diff options
Diffstat (limited to 'nest')
| -rw-r--r-- | nest/config.Y | 7 | ||||
| -rw-r--r-- | nest/limit.h | 49 | ||||
| -rw-r--r-- | nest/proto.c | 702 | ||||
| -rw-r--r-- | nest/protocol.h | 150 | ||||
| -rw-r--r-- | nest/rt-attr.c | 6 | ||||
| -rw-r--r-- | nest/rt-dev.c | 13 | ||||
| -rw-r--r-- | nest/rt-show.c | 73 | ||||
| -rw-r--r-- | nest/rt-table.c | 1902 | ||||
| -rw-r--r-- | nest/rt.h | 206 |
9 files changed, 1804 insertions, 1304 deletions
diff --git a/nest/config.Y b/nest/config.Y index 773673e8..2d73b4c7 100644 --- a/nest/config.Y +++ b/nest/config.Y @@ -112,7 +112,7 @@ proto_postconfig(void) CF_DECLS -CF_KEYWORDS(ROUTER, ID, HOSTNAME, PROTOCOL, TEMPLATE, PREFERENCE, DISABLED, DEBUG, ALL, OFF, DIRECT) +CF_KEYWORDS(ROUTER, ID, HOSTNAME, PROTOCOL, TEMPLATE, PREFERENCE, DISABLED, DEBUG, ALL, OFF, DIRECT, PIPE) CF_KEYWORDS(INTERFACE, IMPORT, EXPORT, FILTER, NONE, VRF, DEFAULT, TABLE, STATES, ROUTES, FILTERS) CF_KEYWORDS(IPV4, IPV6, VPN4, VPN6, ROA4, ROA6, FLOW4, FLOW6, SADR, MPLS) CF_KEYWORDS(RECEIVE, LIMIT, ACTION, WARN, BLOCK, RESTART, DISABLE, KEEP, FILTERED, RPKI) @@ -371,6 +371,7 @@ debug_default: DEBUG PROTOCOLS debug_mask { new_config->proto_default_debug = $3; } | DEBUG CHANNELS debug_mask { new_config->channel_default_debug = $3; } | DEBUG COMMANDS expr { new_config->cli_debug = $3; } + | DEBUG PIPE bool { new_config->pipe_debug = $3; } ; /* MRTDUMP PROTOCOLS is in systep/unix/config.Y */ @@ -850,8 +851,10 @@ CF_CLI(DUMP NEIGHBORS,,, [[Dump neighbor cache]]) { neigh_dump_all(); cli_msg(0, ""); } ; CF_CLI(DUMP ATTRIBUTES,,, [[Dump attribute cache]]) { ea_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 72613f8d..77817888 100644 --- a/nest/proto.c +++ b/nest/proto.c @@ -42,8 +42,7 @@ 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; @@ -51,15 +50,17 @@ 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; } @@ -67,10 +68,46 @@ 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 @@ -140,6 +177,15 @@ 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 @@ -164,12 +210,14 @@ 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; @@ -180,7 +228,6 @@ 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; @@ -202,6 +249,7 @@ 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); } @@ -222,7 +270,7 @@ 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 @@ -231,7 +279,7 @@ 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 @@ -243,71 +291,6 @@ 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->stats.exp_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; @@ -325,14 +308,12 @@ 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 */ @@ -438,32 +419,189 @@ 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); + + memset(&c->export_stats, 0, sizeof(struct channel_export_stats)); - channel_schedule_feed(c, 1); /* Sets ES_FEEDING */ + 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); - c->export_state = ES_DOWN; - c->stats.exp_routes = 0; - bmap_reset(&c->export_map, 1024); + req->hook = NULL; + + 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); @@ -489,23 +627,6 @@ 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) @@ -537,20 +658,11 @@ 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); - memset(&c->stats, 0, sizeof(struct proto_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 @@ -565,9 +677,31 @@ 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) @@ -576,28 +710,21 @@ 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); 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->stats.imp_routes + c->stats.filt_routes) != 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->stats, 0, sizeof(struct proto_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; @@ -616,7 +743,6 @@ 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) @@ -628,20 +754,11 @@ 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: @@ -656,23 +773,24 @@ 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; @@ -697,35 +815,16 @@ 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"); @@ -736,8 +835,8 @@ 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 = { @@ -840,19 +939,19 @@ 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; @@ -964,8 +1063,8 @@ 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); @@ -1517,7 +1616,7 @@ 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 */ @@ -1607,7 +1706,11 @@ 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) @@ -1617,6 +1720,9 @@ 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)) @@ -1735,88 +1841,104 @@ 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->stats.imp_routes + c->stats.filt_routes; + 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->stats.imp_routes > l->limit)) - channel_notify_limit(c, l, PLD_IN, c->stats.imp_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->stats.exp_routes > l->limit)) - channel_notify_limit(c, l, PLD_OUT, c->stats.exp_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 @@ -1868,8 +1990,6 @@ 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)) @@ -1964,38 +2084,58 @@ proto_state_name(struct proto *p) static void channel_show_stats(struct channel *c) { - struct proto_stats *s = &c->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", - s->imp_routes, s->filt_routes, s->exp_routes, s->pref_routes); + in_routes, (rx_routes - in_routes), out_routes, SRI(pref)); else cli_msg(-1006, " Routes: %u imported, %u exported, %u preferred", - s->imp_routes, s->exp_routes, s->pref_routes); - - cli_msg(-1006, " Route change stats: received rejected filtered ignored accepted"); - cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u", - s->imp_updates_received, s->imp_updates_invalid, - s->imp_updates_filtered, s->imp_updates_ignored, - s->imp_updates_accepted); - cli_msg(-1006, " Import withdraws: %10u %10u --- %10u %10u", - s->imp_withdraws_received, s->imp_withdraws_invalid, - s->imp_withdraws_ignored, s->imp_withdraws_accepted); - cli_msg(-1006, " Export updates: %10u %10u %10u --- %10u", - s->exp_updates_received, s->exp_updates_rejected, - s->exp_updates_filtered, s->exp_updates_accepted); - cli_msg(-1006, " Export withdraws: %10u --- --- --- %10u", - s->exp_withdraws_received, s->exp_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 @@ -2003,6 +2143,8 @@ 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)); @@ -2013,9 +2155,9 @@ 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 8f0cc4b4..aeb60ac6 100644 --- a/nest/protocol.h +++ b/nest/protocol.h @@ -13,6 +13,7 @@ #include "lib/resource.h" #include "lib/event.h" #include "nest/rt.h" +#include "nest/limit.h" #include "conf/conf.h" struct iface; @@ -58,7 +59,6 @@ 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_*) */ @@ -114,31 +114,6 @@ struct proto_config { }; /* Protocol statistics */ -struct proto_stats { - /* Import - from protocol to core */ - u32 imp_routes; /* Number of routes successfully imported to the (adjacent) routing table */ - u32 filt_routes; /* Number of routes rejected in import filter but kept in the routing table */ - u32 pref_routes; /* Number of routes selected as best in the (adjacent) routing table */ - u32 imp_updates_received; /* Number of route updates received */ - u32 imp_updates_invalid; /* Number of route updates rejected as invalid */ - u32 imp_updates_filtered; /* Number of route updates rejected by filters */ - u32 imp_updates_ignored; /* Number of route updates rejected as already in route table */ - u32 imp_updates_accepted; /* Number of route updates accepted and imported */ - u32 imp_withdraws_received; /* Number of route withdraws received */ - u32 imp_withdraws_invalid; /* Number of route withdraws rejected as invalid */ - u32 imp_withdraws_ignored; /* Number of route withdraws rejected as already not in route table */ - u32 imp_withdraws_accepted; /* Number of route withdraws accepted and processed */ - - /* Export - from core to protocol */ - u32 exp_routes; /* Number of routes successfully exported to the protocol */ - u32 exp_updates_received; /* Number of route updates received */ - u32 exp_updates_rejected; /* Number of route updates rejected by protocol */ - u32 exp_updates_filtered; /* Number of route updates rejected by filters */ - u32 exp_updates_accepted; /* Number of route updates accepted and exported */ - u32 exp_withdraws_received; /* Number of route withdraws received */ - u32 exp_withdraws_accepted; /* Number of route withdraws accepted and processed */ -}; - struct proto { node n; /* Node in global proto_list */ struct protocol *proto; /* Protocol */ @@ -193,9 +168,9 @@ struct proto { void (*if_notify)(struct proto *, unsigned flags, struct iface *i); void (*ifa_notify)(struct proto *, unsigned flags, struct ifa *a); - void (*rt_notify)(struct proto *, struct channel *, struct network *net, struct rte *new, struct rte *old); + void (*rt_notify)(struct proto *, struct channel *, const net_addr *net, struct rte *new, const struct rte *old); void (*neigh_notify)(struct neighbor *neigh); - int (*preexport)(struct proto *, struct rte *rt); + int (*preexport)(struct channel *, struct rte *rt); void (*reload_routes)(struct channel *); void (*feed_begin)(struct channel *, int initial); void (*feed_end)(struct channel *); @@ -214,10 +189,10 @@ struct proto { int (*rte_recalculate)(struct rtable *, struct network *, struct rte *, struct rte *, struct rte *); int (*rte_better)(struct rte *, struct rte *); int (*rte_mergable)(struct rte *, struct rte *); - struct rte * (*rte_modify)(struct rte *, struct linpool *); + struct rte *(*rte_modify)(struct rte *, struct linpool *); void (*rte_insert)(struct network *, struct rte *); void (*rte_remove)(struct network *, struct rte *); - u32 (*rte_igp_metric)(struct rte *); + u32 (*rte_igp_metric)(const struct rte *); /* Hic sunt protocol-specific data */ }; @@ -257,7 +232,7 @@ 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); @@ -412,19 +387,30 @@ 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 */ @@ -466,6 +452,7 @@ 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 */ @@ -482,7 +469,6 @@ 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; @@ -491,14 +477,39 @@ struct channel { struct rtable *table; const struct filter *in_filter; /* Input filter */ const struct filter *out_filter; /* Output filter */ - struct bmap export_map; /* Keeps track which routes passed 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 proto_stats stats; /* Per-channel protocol statistics */ + 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 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 */ u8 net_type; /* Routing table network type (NET_*), 0 for undefined */ @@ -511,10 +522,7 @@ 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 */ @@ -524,7 +532,7 @@ struct channel { struct rtable *in_table; /* Internal table for received routes */ struct event *reload_event; /* Event responsible for reloading from in_table */ struct fib_iterator reload_fit; /* FIB iterator in in_table used during reloading */ - struct rte *reload_next_rte; /* Route iterator in in_table used during reloading */ + struct rte_storage *reload_next_rte; /* Route iterator in in_table used during reloading */ u8 reload_active; /* Iterator reload_fit is linked */ u8 reload_pending; /* Reloading and another reload is scheduled */ @@ -562,34 +570,34 @@ 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) @@ -607,25 +615,11 @@ 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); void *channel_config_get(const struct channel_class *cc, const char *name, uint net_type, struct proto_config *proto); int channel_reconfigure(struct channel *c, struct channel_config *cf); - -/* Moved from route.h to avoid dependency conflicts */ -static inline void rte_update(struct proto *p, const net_addr *n, rte *new) { rte_update2(p->main_channel, n, new, p->main_source); } - -static inline void -rte_update3(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) -{ - if (c->in_table && !rte_update_in(c, n, new, src)) - return; - - rte_update2(c, n, new, src); -} - - #endif diff --git a/nest/rt-attr.c b/nest/rt-attr.c index 6927751f..31e2057e 100644 --- a/nest/rt-attr.c +++ b/nest/rt-attr.c @@ -172,6 +172,12 @@ struct ea_class ea_gen_flowspec_valid = { .readonly = 1, }; +const char * flowspec_valid_names[FLOWSPEC__MAX] = { + [FLOWSPEC_UNKNOWN] = "unknown", + [FLOWSPEC_VALID] = "", + [FLOWSPEC_INVALID] = "invalid", +}; + pool *rta_pool; static slab *rte_src_slab; diff --git a/nest/rt-dev.c b/nest/rt-dev.c index 63f8a26b..7f45985f 100644 --- a/nest/rt-dev.c +++ b/nest/rt-dev.c @@ -67,12 +67,10 @@ dev_ifa_notify(struct proto *P, uint flags, struct ifa *ad) /* Use iface ID as local source ID */ struct rte_src *src = rt_get_source(P, ad->iface->index); - rte_update2(c, net, NULL, src); + rte_update(c, net, NULL, src); } else if (flags & IF_CHANGE_UP) { - rte *e; - DBG("dev_if_notify: %s:%I going up\n", ad->iface->name, ad->ip); if (cf->check_link && !(ad->iface->flags & IF_LINK_UP)) @@ -91,9 +89,12 @@ dev_ifa_notify(struct proto *P, uint flags, struct ifa *ad) ea_set_attr_u32(&ea, &ea_gen_source, 0, RTS_DEVICE); ea_set_attr_data(&ea, &ea_gen_nexthop, 0, nhad.ad.data, nhad.ad.length); - e = rte_get_temp(rta_lookup(ea), src); - e->pflags = 0; - rte_update2(c, net, e, src); + rte e0 = { + .attrs = ea, + .src = src, + }; + + rte_update(c, net, &e0, src); } } diff --git a/nest/rt-show.c b/nest/rt-show.c index 2cbc500a..f3852d17 100644 --- a/nest/rt-show.c +++ b/nest/rt-show.c @@ -45,9 +45,11 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary ea_list *a = e->attrs; int sync_error = d->kernel ? krt_get_sync_error(d->kernel, e) : 0; void (*get_route_info)(struct rte *, byte *buf); - eattr *nhea = ea_find(a, &ea_gen_nexthop); + eattr *nhea = net_type_match(e->net, NB_DEST) ? + ea_find(a, &ea_gen_nexthop) : NULL; struct nexthop_adata *nhad = nhea ? (struct nexthop_adata *) nhea->u.ptr : NULL; - int dest = NEXTHOP_IS_REACHABLE(nhad) ? RTD_UNICAST : nhad->dest; + int dest = nhad ? (NEXTHOP_IS_REACHABLE(nhad) ? RTD_UNICAST : nhad->dest) : RTD_NONE; + int flowspec_valid = net_is_flow(e->net) ? rt_get_flowspec_valid(e) : FLOWSPEC_UNKNOWN; tm_format_time(tm, &config->tf_route, e->lastmod); ip_addr a_from = ea_get_ip(a, &ea_gen_from, IPA_NONE); @@ -69,8 +71,9 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary if (d->last_table != d->tab) rt_show_table(c, d); - cli_printf(c, -1007, "%-20s %s [%s %s%s]%s%s", ia, rta_dest_name(dest), - e->src->proto->name, tm, from, primary ? (sync_error ? " !" : " *") : "", info); + cli_printf(c, -1007, "%-20s %s [%s %s%s]%s%s", ia, + net_is_flow(e->net) ? flowspec_valid_name(flowspec_valid) : rta_dest_name(dest), + e->src->proto->name, tm, from, primary ? (sync_error ? " !" : " *") : "", info); if (dest == RTD_UNICAST) NEXTHOP_WALK(nh, nhad) @@ -102,10 +105,32 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary ea_show_list(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) { - rte *e, *ee; byte ia[NET_MAX_TEXT_LENGTH+1]; struct channel *ec = d->tab->export_channel; @@ -117,9 +142,9 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) int first_show = 1; int pass = 0; - for (e = n->routes; e; e = e->next) + for (struct rte_storage *er = n->routes; er; er = er->next) { - if (rte_is_filtered(e) != d->filtered) + if (rte_is_filtered(&er->rte) != d->filtered) continue; d->rt_counter++; @@ -129,15 +154,15 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) if (pass) continue; - ee = e; + 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) { - if (!bmap_test(&ec->export_map, ee->id)) + if (!bmap_test(&ec->export_map, e.id)) goto skip; // if (ec->ra_mode != RA_ANY) @@ -146,17 +171,24 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) else if ((d->export_mode == RSEM_EXPORT) && (ec->ra_mode == RA_MERGED)) { /* Special case for merged export */ - rte *rt_free; - e = rt_export_merged(ec, n, &rt_free, c->show_pool, 1); pass = 1; + 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 (!e) - { e = ee; goto skip; } + if (em) + e = *em; + else + goto skip; } else if (d->export_mode) { struct proto *ep = ec->proto; - int ic = ep->preexport ? ep->preexport(ep, e) : 0; + int ic = ep->preexport ? ep->preexport(ec, &e) : 0; if (ec->ra_mode == RA_OPTIMAL || ec->ra_mode == RA_MERGED) pass = 1; @@ -182,7 +214,7 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) } } - if (d->show_protocol && (d->show_protocol != e->src->proto)) + if (d->show_protocol && (d->show_protocol != e.src->proto)) goto skip; if (f_run(d->filter, &e, 0) > F_ACCEPT) @@ -195,18 +227,13 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) else ia[0] = 0; - rt_show_rte(c, ia, e, d, (e->net->routes == ee)); + rt_show_rte(c, ia, &e, d, (n->routes == er)); first_show = 0; } d->show_counter++; skip: - if (e != ee) - { - rte_free(e); - e = ee; - } lp_flush(c->show_pool); if (d->primary_only) @@ -378,7 +405,7 @@ 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 e3dab120..1631e00f 100644 --- a/nest/rt-table.c +++ b/nest/rt-table.c @@ -112,19 +112,56 @@ pool *rt_table_pool; -static slab *rte_slab; static linpool *rte_update_pool; list routing_tables; +list deleted_routing_tables; static void rt_free_hostcache(rtable *tab); static void rt_notify_hostcache(rtable *tab, net *net); static void rt_update_hostcache(rtable *tab); static void rt_next_hop_update(rtable *tab); +static inline void rt_next_hop_resolve_rte(rte *r); +static inline void rt_flowspec_resolve_rte(rte *r, struct channel *c); static inline void rt_prune_table(rtable *tab); static inline void rt_schedule_notify(rtable *tab); static void rt_flowspec_notify(rtable *tab, net *net); -static inline rte *rt_next_hop_update_rte(rtable *tab, rte *old); +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 inline struct rte_storage *rt_next_hop_update_rte(rtable *tab, net *n, rte *old); static struct hostentry *rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep); static void @@ -395,7 +432,7 @@ net_roa_check_ip4_trie(rtable *tab, const net_addr_ip4 *px, u32 asn) net_addr_roa4 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa4(roa, &roa0) && rte_is_valid(r->routes)) + if (net_equal_prefix_roa4(roa, &roa0) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) @@ -422,7 +459,7 @@ net_roa_check_ip4_fib(rtable *tab, const net_addr_ip4 *px, u32 asn) net_addr_roa4 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa4(roa, &n) && rte_is_valid(r->routes)) + if (net_equal_prefix_roa4(roa, &n) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) @@ -455,7 +492,7 @@ net_roa_check_ip6_trie(rtable *tab, const net_addr_ip6 *px, u32 asn) net_addr_roa6 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa6(roa, &roa0) && rte_is_valid(r->routes)) + if (net_equal_prefix_roa6(roa, &roa0) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) @@ -482,7 +519,7 @@ net_roa_check_ip6_fib(rtable *tab, const net_addr_ip6 *px, u32 asn) net_addr_roa6 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa6(roa, &n) && rte_is_valid(r->routes)) + if (net_equal_prefix_roa6(roa, &n) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) @@ -541,105 +578,52 @@ net_roa_check(rtable *tab, const net_addr *n, u32 asn) * @net: network node * @src: route source * - * The rte_find() function returns a route for destination @net - * which is from route source @src. + * The rte_find() function returns a pointer to a route for destination @net + * which is from route source @src. List end pointer is returned if no route is found. */ -rte * +static struct rte_storage ** rte_find(net *net, struct rte_src *src) { - rte *e = net->routes; + struct rte_storage **e = &net->routes; - while (e && e->src != src) - e = e->next; - return e; -} - -/** - * rte_get_temp - get a temporary &rte - * @a: attributes to assign to the new route (a &rta; in case it's - * un-cached, rte_update() will create a cached copy automatically) - * - * Create a temporary &rte and bind it with the attributes @a. - * Also set route preference to the default preference set for - * the protocol. - */ -rte * -rte_get_temp(ea_list *a, struct rte_src *src) -{ - rte *e = sl_alloc(rte_slab); + while ((*e) && (*e)->rte.src != src) + e = &(*e)->next; - e->attrs = a; - e->id = 0; - e->flags = 0; - rt_lock_source(e->src = src); return e; } -rte * -rte_do_cow(rte *r) -{ - rte *e = sl_alloc(rte_slab); - memcpy(e, r, sizeof(rte)); +struct rte_storage * +rte_store(const rte *r, net *net, rtable *tab) +{ + struct rte_storage *e = sl_alloc(tab->rte_slab); - rt_lock_source(e->src); - e->attrs = rta_clone(r->attrs); - e->flags = 0; - return e; -} + e->rte = *r; + e->rte.net = net->n.addr; -/** - * rte_cow_rta - get a private writable copy of &rte with writable &rta - * @r: a route entry to be copied - * @lp: a linpool from which to allocate &rta - * - * rte_cow_rta() takes a &rte and prepares it and associated &rta for - * modification. There are three possibilities: First, both &rte and &rta are - * private copies, in that case they are returned unchanged. Second, &rte is - * private copy, but &rta is cached, in that case &rta is duplicated using - * rta_do_cow(). Third, both &rte is shared and &rta is cached, in that case - * both structures are duplicated by rte_do_cow() and rta_do_cow(). - * - * Note that in the second case, cached &rta loses one reference, while private - * copy created by rta_do_cow() is a shallow copy sharing indirect data (eattrs, - * nexthops, ...) with it. To work properly, original shared &rta should have - * another reference during the life of created private copy. - * - * Result: a pointer to the new writable &rte with writable &rta. - */ -rte * -rte_cow_rta(rte *r) -{ - if (!rta_is_cached(r->attrs)) - return r; + rt_lock_source(e->rte.src); - r = rte_cow(r); + if (ea_is_cached(e->rte.attrs)) + e->rte.attrs = rta_clone(e->rte.attrs); + else + e->rte.attrs = rta_lookup(e->rte.attrs); - /* This looks stupid but should DWIW. */ - rta_free(r->attrs); - return r; + return e; } /** * rte_free - delete a &rte - * @e: &rte to be deleted + * @e: &struct rte_storage to be deleted + * @tab: the table which the rte belongs to * * rte_free() deletes the given &rte from the routing table it's linked to. */ -void -rte_free(rte *e) -{ - rt_unlock_source(e->src); - if (rta_is_cached(e->attrs)) - rta_free(e->attrs); - sl_free(e); -} -static inline void -rte_free_quick(rte *e) +void +rte_free(struct rte_storage *e) { - rt_unlock_source(e->src); - rta_free(e->attrs); + rt_unlock_source(e->rte.src); + rta_free(e->rte.attrs); sl_free(e); } @@ -695,162 +679,187 @@ rte_mergable(rte *pri, rte *sec) } static void -rte_trace(struct channel *c, rte *e, int dir, char *msg) +rte_trace(const char *name, const rte *e, int dir, const char *msg) { - log(L_TRACE "%s.%s %c %s %N %uL %uG %s", - c->proto->name, c->name ?: "?", dir, msg, e->net->n.addr, e->src->private_id, e->src->global_id, + 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(rte_dest(e))); } static inline void -rte_trace_in(uint flag, struct channel *c, rte *e, char *msg) +channel_rte_trace_in(uint flag, struct channel *c, const rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) - rte_trace(c, e, '>', msg); + rte_trace(c->in_req.name, e, '>', msg); } static inline void -rte_trace_out(uint flag, struct channel *c, rte *e, char *msg) +channel_rte_trace_out(uint flag, struct channel *c, const rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) - rte_trace(c, e, '<', msg); + rte_trace(c->out_req.name, e, '<', msg); +} + +static inline void +rt_rte_trace_in(uint flag, struct rt_import_request *req, const rte *e, const char *msg) +{ + if (req->trace_routes & flag) + rte_trace(req->name, e, '>', msg); +} + +#if 0 +// seems to be unused at all +static inline void +rt_rte_trace_out(uint flag, struct rt_export_request *req, const rte *e, const char *msg) +{ + if (req->trace_routes & flag) + rte_trace(req->name, e, '<', msg); +} +#endif + +static uint +rte_feed_count(net *n) +{ + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTE_OR_NULL(e))) + count++; + return count; +} + +static void +rte_feed_obtain(net *n, struct rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTE_OR_NULL(e))) + { + ASSERT_DIE(i < count); + feed[i++] = &e->rte; + } + ASSERT_DIE(i == count); } static rte * -export_filter(struct channel *c, rte *rt0, rte **rt_free, int silent) +export_filter(struct channel *c, rte *rt, int silent) { struct proto *p = c->proto; const struct filter *filter = c->out_filter; - struct proto_stats *stats = &c->stats; - rte *rt; - int v; + struct channel_export_stats *stats = &c->export_stats; - rt = rt0; - *rt_free = NULL; + /* Do nothing if we have already rejected the route */ + if (silent && bmap_test(&c->export_reject_map, rt->id)) + goto reject_noset; - v = p->preexport ? p->preexport(p, rt) : 0; + int v = p->preexport ? p->preexport(c, rt) : 0; if (v < 0) { if (silent) - goto reject; + goto reject_noset; - stats->exp_updates_rejected++; + stats->updates_rejected++; if (v == RIC_REJECT) - rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); - goto reject; + channel_rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); + goto reject_noset; + } if (v > 0) { if (!silent) - rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); + channel_rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); goto accept; } v = filter && ((filter == FILTER_REJECT) || - (f_run(filter, &rt, + (f_run(filter, rt, (silent ? FF_SILENT : 0)) > F_ACCEPT)); if (v) { if (silent) goto reject; - stats->exp_updates_filtered++; - rte_trace_out(D_FILTERS, c, rt, "filtered out"); + stats->updates_filtered++; + channel_rte_trace_out(D_FILTERS, c, rt, "filtered out"); goto reject; } accept: - if (rt != rt0) - *rt_free = rt; + /* We have accepted the route */ + bmap_clear(&c->export_reject_map, rt->id); return rt; reject: + /* We have rejected the route by filter */ + bmap_set(&c->export_reject_map, rt->id); + +reject_noset: /* Discard temporary rte */ - if (rt != rt0) - rte_free(rt); return NULL; } static void -do_rt_notify(struct channel *c, net *net, rte *new, rte *old, int refeed) +do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old) { struct proto *p = c->proto; - struct proto_stats *stats = &c->stats; + struct channel_export_stats *stats = &c->export_stats; - if (refeed && new) + if (c->refeeding && new) c->refeed_count++; - /* Apply export limit */ - struct channel_limit *l = &c->out_limit; - if (l->action && !old && new) - { - if (stats->exp_routes >= l->limit) - channel_notify_limit(c, l, PLD_OUT, stats->exp_routes); - - if (l->state == PLS_BLOCKED) + if (!old && new) + if (CHANNEL_LIMIT_PUSH(c, OUT)) { - stats->exp_updates_rejected++; - rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); + stats->updates_rejected++; + channel_rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); return; } - } + + if (!new && old) + CHANNEL_LIMIT_POP(c, OUT); /* Apply export table */ - struct rte *old_exported = NULL; + struct rte_storage *old_exported = NULL; if (c->out_table) { - if (!rte_update_out(c, net->n.addr, new, old, &old_exported, refeed)) + if (!rte_update_out(c, net, new, old, &old_exported)) + { + channel_rte_trace_out(D_ROUTES, c, new, "idempotent"); return; + } } - else if (c->out_filter == FILTER_ACCEPT) - old_exported = old; if (new) - stats->exp_updates_accepted++; + stats->updates_accepted++; else - stats->exp_withdraws_accepted++; + stats->withdraws_accepted++; if (old) - { bmap_clear(&c->export_map, old->id); - stats->exp_routes--; - } if (new) - { bmap_set(&c->export_map, new->id); - stats->exp_routes++; - } if (p->debug & D_ROUTES) { if (new && old) - rte_trace_out(D_ROUTES, c, new, "replaced"); + channel_rte_trace_out(D_ROUTES, c, new, "replaced"); else if (new) - rte_trace_out(D_ROUTES, c, new, "added"); + channel_rte_trace_out(D_ROUTES, c, new, "added"); else if (old) - rte_trace_out(D_ROUTES, c, old, "removed"); + channel_rte_trace_out(D_ROUTES, c, old, "removed"); } - p->rt_notify(p, c, net, new, old); + p->rt_notify(p, c, net, new, old_exported ? &old_exported->rte : old); if (c->out_table && old_exported) - rte_free_quick(old_exported); + rte_free(old_exported); } static void -rt_notify_basic(struct channel *c, net *net, rte *new, rte *old, int refeed) +rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old) { - // struct proto *p = c->proto; - rte *new_free = NULL; - - if (new) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; - if (new) - new = export_filter(c, new, &new_free, 0); + new = export_filter(c, new, 0); if (old && !bmap_test(&c->export_map, old->id)) old = NULL; @@ -858,133 +867,134 @@ rt_notify_basic(struct channel *c, net *net, rte *new, rte *old, int refeed) if (!new && !old) return; - do_rt_notify(c, net, new, old, refeed); - - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + do_rt_notify(c, net, new, old); } -static void -rt_notify_accepted(struct channel *c, net *net, rte *new_changed, rte *old_changed, int refeed) +void +rt_notify_accepted(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe, + struct rte **feed, uint count) { - // struct proto *p = c->proto; - rte *new_best = NULL; - rte *old_best = NULL; - rte *new_free = NULL; - int new_first = 0; - - /* - * We assume that there are no changes in net route order except (added) - * new_changed and (removed) old_changed. Therefore, the function is not - * compatible with deterministic_med (where nontrivial reordering can happen - * as a result of a route change) and with recomputation of recursive routes - * due to next hop update (where many routes can be changed in one step). - * - * Note that we need this assumption just for optimizations, we could just - * run full new_best recomputation otherwise. - * - * There are three cases: - * feed or old_best is old_changed -> we need to recompute new_best - * old_best is before new_changed -> new_best is old_best, ignore - * old_best is after new_changed -> try new_changed, otherwise old_best - */ + struct channel *c = SKIP_BACK(struct channel, out_req, req); - if (net->routes) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + rte nb0, *new_best = NULL; + const rte *old_best = NULL; - /* Find old_best - either old_changed, or route for net->routes */ - if (old_changed && bmap_test(&c->export_map, old_changed->id)) - old_best = old_changed; - else + for (uint i = 0; i < count; i++) { - for (rte *r = net->routes; rte_is_valid(r); r = r->next) + if (!rte_is_valid(feed[i])) + continue; + + /* Has been already rejected, won't bother with it */ + if (!c->refeeding && bmap_test(&c->export_reject_map, feed[i]->id)) + continue; + + /* Previously exported */ + if (!old_best && bmap_test(&c->export_map, feed[i]->id)) { - if (bmap_test(&c->export_map, r->id)) + /* is still best */ + if (!new_best) { - old_best = r; - break; + DBG("rt_notify_accepted: idempotent\n"); + goto done; } - /* Note if new_changed found before old_best */ - if (r == new_changed) - new_first = 1; + /* is superseded */ + old_best = feed[i]; + break; + } + + /* Have no new best route yet */ + if (!new_best) + { + /* Try this route not seen before */ + nb0 = *feed[i]; + new_best = export_filter(c, &nb0, 0); + DBG("rt_notify_accepted: checking route id %u: %s\n", feed[i]->id, new_best ? "ok" : "no"); } } - /* Find new_best */ - if ((new_changed == old_changed) || (old_best == old_changed)) - { - /* Feed or old_best changed -> find first accepted by filters */ - for (rte *r = net->routes; rte_is_valid(r); r = r->next) - if (new_best = export_filter(c, r, &new_free, 0)) + /* Check obsolete routes for previously exported */ + if (!old_best) + if (rpe && rpe->old && bmap_test(&c->export_map, rpe->old->rte.id)) + old_best = &rpe->old->rte; + +/* for (; rpe; rpe = atomic_load_explicit(&rpe->next, memory_order_relaxed)) + { + if (rpe->old && bmap_test(&hook->accept_map, rpe->old->id)) + { + old_best = &rpe->old.rte; break; - } - else - { - /* Other cases -> either new_changed, or old_best (and nothing changed) */ - if (new_first && (new_changed = export_filter(c, new_changed, &new_free, 0))) - new_best = new_changed; - else - return; - } + } + if (rpe == rpe_last) + break; + } + */ + + /* Nothing to export */ if (!new_best && !old_best) - return; + { + DBG("rt_notify_accepted: nothing to export\n"); + goto done; + } - do_rt_notify(c, net, new_best, old_best, refeed); + do_rt_notify(c, n, new_best, old_best); - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); +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); } rte * -rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int silent) +rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool, int silent) { + _Thread_local static rte rloc; + // struct proto *p = c->proto; struct nexthop_adata *nhs = NULL; - rte *best0, *best, *rt0, *rt, *tmp; - - best0 = net->routes; - *rt_free = NULL; + rte *best0 = feed[0]; + rte *best = NULL; if (!rte_is_valid(best0)) return NULL; - best = export_filter(c, best0, rt_free, silent); + /* Already rejected, no need to re-run the filter */ + if (!c->refeeding && bmap_test(&c->export_reject_map, best0->id)) + return NULL; + + rloc = *best0; + best = export_filter(c, &rloc, silent); + + if (!best) + /* Best route doesn't pass the filter */ + return NULL; - if (!best || !rte_is_reachable(best)) + if (!rte_is_reachable(best)) + /* Unreachable routes can't be merged */ return best; - for (rt0 = best0->next; rt0; rt0 = rt0->next) + for (uint i = 1; i < count; i++) { - if (!rte_mergable(best0, rt0)) + if (!rte_mergable(best0, feed[i])) continue; - rt = export_filter(c, rt0, &tmp, 1); + rte tmp0 = *feed[i]; + rte *tmp = export_filter(c, &tmp0, 1); - if (!rt) + if (!tmp || !rte_is_reachable(tmp)) continue; - if (rte_is_reachable(rt)) - { - eattr *nhea = ea_find(rt->attrs, &ea_gen_nexthop); - ASSERT_DIE(nhea); - - if (nhs) - nhs = nexthop_merge(nhs, (struct nexthop_adata *) nhea->u.ptr, c->merge_limit, pool); - else - nhs = (struct nexthop_adata *) nhea->u.ptr; - } + eattr *nhea = ea_find(tmp->attrs, &ea_gen_nexthop); + ASSERT_DIE(nhea); - if (tmp) - rte_free(tmp); + if (nhs) + nhs = nexthop_merge(nhs, (struct nexthop_adata *) nhea->u.ptr, c->merge_limit, pool); + else + nhs = (struct nexthop_adata *) nhea->u.ptr; } - if (nhs) { eattr *nhea = ea_find(best->attrs, &ea_gen_nexthop); @@ -992,66 +1002,114 @@ rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int nhs = nexthop_merge(nhs, (struct nexthop_adata *) nhea->u.ptr, c->merge_limit, pool); - best = rte_cow_rta(best); ea_set_attr(&best->attrs, EA_LITERAL_DIRECT_ADATA(&ea_gen_nexthop, 0, &nhs->ad)); } - if (best != best0) - *rt_free = best; - return best; } - -static void -rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed, - rte *new_best, rte *old_best, int refeed) +void +rt_notify_merged(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe, + struct rte **feed, uint count) { - // struct proto *p = c->proto; - rte *new_free = NULL; - - /* We assume that all rte arguments are either NULL or rte_is_valid() */ + struct channel *c = SKIP_BACK(struct channel, out_req, req); - /* This check should be done by the caller */ - if (!new_best && !old_best) - return; + // struct proto *p = c->proto; +#if 0 /* TODO: Find whether this check is possible when processing multiple changes at once. */ /* Check whether the change is relevant to the merged route */ if ((new_best == old_best) && (new_changed != old_changed) && !rte_mergable(new_best, new_changed) && !rte_mergable(old_best, old_changed)) return; +#endif - if (new_best) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + rte *old_best = NULL; + /* Find old best route */ + for (uint i = 0; i < count; i++) + if (bmap_test(&c->export_map, feed[i]->id)) + { + old_best = feed[i]; + break; + } + + /* Check obsolete routes for previously exported */ + if (!old_best) + if (rpe && rpe->old && bmap_test(&c->export_map, rpe->old->rte.id)) + old_best = &rpe->old->rte; + +/* for (; rpe; rpe = atomic_load_explicit(&rpe->next, memory_order_relaxed)) + { + if (rpe->old && bmap_test(&hook->accept_map, rpe->old->id)) + { + old_best = &rpe->old.rte; + break; + } + + if (rpe == rpe_last) + break; + } + */ /* Prepare new merged route */ - if (new_best) - new_best = rt_export_merged(c, net, &new_free, rte_update_pool, 0); + rte *new_merged = count ? rt_export_merged(c, feed, count, rte_update_pool, 0) : NULL; - /* Check old merged route */ - if (old_best && !bmap_test(&c->export_map, old_best->id)) - old_best = NULL; + if (new_merged || old_best) + do_rt_notify(c, n, new_merged, old_best); - if (!new_best && !old_best) - return; + /* 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); +} - do_rt_notify(c, net, new_best, old_best, refeed); +void +rt_notify_optimal(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; - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + 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++) + { + 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 - * @type: type of route announcement (RA_UNDEF or RA_ANY) * @net: network in question * @new: the new or changed route * @old: the previous route replaced by the new one @@ -1067,13 +1125,6 @@ rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed * and @new_best and @old_best describes best routes. Other routes are not * affected, but in sorted table the order of other routes might change. * - * Second, There is a bulk change of multiple routes in @net, with shared best - * route selection. In such case separate route changes are described using - * @type of %RA_ANY, with @new and @old specifying the changed route, while - * @new_best and @old_best are NULL. After that, another notification is done - * where @new_best and @old_best are filled (may be the same), but @new and @old - * are NULL. - * * The function announces the change to all associated channels. For each * channel, an appropriate preprocessing is done according to channel &ra_mode. * For example, %RA_OPTIMAL channels receive just changes of best routes. @@ -1088,19 +1139,19 @@ rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed * done outside of scope of rte_announce(). */ static void -rte_announce(rtable *tab, uint type, net *net, rte *new, rte *old, - rte *new_best, rte *old_best) +rte_announce(rtable *tab, net *net, struct rte_storage *new, struct rte_storage *old, + struct rte_storage *new_best, struct rte_storage *old_best) { - if (!rte_is_valid(new)) + if (!rte_is_valid(RTE_OR_NULL(new))) new = NULL; - if (!rte_is_valid(old)) + if (!rte_is_valid(RTE_OR_NULL(old))) old = NULL; - if (!rte_is_valid(new_best)) + if (!rte_is_valid(RTE_OR_NULL(new_best))) new_best = NULL; - if (!rte_is_valid(old_best)) + if (!rte_is_valid(RTE_OR_NULL(old_best))) old_best = NULL; if (!new && !old && !new_best && !old_best) @@ -1109,9 +1160,9 @@ rte_announce(rtable *tab, uint type, net *net, rte *new, rte *old, if (new_best != old_best) { if (new_best) - new_best->sender->stats.pref_routes++; + new_best->rte.sender->stats.pref++; if (old_best) - old_best->sender->stats.pref_routes--; + old_best->rte.sender->stats.pref--; if (tab->hostcache) rt_notify_hostcache(tab, net); @@ -1122,76 +1173,82 @@ rte_announce(rtable *tab, uint type, net *net, rte *new, rte *old, 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; + feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(net, feed, count); + } - if (type && (type != c->ra_mode)) + struct rt_export_hook *eh; + WALK_LIST(eh, tab->exports) + { + if (eh->export_state == TES_STOP) continue; - switch (c->ra_mode) - { - case RA_OPTIMAL: - if (new_best != old_best) - rt_notify_basic(c, net, new_best, old_best, 0); - break; - - case RA_ANY: - if (new != old) - rt_notify_basic(c, net, new, old, 0); - break; - - case RA_ACCEPTED: - rt_notify_accepted(c, net, new, old, 0); - break; + if (new) + eh->stats.updates_received++; + else + eh->stats.withdraws_received++; - case RA_MERGED: - rt_notify_merged(c, net, new, old, new_best, old_best, 0); - break; - } + 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; - net *n = e->net; + const net_addr *n = e->net; - if (!net_validate(n->n.addr)) + if (!net_validate(n)) { log(L_WARN "Ignoring bogus prefix %N received via %s", - n->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } /* FIXME: better handling different nettypes */ - c = !net_is_flow(n->n.addr) ? - net_classify(n->n.addr): (IADDR_HOST | SCOPE_UNIVERSE); + c = !net_is_flow(n) ? + net_classify(n): (IADDR_HOST | SCOPE_UNIVERSE); if ((c < 0) || !(c & IADDR_HOST) || ((c & IADDR_SCOPE_MASK) <= SCOPE_LINK)) { log(L_WARN "Ignoring bogus route %N received via %s", - n->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } - eattr *nhea = ea_find(e->attrs, &ea_gen_nexthop); - int dest = nhea_dest(nhea); - - if (net_type_match(n->n.addr, NB_DEST) == !dest) + if (net_type_match(n, NB_DEST)) { - log(L_WARN "Ignoring route %N with invalid dest %d received via %s", - n->n.addr, dest, e->sender->proto->name); - return 0; - } + eattr *nhea = ea_find(e->attrs, &ea_gen_nexthop); + int dest = nhea_dest(nhea); + + if (dest == RTD_NONE) + { + log(L_WARN "Ignoring route %N with no destination received via %s", + n, ch->proto->name); + return 0; + } - if ((dest == RTD_UNICAST) && - !nexthop_is_sorted((struct nexthop_adata *) nhea->u.ptr)) + if ((dest == RTD_UNICAST) && + !nexthop_is_sorted((struct nexthop_adata *) nhea->u.ptr)) + { + log(L_WARN "Ignoring unsorted multipath route %N received via %s", + n, ch->proto->name); + return 0; + } + } + else if (ea_find(e->attrs, &ea_gen_nexthop)) { - log(L_WARN "Ignoring unsorted multipath route %N received via %s", - n->n.addr, e->sender->proto->name); + log(L_WARN "Ignoring route %N having a nexthop attribute received via %s", + n, ch->proto->name); return 0; } @@ -1212,41 +1269,36 @@ 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 proto_stats *stats = &c->stats; - static struct tbf rl_pipe = TBF_DEFAULT_LOG_LIMITS; - rte *before_old = NULL; - rte *old_best = net->routes; + struct rt_import_stats *stats = &c->stats; + struct rte_storage *old_best_stored = net->routes, *old_stored = NULL; + rte *old_best = old_best_stored ? &old_best_stored->rte : NULL; rte *old = NULL; - rte **k; - k = &net->routes; /* Find and remove original route from the same protocol */ - while (old = *k) + /* Find and remove original route from the same protocol */ + struct rte_storage **before_old = rte_find(net, src); + + if (*before_old) { - if (old->src == src) + old = &(old_stored = (*before_old))->rte; + + /* If there is the same route in the routing table but from + * a different sender, then there are two paths from the + * source protocol to this routing table through transparent + * pipes, which is not allowed. + * We log that and ignore the route. */ + if (old->sender != c) { - /* If there is the same route in the routing table but from - * a different sender, then there are two paths from the - * source protocol to this routing table through transparent - * pipes, which is not allowed. - * - * We log that and ignore the route. If it is withdraw, we - * ignore it completely (there might be 'spurious withdraws', - * see FIXME in do_rte_announce()) - */ - if (old->sender->proto != p) - { - if (new) - { - log_rl(&rl_pipe, L_ERR "Pipe collision detected when sending %N to table %s", - net->n.addr, table->name); - rte_free_quick(new); - } - return; - } + if (!old->generation && !new->generation) + bug("Two protocols claim to author a route with the same rte_src in table %s: %N %s/%u:%u", + c->table->name, net->n.addr, old->src->proto->name, old->src->private_id, old->src->global_id); + + log_rl(&table->rl_pipe, L_ERR "Route source collision in table %s: %N %s/%u:%u", + c->table->name, net->n.addr, old->src->proto->name, old->src->private_id, old->src->global_id); + } if (new && rte_same(old, new)) { @@ -1256,123 +1308,56 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) if (!rte_is_filtered(new)) { - stats->imp_updates_ignored++; - rte_trace_in(D_ROUTES, c, new, "ignored"); + stats->updates_ignored++; + rt_rte_trace_in(D_ROUTES, req, new, "ignored"); } + } - rte_free_quick(new); - return; - } - *k = old->next; - table->rt_count--; - break; - } - k = &old->next; - before_old = old; + *before_old = (*before_old)->next; + table->rt_count--; } - /* Save the last accessed position */ - rte **pos = k; - - if (!old) - before_old = NULL; - if (!old && !new) { - stats->imp_withdraws_ignored++; + stats->withdraws_ignored++; 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->imp_routes + stats->filt_routes; - - if (all_routes >= l->limit) - channel_notify_limit(c, l, PLD_RX, all_routes); - - if (l->state == PLS_BLOCKED) - { - /* In receive limit the situation is simple, old is NULL so - we just free new and exit like nothing happened */ - - stats->imp_updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - rte_free_quick(new); - return; - } - } - - l = &c->in_limit; - if (l->action && !old_ok && new_ok) - { - if (stats->imp_routes >= l->limit) - channel_notify_limit(c, l, PLD_IN, stats->imp_routes); - - if (l->state == PLS_BLOCKED) - { - /* In import limit the situation is more complicated. We - shouldn't just drop the route, we should handle it like - it was filtered. We also have to continue the route - processing if old or new is non-NULL, but we should exit - if both are NULL as this case is probably assumed to be - already handled. */ - - stats->imp_updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - - if (c->in_keep_filtered) - new->flags |= REF_FILTERED; - else - { rte_free_quick(new); new = NULL; } - - /* Note that old && !new could be possible when - c->in_keep_filtered changed in the recent past. */ - - if (!old && !new) - return; - - new_ok = 0; - goto skip_stats1; - } - } - if (new_ok) - stats->imp_updates_accepted++; + stats->updates_accepted++; else if (old_ok) - stats->imp_withdraws_accepted++; + stats->withdraws_accepted++; else - stats->imp_withdraws_ignored++; + stats->withdraws_ignored++; if (old_ok || new_ok) table->last_rt_change = current_time(); - skip_stats1: - - if (new) - rte_is_filtered(new) ? stats->filt_routes++ : stats->imp_routes++; - if (old) - rte_is_filtered(old) ? stats->filt_routes-- : stats->imp_routes--; + struct rte_storage *new_stored = new ? rte_store(new, net, table) : NULL; if (table->config->sorted) { /* If routes are sorted, just insert new route to appropriate position */ - if (new) + if (new_stored) { - if (before_old && !rte_better(new, before_old)) - k = &before_old->next; + struct rte_storage **k; + if ((before_old != &net->routes) && !rte_better(new, &SKIP_BACK(struct rte_storage, next, before_old)->rte)) + k = before_old; else k = &net->routes; for (; *k; k=&(*k)->next) - if (rte_better(new, *k)) + if (rte_better(new, &(*k)->rte)) break; - new->next = *k; - *k = new; + new_stored->next = *k; + *k = new_stored; table->rt_count++; } @@ -1382,16 +1367,17 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) /* If routes are not sorted, find the best route and move it on the first position. There are several optimized cases. */ - if (src->proto->rte_recalculate && src->proto->rte_recalculate(table, net, new, old, old_best)) + if (src->proto->rte_recalculate && + src->proto->rte_recalculate(table, net, new_stored ? &new_stored->rte : NULL, old, old_best)) goto do_recalculate; - if (new && rte_better(new, old_best)) + if (new_stored && rte_better(&new_stored->rte, old_best)) { /* The first case - the new route is cleary optimal, we link it at the first position */ - new->next = net->routes; - net->routes = new; + new_stored->next = net->routes; + net->routes = new_stored; table->rt_count++; } @@ -1405,10 +1391,10 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) do_recalculate: /* Add the new route to the list */ - if (new) + if (new_stored) { - new->next = *pos; - *pos = new; + new_stored->next = *before_old; + *before_old = new_stored; table->rt_count++; } @@ -1416,81 +1402,82 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) /* Find a new optimal route (if there is any) */ if (net->routes) { - rte **bp = &net->routes; - for (k=&(*bp)->next; *k; k=&(*k)->next) - if (rte_better(*k, *bp)) + struct rte_storage **bp = &net->routes; + for (struct rte_storage **k=&(*bp)->next; *k; k=&(*k)->next) + if (rte_better(&(*k)->rte, &(*bp)->rte)) bp = k; /* And relink it */ - rte *best = *bp; + struct rte_storage *best = *bp; *bp = best->next; best->next = net->routes; net->routes = best; } } - else if (new) + else if (new_stored) { /* The third case - the new route is not better than the old best route (therefore old_best != NULL) and the old best route was not removed (therefore old_best == net->routes). We just link the new route to the old/last position. */ - new->next = *pos; - *pos = new; + new_stored->next = *before_old; + *before_old = new_stored; table->rt_count++; } /* The fourth (empty) case - suboptimal route was removed, nothing to do */ } - if (new) + if (new_stored) { - new->lastmod = current_time(); + new_stored->rte.lastmod = current_time(); if (!old) { - new->id = hmap_first_zero(&table->id_map); - hmap_set(&table->id_map, new->id); + new_stored->rte.id = hmap_first_zero(&table->id_map); + hmap_set(&table->id_map, new_stored->rte.id); } else - new->id = old->id; + new_stored->rte.id = old->id; } /* Log the route change */ - if ((c->debug & D_ROUTES) || (p->debug & D_ROUTES)) + if (new_ok) + rt_rte_trace_in(D_ROUTES, req, &new_stored->rte, new_stored == net->routes ? "added [best]" : "added"); + else if (old_ok) { - if (new_ok) - rte_trace(c, new, '>', new == net->routes ? "added [best]" : "added"); - else if (old_ok) - { - if (old != old_best) - rte_trace(c, old, '>', "removed"); - else if (rte_is_ok(net->routes)) - rte_trace(c, old, '>', "removed [replaced]"); - else - rte_trace(c, old, '>', "removed [sole]"); - } + if (old != old_best) + rt_rte_trace_in(D_ROUTES, req, old, "removed"); + else if (net->routes && rte_is_ok(&net->routes->rte)) + rt_rte_trace_in(D_ROUTES, req, old, "removed [replaced]"); + else + rt_rte_trace_in(D_ROUTES, req, old, "removed [sole]"); } /* Propagate the route change */ - rte_announce(table, RA_UNDEF, net, new, old, net->routes, old_best); + rte_announce(table, net, new_stored, old_stored, + net->routes, old_best_stored); if (!net->routes && (table->gc_counter++ >= table->config->gc_max_ops) && (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); + p->rte_insert(net, &new_stored->rte); +#endif if (old) { - if (!new) + if (!new_stored) hmap_clear(&table->id_map, old->id); - rte_free_quick(old); + rte_free(old_stored); } } @@ -1509,184 +1496,158 @@ rte_update_unlock(void) lp_flush(rte_update_pool); } -/** - * rte_update - enter a new update to a routing table - * @table: table to be updated - * @c: channel doing the update - * @net: network node - * @p: protocol submitting the update - * @src: protocol originating the update - * @new: a &rte representing the new route or %NULL for route removal. - * - * This function is called by the routing protocols whenever they discover - * a new route or wish to update/remove an existing route. The right announcement - * sequence is to build route attributes first (either un-cached with @aflags set - * to zero or a cached one using rta_lookup(); in this case please note that - * you need to increase the use count of the attributes yourself by calling - * rta_clone()), call rte_get_temp() to obtain a temporary &rte, fill in all - * the appropriate data and finally submit the new &rte by calling rte_update(). - * - * @src specifies the protocol that originally created the route and the meaning - * of protocol-dependent data of @new. If @new is not %NULL, @src have to be the - * same value as @new->attrs->proto. @p specifies the protocol that called - * rte_update(). In most cases it is the same protocol as @src. rte_update() - * stores @p in @new->sender; - * - * When rte_update() gets any route, it automatically validates it (checks, - * whether the network and next hop address are valid IP addresses and also - * whether a normal routing protocol doesn't try to smuggle a host or link - * scope route to the table), converts all protocol dependent attributes stored - * in the &rte to temporary extended attributes, consults import filters of the - * protocol to see if the route should be accepted and/or its attributes modified, - * stores the temporary attributes back to the &rte. - * - * Now, having a "public" version of the route, we - * automatically find any old route defined by the protocol @src - * for network @n, replace it by the new one (or removing it if @new is %NULL), - * recalculate the optimal route for this destination and finally broadcast - * the change (if any) to all routing protocols by calling rte_announce(). - * - * All memory used for attribute lists and other temporary allocations is taken - * from a special linear pool @rte_update_pool and freed when rte_update() - * finishes. - */ +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_update2(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { - // struct proto *p = c->proto; - struct proto_stats *stats = &c->stats; - const struct filter *filter = c->in_filter; - net *nn; + if (!c->in_req.hook) + return; ASSERT(c->channel_state == CS_UP); + if (c->in_table && !rte_update_in(c, n, new, src)) + return; + + return rte_update_direct(c, n, new, src); +} + +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) { - /* Create a temporary table node */ - nn = alloca(sizeof(net) + n->length); - memset(nn, 0, sizeof(net) + n->length); - net_copy(nn->n.addr, n); + new->net = n; - new->net = nn; - new->sender = c; + int fr; - stats->imp_updates_received++; - if (filter == FILTER_REJECT) + stats->updates_received++; + if ((filter == FILTER_REJECT) || + ((fr = f_run(filter, new, 0)) > F_ACCEPT)) { - stats->imp_updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); - - if (! c->in_keep_filtered) - goto drop; - - /* new is a private copy, i could modify it */ - new->flags |= REF_FILTERED; - } - else if (filter) - { - int fr = f_run(filter, &new, 0); - if (fr > F_ACCEPT) - { - stats->imp_updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); - - if (! c->in_keep_filtered) - goto drop; + stats->updates_filtered++; + channel_rte_trace_in(D_FILTERS, c, new, "filtered out"); + if (c->in_keep_filtered) new->flags |= REF_FILTERED; - } + else + new = NULL; } - rte *new_resolved = rt_next_hop_update_rte(c->table, new); - if (new_resolved) - { - rte_free(new); - new = new_resolved; - } + if (new) + if (net_is_flow(n)) + rt_flowspec_resolve_rte(new, c); + else + rt_next_hop_resolve_rte(new); - /* After all checks, updates and filters have been done, - * validate the route */ - if (!rte_validate(new)) + if (new && !rte_validate(c, new)) { - rte_trace_in(D_FILTERS, c, new, "invalid"); - stats->imp_updates_invalid++; - goto drop; + channel_rte_trace_in(D_FILTERS, c, new, "invalid"); + stats->updates_invalid++; + new = NULL; } - if (!rta_is_cached(new->attrs)) /* Need to copy attributes */ - new->attrs = rta_lookup(new->attrs); - new->flags |= REF_COW; - - /* Use the actual struct network, not the dummy one */ - nn = net_get(c->table, n); - new->net = nn; } else - { - stats->imp_withdraws_received++; + stats->withdraws_received++; - if (!(nn = net_find(c->table, n)) || !src) - { - stats->imp_withdraws_ignored++; - rte_update_unlock(); - return; - } - } - - recalc: - /* And recalculate the best route */ - rte_recalculate(c, nn, new, src); + rte_import(&c->in_req, n, new, src); rte_update_unlock(); - return; +} - drop: - rte_free(new); - new = NULL; - if (nn = net_find(c->table, n)) - goto recalc; +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; - rte_update_unlock(); + net *nn; + if (new) + { + /* Use the actual struct network, not the dummy one */ + nn = net_get(hook->table, n); + new->net = nn->n.addr; + new->sender = hook; + } + else if (!(nn = net_find(hook->table, n))) + { + req->hook->stats.withdraws_ignored++; + return; + } + + /* And recalculate the best route */ + rte_recalculate(hook, nn, new, src); } /* Independent call to rte_announce(), used from next hop recalculation, outside of rte_update(). new must be non-NULL */ static inline void -rte_announce_i(rtable *tab, uint type, net *net, rte *new, rte *old, - rte *new_best, rte *old_best) +rte_announce_i(rtable *tab, net *net, struct rte_storage *new, struct rte_storage *old, + struct rte_storage *new_best, struct rte_storage *old_best) { rte_update_lock(); - rte_announce(tab, type, net, new, old, new_best, old_best); + rte_announce(tab, net, new, old, new_best, old_best); rte_update_unlock(); } static inline void -rte_discard(rte *old) /* Non-filtered route deletion, used during garbage collection */ +rte_discard(net *net, rte *old) /* Non-filtered route deletion, used during garbage collection */ { rte_update_lock(); - rte_recalculate(old->sender, old->net, NULL, old->src); + rte_recalculate(old->sender, net, NULL, old->src); rte_update_unlock(); } /* Modify existing route by protocol hook, used for long-lived graceful restart */ static inline void -rte_modify(rte *old) +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) - { - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - new->flags = (old->flags & ~REF_MODIFY) | REF_COW; - } + new->flags = old->flags & ~REF_MODIFY; - rte_recalculate(old->sender, old->net, new, old->src); + rte_recalculate(old->sender, net, new, old->src); } rte_update_unlock(); @@ -1694,30 +1655,148 @@ rte_modify(rte *old) /* Check rtable for best route to given net whether it would be exported do p */ int -rt_examine(rtable *t, net_addr *a, struct proto *p, const struct filter *filter) +rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filter) { net *n = net_find(t, a); - rte *rt = n ? n->routes : NULL; - if (!rte_is_valid(rt)) + if (!n || !rte_is_valid(RTE_OR_NULL(n->routes))) return 0; + rte rt = n->routes->rte; + rte_update_lock(); /* Rest is stripped down export_filter() */ - int v = p->preexport ? p->preexport(p, rt) : 0; + int v = c->proto->preexport ? c->proto->preexport(c, &rt) : 0; if (v == RIC_PROCESS) v = (f_run(filter, &rt, FF_SILENT) <= F_ACCEPT); - /* Discard temporary rte */ - if (rt != n->routes) - rte_free(rt); - rte_update_unlock(); 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 @@ -1734,14 +1813,13 @@ rt_examine(rtable *t, net_addr *a, struct proto *p, const struct filter *filter) * 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) { - rte *e; - for (e = n->routes; e; e = e->next) - if (e->sender == c) - e->flags |= REF_STALE; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (e->rte.sender == req->hook) + e->rte.flags |= REF_STALE; } FIB_WALK_END; } @@ -1755,17 +1833,16 @@ 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) { - rte *e; - for (e = n->routes; e; e = e->next) - if ((e->sender == c) && (e->flags & REF_STALE)) + for (struct rte_storage *e = n->routes; e; e = e->next) + if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE)) { - e->flags |= REF_DISCARD; + e->rte.flags |= REF_DISCARD; prune = 1; } } @@ -1776,17 +1853,16 @@ 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) { - rte *e; - for (e = n->routes; e; e = e->next) - if ((e->sender == c) && (e->flags & REF_STALE) && !(e->flags & REF_FILTERED)) + for (struct rte_storage *e = n->routes; e; e = e->next) + if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE) && !(e->rte.flags & REF_FILTERED)) { - e->flags |= REF_MODIFY; + e->rte.flags |= REF_MODIFY; prune = 1; } } @@ -1803,12 +1879,11 @@ rt_modify_stale(rtable *t, struct channel *c) * This functions dumps contents of a &rte to debug output. */ void -rte_dump(rte *e) +rte_dump(struct rte_storage *e) { - net *n = e->net; - debug("%-1N ", n->n.addr); - debug("PF=%02x ", e->pflags); - ea_dump(e->attrs); + debug("%-1N ", e->rte.net); + debug("PF=%02x ", e->rte.pflags); + ea_dump(e->rte.attrs); debug("\n"); } @@ -1821,14 +1896,13 @@ rte_dump(rte *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 FIB_WALK(&t->fib, net, n) { - rte *e; - for(e=n->routes; e; e=e->next) + for(struct rte_storage *e=n->routes; e; e=e->next) rte_dump(e); } FIB_WALK_END; @@ -1848,6 +1922,54 @@ 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 @@ -1965,6 +2087,7 @@ 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); } @@ -2110,6 +2233,8 @@ rt_setup(pool *pp, struct rtable_config *cf) rtable *t = ralloc(p, &rt_class); t->rp = p; + t->rte_slab = sl_new(p, sizeof(struct rte_storage)); + t->name = cf->name; t->config = cf; t->addr_type = cf->addr_type; @@ -2124,18 +2249,22 @@ 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(); + t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS; + if (rt_is_flow(t)) { t->flowspec_trie = f_new_trie(lp_new_default(p), 0); @@ -2158,8 +2287,8 @@ rt_init(void) rta_init(); rt_table_pool = rp_new(&root_pool, "Routing tables"); rte_update_pool = lp_new_default(rt_table_pool); - rte_slab = sl_new(rt_table_pool, sizeof(rte)); init_list(&routing_tables); + init_list(&deleted_routing_tables); } @@ -2181,9 +2310,9 @@ static void rt_prune_table(rtable *tab) { struct fib_iterator *fit = &tab->prune_fit; - int limit = 512; + int limit = 2000; - struct channel *c; + struct rt_import_hook *ih; node *n, *x; DBG("Pruning route table %s\n", tab->name); @@ -2197,9 +2326,9 @@ 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; @@ -2215,8 +2344,6 @@ rt_prune_table(rtable *tab) again: FIB_ITERATE_START(&tab->fib, fit, net, n) { - rte *e; - rescan: if (limit <= 0) { @@ -2225,19 +2352,19 @@ again: return; } - for (e=n->routes; e; e=e->next) + for (struct rte_storage *e=n->routes; e; e=e->next) { - if (e->sender->flush_active || (e->flags & REF_DISCARD)) + if ((e->rte.sender->import_state == TIS_FLUSHING) || (e->rte.flags & REF_DISCARD)) { - rte_discard(e); + rte_discard(n, &e->rte); limit--; goto rescan; } - if (e->flags & REF_MODIFY) + if (e->rte.flags & REF_MODIFY) { - rte_modify(e); + rte_modify(n, &e->rte); limit--; goto rescan; @@ -2300,21 +2427,18 @@ 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); + } } /** @@ -2536,24 +2660,30 @@ rta_next_hop_outdated(ea_list *a) ? head : NULL; } -static inline rte * -rt_next_hop_update_rte(rtable *tab UNUSED, rte *old) +static inline struct rte_storage * +rt_next_hop_update_rte(rtable *tab, net *n, rte *old) { struct hostentry_adata *head = rta_next_hop_outdated(old->attrs); if (!head) return NULL; - ea_list *ea = old->attrs; - rta_apply_hostentry(&ea, head); + rte e0 = *old; + rta_apply_hostentry(&e0.attrs, head); - rte *e = sl_alloc(rte_slab); - memcpy(e, old, sizeof(rte)); - e->attrs = rta_lookup(ea); - rt_lock_source(e->src); - - return e; + return rte_store(&e0, n, tab); } +static inline void +rt_next_hop_resolve_rte(rte *r) +{ + eattr *heea = ea_find(r->attrs, &ea_gen_hostentry); + if (!heea) + return; + + struct hostentry_adata *head = (struct hostentry_adata *) heea->u.ptr; + + rta_apply_hostentry(&r->attrs, head); +} #ifdef CONFIG_BGP @@ -2593,7 +2723,7 @@ rta_get_first_asn(ea_list *a) return (e && as_path_get_first_regular(e->u.ptr, &asn)) ? asn : 0; } -int +static inline enum flowspec_valid rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list *a, int interior) { ASSERT(rt_is_ip(tab_ip)); @@ -2602,11 +2732,11 @@ rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list * /* RFC 8955 6. a) Flowspec has defined dst prefix */ if (!net_flow_has_dst_prefix(n)) - return 0; + return FLOWSPEC_INVALID; /* RFC 9117 4.1. Accept AS_PATH is empty (fr */ if (interior && rta_as_path_is_empty(a)) - return 1; + return FLOWSPEC_VALID; /* RFC 8955 6. b) Flowspec and its best-match route have the same originator */ @@ -2620,7 +2750,7 @@ rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list * /* Find best-match BGP unicast route for flowspec dst prefix */ net *nb = net_route(tab_ip, &dst); - rte *rb = nb ? nb->routes : NULL; + const rte *rb = nb ? &nb->routes->rte : NULL; /* Register prefix to trie for tracking further changes */ int max_pxlen = (n->type == NET_FLOW4) ? IP4_MAX_PREFIX_LENGTH : IP6_MAX_PREFIX_LENGTH; @@ -2628,7 +2758,7 @@ rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list * /* No best-match BGP route -> no flowspec */ if (!rb || (rt_get_source_attr(rb) != RTS_BGP)) - return 0; + return FLOWSPEC_INVALID; /* Find ORIGINATOR_ID values */ u32 orig_a = ea_get_int(a, "bgp_originator_id", 0); @@ -2639,17 +2769,17 @@ rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list * ea_get_ip(a, &ea_gen_from, IPA_NONE), ea_get_ip(rb->attrs, &ea_gen_from, IPA_NONE) ))) - return 0; + return FLOWSPEC_INVALID; /* Find ASN of the best-match route, for use in next checks */ u32 asn_b = rta_get_first_asn(rb->attrs); if (!asn_b) - return 0; + return FLOWSPEC_INVALID; /* RFC 9117 4.2. For EBGP, flowspec and its best-match route are from the same AS */ if (!interior && (rta_get_first_asn(a) != asn_b)) - return 0; + return FLOWSPEC_INVALID; /* RFC 8955 6. c) More-specific routes are from the same AS as the best-match route */ TRIE_WALK(tab_ip->trie, subnet, &dst) @@ -2658,100 +2788,134 @@ rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list * if (!nc) continue; - rte *rc = nc->routes; + const rte *rc = &nc->routes->rte; if (rt_get_source_attr(rc) != RTS_BGP) - return 0; + return FLOWSPEC_INVALID; if (rta_get_first_asn(rc->attrs) != asn_b) - return 0; + return FLOWSPEC_INVALID; } TRIE_WALK_END; - return 1; + return FLOWSPEC_VALID; } #endif /* CONFIG_BGP */ -static rte * -rt_flowspec_update_rte(rtable *tab, rte *r) +static struct rte_storage * +rt_flowspec_update_rte(rtable *tab, net *n, rte *r) { #ifdef CONFIG_BGP if (rt_get_source_attr(r) != RTS_BGP) return NULL; - struct bgp_channel *bc = (struct bgp_channel *) r->sender; + struct bgp_channel *bc = (struct bgp_channel *) SKIP_BACK(struct channel, in_req, r->sender->req); if (!bc->base_table) return NULL; - const net_addr *n = r->net->n.addr; - struct bgp_proto *p = (void *) r->src->proto; - int valid = rt_flowspec_check(bc->base_table, tab, n, r->attrs, p->is_interior); - int old = rt_get_flowspec_valid(r); + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, p, bc->c.proto); + + enum flowspec_valid old = rt_get_flowspec_valid(r), + valid = rt_flowspec_check(bc->base_table, tab, n->n.addr, r->attrs, p->is_interior); + if (old == valid) return NULL; - ea_list *a = r->attrs; - ea_set_attr_u32(&a, &ea_gen_flowspec_valid, 0, valid); - - rte *new = sl_alloc(rte_slab); - memcpy(new, r, sizeof(rte)); - new->attrs = ea_lookup(a); + rte new = *r; + ea_set_attr_u32(&new.attrs, &ea_gen_flowspec_valid, 0, valid); - return new; + return rte_store(&new, n, tab); #else return NULL; #endif } +static inline void +rt_flowspec_resolve_rte(rte *r, struct channel *c) +{ +#ifdef CONFIG_BGP + enum flowspec_valid valid, old = rt_get_flowspec_valid(r); + struct bgp_channel *bc = (struct bgp_channel *) c; + + if ( (rt_get_source_attr(r) == RTS_BGP) + && (c->channel == &channel_bgp) + && (bc->base_table)) + { + struct bgp_proto *p = SKIP_BACK(struct bgp_proto, p, bc->c.proto); + valid = rt_flowspec_check( + bc->base_table, + c->in_req.hook->table, + r->net, r->attrs, p->is_interior); + } + else + valid = FLOWSPEC_UNKNOWN; + + if (valid == old) + return; + + if (valid == FLOWSPEC_UNKNOWN) + ea_unset_attr(&r->attrs, 0, &ea_gen_flowspec_valid); + else + ea_set_attr_u32(&r->attrs, &ea_gen_flowspec_valid, 0, valid); +#endif +} static inline int rt_next_hop_update_net(rtable *tab, net *n) { - rte **k, *e, *new, *old_best, **new_best; + struct rte_storage *new; int count = 0; - int free_old_best = 0; + int is_flow = net_is_flow(n->n.addr); - old_best = n->routes; + struct rte_storage *old_best = n->routes; if (!old_best) return 0; - for (k = &n->routes; e = *k; k = &e->next) - { - if (!net_is_flow(n->n.addr)) - new = rt_next_hop_update_rte(tab, e); - else - new = rt_flowspec_update_rte(tab, e); + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) + if (is_flow || rta_next_hop_outdated(e->rte.attrs)) + count++; - if (new) + if (!count) + return 0; + + struct rte_multiupdate { + struct rte_storage *old, *new; + } *updates = alloca(sizeof(struct rte_multiupdate) * count); + + int pos = 0; + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) + if (is_flow || rta_next_hop_outdated(e->rte.attrs)) { - *k = new; + struct rte_storage *new = is_flow + ? rt_flowspec_update_rte(tab, n, &e->rte) + : rt_next_hop_update_rte(tab, n, &e->rte); - rte_trace_in(D_ROUTES, new->sender, new, "updated"); - rte_announce_i(tab, RA_ANY, n, new, e, NULL, NULL); + if (!new) + continue; /* Call a pre-comparison hook */ /* Not really an efficient way to compute this */ - if (e->src->proto->rte_recalculate) - e->src->proto->rte_recalculate(tab, n, new, e, NULL); + if (e->rte.src->proto->rte_recalculate) + e->rte.src->proto->rte_recalculate(tab, n, &new->rte, &e->rte, &old_best->rte); - if (e != old_best) - rte_free_quick(e); - else /* Freeing of the old best rte is postponed */ - free_old_best = 1; + updates[pos++] = (struct rte_multiupdate) { + .old = e, + .new = new, + }; - e = new; - count++; + /* Replace the route in the list */ + new->next = e->next; + *k = e = new; } - } - if (!count) - return 0; + ASSERT_DIE(pos <= count); + count = pos; /* Find the new best route */ - new_best = NULL; - for (k = &n->routes; e = *k; k = &e->next) + struct rte_storage **new_best = NULL; + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) { - if (!new_best || rte_better(e, *new_best)) + if (!new_best || rte_better(&e->rte, &(*new_best)->rte)) new_best = k; } @@ -2764,15 +2928,17 @@ rt_next_hop_update_net(rtable *tab, net *n) n->routes = new; } - /* Announce the new best route */ - if (new != old_best) - rte_trace_in(D_ROUTES, new->sender, new, "updated [best]"); - - /* Propagate changes */ - rte_announce_i(tab, RA_UNDEF, n, NULL, NULL, n->routes, old_best); + /* Announce the changes */ + for (int i=0; i<count; i++) + { + _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]" } }; + 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); + } - if (free_old_best) - rte_free_quick(old_best); + for (int i=0; i<count; i++) + rte_free(updates[i].old); return count; } @@ -2952,19 +3118,6 @@ 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) - rt_notify_accepted(c, n, NULL, NULL, c->refeeding); - else if (c->ra_mode == RA_MERGED) - rt_notify_merged(c, n, NULL, NULL, e, e, c->refeeding); - else /* RA_BASIC */ - rt_notify_basic(c, n, e, e, c->refeeding); - rte_update_unlock(); -} - /** * rt_feed_channel - advertise all routes to a channel * @c: channel to be fed @@ -2974,79 +3127,55 @@ 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) { - rte *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); - 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); - 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); } @@ -3058,16 +3187,10 @@ int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { struct rtable *tab = c->in_table; - rte *old, **pos; net *net; if (new) - { net = net_get(tab, n); - - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - } else { net = net_find(tab, n); @@ -3077,9 +3200,10 @@ rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *sr } /* Find the old rte */ - for (pos = &net->routes; old = *pos; pos = &old->next) - if (old->src == src) + struct rte_storage **pos = rte_find(net, src); + if (*pos) { + rte *old = &(*pos)->rte; if (new && rte_same(old, new)) { /* Refresh the old rte, continue with update to main rtable */ @@ -3092,60 +3216,52 @@ 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 (old == c->reload_next_rte) - c->reload_next_rte = old->next; + if (*pos == c->reload_next_rte) + c->reload_next_rte = (*pos)->next; /* Remove the old rte */ - *pos = old->next; - rte_free_quick(old); + struct rte_storage *del = *pos; + *pos = (*pos)->next; + rte_free(del); tab->rt_count--; + } + else if (new) + { + if (CHANNEL_LIMIT_PUSH(c, RX)) + { + /* Required by rte_trace_in() */ + new->net = n; - break; + channel_rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); + goto drop_update; + } } + else + goto drop_withdraw; if (!new) { - if (!old) - goto drop_withdraw; - if (!net->routes) fib_delete(&tab->fib, net); return 1; } - struct channel_limit *l = &c->rx_limit; - if (l->action && !old) - { - if (tab->rt_count >= l->limit) - channel_notify_limit(c, l, PLD_RX, tab->rt_count); - - if (l->state == PLS_BLOCKED) - { - /* Required by rte_trace_in() */ - new->net = net; - - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - goto drop_update; - } - } - /* Insert the new rte */ - rte *e = rte_do_cow(new); - e->flags |= REF_COW; - e->net = net; - e->sender = c; - e->lastmod = current_time(); + struct rte_storage *e = rte_store(new, net, tab); + e->rte.lastmod = current_time(); e->next = *pos; *pos = e; tab->rt_count++; return 1; drop_update: - c->stats.imp_updates_received++; - c->stats.imp_updates_ignored++; - rte_free(new); + c->import_stats.updates_received++; + c->in_req.hook->stats.updates_ignored++; if (!net->routes) fib_delete(&tab->fib, net); @@ -3153,8 +3269,8 @@ drop_update: return 0; drop_withdraw: - c->stats.imp_withdraws_received++; - c->stats.imp_withdraws_ignored++; + c->import_stats.withdraws_received++; + c->in_req.hook->stats.withdraws_ignored++; return 0; } @@ -3174,7 +3290,7 @@ rt_reload_channel(struct channel *c) } do { - for (rte *e = c->reload_next_rte; e; e = e->next) + for (struct rte_storage *e = c->reload_next_rte; e; e = e->next) { if (max_feed-- <= 0) { @@ -3183,7 +3299,8 @@ rt_reload_channel(struct channel *c) return 0; } - rte_update2(c, e->net->n.addr, rte_do_cow(e), e->src); + rte r = e->rte; + rte_update_direct(c, r.net, &r, r.src); } c->reload_next_rte = NULL; @@ -3226,14 +3343,14 @@ rt_prune_sync(rtable *t, int all) again: FIB_ITERATE_START(&t->fib, &fit, net, n) { - rte *e, **ee = &n->routes; + struct rte_storage *e, **ee = &n->routes; while (e = *ee) { - if (all || (e->flags & (REF_STALE | REF_DISCARD))) + if (all || (e->rte.flags & (REF_STALE | REF_DISCARD))) { *ee = e->next; - rte_free_quick(e); + rte_free(e); t->rt_count--; } else @@ -3256,20 +3373,16 @@ again: */ int -rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, rte **old_exported, int refeed) +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; - rte *old, **pos; net *net; if (new) { net = net_get(tab, n); src = new->src; - - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); } else { @@ -3277,39 +3390,28 @@ rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, rte ** src = old0->src; if (!net) - goto drop_withdraw; + goto drop; } /* Find the old rte */ - for (pos = &net->routes; old = *pos; pos = &old->next) - if ((c->ra_mode != RA_ANY) || (old->src == src)) - { - if (new && rte_same(old, new)) - { - /* REF_STALE / REF_DISCARD not used in export table */ - /* - if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY)) - { - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); - return 1; - } - */ - - goto drop_update; - } + struct rte_storage **pos = (c->ra_mode == RA_ANY) ? rte_find(net, src) : &net->routes; + struct rte_storage *old = NULL; - /* Remove the old rte */ - *pos = old->next; - *old_exported = old; - tab->rt_count--; + if (old = *pos) + { + if (new && rte_same(&(*pos)->rte, new)) + goto drop; - break; - } + /* Remove the old rte */ + *pos = old->next; + *old_exported = old; + tab->rt_count--; + } if (!new) { if (!old) - goto drop_withdraw; + goto drop; if (!net->routes) fib_delete(&tab->fib, net); @@ -3318,23 +3420,43 @@ rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, rte ** } /* Insert the new rte */ - rte *e = rte_do_cow(new); - e->flags |= REF_COW; - e->net = net; - e->sender = c; - e->lastmod = current_time(); + struct rte_storage *e = rte_store(new, net, tab); + e->rte.lastmod = current_time(); e->next = *pos; *pos = e; tab->rt_count++; return 1; -drop_update: - return refeed; - -drop_withdraw: +drop: return 0; } +void +rt_refeed_channel(struct channel *c) +{ + if (!c->out_table) + { + channel_request_feeding(c); + return; + } + + ASSERT_DIE(c->ra_mode != RA_ANY); + + c->proto->feed_begin(c, 0); + + FIB_WALK(&c->out_table->fib, net, n) + { + if (!n->routes) + continue; + + rte e = n->routes->rte; + c->proto->rt_notify(c->proto, c, n->n.addr, &e, NULL); + } + FIB_WALK_END; + + c->proto->feed_end(c); +} + /* * Hostcache @@ -3500,7 +3622,7 @@ if_local_addr(ip_addr a, struct iface *i) } u32 -rt_get_igp_metric(rte *rt) +rt_get_igp_metric(const rte *rt) { eattr *ea = ea_find(rt->attrs, "igp_metric"); @@ -3533,8 +3655,8 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) net *n = net_route(tab, &he_addr); if (n) { - rte *e = n->routes; - ea_list *a = e->attrs; + struct rte_storage *e = n->routes; + ea_list *a = e->rte.attrs; pxlen = n->n.addr->pxlen; if (ea_find(a, &ea_gen_hostentry)) @@ -3566,7 +3688,7 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) he->src = rta_clone(a); he->nexthop_linkable = !direct; - he->igp_metric = rt_get_igp_metric(e); + he->igp_metric = rt_get_igp_metric(&e->rte); } done: @@ -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. */ @@ -20,6 +21,7 @@ struct ea_list; struct protocol; struct proto; +struct channel; struct rte_src; struct symbol; struct timer; @@ -58,16 +60,16 @@ typedef struct rtable { resource r; node n; /* Node in list of all tables */ pool *rp; /* Resource pool to allocate everything from, including itself */ + struct slab *rte_slab; /* Slab to allocate route objects */ 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 pipe_busy; /* Pipe loop detection */ 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; @@ -85,12 +87,14 @@ 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 */ struct f_trie *trie_old; /* Old prefix trie waiting to be freed */ u32 trie_lock_count; /* Prefix trie locked by walks */ u32 trie_old_lock_count; /* Old prefix trie locked by walks */ + struct tbf rl_pipe; /* Rate limiting token buffer for pipe collisions */ list subscribers; /* Subscribers for notifications */ struct timer *settle_timer; /* Settle time for notifications */ @@ -118,7 +122,7 @@ struct rt_flowspec_link { #define NHU_DIRTY 3 typedef struct network { - struct rte *routes; /* Available routes for this network */ + struct rte_storage *routes; /* Available routes for this network */ struct fib_node n; /* FIB flags reserved for kernel syncer */ } net; @@ -148,6 +152,129 @@ struct hostentry { u32 igp_metric; /* Chosen route IGP metric */ }; +struct rte_storage { + struct rte_storage *next; /* Next in chain */ + struct rte rte; /* Route data */ +}; + +#define RTE_COPY(r, l) ((r) ? (((*(l)) = (r)->rte), (l)) : NULL) +#define RTE_OR_NULL(r) ((r) ? &((r)->rte) : NULL) + +/* 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 */ #define RA_OPTIMAL 1 /* Announcement of optimal route change */ @@ -161,6 +288,41 @@ struct hostentry { #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 + * @net: network address + * @rte: a &rte representing the new route + * @src: old route source identifier + * + * This function imports a new route to the appropriate table (via the channel). + * Table keys are @net (obligatory) and @rte->attrs->src. + * Both the @net and @rte pointers can be local. + * + * The route attributes (@rte->attrs) are obligatory. They can be also allocated + * locally. Anyway, if you use an already-cached attribute object, you shall + * call rta_clone() on that object yourself. (This semantics may change in future.) + * + * If the route attributes are local, you may set @rte->attrs->src to NULL, then + * the protocol's default route source will be supplied. + * + * When rte_update() gets a route, it automatically validates it. This includes + * checking for validity of the given network and next hop addresses and also + * checking for host-scope or link-scope routes. Then the import filters are + * processed and if accepted, the route is passed to route table recalculation. + * + * The accepted routes are then inserted into the table, replacing the old route + * for the same @net identified by @src. Then the route is announced + * to all the channels connected to the table using the standard export mechanism. + * Setting @rte to NULL makes this a withdraw, otherwise @rte->src must be the same + * as @src. + * + * All memory used for temporary allocations is taken from a special linpool + * @rte_update_pool and freed when rte_update() finishes. + */ +void rte_update(struct channel *c, const net_addr *net, struct rte *rte, struct rte_src *src); + extern list routing_tables; struct config; @@ -180,34 +342,29 @@ static inline void rt_shutdown(rtable *r) { rfree(r->rp); } static inline net *net_find(rtable *tab, const net_addr *addr) { return (net *) fib_find(&tab->fib, addr); } static inline net *net_find_valid(rtable *tab, const net_addr *addr) -{ net *n = net_find(tab, addr); return (n && rte_is_valid(n->routes)) ? n : NULL; } +{ net *n = net_find(tab, addr); return (n && n->routes && rte_is_valid(&n->routes->rte)) ? n : NULL; } static inline net *net_get(rtable *tab, const net_addr *addr) { return (net *) fib_get(&tab->fib, addr); } net *net_get(rtable *tab, const net_addr *addr); net *net_route(rtable *tab, const net_addr *n); -rte *rte_find(net *net, struct rte_src *src); -rte *rte_get_temp(ea_list *, struct rte_src *src); -void rte_update2(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); -/* rte_update() moved to protocol.h to avoid dependency conflicts */ -int rt_examine(rtable *t, net_addr *a, struct proto *p, const struct filter *filter); -rte *rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int silent); -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); +int rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filter); +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(rte *); -void rte_free(rte *); -rte *rte_do_cow(rte *); -static inline rte * rte_cow(rte *r) { return (r->flags & REF_COW) ? rte_do_cow(r) : r; } -rte *rte_cow_rta(rte *r); +void rte_dump(struct rte_storage *); +void rte_free(struct rte_storage *); +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); -int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); +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, rte **old_exported, int refeed); +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) @@ -247,6 +404,7 @@ 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 */ @@ -290,8 +448,6 @@ struct hostentry_adata { void ea_set_hostentry(ea_list **to, struct rtable *dep, struct rtable *tab, ip_addr gw, ip_addr ll, u32 lnum, u32 labels[lnum]); -int rt_flowspec_check(rtable *tab_ip, rtable *tab_flow, const net_addr *n, ea_list *a, int interior); - /* * Default protocol preferences |