_ip6_sadr_trie(rtable *t, const net_addr_ip6_sadr *n0) { + TRIE_WALK_TO_ROOT_IP6(t->trie, (const net_addr_ip6 *) n0, px) + { + net_addr_ip6_sadr n = NET_ADDR_IP6_SADR(px.prefix, px.pxlen, n0->src_prefix, n0->src_pxlen); + 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 longest matching src prefix. */ + for (struct fib_node *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; + } + TRIE_WALK_TO_ROOT_END; + + return NULL; +} + +static inline net * +net_route_ip4_fib(rtable *t, const net_addr_ip4 *n0) +{ + net_addr_ip4 n; + net_copy_ip4(&n, n0); + 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 net * +net_route_vpn4_fib(rtable *t, const net_addr_vpn4 *n0) +{ + net_addr_vpn4 n; + net_copy_vpn4(&n, n0); - while (r = net_find_valid(t, (net_addr *) n), (!r) && (n->pxlen > 0)) + net *r; + while (r = net_find_valid(t, (net_addr *) &n), (!r) && (n.pxlen > 0)) { - n->pxlen--; - ip4_clrbit(&n->prefix, n->pxlen); + n.pxlen--; + ip4_clrbit(&n.prefix, n.pxlen); } return r; } -static inline void * -net_route_ip6(rtable *t, net_addr_ip6 *n) +static inline net * +net_route_ip6_fib(rtable *t, const net_addr_ip6 *n0) { + net_addr_ip6 n; + net_copy_ip6(&n, n0); + 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; +} - while (r = net_find_valid(t, (net_addr *) n), (!r) && (n->pxlen > 0)) +static inline net * +net_route_vpn6_fib(rtable *t, const net_addr_vpn6 *n0) +{ + net_addr_vpn6 n; + net_copy_vpn6(&n, n0); + + net *r; + while (r = net_find_valid(t, (net_addr *) &n), (!r) && (n.pxlen > 0)) { - n->pxlen--; - ip6_clrbit(&n->prefix, n->pxlen); + n.pxlen--; + ip6_clrbit(&n.prefix, n.pxlen); } return r; } static inline void * -net_route_ip6_sadr(rtable *t, net_addr_ip6_sadr *n) +net_route_ip6_sadr_fib(rtable *t, const net_addr_ip6_sadr *n0) { - struct fib_node *fn; + net_addr_ip6_sadr n; + net_copy_ip6_sadr(&n, n0); while (1) { @@ -135,13 +350,13 @@ net_route_ip6_sadr(rtable *t, net_addr_ip6_sadr *n) /* 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) + match with the longest matching src prefix. */ + for (struct fib_node *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) && + 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); @@ -152,38 +367,52 @@ net_route_ip6_sadr(rtable *t, net_addr_ip6_sadr *n) if (best) return best; - if (!n->dst_pxlen) + if (!n.dst_pxlen) break; - n->dst_pxlen--; - ip6_clrbit(&n->dst_prefix, n->dst_pxlen); + n.dst_pxlen--; + ip6_clrbit(&n.dst_prefix, n.dst_pxlen); } return NULL; } -void * +net * 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: + if (tab->trie) + return net_route_ip4_trie(tab, (net_addr_ip4 *) n); + else + return net_route_ip4_fib (tab, (net_addr_ip4 *) n); + case NET_VPN4: - case NET_ROA4: - return net_route_ip4(tab, (net_addr_ip4 *) n0); + if (tab->trie) + return net_route_vpn4_trie(tab, (net_addr_vpn4 *) n); + else + return net_route_vpn4_fib (tab, (net_addr_vpn4 *) n); case NET_IP6: + if (tab->trie) + return net_route_ip6_trie(tab, (net_addr_ip6 *) n); + else + return net_route_ip6_fib (tab, (net_addr_ip6 *) n); + case NET_VPN6: - case NET_ROA6: - return net_route_ip6(tab, (net_addr_ip6 *) n0); + if (tab->trie) + return net_route_vpn6_trie(tab, (net_addr_vpn6 *) n); + else + return net_route_vpn6_fib (tab, (net_addr_vpn6 *) n); case NET_IP6_SADR: - return net_route_ip6_sadr(tab, (net_addr_ip6_sadr *) n0); + if (tab->trie) + return net_route_ip6_sadr_trie(tab, (net_addr_ip6_sadr *) n); + else + return net_route_ip6_sadr_fib (tab, (net_addr_ip6_sadr *) n); default: return NULL; @@ -192,7 +421,35 @@ net_route(rtable *tab, const net_addr *n) static int -net_roa_check_ip4(rtable *tab, const net_addr_ip4 *px, u32 asn) +net_roa_check_ip4_trie(rtable *tab, const net_addr_ip4 *px, u32 asn) +{ + int anything = 0; + + TRIE_WALK_TO_ROOT_IP4(tab->trie, px, px0) + { + net_addr_roa4 roa0 = NET_ADDR_ROA4(px0.prefix, px0.pxlen, 0, 0); + + struct fib_node *fn; + for (fn = fib_get_chain(&tab->fib, (net_addr *) &roa0); 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, &roa0) && r->routes && rte_is_valid(&r->routes->rte)) + { + anything = 1; + if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) + return ROA_VALID; + } + } + } + TRIE_WALK_TO_ROOT_END; + + return anything ? ROA_INVALID : ROA_UNKNOWN; +} + +static int +net_roa_check_ip4_fib(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; @@ -224,7 +481,35 @@ net_roa_check_ip4(rtable *tab, const net_addr_ip4 *px, u32 asn) } static int -net_roa_check_ip6(rtable *tab, const net_addr_ip6 *px, u32 asn) +net_roa_check_ip6_trie(rtable *tab, const net_addr_ip6 *px, u32 asn) +{ + int anything = 0; + + TRIE_WALK_TO_ROOT_IP6(tab->trie, px, px0) + { + net_addr_roa6 roa0 = NET_ADDR_ROA6(px0.prefix, px0.pxlen, 0, 0); + + struct fib_node *fn; + for (fn = fib_get_chain(&tab->fib, (net_addr *) &roa0); 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, &roa0) && r->routes && rte_is_valid(&r->routes->rte)) + { + anything = 1; + if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) + return ROA_VALID; + } + } + } + TRIE_WALK_TO_ROOT_END; + + return anything ? ROA_INVALID : ROA_UNKNOWN; +} + +static int +net_roa_check_ip6_fib(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; @@ -274,9 +559,19 @@ 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); + { + if (tab->trie) + return net_roa_check_ip4_trie(tab, (const net_addr_ip4 *) n, asn); + else + return net_roa_check_ip4_fib (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); + { + if (tab->trie) + return net_roa_check_ip6_trie(tab, (const net_addr_ip6 *) n, asn); + else + return net_roa_check_ip6_fib (tab, (const net_addr_ip6 *) n, asn); + } else return ROA_UNKNOWN; /* Should not happen */ } @@ -311,10 +606,10 @@ rte_store(const rte *r, net *net, rtable *tab) rt_lock_source(e->rte.src); - if (e->rte.attrs->cached) + if (ea_is_cached(e->rte.attrs)) e->rte.attrs = rta_clone(e->rte.attrs); else - e->rte.attrs = rta_lookup(e->rte.attrs); + e->rte.attrs = rta_lookup(e->rte.attrs, 1); return e; } @@ -328,11 +623,11 @@ rte_store(const rte *r, net *net, rtable *tab) */ void -rte_free(struct rte_storage *e, rtable *tab) +rte_free(struct rte_storage *e) { rt_unlock_source(e->rte.src); rta_free(e->rte.attrs); - sl_free(tab->rte_slab, e); + sl_free(e); } static int /* Actually better or at least as good as */ @@ -345,9 +640,12 @@ rte_better(rte *new, rte *old) if (!rte_is_valid(new)) return 0; - if (new->attrs->pref > old->attrs->pref) + u32 np = rt_get_preference(new); + u32 op = rt_get_preference(old); + + if (np > op) return 1; - if (new->attrs->pref < old->attrs->pref) + if (np < op) return 0; if (new->src->proto->proto != old->src->proto->proto) { @@ -371,7 +669,7 @@ rte_mergable(rte *pri, rte *sec) if (!rte_is_valid(pri) || !rte_is_valid(sec)) return 0; - if (pri->attrs->pref != sec->attrs->pref) + if (rt_get_preference(pri) != rt_get_preference(sec)) return 0; if (pri->src->proto->proto != sec->src->proto->proto) @@ -388,7 +686,7 @@ rte_trace(const char *name, const rte *e, int dir, const char *msg) { log(L_TRACE "%s %c %s %N %uL %uG %s", name, dir, msg, e->net, e->src->private_id, e->src->global_id, - rta_dest_name(e->attrs->dest)); + rta_dest_name(rte_dest(e))); } static inline void @@ -446,7 +744,7 @@ rte_feed_obtain(net *n, struct rte **feed, uint count) } static rte * -export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) +export_filter(struct channel *c, rte *rt, int silent) { struct proto *p = c->proto; const struct filter *filter = c->out_filter; @@ -476,7 +774,7 @@ export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) } v = filter && ((filter == FILTER_REJECT) || - (f_run(filter, rt, pool, + (f_run(filter, rt, (silent ? FF_SILENT : 0)) > F_ACCEPT)); if (v) { @@ -502,12 +800,6 @@ reject_noset: 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, const rte *old) { @@ -528,17 +820,6 @@ do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old) if (!new && old) CHANNEL_LIMIT_POP(c, OUT); - /* Apply export table */ - struct rte_storage *old_exported = NULL; - if (c->out_table) - { - if (!rte_update_out(c, net, new, old, &old_exported)) - { - channel_rte_trace_out(D_ROUTES, c, new, "idempotent"); - return; - } - } - if (new) stats->updates_accepted++; else @@ -560,10 +841,7 @@ do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old) channel_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); + p->rt_notify(p, c, net, new, old); } static void @@ -658,20 +936,13 @@ done: bmap_clear(&c->export_reject_map, rpe->old->rte.id); } - -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, struct rte **feed, uint count, linpool *pool, int silent) { _Thread_local static rte rloc; // struct proto *p = c->proto; - struct nexthop *nhs = NULL; + struct nexthop_adata *nhs = NULL; rte *best0 = feed[0]; rte *best = NULL; @@ -683,7 +954,7 @@ rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool return NULL; rloc = *best0; - best = export_filter_(c, &rloc, pool, silent); + best = export_filter(c, &rloc, silent); if (!best) /* Best route doesn't pass the filter */ @@ -699,23 +970,29 @@ rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool continue; rte tmp0 = *feed[i]; - rte *tmp = export_filter_(c, &tmp0, pool, 1); + rte *tmp = export_filter(c, &tmp0, 1); if (!tmp || !rte_is_reachable(tmp)) continue; - nhs = nexthop_merge_rta(nhs, tmp->attrs, pool, c->merge_limit); + eattr *nhea = ea_find(tmp->attrs, &ea_gen_nexthop); + ASSERT_DIE(nhea); + + if (nhs) + nhs = nexthop_merge(nhs, (struct nexthop_adata *) nhea->u.ptr, c->merge_limit, pool); + else + nhs = (struct nexthop_adata *) nhea->u.ptr; } if (nhs) { - nhs = nexthop_merge_rta(nhs, best->attrs, pool, c->merge_limit); + eattr *nhea = ea_find(best->attrs, &ea_gen_nexthop); + ASSERT_DIE(nhea); - if (nhs->next) - { - best->attrs = rta_cow(best->attrs, pool); - nexthop_link(best->attrs, nhs); - } + nhs = nexthop_merge(nhs, (struct nexthop_adata *) nhea->u.ptr, c->merge_limit, pool); + + ea_set_attr(&best->attrs, + EA_LITERAL_DIRECT_ADATA(&ea_gen_nexthop, 0, &nhs->ad)); } return best; @@ -854,16 +1131,16 @@ static void rte_announce(rtable *tab, 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)) + if (!rte_is_valid(RTE_OR_NULL(new))) new = NULL; - if (!old || !rte_is_valid(&old->rte)) + if (!rte_is_valid(RTE_OR_NULL(old))) old = NULL; - if (!new_best || !rte_is_valid(&new_best->rte)) + if (!rte_is_valid(RTE_OR_NULL(new_best))) new_best = NULL; - if (!old_best || !rte_is_valid(&old_best->rte)) + if (!rte_is_valid(RTE_OR_NULL(old_best))) old_best = NULL; if (!new && !old && !new_best && !old_best) @@ -878,6 +1155,9 @@ rte_announce(rtable *tab, net *net, struct rte_storage *new, struct rte_storage if (tab->hostcache) rt_notify_hostcache(tab, net); + + if (!EMPTY_LIST(tab->flowspec_links)) + rt_flowspec_notify(tab, net); } rt_schedule_notify(tab); @@ -892,11 +1172,33 @@ rte_announce(rtable *tab, net *net, struct rte_storage *new, struct rte_storage } struct rt_export_hook *eh; - WALK_LIST(eh, tab->exports) + WALK_LIST(eh, tab->exporter.hooks) { if (eh->export_state == TES_STOP) continue; + switch (eh->req->addr_mode) + { + case TE_ADDR_NONE: + break; + + case TE_ADDR_IN: + if (!net_in_netX(net->n.addr, eh->req->addr)) + continue; + break; + + case TE_ADDR_EQUAL: + if (!net_equal(net->n.addr, eh->req->addr)) + continue; + break; + + case TE_ADDR_FOR: + bug("Continuos export of best prefix match not implemented yet."); + + default: + bug("Strange table export address mode: %d", eh->req->addr_mode); + } + if (new) eh->stats.updates_received++; else @@ -934,16 +1236,29 @@ rte_validate(struct channel *ch, rte *e) return 0; } - if (net_type_match(n, NB_DEST) == !e->attrs->dest) + if (net_type_match(n, NB_DEST)) { - log(L_WARN "Ignoring route %N with invalid dest %d received via %s", - n, e->attrs->dest, ch->proto->name); - return 0; - } + eattr *nhea = ea_find(e->attrs, &ea_gen_nexthop); + int dest = nhea_dest(nhea); + + if (dest == RTD_NONE) + { + log(L_WARN "Ignoring route %N with no destination received via %s", + n, ch->proto->name); + return 0; + } - if ((e->attrs->dest == RTD_UNICAST) && !nexthop_is_sorted(&(e->attrs->nh))) + if ((dest == RTD_UNICAST) && + !nexthop_is_sorted((struct nexthop_adata *) nhea->u.ptr)) + { + log(L_WARN "Ignoring unsorted multipath route %N received via %s", + n, ch->proto->name); + return 0; + } + } + else if (ea_find(e->attrs, &ea_gen_nexthop)) { - log(L_WARN "Ignoring unsorted multipath route %N received via %s", + log(L_WARN "Ignoring route %N having a nexthop attribute received via %s", n, ch->proto->name); return 0; } @@ -974,6 +1289,14 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr rte *old_best = old_best_stored ? &old_best_stored->rte : NULL; rte *old = NULL; + /* If the new route is identical to the old one, we find the attributes in + * cache and clone these with no performance drop. OTOH, if we were to lookup + * the attributes, such a route definitely hasn't been anywhere yet, + * therefore it's definitely worth the time. */ + struct rte_storage *new_stored = NULL; + if (new) + new = &(new_stored = rte_store(new, net, table))->rte; + /* Find and remove original route from the same protocol */ struct rte_storage **before_old = rte_find(net, src); @@ -996,17 +1319,20 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr c->table->name, net->n.addr, old->src->proto->name, old->src->private_id, old->src->global_id); } - if (new && rte_same(old, new)) + if (new && rte_same(old, &new_stored->rte)) { /* No changes, ignore the new route and refresh the old one */ - - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); + old->stale_cycle = new->stale_cycle; if (!rte_is_filtered(new)) { stats->updates_ignored++; rt_rte_trace_in(D_ROUTES, req, new, "ignored"); } + + /* We need to free the already stored route here before returning */ + rte_free(new_stored); + return; } *before_old = (*before_old)->next; @@ -1019,8 +1345,13 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr return; } - if (req->preimport) - new = req->preimport(req, new, old); + /* If rejected by import limit, we need to pretend there is no route */ + if (req->preimport && (req->preimport(req, new, old) == 0)) + { + rte_free(new_stored); + new_stored = NULL; + new = NULL; + } int new_ok = rte_is_ok(new); int old_ok = rte_is_ok(old); @@ -1035,8 +1366,6 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr if (old_ok || new_ok) table->last_rt_change = current_time(); - struct rte_storage *new_stored = new ? rte_store(new, net, table) : NULL; - if (table->config->sorted) { /* If routes are sorted, just insert new route to appropriate position */ @@ -1173,7 +1502,7 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr if (!new_stored) hmap_clear(&table->id_map, old->id); - rte_free(old_stored, table); + rte_free(old_stored); } } @@ -1192,14 +1521,14 @@ rte_update_unlock(void) lp_flush(rte_update_pool); } -rte * +int channel_preimport(struct rt_import_request *req, rte *new, rte *old) { struct channel *c = SKIP_BACK(struct channel, in_req, req); if (new && !old) if (CHANNEL_LIMIT_PUSH(c, RX)) - return NULL; + return 0; if (!new && old) CHANNEL_LIMIT_POP(c, RX); @@ -1209,22 +1538,20 @@ channel_preimport(struct rt_import_request *req, rte *new, rte *old) if (new_in && !old_in) if (CHANNEL_LIMIT_PUSH(c, IN)) - if (c->in_keep_filtered) + if (c->in_keep & RIK_REJECTED) { new->flags |= REF_FILTERED; - return new; + return 1; } else - return NULL; + return 0; if (!new_in && old_in) CHANNEL_LIMIT_POP(c, IN); - return new; + return 1; } -static void rte_update_direct(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) { @@ -1233,15 +1560,13 @@ rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) ASSERT(c->channel_state == CS_UP); - if (c->in_table && !rte_update_in(c, n, new, src)) - return; - - return rte_update_direct(c, n, new, src); -} + /* The import reloader requires prefilter routes to be the first layer */ + if (new && (c->in_keep & RIK_PREFILTER)) + if (ea_is_cached(new->attrs) && !new->attrs->next) + new->attrs = ea_clone(new->attrs); + else + new->attrs = ea_lookup(new->attrs, 0); -static void -rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) -{ const struct filter *filter = c->in_filter; struct channel_import_stats *stats = &c->import_stats; @@ -1253,29 +1578,49 @@ rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src int fr; stats->updates_received++; - if (!rte_validate(c, new)) - { - channel_rte_trace_in(D_FILTERS, c, new, "invalid"); - stats->updates_invalid++; - new = NULL; - } - else if ((filter == FILTER_REJECT) || - ((fr = f_run(filter, new, rte_update_pool, 0)) > F_ACCEPT)) + if ((filter == FILTER_REJECT) || + ((fr = f_run(filter, new, 0)) > F_ACCEPT)) { stats->updates_filtered++; channel_rte_trace_in(D_FILTERS, c, new, "filtered out"); - if (c->in_keep_filtered) + if (c->in_keep & RIK_REJECTED) new->flags |= REF_FILTERED; else new = NULL; } + + if (new) + if (net_is_flow(n)) + rt_flowspec_resolve_rte(new, c); + else + rt_next_hop_resolve_rte(new); + + if (new && !rte_validate(c, new)) + { + channel_rte_trace_in(D_FILTERS, c, new, "invalid"); + stats->updates_invalid++; + new = NULL; + } + } else stats->withdraws_received++; rte_import(&c->in_req, n, new, src); + /* Now the route attributes are kept by the in-table cached version + * and we may drop the local handle */ + if (new && (c->in_keep & RIK_PREFILTER)) + { + /* There may be some updates on top of the original attribute block */ + ea_list *a = new->attrs; + while (a->next) + a = a->next; + + ea_free(a); + } + rte_update_unlock(); } @@ -1293,6 +1638,9 @@ rte_import(struct rt_import_request *req, const net_addr *n, rte *new, struct rt nn = net_get(hook->table, n); new->net = nn->n.addr; new->sender = hook; + + /* Set the stale cycle */ + new->stale_cycle = hook->stale_set; } else if (!(nn = net_find(hook->table, n))) { @@ -1323,44 +1671,23 @@ rte_discard(net *net, rte *old) /* Non-filtered route deletion, used during garb 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->req->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) + if (!n || !rte_is_valid(RTE_OR_NULL(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); + v = (f_run(filter, &rt, FF_SILENT) <= F_ACCEPT); rte_update_unlock(); @@ -1368,10 +1695,20 @@ rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filte } static void +rt_table_export_done(struct rt_export_hook *hook) +{ + struct rt_exporter *re = hook->table; + struct rtable *tab = SKIP_BACK(struct rtable, exporter, re); + + rt_unlock_table(tab); + DBG("Export hook %p in table %s finished uc=%u\n", hook, tab->name, tab->use_count); +} + +static void rt_export_stopped(void *data) { struct rt_export_hook *hook = data; - rtable *tab = hook->table; + struct rt_exporter *tab = hook->table; /* Unlist */ rem_node(&hook->n); @@ -1379,14 +1716,13 @@ rt_export_stopped(void *data) /* Reporting the channel as stopped. */ hook->stopped(hook->req); + /* Reporting the hook as finished. */ + CALL(tab->done, hook); + /* Freeing the hook together with its coroutine. */ rfree(hook->pool); - rt_unlock_table(tab); - - DBG("Export hook %p in table %s finished uc=%u\n", hook, tab->name, tab->use_count); } - static inline void rt_set_import_state(struct rt_import_hook *hook, u8 state) { @@ -1397,7 +1733,7 @@ rt_set_import_state(struct rt_import_hook *hook, u8 state) hook->req->log_state_change(hook->req, state); } -static inline void +void rt_set_export_state(struct rt_export_hook *hook, u8 state) { hook->last_state_change = current_time(); @@ -1438,54 +1774,114 @@ rt_stop_import(struct rt_import_request *req, void (*stopped)(struct rt_import_r hook->stopped = stopped; } -void -rt_request_export(rtable *tab, struct rt_export_request *req) +static struct rt_export_hook * +rt_table_export_start(struct rt_exporter *re, struct rt_export_request *req) { + rtable *tab = SKIP_BACK(rtable, exporter, re); rt_lock_table(tab); pool *p = rp_new(tab->rp, "Export hook"); - struct rt_export_hook *hook = req->hook = mb_allocz(p, sizeof(struct rt_export_hook)); + struct rt_export_hook *hook = mb_allocz(p, sizeof(struct rt_export_hook)); hook->pool = p; hook->lp = lp_new_default(p); - - hook->req = req; - hook->table = tab; /* stats zeroed by mb_allocz */ + switch (req->addr_mode) + { + case TE_ADDR_IN: + if (tab->trie && net_val_match(tab->addr_type, NB_IP)) + { + hook->walk_state = mb_allocz(p, sizeof (struct f_trie_walk_state)); + hook->walk_lock = rt_lock_trie(tab); + trie_walk_init(hook->walk_state, tab->trie, req->addr); + hook->event = ev_new_init(p, rt_feed_by_trie, hook); + break; + } + /* fall through */ + case TE_ADDR_NONE: + FIB_ITERATE_INIT(&hook->feed_fit, &tab->fib); + hook->event = ev_new_init(p, rt_feed_by_fib, hook); + break; - rt_set_export_state(hook, TES_HUNGRY); + case TE_ADDR_EQUAL: + hook->event = ev_new_init(p, rt_feed_equal, hook); + break; - hook->n = (node) {}; - add_tail(&tab->exports, &hook->n); + case TE_ADDR_FOR: + hook->event = ev_new_init(p, rt_feed_for, hook); + break; - FIB_ITERATE_INIT(&hook->feed_fit, &tab->fib); + default: + bug("Requested an unknown export address mode"); + } DBG("New export hook %p req %p in table %s uc=%u\n", hook, req, tab->name, tab->use_count); - rt_set_export_state(hook, TES_FEEDING); + return hook; +} + +void +rt_request_export(struct rt_exporter *re, struct rt_export_request *req) +{ + struct rt_export_hook *hook = req->hook = re->start(re, req); - hook->event = ev_new_init(p, rt_feed_channel, hook); + hook->req = req; + hook->table = re; + + hook->n = (node) {}; + add_tail(&re->hooks, &hook->n); + + /* Regular export */ + rt_set_export_state(hook, TES_FEEDING); ev_schedule_work(hook->event); } +static void +rt_table_export_stop(struct rt_export_hook *hook) +{ + rtable *tab = SKIP_BACK(rtable, exporter, hook->table); + + if (hook->export_state != TES_FEEDING) + return; + + switch (hook->req->addr_mode) + { + case TE_ADDR_IN: + if (hook->walk_lock) + { + rt_unlock_trie(tab, hook->walk_lock); + hook->walk_lock = NULL; + mb_free(hook->walk_state); + hook->walk_state = NULL; + break; + } + /* fall through */ + case TE_ADDR_NONE: + fit_get(&tab->fib, &hook->feed_fit); + break; + } +} + void rt_stop_export(struct rt_export_request *req, void (*stopped)(struct rt_export_request *)) { ASSERT_DIE(req->hook); struct rt_export_hook *hook = req->hook; - rtable *tab = hook->table; - - /* Stop feeding */ + /* Cancel the feeder event */ ev_postpone(hook->event); - if (hook->export_state == TES_FEEDING) - fit_get(&tab->fib, &hook->feed_fit); + /* Stop feeding from the exporter */ + CALL(hook->table->stop, hook); + /* Reset the event as the stopped event */ hook->event->hook = rt_export_stopped; hook->stopped = stopped; + /* Update export state */ rt_set_export_state(hook, TES_STOP); + + /* Run the stopped event */ ev_schedule(hook->event); } @@ -1504,15 +1900,38 @@ rt_stop_export(struct rt_export_request *req, void (*stopped)(struct rt_export_r * flag in rt_refresh_end() and then removing such routes in the prune loop. */ void -rt_refresh_begin(rtable *t, struct rt_import_request *req) +rt_refresh_begin(struct rt_import_request *req) { - FIB_WALK(&t->fib, net, n) - { - for (struct rte_storage *e = n->routes; e; e = e->next) - if (e->rte.sender == req->hook) - e->rte.flags |= REF_STALE; - } - FIB_WALK_END; + struct rt_import_hook *hook = req->hook; + ASSERT_DIE(hook); + ASSERT_DIE(hook->stale_set == hook->stale_valid); + + /* If the pruning routine is too slow */ + if ((hook->stale_pruned < hook->stale_valid) && (hook->stale_pruned + 128 < hook->stale_valid) + || (hook->stale_pruned > hook->stale_valid) && (hook->stale_pruned > hook->stale_valid + 128)) + { + log(L_WARN "Route refresh flood in table %s", hook->table->name); + FIB_WALK(&hook->table->fib, net, n) + { + for (struct rte_storage *e = n->routes; e; e = e->next) + if (e->rte.sender == req->hook) + e->rte.stale_cycle = 0; + } + FIB_WALK_END; + hook->stale_set = 1; + hook->stale_valid = 0; + hook->stale_pruned = 0; + } + /* Setting a new value of the stale modifier */ + else if (!++hook->stale_set) + { + /* Let's reserve the stale_cycle zero value for always-invalid routes */ + hook->stale_set = 1; + hook->stale_valid = 0; + } + + if (req->trace_routes & D_STATES) + log(L_TRACE "%s: route refresh begin [%u]", req->name, hook->stale_set); } /** @@ -1524,43 +1943,18 @@ rt_refresh_begin(rtable *t, struct rt_import_request *req) * hook. See rt_refresh_begin() for description of refresh cycles. */ void -rt_refresh_end(rtable *t, struct rt_import_request *req) +rt_refresh_end(struct rt_import_request *req) { - int prune = 0; - - FIB_WALK(&t->fib, net, n) - { - for (struct rte_storage *e = n->routes; e; e = e->next) - if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE)) - { - e->rte.flags |= REF_DISCARD; - prune = 1; - } - } - FIB_WALK_END; - - if (prune) - rt_schedule_prune(t); -} + struct rt_import_hook *hook = req->hook; + ASSERT_DIE(hook); -void -rt_modify_stale(rtable *t, struct rt_import_request *req) -{ - int prune = 0; + hook->stale_valid++; + ASSERT_DIE(hook->stale_set == hook->stale_valid); - FIB_WALK(&t->fib, net, n) - { - for (struct rte_storage *e = n->routes; e; e = e->next) - if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE) && !(e->rte.flags & REF_FILTERED)) - { - e->rte.flags |= REF_MODIFY; - prune = 1; - } - } - FIB_WALK_END; + rt_schedule_prune(hook->table); - if (prune) - rt_schedule_prune(t); + if (req->trace_routes & D_STATES) + log(L_TRACE "%s: route refresh end [%u]", req->name, hook->stale_valid); } /** @@ -1574,7 +1968,7 @@ rte_dump(struct rte_storage *e) { debug("%-1N ", e->rte.net); debug("PF=%02x ", e->rte.pflags); - rta_dump(e->rte.attrs); + ea_dump(e->rte.attrs); debug("\n"); } @@ -1638,7 +2032,7 @@ rt_dump_hooks(rtable *tab) } struct rt_export_hook *eh; - WALK_LIST(eh, tab->exports) + WALK_LIST(eh, tab->exporter.hooks) { eh->req->dump_req(eh->req); debug(" Export hook %p requested by %p:" @@ -1789,6 +2183,90 @@ rt_unsubscribe(struct rt_subscription *s) rt_unlock_table(s->tab); } +static struct rt_flowspec_link * +rt_flowspec_find_link(rtable *src, rtable *dst) +{ + struct rt_flowspec_link *ln; + WALK_LIST(ln, src->flowspec_links) + if ((ln->src == src) && (ln->dst == dst)) + return ln; + + return NULL; +} + +void +rt_flowspec_link(rtable *src, rtable *dst) +{ + ASSERT(rt_is_ip(src)); + ASSERT(rt_is_flow(dst)); + + struct rt_flowspec_link *ln = rt_flowspec_find_link(src, dst); + + if (!ln) + { + rt_lock_table(src); + rt_lock_table(dst); + + ln = mb_allocz(src->rp, sizeof(struct rt_flowspec_link)); + ln->src = src; + ln->dst = dst; + add_tail(&src->flowspec_links, &ln->n); + } + + ln->uc++; +} + +void +rt_flowspec_unlink(rtable *src, rtable *dst) +{ + struct rt_flowspec_link *ln = rt_flowspec_find_link(src, dst); + + ASSERT(ln && (ln->uc > 0)); + + ln->uc--; + + if (!ln->uc) + { + rem_node(&ln->n); + mb_free(ln); + + rt_unlock_table(src); + rt_unlock_table(dst); + } +} + +static void +rt_flowspec_notify(rtable *src, net *net) +{ + /* Only IP tables are src links */ + ASSERT(rt_is_ip(src)); + + struct rt_flowspec_link *ln; + WALK_LIST(ln, src->flowspec_links) + { + rtable *dst = ln->dst; + ASSERT(rt_is_flow(dst)); + + /* No need to inspect it further if recalculation is already active */ + if ((dst->nhu_state == NHU_SCHEDULED) || (dst->nhu_state == NHU_DIRTY)) + continue; + + if (trie_match_net(dst->flowspec_trie, net->n.addr)) + rt_schedule_nhu(dst); + } +} + +static void +rt_flowspec_reset_trie(rtable *tab) +{ + linpool *lp = tab->flowspec_trie->lp; + int ipv4 = tab->flowspec_trie->ipv4; + + lp_flush(lp); + tab->flowspec_trie = f_new_trie(lp, 0); + tab->flowspec_trie->ipv4 = ipv4; +} + static void rt_free(resource *_r) { @@ -1797,9 +2275,6 @@ rt_free(resource *_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); @@ -1835,12 +2310,7 @@ static struct resclass rt_class = { 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); + pool *p = rp_newf(pp, "Routing table %s", cf->name); rtable *t = ralloc(p, &rt_class); t->rp = p; @@ -1853,19 +2323,40 @@ rt_setup(pool *pp, struct rtable_config *cf) fib_init(&t->fib, p, t->addr_type, sizeof(net), OFFSETOF(net, n), 0, NULL); - if (!(t->internal = cf->internal)) + if (cf->trie_used) { - init_list(&t->imports); - init_list(&t->exports); - hmap_init(&t->id_map, p, 1024); - hmap_set(&t->id_map, 0); + t->trie = f_new_trie(lp_new_default(p), 0); + t->trie->ipv4 = net_val_match(t->addr_type, NB_IP4 | NB_VPN4 | NB_ROA4); + + t->fib.init = net_init_with_trie; + } + + init_list(&t->flowspec_links); + + t->exporter = (struct rt_exporter) { + .addr_type = t->addr_type, + .start = rt_table_export_start, + .stop = rt_table_export_stop, + .done = rt_table_export_done, + }; + init_list(&t->exporter.hooks); + + init_list(&t->imports); + + hmap_init(&t->id_map, p, 1024); + hmap_set(&t->id_map, 0); - init_list(&t->subscribers); + init_list(&t->subscribers); - t->rt_event = ev_new_init(p, rt_event, t); - t->last_rt_change = t->gc_time = current_time(); + 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; + t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS; + + if (rt_is_flow(t)) + { + t->flowspec_trie = f_new_trie(lp_new_default(p), 0); + t->flowspec_trie->ipv4 = (t->addr_type == NET_FLOW4); } return t; @@ -1906,7 +2397,7 @@ static void rt_prune_table(rtable *tab) { struct fib_iterator *fit = &tab->prune_fit; - int limit = 512; + int limit = 2000; struct rt_import_hook *ih; node *n, *x; @@ -1925,46 +2416,48 @@ rt_prune_table(rtable *tab) WALK_LIST2(ih, n, tab->imports, n) if (ih->import_state == TIS_STOP) rt_set_import_state(ih, TIS_FLUSHING); + else if ((ih->stale_valid != ih->stale_pruning) && (ih->stale_pruning == ih->stale_pruned)) + { + ih->stale_pruning = ih->stale_valid; + + if (ih->req->trace_routes & D_STATES) + log(L_TRACE "%s: table prune after refresh begin [%u]", ih->req->name, ih->stale_pruning); + } FIB_ITERATE_INIT(fit, &tab->fib); tab->prune_state = 2; + + if (tab->prune_trie) + { + /* Init prefix trie pruning */ + tab->trie_new = f_new_trie(lp_new_default(tab->rp), 0); + tab->trie_new->ipv4 = tab->trie->ipv4; + } } again: FIB_ITERATE_START(&tab->fib, fit, net, n) { rescan: + if (limit <= 0) + { + FIB_ITERATE_PUT(fit); + ev_schedule(tab->rt_event); + return; + } + for (struct rte_storage *e=n->routes; e; e=e->next) { - if ((e->rte.sender->import_state == TIS_FLUSHING) || (e->rte.flags & REF_DISCARD)) + struct rt_import_hook *s = e->rte.sender; + if ((s->import_state == TIS_FLUSHING) || + (e->rte.stale_cycle < s->stale_valid) || + (e->rte.stale_cycle > s->stale_set)) { - 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 */ @@ -1973,6 +2466,12 @@ again: fib_delete(&tab->fib, n); goto again; } + + if (tab->trie_new) + { + trie_add_prefix(tab->trie_new, n->n.addr, n->n.addr->pxlen, n->n.addr->pxlen); + limit--; + } } FIB_ITERATE_END; @@ -1986,6 +2485,37 @@ again: /* state change 2->0, 3->1 */ tab->prune_state &= 1; + if (tab->trie_new) + { + /* Finish prefix trie pruning */ + + if (!tab->trie_lock_count) + { + rfree(tab->trie->lp); + } + else + { + ASSERT(!tab->trie_old); + tab->trie_old = tab->trie; + tab->trie_old_lock_count = tab->trie_lock_count; + tab->trie_lock_count = 0; + } + + tab->trie = tab->trie_new; + tab->trie_new = NULL; + tab->prune_trie = 0; + } + else + { + /* Schedule prefix trie pruning */ + if (tab->trie && !tab->trie_old && (tab->trie->prefix_count > (2 * tab->fib.entries))) + { + /* state change 0->1, 2->3 */ + tab->prune_state |= 1; + tab->prune_trie = 1; + } + } + rt_prune_sources(); /* Close flushed channels */ @@ -1998,8 +2528,80 @@ again: mb_free(ih); rt_unlock_table(tab); } + else if (ih->stale_pruning != ih->stale_pruned) + { + ih->stale_pruned = ih->stale_pruning; + if (ih->req->trace_routes & D_STATES) + log(L_TRACE "%s: table prune after refresh end [%u]", ih->req->name, ih->stale_pruned); + } } +/** + * rt_lock_trie - lock a prefix trie of a routing table + * @tab: routing table with prefix trie to be locked + * + * The prune loop may rebuild the prefix trie and invalidate f_trie_walk_state + * structures. Therefore, asynchronous walks should lock the prefix trie using + * this function. That allows the prune loop to rebuild the trie, but postpones + * its freeing until all walks are done (unlocked by rt_unlock_trie()). + * + * Return a current trie that will be locked, the value should be passed back to + * rt_unlock_trie() for unlocking. + * + */ +struct f_trie * +rt_lock_trie(rtable *tab) +{ + ASSERT(tab->trie); + + tab->trie_lock_count++; + return tab->trie; +} + +/** + * rt_unlock_trie - unlock a prefix trie of a routing table + * @tab: routing table with prefix trie to be locked + * @trie: value returned by matching rt_lock_trie() + * + * Done for trie locked by rt_lock_trie() after walk over the trie is done. + * It may free the trie and schedule next trie pruning. + */ +void +rt_unlock_trie(rtable *tab, struct f_trie *trie) +{ + ASSERT(trie); + + if (trie == tab->trie) + { + /* Unlock the current prefix trie */ + ASSERT(tab->trie_lock_count); + tab->trie_lock_count--; + } + else if (trie == tab->trie_old) + { + /* Unlock the old prefix trie */ + ASSERT(tab->trie_old_lock_count); + tab->trie_old_lock_count--; + + /* Free old prefix trie that is no longer needed */ + if (!tab->trie_old_lock_count) + { + rfree(tab->trie_old->lp); + tab->trie_old = NULL; + + /* Kick prefix trie pruning that was postponed */ + if (tab->trie && (tab->trie->prefix_count > (2 * tab->fib.entries))) + { + tab->prune_trie = 1; + rt_schedule_prune(tab); + } + } + } + else + log(L_BUG "Invalid arg to rt_unlock_trie()"); +} + + void rt_preconfig(struct config *c) { @@ -2015,142 +2617,357 @@ rt_preconfig(struct config *c) * triggered by rt_schedule_nhu(). */ -static inline int -rta_next_hop_outdated(rta *a) +void +ea_set_hostentry(ea_list **to, struct rtable *dep, struct rtable *tab, ip_addr gw, ip_addr ll, u32 lnum, u32 labels[lnum]) { - struct hostentry *he = a->hostentry; - - if (!he) - return 0; + struct { + struct adata ad; + struct hostentry *he; + u32 labels[lnum]; + } *head = (void *) tmp_alloc_adata(sizeof *head - sizeof(struct adata)); - if (!he->src) - return a->dest != RTD_UNREACHABLE; + head->he = rt_get_hostentry(tab, gw, ll, dep); + memcpy(head->labels, labels, lnum * sizeof(u32)); - return (a->dest != he->dest) || (a->igp_metric != he->igp_metric) || - (!he->nexthop_linkable) || !nexthop_same(&(a->nh), &(he->src->nh)); + ea_set_attr(to, EA_LITERAL_DIRECT_ADATA( + &ea_gen_hostentry, 0, &head->ad)); } -void -rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls) + +static void +rta_apply_hostentry(ea_list **to, struct hostentry_adata *head) { - a->hostentry = he; - a->dest = he->dest; - a->igp_metric = he->igp_metric; + struct hostentry *he = head->he; + u32 *labels = head->labels; + u32 lnum = (u32 *) (head->ad.data + head->ad.length) - labels; + + ea_set_attr_u32(to, &ea_gen_igp_metric, 0, he->igp_metric); - if (a->dest != RTD_UNICAST) + if (!he->src) { - /* 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)); - } + ea_set_dest(to, 0, RTD_UNREACHABLE); return; } - if (((!mls) || (!mls->len)) && he->nexthop_linkable) + eattr *he_nh_ea = ea_find(he->src, &ea_gen_nexthop); + ASSERT_DIE(he_nh_ea); + + struct nexthop_adata *nhad = (struct nexthop_adata *) he_nh_ea->u.ptr; + int idest = nhea_dest(he_nh_ea); + + if ((idest != RTD_UNICAST) || + !lnum && he->nexthop_linkable) { /* Just link the nexthop chain, no label append happens. */ - memcpy(&(a->nh), &(he->src->nh), nexthop_size(&(he->src->nh))); + ea_copy_attr(to, he->src, &ea_gen_nexthop); return; } - struct nexthop *nhp = NULL, *nhr = NULL; - int skip_nexthop = 0; + uint total_size = OFFSETOF(struct nexthop_adata, nh); - for (struct nexthop *nh = &(he->src->nh); nh; nh = nh->next) + NEXTHOP_WALK(nh, nhad) { - if (skip_nexthop) - skip_nexthop--; - else + if (nh->labels + lnum > MPLS_MAX_LABEL_STACK) { - nhr = nhp; - nhp = (nhp ? (nhp->next = lp_alloc(rte_update_pool, NEXTHOP_MAX_SIZE)) : &(a->nh)); + log(L_WARN "Sum of label stack sizes %d + %d = %d exceedes allowed maximum (%d)", + nh->labels, lnum, nh->labels + lnum, MPLS_MAX_LABEL_STACK); + continue; } - memset(nhp, 0, NEXTHOP_MAX_SIZE); - nhp->iface = nh->iface; - nhp->weight = nh->weight; + total_size += NEXTHOP_SIZE_CNT(nh->labels + lnum); + } - 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) + if (total_size == OFFSETOF(struct nexthop_adata, nh)) + { + log(L_WARN "No valid nexthop remaining, setting route unreachable"); + + struct nexthop_adata nha = { + .ad.length = NEXTHOP_DEST_SIZE, + .dest = RTD_UNREACHABLE, + }; + + ea_set_attr_data(to, &ea_gen_nexthop, 0, &nha.ad.data, nha.ad.length); + return; + } + + struct nexthop_adata *new = (struct nexthop_adata *) tmp_alloc_adata(total_size); + struct nexthop *dest = &new->nh; + + NEXTHOP_WALK(nh, nhad) + { + if (nh->labels + lnum > MPLS_MAX_LABEL_STACK) + continue; + + memcpy(dest, nh, NEXTHOP_SIZE(nh)); + if (lnum) { - nhp->labels = nh->labels; - nhp->labels_orig = 0; - memcpy(nhp->label, nh->label, nh->labels * sizeof(u32)); + memcpy(&(dest->label[dest->labels]), labels, lnum * sizeof labels[0]); + dest->labels += lnum; } if (ipa_nonzero(nh->gw)) - { - nhp->gw = nh->gw; /* Router nexthop */ - nhp->flags |= (nh->flags & RNF_ONLINK); - } + /* Router nexthop */ + dest->flags = (dest->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 */ + dest->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 */ + dest->gw = he->link; /* Device nexthop with link-local address known */ else - nhp->gw = he->addr; /* Device nexthop with link-local address unknown */ + dest->gw = he->addr; /* Device nexthop with link-local address unknown */ + + dest = NEXTHOP_NEXT(dest); } - 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; - } + /* Fix final length */ + new->ad.length = (void *) dest - (void *) new->ad.data; + ea_set_attr(to, EA_LITERAL_DIRECT_ADATA( + &ea_gen_nexthop, 0, &new->ad)); } -static inline struct rte_storage * -rt_next_hop_update_rte(rtable *tab, net *n, rte *old) +static inline struct hostentry_adata * +rta_next_hop_outdated(ea_list *a) { - rta *a = alloca(RTA_MAX_SIZE); - memcpy(a, old->attrs, rta_size(old->attrs)); + /* First retrieve the hostentry */ + eattr *heea = ea_find(a, &ea_gen_hostentry); + if (!heea) + return NULL; + + struct hostentry_adata *head = (struct hostentry_adata *) heea->u.ptr; - mpls_label_stack mls = { .len = a->nh.labels_orig }; - memcpy(mls.stack, &a->nh.label[a->nh.labels - mls.len], mls.len * sizeof(u32)); + /* If no nexthop is present, we have to create one */ + eattr *a_nh_ea = ea_find(a, &ea_gen_nexthop); + if (!a_nh_ea) + return head; - rta_apply_hostentry(a, old->attrs->hostentry, &mls); - a->cached = 0; + struct nexthop_adata *nhad = (struct nexthop_adata *) a_nh_ea->u.ptr; + + /* Shortcut for unresolvable hostentry */ + if (!head->he->src) + return NEXTHOP_IS_REACHABLE(nhad) ? head : NULL; + + /* Comparing our nexthop with the hostentry nexthop */ + eattr *he_nh_ea = ea_find(head->he->src, &ea_gen_nexthop); + + return ( + (ea_get_int(a, &ea_gen_igp_metric, IGP_METRIC_UNKNOWN) != head->he->igp_metric) || + (!head->he->nexthop_linkable) || + (!he_nh_ea != !a_nh_ea) || + (he_nh_ea && a_nh_ea && !adata_same(he_nh_ea->u.ptr, a_nh_ea->u.ptr))) + ? head : NULL; +} + +static inline struct rte_storage * +rt_next_hop_update_rte(rtable *tab, net *n, rte *old) +{ + struct hostentry_adata *head = rta_next_hop_outdated(old->attrs); + if (!head) + return NULL; rte e0 = *old; - e0.attrs = a; + rta_apply_hostentry(&e0.attrs, head); return rte_store(&e0, n, tab); } +static inline void +rt_next_hop_resolve_rte(rte *r) +{ + eattr *heea = ea_find(r->attrs, &ea_gen_hostentry); + if (!heea) + return; + + struct hostentry_adata *head = (struct hostentry_adata *) heea->u.ptr; + + rta_apply_hostentry(&r->attrs, head); +} + +#ifdef CONFIG_BGP + +static inline int +net_flow_has_dst_prefix(const net_addr *n) +{ + ASSUME(net_is_flow(n)); + + if (n->pxlen) + return 1; + + if (n->type == NET_FLOW4) + { + const net_addr_flow4 *n4 = (void *) n; + return (n4->length > sizeof(net_addr_flow4)) && (n4->data[0] == FLOW_TYPE_DST_PREFIX); + } + else + { + const net_addr_flow6 *n6 = (void *) n; + return (n6->length > sizeof(net_addr_flow6)) && (n6->data[0] == FLOW_TYPE_DST_PREFIX); + } +} + +static inline int +rta_as_path_is_empty(ea_list *a) +{ + eattr *e = ea_find(a, "bgp_path"); + return !e || (as_path_getlen(e->u.ptr) == 0); +} + +static inline u32 +rta_get_first_asn(ea_list *a) +{ + eattr *e = ea_find(a, "bgp_path"); + u32 asn; + + return (e && as_path_get_first_regular(e->u.ptr, &asn)) ? asn : 0; +} + +static inline enum flowspec_valid +rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list *a, int interior) +{ + ASSERT(rt_is_ip(tab_ip)); + ASSERT(rt_is_flow(tab_flow)); + ASSERT(tab_ip->trie); + + /* RFC 8955 6. a) Flowspec has defined dst prefix */ + if (!net_flow_has_dst_prefix(n)) + return FLOWSPEC_INVALID; + + /* RFC 9117 4.1. Accept AS_PATH is empty (fr */ + if (interior && rta_as_path_is_empty(a)) + return FLOWSPEC_VALID; + + + /* RFC 8955 6. b) Flowspec and its best-match route have the same originator */ + + /* Find flowspec dst prefix */ + net_addr dst; + if (n->type == NET_FLOW4) + net_fill_ip4(&dst, net4_prefix(n), net4_pxlen(n)); + else + net_fill_ip6(&dst, net6_prefix(n), net6_pxlen(n)); + + /* Find best-match BGP unicast route for flowspec dst prefix */ + net *nb = net_route(tab_ip, &dst); + const rte *rb = nb ? &nb->routes->rte : NULL; + + /* Register prefix to trie for tracking further changes */ + int max_pxlen = (n->type == NET_FLOW4) ? IP4_MAX_PREFIX_LENGTH : IP6_MAX_PREFIX_LENGTH; + trie_add_prefix(tab_flow->flowspec_trie, &dst, (nb ? nb->n.addr->pxlen : 0), max_pxlen); + + /* No best-match BGP route -> no flowspec */ + if (!rb || (rt_get_source_attr(rb) != RTS_BGP)) + return FLOWSPEC_INVALID; + + /* Find ORIGINATOR_ID values */ + u32 orig_a = ea_get_int(a, "bgp_originator_id", 0); + u32 orig_b = ea_get_int(rb->attrs, "bgp_originator_id", 0); + + /* Originator is either ORIGINATOR_ID (if present), or BGP neighbor address (if not) */ + if ((orig_a != orig_b) || (!orig_a && !orig_b && !ipa_equal( + ea_get_ip(a, &ea_gen_from, IPA_NONE), + ea_get_ip(rb->attrs, &ea_gen_from, IPA_NONE) + ))) + return FLOWSPEC_INVALID; + + + /* Find ASN of the best-match route, for use in next checks */ + u32 asn_b = rta_get_first_asn(rb->attrs); + if (!asn_b) + return FLOWSPEC_INVALID; + + /* RFC 9117 4.2. For EBGP, flowspec and its best-match route are from the same AS */ + if (!interior && (rta_get_first_asn(a) != asn_b)) + return FLOWSPEC_INVALID; + + /* RFC 8955 6. c) More-specific routes are from the same AS as the best-match route */ + TRIE_WALK(tab_ip->trie, subnet, &dst) + { + net *nc = net_find_valid(tab_ip, &subnet); + if (!nc) + continue; + + const rte *rc = &nc->routes->rte; + if (rt_get_source_attr(rc) != RTS_BGP) + return FLOWSPEC_INVALID; + + if (rta_get_first_asn(rc->attrs) != asn_b) + return FLOWSPEC_INVALID; + } + TRIE_WALK_END; + + return FLOWSPEC_VALID; +} + +#endif /* CONFIG_BGP */ + +static struct rte_storage * +rt_flowspec_update_rte(rtable *tab, net *n, rte *r) +{ +#ifdef CONFIG_BGP + if (r->generation || (rt_get_source_attr(r) != RTS_BGP)) + return NULL; + + struct bgp_channel *bc = (struct bgp_channel *) SKIP_BACK(struct channel, in_req, r->sender->req); + if (!bc->base_table) + return NULL; + + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, p, bc->c.proto); + + enum flowspec_valid old = rt_get_flowspec_valid(r), + valid = rt_flowspec_check(bc->base_table, tab, n->n.addr, r->attrs, p->is_interior); + + if (old == valid) + return NULL; + + rte new = *r; + ea_set_attr_u32(&new.attrs, &ea_gen_flowspec_valid, 0, valid); + + return rte_store(&new, n, tab); +#else + return NULL; +#endif +} + +static inline void +rt_flowspec_resolve_rte(rte *r, struct channel *c) +{ +#ifdef CONFIG_BGP + enum flowspec_valid valid, old = rt_get_flowspec_valid(r); + struct bgp_channel *bc = (struct bgp_channel *) c; + + if ( (rt_get_source_attr(r) == RTS_BGP) + && (c->channel == &channel_bgp) + && (bc->base_table)) + { + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, p, bc->c.proto); + valid = rt_flowspec_check( + bc->base_table, + c->in_req.hook->table, + r->net, r->attrs, p->is_interior); + } + else + valid = FLOWSPEC_UNKNOWN; + + if (valid == old) + return; + + if (valid == FLOWSPEC_UNKNOWN) + ea_unset_attr(&r->attrs, 0, &ea_gen_flowspec_valid); + else + ea_set_attr_u32(&r->attrs, &ea_gen_flowspec_valid, 0, valid); +#endif +} + static inline int rt_next_hop_update_net(rtable *tab, net *n) { struct rte_storage *new; int count = 0; + int is_flow = net_is_flow(n->n.addr); 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)) + if (is_flow || rta_next_hop_outdated(e->rte.attrs)) count++; if (!count) @@ -2162,9 +2979,14 @@ rt_next_hop_update_net(rtable *tab, net *n) int pos = 0; for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) - if (rta_next_hop_outdated(e->rte.attrs)) + if (is_flow || rta_next_hop_outdated(e->rte.attrs)) { - struct rte_storage *new = rt_next_hop_update_rte(tab, n, &e->rte); + struct rte_storage *new = is_flow + ? rt_flowspec_update_rte(tab, n, &e->rte) + : rt_next_hop_update_rte(tab, n, &e->rte); + + if (!new) + continue; /* Call a pre-comparison hook */ /* Not really an efficient way to compute this */ @@ -2181,7 +3003,8 @@ rt_next_hop_update_net(rtable *tab, net *n) *k = e = new; } - ASSERT_DIE(pos == count); + ASSERT_DIE(pos <= count); + count = pos; /* Find the new best route */ struct rte_storage **new_best = NULL; @@ -2210,7 +3033,7 @@ rt_next_hop_update_net(rtable *tab, net *n) } for (int i=0; i<count; i++) - rte_free(updates[i].old, tab); + rte_free(updates[i].old); return count; } @@ -2228,6 +3051,9 @@ rt_next_hop_update(rtable *tab) { FIB_ITERATE_INIT(fit, &tab->fib); tab->nhu_state = NHU_RUNNING; + + if (tab->flowspec_trie) + rt_flowspec_reset_trie(tab); } FIB_ITERATE_START(&tab->fib, fit, net, n) @@ -2316,6 +3142,22 @@ rt_unlock_table(rtable *r) } } +static int +rt_reconfigure(rtable *tab, struct rtable_config *new, struct rtable_config *old) +{ + if ((new->addr_type != old->addr_type) || + (new->sorted != old->sorted) || + (new->trie_used != old->trie_used)) + return 0; + + DBG("\t%s: same\n", new->name); + new->table = tab; + tab->name = new->name; + tab->config = new; + + return 1; +} + static struct rtable_config * rt_find_table_config(struct config *cf, char *name) { @@ -2345,28 +3187,19 @@ rt_commit(struct config *new, struct config *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); - } - } + rtable *tab = o->table; + if (tab->deleted) + continue; + + r = rt_find_table_config(new, o->name); + if (r && !new->shutdown && rt_reconfigure(tab, r, o)) + continue; + + DBG("\t%s: deleted\n", o->name); + tab->deleted = old; + config_add_obstacle(old); + rt_lock_table(tab); + rt_unlock_table(tab); } } @@ -2381,7 +3214,7 @@ rt_commit(struct config *new, struct config *old) } /** - * rt_feed_channel - advertise all routes to a channel + * rt_feed_by_fib - advertise all routes to a channel by walking a fib * @c: channel to be fed * * This function performs one pass of advertisement of routes to a channel that @@ -2390,7 +3223,7 @@ rt_commit(struct config *new, struct config *old) * order not to monopolize CPU time.) */ static void -rt_feed_channel(void *data) +rt_feed_by_fib(void *data) { struct rt_export_hook *c = data; @@ -2399,7 +3232,9 @@ rt_feed_channel(void *data) ASSERT(c->export_state == TES_FEEDING); - FIB_ITERATE_START(&c->table->fib, fit, net, n) + rtable *tab = SKIP_BACK(rtable, exporter, c->table); + + FIB_ITERATE_START(&tab->fib, fit, net, n) { if (max_feed <= 0) { @@ -2408,315 +3243,134 @@ rt_feed_channel(void *data) return; } - if (c->export_state != TES_FEEDING) - goto done; + ASSERT(c->export_state == TES_FEEDING); - if (c->req->export_bulk) - { - uint count = rte_feed_count(n); - if (count) - { - rte_update_lock(); - rte **feed = alloca(count * sizeof(rte *)); - rte_feed_obtain(n, feed, count); - struct rt_pending_export rpe = { .new_best = n->routes }; - c->req->export_bulk(c->req, n->n.addr, &rpe, feed, count); - max_feed -= count; - rte_update_unlock(); - } - } - else if (n->routes && rte_is_valid(&n->routes->rte)) - { - rte_update_lock(); - struct rt_pending_export rpe = { .new = n->routes, .new_best = n->routes }; - c->req->export_one(c->req, n->n.addr, &rpe); - max_feed--; - rte_update_unlock(); - } + if ((c->req->addr_mode == TE_ADDR_NONE) || net_in_netX(n->n.addr, c->req->addr)) + max_feed -= rt_feed_net(c, n); } FIB_ITERATE_END; -done: rt_set_export_state(c, TES_READY); } - -/* - * Import table - */ - -int -rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +static void +rt_feed_by_trie(void *data) { - 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; - } - - if (!new) - CHANNEL_LIMIT_POP(c, RX); + struct rt_export_hook *c = data; + rtable *tab = SKIP_BACK(rtable, exporter, c->table); - /* Move iterator if needed */ - if (*pos == c->reload_next_rte) - c->reload_next_rte = (*pos)->next; + ASSERT_DIE(c->walk_state); + struct f_trie_walk_state *ws = c->walk_state; - /* Remove the old rte */ - struct rte_storage *del = *pos; - *pos = (*pos)->next; - rte_free(del, tab); - tab->rt_count--; - } - else if (new) - { - if (CHANNEL_LIMIT_PUSH(c, RX)) - { - /* Required by rte_trace_in() */ - new->net = n; + int max_feed = 256; - channel_rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - goto drop_update; - } - } - else - goto drop_withdraw; + ASSERT_DIE(c->export_state == TES_FEEDING); - if (!new) + net_addr addr; + while (trie_walk_next(ws, &addr)) { - if (!net->routes) - fib_delete(&tab->fib, net); - - return 1; - } + net *n = net_find(tab, &addr); + if (!n) + continue; - /* 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; + if ((max_feed -= rt_feed_net(c, n)) <= 0) + return; -drop_update: - c->import_stats.updates_received++; - c->in_req.hook->stats.updates_ignored++; + ASSERT_DIE(c->export_state == TES_FEEDING); + } - if (!net->routes) - fib_delete(&tab->fib, net); + rt_unlock_trie(tab, c->walk_lock); + c->walk_lock = NULL; - return 0; + mb_free(c->walk_state); + c->walk_state = NULL; -drop_withdraw: - c->import_stats.withdraws_received++; - c->in_req.hook->stats.withdraws_ignored++; - return 0; + rt_set_export_state(c, TES_READY); } -int -rt_reload_channel(struct channel *c) +static void +rt_feed_equal(void *data) { - struct rtable *tab = c->in_table; - struct fib_iterator *fit = &c->reload_fit; - int max_feed = 64; - - ASSERT(c->channel_state == CS_UP); + struct rt_export_hook *c = data; + rtable *tab = SKIP_BACK(rtable, exporter, c->table); - if (!c->reload_active) - { - FIB_ITERATE_INIT(fit, &tab->fib); - c->reload_active = 1; - } + ASSERT_DIE(c->export_state == TES_FEEDING); + ASSERT_DIE(c->req->addr_mode == TE_ADDR_EQUAL); - 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; - } + net *n = net_find(tab, c->req->addr); + if (n) + rt_feed_net(c, n); - rte r = e->rte; - rte_update_direct(c, r.net, &r, r.src); - } + rt_set_export_state(c, TES_READY); +} - c->reload_next_rte = NULL; +static void +rt_feed_for(void *data) +{ + struct rt_export_hook *c = data; + rtable *tab = SKIP_BACK(rtable, exporter, c->table); - 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); + ASSERT_DIE(c->export_state == TES_FEEDING); + ASSERT_DIE(c->req->addr_mode == TE_ADDR_FOR); - c->reload_active = 0; - return 1; -} + net *n = net_route(tab, c->req->addr); + if (n) + rt_feed_net(c, n); -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; - } + rt_set_export_state(c, TES_READY); } -void -rt_prune_sync(rtable *t, int all) +static uint +rt_feed_net(struct rt_export_hook *c, net *n) { - 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; + if (c->req->export_bulk) + { + uint count = rte_feed_count(n); + if (count) + { + rte_update_lock(); + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(n, feed, count); + struct rt_pending_export rpe = { .new_best = n->routes }; + c->req->export_bulk(c->req, n->n.addr, &rpe, feed, count); + rte_update_unlock(); + } + return count; + } - while (e = *ee) - { - if (all || (e->rte.flags & (REF_STALE | REF_DISCARD))) + if (n->routes && rte_is_valid(&n->routes->rte)) { - *ee = e->next; - rte_free(e, t); - t->rt_count--; + rte_update_lock(); + struct rt_pending_export rpe = { .new = n->routes, .new_best = n->routes }; + c->req->export_one(c->req, n->n.addr, &rpe); + rte_update_unlock(); + return 1; } - else - ee = &e->next; - } - if (all || !n->routes) - { - FIB_ITERATE_PUT(&fit); - fib_delete(&t->fib, n); - goto again; - } - } - FIB_ITERATE_END; + return 0; } /* - * Export table + * Import table */ -int -rte_update_out(struct channel *c, const net_addr *n, rte *new, const 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) +void channel_reload_export_bulk(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) { - if (!c->out_table) - { - channel_request_feeding(c); - return; - } - - ASSERT_DIE(c->ra_mode != RA_ANY); - - c->proto->feed_begin(c, 0); + struct channel *c = SKIP_BACK(struct channel, reload_req, req); - 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; + for (uint i=0; i<count; i++) + if (feed[i]->sender == c->in_req.hook) + { + /* Strip the later attribute layers */ + rte new = *feed[i]; + while (new.attrs->next) + new.attrs = new.attrs->next; - c->proto->feed_end(c); + /* And reload the route */ + rte_update(c, net, &new, new.src); + } } @@ -2815,7 +3469,7 @@ hc_delete_hostentry(struct hostcache *hc, pool *p, struct hostentry *he) rem_node(&he->ln); hc_remove(hc, he); - sl_free(hc->slab, he); + sl_free(he); hc->hash_items--; if (hc->hash_items < hc->hash_min) @@ -2832,7 +3486,7 @@ rt_init_hostcache(rtable *tab) 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->lp = lp_new(tab->rp); hc->trie = f_new_trie(hc->lp, 0); tab->hostcache = hc; @@ -2884,14 +3538,14 @@ if_local_addr(ip_addr a, struct iface *i) } u32 -rt_get_igp_metric(rte *rt) +rt_get_igp_metric(const rte *rt) { - eattr *ea = ea_find(rt->attrs->eattrs, EA_GEN_IGP_METRIC); + eattr *ea = ea_find(rt->attrs, "igp_metric"); if (ea) return ea->u.data; - if (rt->attrs->source == RTS_DEVICE) + if (rt_get_source_attr(rt) == RTS_DEVICE) return 0; if (rt->src->proto->rte_igp_metric) @@ -2903,13 +3557,12 @@ rt_get_igp_metric(rte *rt) static int rt_update_hostentry(rtable *tab, struct hostentry *he) { - rta *old_src = he->src; + ea_list *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; @@ -2919,10 +3572,10 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) if (n) { struct rte_storage *e = n->routes; - rta *a = e->rte.attrs; + ea_list *a = e->rte.attrs; pxlen = n->n.addr->pxlen; - if (a->hostentry) + if (ea_find(a, &ea_gen_hostentry)) { /* Recursive route should not depend on another recursive route */ log(L_WARN "Next hop address %I resolvable through recursive route for %N", @@ -2930,9 +3583,12 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) goto done; } - if (a->dest == RTD_UNICAST) - { - for (struct nexthop *nh = &(a->nh); nh; nh = nh->next) + eattr *nhea = ea_find(a, &ea_gen_nexthop); + ASSERT_DIE(nhea); + struct nexthop_adata *nhad = (void *) nhea->u.ptr; + + if (NEXTHOP_IS_REACHABLE(nhad)) + NEXTHOP_WALK(nh, nhad) if (ipa_zero(nh->gw)) { if (if_local_addr(he->addr, nh->iface)) @@ -2945,10 +3601,8 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) direct++; } - } he->src = rta_clone(a); - he->dest = a->dest; he->nexthop_linkable = !direct; he->igp_metric = rt_get_igp_metric(&e->rte); } @@ -2988,7 +3642,7 @@ rt_update_hostcache(rtable *tab) tab->hcu_scheduled = 0; } -struct hostentry * +static struct hostentry * rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep) { struct hostentry *he; |