diff options
| -rw-r--r-- | filter/filter.c | 34 | ||||
| -rw-r--r-- | nest/config.Y | 4 | ||||
| -rw-r--r-- | nest/limit.h | 49 | ||||
| -rw-r--r-- | nest/proto.c | 708 | ||||
| -rw-r--r-- | nest/protocol.h | 123 | ||||
| -rw-r--r-- | nest/route.h | 147 | ||||
| -rw-r--r-- | nest/rt-show.c | 36 | ||||
| -rw-r--r-- | nest/rt-table.c | 907 | ||||
| -rw-r--r-- | proto/bgp/bgp.c | 24 | ||||
| -rw-r--r-- | proto/pipe/pipe.c | 50 | ||||
| -rw-r--r-- | proto/rpki/packets.c | 19 | ||||
| -rw-r--r-- | proto/rpki/rpki.c | 9 | ||||
| -rw-r--r-- | proto/static/static.c | 11 | ||||
| -rw-r--r-- | sysdep/unix/krt.c | 52 | ||||
| -rw-r--r-- | sysdep/unix/krt.h | 1 |
15 files changed, 1360 insertions, 814 deletions
diff --git a/filter/filter.c b/filter/filter.c index 6f1e6ea0..625b3ade 100644 --- a/filter/filter.c +++ b/filter/filter.c @@ -425,6 +425,23 @@ filter_commit(struct config *new, struct config *old) } } +void channel_filter_dump(const struct filter *f) +{ + if (f == FILTER_ACCEPT) + debug(" ALL"); + else if (f == FILTER_REJECT) + debug(" NONE"); + else if (f == FILTER_UNDEF) + debug(" UNDEF"); + else if (f->sym) { + ASSERT(f->sym->filter == f); + debug(" named filter %s", f->sym->name); + } else { + debug("\n"); + f_dump_line(f->root, 2); + } +} + void filters_dump_all(void) { struct symbol *sym; @@ -444,19 +461,10 @@ void filters_dump_all(void) struct channel *c; WALK_LIST(c, sym->proto->proto->channels) { debug(" Channel %s (%s) IMPORT", c->name, net_label[c->net_type]); - if (c->in_filter == FILTER_ACCEPT) - debug(" ALL\n"); - else if (c->in_filter == FILTER_REJECT) - debug(" NONE\n"); - else if (c->in_filter == FILTER_UNDEF) - debug(" UNDEF\n"); - else if (c->in_filter->sym) { - ASSERT(c->in_filter->sym->filter == c->in_filter); - debug(" named filter %s\n", c->in_filter->sym->name); - } else { - debug("\n"); - f_dump_line(c->in_filter->root, 2); - } + channel_filter_dump(c->in_filter); + debug(" EXPORT", c->name, net_label[c->net_type]); + channel_filter_dump(c->out_filter); + debug("\n"); } } } diff --git a/nest/config.Y b/nest/config.Y index 1bebc084..7d266bbc 100644 --- a/nest/config.Y +++ b/nest/config.Y @@ -851,8 +851,10 @@ CF_CLI(DUMP NEIGHBORS,,, [[Dump neighbor cache]]) { neigh_dump_all(); cli_msg(0, ""); } ; CF_CLI(DUMP ATTRIBUTES,,, [[Dump attribute cache]]) { rta_dump_all(); cli_msg(0, ""); } ; -CF_CLI(DUMP ROUTES,,, [[Dump routing table]]) +CF_CLI(DUMP ROUTES,,, [[Dump routes]]) { rt_dump_all(); cli_msg(0, ""); } ; +CF_CLI(DUMP TABLES,,, [[Dump table connections]]) +{ rt_dump_hooks_all(); cli_msg(0, ""); } ; CF_CLI(DUMP PROTOCOLS,,, [[Dump protocol information]]) { protos_dump_all(); cli_msg(0, ""); } ; CF_CLI(DUMP FILTER ALL,,, [[Dump all filters in linearized form]]) diff --git a/nest/limit.h b/nest/limit.h new file mode 100644 index 00000000..5838ad3b --- /dev/null +++ b/nest/limit.h @@ -0,0 +1,49 @@ +/* + * BIRD Internet Routing Daemon -- Limits + * + * (c) 1998--2000 Martin Mares <mj@ucw.cz> + * (c) 2021 Maria Matejka <mq@jmq.cz> + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _BIRD_LIMIT_H_ +#define _BIRD_LIMIT_H_ + +struct limit { + u32 max; + u32 count; + int (*action)(struct limit *, void *data); +}; + +static inline int limit_do_action(struct limit *l, void *data) +{ + return l->action ? l->action(l, data) : 1; +} + +static inline int limit_push(struct limit *l, void *data) +{ + if ((l->count >= l->max) && limit_do_action(l, data)) + return 1; + + l->count++; + return 0; +} + +static inline void limit_pop(struct limit *l) +{ + --l->count; +} + +static inline void limit_reset(struct limit *l) +{ + l->count = 0; +} + +static inline void limit_update(struct limit *l, void *data, u32 max) +{ + if (l->count > (l->max = max)) + limit_do_action(l, data); +} + +#endif diff --git a/nest/proto.c b/nest/proto.c index c7e25209..09582d2e 100644 --- a/nest/proto.c +++ b/nest/proto.c @@ -43,8 +43,7 @@ static int graceful_restart_state; static u32 graceful_restart_locks; static char *p_states[] = { "DOWN", "START", "UP", "STOP" }; -static char *c_states[] = { "DOWN", "START", "UP", "FLUSHING" }; -static char *e_states[] = { "DOWN", "FEEDING", "READY" }; +static char *c_states[] = { "DOWN", "START", "UP", "STOP", "RESTART" }; extern struct protocol proto_unix_iface; @@ -52,15 +51,17 @@ static void channel_request_reload(struct channel *c); static void proto_shutdown_loop(timer *); static void proto_rethink_goal(struct proto *p); static char *proto_state_name(struct proto *p); -static void channel_verify_limits(struct channel *c); -static inline void channel_reset_limit(struct channel_limit *l); - +static void channel_init_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf); +static void channel_update_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf); +static void channel_reset_limit(struct channel *c, struct limit *l, int dir); +static void channel_feed_end(struct channel *c); +static void channel_export_stopped(struct rt_export_request *req); static inline int proto_is_done(struct proto *p) { return (p->proto_state == PS_DOWN) && (p->active_channels == 0); } static inline int channel_is_active(struct channel *c) -{ return (c->channel_state == CS_START) || (c->channel_state == CS_UP); } +{ return (c->channel_state != CS_DOWN); } static inline int channel_reloadable(struct channel *c) { return c->proto->reload_routes && c->reloadable; } @@ -68,10 +69,46 @@ static inline int channel_reloadable(struct channel *c) static inline void channel_log_state_change(struct channel *c) { - if (c->export_state) - CD(c, "State changed to %s/%s", c_states[c->channel_state], e_states[c->export_state]); - else - CD(c, "State changed to %s", c_states[c->channel_state]); + CD(c, "State changed to %s", c_states[c->channel_state]); +} + +void +channel_import_log_state_change(struct rt_import_request *req, u8 state) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); + CD(c, "Channel import state changed to %s", rt_import_state_name(state)); +} + +void +channel_export_log_state_change(struct rt_export_request *req, u8 state) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + CD(c, "Channel export state changed to %s", rt_export_state_name(state)); + + switch (state) + { + case TES_FEEDING: + if (c->proto->feed_begin) + c->proto->feed_begin(c, !c->refeeding); + break; + case TES_READY: + channel_feed_end(c); + break; + } +} + +static void +channel_dump_import_req(struct rt_import_request *req) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); + debug(" Channel %s.%s import request %p\n", c->proto->name, c->name, req); +} + +static void +channel_dump_export_req(struct rt_export_request *req) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + debug(" Channel %s.%s export request %p\n", c->proto->name, c->name, req); } static void @@ -141,6 +178,15 @@ proto_find_channel_by_name(struct proto *p, const char *n) return NULL; } +rte * channel_preimport(struct rt_import_request *req, rte *new, rte *old); + +void rt_notify_optimal(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe); +void rt_notify_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe); +void rt_feed_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); +void rt_notify_accepted(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); +void rt_notify_merged(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); + + /** * proto_add_channel - connect protocol to a routing table * @p: protocol instance @@ -165,12 +211,14 @@ proto_add_channel(struct proto *p, struct channel_config *cf) c->channel = cf->channel; c->proto = p; c->table = cf->table->table; + rt_lock_table(c->table); c->in_filter = cf->in_filter; c->out_filter = cf->out_filter; - c->rx_limit = cf->rx_limit; - c->in_limit = cf->in_limit; - c->out_limit = cf->out_limit; + + channel_init_limit(c, &c->rx_limit, PLD_RX, &cf->rx_limit); + channel_init_limit(c, &c->in_limit, PLD_IN, &cf->in_limit); + channel_init_limit(c, &c->out_limit, PLD_OUT, &cf->out_limit); c->net_type = cf->net_type; c->ra_mode = cf->ra_mode; @@ -181,7 +229,6 @@ proto_add_channel(struct proto *p, struct channel_config *cf) c->rpki_reload = cf->rpki_reload; c->channel_state = CS_DOWN; - c->export_state = ES_DOWN; c->last_state_change = current_time(); c->reloadable = 1; @@ -203,6 +250,7 @@ proto_remove_channel(struct proto *p UNUSED, struct channel *c) CD(c, "Removed", c->name); + rt_unlock_table(c->table); rem_node(&c->n); mb_free(c); } @@ -223,7 +271,7 @@ proto_pause_channels(struct proto *p) struct channel *c; WALK_LIST(c, p->channels) if (!c->disabled && channel_is_active(c)) - channel_set_state(c, CS_START); + channel_set_state(c, CS_PAUSE); } static void @@ -232,7 +280,7 @@ proto_stop_channels(struct proto *p) struct channel *c; WALK_LIST(c, p->channels) if (!c->disabled && channel_is_active(c)) - channel_set_state(c, CS_FLUSHING); + channel_set_state(c, CS_STOP); } static void @@ -244,71 +292,6 @@ proto_remove_channels(struct proto *p) } static void -channel_schedule_feed(struct channel *c, int initial) -{ - // DBG("%s: Scheduling meal\n", p->name); - ASSERT(c->channel_state == CS_UP); - - c->export_state = ES_FEEDING; - c->refeeding = !initial; - - ev_schedule_work(c->feed_event); -} - -static void -channel_feed_loop(void *ptr) -{ - struct channel *c = ptr; - - if (c->export_state != ES_FEEDING) - return; - - /* Start feeding */ - if (!c->feed_active) - { - if (c->proto->feed_begin) - c->proto->feed_begin(c, !c->refeeding); - - c->refeed_pending = 0; - } - - // DBG("Feeding protocol %s continued\n", p->name); - if (!rt_feed_channel(c)) - { - ev_schedule_work(c->feed_event); - return; - } - - /* Reset export limit if the feed ended with acceptable number of exported routes */ - struct channel_limit *l = &c->out_limit; - if (c->refeeding && - (l->state == PLS_BLOCKED) && - (c->refeed_count <= l->limit) && - (c->export_stats.routes <= l->limit)) - { - log(L_INFO "Protocol %s resets route export limit (%u)", c->proto->name, l->limit); - channel_reset_limit(&c->out_limit); - - /* Continue in feed - it will process routing table again from beginning */ - c->refeed_count = 0; - ev_schedule_work(c->feed_event); - return; - } - - // DBG("Feeding protocol %s finished\n", p->name); - c->export_state = ES_READY; - channel_log_state_change(c); - - if (c->proto->feed_end) - c->proto->feed_end(c); - - /* Restart feeding */ - if (c->refeed_pending) - channel_request_feeding(c); -} - - -static void channel_roa_in_changed(struct rt_subscription *s) { struct channel *c = s->data; @@ -326,14 +309,12 @@ static void channel_roa_out_changed(struct rt_subscription *s) { struct channel *c = s->data; - int active = (c->export_state == ES_FEEDING); + CD(c, "Feeding triggered by RPKI change"); - CD(c, "Feeding triggered by RPKI change%s", active ? " - already active" : ""); + c->refeed_pending = 1; - if (!active) - channel_request_feeding(c); - else - c->refeed_pending = 1; + if (c->out_req.hook) + rt_stop_export(&c->out_req, channel_export_stopped); } /* Temporary code, subscriptions should be changed to resources */ @@ -445,33 +426,189 @@ channel_roa_unsubscribe_all(struct channel *c) } static void +channel_start_import(struct channel *c) +{ + if (c->in_req.hook) + { + log(L_WARN "%s.%s: Attempted to start channel's already started import", c->proto->name, c->name); + return; + } + + int nlen = strlen(c->name) + strlen(c->proto->name) + 2; + char *rn = mb_allocz(c->proto->pool, nlen); + bsprintf(rn, "%s.%s", c->proto->name, c->name); + + c->in_req = (struct rt_import_request) { + .name = rn, + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_dump_import_req, + .log_state_change = channel_import_log_state_change, + .preimport = channel_preimport, + .rte_modify = c->proto->rte_modify, + }; + + ASSERT(c->channel_state == CS_UP); + + channel_reset_limit(c, &c->rx_limit, PLD_RX); + channel_reset_limit(c, &c->in_limit, PLD_IN); + + memset(&c->import_stats, 0, sizeof(struct channel_import_stats)); + + DBG("%s.%s: Channel start import req=%p\n", c->proto->name, c->name, &c->in_req); + rt_request_import(c->table, &c->in_req); +} + +static void channel_start_export(struct channel *c) { + if (c->out_req.hook) + { + log(L_WARN "%s.%s: Attempted to start channel's already started export", c->proto->name, c->name); + return; + } + ASSERT(c->channel_state == CS_UP); - ASSERT(c->export_state == ES_DOWN); + int nlen = strlen(c->name) + strlen(c->proto->name) + 2; + char *rn = mb_allocz(c->proto->pool, nlen); + bsprintf(rn, "%s.%s", c->proto->name, c->name); + + c->out_req = (struct rt_export_request) { + .name = rn, + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_dump_export_req, + .log_state_change = channel_export_log_state_change, + }; + + bmap_init(&c->export_map, c->proto->pool, 1024); + bmap_init(&c->export_reject_map, c->proto->pool, 1024); + + channel_reset_limit(c, &c->out_limit, PLD_OUT); - channel_schedule_feed(c, 1); /* Sets ES_FEEDING */ + memset(&c->export_stats, 0, sizeof(struct channel_export_stats)); + + switch (c->ra_mode) { + case RA_OPTIMAL: + c->out_req.export_one = rt_notify_optimal; + break; + case RA_ANY: + c->out_req.export_one = rt_notify_any; + c->out_req.export_bulk = rt_feed_any; + break; + case RA_ACCEPTED: + c->out_req.export_bulk = rt_notify_accepted; + break; + case RA_MERGED: + c->out_req.export_bulk = rt_notify_merged; + break; + default: + bug("Unknown route announcement mode"); + } + + DBG("%s.%s: Channel start export req=%p\n", c->proto->name, c->name, &c->out_req); + rt_request_export(c->table, &c->out_req); } static void -channel_stop_export(struct channel *c) +channel_check_stopped(struct channel *c) { - /* Need to abort feeding */ - if (c->export_state == ES_FEEDING) - rt_feed_channel_abort(c); + switch (c->channel_state) + { + case CS_STOP: + if (c->out_req.hook || c->in_req.hook) + return; + + channel_set_state(c, CS_DOWN); + ev_schedule(c->proto->event); + + break; + case CS_PAUSE: + if (c->out_req.hook) + return; + + channel_set_state(c, CS_START); + break; + default: + bug("Stopped channel in a bad state: %d", c->channel_state); + } + + DBG("%s.%s: Channel requests/hooks stopped (in state %s)\n", c->proto->name, c->name, c_states[c->channel_state]); +} + +void +channel_import_stopped(struct rt_import_request *req) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); + + req->hook = NULL; - c->export_state = ES_DOWN; - c->export_stats.routes = 0; - bmap_reset(&c->export_map, 1024); - bmap_reset(&c->export_reject_map, 1024); + if (c->in_table) + rt_prune_sync(c->in_table, 1); + + mb_free(c->in_req.name); + c->in_req.name = NULL; + + channel_check_stopped(c); } +static void +channel_export_stopped(struct rt_export_request *req) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + + /* The hook has already stopped */ + req->hook = NULL; + + if (c->refeed_pending) + { + c->refeeding = 1; + c->refeed_pending = 0; + rt_request_export(c->table, req); + return; + } + + /* Free the routes from out_table */ + if (c->out_table) + rt_prune_sync(c->out_table, 1); + + mb_free(c->out_req.name); + c->out_req.name = NULL; + + channel_check_stopped(c); +} + +static void +channel_feed_end(struct channel *c) +{ + struct rt_export_request *req = &c->out_req; + + /* Reset export limit if the feed ended with acceptable number of exported routes */ + struct limit *l = &c->out_limit; + if (c->refeeding && + (c->limit_active & (1 << PLD_OUT)) && + (c->refeed_count <= l->max) && + (l->count <= l->max)) + { + log(L_INFO "Protocol %s resets route export limit (%u)", c->proto->name, l->max); + + c->refeed_pending = 1; + rt_stop_export(req, channel_export_stopped); + return; + } + + if (c->proto->feed_end) + c->proto->feed_end(c); + + if (c->refeed_pending) + rt_stop_export(req, channel_export_stopped); + else + c->refeeding = 0; +} /* Called by protocol for reload from in_table */ void channel_schedule_reload(struct channel *c) { - ASSERT(c->channel_state == CS_UP); + ASSERT(c->in_req.hook); rt_reload_channel_abort(c); ev_schedule_work(c->reload_event); @@ -497,23 +634,6 @@ channel_reload_loop(void *ptr) channel_request_reload(c); } -static void -channel_reset_import(struct channel *c) -{ - /* Need to abort feeding */ - ev_postpone(c->reload_event); - rt_reload_channel_abort(c); - - rt_prune_sync(c->in_table, 1); -} - -static void -channel_reset_export(struct channel *c) -{ - /* Just free the routes */ - rt_prune_sync(c->out_table, 1); -} - /* Called by protocol to activate in_table */ void channel_setup_in_table(struct channel *c) @@ -545,22 +665,11 @@ channel_setup_out_table(struct channel *c) static void channel_do_start(struct channel *c) { - rt_lock_table(c->table); - add_tail(&c->table->channels, &c->table_node); c->proto->active_channels++; - c->feed_event = ev_new_init(c->proto->pool, channel_feed_loop, c); - - bmap_init(&c->export_map, c->proto->pool, 1024); - bmap_init(&c->export_reject_map, c->proto->pool, 1024); - memset(&c->export_stats, 0, sizeof(struct export_stats)); - memset(&c->import_stats, 0, sizeof(struct import_stats)); - - channel_reset_limit(&c->rx_limit); - channel_reset_limit(&c->in_limit); - channel_reset_limit(&c->out_limit); - CALL(c->channel->start, c); + + channel_start_import(c); } static void @@ -575,9 +684,31 @@ channel_do_up(struct channel *c) } static void -channel_do_flush(struct channel *c) +channel_do_pause(struct channel *c) +{ + /* Need to abort feeding */ + if (c->reload_event) + { + ev_postpone(c->reload_event); + rt_reload_channel_abort(c); + } + + /* Stop export */ + if (c->out_req.hook) + rt_stop_export(&c->out_req, channel_export_stopped); + + channel_roa_unsubscribe_all(c); + + bmap_free(&c->export_map); + bmap_free(&c->export_reject_map); +} + +static void +channel_do_stop(struct channel *c) { - rt_schedule_prune(c->table); + /* Stop import */ + if (c->in_req.hook) + rt_stop_import(&c->in_req, channel_import_stopped); c->gr_wait = 0; if (c->gr_lock) @@ -586,30 +717,21 @@ channel_do_flush(struct channel *c) CALL(c->channel->shutdown, c); /* This have to be done in here, as channel pool is freed before channel_do_down() */ - bmap_free(&c->export_map); - bmap_free(&c->export_reject_map); c->in_table = NULL; c->reload_event = NULL; c->out_table = NULL; - - channel_roa_unsubscribe_all(c); } static void channel_do_down(struct channel *c) { - ASSERT(!c->feed_active && !c->reload_active); + ASSERT(!c->reload_active); - rem_node(&c->table_node); - rt_unlock_table(c->table); c->proto->active_channels--; - if ((c->import_stats.routes + c->import_stats.filtered) != 0) - log(L_ERR "%s: Channel %s is down but still has some routes", c->proto->name, c->name); - // bmap_free(&c->export_map); - memset(&c->import_stats, 0, sizeof(struct import_stats)); - memset(&c->export_stats, 0, sizeof(struct export_stats)); + memset(&c->import_stats, 0, sizeof(struct channel_import_stats)); + memset(&c->export_stats, 0, sizeof(struct channel_export_stats)); c->in_table = NULL; c->reload_event = NULL; @@ -628,7 +750,6 @@ void channel_set_state(struct channel *c, uint state) { uint cs = c->channel_state; - uint es = c->export_state; DBG("%s reporting channel %s state transition %s -> %s\n", c->proto->name, c->name, c_states[cs], c_states[state]); if (state == cs) @@ -640,20 +761,11 @@ channel_set_state(struct channel *c, uint state) switch (state) { case CS_START: - ASSERT(cs == CS_DOWN || cs == CS_UP); + ASSERT(cs == CS_DOWN || cs == CS_PAUSE); if (cs == CS_DOWN) channel_do_start(c); - if (es != ES_DOWN) - channel_stop_export(c); - - if (c->in_table && (cs == CS_UP)) - channel_reset_import(c); - - if (c->out_table && (cs == CS_UP)) - channel_reset_export(c); - break; case CS_UP: @@ -668,23 +780,24 @@ channel_set_state(struct channel *c, uint state) channel_do_up(c); break; - case CS_FLUSHING: - ASSERT(cs == CS_START || cs == CS_UP); + case CS_PAUSE: + ASSERT(cs == CS_UP); - if (es != ES_DOWN) - channel_stop_export(c); + if (cs == CS_UP) + channel_do_pause(c); + break; - if (c->in_table && (cs == CS_UP)) - channel_reset_import(c); + case CS_STOP: + ASSERT(cs == CS_UP || cs == CS_START || cs == CS_PAUSE); - if (c->out_table && (cs == CS_UP)) - channel_reset_export(c); + if (cs == CS_UP) + channel_do_pause(c); - channel_do_flush(c); + channel_do_stop(c); break; case CS_DOWN: - ASSERT(cs == CS_FLUSHING); + ASSERT(cs == CS_STOP); channel_do_down(c); break; @@ -709,35 +822,16 @@ channel_set_state(struct channel *c, uint state) void channel_request_feeding(struct channel *c) { - ASSERT(c->channel_state == CS_UP); - - CD(c, "Feeding requested"); - - /* Do nothing if we are still waiting for feeding */ - if (c->export_state == ES_DOWN) - return; - - /* If we are already feeding, we want to restart it */ - if (c->export_state == ES_FEEDING) - { - /* Unless feeding is in initial state */ - if (!c->feed_active) - return; + ASSERT(c->out_req.hook); - rt_feed_channel_abort(c); - } - - /* Track number of exported routes during refeed */ - c->refeed_count = 0; - - channel_schedule_feed(c, 0); /* Sets ES_FEEDING */ - channel_log_state_change(c); + c->refeed_pending = 1; + rt_stop_export(&c->out_req, channel_export_stopped); } static void channel_request_reload(struct channel *c) { - ASSERT(c->channel_state == CS_UP); + ASSERT(c->in_req.hook); ASSERT(channel_reloadable(c)); CD(c, "Reload requested"); @@ -748,8 +842,8 @@ channel_request_reload(struct channel *c) * Should this be done before reload_routes() hook? * Perhaps, but routes are updated asynchronously. */ - channel_reset_limit(&c->rx_limit); - channel_reset_limit(&c->in_limit); + channel_reset_limit(c, &c->rx_limit, PLD_RX); + channel_reset_limit(c, &c->in_limit, PLD_IN); } const struct channel_class channel_basic = { @@ -852,19 +946,19 @@ channel_reconfigure(struct channel *c, struct channel_config *cf) /* Reconfigure channel fields */ c->in_filter = cf->in_filter; c->out_filter = cf->out_filter; - c->rx_limit = cf->rx_limit; - c->in_limit = cf->in_limit; - c->out_limit = cf->out_limit; + + channel_update_limit(c, &c->rx_limit, PLD_RX, &cf->rx_limit); + channel_update_limit(c, &c->in_limit, PLD_IN, &cf->in_limit); + channel_update_limit(c, &c->out_limit, PLD_OUT, &cf->out_limit); // c->ra_mode = cf->ra_mode; c->merge_limit = cf->merge_limit; c->preference = cf->preference; c->debug = cf->debug; + c->in_req.trace_routes = c->out_req.trace_routes = c->debug | c->proto->debug; c->in_keep_filtered = cf->in_keep_filtered; c->rpki_reload = cf->rpki_reload; - channel_verify_limits(c); - /* Execute channel-specific reconfigure hook */ if (c->channel->reconfigure && !c->channel->reconfigure(c, cf, &import_changed, &export_changed)) return 0; @@ -976,8 +1070,8 @@ proto_event(void *ptr) if (proto_is_done(p)) { - if (p->proto->cleanup) - p->proto->cleanup(p); + rfree(p->pool); + p->pool = NULL; p->active = 0; proto_log_state_change(p); @@ -1529,7 +1623,7 @@ graceful_restart_done(timer *t UNUSED) WALK_LIST(c, p->channels) { /* Resume postponed export of routes */ - if ((c->channel_state == CS_UP) && c->gr_wait && c->proto->rt_notify) + if ((c->channel_state == CS_UP) && c->gr_wait && p->rt_notify) channel_start_export(c); /* Cleanup */ @@ -1619,7 +1713,11 @@ protos_dump_all(void) struct proto *p; WALK_LIST(p, proto_list) { - debug(" protocol %s state %s\n", p->name, p_states[p->proto_state]); +#define DPF(x) (p->x ? " " #x : "") + debug(" protocol %s (%p) state %s with %d active channels flags: %s%s%s%s%s\n", + p->name, p, p_states[p->proto_state], p->active_channels, + DPF(disabled), DPF(active), DPF(do_start), DPF(do_stop), DPF(reconfiguring)); +#undef DPF struct channel *c; WALK_LIST(c, p->channels) @@ -1629,6 +1727,9 @@ protos_dump_all(void) debug("\tInput filter: %s\n", filter_name(c->in_filter)); if (c->out_filter) debug("\tOutput filter: %s\n", filter_name(c->out_filter)); + debug("\tChannel state: %s/%s/%s\n", c_states[c->channel_state], + c->in_req.hook ? rt_import_state_name(rt_import_get_state(c->in_req.hook)) : "-", + c->out_req.hook ? rt_export_state_name(rt_export_get_state(c->out_req.hook)) : "-"); } if (p->proto->dump && (p->proto_state != PS_DOWN)) @@ -1785,88 +1886,104 @@ proto_set_message(struct proto *p, char *msg, int len) } -static const char * -channel_limit_name(struct channel_limit *l) -{ - const char *actions[] = { - [PLA_WARN] = "warn", - [PLA_BLOCK] = "block", - [PLA_RESTART] = "restart", - [PLA_DISABLE] = "disable", - }; +static const char * channel_limit_name[] = { + [PLA_WARN] = "warn", + [PLA_BLOCK] = "block", + [PLA_RESTART] = "restart", + [PLA_DISABLE] = "disable", +}; - return actions[l->action]; -} -/** - * channel_notify_limit: notify about limit hit and take appropriate action - * @c: channel - * @l: limit being hit - * @dir: limit direction (PLD_*) - * @rt_count: the number of routes - * - * The function is called by the route processing core when limit @l - * is breached. It activates the limit and tooks appropriate action - * according to @l->action. - */ -void -channel_notify_limit(struct channel *c, struct channel_limit *l, int dir, u32 rt_count) +static void +channel_log_limit(struct channel *c, struct limit *l, int dir) { const char *dir_name[PLD_MAX] = { "receive", "import" , "export" }; - const byte dir_down[PLD_MAX] = { PDC_RX_LIMIT_HIT, PDC_IN_LIMIT_HIT, PDC_OUT_LIMIT_HIT }; - struct proto *p = c->proto; + log(L_WARN "Channel %s.%s hits route %s limit (%d), action: %s", + c->proto->name, c->name, dir_name[dir], l->max, channel_limit_name[c->limit_actions[dir]]); +} - if (l->state == PLS_BLOCKED) +static void +channel_activate_limit(struct channel *c, struct limit *l, int dir) +{ + if (c->limit_active & (1 << dir)) return; - /* For warning action, we want the log message every time we hit the limit */ - if (!l->state || ((l->action == PLA_WARN) && (rt_count == l->limit))) - log(L_WARN "Protocol %s hits route %s limit (%d), action: %s", - p->name, dir_name[dir], l->limit, channel_limit_name(l)); + c->limit_active |= (1 << dir); + channel_log_limit(c, l, dir); +} - switch (l->action) - { - case PLA_WARN: - l->state = PLS_ACTIVE; - break; +static int +channel_limit_warn(struct limit *l, void *data) +{ + struct channel_limit_data *cld = data; + struct channel *c = cld->c; + int dir = cld->dir; - case PLA_BLOCK: - l->state = PLS_BLOCKED; - break; + channel_log_limit(c, l, dir); - case PLA_RESTART: - case PLA_DISABLE: - l->state = PLS_BLOCKED; - if (p->proto_state == PS_UP) - proto_schedule_down(p, l->action == PLA_RESTART, dir_down[dir]); - break; - } + return 0; } -static void -channel_verify_limits(struct channel *c) +static int +channel_limit_block(struct limit *l, void *data) +{ + struct channel_limit_data *cld = data; + struct channel *c = cld->c; + int dir = cld->dir; + + channel_activate_limit(c, l, dir); + + return 1; +} + +static const byte chl_dir_down[PLD_MAX] = { PDC_RX_LIMIT_HIT, PDC_IN_LIMIT_HIT, PDC_OUT_LIMIT_HIT }; + +static int +channel_limit_down(struct limit *l, void *data) { - struct channel_limit *l; - u32 all_routes = c->import_stats.routes + c->import_stats.filtered; + struct channel_limit_data *cld = data; + struct channel *c = cld->c; + struct proto *p = c->proto; + int dir = cld->dir; - l = &c->rx_limit; - if (l->action && (all_routes > l->limit)) - channel_notify_limit(c, l, PLD_RX, all_routes); + channel_activate_limit(c, l, dir); - l = &c->in_limit; - if (l->action && (c->import_stats.routes > l->limit)) - channel_notify_limit(c, l, PLD_IN, c->import_stats.routes); + if (p->proto_state == PS_UP) + proto_schedule_down(p, c->limit_actions[dir] == PLA_RESTART, chl_dir_down[dir]); - l = &c->out_limit; - if (l->action && (c->export_stats.routes > l->limit)) - channel_notify_limit(c, l, PLD_OUT, c->export_stats.routes); + return 1; } -static inline void -channel_reset_limit(struct channel_limit *l) +static int (*channel_limit_action[])(struct limit *, void *) = { + [PLA_NONE] = NULL, + [PLA_WARN] = channel_limit_warn, + [PLA_BLOCK] = channel_limit_block, + [PLA_RESTART] = channel_limit_down, + [PLA_DISABLE] = channel_limit_down, +}; + +static void +channel_update_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf) +{ + l->action = channel_limit_action[cf->action]; + c->limit_actions[dir] = cf->action; + + struct channel_limit_data cld = { .c = c, .dir = dir }; + limit_update(l, &cld, cf->action ? cf->limit : ~((u32) 0)); +} + +static void +channel_init_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf) +{ + channel_reset_limit(c, l, dir); + channel_update_limit(c, l, dir, cf); +} + +static void +channel_reset_limit(struct channel *c, struct limit *l, int dir) { - if (l->action) - l->state = PLS_INITIAL; + limit_reset(l); + c->limit_active &= ~(1 << dir); } static inline void @@ -1918,8 +2035,6 @@ proto_do_down(struct proto *p) { p->down_code = 0; neigh_prune(); - rfree(p->pool); - p->pool = NULL; /* Shutdown is finished in the protocol event */ if (proto_is_done(p)) @@ -2014,39 +2129,58 @@ proto_state_name(struct proto *p) static void channel_show_stats(struct channel *c) { - struct import_stats *is = &c->import_stats; - struct export_stats *es = &c->export_stats; + struct channel_import_stats *ch_is = &c->import_stats; + struct channel_export_stats *ch_es = &c->export_stats; + struct rt_import_stats *rt_is = c->in_req.hook ? &c->in_req.hook->stats : NULL; + struct rt_export_stats *rt_es = c->out_req.hook ? &c->out_req.hook->stats : NULL; + +#define SON(ie, item) ((ie) ? (ie)->item : 0) +#define SCI(item) SON(ch_is, item) +#define SCE(item) SON(ch_es, item) +#define SRI(item) SON(rt_is, item) +#define SRE(item) SON(rt_es, item) + + u32 rx_routes = c->rx_limit.count; + u32 in_routes = c->in_limit.count; + u32 out_routes = c->out_limit.count; if (c->in_keep_filtered) cli_msg(-1006, " Routes: %u imported, %u filtered, %u exported, %u preferred", - is->routes, is->filtered, es->routes, is->pref); + in_routes, (rx_routes - in_routes), out_routes, SRI(pref)); else cli_msg(-1006, " Routes: %u imported, %u exported, %u preferred", - is->routes, es->routes, is->pref); - - cli_msg(-1006, " Route change stats: received rejected filtered ignored accepted"); - cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u", - is->updates_received, is->updates_invalid, - is->updates_filtered, is->updates_ignored, - is->updates_accepted); - cli_msg(-1006, " Import withdraws: %10u %10u --- %10u %10u", - is->withdraws_received, is->withdraws_invalid, - is->withdraws_ignored, is->withdraws_accepted); - cli_msg(-1006, " Export updates: %10u %10u %10u --- %10u", - es->updates_received, es->updates_rejected, - es->updates_filtered, es->updates_accepted); - cli_msg(-1006, " Export withdraws: %10u --- --- --- %10u", - es->withdraws_received, es->withdraws_accepted); + in_routes, out_routes, SRI(pref)); + + cli_msg(-1006, " Route change stats: received rejected filtered ignored RX limit IN limit accepted"); + cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u %10u %10u", + SCI(updates_received), SCI(updates_invalid), + SCI(updates_filtered), SRI(updates_ignored), + SCI(updates_limited_rx), SCI(updates_limited_in), + SRI(updates_accepted)); + cli_msg(-1006, " Import withdraws: %10u %10u --- %10u --- %10u", + SCI(withdraws_received), SCI(withdraws_invalid), + SRI(withdraws_ignored), SRI(withdraws_accepted)); + cli_msg(-1006, " Export updates: %10u %10u %10u --- %10u %10u", + SRE(updates_received), SCE(updates_rejected), + SCE(updates_filtered), SCE(updates_limited), SCE(updates_accepted)); + cli_msg(-1006, " Export withdraws: %10u --- --- --- ---%10u", + SRE(withdraws_received), SCE(withdraws_accepted)); + +#undef SRI +#undef SRE +#undef SCI +#undef SCE +#undef SON } void -channel_show_limit(struct channel_limit *l, const char *dsc) +channel_show_limit(struct limit *l, const char *dsc, int active, int action) { if (!l->action) return; - cli_msg(-1006, " %-16s%d%s", dsc, l->limit, l->state ? " [HIT]" : ""); - cli_msg(-1006, " Action: %s", channel_limit_name(l)); + cli_msg(-1006, " %-16s%d%s", dsc, l->max, active ? " [HIT]" : ""); + cli_msg(-1006, " Action: %s", channel_limit_name[action]); } void @@ -2054,6 +2188,8 @@ channel_show_info(struct channel *c) { cli_msg(-1006, " Channel %s", c->name); cli_msg(-1006, " State: %s", c_states[c->channel_state]); + cli_msg(-1006, " Import state: %s", rt_import_state_name(rt_import_get_state(c->in_req.hook))); + cli_msg(-1006, " Export state: %s", rt_export_state_name(rt_export_get_state(c->out_req.hook))); cli_msg(-1006, " Table: %s", c->table->name); cli_msg(-1006, " Preference: %d", c->preference); cli_msg(-1006, " Input filter: %s", filter_name(c->in_filter)); @@ -2064,9 +2200,9 @@ channel_show_info(struct channel *c) c->gr_lock ? " pending" : "", c->gr_wait ? " waiting" : ""); - channel_show_limit(&c->rx_limit, "Receive limit:"); - channel_show_limit(&c->in_limit, "Import limit:"); - channel_show_limit(&c->out_limit, "Export limit:"); + channel_show_limit(&c->rx_limit, "Receive limit:", c->limit_active & (1 << PLD_RX), c->limit_actions[PLD_RX]); + channel_show_limit(&c->in_limit, "Import limit:", c->limit_active & (1 << PLD_IN), c->limit_actions[PLD_IN]); + channel_show_limit(&c->out_limit, "Export limit:", c->limit_active & (1 << PLD_OUT), c->limit_actions[PLD_OUT]); if (c->channel_state != CS_DOWN) channel_show_stats(c); diff --git a/nest/protocol.h b/nest/protocol.h index 4c87c72d..7447cbf0 100644 --- a/nest/protocol.h +++ b/nest/protocol.h @@ -13,6 +13,7 @@ #include "lib/resource.h" #include "lib/event.h" #include "nest/route.h" +#include "nest/limit.h" #include "conf/conf.h" struct iface; @@ -76,7 +77,6 @@ struct protocol { void (*dump)(struct proto *); /* Debugging dump */ int (*start)(struct proto *); /* Start the instance */ int (*shutdown)(struct proto *); /* Stop the instance */ - void (*cleanup)(struct proto *); /* Called after shutdown when protocol became hungry/down */ void (*get_status)(struct proto *, byte *buf); /* Get instance status (for `show protocols' command) */ void (*get_route_info)(struct rte *, byte *buf); /* Get route information (for `show route' command) */ int (*get_attr)(const struct eattr *, byte *buf, int buflen); /* ASCIIfy dynamic attribute (returns GA_*) */ @@ -132,33 +132,6 @@ struct proto_config { }; /* Protocol statistics */ -struct import_stats { - /* Import - from protocol to core */ - u32 routes; /* Number of routes successfully imported to the (adjacent) routing table */ - u32 filtered; /* Number of routes rejected in import filter but kept in the routing table */ - u32 pref; /* Number of routes selected as best in the (adjacent) routing table */ - u32 updates_received; /* Number of route updates received */ - u32 updates_invalid; /* Number of route updates rejected as invalid */ - u32 updates_filtered; /* Number of route updates rejected by filters */ - u32 updates_ignored; /* Number of route updates rejected as already in route table */ - u32 updates_accepted; /* Number of route updates accepted and imported */ - u32 withdraws_received; /* Number of route withdraws received */ - u32 withdraws_invalid; /* Number of route withdraws rejected as invalid */ - u32 withdraws_ignored; /* Number of route withdraws rejected as already not in route table */ - u32 withdraws_accepted; /* Number of route withdraws accepted and processed */ -}; - -struct export_stats { - /* Export - from core to protocol */ - u32 routes; /* Number of routes successfully exported to the protocol */ - u32 updates_received; /* Number of route updates received */ - u32 updates_rejected; /* Number of route updates rejected by protocol */ - u32 updates_filtered; /* Number of route updates rejected by filters */ - u32 updates_accepted; /* Number of route updates accepted and exported */ - u32 withdraws_received; /* Number of route withdraws received */ - u32 withdraws_accepted; /* Number of route withdraws accepted and processed */ -}; - struct proto { node n; /* Node in global proto_list */ struct protocol *proto; /* Protocol */ @@ -277,7 +250,7 @@ void channel_graceful_restart_unlock(struct channel *c); #define DEFAULT_GR_WAIT 240 -void channel_show_limit(struct channel_limit *l, const char *dsc); +void channel_show_limit(struct limit *l, const char *dsc, int active, int action); void channel_show_info(struct channel *c); void channel_cmd_debug(struct channel *c, uint mask); @@ -432,19 +405,30 @@ extern struct proto_config *cf_dev_proto; #define PLA_RESTART 4 /* Force protocol restart */ #define PLA_DISABLE 5 /* Shutdown and disable protocol */ -#define PLS_INITIAL 0 /* Initial limit state after protocol start */ -#define PLS_ACTIVE 1 /* Limit was hit */ -#define PLS_BLOCKED 2 /* Limit is active and blocking new routes */ - struct channel_limit { u32 limit; /* Maximum number of prefixes */ u8 action; /* Action to take (PLA_*) */ - u8 state; /* State of limit (PLS_*) */ }; -void channel_notify_limit(struct channel *c, struct channel_limit *l, int dir, u32 rt_count); +struct channel_limit_data { + struct channel *c; + int dir; +}; + +#define CLP__RX(_c) (&(_c)->rx_limit) +#define CLP__IN(_c) (&(_c)->in_limit) +#define CLP__OUT(_c) (&(_c)->out_limit) +#if 0 +#define CHANNEL_LIMIT_LOG(_c, _dir, _op) log(L_TRACE "%s.%s: %s limit %s %u", (_c)->proto->name, (_c)->name, #_dir, _op, (CLP__##_dir(_c))->count) +#else +#define CHANNEL_LIMIT_LOG(_c, _dir, _op) +#endif + +#define CHANNEL_LIMIT_PUSH(_c, _dir) ({ CHANNEL_LIMIT_LOG(_c, _dir, "push from"); struct channel_limit_data cld = { .c = (_c), .dir = PLD_##_dir }; limit_push(CLP__##_dir(_c), &cld); }) +#define CHANNEL_LIMIT_POP(_c, _dir) ({ limit_pop(CLP__##_dir(_c)); CHANNEL_LIMIT_LOG(_c, _dir, "pop to"); }) + /* * Channels */ @@ -486,6 +470,7 @@ struct channel_config { struct proto_config *parent; /* Where channel is defined (proto or template) */ struct rtable_config *table; /* Table we're attached to */ const struct filter *in_filter, *out_filter; /* Attached filters */ + struct channel_limit rx_limit; /* Limit for receiving routes from protocol (relevant when in_keep_filtered is active) */ struct channel_limit in_limit; /* Limit for importing routes from protocol */ @@ -502,7 +487,6 @@ struct channel_config { struct channel { node n; /* Node in proto->channels */ - node table_node; /* Node in table->channels */ const char *name; /* Channel name (may be NULL) */ const struct channel_class *channel; @@ -513,14 +497,36 @@ struct channel { const struct filter *out_filter; /* Output filter */ struct bmap export_map; /* Keeps track which routes were really exported */ struct bmap export_reject_map; /* Keeps track which routes were rejected by export filter */ - struct channel_limit rx_limit; /* Receive limit (for in_keep_filtered) */ - struct channel_limit in_limit; /* Input limit */ - struct channel_limit out_limit; /* Output limit */ - struct event *feed_event; /* Event responsible for feeding */ - struct fib_iterator feed_fit; /* Routing table iterator used during feeding */ - struct import_stats import_stats; /* Import statistics */ - struct export_stats export_stats; /* Export statistics */ + struct limit rx_limit; /* Receive limit (for in_keep_filtered) */ + struct limit in_limit; /* Input limit */ + struct limit out_limit; /* Output limit */ + + u8 limit_actions[PLD_MAX]; /* Limit actions enum */ + u8 limit_active; /* Flags for active limits */ + + struct channel_import_stats { + /* Import - from protocol to core */ + u32 updates_received; /* Number of route updates received */ + u32 updates_invalid; /* Number of route updates rejected as invalid */ + u32 updates_filtered; /* Number of route updates rejected by filters */ + u32 updates_limited_rx; /* Number of route updates exceeding the rx_limit */ + u32 updates_limited_in; /* Number of route updates exceeding the in_limit */ + u32 withdraws_received; /* Number of route withdraws received */ + u32 withdraws_invalid; /* Number of route withdraws rejected as invalid */ + } import_stats; + + struct channel_export_stats { + /* Export - from core to protocol */ + u32 updates_rejected; /* Number of route updates rejected by protocol */ + u32 updates_filtered; /* Number of route updates rejected by filters */ + u32 updates_accepted; /* Number of route updates accepted and exported */ + u32 updates_limited; /* Number of route updates exceeding the out_limit */ + u32 withdraws_accepted; /* Number of route withdraws accepted and processed */ + } export_stats; + + struct rt_import_request in_req; /* Table import connection */ + struct rt_export_request out_req; /* Table export connection */ u32 refeed_count; /* Number of routes exported during refeed regardless of out_limit */ @@ -534,10 +540,7 @@ struct channel { u8 stale; /* Used in reconfiguration */ u8 channel_state; - u8 export_state; /* Route export state (ES_*, see below) */ - u8 feed_active; - u8 flush_active; - u8 refeeding; /* We are refeeding (valid only if export_state == ES_FEEDING) */ + u8 refeeding; /* Refeeding the channel. */ u8 reloadable; /* Hook reload_routes() is allowed on the channel */ u8 gr_lock; /* Graceful restart mechanism should wait for this channel */ u8 gr_wait; /* Route export to channel is postponed until graceful restart */ @@ -585,34 +588,34 @@ struct channel { * restricted by that and is on volition of the protocol. Generally, channels * are opened in protocols' start() hooks when going to PS_UP. * - * CS_FLUSHING - The transitional state between initialized channel and closed + * CS_STOP - The transitional state between initialized channel and closed * channel. The channel is still initialized, but no route exchange is allowed. * Instead, the associated table is running flush loop to remove routes imported * through the channel. After that, the channel changes state to CS_DOWN and * is detached from the table (the table is unlocked and the channel is unlinked - * from it). Unlike other states, the CS_FLUSHING state is not explicitly + * from it). Unlike other states, the CS_STOP state is not explicitly * entered or left by the protocol. A protocol may request to close a channel * (by calling channel_close()), which causes the channel to change state to - * CS_FLUSHING and later to CS_DOWN. Also note that channels are closed + * CS_STOP and later to CS_DOWN. Also note that channels are closed * automatically by the core when the protocol is going down. * + * CS_PAUSE - Almost the same as CS_STOP, just the table import is kept and + * the table export is stopped before transitioning to CS_START. + * * Allowed transitions: * * CS_DOWN -> CS_START / CS_UP - * CS_START -> CS_UP / CS_FLUSHING - * CS_UP -> CS_START / CS_FLUSHING - * CS_FLUSHING -> CS_DOWN (automatic) + * CS_START -> CS_UP / CS_STOP + * CS_UP -> CS_PAUSE / CS_STOP + * CS_PAUSE -> CS_START (automatic) + * CS_STOP -> CS_DOWN (automatic) */ #define CS_DOWN 0 #define CS_START 1 #define CS_UP 2 -#define CS_FLUSHING 3 - -#define ES_DOWN 0 -#define ES_FEEDING 1 -#define ES_READY 2 - +#define CS_STOP 3 +#define CS_PAUSE 4 struct channel_config *proto_cf_find_channel(struct proto_config *p, uint net_type); static inline struct channel_config *proto_cf_main_channel(struct proto_config *pc) @@ -630,7 +633,7 @@ void channel_schedule_reload(struct channel *c); static inline void channel_init(struct channel *c) { channel_set_state(c, CS_START); } static inline void channel_open(struct channel *c) { channel_set_state(c, CS_UP); } -static inline void channel_close(struct channel *c) { channel_set_state(c, CS_FLUSHING); } +static inline void channel_close(struct channel *c) { channel_set_state(c, CS_STOP); } void channel_request_feeding(struct channel *c); void *channel_config_new(const struct channel_class *cc, const char *name, uint net_type, struct proto_config *proto); diff --git a/nest/route.h b/nest/route.h index 2543aaf6..ab8ee258 100644 --- a/nest/route.h +++ b/nest/route.h @@ -2,6 +2,7 @@ * BIRD Internet Routing Daemon -- Routing Table * * (c) 1998--2000 Martin Mares <mj@ucw.cz> + * (c) 2019--2021 Maria Matejka <mq@jmq.cz> * * Can be freely distributed and used under the terms of the GNU GPL. */ @@ -17,6 +18,7 @@ struct ea_list; struct protocol; struct proto; +struct channel; struct rte_src; struct symbol; struct timer; @@ -165,12 +167,12 @@ typedef struct rtable { struct fib fib; struct f_trie *trie; /* Trie of prefixes defined in fib */ char *name; /* Name of this table */ - list channels; /* List of attached channels (struct channel) */ uint addr_type; /* Type of address data stored in table (NET_*) */ int use_count; /* Number of protocols using this table */ u32 rt_count; /* Number of routes in the table */ - byte internal; /* Internal table of a protocol */ + list imports; /* Registered route importers */ + list exports; /* Registered route exporters */ struct hmap id_map; struct hostcache *hostcache; @@ -188,6 +190,7 @@ typedef struct rtable { byte prune_trie; /* Prune prefix trie during next table prune */ byte hcu_scheduled; /* Hostcache update is scheduled */ byte nhu_state; /* Next Hop Update state */ + byte internal; /* This table is internal for some other object */ struct fib_iterator prune_fit; /* Rtable prune FIB iterator */ struct fib_iterator nhu_fit; /* Next Hop Update FIB iterator */ struct f_trie *trie_new; /* New prefix trie defined during pruning */ @@ -257,7 +260,7 @@ typedef struct rte { struct rta *attrs; /* Attributes of this route */ const net_addr *net; /* Network this RTE belongs to */ struct rte_src *src; /* Route source that created the route */ - struct channel *sender; /* Channel used to send the route to the routing table */ + struct rt_import_hook *sender; /* Import hook used to send the route to the routing table */ btime lastmod; /* Last modified (set by table) */ u32 id; /* Table specific route id */ byte flags; /* Table-specific flags */ @@ -281,14 +284,129 @@ struct rte_storage { #define REF_MODIFY 16 /* Route is scheduled for modify */ /* Route is valid for propagation (may depend on other flags in the future), accepts NULL */ -static inline int rte_is_valid_rte(rte *r) { return r && !(r->flags & REF_FILTERED); } -static inline int rte_is_valid_storage(struct rte_storage *r) { return r && rte_is_valid_rte(&r->rte); } +static inline int rte_is_valid_rte(const rte *r) { return r && !(r->flags & REF_FILTERED); } +static inline int rte_is_valid_storage(const struct rte_storage *r) { return r && rte_is_valid_rte(&r->rte); } #define rte_is_valid(r) _Generic((*r), rte: rte_is_valid_rte, struct rte_storage: rte_is_valid_storage)(r) /* Route just has REF_FILTERED flag */ -static inline int rte_is_filtered(rte *r) { return !!(r->flags & REF_FILTERED); } +static inline int rte_is_filtered(const rte *r) { return !!(r->flags & REF_FILTERED); } + + +/* Table-channel connections */ + +struct rt_import_request { + struct rt_import_hook *hook; /* The table part of importer */ + char *name; + u8 trace_routes; + + void (*dump_req)(struct rt_import_request *req); + void (*log_state_change)(struct rt_import_request *req, u8 state); + /* Preimport is called when the @new route is just-to-be inserted, replacing @old. + * Return a route (may be different or modified in-place) to continue or NULL to withdraw. */ + struct rte *(*preimport)(struct rt_import_request *req, struct rte *new, struct rte *old); + struct rte *(*rte_modify)(struct rte *, struct linpool *); +}; + +struct rt_import_hook { + node n; + rtable *table; /* The connected table */ + struct rt_import_request *req; /* The requestor */ + + struct rt_import_stats { + /* Import - from protocol to core */ + u32 pref; /* Number of routes selected as best in the (adjacent) routing table */ + u32 updates_ignored; /* Number of route updates rejected as already in route table */ + u32 updates_accepted; /* Number of route updates accepted and imported */ + u32 withdraws_ignored; /* Number of route withdraws rejected as already not in route table */ + u32 withdraws_accepted; /* Number of route withdraws accepted and processed */ + } stats; + + btime last_state_change; /* Time of last state transition */ + + u8 import_state; /* IS_* */ + + void (*stopped)(struct rt_import_request *); /* Stored callback when import is stopped */ +}; + +struct rt_pending_export { + struct rte_storage *new, *new_best, *old, *old_best; +}; + +struct rt_export_request { + struct rt_export_hook *hook; /* Table part of the export */ + char *name; + u8 trace_routes; + + /* There are two methods of export. You can either request feeding every single change + * or feeding the whole route feed. In case of regular export, &export_one is preferred. + * Anyway, when feeding, &export_bulk is preferred, falling back to &export_one. + * Thus, for RA_OPTIMAL, &export_one is only set, + * for RA_MERGED and RA_ACCEPTED, &export_bulk is only set + * and for RA_ANY, both are set to accomodate for feeding all routes but receiving single changes + */ + void (*export_one)(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe); + void (*export_bulk)(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); + + void (*dump_req)(struct rt_export_request *req); + void (*log_state_change)(struct rt_export_request *req, u8); +}; + +struct rt_export_hook { + node n; + rtable *table; /* The connected table */ + + pool *pool; + linpool *lp; + + struct rt_export_request *req; /* The requestor */ + + struct rt_export_stats { + /* Export - from core to protocol */ + u32 updates_received; /* Number of route updates received */ + u32 withdraws_received; /* Number of route withdraws received */ + } stats; + + struct fib_iterator feed_fit; /* Routing table iterator used during feeding */ + + btime last_state_change; /* Time of last state transition */ + + u8 refeed_pending; /* Refeeding and another refeed is scheduled */ + u8 export_state; /* Route export state (TES_*, see below) */ + + struct event *event; /* Event running all the export operations */ + + void (*stopped)(struct rt_export_request *); /* Stored callback when export is stopped */ +}; + +#define TIS_DOWN 0 +#define TIS_UP 1 +#define TIS_STOP 2 +#define TIS_FLUSHING 3 +#define TIS_WAITING 4 +#define TIS_CLEARED 5 +#define TIS_MAX 6 + +#define TES_DOWN 0 +#define TES_HUNGRY 1 +#define TES_FEEDING 2 +#define TES_READY 3 +#define TES_STOP 4 +#define TES_MAX 5 + +void rt_request_import(rtable *tab, struct rt_import_request *req); +void rt_request_export(rtable *tab, struct rt_export_request *req); + +void rt_stop_import(struct rt_import_request *, void (*stopped)(struct rt_import_request *)); +void rt_stop_export(struct rt_export_request *, void (*stopped)(struct rt_export_request *)); + +const char *rt_import_state_name(u8 state); +const char *rt_export_state_name(u8 state); + +static inline u8 rt_import_get_state(struct rt_import_hook *ih) { return ih ? ih->import_state : TIS_DOWN; } +static inline u8 rt_export_get_state(struct rt_export_hook *eh) { return eh ? eh->export_state : TES_DOWN; } +void rte_import(struct rt_import_request *req, const net_addr *net, rte *new, struct rte_src *src); /* Types of route announcement, also used as flags */ #define RA_UNDEF 0 /* Undefined RA type */ @@ -303,6 +421,7 @@ static inline int rte_is_filtered(rte *r) { return !!(r->flags & REF_FILTERED); #define RIC_REJECT -1 /* Rejected by protocol */ #define RIC_DROP -2 /* Silently dropped by protocol */ +#define rte_update channel_rte_import /** * rte_update - enter a new update to a routing table * @c: channel doing the update @@ -362,23 +481,24 @@ net *net_get(rtable *tab, const net_addr *addr); net *net_route(rtable *tab, const net_addr *n); int net_roa_check(rtable *tab, const net_addr *n, u32 asn); int rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filter); -rte *rt_export_merged_show(struct channel *c, net *n, linpool *pool); -void rt_refresh_begin(rtable *t, struct channel *c); -void rt_refresh_end(rtable *t, struct channel *c); -void rt_modify_stale(rtable *t, struct channel *c); +rte *rt_export_merged(struct channel *c, rte ** feed, uint count, linpool *pool, int silent); +void rt_refresh_begin(rtable *t, struct rt_import_request *); +void rt_refresh_end(rtable *t, struct rt_import_request *); +void rt_modify_stale(rtable *t, struct rt_import_request *); void rt_schedule_prune(rtable *t); void rte_dump(struct rte_storage *); void rte_free(struct rte_storage *, rtable *); struct rte_storage *rte_store(const rte *, net *net, rtable *); void rt_dump(rtable *); void rt_dump_all(void); -int rt_feed_channel(struct channel *c); -void rt_feed_channel_abort(struct channel *c); +void rt_dump_hooks(rtable *); +void rt_dump_hooks_all(void); int rt_reload_channel(struct channel *c); void rt_reload_channel_abort(struct channel *c); void rt_refeed_channel(struct channel *c); void rt_prune_sync(rtable *t, int all); -int rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old, struct rte_storage **old_exported); +int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); +int rte_update_out(struct channel *c, const net_addr *n, rte *new, const rte *old, struct rte_storage **old_exported); struct rtable_config *rt_new_table(struct symbol *s, uint addr_type); static inline int rt_is_ip(rtable *tab) @@ -418,6 +538,7 @@ struct rt_show_data { struct channel *export_channel; struct config *running_on_config; struct krt_proto *kernel; + struct rt_export_hook *kernel_export_hook; int export_mode, addr_mode, primary_only, filtered, stats; int table_open; /* Iteration (fit) is open */ diff --git a/nest/rt-show.c b/nest/rt-show.c index 117d6e00..aadb679e 100644 --- a/nest/rt-show.c +++ b/nest/rt-show.c @@ -99,6 +99,29 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary rta_show(c, a); } +static uint +rte_feed_count(net *n) +{ + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTE_OR_NULL(e))) + count++; + return count; +} + +static void +rte_feed_obtain(net *n, rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTE_OR_NULL(e))) + { + ASSERT_DIE(i < count); + feed[i++] = &e->rte; + } + ASSERT_DIE(i == count); +} + static void rt_show_net(struct cli *c, net *n, struct rt_show_data *d) { @@ -128,7 +151,7 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) struct rte e = er->rte; /* Export channel is down, do not try to export routes to it */ - if (ec && (ec->export_state == ES_DOWN)) + if (ec && !ec->out_req.hook) goto skip; if (d->export_mode == RSEM_EXPORTED) @@ -143,7 +166,14 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) { /* Special case for merged export */ pass = 1; - rte *em = rt_export_merged_show(ec, n, c->show_pool); + uint count = rte_feed_count(n); + if (!count) + goto skip; + + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(n, feed, count); + rte *em = rt_export_merged(ec, feed, count, c->show_pool, 1); + if (em) e = *em; else @@ -369,7 +399,7 @@ rt_show_get_default_tables(struct rt_show_data *d) { WALK_LIST(c, d->export_protocol->channels) { - if (c->export_state == ES_DOWN) + if (!c->out_req.hook) continue; tab = rt_show_add_table(d, c->table); diff --git a/nest/rt-table.c b/nest/rt-table.c index 4c3db55d..81929a0b 100644 --- a/nest/rt-table.c +++ b/nest/rt-table.c @@ -115,13 +115,7 @@ pool *rt_table_pool; static linpool *rte_update_pool; list routing_tables; - -struct rt_pending_export { - struct rte_storage *new; /* New route */ - struct rte_storage *new_best; /* New best route */ - struct rte_storage *old; /* Old route */ - struct rte_storage *old_best; /* Old best route */ -}; +list deleted_routing_tables; static void rt_free_hostcache(rtable *tab); static void rt_notify_hostcache(rtable *tab, net *net); @@ -130,6 +124,41 @@ 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_feed_channel(void *); + +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_HUNGRY] = "HUNGRY", + [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 void @@ -644,26 +673,43 @@ rte_mergable(rte *pri, rte *sec) } static void -rte_trace(struct channel *c, rte *e, int dir, const 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, e->src->private_id, e->src->global_id, + log(L_TRACE "%s %c %s %N %uL %uG %s", + name, dir, msg, e->net, e->src->private_id, e->src->global_id, rta_dest_name(e->attrs->dest)); } static inline void -rte_trace_in(uint flag, struct channel *c, rte *e, const 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, const 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) @@ -693,7 +739,7 @@ export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) { struct proto *p = c->proto; const struct filter *filter = c->out_filter; - struct export_stats *stats = &c->export_stats; + struct channel_export_stats *stats = &c->export_stats; /* Do nothing if we have already rejected the route */ if (silent && bmap_test(&c->export_reject_map, rt->id)) @@ -707,14 +753,14 @@ export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) stats->updates_rejected++; if (v == RIC_REJECT) - rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); + 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; } @@ -727,7 +773,7 @@ export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) goto reject; stats->updates_filtered++; - rte_trace_out(D_FILTERS, c, rt, "filtered out"); + channel_rte_trace_out(D_FILTERS, c, rt, "filtered out"); goto reject; } @@ -741,7 +787,6 @@ export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) bmap_set(&c->export_reject_map, rt->id); reject_noset: - /* Invalidate the route */ /* Discard temporary rte */ return NULL; } @@ -753,28 +798,24 @@ export_filter(struct channel *c, rte *rt, int silent) } static void -do_rt_notify(struct channel *c, const net_addr *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 export_stats *stats = &c->export_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->routes >= l->limit) - channel_notify_limit(c, l, PLD_OUT, stats->routes); - - if (l->state == PLS_BLOCKED) + if (!old && new) + if (CHANNEL_LIMIT_PUSH(c, OUT)) { stats->updates_rejected++; - rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); + channel_rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); return; } - } + + if (!new && old) + CHANNEL_LIMIT_POP(c, OUT); /* Apply export table */ struct rte_storage *old_exported = NULL; @@ -782,7 +823,7 @@ do_rt_notify(struct channel *c, const net_addr *net, rte *new, rte *old, int ref { if (!rte_update_out(c, net, new, old, &old_exported)) { - rte_trace_out(D_ROUTES, c, new, "idempotent"); + channel_rte_trace_out(D_ROUTES, c, new, "idempotent"); return; } } @@ -793,25 +834,19 @@ do_rt_notify(struct channel *c, const net_addr *net, rte *new, rte *old, int ref stats->withdraws_accepted++; if (old) - { bmap_clear(&c->export_map, old->id); - stats->routes--; - } if (new) - { bmap_set(&c->export_map, new->id); - stats->routes++; - } if (p->debug & D_ROUTES) { if (new && old) - rte_trace_out(D_ROUTES, c, new, "replaced"); + 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_exported ? &old_exported->rte : old); @@ -821,14 +856,9 @@ do_rt_notify(struct channel *c, const net_addr *net, rte *new, rte *old, int ref } static void -rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old, int refeed) +rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old) { if (new) - c->export_stats.updates_received++; - else - c->export_stats.withdraws_received++; - - if (new) new = export_filter(c, new, 0); if (old && !bmap_test(&c->export_map, old->id)) @@ -837,14 +867,17 @@ rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old, int if (!new && !old) return; - do_rt_notify(c, net, new, old, refeed); + do_rt_notify(c, net, new, old); } -static void -rt_notify_accepted(struct channel *c, const net_addr *n, struct rt_pending_export *rpe, - struct rte **feed, uint count, int refeed) +void +rt_notify_accepted(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe, + struct rte **feed, uint count) { - rte nb0, *new_best = NULL, *old_best = NULL; + 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++) { @@ -852,7 +885,7 @@ rt_notify_accepted(struct channel *c, const net_addr *n, struct rt_pending_expor continue; /* Has been already rejected, won't bother with it */ - if (!refeed && bmap_test(&c->export_reject_map, feed[i]->id)) + if (!c->refeeding && bmap_test(&c->export_reject_map, feed[i]->id)) continue; /* Previously exported */ @@ -862,7 +895,7 @@ rt_notify_accepted(struct channel *c, const net_addr *n, struct rt_pending_expor if (!new_best) { DBG("rt_notify_accepted: idempotent\n"); - return; + goto done; } /* is superseded */ @@ -902,10 +935,16 @@ rt_notify_accepted(struct channel *c, const net_addr *n, struct rt_pending_expor if (!new_best && !old_best) { DBG("rt_notify_accepted: nothing to export\n"); - return; + goto done; } - do_rt_notify(c, n, new_best, old_best, refeed); + do_rt_notify(c, n, new_best, old_best); + +done: + /* Drop the old stored rejection if applicable. + * new->id == old->id happens when updating hostentries. */ + if (rpe && rpe->old && (!rpe->new || (rpe->new->rte.id != rpe->old->rte.id))) + bmap_clear(&c->export_reject_map, rpe->old->rte.id); } @@ -915,20 +954,21 @@ nexthop_merge_rta(struct nexthop *nhs, rta *a, linpool *pool, int max) return nexthop_merge(nhs, &(a->nh), 1, 0, max, pool); } -static rte * -rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool, int silent, int refeed) +rte * +rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool, int silent) { _Thread_local static rte rloc; // struct proto *p = c->proto; struct nexthop *nhs = NULL; - rte *best0 = feed[0], *best = NULL; + rte *best0 = feed[0]; + rte *best = NULL; if (!rte_is_valid(best0)) return NULL; /* Already rejected, no need to re-run the filter */ - if (!refeed && bmap_test(&c->export_reject_map, best0->id)) + if (!c->refeeding && bmap_test(&c->export_reject_map, best0->id)) return NULL; rloc = *best0; @@ -970,19 +1010,12 @@ rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool return best; } -rte * -rt_export_merged_show(struct channel *c, net *n, linpool *pool) +void +rt_notify_merged(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe, + struct rte **feed, uint count) { - uint count = rte_feed_count(n); - rte **feed = alloca(count * sizeof(rte *)); - rte_feed_obtain(n, feed, count); - return rt_export_merged(c, feed, count, pool, 1, 0); -} + struct channel *c = SKIP_BACK(struct channel, out_req, req); -static void -rt_notify_merged(struct channel *c, const net_addr *n, struct rt_pending_export *rpe, - struct rte **feed, uint count, int refeed) -{ // struct proto *p = c->proto; #if 0 /* TODO: Find whether this check is possible when processing multiple changes at once. */ @@ -1022,28 +1055,59 @@ rt_notify_merged(struct channel *c, const net_addr *n, struct rt_pending_export */ /* Prepare new merged route */ - rte *new_merged = count ? rt_export_merged(c, feed, count, rte_update_pool, 0, refeed) : NULL; + rte *new_merged = count ? rt_export_merged(c, feed, count, rte_update_pool, 0) : NULL; - if (!new_merged && !old_best) - return; + if (new_merged || old_best) + do_rt_notify(c, n, new_merged, old_best); - do_rt_notify(c, n, new_merged, old_best, refeed); + /* Drop the old stored rejection if applicable. + * new->id == old->id happens when updating hostentries. */ + if (rpe && rpe->old && (!rpe->new || (rpe->new->rte.id != rpe->old->rte.id))) + bmap_clear(&c->export_reject_map, rpe->old->rte.id); } -static void -rt_notify_bulk(struct channel *c, const net_addr *n, struct rt_pending_export *rpe, - struct rte **feed, uint count, int refeed) +void +rt_notify_optimal(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) { - switch (c->ra_mode) + struct channel *c = SKIP_BACK(struct channel, out_req, req); + rte n0; + + if (rpe->new_best != rpe->old_best) + rt_notify_basic(c, net, RTE_COPY(rpe->new_best, &n0), RTE_OR_NULL(rpe->old_best)); + + /* Drop the old stored rejection if applicable. + * new->id == old->id happens when updating hostentries. */ + if (rpe->old && (!rpe->new || (rpe->new->rte.id != rpe->old->rte.id))) + bmap_clear(&c->export_reject_map, rpe->old->rte.id); +} + +void +rt_notify_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + rte n0; + + if (rpe->new != rpe->old) + rt_notify_basic(c, net, RTE_COPY(rpe->new, &n0), RTE_OR_NULL(rpe->old)); + + /* Drop the old stored rejection if applicable. + * new->id == old->id happens when updating hostentries. */ + if (rpe->old && (!rpe->new || (rpe->new->rte.id != rpe->old->rte.id))) + bmap_clear(&c->export_reject_map, rpe->old->rte.id); +} + +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++) { - case RA_ACCEPTED: - return rt_notify_accepted(c, n, rpe, feed, count, refeed); - case RA_MERGED: - return rt_notify_merged(c, n, rpe, feed, count, refeed); + rte n0 = *feed[i]; + rt_notify_basic(c, net, &n0, NULL); } } - /** * rte_announce - announce a routing table change * @tab: table the route has been added to @@ -1097,9 +1161,9 @@ rte_announce(rtable *tab, net *net, struct rte_storage *new, struct rte_storage if (new_best != old_best) { if (new_best) - new_best->rte.sender->import_stats.pref++; + new_best->rte.sender->stats.pref++; if (old_best) - old_best->rte.sender->import_stats.pref--; + old_best->rte.sender->stats.pref--; if (tab->hostcache) rt_notify_hostcache(tab, net); @@ -1110,46 +1174,37 @@ rte_announce(rtable *tab, net *net, struct rte_storage *new, struct rte_storage rt_schedule_notify(tab); - struct channel *c; node *n; - WALK_LIST2(c, n, tab->channels, table_node) + struct rt_pending_export rpe = { .new = new, .old = old, .new_best = new_best, .old_best = old_best }; + uint count = rte_feed_count(net); + rte **feed = NULL; + if (count) { - if (c->export_state == ES_DOWN) - continue; - - rte n0; - switch (c->ra_mode) - { - case RA_OPTIMAL: - if (new_best != old_best) - rt_notify_basic(c, net->n.addr, RTE_COPY(new_best, &n0), RTE_OR_NULL(old_best), 0); - break; + feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(net, feed, count); + } - case RA_ANY: - if (new != old) - rt_notify_basic(c, net->n.addr, RTE_COPY(new, &n0), RTE_OR_NULL(old), 0); - break; + struct rt_export_hook *eh; + WALK_LIST(eh, tab->exports) + { + if (eh->export_state == TES_STOP) + continue; - case RA_ACCEPTED: - case RA_MERGED: - { - struct rt_pending_export rpe = { .new = new, .old = old, .new_best = new_best, .old_best = old_best }; - uint count = rte_feed_count(net); - rte **feed = alloca(count * sizeof(rte *)); - rte_feed_obtain(net, feed, count); - rt_notify_bulk(c, net->n.addr, &rpe, feed, count, 0); - break; - } - } + if (new) + eh->stats.updates_received++; + else + eh->stats.withdraws_received++; - /* Drop the old stored rejection if applicable. - * new->id == old->id happens when updating hostentries. */ - if (old && (!new || (new->rte.id != old->rte.id))) - bmap_clear(&c->export_reject_map, old->rte.id); + if (eh->req->export_one) + eh->req->export_one(eh->req, net->n.addr, &rpe); + else if (eh->req->export_bulk) + eh->req->export_bulk(eh->req, net->n.addr, &rpe, feed, count); + else + bug("Export request must always provide an export method"); } } static inline int -rte_validate(rte *e) +rte_validate(struct channel *ch, rte *e) { int c; const net_addr *n = e->net; @@ -1157,7 +1212,7 @@ rte_validate(rte *e) if (!net_validate(n)) { log(L_WARN "Ignoring bogus prefix %N received via %s", - n, e->sender->proto->name); + n, ch->proto->name); return 0; } @@ -1167,7 +1222,7 @@ rte_validate(rte *e) if ((c < 0) || !(c & IADDR_HOST) || ((c & IADDR_SCOPE_MASK) <= SCOPE_LINK)) { log(L_WARN "Ignoring bogus route %N received via %s", - n, e->sender->proto->name); + n, ch->proto->name); return 0; } @@ -1178,14 +1233,14 @@ rte_validate(rte *e) return 1; log(L_WARN "Ignoring route %N with invalid dest %d received via %s", - n, e->attrs->dest, e->sender->proto->name); + n, e->attrs->dest, ch->proto->name); return 0; } if ((e->attrs->dest == RTD_UNICAST) && !nexthop_is_sorted(&(e->attrs->nh))) { log(L_WARN "Ignoring unsorted multipath route %N received via %s", - n, e->sender->proto->name); + n, ch->proto->name); return 0; } @@ -1206,11 +1261,11 @@ 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) +rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *src) { - struct proto *p = c->proto; + struct rt_import_request *req = c->req; struct rtable *table = c->table; - struct import_stats *stats = &c->import_stats; + 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; @@ -1227,7 +1282,7 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) * source protocol to this routing table through transparent * pipes, which is not allowed. * We log that and ignore the route. */ - if (old->sender->proto != p) + if (old->sender != c) { 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", @@ -1235,8 +1290,6 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) log_rl(&table->rl_pipe, L_ERR "Route source collision in table %s: %N %s/%u:%u", c->table->name, net->n.addr, old->src->proto->name, old->src->private_id, old->src->global_id); - - return; } if (new && rte_same(old, new)) @@ -1248,13 +1301,10 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) if (!rte_is_filtered(new)) { stats->updates_ignored++; - rte_trace_in(D_ROUTES, c, new, "ignored"); + rt_rte_trace_in(D_ROUTES, req, new, "ignored"); } - - return; } - *before_old = (*before_old)->next; table->rt_count--; } @@ -1265,62 +1315,12 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) return; } + if (req->preimport) + new = req->preimport(req, new, old); + int new_ok = rte_is_ok(new); int old_ok = rte_is_ok(old); - struct channel_limit *l = &c->rx_limit; - if (l->action && !old && new && !c->in_table) - { - u32 all_routes = stats->routes + stats->filtered; - - if (all_routes >= l->limit) - channel_notify_limit(c, l, PLD_RX, all_routes); - - if (l->state == PLS_BLOCKED) - { - /* In receive limit the situation is simple, old is NULL so - we just free new and exit like nothing happened */ - - stats->updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - return; - } - } - - l = &c->in_limit; - if (l->action && !old_ok && new_ok) - { - if (stats->routes >= l->limit) - channel_notify_limit(c, l, PLD_IN, stats->routes); - - if (l->state == PLS_BLOCKED) - { - /* In import limit the situation is more complicated. We - shouldn't just drop the route, we should handle it like - it was filtered. We also have to continue the route - processing if old or new is non-NULL, but we should exit - if both are NULL as this case is probably assumed to be - already handled. */ - - stats->updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - - if (c->in_keep_filtered) - new->flags |= REF_FILTERED; - else - new = NULL; - - /* Note that old && !new could be possible when - c->in_keep_filtered changed in the recent past. */ - - if (!old && !new) - return; - - new_ok = 0; - goto skip_stats1; - } - } - if (new_ok) stats->updates_accepted++; else if (old_ok) @@ -1331,14 +1331,8 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) if (old_ok || new_ok) table->last_rt_change = current_time(); - skip_stats1:; struct rte_storage *new_stored = new ? rte_store(new, net, table) : NULL; - if (new) - rte_is_filtered(new) ? stats->filtered++ : stats->routes++; - if (old) - rte_is_filtered(old) ? stats->filtered-- : stats->routes--; - if (table->config->sorted) { /* If routes are sorted, just insert new route to appropriate position */ @@ -1441,19 +1435,16 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) } /* 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_stored->rte, '>', new_stored == net->routes ? "added [best]" : "added"); - else if (old_ok) - { - if (old != old_best) - rte_trace(c, old, '>', "removed"); - else if (net->routes && rte_is_ok(&net->routes->rte)) - rte_trace(c, old, '>', "removed [replaced]"); - else - rte_trace(c, old, '>', "removed [sole]"); - } + 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 */ @@ -1465,10 +1456,13 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) (table->gc_time + table->config->gc_min_time <= current_time())) rt_schedule_prune(table); +#if 0 + /* Enable and reimplement these callbacks if anybody wants to use them */ if (old_ok && p->rte_remove) p->rte_remove(net, old); if (new_ok && p->rte_insert) p->rte_insert(net, &new_stored->rte); +#endif if (old) { @@ -1494,89 +1488,116 @@ rte_update_unlock(void) lp_flush(rte_update_pool); } -static int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); +rte * +channel_preimport(struct rt_import_request *req, rte *new, rte *old) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); + + if (new && !old) + if (CHANNEL_LIMIT_PUSH(c, RX)) + return NULL; + + if (!new && old) + CHANNEL_LIMIT_POP(c, RX); + + int new_in = new && !rte_is_filtered(new); + int old_in = old && !rte_is_filtered(old); + + if (new_in && !old_in) + if (CHANNEL_LIMIT_PUSH(c, IN)) + if (c->in_keep_filtered) + { + new->flags |= REF_FILTERED; + return new; + } + else + return NULL; + + if (!new_in && old_in) + CHANNEL_LIMIT_POP(c, IN); + + return new; +} + +static void rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); void rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { + if (!c->in_req.hook) + return; + + ASSERT(c->channel_state == CS_UP); + if (c->in_table && !rte_update_in(c, n, new, src)) return; - struct import_stats *stats = &c->import_stats; - const struct filter *filter = c->in_filter; - net *nn; + return rte_update_direct(c, n, new, src); +} - ASSERT(c->channel_state == CS_UP); +static void +rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +{ + const struct filter *filter = c->in_filter; + struct channel_import_stats *stats = &c->import_stats; rte_update_lock(); if (new) { new->net = n; - new->sender = c; + + int fr; stats->updates_received++; - if (!rte_validate(new)) + if (!rte_validate(c, new)) { - rte_trace_in(D_FILTERS, c, new, "invalid"); + channel_rte_trace_in(D_FILTERS, c, new, "invalid"); stats->updates_invalid++; - goto drop; + new = NULL; } - - if (filter == FILTER_REJECT) + else if ((filter == FILTER_REJECT) || + ((fr = f_run(filter, new, rte_update_pool, 0)) > F_ACCEPT)) { stats->updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); + channel_rte_trace_in(D_FILTERS, c, new, "filtered out"); - if (! c->in_keep_filtered) - goto drop; - - /* new is a private copy, i could modify it */ - new->flags |= REF_FILTERED; + if (c->in_keep_filtered) + new->flags |= REF_FILTERED; + else + new = NULL; } - else if (filter) - { - int fr = f_run(filter, new, rte_update_pool, 0); - if (fr > F_ACCEPT) - { - stats->updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); + } + else + stats->withdraws_received++; - if (! c->in_keep_filtered) - goto drop; + rte_import(&c->in_req, n, new, src); - new->flags |= REF_FILTERED; - } - } + rte_update_unlock(); +} + +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; + net *nn; + if (new) + { /* Use the actual struct network, not the dummy one */ - nn = net_get(c->table, n); + nn = net_get(hook->table, n); new->net = nn->n.addr; + new->sender = hook; } - else + else if (!(nn = net_find(hook->table, n))) { - stats->withdraws_received++; - - if (!(nn = net_find(c->table, n)) || !src) - { - stats->withdraws_ignored++; - rte_update_unlock(); - return; - } + req->hook->stats.withdraws_ignored++; + return; } - recalc: /* And recalculate the best route */ - rte_recalculate(c, nn, new, src); - - rte_update_unlock(); - return; - - drop: - new = NULL; - if (nn = net_find(c->table, n)) - goto recalc; - - rte_update_unlock(); + rte_recalculate(hook, nn, new, src); } /* Independent call to rte_announce(), used from next hop @@ -1604,7 +1625,7 @@ rte_modify(net *net, rte *old) { rte_update_lock(); - rte *new = old->sender->proto->rte_modify(old, rte_update_pool); + rte *new = old->sender->req->rte_modify(old, rte_update_pool); if (new != old) { if (new) @@ -1639,6 +1660,127 @@ rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filte return v > 0; } +static void +rt_export_stopped(void *data) +{ + struct rt_export_hook *hook = data; + rtable *tab = hook->table; + + /* Unlist */ + rem_node(&hook->n); + + /* Reporting the channel as stopped. */ + hook->stopped(hook->req); + + /* Freeing the hook together with its coroutine. */ + rfree(hook->pool); + rt_unlock_table(tab); + + DBG("Export hook %p in table %s finished uc=%u\n", hook, tab->name, tab->use_count); +} + + +static inline void +rt_set_import_state(struct rt_import_hook *hook, u8 state) +{ + hook->last_state_change = current_time(); + hook->import_state = state; + + if (hook->req->log_state_change) + hook->req->log_state_change(hook->req, state); +} + +static inline void +rt_set_export_state(struct rt_export_hook *hook, u8 state) +{ + hook->last_state_change = current_time(); + hook->export_state = state; + + if (hook->req->log_state_change) + hook->req->log_state_change(hook->req, state); +} + +void +rt_request_import(rtable *tab, struct rt_import_request *req) +{ + rt_lock_table(tab); + + struct rt_import_hook *hook = req->hook = mb_allocz(tab->rp, sizeof(struct rt_import_hook)); + + DBG("Lock table %s for import %p req=%p uc=%u\n", tab->name, hook, req, tab->use_count); + + hook->req = req; + hook->table = tab; + + rt_set_import_state(hook, TIS_UP); + + hook->n = (node) {}; + 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; + + rt_schedule_prune(hook->table); + + rt_set_import_state(hook, TIS_STOP); + + hook->stopped = stopped; +} + +void +rt_request_export(rtable *tab, struct rt_export_request *req) +{ + rt_lock_table(tab); + + pool *p = rp_new(tab->rp, "Export hook"); + struct rt_export_hook *hook = req->hook = mb_allocz(p, sizeof(struct rt_export_hook)); + hook->pool = p; + hook->lp = lp_new_default(p); + + hook->req = req; + hook->table = tab; + + /* stats zeroed by mb_allocz */ + + rt_set_export_state(hook, TES_HUNGRY); + + hook->n = (node) {}; + add_tail(&tab->exports, &hook->n); + + FIB_ITERATE_INIT(&hook->feed_fit, &tab->fib); + + DBG("New export hook %p req %p in table %s uc=%u\n", hook, req, tab->name, tab->use_count); + + rt_set_export_state(hook, TES_FEEDING); + + hook->event = ev_new_init(p, rt_feed_channel, hook); + ev_schedule_work(hook->event); +} + +void +rt_stop_export(struct rt_export_request *req, void (*stopped)(struct rt_export_request *)) +{ + ASSERT_DIE(req->hook); + struct rt_export_hook *hook = req->hook; + + rtable *tab = hook->table; + + /* Stop feeding */ + ev_postpone(hook->event); + + if (hook->export_state == TES_FEEDING) + fit_get(&tab->fib, &hook->feed_fit); + + hook->event->hook = rt_export_stopped; + hook->stopped = stopped; + + rt_set_export_state(hook, TES_STOP); + ev_schedule(hook->event); +} /** * rt_refresh_begin - start a refresh cycle @@ -1655,12 +1797,12 @@ 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(rtable *t, struct rt_import_request *req) { FIB_WALK(&t->fib, net, n) { for (struct rte_storage *e = n->routes; e; e = e->next) - if (e->rte.sender == c) + if (e->rte.sender == req->hook) e->rte.flags |= REF_STALE; } FIB_WALK_END; @@ -1675,14 +1817,14 @@ 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(rtable *t, struct rt_import_request *req) { int prune = 0; FIB_WALK(&t->fib, net, n) { for (struct rte_storage *e = n->routes; e; e = e->next) - if ((e->rte.sender == c) && (e->rte.flags & REF_STALE)) + if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE)) { e->rte.flags |= REF_DISCARD; prune = 1; @@ -1695,14 +1837,14 @@ rt_refresh_end(rtable *t, struct channel *c) } void -rt_modify_stale(rtable *t, struct channel *c) +rt_modify_stale(rtable *t, struct rt_import_request *req) { int prune = 0; FIB_WALK(&t->fib, net, n) { for (struct rte_storage *e = n->routes; e; e = e->next) - if ((e->rte.sender == c) && (e->rte.flags & REF_STALE) && !(e->rte.flags & REF_FILTERED)) + if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE) && !(e->rte.flags & REF_FILTERED)) { e->rte.flags |= REF_MODIFY; prune = 1; @@ -1738,7 +1880,7 @@ rte_dump(struct rte_storage *e) void rt_dump(rtable *t) { - debug("Dump of routing table <%s>\n", t->name); + debug("Dump of routing table <%s>%s\n", t->name, t->deleted ? " (deleted)" : ""); #ifdef DEBUGGING fib_check(&t->fib); #endif @@ -1764,6 +1906,54 @@ 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); +} + +void +rt_dump_hooks(rtable *tab) +{ + debug("Dump of hooks in routing table <%s>%s\n", tab->name, tab->deleted ? " (deleted)" : ""); + debug(" nhu_state=%u hcu_scheduled=%u use_count=%d rt_count=%u\n", + tab->nhu_state, tab->hcu_scheduled, 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_export_hook *eh; + WALK_LIST(eh, tab->exports) + { + eh->req->dump_req(eh->req); + debug(" Export hook %p requested by %p:" + " refeed_pending=%u last_state_change=%t export_state=%u stopped=%p\n", + eh, eh->req, eh->refeed_pending, eh->last_state_change, eh->export_state, eh->stopped); + } + debug("\n"); +} + +void +rt_dump_hooks_all(void) +{ + rtable *t; + node *n; + + 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 @@ -1881,6 +2071,7 @@ rt_subscribe(rtable *tab, struct rt_subscription *s) { s->tab = tab; rt_lock_table(tab); + DBG("rt_subscribe(%s)\n", tab->name); add_tail(&tab->subscribers, &s->n); } @@ -2047,15 +2238,17 @@ 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); if (!(t->internal = cf->internal)) { + init_list(&t->imports); + init_list(&t->exports); hmap_init(&t->id_map, p, 1024); hmap_set(&t->id_map, 0); + init_list(&t->subscribers); + t->rt_event = ev_new_init(p, rt_event, t); t->last_rt_change = t->gc_time = current_time(); @@ -2084,6 +2277,7 @@ rt_init(void) rt_table_pool = rp_new(&root_pool, "Routing tables"); rte_update_pool = lp_new_default(rt_table_pool); init_list(&routing_tables); + init_list(&deleted_routing_tables); } @@ -2107,7 +2301,7 @@ rt_prune_table(rtable *tab) struct fib_iterator *fit = &tab->prune_fit; int limit = 512; - struct channel *c; + struct rt_import_hook *ih; node *n, *x; DBG("Pruning route table %s\n", tab->name); @@ -2121,9 +2315,9 @@ 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); FIB_ITERATE_INIT(fit, &tab->fib); tab->prune_state = 2; @@ -2149,7 +2343,7 @@ again: for (struct rte_storage *e=n->routes; e; e=e->next) { - if (e->rte.sender->flush_active || (e->rte.flags & REF_DISCARD)) + if ((e->rte.sender->import_state == TIS_FLUSHING) || (e->rte.flags & REF_DISCARD)) { rte_discard(n, &e->rte); limit--; @@ -2222,21 +2416,18 @@ again: } } - if (tab->prune_state > 0) - ev_schedule(tab->rt_event); - - /* FIXME: This should be handled in a better way */ rt_prune_sources(); /* Close flushed channels */ - WALK_LIST2_DELSAFE(c, n, x, tab->channels, table_node) - if (c->flush_active) - { - c->flush_active = 0; - channel_set_state(c, CS_DOWN); - } - - return; + WALK_LIST2_DELSAFE(ih, n, x, tab->imports, n) + if (ih->import_state == TIS_FLUSHING) + { + rt_set_import_state(ih, TIS_CLEARED); + ih->stopped(ih->req); + rem_node(&ih->n); + mb_free(ih); + rt_unlock_table(tab); + } } /** @@ -2665,7 +2856,7 @@ rt_next_hop_update_net(rtable *tab, net *n) { _Bool nb = (new == updates[i].new), ob = (old_best == updates[i].old); const char *best_indicator[2][2] = { { "updated", "updated [-best]" }, { "updated [+best]", "updated [best]" } }; - rte_trace_in(D_ROUTES, new->rte.sender, &updates[i].new->rte, best_indicator[nb][ob]); + rt_rte_trace_in(D_ROUTES, updates[i].new->rte.sender->req, &updates[i].new->rte, best_indicator[nb][ob]); rte_announce_i(tab, n, updates[i].new, updates[i].old, new, old_best); } @@ -2850,25 +3041,6 @@ rt_commit(struct config *new, struct config *old) DBG("\tdone\n"); } -static inline void -do_feed_channel(struct channel *c, net *n, rte *e) -{ - rte_update_lock(); - if ((c->ra_mode == RA_ACCEPTED) || (c->ra_mode == RA_MERGED)) - { - uint count = rte_feed_count(n); - rte **feed = alloca(count * sizeof(rte *)); - rte_feed_obtain(n, feed, count); - rt_notify_bulk(c, n->n.addr, NULL, feed, count, c->refeeding); - } - else /* RA_BASIC */ - { - rte e0 = *e; - rt_notify_basic(c, n->n.addr, &e0, &e0, c->refeeding); - } - rte_update_unlock(); -} - /** * rt_feed_channel - advertise all routes to a channel * @c: channel to be fed @@ -2878,79 +3050,55 @@ do_feed_channel(struct channel *c, net *n, rte *e) * 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) +static void +rt_feed_channel(void *data) { + struct rt_export_hook *c = data; + struct fib_iterator *fit = &c->feed_fit; int max_feed = 256; - ASSERT(c->export_state == ES_FEEDING); - - if (!c->feed_active) - { - FIB_ITERATE_INIT(fit, &c->table->fib); - c->feed_active = 1; - } + ASSERT(c->export_state == TES_FEEDING); FIB_ITERATE_START(&c->table->fib, fit, net, n) { - struct rte_storage *e = n->routes; if (max_feed <= 0) { FIB_ITERATE_PUT(fit); - return 0; + ev_schedule_work(c->event); + return; } - 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->rte); - max_feed--; - } - - if (c->ra_mode == RA_ANY) - for(e = n->routes; e; e = e->next) - { - /* In the meantime, the protocol may fell down */ - if (c->export_state != ES_FEEDING) - goto done; - - if (!rte_is_valid(e)) - continue; + if (c->export_state != TES_FEEDING) + goto done; - do_feed_channel(c, n, &e->rte); - max_feed--; - } + if (c->req->export_bulk) + { + uint count = rte_feed_count(n); + if (count) + { + rte_update_lock(); + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(n, feed, count); + struct rt_pending_export rpe = { .new_best = n->routes }; + c->req->export_bulk(c->req, n->n.addr, &rpe, feed, count); + max_feed -= count; + rte_update_unlock(); + } + } + else if (n->routes && rte_is_valid(&n->routes->rte)) + { + rte_update_lock(); + struct rt_pending_export rpe = { .new = n->routes, .new_best = n->routes }; + c->req->export_one(c->req, n->n.addr, &rpe); + max_feed--; + rte_update_unlock(); + } } FIB_ITERATE_END; done: - c->feed_active = 0; - return 1; -} - -/** - * rt_feed_baby_abort - abort protocol feeding - * @c: channel - * - * This function is called by the protocol code when the protocol stops or - * ceases to exist during the feeding. - */ -void -rt_feed_channel_abort(struct channel *c) -{ - if (c->feed_active) - { - /* Unlink the iterator */ - fit_get(&c->table->fib, &c->feed_fit); - c->feed_active = 0; - } + rt_set_export_state(c, TES_READY); } @@ -2958,7 +3106,7 @@ rt_feed_channel_abort(struct channel *c) * Import table */ -static int +int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { struct rtable *tab = c->in_table; @@ -2991,6 +3139,9 @@ rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *sr goto drop_update; } + if (!new) + CHANNEL_LIMIT_POP(c, RX); + /* Move iterator if needed */ if (*pos == c->reload_next_rte) c->reload_next_rte = (*pos)->next; @@ -3001,7 +3152,18 @@ rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *sr rte_free(del, tab); tab->rt_count--; } - else if (!new) + else if (new) + { + if (CHANNEL_LIMIT_PUSH(c, RX)) + { + /* Required by rte_trace_in() */ + new->net = n; + + channel_rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); + goto drop_update; + } + } + else goto drop_withdraw; if (!new) @@ -3012,25 +3174,8 @@ rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *sr return 1; } - struct channel_limit *l = &c->rx_limit; - if (l->action && !*pos) - { - if (tab->rt_count >= l->limit) - channel_notify_limit(c, l, PLD_RX, tab->rt_count); - - if (l->state == PLS_BLOCKED) - { - /* Required by rte_trace_in() */ - new->net = n; - - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - goto drop_update; - } - } - /* Insert the new rte */ struct rte_storage *e = rte_store(new, net, tab); - e->rte.sender = c; e->rte.lastmod = current_time(); e->next = *pos; *pos = e; @@ -3039,7 +3184,7 @@ rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *sr drop_update: c->import_stats.updates_received++; - c->import_stats.updates_ignored++; + c->in_req.hook->stats.updates_ignored++; if (!net->routes) fib_delete(&tab->fib, net); @@ -3048,7 +3193,7 @@ drop_update: drop_withdraw: c->import_stats.withdraws_received++; - c->import_stats.withdraws_ignored++; + c->in_req.hook->stats.withdraws_ignored++; return 0; } @@ -3078,7 +3223,7 @@ rt_reload_channel(struct channel *c) } rte r = e->rte; - rte_update(c, r.net, &r, r.src); + rte_update_direct(c, r.net, &r, r.src); } c->reload_next_rte = NULL; @@ -3151,7 +3296,7 @@ again: */ int -rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, struct rte_storage **old_exported) +rte_update_out(struct channel *c, const net_addr *n, rte *new, const rte *old0, struct rte_storage **old_exported) { struct rtable *tab = c->out_table; struct rte_src *src; diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c index 52400762..f52eb2f0 100644 --- a/proto/bgp/bgp.c +++ b/proto/bgp/bgp.c @@ -760,25 +760,25 @@ bgp_handle_graceful_restart(struct bgp_proto *p) { case BGP_GRS_NONE: c->gr_active = BGP_GRS_ACTIVE; - rt_refresh_begin(c->c.table, &c->c); + rt_refresh_begin(c->c.table, &c->c.in_req); break; case BGP_GRS_ACTIVE: - rt_refresh_end(c->c.table, &c->c); - rt_refresh_begin(c->c.table, &c->c); + rt_refresh_end(c->c.table, &c->c.in_req); + rt_refresh_begin(c->c.table, &c->c.in_req); break; case BGP_GRS_LLGR: - rt_refresh_begin(c->c.table, &c->c); - rt_modify_stale(c->c.table, &c->c); + rt_refresh_begin(c->c.table, &c->c.in_req); + rt_modify_stale(c->c.table, &c->c.in_req); break; } } else { /* Just flush the routes */ - rt_refresh_begin(c->c.table, &c->c); - rt_refresh_end(c->c.table, &c->c); + rt_refresh_begin(c->c.table, &c->c.in_req); + rt_refresh_end(c->c.table, &c->c.in_req); } /* Reset bucket and prefix tables */ @@ -819,7 +819,7 @@ bgp_graceful_restart_done(struct bgp_channel *c) BGP_TRACE(D_EVENTS, "Neighbor graceful restart done"); tm_stop(c->stale_timer); - rt_refresh_end(c->c.table, &c->c); + rt_refresh_end(c->c.table, &c->c.in_req); } /** @@ -861,7 +861,7 @@ bgp_graceful_restart_timeout(timer *t) /* Channel is in GR, and supports LLGR -> start LLGR */ c->gr_active = BGP_GRS_LLGR; tm_start(c->stale_timer, c->stale_time S); - rt_modify_stale(c->c.table, &c->c); + rt_modify_stale(c->c.table, &c->c.in_req); } } else @@ -899,10 +899,10 @@ bgp_refresh_begin(struct bgp_channel *c) { log(L_WARN "%s: BEGIN-OF-RR received before END-OF-RIB, ignoring", p->p.name); return; } c->load_state = BFS_REFRESHING; - rt_refresh_begin(c->c.table, &c->c); + rt_refresh_begin(c->c.table, &c->c.in_req); if (c->c.in_table) - rt_refresh_begin(c->c.in_table, &c->c); + rt_refresh_begin(c->c.in_table, &c->c.in_req); } /** @@ -923,7 +923,7 @@ bgp_refresh_end(struct bgp_channel *c) { log(L_WARN "%s: END-OF-RR received without prior BEGIN-OF-RR, ignoring", p->p.name); return; } c->load_state = BFS_NONE; - rt_refresh_end(c->c.table, &c->c); + rt_refresh_end(c->c.table, &c->c.in_req); if (c->c.in_table) rt_prune_sync(c->c.in_table, 0); diff --git a/proto/pipe/pipe.c b/proto/pipe/pipe.c index 55a0b526..bbcd8f0d 100644 --- a/proto/pipe/pipe.c +++ b/proto/pipe/pipe.c @@ -82,7 +82,7 @@ pipe_preexport(struct channel *c, rte *e) struct pipe_proto *p = (void *) c->proto; /* Avoid direct loopbacks */ - if (e->sender == c) + if (e->sender == c->in_req.hook) return -1; /* Indirection check */ @@ -212,10 +212,18 @@ pipe_get_status(struct proto *P, byte *buf) static void pipe_show_stats(struct pipe_proto *p) { - struct import_stats *s1i = &p->pri->import_stats; - struct export_stats *s1e = &p->pri->export_stats; - struct import_stats *s2i = &p->sec->import_stats; - struct export_stats *s2e = &p->sec->export_stats; + struct channel_import_stats *s1i = &p->pri->import_stats; + struct channel_export_stats *s1e = &p->pri->export_stats; + struct channel_import_stats *s2i = &p->sec->import_stats; + struct channel_export_stats *s2e = &p->sec->export_stats; + + struct rt_import_stats *rs1i = p->pri->in_req.hook ? &p->pri->in_req.hook->stats : NULL; + struct rt_export_stats *rs1e = p->pri->out_req.hook ? &p->pri->out_req.hook->stats : NULL; + struct rt_import_stats *rs2i = p->sec->in_req.hook ? &p->sec->in_req.hook->stats : NULL; + struct rt_export_stats *rs2e = p->sec->out_req.hook ? &p->sec->out_req.hook->stats : NULL; + + u32 pri_routes = p->pri->in_limit.count; + u32 sec_routes = p->sec->in_limit.count; /* * Pipe stats (as anything related to pipes) are a bit tricky. There @@ -239,24 +247,22 @@ pipe_show_stats(struct pipe_proto *p) */ cli_msg(-1006, " Routes: %u imported, %u exported", - s1i->routes, s2i->routes); + pri_routes, sec_routes); cli_msg(-1006, " Route change stats: received rejected filtered ignored accepted"); cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u", - s2e->updates_received, s2e->updates_rejected + s1i->updates_invalid, - s2e->updates_filtered, s1i->updates_ignored, s1i->updates_accepted); + rs2e->updates_received, s2e->updates_rejected + s1i->updates_invalid, + s2e->updates_filtered, rs1i->updates_ignored, rs1i->updates_accepted); cli_msg(-1006, " Import withdraws: %10u %10u --- %10u %10u", - s2e->withdraws_received, s1i->withdraws_invalid, - s1i->withdraws_ignored, s1i->withdraws_accepted); + rs2e->withdraws_received, s1i->withdraws_invalid, + rs1i->withdraws_ignored, rs1i->withdraws_accepted); cli_msg(-1006, " Export updates: %10u %10u %10u %10u %10u", - s1e->updates_received, s1e->updates_rejected + s2i->updates_invalid, - s1e->updates_filtered, s2i->updates_ignored, s2i->updates_accepted); + rs1e->updates_received, s1e->updates_rejected + s2i->updates_invalid, + s1e->updates_filtered, rs2i->updates_ignored, rs2i->updates_accepted); cli_msg(-1006, " Export withdraws: %10u %10u --- %10u %10u", - s1e->withdraws_received, s2i->withdraws_invalid, - s2i->withdraws_ignored, s2i->withdraws_accepted); + rs1e->withdraws_received, s2i->withdraws_invalid, + rs2i->withdraws_ignored, rs2i->withdraws_accepted); } -static const char *pipe_feed_state[] = { [ES_DOWN] = "down", [ES_FEEDING] = "feed", [ES_READY] = "up" }; - static void pipe_show_proto_info(struct proto *P) { @@ -265,13 +271,17 @@ pipe_show_proto_info(struct proto *P) cli_msg(-1006, " Channel %s", "main"); cli_msg(-1006, " Table: %s", p->pri->table->name); cli_msg(-1006, " Peer table: %s", p->sec->table->name); - cli_msg(-1006, " Import state: %s", pipe_feed_state[p->sec->export_state]); - cli_msg(-1006, " Export state: %s", pipe_feed_state[p->pri->export_state]); + cli_msg(-1006, " Import state: %s", rt_export_state_name(rt_export_get_state(p->sec->out_req.hook))); + cli_msg(-1006, " Export state: %s", rt_export_state_name(rt_export_get_state(p->pri->out_req.hook))); cli_msg(-1006, " Import filter: %s", filter_name(p->sec->out_filter)); cli_msg(-1006, " Export filter: %s", filter_name(p->pri->out_filter)); - channel_show_limit(&p->pri->in_limit, "Import limit:"); - channel_show_limit(&p->sec->in_limit, "Export limit:"); + + + channel_show_limit(&p->pri->in_limit, "Import limit:", + (p->pri->limit_active & (1 << PLD_IN)), p->pri->limit_actions[PLD_IN]); + channel_show_limit(&p->sec->in_limit, "Export limit:", + (p->sec->limit_active & (1 << PLD_IN)), p->sec->limit_actions[PLD_IN]); if (P->proto_state != PS_DOWN) pipe_show_stats(p); diff --git a/proto/rpki/packets.c b/proto/rpki/packets.c index d246dd50..4a52b54b 100644 --- a/proto/rpki/packets.c +++ b/proto/rpki/packets.c @@ -661,9 +661,9 @@ rpki_handle_cache_response_pdu(struct rpki_cache *cache, const struct pdu_cache_ * a refresh cycle. */ if (cache->p->roa4_channel) - rt_refresh_begin(cache->p->roa4_channel->table, cache->p->roa4_channel); + rt_refresh_begin(cache->p->roa4_channel->table, &cache->p->roa4_channel->in_req); if (cache->p->roa6_channel) - rt_refresh_begin(cache->p->roa6_channel->table, cache->p->roa6_channel); + rt_refresh_begin(cache->p->roa6_channel->table, &cache->p->roa6_channel->in_req); cache->p->refresh_channels = 1; } @@ -846,9 +846,9 @@ rpki_handle_end_of_data_pdu(struct rpki_cache *cache, const struct pdu_end_of_da { cache->p->refresh_channels = 0; if (cache->p->roa4_channel) - rt_refresh_end(cache->p->roa4_channel->table, cache->p->roa4_channel); + rt_refresh_end(cache->p->roa4_channel->table, &cache->p->roa4_channel->in_req); if (cache->p->roa6_channel) - rt_refresh_end(cache->p->roa6_channel->table, cache->p->roa6_channel); + rt_refresh_end(cache->p->roa6_channel->table, &cache->p->roa6_channel->in_req); } cache->last_update = current_time(); @@ -924,6 +924,9 @@ rpki_rx_hook(struct birdsock *sk, uint size) struct rpki_cache *cache = sk->data; struct rpki_proto *p = cache->p; + if ((p->p.proto_state == PS_DOWN) || (p->cache != cache)) + return 0; + byte *pkt_start = sk->rbuf; byte *end = pkt_start + size; @@ -980,6 +983,8 @@ rpki_err_hook(struct birdsock *sk, int error_num) CACHE_TRACE(D_EVENTS, cache, "The other side closed a connection"); } + if (cache->p->cache != cache) + return; rpki_cache_change_state(cache, RPKI_CS_ERROR_TRANSPORT); } @@ -999,6 +1004,9 @@ rpki_tx_hook(sock *sk) { struct rpki_cache *cache = sk->data; + if (cache->p->cache != cache) + return; + while (rpki_fire_tx(cache) > 0) ; } @@ -1008,6 +1016,9 @@ rpki_connected_hook(sock *sk) { struct rpki_cache *cache = sk->data; + if (cache->p->cache != cache) + return; + CACHE_TRACE(D_EVENTS, cache, "Connected"); proto_notify_state(&cache->p->p, PS_UP); diff --git a/proto/rpki/rpki.c b/proto/rpki/rpki.c index 72fbc967..95066499 100644 --- a/proto/rpki/rpki.c +++ b/proto/rpki/rpki.c @@ -384,6 +384,9 @@ rpki_refresh_hook(timer *tm) { struct rpki_cache *cache = tm->data; + if (cache->p->cache != cache) + return; + CACHE_DBG(cache, "%s", rpki_cache_state_to_str(cache->state)); switch (cache->state) @@ -430,6 +433,9 @@ rpki_retry_hook(timer *tm) { struct rpki_cache *cache = tm->data; + if (cache->p->cache != cache) + return; + CACHE_DBG(cache, "%s", rpki_cache_state_to_str(cache->state)); switch (cache->state) @@ -475,6 +481,9 @@ rpki_expire_hook(timer *tm) { struct rpki_cache *cache = tm->data; + if (cache->p->cache != cache) + return; + if (!cache->last_update) return; diff --git a/proto/static/static.c b/proto/static/static.c index 6e258f6f..f8ac9c81 100644 --- a/proto/static/static.c +++ b/proto/static/static.c @@ -507,19 +507,13 @@ static_shutdown(struct proto *P) WALK_LIST(r, cf->routes) static_reset_rte(p, r); - return PS_DOWN; -} - -static void -static_cleanup(struct proto *P) -{ - struct static_proto *p = (void *) P; - if (p->igp_table_ip4) rt_unlock_table(p->igp_table_ip4); if (p->igp_table_ip6) rt_unlock_table(p->igp_table_ip6); + + return PS_DOWN; } static void @@ -777,7 +771,6 @@ struct protocol proto_static = { .dump = static_dump, .start = static_start, .shutdown = static_shutdown, - .cleanup = static_cleanup, .reconfigure = static_reconfigure, .copy_config = static_copy_config, .get_route_info = static_get_route_info, diff --git a/sysdep/unix/krt.c b/sysdep/unix/krt.c index b98e7ec0..40a58442 100644 --- a/sysdep/unix/krt.c +++ b/sysdep/unix/krt.c @@ -542,21 +542,27 @@ krt_is_installed(struct krt_proto *p, net *n) return n->routes && bmap_test(&p->p.main_channel->export_map, n->routes->rte.id); } -static void -krt_flush_routes(struct krt_proto *p) +static uint +rte_feed_count(net *n) { - struct rtable *t = p->p.main_channel->table; + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTE_OR_NULL(e))) + count++; + return count; +} - KRT_TRACE(p, D_EVENTS, "Flushing kernel routes"); - FIB_WALK(&t->fib, net, n) +static void +rte_feed_obtain(net *n, rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTE_OR_NULL(e))) { - if (krt_is_installed(p, n)) - { - /* FIXME: this does not work if gw is changed in export filter */ - krt_replace_rte(p, n->n.addr, NULL, &n->routes->rte); - } + ASSERT_DIE(i < count); + feed[i++] = &e->rte; } - FIB_WALK_END; + ASSERT_DIE(i == count); } static struct rte * @@ -566,7 +572,15 @@ krt_export_net(struct krt_proto *p, net *net) const struct filter *filter = c->out_filter; if (c->ra_mode == RA_MERGED) - return rt_export_merged_show(c, net, krt_filter_lp); + { + uint count = rte_feed_count(net); + if (!count) + return NULL; + + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(net, feed, count); + return rt_export_merged(c, feed, count, krt_filter_lp, 1); + } static _Thread_local rte rt; rt = net->routes->rte; @@ -637,6 +651,9 @@ krt_got_route(struct krt_proto *p, rte *e, s8 src) #endif /* The rest is for KRT_SRC_BIRD (or KRT_SRC_UNKNOWN) */ + /* Deleting all routes if flush is requested */ + if (p->flush_routes) + goto delete; /* We wait for the initial feed to have correct installed state */ if (!p->ready) @@ -729,6 +746,17 @@ krt_prune(struct krt_proto *p) p->initialized = 1; } +static void +krt_flush_routes(struct krt_proto *p) +{ + KRT_TRACE(p, D_EVENTS, "Flushing kernel routes"); + p->flush_routes = 1; + krt_init_scan(p); + krt_do_scan(p); + /* No prune! */ + p->flush_routes = 0; +} + void krt_got_route_async(struct krt_proto *p, rte *e, int new, s8 src) { diff --git a/sysdep/unix/krt.h b/sysdep/unix/krt.h index cd4bd07d..f6ad6fde 100644 --- a/sysdep/unix/krt.h +++ b/sysdep/unix/krt.h @@ -66,6 +66,7 @@ struct krt_proto { byte ready; /* Initial feed has been finished */ byte initialized; /* First scan has been finished */ byte reload; /* Next scan is doing reload */ + byte flush_routes; /* Scanning to flush */ }; extern pool *krt_pool; |