diff options
Diffstat (limited to 'nest/rt-table.c')
| -rw-r--r-- | nest/rt-table.c | 4057 |
1 files changed, 2509 insertions, 1548 deletions
diff --git a/nest/rt-table.c b/nest/rt-table.c index abb29fe1..95248635 100644 --- a/nest/rt-table.c +++ b/nest/rt-table.c @@ -43,10 +43,10 @@ * 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 + * an auxiliary export request, which 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 + * as an 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() @@ -60,15 +60,14 @@ * 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). + * share common net prefix than routes sharing a common next hop). Every dst + * table has its own export request in every src table. 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 + * When a best route changes in the src table, the notification mechanism is + * invoked by the export request which checks its dst table's trie to see + * whether the change is relevant, and if so, 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. @@ -83,15 +82,14 @@ * 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. + * simplest. It is also an auxiliary export request belonging to the + * appropriate channel, triggering its reload/refeed timer after a settle time. */ #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" @@ -105,32 +103,96 @@ #include "lib/string.h" #include "lib/alloca.h" #include "lib/flowspec.h" +#include "lib/idm.h" #ifdef CONFIG_BGP #include "proto/bgp/bgp.h" #endif -pool *rt_table_pool; +#include <stdatomic.h> -static slab *rte_slab; -static linpool *rte_update_pool; +pool *rt_table_pool; list routing_tables; +list deleted_routing_tables; + +struct rt_cork rt_cork; -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); -static void rt_flowspec_notify(rtable *tab, net *net); -static void rt_kick_prune_timer(rtable *tab); +/* Data structures for export journal */ +#define RT_PENDING_EXPORT_ITEMS (page_size - sizeof(struct rt_export_block)) / sizeof(struct rt_pending_export) +struct rt_export_block { + node n; + _Atomic u32 end; + _Atomic _Bool not_last; + struct rt_pending_export export[]; +}; + +static void rt_free_hostcache(struct rtable_private *tab); +static void rt_update_hostcache(void *tab); +static void rt_next_hop_update(void *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(struct rtable_private *tab); +static void rt_kick_prune_timer(struct rtable_private *tab); +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 void rt_check_cork_low(struct rtable_private *tab); +static void rt_check_cork_high(struct rtable_private *tab); +static void rt_cork_release_hook(void *); +static void rt_delete(void *); + +static void rt_export_used(struct rt_table_exporter *); +static void rt_export_cleanup(struct rtable_private *tab); + +static int rte_same(rte *x, rte *y); + +const char *rt_import_state_name_array[TIS_MAX] = { + [TIS_DOWN] = "DOWN", + [TIS_UP] = "UP", + [TIS_STOP] = "STOP", + [TIS_FLUSHING] = "FLUSHING", + [TIS_WAITING] = "WAITING", + [TIS_CLEARED] = "CLEARED", +}; + +const char *rt_export_state_name_array[TES_MAX] = { + [TES_DOWN] = "DOWN", + [TES_FEEDING] = "FEEDING", + [TES_READY] = "READY", + [TES_STOP] = "STOP" +}; + +const char *rt_import_state_name(u8 state) +{ + if (state >= TIS_MAX) + return "!! INVALID !!"; + else + return rt_import_state_name_array[state]; +} + +const char *rt_export_state_name(u8 state) +{ + if (state >= TES_MAX) + return "!! INVALID !!"; + else + return rt_export_state_name_array[state]; +} + +static struct hostentry *rt_get_hostentry(struct rtable_private *tab, ip_addr a, ip_addr ll, rtable *dep); + +#define rt_trace(tab, level, fmt, args...) do {\ + struct rtable_private *t = (tab); \ + if (t->config->debug & (level)) \ + log(L_TRACE "%s: " fmt, t->name, ##args); \ +} while (0) static void net_init_with_trie(struct fib *f, void *N) { - rtable *tab = SKIP_BACK(rtable, fib, f); + struct rtable_private *tab = SKIP_BACK(struct rtable_private, fib, f); net *n = N; if (tab->trie) @@ -141,7 +203,7 @@ net_init_with_trie(struct fib *f, void *N) } static inline net * -net_route_ip4_trie(rtable *t, const net_addr_ip4 *n0) +net_route_ip4_trie(struct rtable_private *t, const net_addr_ip4 *n0) { TRIE_WALK_TO_ROOT_IP4(t->trie, n0, n) { @@ -155,7 +217,7 @@ net_route_ip4_trie(rtable *t, const net_addr_ip4 *n0) } static inline net * -net_route_vpn4_trie(rtable *t, const net_addr_vpn4 *n0) +net_route_vpn4_trie(struct rtable_private *t, const net_addr_vpn4 *n0) { TRIE_WALK_TO_ROOT_IP4(t->trie, (const net_addr_ip4 *) n0, px) { @@ -171,7 +233,7 @@ net_route_vpn4_trie(rtable *t, const net_addr_vpn4 *n0) } static inline net * -net_route_ip6_trie(rtable *t, const net_addr_ip6 *n0) +net_route_ip6_trie(struct rtable_private *t, const net_addr_ip6 *n0) { TRIE_WALK_TO_ROOT_IP6(t->trie, n0, n) { @@ -185,7 +247,7 @@ net_route_ip6_trie(rtable *t, const net_addr_ip6 *n0) } static inline net * -net_route_vpn6_trie(rtable *t, const net_addr_vpn6 *n0) +net_route_vpn6_trie(struct rtable_private *t, const net_addr_vpn6 *n0) { TRIE_WALK_TO_ROOT_IP6(t->trie, (const net_addr_ip6 *) n0, px) { @@ -201,7 +263,7 @@ net_route_vpn6_trie(rtable *t, const net_addr_vpn6 *n0) } static inline void * -net_route_ip6_sadr_trie(rtable *t, const net_addr_ip6_sadr *n0) +net_route_ip6_sadr_trie(struct rtable_private *t, const net_addr_ip6_sadr *n0) { TRIE_WALK_TO_ROOT_IP6(t->trie, (const net_addr_ip6 *) n0, px) { @@ -234,7 +296,7 @@ net_route_ip6_sadr_trie(rtable *t, const net_addr_ip6_sadr *n0) } static inline net * -net_route_ip4_fib(rtable *t, const net_addr_ip4 *n0) +net_route_ip4_fib(struct rtable_private *t, const net_addr_ip4 *n0) { net_addr_ip4 n; net_copy_ip4(&n, n0); @@ -250,7 +312,7 @@ net_route_ip4_fib(rtable *t, const net_addr_ip4 *n0) } static inline net * -net_route_vpn4_fib(rtable *t, const net_addr_vpn4 *n0) +net_route_vpn4_fib(struct rtable_private *t, const net_addr_vpn4 *n0) { net_addr_vpn4 n; net_copy_vpn4(&n, n0); @@ -266,7 +328,7 @@ net_route_vpn4_fib(rtable *t, const net_addr_vpn4 *n0) } static inline net * -net_route_ip6_fib(rtable *t, const net_addr_ip6 *n0) +net_route_ip6_fib(struct rtable_private *t, const net_addr_ip6 *n0) { net_addr_ip6 n; net_copy_ip6(&n, n0); @@ -282,7 +344,7 @@ net_route_ip6_fib(rtable *t, const net_addr_ip6 *n0) } static inline net * -net_route_vpn6_fib(rtable *t, const net_addr_vpn6 *n0) +net_route_vpn6_fib(struct rtable_private *t, const net_addr_vpn6 *n0) { net_addr_vpn6 n; net_copy_vpn6(&n, n0); @@ -298,7 +360,7 @@ net_route_vpn6_fib(rtable *t, const net_addr_vpn6 *n0) } static inline void * -net_route_ip6_sadr_fib(rtable *t, const net_addr_ip6_sadr *n0) +net_route_ip6_sadr_fib(struct rtable_private *t, const net_addr_ip6_sadr *n0) { net_addr_ip6_sadr n; net_copy_ip6_sadr(&n, n0); @@ -338,7 +400,7 @@ net_route_ip6_sadr_fib(rtable *t, const net_addr_ip6_sadr *n0) } net * -net_route(rtable *tab, const net_addr *n) +net_route(struct rtable_private *tab, const net_addr *n) { ASSERT(tab->addr_type == n->type); @@ -381,7 +443,7 @@ net_route(rtable *tab, const net_addr *n) static int -net_roa_check_ip4_trie(rtable *tab, const net_addr_ip4 *px, u32 asn) +net_roa_check_ip4_trie(struct rtable_private *tab, const net_addr_ip4 *px, u32 asn) { int anything = 0; @@ -395,7 +457,7 @@ net_roa_check_ip4_trie(rtable *tab, const net_addr_ip4 *px, u32 asn) net_addr_roa4 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa4(roa, &roa0) && rte_is_valid(r->routes)) + 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)) @@ -409,7 +471,7 @@ net_roa_check_ip4_trie(rtable *tab, const net_addr_ip4 *px, u32 asn) } static int -net_roa_check_ip4_fib(rtable *tab, const net_addr_ip4 *px, u32 asn) +net_roa_check_ip4_fib(struct rtable_private *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; @@ -422,7 +484,7 @@ net_roa_check_ip4_fib(rtable *tab, const net_addr_ip4 *px, u32 asn) net_addr_roa4 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa4(roa, &n) && rte_is_valid(r->routes)) + 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)) @@ -441,7 +503,7 @@ net_roa_check_ip4_fib(rtable *tab, const net_addr_ip4 *px, u32 asn) } static int -net_roa_check_ip6_trie(rtable *tab, const net_addr_ip6 *px, u32 asn) +net_roa_check_ip6_trie(struct rtable_private *tab, const net_addr_ip6 *px, u32 asn) { int anything = 0; @@ -455,7 +517,7 @@ net_roa_check_ip6_trie(rtable *tab, const net_addr_ip6 *px, u32 asn) net_addr_roa6 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa6(roa, &roa0) && rte_is_valid(r->routes)) + 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)) @@ -469,7 +531,7 @@ net_roa_check_ip6_trie(rtable *tab, const net_addr_ip6 *px, u32 asn) } static int -net_roa_check_ip6_fib(rtable *tab, const net_addr_ip6 *px, u32 asn) +net_roa_check_ip6_fib(struct rtable_private *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; @@ -482,7 +544,7 @@ net_roa_check_ip6_fib(rtable *tab, const net_addr_ip6 *px, u32 asn) net_addr_roa6 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa6(roa, &n) && rte_is_valid(r->routes)) + 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)) @@ -516,24 +578,30 @@ net_roa_check_ip6_fib(rtable *tab, const net_addr_ip6 *px, u32 asn) * must have type NET_IP4 or NET_IP6, respectively. */ int -net_roa_check(rtable *tab, const net_addr *n, u32 asn) +net_roa_check(rtable *tp, const net_addr *n, u32 asn) { - if ((tab->addr_type == NET_ROA4) && (n->type == NET_IP4)) - { - 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)) + int out = ROA_UNKNOWN; + + RT_LOCKED(tp, tab) { - if (tab->trie) - return net_roa_check_ip6_trie(tab, (const net_addr_ip6 *) n, asn); + if ((tab->addr_type == NET_ROA4) && (n->type == NET_IP4)) + { + if (tab->trie) + out = net_roa_check_ip4_trie(tab, (const net_addr_ip4 *) n, asn); + else + out = net_roa_check_ip4_fib (tab, (const net_addr_ip4 *) n, asn); + } + else if ((tab->addr_type == NET_ROA6) && (n->type == NET_IP6)) + { + if (tab->trie) + out = net_roa_check_ip6_trie(tab, (const net_addr_ip6 *) n, asn); + else + out = net_roa_check_ip6_fib (tab, (const net_addr_ip6 *) n, asn); + } else - return net_roa_check_ip6_fib (tab, (const net_addr_ip6 *) n, asn); + out = ROA_UNKNOWN; /* Should not happen */ } - else - return ROA_UNKNOWN; /* Should not happen */ + return out; } /** @@ -541,83 +609,53 @@ net_roa_check(rtable *tab, const net_addr *n, u32 asn) * @net: network node * @src: route source * - * The rte_find() function returns a route for destination @net - * which is from route source @src. + * 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. */ -rte * +static struct rte_storage ** rte_find(net *net, struct rte_src *src) { - rte *e = net->routes; + struct rte_storage **e = &net->routes; + + while ((*e) && (*e)->rte.src != src) + e = &(*e)->next; - while (e && e->src != src) - e = e->next; return e; } -/** - * rte_get_temp - get a temporary &rte - * @a: attributes to assign to the new route (a &rta; in case it's - * un-cached, rte_update() will create a cached copy automatically) - * - * Create a temporary &rte and bind it with the attributes @a. - * Also set route preference to the default preference set for - * the protocol. - */ -rte * -rte_get_temp(rta *a, struct rte_src *src) + +struct rte_storage * +rte_store(const rte *r, net *net, struct rtable_private *tab) { - rte *e = sl_alloc(rte_slab); + struct rte_storage *e = sl_alloc(tab->rte_slab); - e->attrs = a; - e->id = 0; - e->flags = 0; - rt_lock_source(e->src = src); - return e; -} + e->rte = *r; + e->rte.net = net->n.addr; -rte * -rte_do_cow(rte *r) -{ - rte *e = sl_alloc(rte_slab); + rt_lock_source(e->rte.src); - memcpy(e, r, sizeof(rte)); + if (ea_is_cached(e->rte.attrs)) + e->rte.attrs = rta_clone(e->rte.attrs); + else + e->rte.attrs = rta_lookup(e->rte.attrs, 1); - rt_lock_source(e->src); - e->attrs = rta_clone(r->attrs); - e->flags = 0; return e; } /** - * rte_cow_rta - get a private writable copy of &rte with writable &rta - * @r: a route entry to be copied - * @lp: a linpool from which to allocate &rta - * - * rte_cow_rta() takes a &rte and prepares it and associated &rta for - * modification. There are three possibilities: First, both &rte and &rta are - * private copies, in that case they are returned unchanged. Second, &rte is - * private copy, but &rta is cached, in that case &rta is duplicated using - * rta_do_cow(). Third, both &rte is shared and &rta is cached, in that case - * both structures are duplicated by rte_do_cow() and rta_do_cow(). - * - * Note that in the second case, cached &rta loses one reference, while private - * copy created by rta_do_cow() is a shallow copy sharing indirect data (eattrs, - * nexthops, ...) with it. To work properly, original shared &rta should have - * another reference during the life of created private copy. + * rte_free - delete a &rte + * @e: &struct rte_storage to be deleted + * @tab: the table which the rte belongs to * - * Result: a pointer to the new writable &rte with writable &rta. + * rte_free() deletes the given &rte from the routing table it's linked to. */ -rte * -rte_cow_rta(rte *r, linpool *lp) -{ - if (!rta_is_cached(r->attrs)) - return r; - r = rte_cow(r); - rta *a = rta_do_cow(r->attrs, lp); - rta_free(r->attrs); - r->attrs = a; - return r; +void +rte_free(struct rte_storage *e) +{ + rt_unlock_source(e->rte.src); + rta_free(e->rte.attrs); + sl_free(e); } static int /* Actually better or at least as good as */ @@ -630,20 +668,23 @@ 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) + if (new->src->owner->class != old->src->owner->class) { /* * 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; + return new->src->owner->class > old->src->owner->class; } - if (better = new->src->proto->rte_better) + if (better = new->src->owner->class->rte_better) return better(new, old); return 0; } @@ -656,172 +697,197 @@ 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) + if (pri->src->owner->class != sec->src->owner->class) return 0; - if (mergable = pri->src->proto->rte_mergable) + if (mergable = pri->src->owner->class->rte_mergable) return mergable(pri, sec); return 0; } static void -rte_trace(struct channel *c, rte *e, int dir, char *msg) +rte_trace(const char *name, const 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->n.addr, e->src->private_id, e->src->global_id, - rta_dest_name(e->attrs->dest)); + log(L_TRACE "%s %c %s %N src %uL %uG %uS id %u %s", + name, dir, msg, e->net, + e->src->private_id, e->src->global_id, e->stale_cycle, e->id, + rta_dest_name(rte_dest(e))); } static inline void -rte_trace_in(uint flag, struct channel *c, rte *e, char *msg) +channel_rte_trace_in(uint flag, struct channel *c, const rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) - rte_trace(c, e, '>', msg); + rte_trace(c->in_req.name, e, '>', msg); } static inline void -rte_trace_out(uint flag, struct channel *c, rte *e, char *msg) +channel_rte_trace_out(uint flag, struct channel *c, const rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) - rte_trace(c, e, '<', msg); + rte_trace(c->out_req.name, e, '<', msg); +} + +static inline void +rt_rte_trace_in(uint flag, struct rt_import_request *req, const rte *e, const char *msg) +{ + if (req->trace_routes & flag) + rte_trace(req->name, e, '>', msg); +} + +#if 0 +// seems to be unused at all +static inline void +rt_rte_trace_out(uint flag, struct rt_export_request *req, const rte *e, const char *msg) +{ + if (req->trace_routes & flag) + rte_trace(req->name, e, '<', msg); +} +#endif + +static uint +rte_feed_count(net *n) +{ + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + count++; + + return count; +} + +static void +rte_feed_obtain(net *n, struct rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + { + ASSERT_DIE(i < count); + feed[i++] = &e->rte; + } + + ASSERT_DIE(i == count); } static rte * -export_filter_(struct channel *c, rte *rt0, rte **rt_free, 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; - struct proto_stats *stats = &c->stats; - rte *rt; - int v; + struct channel_export_stats *stats = &c->export_stats; - rt = rt0; - *rt_free = NULL; + /* Do nothing if we have already rejected the route */ + if (silent && bmap_test(&c->export_reject_map, rt->id)) + goto reject_noset; - v = p->preexport ? p->preexport(c, rt) : 0; + int v = p->preexport ? p->preexport(c, rt) : 0; if (v < 0) { if (silent) - goto reject; + goto reject_noset; - stats->exp_updates_rejected++; + stats->updates_rejected++; if (v == RIC_REJECT) - rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); - goto reject; + channel_rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); + goto reject_noset; + } if (v > 0) { if (!silent) - rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); + channel_rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); goto accept; } v = filter && ((filter == FILTER_REJECT) || - (f_run(filter, &rt, pool, + (f_run(filter, rt, (silent ? FF_SILENT : 0)) > F_ACCEPT)); if (v) { if (silent) goto reject; - stats->exp_updates_filtered++; - rte_trace_out(D_FILTERS, c, rt, "filtered out"); + stats->updates_filtered++; + channel_rte_trace_out(D_FILTERS, c, rt, "filtered out"); goto reject; } accept: - if (rt != rt0) - *rt_free = rt; + /* We have accepted the route */ + bmap_clear(&c->export_reject_map, rt->id); return rt; reject: + /* We have rejected the route by filter */ + bmap_set(&c->export_reject_map, rt->id); + +reject_noset: /* Discard temporary rte */ - if (rt != rt0) - rte_free(rt); return NULL; } -static inline rte * -export_filter(struct channel *c, rte *rt0, rte **rt_free, int silent) -{ - return export_filter_(c, rt0, rt_free, rte_update_pool, silent); -} - static void -do_rt_notify(struct channel *c, net *net, rte *new, rte *old, int refeed) +do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old) { struct proto *p = c->proto; - struct proto_stats *stats = &c->stats; + struct channel_export_stats *stats = &c->export_stats; - if (refeed && new) + if (c->refeeding && new) c->refeed_count++; - /* Apply export limit */ - struct channel_limit *l = &c->out_limit; - if (l->action && !old && new) - { - if (stats->exp_routes >= l->limit) - channel_notify_limit(c, l, PLD_OUT, stats->exp_routes); - - if (l->state == PLS_BLOCKED) + if (!old && new) + if (CHANNEL_LIMIT_PUSH(c, OUT)) { - stats->exp_updates_rejected++; - rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); + stats->updates_rejected++; + channel_rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); return; } - } - /* Apply export table */ - if (c->out_table && !rte_update_out(c, net->n.addr, new, old, refeed)) - return; + if (!new && old) + CHANNEL_LIMIT_POP(c, OUT); if (new) - stats->exp_updates_accepted++; + stats->updates_accepted++; else - stats->exp_withdraws_accepted++; + stats->withdraws_accepted++; if (old) - { bmap_clear(&c->export_map, old->id); - stats->exp_routes--; - } if (new) - { bmap_set(&c->export_map, new->id); - stats->exp_routes++; - } if (p->debug & D_ROUTES) { if (new && old) - rte_trace_out(D_ROUTES, c, new, "replaced"); + channel_rte_trace_out(D_ROUTES, c, new, "replaced"); else if (new) - rte_trace_out(D_ROUTES, c, new, "added"); + channel_rte_trace_out(D_ROUTES, c, new, "added"); else if (old) - rte_trace_out(D_ROUTES, c, old, "removed"); + channel_rte_trace_out(D_ROUTES, c, old, "removed"); } p->rt_notify(p, c, net, new, old); } static void -rt_notify_basic(struct channel *c, net *net, rte *new, rte *old, int refeed) +rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old) { - // struct proto *p = c->proto; - rte *new_free = NULL; - - if (new) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + if (new && old && rte_same(new, old)) + { + if ((new->id != old->id) && bmap_test(&c->export_map, old->id)) + { + bmap_set(&c->export_map, new->id); + bmap_clear(&c->export_map, old->id); + } + return; + } if (new) - new = export_filter(c, new, &new_free, 0); + new = export_filter(c, new, 0); if (old && !bmap_test(&c->export_map, old->id)) old = NULL; @@ -829,197 +895,352 @@ rt_notify_basic(struct channel *c, net *net, rte *new, rte *old, int refeed) if (!new && !old) return; - do_rt_notify(c, net, new, old, refeed); - - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + do_rt_notify(c, net, new, old); } static void -rt_notify_accepted(struct channel *c, net *net, rte *new_changed, rte *old_changed, int refeed) +channel_rpe_mark_seen(struct rt_export_request *req, struct rt_pending_export *rpe) { - // struct proto *p = c->proto; - rte *new_best = NULL; - rte *old_best = NULL; - rte *new_free = 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 - */ + struct channel *c = SKIP_BACK(struct channel, out_req, req); - if (net->routes) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + rpe_mark_seen(req->hook, rpe); + if (rpe->old) + bmap_clear(&c->export_reject_map, rpe->old->rte.id); +} - /* 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 +void +rt_notify_accepted(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *first, + struct rte **feed, uint count) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + + rte nb0, *new_best = NULL; + const rte *old_best = NULL; + + for (uint i = 0; i < count; i++) { - for (rte *r = net->routes; rte_is_valid(r); r = r->next) + if (!rte_is_valid(feed[i])) + continue; + + /* Has been already rejected, won't bother with it */ + if (!c->refeeding && bmap_test(&c->export_reject_map, feed[i]->id)) + continue; + + /* Previously exported */ + if (!old_best && bmap_test(&c->export_map, feed[i]->id)) { - if (bmap_test(&c->export_map, r->id)) + /* is still best */ + if (!new_best) { - old_best = r; - break; + DBG("rt_notify_accepted: idempotent\n"); + goto done; } - /* Note if new_changed found before old_best */ - if (r == new_changed) - new_first = 1; + /* is superseded */ + old_best = feed[i]; + break; } - } - /* Find new_best */ - if ((new_changed == old_changed) || (old_best == old_changed)) - { - /* Feed or old_best changed -> find first accepted by filters */ - for (rte *r = net->routes; rte_is_valid(r); r = r->next) - if (new_best = export_filter(c, r, &new_free, 0)) - break; + /* Have no new best route yet */ + if (!new_best) + { + /* Try this route not seen before */ + nb0 = *feed[i]; + new_best = export_filter(c, &nb0, 0); + DBG("rt_notify_accepted: checking route id %u: %s\n", feed[i]->id, new_best ? "ok" : "no"); + } } - else + +done: + /* Check obsolete routes for previously exported */ + RPE_WALK(first, rpe, NULL) { - /* Other cases -> either new_changed, or old_best (and nothing changed) */ - if (new_first && (new_changed = export_filter(c, new_changed, &new_free, 0))) - new_best = new_changed; - else - return; + channel_rpe_mark_seen(req, rpe); + if (rpe->old) + { + if (bmap_test(&c->export_map, rpe->old->rte.id)) + { + ASSERT_DIE(old_best == NULL); + old_best = &rpe->old->rte; + } + } } - if (!new_best && !old_best) - return; - - do_rt_notify(c, net, new_best, old_best, refeed); - - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); -} - - -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); + /* Nothing to export */ + if (new_best || old_best) + do_rt_notify(c, n, new_best, old_best); + else + DBG("rt_notify_accepted: nothing to export\n"); } rte * -rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int silent) +rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool, int silent) { - // struct proto *p = c->proto; - struct nexthop *nhs = NULL; - rte *best0, *best, *rt0, *rt, *tmp; + _Thread_local static rte rloc; - best0 = net->routes; - *rt_free = NULL; + // struct proto *p = c->proto; + struct nexthop_adata *nhs = NULL; + rte *best0 = feed[0]; + rte *best = NULL; if (!rte_is_valid(best0)) return NULL; - best = export_filter_(c, best0, rt_free, pool, silent); + /* Already rejected, no need to re-run the filter */ + if (!c->refeeding && bmap_test(&c->export_reject_map, best0->id)) + return NULL; + + rloc = *best0; + best = export_filter(c, &rloc, silent); - if (!best || !rte_is_reachable(best)) + if (!best) + /* Best route doesn't pass the filter */ + return NULL; + + if (!rte_is_reachable(best)) + /* Unreachable routes can't be merged */ return best; - for (rt0 = best0->next; rt0; rt0 = rt0->next) + for (uint i = 1; i < count; i++) { - if (!rte_mergable(best0, rt0)) + if (!rte_mergable(best0, feed[i])) continue; - rt = export_filter_(c, rt0, &tmp, pool, 1); + rte tmp0 = *feed[i]; + rte *tmp = export_filter(c, &tmp0, 1); - if (!rt) + if (!tmp || !rte_is_reachable(tmp)) continue; - if (rte_is_reachable(rt)) - nhs = nexthop_merge_rta(nhs, rt->attrs, pool, c->merge_limit); + eattr *nhea = ea_find(tmp->attrs, &ea_gen_nexthop); + ASSERT_DIE(nhea); - if (tmp) - rte_free(tmp); + 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 = rte_cow_rta(best, pool); - nexthop_link(best->attrs, nhs); - } - } + nhs = nexthop_merge(nhs, (struct nexthop_adata *) nhea->u.ptr, c->merge_limit, pool); - if (best != best0) - *rt_free = best; + ea_set_attr(&best->attrs, + EA_LITERAL_DIRECT_ADATA(&ea_gen_nexthop, 0, &nhs->ad)); + } 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) +void +rt_notify_merged(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *first, + struct rte **feed, uint count) { - // struct proto *p = c->proto; - rte *new_free = NULL; - - /* We assume that all rte arguments are either NULL or rte_is_valid() */ + struct channel *c = SKIP_BACK(struct channel, out_req, req); - /* This check should be done by the caller */ - if (!new_best && !old_best) - return; + // struct proto *p = c->proto; +#if 0 /* TODO: Find whether this check is possible when processing multiple changes at once. */ /* 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; +#endif - if (new_best) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + rte *old_best = NULL; + /* Find old best route */ + for (uint i = 0; i < count; i++) + if (bmap_test(&c->export_map, feed[i]->id)) + { + old_best = feed[i]; + break; + } + + /* Check obsolete routes for previously exported */ + RPE_WALK(first, rpe, NULL) + { + channel_rpe_mark_seen(req, rpe); + if (rpe->old) + { + if (bmap_test(&c->export_map, rpe->old->rte.id)) + { + ASSERT_DIE(old_best == NULL); + old_best = &rpe->old->rte; + } + } + } /* Prepare new merged route */ - if (new_best) - new_best = rt_export_merged(c, net, &new_free, rte_update_pool, 0); + rte *new_merged = count ? rt_export_merged(c, feed, count, tmp_linpool, 0) : NULL; + + if (new_merged || old_best) + do_rt_notify(c, n, new_merged, old_best); +} + +void +rt_notify_optimal(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *first) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + rte *o = RTE_VALID_OR_NULL(first->old_best); + struct rte_storage *new_best = first->new_best; - /* Check old merged route */ - if (old_best && !bmap_test(&c->export_map, old_best->id)) - old_best = NULL; + RPE_WALK(first, rpe, NULL) + { + channel_rpe_mark_seen(req, rpe); + new_best = rpe->new_best; + } + + rte n0 = RTE_COPY_VALID(new_best); + if (n0.src || o) + rt_notify_basic(c, net, n0.src ? &n0 : NULL, o); +} - if (!new_best && !old_best) +void +rt_notify_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *first) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + + rte *n = RTE_VALID_OR_NULL(first->new); + rte *o = RTE_VALID_OR_NULL(first->old); + + if (!n && !o) + { + channel_rpe_mark_seen(req, first); return; + } - do_rt_notify(c, net, new_best, old_best, refeed); + struct rte_src *src = n ? n->src : o->src; + struct rte_storage *new_latest = first->new; - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + RPE_WALK(first, rpe, src) + { + channel_rpe_mark_seen(req, rpe); + new_latest = rpe->new; + } + + rte n0 = RTE_COPY_VALID(new_latest); + if (n0.src || o) + rt_notify_basic(c, net, n0.src ? &n0 : NULL, o); } +void +rt_feed_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + + for (uint i=0; i<count; i++) + if (rte_is_valid(feed[i])) + { + rte n0 = *feed[i]; + rt_notify_basic(c, net, &n0, NULL); + } +} + +void +rpe_mark_seen(struct rt_export_hook *hook, struct rt_pending_export *rpe) +{ + bmap_set(&hook->seq_map, rpe->seq); +} + +struct rt_pending_export * +rpe_next(struct rt_pending_export *rpe, struct rte_src *src) +{ + struct rt_pending_export *next = atomic_load_explicit(&rpe->next, memory_order_acquire); + + if (!next) + return NULL; + + if (!src) + return next; + + while (rpe = next) + if (src == (rpe->new ? rpe->new->rte.src : rpe->old->rte.src)) + return rpe; + else + next = atomic_load_explicit(&rpe->next, memory_order_acquire); + + return NULL; +} + +static struct rt_pending_export * rt_next_export_fast(struct rt_pending_export *last); +static int +rte_export(struct rt_table_export_hook *th, struct rt_pending_export *rpe) +{ + rtable *tab = RT_PUB(SKIP_BACK(struct rtable_private, exporter, th->table)); + struct rt_export_hook *hook = &th->h; + if (bmap_test(&hook->seq_map, rpe->seq)) + goto ignore; /* Seen already */ + + const net_addr *n = rpe->new_best ? rpe->new_best->rte.net : rpe->old_best->rte.net; + + switch (hook->req->addr_mode) + { + case TE_ADDR_NONE: + break; + + case TE_ADDR_IN: + if (!net_in_netX(n, hook->req->addr)) + goto ignore; + break; + + case TE_ADDR_EQUAL: + if (!net_equal(n, hook->req->addr)) + goto ignore; + break; + + case TE_ADDR_FOR: + bug("Continuos export of best prefix match not implemented yet."); + + default: + bug("Strange table export address mode: %d", hook->req->addr_mode); + } + + if (rpe->new) + hook->stats.updates_received++; + else + hook->stats.withdraws_received++; + + if (hook->req->export_one) + hook->req->export_one(hook->req, n, rpe); + else if (hook->req->export_bulk) + { + net *net = SKIP_BACK(struct network, n.addr, (net_addr (*)[0]) n); + RT_LOCK(tab); + uint count = rte_feed_count(net); + rte **feed = NULL; + if (count) + { + feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(net, feed, count); + } + RT_UNLOCK(tab); + hook->req->export_bulk(hook->req, n, rpe, feed, count); + } + else + bug("Export request must always provide an export method"); + +ignore: + /* Get the next export if exists */ + th->rpe_next = rt_next_export_fast(rpe); + + /* The last block may be available to free */ + int used = (PAGE_HEAD(th->rpe_next) != PAGE_HEAD(rpe)); + + /* Releasing this export for cleanup routine */ + DBG("store hook=%p last_export=%p seq=%lu\n", hook, rpe, rpe->seq); + atomic_store_explicit(&th->last_export, rpe, memory_order_release); + + return used; +} /** * 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 @@ -1035,13 +1256,6 @@ rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed * 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. @@ -1056,139 +1270,310 @@ rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed * done outside of scope of rte_announce(). */ static void -rte_announce(rtable *tab, uint type, net *net, rte *new, rte *old, - rte *new_best, rte *old_best) +rte_announce(struct rtable_private *tab, net *net, struct rte_storage *new, struct rte_storage *old, + struct rte_storage *new_best, struct rte_storage *old_best) { - if (!rte_is_valid(new)) - new = NULL; + int new_best_valid = rte_is_valid(RTE_OR_NULL(new_best)); + int old_best_valid = rte_is_valid(RTE_OR_NULL(old_best)); - if (!rte_is_valid(old)) - old = NULL; + if ((new == old) && (new_best == old_best)) + return; - if (!rte_is_valid(new_best)) - new_best = NULL; + if (new_best_valid) + new_best->rte.sender->stats.pref++; + if (old_best_valid) + old_best->rte.sender->stats.pref--; - if (!rte_is_valid(old_best)) - old_best = NULL; + if (EMPTY_LIST(tab->exporter.e.hooks) && EMPTY_LIST(tab->exporter.pending)) + { + /* No export hook and no pending exports to cleanup. We may free the route immediately. */ + if (!old) + return; - if (!new && !old && !new_best && !old_best) + hmap_clear(&tab->id_map, old->rte.id); + rte_free(old); return; + } + + /* Get the pending export structure */ + struct rt_export_block *rpeb = NULL, *rpebsnl = NULL; + u32 end = 0; - if (new_best != old_best) + if (!EMPTY_LIST(tab->exporter.pending)) { - if (new_best) - new_best->sender->stats.pref_routes++; - if (old_best) - old_best->sender->stats.pref_routes--; + rpeb = TAIL(tab->exporter.pending); + end = atomic_load_explicit(&rpeb->end, memory_order_relaxed); + if (end >= RT_PENDING_EXPORT_ITEMS) + { + ASSERT_DIE(end == RT_PENDING_EXPORT_ITEMS); + rpebsnl = rpeb; - if (tab->hostcache) - rt_notify_hostcache(tab, net); + rpeb = NULL; + end = 0; + } + } - if (!EMPTY_LIST(tab->flowspec_links)) - rt_flowspec_notify(tab, net); + if (!rpeb) + { + rpeb = alloc_page(); + *rpeb = (struct rt_export_block) {}; + add_tail(&tab->exporter.pending, &rpeb->n); } - rt_schedule_notify(tab); + /* Fill the pending export */ + struct rt_pending_export *rpe = &rpeb->export[rpeb->end]; + *rpe = (struct rt_pending_export) { + .new = new, + .new_best = new_best, + .old = old, + .old_best = old_best, + .seq = tab->exporter.next_seq++, + }; + + DBGL("rte_announce: table=%s net=%N new=%p id %u from %s old=%p id %u from %s new_best=%p id %u old_best=%p id %u seq=%lu", + tab->name, net->n.addr, + new, new ? new->rte.id : 0, new ? new->rte.sender->req->name : NULL, + old, old ? old->rte.id : 0, old ? old->rte.sender->req->name : NULL, + new_best, old_best, rpe->seq); + + ASSERT_DIE(atomic_fetch_add_explicit(&rpeb->end, 1, memory_order_release) == end); - struct channel *c; node *n; - WALK_LIST2(c, n, tab->channels, table_node) + if (rpebsnl) { - if (c->export_state == ES_DOWN) - continue; + _Bool f = 0; + ASSERT_DIE(atomic_compare_exchange_strong_explicit(&rpebsnl->not_last, &f, 1, + memory_order_release, memory_order_relaxed)); + } - if (type && (type != c->ra_mode)) - continue; + /* Append to the same-network squasher list */ + if (net->last) + { + struct rt_pending_export *rpenull = NULL; + ASSERT_DIE(atomic_compare_exchange_strong_explicit( + &net->last->next, &rpenull, rpe, + memory_order_relaxed, + memory_order_relaxed)); + + } + + net->last = rpe; + + if (!net->first) + net->first = rpe; + + if (tab->exporter.first == NULL) + tab->exporter.first = rpe; + + rt_check_cork_high(tab); +} + +static struct rt_pending_export * +rt_next_export_fast(struct rt_pending_export *last) +{ + /* Get the whole export block and find our position in there. */ + struct rt_export_block *rpeb = PAGE_HEAD(last); + u32 pos = (last - &rpeb->export[0]); + u32 end = atomic_load_explicit(&rpeb->end, memory_order_acquire); + ASSERT_DIE(pos < end); + + /* Next is in the same block. */ + if (++pos < end) + return &rpeb->export[pos]; + + /* There is another block. */ + if (atomic_load_explicit(&rpeb->not_last, memory_order_acquire)) + { + /* This is OK to do non-atomically because of the not_last flag. */ + rpeb = NODE_NEXT(rpeb); + return &rpeb->export[0]; + } + + /* There is nothing more. */ + return NULL; +} + +static struct rt_pending_export * +rt_next_export(struct rt_table_export_hook *hook, struct rt_table_exporter *tab) +{ + ASSERT_DIE(RT_IS_LOCKED(SKIP_BACK(struct rtable_private, exporter, tab))); + + /* As the table is locked, it is safe to reload the last export pointer */ + struct rt_pending_export *last = atomic_load_explicit(&hook->last_export, memory_order_acquire); + + /* It is still valid, let's reuse it */ + if (last) + return rt_next_export_fast(last); + + /* No, therefore we must process the table's first pending export */ + else + return tab->first; +} - switch (c->ra_mode) +static inline void +rt_send_export_event(struct rt_export_hook *hook) +{ + ev_send(hook->req->list, &hook->event); +} + +static void +rt_announce_exports(timer *tm) +{ + RT_LOCKED((rtable *) tm->data, tab) + if (!EMPTY_LIST(tab->exporter.pending)) { - case RA_OPTIMAL: - if (new_best != old_best) - rt_notify_basic(c, net, new_best, old_best, 0); - break; + struct rt_export_hook *c; node *n; + WALK_LIST2(c, n, tab->exporter.e.hooks, n) + { + if (atomic_load_explicit(&c->export_state, memory_order_acquire) != TES_READY) + continue; - case RA_ANY: - if (new != old) - rt_notify_basic(c, net, new, old, 0); - break; + rt_send_export_event(c); + } + } +} - case RA_ACCEPTED: - rt_notify_accepted(c, net, new, old, 0); - break; +static void +rt_import_announce_exports(void *_hook) +{ + struct rt_import_hook *hook = _hook; + RT_LOCKED(hook->table, tab) + { + if (hook->import_state == TIS_CLEARED) + { + void (*stopped)(struct rt_import_request *) = hook->stopped; + struct rt_import_request *req = hook->req; + req->hook = NULL; - case RA_MERGED: - rt_notify_merged(c, net, new, old, new_best, old_best, 0); - break; + rt_trace(tab, D_EVENTS, "Hook %s stopped", req->name); + rem_node(&hook->n); + mb_free(hook); + rt_unlock_table(tab); + RT_UNLOCK(tab); + + stopped(req); + return; + } + + rt_trace(tab, D_EVENTS, "Announcing exports after imports from %s", hook->req->name); + + if (!tm_active(tab->exporter.export_timer)) + tm_start(tab->exporter.export_timer, tab->config->export_settle_time); + } +} + +static struct rt_pending_export * +rt_last_export(struct rt_table_exporter *tab) +{ + struct rt_pending_export *rpe = NULL; + + if (!EMPTY_LIST(tab->pending)) + { + /* We'll continue processing exports from this export on */ + struct rt_export_block *reb = TAIL(tab->pending); + ASSERT_DIE(reb->end); + rpe = &reb->export[reb->end - 1]; + } + + return rpe; +} + +#define RT_EXPORT_BULK 1024 + +static void +rt_export_hook(void *_data) +{ + struct rt_table_export_hook *c = _data; + rtable *tab = SKIP_BACK(rtable, priv.exporter, c->table); + + ASSERT_DIE(atomic_load_explicit(&c->h.export_state, memory_order_relaxed) == TES_READY); + + if (!c->rpe_next) + { + RT_LOCK(tab); + c->rpe_next = rt_next_export(c, c->table); + + if (!c->rpe_next) + { + rt_export_used(c->table); + RT_UNLOCK(tab); + return; } + + RT_UNLOCK(tab); } + + int used = 0; + + /* Process the export */ + for (uint i=0; i<RT_EXPORT_BULK; i++) + { + used += rte_export(c, c->rpe_next); + + if (!c->rpe_next) + break; + } + + if (used) + RT_LOCKED(tab, _) + rt_export_used(c->table); + + rt_send_export_event(&c->h); } + static inline int -rte_validate(rte *e) +rte_validate(struct channel *ch, rte *e) { int c; - net *n = e->net; + const net_addr *n = e->net; - if (!net_validate(n->n.addr)) + if (!net_validate(n)) { log(L_WARN "Ignoring bogus prefix %N received via %s", - n->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } /* FIXME: better handling different nettypes */ - c = !net_is_flow(n->n.addr) ? - net_classify(n->n.addr): (IADDR_HOST | SCOPE_UNIVERSE); + 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->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } - if (net_type_match(n->n.addr, NB_DEST) == !e->attrs->dest) + if (net_type_match(n, NB_DEST)) { - /* Exception for flowspec that failed validation */ - if (net_is_flow(n->n.addr) && (e->attrs->dest == RTD_UNREACHABLE)) - return 1; + eattr *nhea = ea_find(e->attrs, &ea_gen_nexthop); + int dest = nhea_dest(nhea); - log(L_WARN "Ignoring route %N with invalid dest %d received via %s", - n->n.addr, e->attrs->dest, e->sender->proto->name); - return 0; - } + 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", - n->n.addr, e->sender->proto->name); + log(L_WARN "Ignoring route %N having a nexthop attribute received via %s", + n, ch->proto->name); return 0; } return 1; } -/** - * rte_free - delete a &rte - * @e: &rte to be deleted - * - * rte_free() deletes the given &rte from the routing table it's linked to. - */ -void -rte_free(rte *e) -{ - rt_unlock_source(e->src); - if (rta_is_cached(e->attrs)) - rta_free(e->attrs); - sl_free(e); -} - -static inline void -rte_free_quick(rte *e) -{ - rt_unlock_source(e->src); - rta_free(e->attrs); - sl_free(e); -} - static int rte_same(rte *x, rte *y) { @@ -1202,168 +1587,109 @@ rte_same(rte *x, rte *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) +static int +rte_recalculate(struct rtable_private *table, struct rt_import_hook *c, net *net, rte *new, struct rte_src *src) { - struct proto *p = c->proto; - struct rtable *table = c->table; - struct proto_stats *stats = &c->stats; - static struct tbf rl_pipe = TBF_DEFAULT_LOG_LIMITS; - rte *before_old = NULL; - rte *old_best = net->routes; + struct rt_import_request *req = c->req; + struct rt_import_stats *stats = &c->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; - rte **k; - k = &net->routes; /* Find and remove original route from the same protocol */ - while (old = *k) + /* 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); + + if (*before_old) { - if (old->src == src) + 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 != c) { - /* 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 it is withdraw, we - * ignore it completely (there might be 'spurious withdraws', - * see FIXME in do_rte_announce()) - */ - if (old->sender->proto != p) - { - if (new) - { - log_rl(&rl_pipe, L_ERR "Pipe collision detected when sending %N to table %s", - net->n.addr, table->name); - rte_free_quick(new); - } - return; - } + 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->owner->name, old->src->private_id, old->src->global_id); - if (new && rte_same(old, new)) + 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->owner->name, old->src->private_id, old->src->global_id); + } + + 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->imp_updates_ignored++; - rte_trace_in(D_ROUTES, c, new, "ignored"); + stats->updates_ignored++; + rt_rte_trace_in(D_ROUTES, req, new, "ignored"); } - rte_free_quick(new); - return; - } - *k = old->next; - table->rt_count--; - break; - } - k = &old->next; - before_old = old; - } - - /* Save the last accessed position */ - rte **pos = k; + /* We need to free the already stored route here before returning */ + rte_free(new_stored); + return 0; + } - if (!old) - before_old = NULL; + *before_old = (*before_old)->next; + table->rt_count--; + } if (!old && !new) { - stats->imp_withdraws_ignored++; - return; + stats->withdraws_ignored++; + return 0; } + /* 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); - struct channel_limit *l = &c->rx_limit; - if (l->action && !old && new && !c->in_table) - { - u32 all_routes = stats->imp_routes + stats->filt_routes; - - 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->imp_updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - rte_free_quick(new); - return; - } - } - - l = &c->in_limit; - if (l->action && !old_ok && new_ok) - { - if (stats->imp_routes >= l->limit) - channel_notify_limit(c, l, PLD_IN, stats->imp_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->imp_updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - - if (c->in_keep_filtered) - new->flags |= REF_FILTERED; - else - { rte_free_quick(new); 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->imp_updates_accepted++; + stats->updates_accepted++; else if (old_ok) - stats->imp_withdraws_accepted++; + stats->withdraws_accepted++; else - stats->imp_withdraws_ignored++; + stats->withdraws_ignored++; if (old_ok || new_ok) table->last_rt_change = current_time(); - skip_stats1: - - if (new) - rte_is_filtered(new) ? stats->filt_routes++ : stats->imp_routes++; - if (old) - rte_is_filtered(old) ? stats->filt_routes-- : stats->imp_routes--; - if (table->config->sorted) { /* If routes are sorted, just insert new route to appropriate position */ - if (new) + if (new_stored) { - if (before_old && !rte_better(new, before_old)) - k = &before_old->next; + 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)) + if (rte_better(new, &(*k)->rte)) break; - new->next = *k; - *k = new; + new_stored->next = *k; + *k = new_stored; table->rt_count++; } @@ -1373,16 +1699,17 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) /* 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, old, old_best)) + if (src->owner->rte_recalculate && + src->owner->rte_recalculate(table, net, new_stored ? &new_stored->rte : NULL, old, old_best)) goto do_recalculate; - if (new && rte_better(new, old_best)) + 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->next = net->routes; - net->routes = new; + new_stored->next = net->routes; + net->routes = new_stored; table->rt_count++; } @@ -1396,10 +1723,10 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) do_recalculate: /* Add the new route to the list */ - if (new) + if (new_stored) { - new->next = *pos; - *pos = new; + new_stored->next = *before_old; + *before_old = new_stored; table->rt_count++; } @@ -1407,298 +1734,465 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) /* Find a new optimal route (if there is any) */ if (net->routes) { - rte **bp = &net->routes; - for (k=&(*bp)->next; *k; k=&(*k)->next) - if (rte_better(*k, *bp)) + 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 */ - rte *best = *bp; + struct rte_storage *best = *bp; *bp = best->next; best->next = net->routes; net->routes = best; } } - else if (new) + 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->next = *pos; - *pos = new; + 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) + if (new_stored) { - new->lastmod = current_time(); - - if (!old) - { - new->id = hmap_first_zero(&table->id_map); - hmap_set(&table->id_map, new->id); - } - else - new->id = old->id; + new_stored->rte.lastmod = current_time(); + new_stored->rte.id = hmap_first_zero(&table->id_map); + hmap_set(&table->id_map, new_stored->rte.id); } /* Log the route change */ - if ((c->debug & D_ROUTES) || (p->debug & D_ROUTES)) + if (new_ok) + rt_rte_trace_in(D_ROUTES, req, &new_stored->rte, new_stored == net->routes ? "added [best]" : "added"); + else if (old_ok) { - if (new_ok) - rte_trace(c, new, '>', new == net->routes ? "added [best]" : "added"); - else if (old_ok) - { - if (old != old_best) - rte_trace(c, old, '>', "removed"); - else if (rte_is_ok(net->routes)) - rte_trace(c, old, '>', "removed [replaced]"); - else - rte_trace(c, old, '>', "removed [sole]"); - } + if (old != old_best) + rt_rte_trace_in(D_ROUTES, req, old, "removed"); + else if (net->routes && rte_is_ok(&net->routes->rte)) + rt_rte_trace_in(D_ROUTES, req, old, "removed [replaced]"); + else + rt_rte_trace_in(D_ROUTES, req, old, "removed [sole]"); } /* Propagate the route change */ - rte_announce(table, RA_UNDEF, net, new, old, net->routes, old_best); + rte_announce(table, net, new_stored, old_stored, + net->routes, old_best_stored); - if (!net->routes && - (table->gc_counter++ >= table->config->gc_threshold)) - rt_kick_prune_timer(table); + return 1; +} - if (old_ok && p->rte_remove) - p->rte_remove(net, old); - if (new_ok && p->rte_insert) - p->rte_insert(net, new); +int +channel_preimport(struct rt_import_request *req, rte *new, rte *old) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); - if (old) - { - if (!new) - hmap_clear(&table->id_map, old->id); + if (new && !old) + if (CHANNEL_LIMIT_PUSH(c, RX)) + return 0; - rte_free_quick(old); - } -} + if (!new && old) + CHANNEL_LIMIT_POP(c, RX); -static int rte_update_nest_cnt; /* Nesting counter to allow recursive updates */ + int new_in = new && !rte_is_filtered(new); + int old_in = old && !rte_is_filtered(old); -static inline void -rte_update_lock(void) -{ - rte_update_nest_cnt++; -} + if (new_in && !old_in) + if (CHANNEL_LIMIT_PUSH(c, IN)) + if (c->in_keep & RIK_REJECTED) + { + new->flags |= REF_FILTERED; + return 1; + } + else + return 0; -static inline void -rte_update_unlock(void) -{ - if (!--rte_update_nest_cnt) - lp_flush(rte_update_pool); -} + if (!new_in && old_in) + CHANNEL_LIMIT_POP(c, IN); -/** - * rte_update - enter a new update to a routing table - * @table: table to be updated - * @c: channel doing the update - * @net: network node - * @p: protocol submitting the update - * @src: protocol originating the update - * @new: a &rte representing the new route or %NULL for route removal. - * - * This function is called by the routing protocols whenever they discover - * a new route or wish to update/remove an existing route. The right announcement - * sequence is to build route attributes first (either un-cached with @aflags set - * to zero or a cached one using rta_lookup(); in this case please note that - * you need to increase the use count of the attributes yourself by calling - * rta_clone()), call rte_get_temp() to obtain a temporary &rte, fill in all - * the appropriate data and finally submit the new &rte by calling rte_update(). - * - * @src specifies the protocol that originally created the route and the meaning - * of protocol-dependent data of @new. If @new is not %NULL, @src have to be the - * same value as @new->attrs->proto. @p specifies the protocol that called - * rte_update(). In most cases it is the same protocol as @src. rte_update() - * stores @p in @new->sender; - * - * When rte_update() gets any route, it automatically validates it (checks, - * whether the network and next hop address are valid IP addresses and also - * whether a normal routing protocol doesn't try to smuggle a host or link - * scope route to the table), converts all protocol dependent attributes stored - * in the &rte to temporary extended attributes, consults import filters of the - * protocol to see if the route should be accepted and/or its attributes modified, - * stores the temporary attributes back to the &rte. - * - * Now, having a "public" version of the route, we - * automatically find any old route defined by the protocol @src - * for network @n, replace it by the new one (or removing it if @new is %NULL), - * recalculate the optimal route for this destination and finally broadcast - * the change (if any) to all routing protocols by calling rte_announce(). - * - * All memory used for attribute lists and other temporary allocations is taken - * from a special linear pool @rte_update_pool and freed when rte_update() - * finishes. - */ + return 1; +} void -rte_update2(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { - // struct proto *p = c->proto; - struct proto_stats *stats = &c->stats; - const struct filter *filter = c->in_filter; - net *nn; + if (!c->in_req.hook) + return; ASSERT(c->channel_state == CS_UP); - rte_update_lock(); + /* 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); + + const struct filter *filter = c->in_filter; + struct channel_import_stats *stats = &c->import_stats; + if (new) { - /* Create a temporary table node */ - nn = alloca(sizeof(net) + n->length); - memset(nn, 0, sizeof(net) + n->length); - net_copy(nn->n.addr, n); + new->net = n; - new->net = nn; - new->sender = c; + int fr; - stats->imp_updates_received++; - if (!rte_validate(new)) + stats->updates_received++; + if ((filter == FILTER_REJECT) || + ((fr = f_run(filter, new, 0)) > F_ACCEPT)) { - rte_trace_in(D_FILTERS, c, new, "invalid"); - stats->imp_updates_invalid++; - goto drop; + stats->updates_filtered++; + channel_rte_trace_in(D_FILTERS, c, new, "filtered out"); + + if (c->in_keep & RIK_REJECTED) + new->flags |= REF_FILTERED; + else + new = NULL; } - if (filter == FILTER_REJECT) + 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)) { - stats->imp_updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); + channel_rte_trace_in(D_FILTERS, c, new, "invalid"); + stats->updates_invalid++; + new = NULL; + } - if (! c->in_keep_filtered) - goto drop; + } + else + stats->withdraws_received++; - /* 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->imp_updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); + rte_import(&c->in_req, n, new, src); - if (! c->in_keep_filtered) - goto drop; + /* 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; - new->flags |= REF_FILTERED; - } - } - if (!rta_is_cached(new->attrs)) /* Need to copy attributes */ - new->attrs = rta_lookup(new->attrs); - new->flags |= REF_COW; + ea_free(a); + } + +} + +void +rte_import(struct rt_import_request *req, const net_addr *n, rte *new, struct rte_src *src) +{ + struct rt_import_hook *hook = req->hook; + if (!hook) + return; + RT_LOCKED(hook->table, tab) + { + net *nn; + if (new) + { /* Use the actual struct network, not the dummy one */ - nn = net_get(c->table, n); - new->net = nn; + nn = net_get(tab, n); + new->net = nn->n.addr; + new->sender = hook; + + /* Set the stale cycle */ + new->stale_cycle = hook->stale_set; } - else + else if (!(nn = net_find(tab, n))) { - stats->imp_withdraws_received++; - - if (!(nn = net_find(c->table, n)) || !src) - { - stats->imp_withdraws_ignored++; - rte_update_unlock(); - return; - } + req->hook->stats.withdraws_ignored++; + RT_RETURN(tab); } - recalc: - /* And recalculate the best route */ - rte_recalculate(c, nn, new, src); + /* Recalculate the best route */ + if (rte_recalculate(tab, hook, nn, new, src)) + ev_send(req->list, &hook->announce_event); + } +} - rte_update_unlock(); - return; +/* Check rtable for best route to given net whether it would be exported do p */ +int +rt_examine(rtable *tp, net_addr *a, struct channel *c, const struct filter *filter) +{ + rte rt = {}; - drop: - rte_free(new); - new = NULL; - if (nn = net_find(c->table, n)) - goto recalc; + RT_LOCKED(tp, t) + { + net *n = net_find(t, a); + if (n) + rt = RTE_COPY_VALID(n->routes); + } + + if (!rt.src) + return 0; + + int v = c->proto->preexport ? c->proto->preexport(c, &rt) : 0; + if (v == RIC_PROCESS) + v = (f_run(filter, &rt, FF_SILENT) <= F_ACCEPT); - rte_update_unlock(); + return v > 0; } -/* 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, rte *new, rte *old, - rte *new_best, rte *old_best) +static void +rt_table_export_done(void *hh) { - rte_update_lock(); - rte_announce(tab, type, net, new, old, new_best, old_best); - rte_update_unlock(); + struct rt_table_export_hook *hook = hh; + struct rt_export_request *req = hook->h.req; + void (*stopped)(struct rt_export_request *) = hook->h.stopped; + rtable *t = SKIP_BACK(rtable, priv.exporter, hook->table); + + RT_LOCKED(t, tab) + { + DBG("Export hook %p in table %s finished uc=%u\n", hook, tab->name, tab->use_count); + + /* Drop pending exports */ + rt_export_used(&tab->exporter); + + /* Do the common code; this frees the hook */ + rt_export_stopped(&hook->h); + } + + /* Report the channel as stopped. */ + CALL(stopped, req); + + /* Unlock the table; this may free it */ + rt_unlock_table(t); } -static inline void -rte_discard(rte *old) /* Non-filtered route deletion, used during garbage collection */ +void +rt_export_stopped(struct rt_export_hook *hook) { - rte_update_lock(); - rte_recalculate(old->sender, old->net, NULL, old->src); - rte_update_unlock(); + /* Unlink from the request */ + hook->req->hook = NULL; + + /* Unlist */ + rem_node(&hook->n); + + /* Free the hook itself together with its pool */ + rfree(hook->pool); } -/* Modify existing route by protocol hook, used for long-lived graceful restart */ static inline void -rte_modify(rte *old) +rt_set_import_state(struct rt_import_hook *hook, u8 state) { - rte_update_lock(); + hook->last_state_change = current_time(); + hook->import_state = state; - rte *new = old->sender->proto->rte_modify(old, rte_update_pool); - if (new != old) + CALL(hook->req->log_state_change, hook->req, state); +} + +void +rt_set_export_state(struct rt_export_hook *hook, u8 state) +{ + hook->last_state_change = current_time(); + atomic_store_explicit(&hook->export_state, state, memory_order_release); + + CALL(hook->req->log_state_change, hook->req, state); +} + +void +rt_request_import(rtable *t, struct rt_import_request *req) +{ + RT_LOCKED(t, tab) { - if (new) - { - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - new->flags = (old->flags & ~REF_MODIFY) | REF_COW; - } + rt_lock_table(tab); + + struct rt_import_hook *hook = req->hook = mb_allocz(tab->rp, sizeof(struct rt_import_hook)); + + hook->announce_event = (event) { .hook = rt_import_announce_exports, .data = hook }; + + DBG("Lock table %s for import %p req=%p uc=%u\n", tab->name, hook, req, tab->use_count); - rte_recalculate(old->sender, old->net, new, old->src); + hook->req = req; + hook->table = t; + + rt_set_import_state(hook, TIS_UP); + add_tail(&tab->imports, &hook->n); } +} + +void +rt_stop_import(struct rt_import_request *req, void (*stopped)(struct rt_import_request *)) +{ + ASSERT_DIE(req->hook); + struct rt_import_hook *hook = req->hook; - rte_update_unlock(); + RT_LOCKED(hook->table, tab) + { + rt_schedule_prune(tab); + rt_set_import_state(hook, TIS_STOP); + hook->stopped = stopped; + } } -/* 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) +static void +rt_table_export_start_locked(struct rtable_private *tab, struct rt_export_request *req) { - struct proto *p = c->proto; - net *n = net_find(t, a); - rte *rt = n ? n->routes : NULL; + struct rt_exporter *re = &tab->exporter.e; + rt_lock_table(tab); - if (!rte_is_valid(rt)) - return 0; + req->hook = rt_alloc_export(re, sizeof(struct rt_table_export_hook)); + req->hook->req = req; - rte_update_lock(); + struct rt_table_export_hook *hook = SKIP_BACK(struct rt_table_export_hook, h, req->hook); - /* Rest is stripped down export_filter() */ - int v = p->preexport ? p->preexport(c, rt) : 0; - if (v == RIC_PROCESS) - v = (f_run(filter, &rt, rte_update_pool, FF_SILENT) <= F_ACCEPT); + /* 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(hook->h.pool, sizeof (struct f_trie_walk_state)); + hook->walk_lock = rt_lock_trie(tab); + trie_walk_init(hook->walk_state, tab->trie, req->addr); + hook->h.event.hook = rt_feed_by_trie; + hook->walk_last.type = 0; + break; + } + /* fall through */ + case TE_ADDR_NONE: + FIB_ITERATE_INIT(&hook->feed_fit, &tab->fib); + hook->h.event.hook = rt_feed_by_fib; + break; - /* Discard temporary rte */ - if (rt != n->routes) - rte_free(rt); + case TE_ADDR_EQUAL: + hook->h.event.hook = rt_feed_equal; + break; - rte_update_unlock(); + case TE_ADDR_FOR: + hook->h.event.hook = rt_feed_for; + break; - return v > 0; + 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); + + struct rt_pending_export *rpe = rt_last_export(hook->table); + DBG("store hook=%p last_export=%p seq=%lu\n", hook, rpe, rpe ? rpe->seq : 0); + atomic_store_explicit(&hook->last_export, rpe, memory_order_relaxed); + + rt_init_export(re, req->hook); } +static void +rt_table_export_start(struct rt_exporter *re, struct rt_export_request *req) +{ + RT_LOCKED(SKIP_BACK(rtable, priv.exporter.e, re), tab) + rt_table_export_start_locked(tab, req); +} + +void rt_request_export(rtable *t, struct rt_export_request *req) +{ + RT_LOCKED(t, tab) + rt_table_export_start_locked(tab, req); /* Is locked inside */ +} + +void +rt_request_export_other(struct rt_exporter *re, struct rt_export_request *req) +{ + return re->class->start(re, req); +} + +struct rt_export_hook * +rt_alloc_export(struct rt_exporter *re, uint size) +{ + pool *p = rp_new(re->rp, "Export hook"); + struct rt_export_hook *hook = mb_allocz(p, size); + + hook->pool = p; + hook->table = re; + + return hook; +} + +void +rt_init_export(struct rt_exporter *re, struct rt_export_hook *hook) +{ + hook->event.data = hook; + + bmap_init(&hook->seq_map, hook->pool, 1024); + + hook->n = (node) {}; + add_tail(&re->hooks, &hook->n); + + /* Regular export */ + rt_set_export_state(hook, TES_FEEDING); + rt_send_export_event(hook); +} + +static void +rt_table_export_stop_locked(struct rt_export_hook *hh) +{ + struct rt_table_export_hook *hook = SKIP_BACK(struct rt_table_export_hook, h, hh); + struct rtable_private *tab = SKIP_BACK(struct rtable_private, exporter, hook->table); + + if (atomic_load_explicit(&hh->export_state, memory_order_relaxed) == TES_FEEDING) + switch (hh->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; + } +} + +static void +rt_table_export_stop(struct rt_export_hook *hh) +{ + struct rt_table_export_hook *hook = SKIP_BACK(struct rt_table_export_hook, h, hh); + rtable *t = SKIP_BACK(rtable, priv.exporter, hook->table); + if (RT_IS_LOCKED(t)) + rt_table_export_stop_locked(hh); + else + RT_LOCKED(t, tab) + rt_table_export_stop_locked(hh); +} + +void +rt_stop_export(struct rt_export_request *req, void (*stopped)(struct rt_export_request *)) +{ + ASSERT_DIE(birdloop_inside(req->list->loop)); + ASSERT_DIE(req->hook); + struct rt_export_hook *hook = req->hook; + + /* Cancel the feeder event */ + ev_postpone(&hook->event); + + /* Stop feeding from the exporter */ + CALL(hook->table->class->stop, hook); + + /* Reset the event as the stopped event */ + hook->event.hook = hook->table->class->done; + hook->stopped = stopped; + + /* Update export state */ + rt_set_export_state(hook, TES_STOP); + + /* Run the stopped event */ + rt_send_export_event(hook); +} /** * rt_refresh_begin - start a refresh cycle @@ -1715,16 +2209,43 @@ rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filte * flag in rt_refresh_end() and then removing such routes in the prune loop. */ void -rt_refresh_begin(rtable *t, struct channel *c) +rt_refresh_begin(struct rt_import_request *req) { - FIB_WALK(&t->fib, net, n) - { - rte *e; - for (e = n->routes; e; e = e->next) - if (e->sender == c) - e->flags |= REF_STALE; - } - FIB_WALK_END; + struct rt_import_hook *hook = req->hook; + ASSERT_DIE(hook); + ASSERT_DIE(hook->stale_set == hook->stale_valid); + + RT_LOCKED(hook->table, tab) + { + + /* 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(&tab->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); + + } } /** @@ -1736,45 +2257,21 @@ rt_refresh_begin(rtable *t, struct channel *c) * hook. See rt_refresh_begin() for description of refresh cycles. */ void -rt_refresh_end(rtable *t, struct channel *c) +rt_refresh_end(struct rt_import_request *req) { - int prune = 0; - - FIB_WALK(&t->fib, net, n) - { - rte *e; - for (e = n->routes; e; e = e->next) - if ((e->sender == c) && (e->flags & REF_STALE)) - { - e->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 channel *c) -{ - int prune = 0; + RT_LOCKED(hook->table, tab) + { + hook->stale_valid++; + ASSERT_DIE(hook->stale_set == hook->stale_valid); - FIB_WALK(&t->fib, net, n) - { - rte *e; - for (e = n->routes; e; e = e->next) - if ((e->sender == c) && (e->flags & REF_STALE) && !(e->flags & REF_FILTERED)) - { - e->flags |= REF_MODIFY; - prune = 1; - } - } - FIB_WALK_END; + rt_schedule_prune(tab); - 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); + } } /** @@ -1784,12 +2281,11 @@ rt_modify_stale(rtable *t, struct channel *c) * This functions dumps contents of a &rte to debug output. */ void -rte_dump(rte *e) +rte_dump(struct rte_storage *e) { - net *n = e->net; - debug("%-1N ", n->n.addr); - debug("PF=%02x ", e->pflags); - rta_dump(e->attrs); + debug("%-1N ", e->rte.net); + debug("PF=%02x ", e->rte.pflags); + ea_dump(e->rte.attrs); debug("\n"); } @@ -1800,20 +2296,24 @@ rte_dump(rte *e) * This function dumps contents of a given routing table to debug output. */ void -rt_dump(rtable *t) +rt_dump(rtable *tp) { - debug("Dump of routing table <%s>\n", t->name); + RT_LOCKED(tp, t) + { + + debug("Dump of routing table <%s>%s\n", t->name, t->deleted ? " (deleted)" : ""); #ifdef DEBUGGING fib_check(&t->fib); #endif FIB_WALK(&t->fib, net, n) { - rte *e; - for(e=n->routes; e; e=e->next) + for(struct rte_storage *e=n->routes; e; e=e->next) rte_dump(e); } FIB_WALK_END; debug("\n"); + + } } /** @@ -1829,33 +2329,89 @@ rt_dump_all(void) WALK_LIST2(t, n, routing_tables, n) rt_dump(t); + + WALK_LIST2(t, n, deleted_routing_tables, n) + rt_dump(t); } -static inline void -rt_schedule_hcu(rtable *tab) +void +rt_dump_hooks(rtable *tp) { - if (tab->hcu_scheduled) - return; + RT_LOCKED(tp, tab) + { - tab->hcu_scheduled = 1; - ev_schedule(tab->rt_event); + debug("Dump of hooks in routing table <%s>%s\n", tab->name, tab->deleted ? " (deleted)" : ""); + debug(" nhu_state=%u use_count=%d rt_count=%u\n", + tab->nhu_state, tab->use_count, tab->rt_count); + debug(" last_rt_change=%t gc_time=%t gc_counter=%d prune_state=%u\n", + tab->last_rt_change, tab->gc_time, tab->gc_counter, tab->prune_state); + + struct rt_import_hook *ih; + WALK_LIST(ih, tab->imports) + { + ih->req->dump_req(ih->req); + debug(" Import hook %p requested by %p: pref=%u" + " last_state_change=%t import_state=%u stopped=%p\n", + ih, ih->req, ih->stats.pref, + ih->last_state_change, ih->import_state, ih->stopped); + } + + struct rt_table_export_hook *eh; + WALK_LIST(eh, tab->exporter.e.hooks) + { + eh->h.req->dump_req(eh->h.req); + debug(" Export hook %p requested by %p:" + " refeed_pending=%u last_state_change=%t export_state=%u\n", + eh, eh->h.req, eh->refeed_pending, eh->h.last_state_change, + atomic_load_explicit(&eh->h.export_state, memory_order_relaxed)); + } + debug("\n"); + + } } -static inline void -rt_schedule_nhu(rtable *tab) +void +rt_dump_hooks_all(void) { - if (tab->nhu_state == NHU_CLEAN) - ev_schedule(tab->rt_event); + rtable *t; + node *n; - /* state change: - * NHU_CLEAN -> NHU_SCHEDULED - * NHU_RUNNING -> NHU_DIRTY - */ - tab->nhu_state |= NHU_SCHEDULED; + debug("Dump of all table hooks\n"); + + WALK_LIST2(t, n, routing_tables, n) + rt_dump_hooks(t); + + WALK_LIST2(t, n, deleted_routing_tables, n) + rt_dump_hooks(t); +} + +static inline void +rt_schedule_nhu(struct rtable_private *tab) +{ + if (tab->nhu_corked) + { + if (!(tab->nhu_corked & NHU_SCHEDULED)) + { + tab->nhu_corked |= NHU_SCHEDULED; + rt_lock_table(tab); + } + } + else if (!(tab->nhu_state & NHU_SCHEDULED)) + { + rt_trace(tab, D_EVENTS, "Scheduling NHU"); + rt_lock_table(tab); + + /* state change: + * NHU_CLEAN -> NHU_SCHEDULED + * NHU_RUNNING -> NHU_DIRTY + */ + if ((tab->nhu_state |= NHU_SCHEDULED) == NHU_SCHEDULED) + ev_schedule(tab->nhu_event); + } } void -rt_schedule_prune(rtable *tab) +rt_schedule_prune(struct rtable_private *tab) { if (tab->prune_state == 0) ev_schedule(tab->rt_event); @@ -1864,38 +2420,49 @@ rt_schedule_prune(rtable *tab) tab->prune_state |= 1; } +static void +rt_export_used(struct rt_table_exporter *e) +{ + struct rtable_private *tab = SKIP_BACK(struct rtable_private, exporter, e); + ASSERT_DIE(RT_IS_LOCKED(tab)); + + rt_trace(tab, D_EVENTS, "Export cleanup requested"); + + if (tab->export_used) + return; + + tab->export_used = 1; + ev_schedule(tab->rt_event); +} static void rt_event(void *ptr) { - rtable *tab = ptr; + RT_LOCKED((rtable *) ptr, tab) + { rt_lock_table(tab); - if (tab->hcu_scheduled) - rt_update_hostcache(tab); - - if (tab->nhu_state) - rt_next_hop_update(tab); + if (tab->export_used) + rt_export_cleanup(tab); if (tab->prune_state) rt_prune_table(tab); rt_unlock_table(tab); + } } - static void rt_prune_timer(timer *t) { - rtable *tab = t->data; - - if (tab->gc_counter >= tab->config->gc_threshold) - rt_schedule_prune(tab); + RT_LOCKED((rtable *) t->data, tab) + if (tab->gc_counter >= tab->config->gc_threshold) + rt_schedule_prune(tab); } static void -rt_kick_prune_timer(rtable *tab) +rt_kick_prune_timer(struct rtable_private *tab) { /* Return if prune is already scheduled */ if (tm_active(tab->prune_timer) || (tab->prune_state & 1)) @@ -1908,152 +2475,132 @@ rt_kick_prune_timer(rtable *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) +rt_flowspec_export_one(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *first) { - rtable *tab = t->data; - - if (!tab->base_settle_time) - return; - - btime settled_time = rt_settled_time(tab); - if (current_time() < settled_time) + struct rt_flowspec_link *ln = SKIP_BACK(struct rt_flowspec_link, req, req); + rtable *dst_pub = ln->dst; + ASSUME(rt_is_flow(dst_pub)); + struct rtable_private *dst = RT_LOCK(dst_pub); + + /* No need to inspect it further if recalculation is already scheduled */ + if ((dst->nhu_state == NHU_SCHEDULED) || (dst->nhu_state == NHU_DIRTY) + || !trie_match_net(dst->flowspec_trie, net)) { - tm_set(tab->settle_timer, settled_time); + RT_UNLOCK(dst_pub); + rpe_mark_seen_all(req->hook, first, NULL); 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; + /* This net may affect some flowspecs, check the actual change */ + rte *o = RTE_VALID_OR_NULL(first->old_best); + struct rte_storage *new_best = first->new_best; - if (tab->base_settle_time) - return; + RPE_WALK(first, rpe, NULL) + { + rpe_mark_seen(req->hook, rpe); + new_best = rpe->new_best; + } - rt_kick_settle_timer(tab); -} + /* Yes, something has actually changed. Schedule the update. */ + if (o != RTE_VALID_OR_NULL(new_best)) + rt_schedule_nhu(dst); -void -rt_subscribe(rtable *tab, struct rt_subscription *s) -{ - s->tab = tab; - rt_lock_table(tab); - add_tail(&tab->subscribers, &s->n); + RT_UNLOCK(dst_pub); } -void -rt_unsubscribe(struct rt_subscription *s) +static void +rt_flowspec_dump_req(struct rt_export_request *req) { - rem_node(&s->n); - rt_unlock_table(s->tab); + struct rt_flowspec_link *ln = SKIP_BACK(struct rt_flowspec_link, req, req); + debug(" Flowspec link for table %s (%p)\n", ln->dst->name, req); } static struct rt_flowspec_link * -rt_flowspec_find_link(rtable *src, rtable *dst) +rt_flowspec_find_link(struct rtable_private *src, rtable *dst) { - struct rt_flowspec_link *ln; - WALK_LIST(ln, src->flowspec_links) - if ((ln->src == src) && (ln->dst == dst)) - return ln; + struct rt_table_export_hook *hook; node *n; + WALK_LIST2(hook, n, src->exporter.e.hooks, h.n) + switch (atomic_load_explicit(&hook->h.export_state, memory_order_acquire)) + { + case TES_FEEDING: + case TES_READY: + if (hook->h.req->export_one == rt_flowspec_export_one) + { + struct rt_flowspec_link *ln = SKIP_BACK(struct rt_flowspec_link, req, hook->h.req); + if (ln->dst == dst) + return ln; + } + } return NULL; } void -rt_flowspec_link(rtable *src, rtable *dst) +rt_flowspec_link(rtable *src_pub, rtable *dst_pub) { - ASSERT(rt_is_ip(src)); - ASSERT(rt_is_flow(dst)); + ASSERT(rt_is_ip(src_pub)); + ASSERT(rt_is_flow(dst_pub)); - struct rt_flowspec_link *ln = rt_flowspec_find_link(src, dst); + int lock_dst = 0; - if (!ln) + RT_LOCKED(src_pub, src) { - rt_lock_table(src); - rt_lock_table(dst); + struct rt_flowspec_link *ln = rt_flowspec_find_link(src, dst_pub); + + if (!ln) + { + pool *p = src->rp; + ln = mb_allocz(p, sizeof(struct rt_flowspec_link)); + ln->src = src_pub; + ln->dst = dst_pub; + ln->req = (struct rt_export_request) { + .name = mb_sprintf(p, "%s.flowspec.notifier", dst_pub->name), + .list = &global_work_list, + .trace_routes = src->config->debug, + .dump_req = rt_flowspec_dump_req, + .export_one = rt_flowspec_export_one, + }; + + rt_table_export_start_locked(src, &ln->req); + + lock_dst = 1; + } - 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++; } - ln->uc++; + if (lock_dst) + rt_lock_table(dst_pub); } -void -rt_flowspec_unlink(rtable *src, rtable *dst) +static void +rt_flowspec_link_stopped(struct rt_export_request *req) { - 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); + struct rt_flowspec_link *ln = SKIP_BACK(struct rt_flowspec_link, req, req); + rtable *dst = ln->dst; - rt_unlock_table(src); - rt_unlock_table(dst); - } + mb_free(ln); + rt_unlock_table(dst); } -static void -rt_flowspec_notify(rtable *src, net *net) +void +rt_flowspec_unlink(rtable *src, rtable *dst) { - /* Only IP tables are src links */ - ASSERT(rt_is_ip(src)); - struct rt_flowspec_link *ln; - WALK_LIST(ln, src->flowspec_links) + RT_LOCKED(src, t) { - rtable *dst = ln->dst; - ASSERT(rt_is_flow(dst)); + ln = rt_flowspec_find_link(t, dst); - /* No need to inspect it further if recalculation is already active */ - if ((dst->nhu_state == NHU_SCHEDULED) || (dst->nhu_state == NHU_DIRTY)) - continue; + ASSERT(ln && (ln->uc > 0)); - if (trie_match_net(dst->flowspec_trie, net->n.addr)) - rt_schedule_nhu(dst); + if (!--ln->uc) + rt_stop_export(&ln->req, rt_flowspec_link_stopped); } } static void -rt_flowspec_reset_trie(rtable *tab) +rt_flowspec_reset_trie(struct rtable_private *tab) { linpool *lp = tab->flowspec_trie->lp; int ipv4 = tab->flowspec_trie->ipv4; @@ -2066,14 +2613,13 @@ rt_flowspec_reset_trie(rtable *tab) static void rt_free(resource *_r) { - rtable *r = (rtable *) _r; + struct rtable_private *r = SKIP_BACK(struct rtable_private, r, _r); + + DOMAIN_FREE(rtable, r->lock); 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); @@ -2084,7 +2630,6 @@ rt_free(resource *_r) fib_free(&r->fib); hmap_free(&r->id_map); rfree(r->rt_event); - rfree(r->settle_timer); mb_free(r); */ } @@ -2092,31 +2637,54 @@ rt_free(resource *_r) static void rt_res_dump(resource *_r) { - rtable *r = (rtable *) _r; + struct rtable_private *r = SKIP_BACK(struct rtable_private, r, _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), + .size = sizeof(rtable), .free = rt_free, .dump = rt_res_dump, .lookup = NULL, .memsize = NULL, }; +static const struct rt_exporter_class rt_table_exporter_class = { + .start = rt_table_export_start, + .stop = rt_table_export_stop, + .done = rt_table_export_done, +}; + +void +rt_exporter_init(struct rt_exporter *e) +{ + init_list(&e->hooks); +} + +static struct idm rtable_idm; +uint rtable_max_id = 0; + rtable * rt_setup(pool *pp, struct rtable_config *cf) { pool *p = rp_newf(pp, "Routing table %s", cf->name); - rtable *t = ralloc(p, &rt_class); + struct rtable_private *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; + t->id = idm_alloc(&rtable_idm); + if (t->id >= rtable_max_id) + rtable_max_id = t->id + 1; + + t->lock = DOMAIN_NEW(rtable, t->name); fib_init(&t->fib, p, t->addr_type, sizeof(net), OFFSETOF(net, n), 0, NULL); @@ -2128,27 +2696,41 @@ rt_setup(pool *pp, struct rtable_config *cf) t->fib.init = net_init_with_trie; } - init_list(&t->channels); - init_list(&t->flowspec_links); - init_list(&t->subscribers); + init_list(&t->imports); - if (!(t->internal = cf->internal)) - { - hmap_init(&t->id_map, p, 1024); - hmap_set(&t->id_map, 0); + hmap_init(&t->id_map, p, 1024); + hmap_set(&t->id_map, 0); - t->rt_event = ev_new_init(p, rt_event, t); - t->prune_timer = tm_new_init(p, rt_prune_timer, t, 0, 0); - t->last_rt_change = t->gc_time = current_time(); + t->rt_event = ev_new_init(p, rt_event, t); + t->nhu_event = ev_new_init(p, rt_next_hop_update, t); + t->prune_timer = tm_new_init(p, rt_prune_timer, t, 0, 0); + t->last_rt_change = t->gc_time = current_time(); - 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); - } + t->exporter = (struct rt_table_exporter) { + .e = { + .class = &rt_table_exporter_class, + .addr_type = t->addr_type, + .rp = t->rp, + }, + .export_timer = tm_new_init(p, rt_announce_exports, t, 0, 0), + .next_seq = 1, + }; + + rt_exporter_init(&t->exporter.e); + + init_list(&t->exporter.pending); + + t->cork_threshold = cf->cork_threshold; + + t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS; + + if (rt_is_flow(RT_PUB(t))) + { + t->flowspec_trie = f_new_trie(lp_new_default(p), 0); + t->flowspec_trie->ipv4 = (t->addr_type == NET_FLOW4); } - return t; + return RT_PUB(t); } /** @@ -2162,9 +2744,11 @@ rt_init(void) { rta_init(); rt_table_pool = rp_new(&root_pool, "Routing tables"); - rte_update_pool = lp_new_default(rt_table_pool); - rte_slab = sl_new(rt_table_pool, sizeof(rte)); init_list(&routing_tables); + init_list(&deleted_routing_tables); + ev_init_list(&rt_cork.queue, &main_birdloop, "Route cork release"); + rt_cork.run = (event) { .hook = rt_cork_release_hook }; + idm_init(&rtable_idm, rt_table_pool, 256); } @@ -2183,15 +2767,15 @@ rt_init(void) * iteration. */ static void -rt_prune_table(rtable *tab) +rt_prune_table(struct rtable_private *tab) { struct fib_iterator *fit = &tab->prune_fit; int limit = 2000; - struct channel *c; + struct rt_import_hook *ih; node *n, *x; - DBG("Pruning route table %s\n", tab->name); + rt_trace(tab, D_STATES, "Pruning"); #ifdef DEBUGGING fib_check(&tab->fib); #endif @@ -2202,9 +2786,16 @@ rt_prune_table(rtable *tab) 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; + 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; @@ -2223,8 +2814,6 @@ rt_prune_table(rtable *tab) again: FIB_ITERATE_START(&tab->fib, fit, net, n) { - rte *e; - rescan: if (limit <= 0) { @@ -2233,26 +2822,21 @@ again: return; } - for (e=n->routes; e; e=e->next) + for (struct rte_storage *e=n->routes; e; e=e->next) { - if (e->sender->flush_active || (e->flags & REF_DISCARD)) - { - rte_discard(e); - limit--; - - goto rescan; - } - - if (e->flags & REF_MODIFY) + 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)) { - rte_modify(e); + rte_recalculate(tab, e->rte.sender, n, NULL, e->rte.src); limit--; goto rescan; } } - if (!n->routes) /* Orphaned FIB entry */ + if (!n->routes && !n->first) /* Orphaned FIB entry */ { FIB_ITERATE_PUT(fit); fib_delete(&tab->fib, n); @@ -2267,6 +2851,10 @@ again: } FIB_ITERATE_END; + rt_trace(tab, D_EVENTS, "Prune done, scheduling export timer"); + if (!tm_active(tab->exporter.export_timer)) + tm_start(tab->exporter.export_timer, tab->config->export_settle_time); + #ifdef DEBUGGING fib_check(&tab->fib); #endif @@ -2305,21 +2893,203 @@ again: } } - if (tab->prune_state > 0) - ev_schedule(tab->rt_event); - - /* FIXME: This should be handled in a better way */ - rt_prune_sources(); + uint flushed_channels = 0; /* Close flushed channels */ - WALK_LIST2_DELSAFE(c, n, x, tab->channels, table_node) - if (c->flush_active) + WALK_LIST2_DELSAFE(ih, n, x, tab->imports, n) + if (ih->import_state == TIS_FLUSHING) + { + ih->flush_seq = tab->exporter.next_seq; + rt_set_import_state(ih, TIS_WAITING); + flushed_channels++; + } + 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); + } + + /* In some cases, we may want to directly proceed to export cleanup */ + if (EMPTY_LIST(tab->exporter.e.hooks) && flushed_channels) + rt_export_cleanup(tab); +} + +static void +rt_export_cleanup(struct rtable_private *tab) +{ + tab->export_used = 0; + + u64 min_seq = ~((u64) 0); + struct rt_pending_export *last_export_to_free = NULL; + struct rt_pending_export *first = tab->exporter.first; + int want_prune = 0; + + struct rt_table_export_hook *eh; + node *n; + WALK_LIST2(eh, n, tab->exporter.e.hooks, h.n) + { + switch (atomic_load_explicit(&eh->h.export_state, memory_order_acquire)) + { + case TES_DOWN: + continue; + + case TES_READY: + { + struct rt_pending_export *last = atomic_load_explicit(&eh->last_export, memory_order_acquire); + if (!last) + /* No last export means that the channel has exported nothing since last cleanup */ + goto done; + + else if (min_seq > last->seq) + { + min_seq = last->seq; + last_export_to_free = last; + } + continue; + } + + default: + /* It's only safe to cleanup when the export state is idle or regular. No feeding or stopping allowed. */ + goto done; + } + } + + tab->exporter.first = last_export_to_free ? rt_next_export_fast(last_export_to_free) : NULL; + + rt_trace(tab, D_STATES, "Export cleanup, old exporter.first seq %lu, new %lu, min_seq %ld", + first ? first->seq : 0, + tab->exporter.first ? tab->exporter.first->seq : 0, + min_seq); + + WALK_LIST2(eh, n, tab->exporter.e.hooks, h.n) + { + if (atomic_load_explicit(&eh->h.export_state, memory_order_acquire) != TES_READY) + continue; + + struct rt_pending_export *last = atomic_load_explicit(&eh->last_export, memory_order_acquire); + if (last == last_export_to_free) + { + /* This may fail when the channel managed to export more inbetween. This is OK. */ + atomic_compare_exchange_strong_explicit( + &eh->last_export, &last, NULL, + memory_order_release, + memory_order_relaxed); + + DBG("store hook=%p last_export=NULL\n", eh); + } + } + + while (first && (first->seq <= min_seq)) + { + ASSERT_DIE(first->new || first->old); + + const net_addr *n = first->new ? + first->new->rte.net : + first->old->rte.net; + net *net = SKIP_BACK(struct network, n.addr, (net_addr (*)[0]) n); + + ASSERT_DIE(net->first == first); + + if (first == net->last) + /* The only export here */ + net->last = net->first = NULL; + else + /* First is now the next one */ + net->first = atomic_load_explicit(&first->next, memory_order_relaxed); + + want_prune += !net->routes && !net->first; + + /* For now, the old route may be finally freed */ + if (first->old) + { + rt_rte_trace_in(D_ROUTES, first->old->rte.sender->req, &first->old->rte, "freed"); + hmap_clear(&tab->id_map, first->old->rte.id); + rte_free(first->old); + } + +#ifdef LOCAL_DEBUG + memset(first, 0xbd, sizeof(struct rt_pending_export)); +#endif + + struct rt_export_block *reb = HEAD(tab->exporter.pending); + ASSERT_DIE(reb == PAGE_HEAD(first)); + + u32 pos = (first - &reb->export[0]); + u32 end = atomic_load_explicit(&reb->end, memory_order_relaxed); + ASSERT_DIE(pos < end); + + struct rt_pending_export *next = NULL; + + if (++pos < end) + next = &reb->export[pos]; + else + { + rem_node(&reb->n); + +#ifdef LOCAL_DEBUG + memset(reb, 0xbe, page_size); +#endif + + free_page(reb); + + if (EMPTY_LIST(tab->exporter.pending)) + { + rt_trace(tab, D_EVENTS, "Resetting export seq"); + + node *n; + WALK_LIST2(eh, n, tab->exporter.e.hooks, h.n) + { + if (atomic_load_explicit(&eh->h.export_state, memory_order_acquire) != TES_READY) + continue; + + ASSERT_DIE(atomic_load_explicit(&eh->last_export, memory_order_acquire) == NULL); + bmap_reset(&eh->h.seq_map, 1024); + } + + tab->exporter.next_seq = 1; + } + else { - c->flush_active = 0; - channel_set_state(c, CS_DOWN); + reb = HEAD(tab->exporter.pending); + next = &reb->export[0]; } + } + + first = next; + } - return; + rt_check_cork_low(tab); + +done:; + struct rt_import_hook *ih; node *x; + WALK_LIST2_DELSAFE(ih, n, x, tab->imports, n) + if (ih->import_state == TIS_WAITING) + if (!first || (first->seq >= ih->flush_seq)) + { + ih->import_state = TIS_CLEARED; + ev_send(ih->req->list, &ih->announce_event); + } + + if ((tab->gc_counter += want_prune) >= tab->config->gc_threshold) + rt_kick_prune_timer(tab); + + if (tab->export_used) + ev_schedule(tab->rt_event); + + + if (EMPTY_LIST(tab->exporter.pending) && tm_active(tab->exporter.export_timer)) + tm_stop(tab->exporter.export_timer); +} + +static void +rt_cork_release_hook(void *data UNUSED) +{ + do synchronize_rcu(); + while ( + !atomic_load_explicit(&rt_cork.active, memory_order_acquire) && + ev_run_list(&rt_cork.queue) + ); } /** @@ -2336,7 +3106,7 @@ again: * */ struct f_trie * -rt_lock_trie(rtable *tab) +rt_lock_trie(struct rtable_private *tab) { ASSERT(tab->trie); @@ -2353,7 +3123,7 @@ rt_lock_trie(rtable *tab) * It may free the trie and schedule next trie pruning. */ void -rt_unlock_trie(rtable *tab, struct f_trie *trie) +rt_unlock_trie(struct rtable_private *tab, struct f_trie *trie) { ASSERT(trie); @@ -2393,8 +3163,8 @@ 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); + c->def_tables[NET_IP4] = cf_define_symbol(cf_get_symbol("master4"), SYM_TABLE, table, NULL); + c->def_tables[NET_IP6] = cf_define_symbol(cf_get_symbol("master6"), SYM_TABLE, table, NULL); } void @@ -2409,6 +3179,13 @@ rt_postconfig(struct config *c) WALK_LIST(rc, c->tables) if (rc->gc_period == (uint) -1) rc->gc_period = (uint) def_gc_period; + + for (uint net_type = 0; net_type < NET_MAX; net_type++) + if (c->def_tables[net_type] && !c->def_tables[net_type]->table) + { + c->def_tables[net_type]->class = SYM_VOID; + c->def_tables[net_type] = NULL; + } } @@ -2418,135 +3195,168 @@ rt_postconfig(struct config *c) */ void -rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls) +ea_set_hostentry(ea_list **to, rtable *dep, rtable *src, ip_addr gw, ip_addr ll, u32 lnum, u32 labels[lnum]) +{ + struct { + struct adata ad; + struct hostentry *he; + u32 labels[lnum]; + } *head = (void *) tmp_alloc_adata(sizeof *head - sizeof(struct adata)); + + RT_LOCKED(src, tab) + head->he = rt_get_hostentry(tab, gw, ll, dep); + memcpy(head->labels, labels, lnum * sizeof(u32)); + + ea_set_attr(to, EA_LITERAL_DIRECT_ADATA( + &ea_gen_hostentry, 0, &head->ad)); +} + + +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; - if (a->dest != RTD_UNICAST) + ea_set_attr_u32(to, &ea_gen_igp_metric, 0, he->igp_metric); + + 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 int -rta_next_hop_outdated(rta *a) +static inline struct hostentry_adata * +rta_next_hop_outdated(ea_list *a) { - struct hostentry *he = a->hostentry; + /* First retrieve the hostentry */ + eattr *heea = ea_find(a, &ea_gen_hostentry); + if (!heea) + return NULL; - if (!he) - return 0; + struct hostentry_adata *head = (struct hostentry_adata *) heea->u.ptr; - if (!he->src) - return a->dest != RTD_UNREACHABLE; + /* 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; + + 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; - return (a->dest != he->dest) || (a->igp_metric != he->igp_metric) || - (!he->nexthop_linkable) || !nexthop_same(&(a->nh), &(he->src->nh)); + /* 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 rte * -rt_next_hop_update_rte(rtable *tab UNUSED, rte *old) +static inline int +rt_next_hop_update_rte(rte *old, rte *new) { - if (!rta_next_hop_outdated(old->attrs)) - return NULL; - - rta *a = alloca(RTA_MAX_SIZE); - memcpy(a, old->attrs, rta_size(old->attrs)); + struct hostentry_adata *head = rta_next_hop_outdated(old->attrs); + if (!head) + return 0; - mpls_label_stack mls = { .len = a->nh.labels_orig }; - memcpy(mls.stack, &a->nh.label[a->nh.labels - mls.len], mls.len * sizeof(u32)); + *new = *old; + rta_apply_hostentry(&new->attrs, head); + return 1; +} - rta_apply_hostentry(a, old->attrs->hostentry, &mls); - a->cached = 0; +static inline void +rt_next_hop_resolve_rte(rte *r) +{ + eattr *heea = ea_find(r->attrs, &ea_gen_hostentry); + if (!heea) + return; - rte *e = sl_alloc(rte_slab); - memcpy(e, old, sizeof(rte)); - e->attrs = rta_lookup(a); - rt_lock_source(e->src); + struct hostentry_adata *head = (struct hostentry_adata *) heea->u.ptr; - return e; + rta_apply_hostentry(&r->attrs, head); } - #ifdef CONFIG_BGP static inline int @@ -2570,35 +3380,34 @@ net_flow_has_dst_prefix(const net_addr *n) } static inline int -rta_as_path_is_empty(rta *a) +rta_as_path_is_empty(ea_list *a) { - eattr *e = ea_find(a->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); + eattr *e = ea_find(a, "bgp_path"); return !e || (as_path_getlen(e->u.ptr) == 0); } static inline u32 -rta_get_first_asn(rta *a) +rta_get_first_asn(ea_list *a) { - eattr *e = ea_find(a->eattrs, EA_CODE(PROTOCOL_BGP, BA_AS_PATH)); + eattr *e = ea_find(a, "bgp_path"); u32 asn; return (e && as_path_get_first_regular(e->u.ptr, &asn)) ? asn : 0; } -int -rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, rta *a, int interior) +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 0; + return FLOWSPEC_INVALID; /* RFC 9117 4.1. Accept AS_PATH is empty (fr */ if (interior && rta_as_path_is_empty(a)) - return 1; + return FLOWSPEC_VALID; /* RFC 8955 6. b) Flowspec and its best-match route have the same originator */ @@ -2610,146 +3419,239 @@ rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, rta *a, i 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); - rte *rb = nb ? nb->routes : NULL; + rte rb = {}; + net_addr_union nau; + RT_LOCKED(tab_ip, tip) + { + ASSERT(tip->trie); + /* Find best-match BGP unicast route for flowspec dst prefix */ + net *nb = net_route(tip, &dst); + if (nb) + { + rb = RTE_COPY_VALID(nb->routes); + rta_clone(rb.attrs); + net_copy(&nau.n, nb->n.addr); + rb.net = &nau.n; + } + } /* 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); + RT_LOCKED(tab_flow, tfl) + trie_add_prefix(tfl->flowspec_trie, &dst, (rb.net ? rb.net->pxlen : 0), max_pxlen); /* No best-match BGP route -> no flowspec */ - if (!rb || (rb->attrs->source != RTS_BGP)) - return 0; + if (!rb.attrs || (rt_get_source_attr(&rb) != RTS_BGP)) + return FLOWSPEC_INVALID; /* Find ORIGINATOR_ID values */ - u32 orig_a = ea_get_int(a->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGINATOR_ID), 0); - u32 orig_b = ea_get_int(rb->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGINATOR_ID), 0); + 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(a->from, rb->attrs->from))) - return 0; + 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); + u32 asn_b = rta_get_first_asn(rb.attrs); if (!asn_b) - return 0; + 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 0; + 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) + RT_LOCKED(tab_ip, tip) { - net *nc = net_find_valid(tab_ip, &subnet); - if (!nc) - continue; + TRIE_WALK(tip->trie, subnet, &dst) + { + net *nc = net_find_valid(tip, &subnet); + if (!nc) + continue; - rte *rc = nc->routes; - if (rc->attrs->source != RTS_BGP) - return 0; + const rte *rc = &nc->routes->rte; + if (rt_get_source_attr(rc) != RTS_BGP) + RT_RETURN(tip, FLOWSPEC_INVALID); - if (rta_get_first_asn(rc->attrs) != asn_b) - return 0; + if (rta_get_first_asn(rc->attrs) != asn_b) + RT_RETURN(tip, FLOWSPEC_INVALID); + } + TRIE_WALK_END; } - TRIE_WALK_END; - return 1; + return FLOWSPEC_VALID; } #endif /* CONFIG_BGP */ -static rte * -rt_flowspec_update_rte(rtable *tab, rte *r) +static int +rt_flowspec_update_rte(rtable *tab, rte *r, rte *new) { #ifdef CONFIG_BGP - if ((r->attrs->source != RTS_BGP) || (r->sender->proto != r->src->proto)) - return NULL; + if (r->generation || (rt_get_source_attr(r) != RTS_BGP)) + return 0; - struct bgp_channel *bc = (struct bgp_channel *) r->sender; + struct bgp_channel *bc = (struct bgp_channel *) SKIP_BACK(struct channel, in_req, r->sender->req); if (!bc->base_table) - return NULL; - - const net_addr *n = r->net->n.addr; - struct bgp_proto *p = (void *) r->src->proto; - int valid = rt_flowspec_check(bc->base_table, tab, n, r->attrs, p->is_interior); - int dest = valid ? RTD_NONE : RTD_UNREACHABLE; + return 0; - if (dest == r->attrs->dest) - return NULL; + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, p, bc->c.proto); - rta *a = alloca(RTA_MAX_SIZE); - memcpy(a, r->attrs, rta_size(r->attrs)); - a->dest = dest; - a->cached = 0; + enum flowspec_valid old = rt_get_flowspec_valid(r), + valid = rt_flowspec_check(bc->base_table, tab, r->net, r->attrs, p->is_interior); - rte *new = sl_alloc(rte_slab); - memcpy(new, r, sizeof(rte)); - new->attrs = rta_lookup(a); - rt_lock_source(new->src); + if (old == valid) + return 0; - return new; + *new = *r; + ea_set_attr_u32(&new->attrs, &ea_gen_flowspec_valid, 0, valid); + return 1; #else - return NULL; + return 0; #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) +rt_next_hop_update_net(struct rtable_private *tab, net *n) { - rte **k, *e, *new, *old_best, **new_best; - int count = 0; - int free_old_best = 0; + uint count = 0; + int is_flow = net_is_flow(n->n.addr); - old_best = n->routes; + struct rte_storage *old_best = n->routes; if (!old_best) return 0; - for (k = &n->routes; e = *k; k = &e->next) + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) + count++; + + if (!count) + return 0; + + struct rte_multiupdate { + struct rte_storage *old, *new_stored; + rte new; + } *updates = tmp_allocz(sizeof(struct rte_multiupdate) * (count+1)); + + struct rt_pending_export *last_pending = n->last; + + uint pos = 0; + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) + updates[pos++].old = e; + + /* This is an exceptional place where table can be unlocked while keeping its data: + * the reason why this is safe is that NHU must be always run from the same + * thread as cleanup routines, therefore the only real problem may arise when + * some importer does a change on this particular net (destination) while NHU + * is being computed. Statistically, this should almost never happen. In such + * case, we just drop all the computed changes and do it once again. + * */ + RT_UNLOCK(tab); + + uint mod = 0; + if (is_flow) + for (uint i = 0; i < pos; i++) + mod += rt_flowspec_update_rte(RT_PUB(tab), &updates[i].old->rte, &updates[i].new); + + else + for (uint i = 0; i < pos; i++) + mod += rt_next_hop_update_rte(&updates[i].old->rte, &updates[i].new); + + RT_LOCK(RT_PUB(tab)); + + if (!mod) + return 0; + + /* Something has changed inbetween, retry NHU. */ + if (last_pending != n->last) + return rt_next_hop_update_net(tab, n); + + /* Now we reconstruct the original linked list */ + struct rte_storage **nptr = &n->routes; + for (uint i = 0; i < pos; i++) { - if (!net_is_flow(n->n.addr)) - new = rt_next_hop_update_rte(tab, e); + updates[i].old->next = NULL; + + struct rte_storage *put; + if (updates[i].new.attrs) + put = updates[i].new_stored = rte_store(&updates[i].new, n, tab); else - new = rt_flowspec_update_rte(tab, e); + put = updates[i].old; - if (new) - { - *k = new; + *nptr = put; + nptr = &put->next; + } + *nptr = NULL; - rte_trace_in(D_ROUTES, new->sender, new, "updated"); - rte_announce_i(tab, RA_ANY, n, new, e, NULL, NULL); + /* Call the pre-comparison hooks */ + for (uint i = 0; i < pos; i++) + if (updates[i].new_stored) + { + /* Get a new ID for the route */ + updates[i].new_stored->rte.lastmod = current_time(); + updates[i].new_stored->rte.id = hmap_first_zero(&tab->id_map); + hmap_set(&tab->id_map, updates[i].new_stored->rte.id); /* Call a pre-comparison hook */ /* Not really an efficient way to compute this */ - if (e->src->proto->rte_recalculate) - e->src->proto->rte_recalculate(tab, n, new, e, NULL); - - if (e != old_best) - rte_free_quick(e); - else /* Freeing of the old best rte is postponed */ - free_old_best = 1; - - e = new; - count++; + if (updates[i].old->rte.src->owner->rte_recalculate) + updates[i].old->rte.src->owner->rte_recalculate(tab, n, &updates[i].new_stored->rte, &updates[i].old->rte, &old_best->rte); } - } - if (!count) - return 0; +#if DEBUGGING + { + uint t = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + t++; + ASSERT_DIE(t == pos); + ASSERT_DIE(pos == count); + } +#endif /* Find the new best route */ - new_best = NULL; - for (k = &n->routes; e = *k; k = &e->next) + 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, *new_best)) + 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; + struct rte_storage *new = *new_best; if (new != n->routes) { *new_best = new->next; @@ -2757,84 +3659,155 @@ rt_next_hop_update_net(rtable *tab, net *n) n->routes = new; } - /* Announce the new best route */ - if (new != old_best) - rte_trace_in(D_ROUTES, new->sender, new, "updated [best]"); + uint total = 0; + /* Announce the changes */ + for (uint i=0; i<count; i++) + { + if (!updates[i].new_stored) + continue; - /* Propagate changes */ - rte_announce_i(tab, RA_UNDEF, n, NULL, NULL, n->routes, old_best); + _Bool nb = (new->rte.src == updates[i].new.src), ob = (i == 0); + const char *best_indicator[2][2] = { + { "autoupdated", "autoupdated [-best]" }, + { "autoupdated [+best]", "autoupdated [best]" } + }; + rt_rte_trace_in(D_ROUTES, updates[i].new.sender->req, &updates[i].new, best_indicator[nb][ob]); + rte_announce(tab, n, updates[i].new_stored, updates[i].old, new, old_best); - if (free_old_best) - rte_free_quick(old_best); + total++; + } - return count; + return total; } static void -rt_next_hop_update(rtable *tab) +rt_next_hop_update(void *_tab) { + RT_LOCKED((rtable *) _tab, tab) + { + + /* If called from an uncork hook, reset the state */ + if (tab->nhu_corked) + { + ASSERT_DIE(tab->nhu_state == 0); + tab->nhu_state = tab->nhu_corked; + tab->nhu_corked = 0; + rt_trace(tab, D_STATES, "Next hop updater uncorked"); + } + + if (!tab->nhu_state) + bug("Called NHU event for no reason in table %s", tab->name); + + /* Check corkedness */ + if (rt_cork_check(tab->nhu_event)) + { + rt_trace(tab, D_STATES, "Next hop updater corked"); + if ((tab->nhu_state & NHU_RUNNING) + && !EMPTY_LIST(tab->exporter.pending) + && !tm_active(tab->exporter.export_timer)) + tm_start(tab->exporter.export_timer, tab->config->export_settle_time); + + tab->nhu_corked = tab->nhu_state; + tab->nhu_state = 0; + RT_RETURN(tab); + } + struct fib_iterator *fit = &tab->nhu_fit; int max_feed = 32; - if (tab->nhu_state == NHU_CLEAN) - return; - + /* Initialize a new run */ if (tab->nhu_state == NHU_SCHEDULED) - { - FIB_ITERATE_INIT(fit, &tab->fib); - tab->nhu_state = NHU_RUNNING; + { + FIB_ITERATE_INIT(fit, &tab->fib); + tab->nhu_state = NHU_RUNNING; - if (tab->flowspec_trie) - rt_flowspec_reset_trie(tab); - } + if (tab->flowspec_trie) + rt_flowspec_reset_trie(tab); + } + /* Walk the fib one net after another */ FIB_ITERATE_START(&tab->fib, fit, net, n) { if (max_feed <= 0) { FIB_ITERATE_PUT(fit); - ev_schedule(tab->rt_event); - return; + ev_schedule(tab->nhu_event); + RT_RETURN(tab); } + lp_state lps; + lp_save(tmp_linpool, &lps); max_feed -= rt_next_hop_update_net(tab, n); + lp_restore(tmp_linpool, &lps); } FIB_ITERATE_END; + /* Finished NHU, cleanup */ + rt_trace(tab, D_EVENTS, "NHU done, scheduling export timer"); + + if (!tm_active(tab->exporter.export_timer)) + tm_start(tab->exporter.export_timer, tab->config->export_settle_time); + /* State change: * NHU_DIRTY -> NHU_SCHEDULED * NHU_RUNNING -> NHU_CLEAN */ - tab->nhu_state &= 1; + if ((tab->nhu_state &= NHU_SCHEDULED) == NHU_SCHEDULED) + ev_schedule(tab->nhu_event); - if (tab->nhu_state != NHU_CLEAN) - ev_schedule(tab->rt_event); + rt_unlock_table(tab); + + } } +void +rt_new_default_table(struct symbol *s) +{ + for (uint addr_type = 0; addr_type < NET_MAX; addr_type++) + if (s == new_config->def_tables[addr_type]) + { + s->table = rt_new_table(s, addr_type); + return; + } + + bug("Requested an unknown new default table: %s", s->name); +} struct rtable_config * -rt_new_table(struct symbol *s, uint addr_type) +rt_get_default_table(struct config *cf, 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 symbol *ts = cf->def_tables[addr_type]; + if (!ts) + return NULL; + + if (!ts->table) + rt_new_default_table(ts); + return ts->table; +} + +struct rtable_config * +rt_new_table(struct symbol *s, uint addr_type) +{ struct rtable_config *c = cfg_allocz(sizeof(struct rtable_config)); - cf_define_symbol(s, SYM_TABLE, table, c); + if (s == new_config->def_tables[addr_type]) + s->table = c; + else + cf_define_symbol(s, SYM_TABLE, table, c); + c->name = s->name; c->addr_type = addr_type; c->gc_threshold = 1000; c->gc_period = (uint) -1; /* set in rt_postconfig() */ - c->min_settle_time = 1 S; - c->max_settle_time = 20 S; + c->cork_threshold.low = 128; + c->cork_threshold.high = 512; + c->debug = new_config->table_debug; 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; + new_config->def_tables[addr_type] = s; return c; } @@ -2848,8 +3821,9 @@ rt_new_table(struct symbol *s, uint addr_type) * configuration. */ void -rt_lock_table(rtable *r) +rt_lock_table_priv(struct rtable_private *r, const char *file, uint line) { + rt_trace(r, D_STATES, "Locked at %s:%d", file, line); r->use_count++; } @@ -2862,20 +3836,60 @@ rt_lock_table(rtable *r) * for deletion by configuration changes. */ void -rt_unlock_table(rtable *r) +rt_unlock_table_priv(struct rtable_private *r, const char *file, uint line) { + rt_trace(r, D_STATES, "Unlocked at %s:%d", file, line); if (!--r->use_count && r->deleted) - { - struct config *conf = r->deleted; + /* Schedule the delete event to finish this up */ + ev_send(&global_event_list, ev_new_init(r->rp, rt_delete, r)); +} - /* Delete the routing table by freeing its pool */ - rt_shutdown(r); - config_del_obstacle(conf); - } +static void +rt_delete(void *tab_) +{ + /* We assume that nobody holds the table reference now as use_count is zero. + * Anyway the last holder may still hold the lock. Therefore we lock and + * unlock it the last time to be sure that nobody is there. */ + struct rtable_private *tab = RT_LOCK((rtable *) tab_); + struct config *conf = tab->deleted; + + RT_UNLOCK(RT_PUB(tab)); + + rfree(tab->rp); + config_del_obstacle(conf); +} + + +static void +rt_check_cork_low(struct rtable_private *tab) +{ + if (!tab->cork_active) + return; + + if (!tab->exporter.first || (tab->exporter.first->seq + tab->cork_threshold.low > tab->exporter.next_seq)) + { + tab->cork_active = 0; + rt_cork_release(); + + rt_trace(tab, D_STATES, "Uncorked"); + } } +static void +rt_check_cork_high(struct rtable_private *tab) +{ + if (!tab->cork_active && tab->exporter.first && (tab->exporter.first->seq + tab->cork_threshold.high <= tab->exporter.next_seq)) + { + tab->cork_active = 1; + rt_cork_acquire(); + + rt_trace(tab, D_STATES, "Corked"); + } +} + + static int -rt_reconfigure(rtable *tab, struct rtable_config *new, struct rtable_config *old) +rt_reconfigure(struct rtable_private *tab, struct rtable_config *new, struct rtable_config *old) { if ((new->addr_type != old->addr_type) || (new->sorted != old->sorted) || @@ -2883,10 +3897,26 @@ rt_reconfigure(rtable *tab, struct rtable_config *new, struct rtable_config *old return 0; DBG("\t%s: same\n", new->name); - new->table = tab; + new->table = RT_PUB(tab); tab->name = new->name; tab->config = new; + if (tab->hostcache) + tab->hostcache->req.trace_routes = new->debug; + + struct rt_table_export_hook *hook; node *n; + WALK_LIST2(hook, n, tab->exporter.e.hooks, h.n) + if (hook->h.req->export_one == rt_flowspec_export_one) + hook->h.req->trace_routes = new->debug; + + tab->cork_threshold = new->cork_threshold; + + if (new->cork_threshold.high != old->cork_threshold.high) + rt_check_cork_high(tab); + + if (new->cork_threshold.low != old->cork_threshold.low) + rt_check_cork_low(tab); + return 1; } @@ -2919,19 +3949,32 @@ rt_commit(struct config *new, struct config *old) { WALK_LIST(o, old->tables) { - rtable *tab = o->table; + struct rtable_private *tab = RT_LOCK(o->table); + if (tab->deleted) + { + RT_UNLOCK(tab); continue; + } r = rt_find_table_config(new, o->name); if (r && !new->shutdown && rt_reconfigure(tab, r, o)) + { + RT_UNLOCK(tab); continue; + } DBG("\t%s: deleted\n", o->name); tab->deleted = old; config_add_obstacle(old); rt_lock_table(tab); + + if (tab->hostcache) + rt_stop_export(&tab->hostcache->req, NULL); + rt_unlock_table(tab); + + RT_UNLOCK(tab); } } @@ -2945,387 +3988,242 @@ rt_commit(struct config *new, struct config *old) DBG("\tdone\n"); } -static inline void -do_feed_channel(struct channel *c, net *n, rte *e) +static void +rt_feed_done(struct rt_export_hook *c) { - 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 */ - rt_notify_basic(c, n, e, e, c->refeeding); - rte_update_unlock(); -} + c->event.hook = rt_export_hook; -/** - * 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; + rt_set_export_state(c, TES_READY); - ASSERT(c->export_state == ES_FEEDING); + rt_send_export_event(c); +} - if (!c->feed_active) - { - FIB_ITERATE_INIT(fit, &c->table->fib); - c->feed_active = 1; - } +#define MAX_FEED_BLOCK 1024 +typedef struct { + uint cnt, pos; + union { + struct rt_pending_export *rpe; + struct { + rte **feed; + uint *start; + }; + }; +} rt_feed_block; - FIB_ITERATE_START(&c->table->fib, fit, net, n) +static int +rt_prepare_feed(struct rt_table_export_hook *c, net *n, rt_feed_block *b) +{ + if (n->routes) + { + if (c->h.req->export_bulk) { - rte *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 (rte_is_valid(e)) - { - /* In the meantime, the protocol may fell down */ - if (c->export_state != ES_FEEDING) - goto done; - - do_feed_channel(c, n, e); - max_feed--; - } + uint cnt = rte_feed_count(n); + if (b->cnt && (b->cnt + cnt > MAX_FEED_BLOCK)) + return 0; - 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 (!b->cnt) + { + b->feed = tmp_alloc(sizeof(rte *) * MAX(MAX_FEED_BLOCK, cnt)); + b->start = tmp_alloc(sizeof(uint) * ((cnt >= MAX_FEED_BLOCK) ? 2 : (MAX_FEED_BLOCK + 2 - cnt))); + } - if (!rte_is_valid(e)) - continue; + rte_feed_obtain(n, &b->feed[b->cnt], cnt); + b->start[b->pos++] = b->cnt; + b->cnt += cnt; + } + else if (b->pos == MAX_FEED_BLOCK) + return 0; + else + { + if (!b->pos) + b->rpe = tmp_alloc(sizeof(struct rt_pending_export) * MAX_FEED_BLOCK); - do_feed_channel(c, n, e); - max_feed--; - } + b->rpe[b->pos++] = (struct rt_pending_export) { .new = n->routes, .new_best = n->routes }; } - FIB_ITERATE_END; + } -done: - c->feed_active = 0; + rpe_mark_seen_all(&c->h, n->first, NULL); 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) +static void +rt_process_feed(struct rt_table_export_hook *c, rt_feed_block *b) { - if (c->feed_active) + if (!b->pos) + return; + + if (c->h.req->export_bulk) + { + b->start[b->pos] = b->cnt; + for (uint p = 0; p < b->pos; p++) { - /* Unlink the iterator */ - fit_get(&c->table->fib, &c->feed_fit); - c->feed_active = 0; + rte **feed = &b->feed[b->start[p]]; + c->h.req->export_bulk(c->h.req, feed[0]->net, NULL, feed, b->start[p+1] - b->start[p]); } + } + else + for (uint p = 0; p < b->pos; p++) + c->h.req->export_one(c->h.req, b->rpe[p].new->rte.net, &b->rpe[p]); } - -/* - * Import table +/** + * 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 + * 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 -rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +static void +rt_feed_by_fib(void *data) { - struct rtable *tab = c->in_table; - rte *old, **pos; - net *net; + struct rt_table_export_hook *c = data; + struct fib_iterator *fit = &c->feed_fit; + rt_feed_block block = {}; - if (new) - { - net = net_get(tab, n); + ASSERT(atomic_load_explicit(&c->h.export_state, memory_order_relaxed) == TES_FEEDING); - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - } - else + RT_LOCKED(RT_PUB(SKIP_BACK(struct rtable_private, exporter, c->table)), tab) { - net = net_find(tab, n); - if (!net) - goto drop_withdraw; - } - - /* Find the old rte */ - for (pos = &net->routes; old = *pos; pos = &old->next) - if (old->src == src) + FIB_ITERATE_START(&tab->fib, fit, net, n) { - if (new && rte_same(old, new)) + if ((c->h.req->addr_mode == TE_ADDR_NONE) || net_in_netX(n->n.addr, c->h.req->addr)) { - /* Refresh the old rte, continue with update to main rtable */ - if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY)) + if (atomic_load_explicit(&c->h.export_state, memory_order_acquire) != TES_FEEDING) + return; + + if (!rt_prepare_feed(c, n, &block)) { - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); - return 1; + FIB_ITERATE_PUT(fit); + RT_UNLOCK(tab); + rt_process_feed(c, &block); + rt_send_export_event(&c->h); + return; } - - goto drop_update; } + } + FIB_ITERATE_END; + } - /* Move iterator if needed */ - if (old == c->reload_next_rte) - c->reload_next_rte = old->next; - - /* Remove the old rte */ - *pos = old->next; - rte_free_quick(old); - tab->rt_count--; + rt_process_feed(c, &block); + rt_feed_done(&c->h); +} - break; - } +static void +rt_feed_by_trie(void *data) +{ + struct rt_table_export_hook *c = data; + rt_feed_block block = {}; - if (!new) + RT_LOCKED(RT_PUB(SKIP_BACK(struct rtable_private, exporter, c->table)), tab) { - if (!old) - goto drop_withdraw; - if (!net->routes) - fib_delete(&tab->fib, net); + ASSERT_DIE(c->walk_state); + struct f_trie_walk_state *ws = c->walk_state; - return 1; - } + ASSERT(atomic_load_explicit(&c->h.export_state, memory_order_relaxed) == TES_FEEDING); - struct channel_limit *l = &c->rx_limit; - if (l->action && !old) - { - if (tab->rt_count >= l->limit) - channel_notify_limit(c, l, PLD_RX, tab->rt_count); + do { + if (!c->walk_last.type) + continue; - if (l->state == PLS_BLOCKED) - { - /* Required by rte_trace_in() */ - new->net = net; + net *n = net_find(tab, &c->walk_last); + if (!n) + continue; + + if (atomic_load_explicit(&c->h.export_state, memory_order_acquire) != TES_FEEDING) + RT_RETURN(tab); - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - goto drop_update; + if (!rt_prepare_feed(c, n, &block)) + { + RT_UNLOCK(tab); + rt_process_feed(c, &block); + rt_send_export_event(&c->h); + return; } } + while (trie_walk_next(ws, &c->walk_last)); - /* Insert the new rte */ - rte *e = rte_do_cow(new); - e->flags |= REF_COW; - e->net = net; - e->sender = c; - e->lastmod = current_time(); - e->next = *pos; - *pos = e; - tab->rt_count++; - return 1; + rt_unlock_trie(tab, c->walk_lock); + c->walk_lock = NULL; -drop_update: - c->stats.imp_updates_received++; - c->stats.imp_updates_ignored++; - rte_free(new); + mb_free(c->walk_state); + c->walk_state = NULL; - if (!net->routes) - fib_delete(&tab->fib, net); + c->walk_last.type = 0; - return 0; + } -drop_withdraw: - c->stats.imp_withdraws_received++; - c->stats.imp_withdraws_ignored++; - return 0; + rt_process_feed(c, &block); + rt_feed_done(&c->h); } -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_table_export_hook *c = data; + rt_feed_block block = {}; + net *n; - if (!c->reload_active) + RT_LOCKED(RT_PUB(SKIP_BACK(struct rtable_private, exporter, c->table)), tab) { - FIB_ITERATE_INIT(fit, &tab->fib); - c->reload_active = 1; - } - - do { - for (rte *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_update2(c, e->net->n.addr, rte_do_cow(e), e->src); - } + ASSERT_DIE(atomic_load_explicit(&c->h.export_state, memory_order_relaxed) == TES_FEEDING); + ASSERT_DIE(c->h.req->addr_mode == TE_ADDR_EQUAL); - 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; + if (n = net_find(tab, c->h.req->addr)) + ASSERT_DIE(rt_prepare_feed(c, n, &block)); } - while (c->reload_next_rte); - c->reload_active = 0; - return 1; -} + if (n) + rt_process_feed(c, &block); -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_feed_done(&c->h); } -void -rt_prune_sync(rtable *t, int all) +static void +rt_feed_for(void *data) { - struct fib_iterator fit; + struct rt_table_export_hook *c = data; + rt_feed_block block = {}; + net *n; - FIB_ITERATE_INIT(&fit, &t->fib); - -again: - FIB_ITERATE_START(&t->fib, &fit, net, n) + RT_LOCKED(RT_PUB(SKIP_BACK(struct rtable_private, exporter, c->table)), tab) { - rte *e, **ee = &n->routes; - - while (e = *ee) - { - if (all || (e->flags & (REF_STALE | REF_DISCARD))) - { - *ee = e->next; - rte_free_quick(e); - t->rt_count--; - } - else - ee = &e->next; - } + ASSERT_DIE(atomic_load_explicit(&c->h.export_state, memory_order_relaxed) == TES_FEEDING); + ASSERT_DIE(c->h.req->addr_mode == TE_ADDR_FOR); - if (all || !n->routes) - { - FIB_ITERATE_PUT(&fit); - fib_delete(&t->fib, n); - goto again; - } + if (n = net_route(tab, c->h.req->addr)) + ASSERT_DIE(rt_prepare_feed(c, n, &block)); } - FIB_ITERATE_END; + + if (n) + rt_process_feed(c, &block); + + rt_feed_done(&c->h); } /* - * Export table + * Import table */ -int -rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, int refeed) +void channel_reload_export_bulk(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) { - struct rtable *tab = c->out_table; - struct rte_src *src; - rte *old, **pos; - net *net; - - if (new) - { - net = net_get(tab, n); - src = new->src; - - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - } - else - { - net = net_find(tab, n); - src = old0->src; - - if (!net) - goto drop_withdraw; - } + struct channel *c = SKIP_BACK(struct channel, reload_req, req); - /* Find the old rte */ - for (pos = &net->routes; old = *pos; pos = &old->next) - if ((c->ra_mode != RA_ANY) || (old->src == src)) + for (uint i=0; i<count; i++) + if (feed[i]->sender == c->in_req.hook) { - if (new && rte_same(old, new)) - { - /* REF_STALE / REF_DISCARD not used in export table */ - /* - if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY)) - { - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); - return 1; - } - */ - - goto drop_update; - } + /* Strip the later attribute layers */ + rte new = *feed[i]; + while (new.attrs->next) + new.attrs = new.attrs->next; - /* Remove the old rte */ - *pos = old->next; - rte_free_quick(old); - tab->rt_count--; - - break; + /* And reload the route */ + rte_update(c, net, &new, new.src); } - - if (!new) - { - if (!old) - goto drop_withdraw; - - if (!net->routes) - fib_delete(&tab->fib, net); - - return 1; - } - - /* Insert the new rte */ - rte *e = rte_do_cow(new); - e->flags |= REF_COW; - e->net = net; - e->sender = c; - e->lastmod = current_time(); - e->next = *pos; - *pos = e; - tab->rt_count++; - return 1; - -drop_update: - return refeed; - -drop_withdraw: - return 0; } @@ -3432,7 +4330,41 @@ hc_delete_hostentry(struct hostcache *hc, pool *p, struct hostentry *he) } static void -rt_init_hostcache(rtable *tab) +hc_notify_dump_req(struct rt_export_request *req) +{ + debug(" Table %s (%p)\n", req->name, req); +} + +static void +hc_notify_export_one(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *first) +{ + struct hostcache *hc = SKIP_BACK(struct hostcache, req, req); + + /* No interest in this update, mark seen only */ + if (ev_active(&hc->update) || !trie_match_net(hc->trie, net)) + { + rpe_mark_seen_all(req->hook, first, NULL); + return; + } + + /* This net may affect some hostentries, check the actual change */ + rte *o = RTE_VALID_OR_NULL(first->old_best); + struct rte_storage *new_best = first->new_best; + + RPE_WALK(first, rpe, NULL) + { + rpe_mark_seen(req->hook, rpe); + new_best = rpe->new_best; + } + + /* Yes, something has actually changed. Do the hostcache update. */ + if (o != RTE_VALID_OR_NULL(new_best)) + ev_schedule_work(&hc->update); +} + + +static void +rt_init_hostcache(struct rtable_private *tab) { struct hostcache *hc = mb_allocz(tab->rp, sizeof(struct hostcache)); init_list(&hc->hostentries); @@ -3444,11 +4376,26 @@ rt_init_hostcache(rtable *tab) hc->lp = lp_new(tab->rp); hc->trie = f_new_trie(hc->lp, 0); + hc->update = (event) { + .hook = rt_update_hostcache, + .data = tab, + }; + + hc->req = (struct rt_export_request) { + .name = mb_sprintf(tab->rp, "%s.hcu.notifier", tab->name), + .list = &global_work_list, + .trace_routes = tab->config->debug, + .dump_req = hc_notify_dump_req, + .export_one = hc_notify_export_one, + }; + + rt_table_export_start_locked(tab, &hc->req); + tab->hostcache = hc; } static void -rt_free_hostcache(rtable *tab) +rt_free_hostcache(struct rtable_private *tab) { struct hostcache *hc = tab->hostcache; @@ -3470,16 +4417,6 @@ rt_free_hostcache(rtable *tab) */ } -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) { @@ -3493,32 +4430,31 @@ 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) - return rt->src->proto->rte_igp_metric(rt); + if (rt->src->owner->class->rte_igp_metric) + return rt->src->owner->class->rte_igp_metric(rt); return IGP_METRIC_UNKNOWN; } static int -rt_update_hostentry(rtable *tab, struct hostentry *he) +rt_update_hostentry(struct rtable_private *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; @@ -3527,11 +4463,14 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) net *n = net_route(tab, &he_addr); if (n) { - rte *e = n->routes; - rta *a = e->attrs; - pxlen = n->n.addr->pxlen; - - if (a->hostentry) + struct rte_storage *e = n->routes; + ea_list *a = e->rte.attrs; + u32 pref = rt_get_preference(&e->rte); + + for (struct rte_storage *ee = n->routes; ee; ee = ee->next) + if (rte_is_valid(&ee->rte) && + (rt_get_preference(&ee->rte) >= pref) && + ea_find(ee->rte.attrs, &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", @@ -3539,9 +4478,14 @@ 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) + pxlen = n->n.addr->pxlen; + + 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)) @@ -3554,12 +4498,10 @@ 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); + he->igp_metric = rt_get_igp_metric(&e->rte); } done: @@ -3571,9 +4513,24 @@ done: } static void -rt_update_hostcache(rtable *tab) +rt_update_hostcache(void *data) { + rtable **nhu_pending; + + RT_LOCKED((rtable *) data, tab) + { + struct hostcache *hc = tab->hostcache; + + if (rt_cork_check(&hc->update)) + { + rt_trace(tab, D_STATES, "Hostcache update corked"); + RT_RETURN(tab); + } + + /* Destination schedule map */ + nhu_pending = tmp_allocz(sizeof(rtable *) * rtable_max_id); + struct hostentry *he; node *n, *x; @@ -3591,14 +4548,18 @@ rt_update_hostcache(rtable *tab) } if (rt_update_hostentry(tab, he)) - rt_schedule_nhu(he->tab); + nhu_pending[he->tab->id] = he->tab; } + } - tab->hcu_scheduled = 0; + for (uint i=0; i<rtable_max_id; i++) + if (nhu_pending[i]) + RT_LOCKED(nhu_pending[i], dst) + rt_schedule_nhu(dst); } -struct hostentry * -rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep) +static struct hostentry * +rt_get_hostentry(struct rtable_private *tab, ip_addr a, ip_addr ll, rtable *dep) { struct hostentry *he; |