summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMaria Matejka <mq@ucw.cz>2022-07-12 14:46:06 +0200
committerMaria Matejka <mq@ucw.cz>2022-07-12 14:46:06 +0200
commit1df20989c191353d9c6da4d8679f8cb1ead49f7f (patch)
treea40097fcf4687f1b9d33a3ba794e38898bfbab4f
parent44f26c49f966ca842ff9af55468de0b98c44b73e (diff)
Revert "Special table hooks rectified."
This reverts commit 44f26c49f966ca842ff9af55468de0b98c44b73e.
-rw-r--r--lib/lists.c9
-rw-r--r--lib/lists.h1
-rw-r--r--nest/config.Y4
-rw-r--r--nest/proto.c426
-rw-r--r--nest/protocol.h23
-rw-r--r--nest/route.h35
-rw-r--r--nest/rt-attr.c3
-rw-r--r--nest/rt-show.c7
-rw-r--r--nest/rt-table.c593
-rw-r--r--proto/bgp/attrs.c50
-rw-r--r--proto/bgp/bgp.c94
-rw-r--r--proto/bgp/bgp.h3
-rw-r--r--proto/bgp/packets.c2
-rw-r--r--proto/rpki/packets.c8
-rw-r--r--sysdep/unix/krt.c256
-rw-r--r--sysdep/unix/krt.h4
16 files changed, 841 insertions, 677 deletions
diff --git a/lib/lists.c b/lib/lists.c
index fe2b692b..200576cf 100644
--- a/lib/lists.c
+++ b/lib/lists.c
@@ -109,15 +109,6 @@ add_head(list *l, node *n)
l->head = n;
}
-LIST_INLINE void
-self_link(node *n)
-{
- ASSUME(n->prev == NULL);
- ASSUME(n->next == NULL);
-
- n->prev = n->next = n;
-}
-
/**
* insert_node - insert a node to a list
* @n: a new list node
diff --git a/lib/lists.h b/lib/lists.h
index 64b4a981..479f4ed1 100644
--- a/lib/lists.h
+++ b/lib/lists.h
@@ -78,7 +78,6 @@ typedef union list { /* In fact two overlayed nodes */
#define LIST_INLINE
void add_tail(list *, node *);
void add_head(list *, node *);
-void self_link(node *);
void rem_node(node *);
void add_tail_list(list *, list *);
void init_list(list *);
diff --git a/nest/config.Y b/nest/config.Y
index a56b25be..29d6b0db 100644
--- a/nest/config.Y
+++ b/nest/config.Y
@@ -644,12 +644,12 @@ r_args:
}
| r_args IMPORT TABLE channel_arg {
if (!$4->in_table) cf_error("No import table in channel %s.%s", $4->proto->name, $4->name);
- rt_show_add_table($$, $4->in_table->tab);
+ rt_show_add_table($$, $4->in_table);
$$->tables_defined_by = RSD_TDB_DIRECT;
}
| r_args EXPORT TABLE channel_arg {
if (!$4->out_table) cf_error("No export table in channel %s.%s", $4->proto->name, $4->name);
- rt_show_add_table($$, $4->out_table->tab);
+ rt_show_add_table($$, $4->out_table);
$$->tables_defined_by = RSD_TDB_DIRECT;
}
| r_args FILTER filter {
diff --git a/nest/proto.c b/nest/proto.c
index cf448fd9..09582d2e 100644
--- a/nest/proto.c
+++ b/nest/proto.c
@@ -47,7 +47,7 @@ static char *c_states[] = { "DOWN", "START", "UP", "STOP", "RESTART" };
extern struct protocol proto_unix_iface;
-static void channel_aux_request_refeed(struct channel_aux_table *cat);
+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);
@@ -88,9 +88,7 @@ channel_export_log_state_change(struct rt_export_request *req, u8 state)
switch (state)
{
case TES_FEEDING:
- if (c->out_table)
- rt_refresh_begin(&c->out_table->push);
- else if (c->proto->feed_begin)
+ if (c->proto->feed_begin)
c->proto->feed_begin(c, !c->refeeding);
break;
case TES_READY:
@@ -181,7 +179,6 @@ proto_find_channel_by_name(struct proto *p, const char *n)
}
rte * channel_preimport(struct rt_import_request *req, rte *new, rte *old);
-rte * channel_in_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);
@@ -298,10 +295,14 @@ static void
channel_roa_in_changed(struct rt_subscription *s)
{
struct channel *c = s->data;
+ int active = c->reload_event && ev_active(c->reload_event);
- CD(c, "Reload triggered by RPKI change");
+ CD(c, "Reload triggered by RPKI change%s", active ? " - already active" : "");
- channel_request_reload(c);
+ if (!active)
+ channel_request_reload(c);
+ else
+ c->reload_pending = 1;
}
static void
@@ -443,6 +444,7 @@ channel_start_import(struct channel *c)
.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);
@@ -461,8 +463,7 @@ channel_start_export(struct channel *c)
{
if (c->out_req.hook)
{
- c->restart_export = 1;
- log(L_WARN "%s.%s: Fast channel export restart", c->proto->name, c->name);
+ log(L_WARN "%s.%s: Attempted to start channel's already started export", c->proto->name, c->name);
return;
}
@@ -513,7 +514,7 @@ channel_check_stopped(struct channel *c)
switch (c->channel_state)
{
case CS_STOP:
- if (c->out_req.hook || c->in_req.hook || c->out_table || c->in_table)
+ if (c->out_req.hook || c->in_req.hook)
return;
channel_set_state(c, CS_DOWN);
@@ -540,6 +541,9 @@ channel_import_stopped(struct rt_import_request *req)
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;
@@ -562,16 +566,14 @@ channel_export_stopped(struct rt_export_request *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;
- if (c->restart_export)
- {
- c->restart_export = 0;
- channel_start_export(c);
- }
- else
- channel_check_stopped(c);
+ channel_check_stopped(c);
}
static void
@@ -593,296 +595,72 @@ channel_feed_end(struct channel *c)
return;
}
- if (c->out_table)
- rt_refresh_end(&c->out_table->push);
- else if (c->proto->feed_end)
+ if (c->proto->feed_end)
c->proto->feed_end(c);
if (c->refeed_pending)
rt_stop_export(req, channel_export_stopped);
-}
-
-#define CHANNEL_AUX_TABLE_DUMP_REQ(inout, imex, pgimex, pushget) static void \
- channel_##inout##_##pushget##_dump_req(struct rt_##pgimex##_request *req) { \
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, pushget, req); \
- debug(" Channel %s.%s " #imex " table " #pushget " request %p\n", cat->c->proto->name, cat->c->name, req); }
-
-CHANNEL_AUX_TABLE_DUMP_REQ(in, import, import, push)
-CHANNEL_AUX_TABLE_DUMP_REQ(in, import, export, get)
-CHANNEL_AUX_TABLE_DUMP_REQ(out, export, import, push)
-CHANNEL_AUX_TABLE_DUMP_REQ(out, export, export, get)
-
-#undef CHANNEL_AUX_TABLE_DUMP_REQ
-
-static uint channel_aux_imex(struct channel_aux_table *cat)
-{
- if (cat->c->in_table == cat)
- return 0;
- else if (cat->c->out_table == cat)
- return 1;
- else
- bug("Channel aux table must be in_table or out_table");
-}
-
-static void
-channel_aux_stopped(void *data)
-{
- struct channel_aux_table *cat = data;
- struct channel *c = cat->c;
-
- if (channel_aux_imex(cat))
- c->out_table = NULL;
else
- c->in_table = NULL;
-
- mb_free(cat);
- return channel_check_stopped(c);
-}
-
-static void
-channel_aux_import_stopped(struct rt_import_request *req)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, push, req);
- ASSERT_DIE(cat->stop);
-}
-
-static void
-channel_aux_export_stopped(struct rt_export_request *req)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, push, req);
- req->hook = NULL;
-
- if (cat->refeed_pending && !cat->stop)
- {
- cat->refeed_pending = 0;
- rt_request_export(cat->tab, req);
- }
- else
- ASSERT_DIE(cat->stop);
+ c->refeeding = 0;
}
-static void
-channel_aux_stop(struct channel_aux_table *cat)
+/* Called by protocol for reload from in_table */
+void
+channel_schedule_reload(struct channel *c)
{
- cat->stop = 1;
-
- rt_stop_import(&cat->push, channel_aux_import_stopped);
- rt_stop_export(&cat->get, channel_aux_export_stopped);
+ ASSERT(c->in_req.hook);
- rt_lock_table(cat->tab);
- cat->tab->deleted = channel_aux_stopped;
- cat->tab->del_data = cat;
- rt_unlock_table(cat->tab);
+ rt_reload_channel_abort(c);
+ ev_schedule_work(c->reload_event);
}
static void
-channel_push_log_state_change(struct rt_import_request *req, u8 state)
+channel_reload_loop(void *ptr)
{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, push, req);
- const char *imex = channel_aux_imex(cat) ? "export" : "import";
- CD(cat->c, "Channel %s table import state changed to %s", imex, rt_import_state_name(state));
-}
+ struct channel *c = ptr;
-static void
-channel_get_log_state_change(struct rt_export_request *req, u8 state)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
- const char *imex = channel_aux_imex(cat) ? "export" : "import";
- CD(cat->c, "Channel %s table export state changed to %s", imex, rt_export_state_name(state));
+ /* Start reload */
+ if (!c->reload_active)
+ c->reload_pending = 0;
- switch (state)
+ if (!rt_reload_channel(c))
{
- case TES_FEEDING:
- if (imex && cat->c->proto->feed_begin)
- cat->c->proto->feed_begin(cat->c, !cat->c->refeeding);
- else if (!imex)
- rt_refresh_begin(&cat->c->in_req);
- break;
-
- case TES_READY:
- if (imex && cat->c->proto->feed_end)
- cat->c->proto->feed_end(cat->c);
- else if (!imex)
- rt_refresh_end(&cat->c->in_req);
-
- if (cat->refeed_pending)
- rt_stop_export(&cat->get, channel_aux_export_stopped);
-
- break;
- }
-}
-
-void rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src);
-
-static void
-channel_in_export_one_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
-
- if (!rpe->new && !rpe->old)
- return;
-
- rte n0;
- struct rte_src *src = rpe->new ? rpe->new->rte.src : rpe->old->rte.src;
- rte_update_direct(cat->c, net, RTES_CLONE(rpe->new, &n0), src);
-}
-
-static void
-channel_in_export_one_best(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
-
- if (!rpe->new && !rpe->old)
+ ev_schedule_work(c->reload_event);
return;
-
- rte n0;
- struct rte_src *src = rpe->old_best ? rpe->old_best->rte.src : rpe->new_best->rte.src;
- rte_update_direct(cat->c, net, RTES_CLONE(rpe->new_best, &n0), src);
-}
-
-static void
-channel_in_export_bulk_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
- for (uint i=0; i<count; i++)
- {
- rte n0 = *feed[i];
- rte_update_direct(cat->c, net, &n0, n0.src);
}
-}
-
-static void
-channel_in_export_bulk_best(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
- if (!count)
- return;
-
- rte n0 = *feed[0];
- rte_update_direct(cat->c, net, &n0, n0.src);
-}
-
-void do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte *old);
-
-static void
-channel_out_export_one_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
- rte n0;
- do_rt_notify_direct(cat->c, net, RTES_CLONE(rpe->new, &n0), RTES_OR_NULL(rpe->old));
-}
-
-static void
-channel_out_export_one_best(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
- rte n0;
- do_rt_notify_direct(cat->c, net, RTES_CLONE(rpe->new_best, &n0), RTES_OR_NULL(rpe->old_best));
-}
-static void
-channel_out_export_bulk(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req);
- if (cat->c->ra_mode != RA_ANY)
- ASSERT_DIE(count <= 1);
-
- for (uint i=0; i<count; i++)
- {
- rte n0 = *feed[i];
- do_rt_notify_direct(cat->c, net, &n0, NULL);
- }
+ /* Restart reload */
+ if (c->reload_pending)
+ channel_request_reload(c);
}
/* Called by protocol to activate in_table */
void
-channel_setup_in_table(struct channel *c, int best)
+channel_setup_in_table(struct channel *c)
{
- int nlen = sizeof("import") + strlen(c->name) + strlen(c->proto->name) + 3;
-
- struct {
- struct channel_aux_table cat;
- struct rtable_config tab_cf;
- char name[0];
- } *cat = mb_allocz(c->proto->pool, sizeof(*cat) + nlen);
-
- bsprintf(cat->name, "%s.%s.import", c->proto->name, c->name);
+ struct rtable_config *cf = mb_allocz(c->proto->pool, sizeof(struct rtable_config));
- cat->tab_cf.name = cat->name;
- cat->tab_cf.addr_type = c->net_type;
-
- c->in_table = &cat->cat;
- c->in_table->push = (struct rt_import_request) {
- .name = cat->name,
- .trace_routes = c->debug | c->proto->debug,
- .dump_req = channel_in_push_dump_req,
- .log_state_change = channel_push_log_state_change,
- .preimport = channel_in_preimport,
- };
- c->in_table->get = (struct rt_export_request) {
- .name = cat->name,
- .trace_routes = c->debug | c->proto->debug,
- .dump_req = channel_in_get_dump_req,
- .log_state_change = channel_get_log_state_change,
- .export_one = best ? channel_in_export_one_best : channel_in_export_one_any,
- .export_bulk = best ? channel_in_export_bulk_best : channel_in_export_bulk_any,
- };
+ cf->name = "import";
+ cf->addr_type = c->net_type;
+ cf->internal = 1;
- c->in_table->c = c;
- c->in_table->tab = rt_setup(c->proto->pool, &cat->tab_cf);
- self_link(&c->in_table->tab->n);
+ c->in_table = rt_setup(c->proto->pool, cf);
- rt_request_import(c->in_table->tab, &c->in_table->push);
- rt_request_export(c->in_table->tab, &c->in_table->get);
+ c->reload_event = ev_new_init(c->proto->pool, channel_reload_loop, c);
}
/* Called by protocol to activate out_table */
void
channel_setup_out_table(struct channel *c)
{
- int nlen = sizeof("export") + strlen(c->name) + strlen(c->proto->name) + 3;
-
- struct {
- struct channel_aux_table cat;
- struct rtable_config tab_cf;
- char name[0];
- } *cat = mb_allocz(c->proto->pool, sizeof(*cat) + nlen);
-
- bsprintf(cat->name, "%s.%s.export", c->proto->name, c->name);
+ struct rtable_config *cf = mb_allocz(c->proto->pool, sizeof(struct rtable_config));
+ cf->name = "export";
+ cf->addr_type = c->net_type;
+ cf->internal = 1;
- cat->tab_cf.name = cat->name;
- cat->tab_cf.addr_type = c->net_type;
-
- c->out_table = &cat->cat;
- c->out_table->push = (struct rt_import_request) {
- .name = cat->name,
- .trace_routes = c->debug | c->proto->debug,
- .dump_req = channel_out_push_dump_req,
- .log_state_change = channel_push_log_state_change,
- };
- c->out_table->get = (struct rt_export_request) {
- .name = cat->name,
- .trace_routes = c->debug | c->proto->debug,
- .dump_req = channel_out_get_dump_req,
- .log_state_change = channel_get_log_state_change,
- .export_one = (c->ra_mode == RA_ANY) ? channel_out_export_one_any : channel_out_export_one_best,
- .export_bulk = channel_out_export_bulk,
- };
-
- c->out_table->c = c;
- c->out_table->tab = rt_setup(c->proto->pool, &cat->tab_cf);
- self_link(&c->out_table->tab->n);
-
- rt_request_import(c->out_table->tab, &c->out_table->push);
- rt_request_export(c->out_table->tab, &c->out_table->get);
+ c->out_table = rt_setup(c->proto->pool, cf);
}
-static void
-channel_aux_request_refeed(struct channel_aux_table *cat)
-{
- cat->refeed_pending = 1;
- rt_stop_export(&cat->get, channel_aux_export_stopped);
-}
static void
channel_do_start(struct channel *c)
@@ -908,12 +686,16 @@ channel_do_up(struct channel *c)
static void
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);
- c->refeeding = 0;
- }
channel_roa_unsubscribe_all(c);
@@ -924,13 +706,6 @@ channel_do_pause(struct channel *c)
static void
channel_do_stop(struct channel *c)
{
- /* Drop auxiliary tables */
- if (c->in_table)
- channel_aux_stop(c->in_table);
-
- if (c->out_table)
- channel_aux_stop(c->out_table);
-
/* Stop import */
if (c->in_req.hook)
rt_stop_import(&c->in_req, channel_import_stopped);
@@ -941,13 +716,16 @@ channel_do_stop(struct channel *c)
CALL(c->channel->shutdown, c);
- channel_roa_unsubscribe_all(c);
+ /* This have to be done in here, as channel pool is freed before channel_do_down() */
+ c->in_table = NULL;
+ c->reload_event = NULL;
+ c->out_table = NULL;
}
static void
channel_do_down(struct channel *c)
{
- ASSERT(!c->out_req.hook && !c->in_req.hook && !c->out_table && !c->in_table);
+ ASSERT(!c->reload_active);
c->proto->active_channels--;
@@ -955,11 +733,13 @@ channel_do_down(struct channel *c)
memset(&c->import_stats, 0, sizeof(struct channel_import_stats));
memset(&c->export_stats, 0, sizeof(struct channel_export_stats));
- CALL(c->channel->cleanup, c);
+ c->in_table = NULL;
+ c->reload_event = NULL;
+ c->out_table = NULL;
- /* This have to be done in here, as channel pool is freed before channel_do_down() */
- bmap_free(&c->export_map);
- bmap_free(&c->export_reject_map);
+ /* The in_table and out_table are going to be freed by freeing their resource pools. */
+
+ CALL(c->channel->cleanup, c);
/* Schedule protocol shutddown */
if (proto_is_done(c->proto))
@@ -989,7 +769,7 @@ channel_set_state(struct channel *c, uint state)
break;
case CS_UP:
- ASSERT(cs == CS_DOWN || cs == CS_START || cs == CS_PAUSE);
+ ASSERT(cs == CS_DOWN || cs == CS_START);
if (cs == CS_DOWN)
channel_do_start(c);
@@ -1039,27 +819,16 @@ channel_set_state(struct channel *c, uint state)
* completed, it will switch back to ES_READY. This function can be called
* even when feeding is already running, in that case it is restarted.
*/
-static void
-channel_request_table_feeding(struct channel *c)
-{
- ASSERT(c->out_req.hook);
-
- c->refeed_pending = 1;
- rt_stop_export(&c->out_req, channel_export_stopped);
-}
-
void
channel_request_feeding(struct channel *c)
{
ASSERT(c->out_req.hook);
- if (c->out_table)
- channel_aux_request_refeed(c->out_table);
- else
- channel_request_table_feeding(c);
+ c->refeed_pending = 1;
+ rt_stop_export(&c->out_req, channel_export_stopped);
}
-void
+static void
channel_request_reload(struct channel *c)
{
ASSERT(c->in_req.hook);
@@ -1067,29 +836,14 @@ channel_request_reload(struct channel *c)
CD(c, "Reload requested");
- if (c->in_table)
- channel_aux_request_refeed(c->in_table);
- else
- c->proto->reload_routes(c);
-}
+ c->proto->reload_routes(c);
-void
-channel_refresh_begin(struct channel *c)
-{
- CD(c, "Channel route refresh begin");
- if (c->in_table)
- rt_refresh_begin(&c->in_table->push);
- else
- rt_refresh_begin(&c->in_req);
-}
-
-void
-channel_refresh_end(struct channel *c)
-{
- if (c->in_table)
- rt_refresh_end(&c->in_table->push);
- else
- rt_refresh_end(&c->in_req);
+ /*
+ * Should this be done before reload_routes() hook?
+ * Perhaps, but routes are updated asynchronously.
+ */
+ channel_reset_limit(c, &c->rx_limit, PLD_RX);
+ channel_reset_limit(c, &c->in_limit, PLD_IN);
}
const struct channel_class channel_basic = {
@@ -1247,7 +1001,7 @@ channel_reconfigure(struct channel *c, struct channel_config *cf)
channel_request_reload(c);
if (export_changed)
- channel_request_table_feeding(c);
+ channel_request_feeding(c);
done:
CD(c, "Reconfigured");
@@ -1960,7 +1714,7 @@ protos_dump_all(void)
WALK_LIST(p, proto_list)
{
#define DPF(x) (p->x ? " " #x : "")
- debug(" protocol %s (%p) state %s with %d active channels flags: %s%s%s%s\n",
+ 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
@@ -1976,20 +1730,6 @@ protos_dump_all(void)
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 (c->in_table)
- {
- debug("\tInput aux table:\n");
- rt_dump_hooks(c->in_table->tab);
- rt_dump(c->in_table->tab);
- debug("\tEnd of input aux table.\n");
- }
- if (c->out_table)
- {
- debug("\tOutput aux table:\n");
- rt_dump_hooks(c->in_table->tab);
- rt_dump(c->in_table->tab);
- debug("\tEnd of output aux table.\n");
- }
}
if (p->proto->dump && (p->proto_state != PS_DOWN))
@@ -2411,11 +2151,11 @@ channel_show_stats(struct channel *c)
cli_msg(-1006, " Routes: %u imported, %u exported, %u preferred",
in_routes, out_routes, SRI(pref));
- cli_msg(-1006, " Route change stats: received rejected filtered ignored limited accepted");
- cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u %10u",
+ 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),
+ 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),
diff --git a/nest/protocol.h b/nest/protocol.h
index f9996b18..7447cbf0 100644
--- a/nest/protocol.h
+++ b/nest/protocol.h
@@ -207,6 +207,7 @@ 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 *);
void (*rte_insert)(struct network *, struct rte *);
void (*rte_remove)(struct network *, struct rte *);
u32 (*rte_igp_metric)(struct rte *);
@@ -543,29 +544,24 @@ struct 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 */
- u8 restart_export; /* Route export should restart as soon as it stops */
btime last_state_change; /* Time of last state transition */
- struct channel_aux_table *in_table; /* Internal table for received routes */
+ 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_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 */
u8 refeed_pending; /* Refeeding and another refeed is scheduled */
u8 rpki_reload; /* RPKI changes trigger channel reload */
- struct channel_aux_table *out_table; /* Internal table for exported routes */
+ struct rtable *out_table; /* Internal table for exported routes */
list roa_subscriptions; /* List of active ROA table subscriptions based on filters roa_check() */
};
-struct channel_aux_table {
- struct channel *c;
- struct rt_import_request push;
- struct rt_export_request get;
- rtable *tab;
- u8 stop;
- u8 refeed_pending;
-};
/*
* Channel states
@@ -631,7 +627,7 @@ struct channel *proto_add_channel(struct proto *p, struct channel_config *cf);
int proto_configure_channel(struct proto *p, struct channel **c, struct channel_config *cf);
void channel_set_state(struct channel *c, uint state);
-void channel_setup_in_table(struct channel *c, int best);
+void channel_setup_in_table(struct channel *c);
void channel_setup_out_table(struct channel *c);
void channel_schedule_reload(struct channel *c);
@@ -640,9 +636,6 @@ 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_STOP); }
void channel_request_feeding(struct channel *c);
-void channel_request_reload(struct channel *c);
-void channel_refresh_begin(struct channel *c);
-void channel_refresh_end(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);
diff --git a/nest/route.h b/nest/route.h
index cb66be2a..d0568133 100644
--- a/nest/route.h
+++ b/nest/route.h
@@ -150,6 +150,7 @@ struct rtable_config {
int gc_max_ops; /* Maximum number of operations before GC is run */
int gc_min_time; /* Minimum time between two consecutive GC runs */
byte sorted; /* Routes of network are sorted according to rte_better() */
+ byte internal; /* Internal table of a protocol */
btime min_settle_time; /* Minimum settle time for notifications */
btime max_settle_time; /* Maximum settle time for notifications */
};
@@ -171,8 +172,10 @@ typedef struct rtable {
struct hmap id_map;
struct hostcache *hostcache;
struct rtable_config *config; /* Configuration of this table */
- void (*deleted)(void *); /* Table should free itself. Call this when it is done. */
- void *del_data;
+ struct config *deleted; /* Table doesn't exist in current configuration,
+ * delete as soon as use_count becomes 0 and remove
+ * obstacle from this routing table.
+ */
struct event *rt_event; /* Routing table event */
btime last_rt_change; /* Last time when route changed */
btime base_settle_time; /* Start time of rtable settling interval */
@@ -181,6 +184,7 @@ typedef struct rtable {
byte prune_state; /* Table prune state, 1 -> scheduled, 2-> running */
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 tbf rl_pipe; /* Rate limiting token buffer for pipe collisions */
@@ -245,7 +249,6 @@ typedef struct rte {
u8 generation; /* If this route import is based on other previously exported route,
this value should be 1 + MAX(generation of the parent routes).
Otherwise the route is independent and this value is zero. */
- u8 stale_cycle; /* Auxiliary value for route refresh */
} rte;
struct rte_storage {
@@ -253,11 +256,13 @@ struct rte_storage {
struct rte rte; /* Route data */
};
-#define RTES_CLONE(r, l) ((r) ? (((*(l)) = (r)->rte), (l)) : NULL)
-#define RTES_OR_NULL(r) ((r) ? &((r)->rte) : NULL)
+#define RTE_COPY(r, l) ((r) ? (((*(l)) = (r)->rte), (l)) : NULL)
+#define RTE_OR_NULL(r) ((r) ? &((r)->rte) : NULL)
#define REF_FILTERED 2 /* Route is rejected by import filter */
-#define REF_USE_STALE 4 /* Do not reset route's stale_cycle to the actual value */
+#define REF_STALE 4 /* Route is stale in a refresh cycle */
+#define REF_DISCARD 8 /* Route is scheduled for discard */
+#define REF_MODIFY 16 /* Route is scheduled for modify */
/* Route is valid for propagation (may depend on other flags in the future), accepts NULL */
static inline int rte_is_valid(const rte *r) { return r && !(r->flags & REF_FILTERED); }
@@ -278,6 +283,7 @@ struct rt_import_request {
/* 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 {
@@ -297,10 +303,6 @@ struct rt_import_hook {
btime last_state_change; /* Time of last state transition */
u8 import_state; /* IS_* */
- u8 stale_set; /* Set this stale_cycle to imported routes */
- u8 stale_valid; /* Routes with this stale_cycle and bigger are considered valid */
- u8 stale_pruned; /* Last prune finished when this value was set at stale_valid */
- u8 stale_pruning; /* Last prune started when this value was set at stale_valid */
void (*stopped)(struct rt_import_request *); /* Stored callback when import is stopped */
};
@@ -453,9 +455,9 @@ void *net_route(rtable *tab, const net_addr *n);
int net_roa_check(rtable *tab, const net_addr *n, u32 asn);
int rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filter);
rte *rt_export_merged(struct channel *c, rte ** feed, uint count, linpool *pool, int silent);
-
-void rt_refresh_begin(struct rt_import_request *);
-void rt_refresh_end(struct rt_import_request *);
+void rt_refresh_begin(rtable *t, struct rt_import_request *);
+void rt_refresh_end(rtable *t, struct rt_import_request *);
+void rt_modify_stale(rtable *t, struct rt_import_request *);
void rt_schedule_prune(rtable *t);
void rte_dump(struct rte_storage *);
void rte_free(struct rte_storage *, rtable *);
@@ -464,9 +466,15 @@ void rt_dump(rtable *);
void rt_dump_all(void);
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_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);
+
/* Default limit for ECMP next hops, defined in sysdep code */
extern const int rt_default_ecmp;
@@ -781,7 +789,6 @@ void rta__free(rta *r);
static inline void rta_free(rta *r) { if (r && !--r->uc) rta__free(r); }
rta *rta_do_cow(rta *o, linpool *lp);
static inline rta * rta_cow(rta *r, linpool *lp) { return rta_is_cached(r) ? rta_do_cow(r, lp) : r; }
-static inline void rta_uncache(rta *r) { r->cached = 0; r->uc = 0; }
void rta_dump(rta *);
void rta_dump_all(void);
void rta_show(struct cli *, rta *);
diff --git a/nest/rt-attr.c b/nest/rt-attr.c
index 77fd3c3b..1bece201 100644
--- a/nest/rt-attr.c
+++ b/nest/rt-attr.c
@@ -1245,7 +1245,8 @@ rta_do_cow(rta *o, linpool *lp)
memcpy(*nhn, nho, nexthop_size(nho));
nhn = &((*nhn)->next);
}
- rta_uncache(r);
+ r->cached = 0;
+ r->uc = 0;
return r;
}
diff --git a/nest/rt-show.c b/nest/rt-show.c
index d942b8e1..235d72e4 100644
--- a/nest/rt-show.c
+++ b/nest/rt-show.c
@@ -95,10 +95,7 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary
}
if (d->verbose)
- {
- cli_printf(c, -1008, "\tInternal route ID: %uL %uG %uS", e->src->private_id, e->src->global_id, e->stale_cycle);
rta_show(c, a);
- }
}
static uint
@@ -106,7 +103,7 @@ rte_feed_count(net *n)
{
uint count = 0;
for (struct rte_storage *e = n->routes; e; e = e->next)
- if (rte_is_valid(RTES_OR_NULL(e)))
+ if (rte_is_valid(RTE_OR_NULL(e)))
count++;
return count;
}
@@ -116,7 +113,7 @@ 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(RTES_OR_NULL(e)))
+ if (rte_is_valid(RTE_OR_NULL(e)))
{
ASSERT_DIE(i < count);
feed[i++] = &e->rte;
diff --git a/nest/rt-table.c b/nest/rt-table.c
index 2f480992..e7ff2816 100644
--- a/nest/rt-table.c
+++ b/nest/rt-table.c
@@ -50,6 +50,7 @@ pool *rt_table_pool;
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);
@@ -385,11 +386,9 @@ rte_mergable(rte *pri, rte *sec)
static void
rte_trace(const char *name, const rte *e, int dir, const char *msg)
{
- log(L_TRACE "%s %c %s %N src %uL %uG %uS %s%s",
- name, dir, msg, e->net,
- e->src->private_id, e->src->global_id, e->stale_cycle,
- rta_dest_name(e->attrs->dest),
- rte_is_filtered(e) ? " (filtered)" : "");
+ log(L_TRACE "%s %c %s %N %uL %uG %s",
+ name, dir, msg, e->net, e->src->private_id, e->src->global_id,
+ rta_dest_name(e->attrs->dest));
}
static inline void
@@ -428,7 +427,7 @@ rte_feed_count(net *n)
{
uint count = 0;
for (struct rte_storage *e = n->routes; e; e = e->next)
- if (rte_is_valid(RTES_OR_NULL(e)))
+ if (rte_is_valid(RTE_OR_NULL(e)))
count++;
return count;
}
@@ -438,7 +437,7 @@ 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(RTES_OR_NULL(e)))
+ if (rte_is_valid(RTE_OR_NULL(e)))
{
ASSERT_DIE(i < count);
feed[i++] = &e->rte;
@@ -509,11 +508,10 @@ export_filter(struct channel *c, rte *rt, int silent)
return export_filter_(c, rt, rte_update_pool, silent);
}
-void do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte *old);
-
static void
do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old)
{
+ struct proto *p = c->proto;
struct channel_export_stats *stats = &c->export_stats;
if (c->refeeding && new)
@@ -530,31 +528,28 @@ do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old)
if (!new && old)
CHANNEL_LIMIT_POP(c, OUT);
- /* Store route export state */
- if (old)
- bmap_clear(&c->export_map, old->id);
-
- if (new)
- bmap_set(&c->export_map, new->id);
-
/* Apply export table */
+ struct rte_storage *old_exported = NULL;
if (c->out_table)
- rte_import(&c->out_table->push, net, new, old ? old->src : new->src);
- else
- do_rt_notify_direct(c, net, new, old);
-}
-
-void
-do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte *old)
-{
- struct proto *p = c->proto;
- struct channel_export_stats *stats = &c->export_stats;
+ {
+ if (!rte_update_out(c, net, new, old, &old_exported))
+ {
+ channel_rte_trace_out(D_ROUTES, c, new, "idempotent");
+ return;
+ }
+ }
if (new)
stats->updates_accepted++;
else
stats->withdraws_accepted++;
+ if (old)
+ bmap_clear(&c->export_map, old->id);
+
+ if (new)
+ bmap_set(&c->export_map, new->id);
+
if (p->debug & D_ROUTES)
{
if (new && old)
@@ -565,7 +560,10 @@ do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte
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(old_exported, c->out_table);
}
static void
@@ -786,7 +784,7 @@ rt_notify_optimal(struct rt_export_request *req, const net_addr *net, struct rt_
rte n0;
if (rpe->new_best != rpe->old_best)
- rt_notify_basic(c, net, RTES_CLONE(rpe->new_best, &n0), RTES_OR_NULL(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. */
@@ -801,7 +799,7 @@ rt_notify_any(struct rt_export_request *req, const net_addr *net, struct rt_pend
rte n0;
if (rpe->new != rpe->old)
- rt_notify_basic(c, net, RTES_CLONE(rpe->new, &n0), RTES_OR_NULL(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. */
@@ -976,10 +974,6 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr
rte *old_best = old_best_stored ? &old_best_stored->rte : NULL;
rte *old = NULL;
- /* Set the stale cycle unless already set */
- if (new && !(new->flags & REF_USE_STALE))
- new->stale_cycle = c->stale_set;
-
/* Find and remove original route from the same protocol */
struct rte_storage **before_old = rte_find(net, src);
@@ -1005,7 +999,8 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr
if (new && rte_same(old, new))
{
/* No changes, ignore the new route and refresh the old one */
- old->stale_cycle = new->stale_cycle;
+
+ old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY);
if (!rte_is_filtered(new))
{
@@ -1143,23 +1138,18 @@ rte_recalculate(struct rt_import_hook *c, net *net, rte *new, struct rte_src *sr
new_stored->rte.id = old->id;
}
- _Bool nb = (new_stored == net->routes);
- _Bool ob = (old_best == old);
-
/* Log the route change */
- if (new_ok && old_ok)
+ if (new_ok)
+ rt_rte_trace_in(D_ROUTES, req, &new_stored->rte, new_stored == net->routes ? "added [best]" : "added");
+ else if (old_ok)
{
- const char *best_indicator[2][2] = { { "updated", "updated [-best]" }, { "updated [+best]", "updated [best]" } };
- rt_rte_trace_in(D_ROUTES, req, &new_stored->rte, best_indicator[nb][ob]);
+ 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]");
}
- else if (new_ok)
- rt_rte_trace_in(D_ROUTES, req, &new_stored->rte,
- (!net->routes->next || !rte_is_ok(&net->routes->next->rte)) ? "added [sole]" :
- nb ? "added [best]" : "added");
- else if (old_ok)
- rt_rte_trace_in(D_ROUTES, req, old,
- (!net->routes || !rte_is_ok(&net->routes->rte)) ? "removed [sole]" :
- ob ? "removed [best]" : "removed");
/* Propagate the route change */
rte_announce(table, net, new_stored, old_stored,
@@ -1207,15 +1197,12 @@ channel_preimport(struct rt_import_request *req, rte *new, rte *old)
{
struct channel *c = SKIP_BACK(struct channel, in_req, req);
- if (!c->in_table)
- {
- if (new && !old)
- if (CHANNEL_LIMIT_PUSH(c, RX))
- return NULL;
+ if (new && !old)
+ if (CHANNEL_LIMIT_PUSH(c, RX))
+ return NULL;
- if (!new && old)
- CHANNEL_LIMIT_POP(c, RX);
- }
+ if (!new && old)
+ CHANNEL_LIMIT_POP(c, RX);
int new_in = new && !rte_is_filtered(new);
int old_in = old && !rte_is_filtered(old);
@@ -1236,22 +1223,7 @@ channel_preimport(struct rt_import_request *req, rte *new, rte *old)
return new;
}
-rte *
-channel_in_preimport(struct rt_import_request *req, rte *new, rte *old)
-{
- struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, push, req);
-
- if (new && !old)
- if (CHANNEL_LIMIT_PUSH(cat->c, RX))
- return NULL;
-
- if (!new && old)
- CHANNEL_LIMIT_POP(cat->c, RX);
-
- return new;
-}
-
-void rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src);
+static void rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src);
void
rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src)
@@ -1261,14 +1233,13 @@ rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src)
ASSERT(c->channel_state == CS_UP);
+ if (c->in_table && !rte_update_in(c, n, new, src))
+ return;
- if (c->in_table)
- rte_import(&c->in_table->push, n, new, src);
- else
- rte_update_direct(c, n, new, src);
+ return rte_update_direct(c, n, new, src);
}
-void
+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;
@@ -1352,6 +1323,24 @@ rte_discard(net *net, rte *old) /* Non-filtered route deletion, used during garb
rte_update_unlock();
}
+/* Modify existing route by protocol hook, used for long-lived graceful restart */
+static inline void
+rte_modify(net *net, rte *old)
+{
+ rte_update_lock();
+
+ rte *new = old->sender->req->rte_modify(old, rte_update_pool);
+ if (new != old)
+ {
+ if (new)
+ new->flags = old->flags & ~REF_MODIFY;
+
+ rte_recalculate(old->sender, net, new, old->src);
+ }
+
+ rte_update_unlock();
+}
+
/* Check rtable for best route to given net whether it would be exported do p */
int
rt_examine(rtable *t, net_addr *a, struct channel *c, const struct filter *filter)
@@ -1430,9 +1419,6 @@ rt_request_import(rtable *tab, struct rt_import_request *req)
hook->req = req;
hook->table = tab;
- if (!hook->stale_set)
- hook->stale_set = hook->stale_valid = hook->stale_pruning = hook->stale_pruned = 1;
-
rt_set_import_state(hook, TIS_UP);
hook->n = (node) {};
@@ -1513,41 +1499,20 @@ rt_stop_export(struct rt_export_request *req, void (*stopped)(struct rt_export_r
* routes to the routing table (by rte_update()). After that, all protocol
* routes (more precisely routes with @c as @sender) not sent during the
* refresh cycle but still in the table from the past are pruned. This is
- * implemented by setting rte->stale_cycle to req->stale_set in rte_update()
- * and then dropping all routes with old stale_cycle values in table prune loop. */
+ * implemented by marking all related routes as stale by REF_STALE flag in
+ * rt_refresh_begin(), then marking all related stale routes with REF_DISCARD
+ * flag in rt_refresh_end() and then removing such routes in the prune loop.
+ */
void
-rt_refresh_begin(struct rt_import_request *req)
+rt_refresh_begin(rtable *t, struct rt_import_request *req)
{
- struct rt_import_hook *hook = req->hook;
- ASSERT_DIE(hook);
-
- ASSERT_DIE(hook->stale_set == hook->stale_valid);
-
- /* If the pruning routine is too slow */
- if ((hook->stale_pruned < hook->stale_valid) && (hook->stale_pruned + 128 < hook->stale_valid)
- || (hook->stale_pruned > hook->stale_valid) && (hook->stale_pruned > hook->stale_valid + 128))
- {
- log(L_WARN "Route refresh flood in table %s", hook->table->name);
- FIB_WALK(&hook->table->fib, net, n)
- {
- for (struct rte_storage *e = n->routes; e; e = e->next)
- if (e->rte.sender == req->hook)
- e->rte.stale_cycle = 0;
- }
- FIB_WALK_END;
- hook->stale_set = 1;
- hook->stale_valid = 0;
- hook->stale_pruned = 0;
- }
- else if (!++hook->stale_set)
- {
- /* Let's reserve the stale_cycle zero value for always-invalid routes */
- hook->stale_set = 1;
- hook->stale_valid = 0;
- }
-
- if (req->trace_routes & D_STATES)
- log(L_TRACE "%s: route refresh begin [%u]", req->name, hook->stale_set);
+ FIB_WALK(&t->fib, net, n)
+ {
+ for (struct rte_storage *e = n->routes; e; e = e->next)
+ if (e->rte.sender == req->hook)
+ e->rte.flags |= REF_STALE;
+ }
+ FIB_WALK_END;
}
/**
@@ -1559,18 +1524,43 @@ rt_refresh_begin(struct rt_import_request *req)
* hook. See rt_refresh_begin() for description of refresh cycles.
*/
void
-rt_refresh_end(struct rt_import_request *req)
+rt_refresh_end(rtable *t, struct rt_import_request *req)
{
- struct rt_import_hook *hook = req->hook;
- ASSERT_DIE(hook);
+ int prune = 0;
- hook->stale_valid++;
- ASSERT_DIE(hook->stale_set == hook->stale_valid);
+ FIB_WALK(&t->fib, net, n)
+ {
+ for (struct rte_storage *e = n->routes; e; e = e->next)
+ if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE))
+ {
+ e->rte.flags |= REF_DISCARD;
+ prune = 1;
+ }
+ }
+ FIB_WALK_END;
- rt_schedule_prune(hook->table);
+ if (prune)
+ rt_schedule_prune(t);
+}
- if (req->trace_routes & D_STATES)
- log(L_TRACE "%s: route refresh end [%u]", req->name, hook->stale_valid);
+void
+rt_modify_stale(rtable *t, struct rt_import_request *req)
+{
+ int prune = 0;
+
+ FIB_WALK(&t->fib, net, n)
+ {
+ for (struct rte_storage *e = n->routes; e; e = e->next)
+ if ((e->rte.sender == req->hook) && (e->rte.flags & REF_STALE) && !(e->rte.flags & REF_FILTERED))
+ {
+ e->rte.flags |= REF_MODIFY;
+ prune = 1;
+ }
+ }
+ FIB_WALK_END;
+
+ if (prune)
+ rt_schedule_prune(t);
}
/**
@@ -1623,6 +1613,9 @@ 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
@@ -1665,6 +1658,9 @@ rt_dump_hooks_all(void)
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
@@ -1800,7 +1796,9 @@ rt_free(resource *_r)
DBG("Deleting routing table %s\n", r->name);
ASSERT_DIE(r->use_count == 0);
- ASSERT_DIE(r->deleted);
+
+ if (r->internal)
+ return;
r->config->table = NULL;
rem_node(&r->n);
@@ -1855,17 +1853,20 @@ rt_setup(pool *pp, struct rtable_config *cf)
fib_init(&t->fib, p, t->addr_type, sizeof(net), OFFSETOF(net, n), 0, NULL);
- init_list(&t->imports);
- init_list(&t->exports);
- hmap_init(&t->id_map, p, 1024);
- hmap_set(&t->id_map, 0);
+ 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);
+ init_list(&t->subscribers);
- t->rt_event = ev_new_init(p, rt_event, t);
- t->last_rt_change = t->gc_time = current_time();
+ t->rt_event = ev_new_init(p, rt_event, t);
+ t->last_rt_change = t->gc_time = current_time();
- t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS;
+ t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS;
+ }
return t;
}
@@ -1883,6 +1884,7 @@ rt_init(void)
rt_table_pool = rp_new(&root_pool, "Routing tables");
rte_update_pool = lp_new_default(rt_table_pool);
init_list(&routing_tables);
+ init_list(&deleted_routing_tables);
}
@@ -1923,13 +1925,6 @@ rt_prune_table(rtable *tab)
WALK_LIST2(ih, n, tab->imports, n)
if (ih->import_state == TIS_STOP)
rt_set_import_state(ih, TIS_FLUSHING);
- else if ((ih->stale_valid != ih->stale_pruning) && (ih->stale_pruning == ih->stale_pruned))
- {
- ih->stale_pruning = ih->stale_valid;
-
- if (ih->req->trace_routes & D_STATES)
- log(L_TRACE "%s: table prune after refresh begin [%u]", ih->req->name, ih->stale_pruning);
- }
FIB_ITERATE_INIT(fit, &tab->fib);
tab->prune_state = 2;
@@ -1941,11 +1936,7 @@ again:
rescan:
for (struct rte_storage *e=n->routes; e; e=e->next)
{
- struct rt_import_hook *s = e->rte.sender;
-
- if ((s->import_state == TIS_FLUSHING) ||
- (e->rte.stale_cycle < s->stale_valid) ||
- (e->rte.stale_cycle > s->stale_set))
+ if ((e->rte.sender->import_state == TIS_FLUSHING) || (e->rte.flags & REF_DISCARD))
{
if (limit <= 0)
{
@@ -1959,6 +1950,21 @@ again:
goto rescan;
}
+
+ if (e->rte.flags & REF_MODIFY)
+ {
+ if (limit <= 0)
+ {
+ FIB_ITERATE_PUT(fit);
+ ev_schedule(tab->rt_event);
+ return;
+ }
+
+ rte_modify(n, &e->rte);
+ limit--;
+
+ goto rescan;
+ }
}
if (!n->routes) /* Orphaned FIB entry */
@@ -1992,13 +1998,6 @@ again:
mb_free(ih);
rt_unlock_table(tab);
}
- else if (ih->stale_pruning != ih->stale_pruned)
- {
- ih->stale_pruned = ih->stale_pruning;
-
- if (ih->req->trace_routes & D_STATES)
- log(L_TRACE "%s: table prune after refresh end [%u]", ih->req->name, ih->stale_pruned);
- }
}
void
@@ -2205,10 +2204,7 @@ rt_next_hop_update_net(rtable *tab, net *n)
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] = {
- { "autoupdated", "autoupdated [-best]" },
- { "autoupdated [+best]", "autoupdated [best]" }
- };
+ 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);
}
@@ -2312,12 +2308,11 @@ rt_unlock_table(rtable *r)
{
if (!--r->use_count && r->deleted)
{
- void *del_data = r->del_data;
- void (*deleted)(void *) = r->deleted;
+ struct config *conf = r->deleted;
/* Delete the routing table by freeing its pool */
rt_shutdown(r);
- deleted(del_data);
+ config_del_obstacle(conf);
}
}
@@ -2328,8 +2323,6 @@ rt_find_table_config(struct config *cf, char *name)
return (sym && (sym->class == SYM_TABLE)) ? sym->table : NULL;
}
-static void rt_config_del_obstacle(void *data) { config_del_obstacle(data); }
-
/**
* rt_commit - commit new routing table configuration
* @new: new configuration
@@ -2368,10 +2361,9 @@ rt_commit(struct config *new, struct config *old)
else
{
DBG("\t%s: deleted\n", o->name);
- rt_lock_table(ot);
- ot->deleted = rt_config_del_obstacle;
- ot->del_data = old;
+ ot->deleted = old;
config_add_obstacle(old);
+ rt_lock_table(ot);
rt_unlock_table(ot);
}
}
@@ -2450,6 +2442,285 @@ done:
/*
+ * Import table
+ */
+
+int
+rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src)
+{
+ struct rtable *tab = c->in_table;
+ net *net;
+
+ if (new)
+ net = net_get(tab, n);
+ else
+ {
+ net = net_find(tab, n);
+
+ if (!net)
+ goto drop_withdraw;
+ }
+
+ /* Find the old rte */
+ struct rte_storage **pos = rte_find(net, src);
+ if (*pos)
+ {
+ rte *old = &(*pos)->rte;
+ if (new && rte_same(old, new))
+ {
+ /* Refresh the old rte, continue with update to main rtable */
+ if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY))
+ {
+ old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY);
+ return 1;
+ }
+
+ goto drop_update;
+ }
+
+ if (!new)
+ CHANNEL_LIMIT_POP(c, RX);
+
+ /* Move iterator if needed */
+ if (*pos == c->reload_next_rte)
+ c->reload_next_rte = (*pos)->next;
+
+ /* Remove the old rte */
+ struct rte_storage *del = *pos;
+ *pos = (*pos)->next;
+ rte_free(del, tab);
+ tab->rt_count--;
+ }
+ else if (new)
+ {
+ if (CHANNEL_LIMIT_PUSH(c, RX))
+ {
+ /* Required by rte_trace_in() */
+ new->net = n;
+
+ channel_rte_trace_in(D_FILTERS, c, new, "ignored [limit]");
+ goto drop_update;
+ }
+ }
+ else
+ goto drop_withdraw;
+
+ if (!new)
+ {
+ if (!net->routes)
+ fib_delete(&tab->fib, net);
+
+ return 1;
+ }
+
+ /* Insert the new rte */
+ struct rte_storage *e = rte_store(new, net, tab);
+ e->rte.lastmod = current_time();
+ e->next = *pos;
+ *pos = e;
+ tab->rt_count++;
+ return 1;
+
+drop_update:
+ c->import_stats.updates_received++;
+ c->in_req.hook->stats.updates_ignored++;
+
+ if (!net->routes)
+ fib_delete(&tab->fib, net);
+
+ return 0;
+
+drop_withdraw:
+ c->import_stats.withdraws_received++;
+ c->in_req.hook->stats.withdraws_ignored++;
+ return 0;
+}
+
+int
+rt_reload_channel(struct channel *c)
+{
+ struct rtable *tab = c->in_table;
+ struct fib_iterator *fit = &c->reload_fit;
+ int max_feed = 64;
+
+ ASSERT(c->channel_state == CS_UP);
+
+ if (!c->reload_active)
+ {
+ FIB_ITERATE_INIT(fit, &tab->fib);
+ c->reload_active = 1;
+ }
+
+ do {
+ for (struct rte_storage *e = c->reload_next_rte; e; e = e->next)
+ {
+ if (max_feed-- <= 0)
+ {
+ c->reload_next_rte = e;
+ debug("%s channel reload burst split (max_feed=%d)", c->proto->name, max_feed);
+ return 0;
+ }
+
+ rte r = e->rte;
+ rte_update_direct(c, r.net, &r, r.src);
+ }
+
+ c->reload_next_rte = NULL;
+
+ FIB_ITERATE_START(&tab->fib, fit, net, n)
+ {
+ if (c->reload_next_rte = n->routes)
+ {
+ FIB_ITERATE_PUT_NEXT(fit, &tab->fib);
+ break;
+ }
+ }
+ FIB_ITERATE_END;
+ }
+ while (c->reload_next_rte);
+
+ c->reload_active = 0;
+ return 1;
+}
+
+void
+rt_reload_channel_abort(struct channel *c)
+{
+ if (c->reload_active)
+ {
+ /* Unlink the iterator */
+ fit_get(&c->in_table->fib, &c->reload_fit);
+ c->reload_next_rte = NULL;
+ c->reload_active = 0;
+ }
+}
+
+void
+rt_prune_sync(rtable *t, int all)
+{
+ struct fib_iterator fit;
+
+ FIB_ITERATE_INIT(&fit, &t->fib);
+
+again:
+ FIB_ITERATE_START(&t->fib, &fit, net, n)
+ {
+ struct rte_storage *e, **ee = &n->routes;
+
+ while (e = *ee)
+ {
+ if (all || (e->rte.flags & (REF_STALE | REF_DISCARD)))
+ {
+ *ee = e->next;
+ rte_free(e, t);
+ t->rt_count--;
+ }
+ else
+ ee = &e->next;
+ }
+
+ if (all || !n->routes)
+ {
+ FIB_ITERATE_PUT(&fit);
+ fib_delete(&t->fib, n);
+ goto again;
+ }
+ }
+ FIB_ITERATE_END;
+}
+
+
+/*
+ * Export table
+ */
+
+int
+rte_update_out(struct channel *c, const net_addr *n, rte *new, const rte *old0, struct rte_storage **old_exported)
+{
+ struct rtable *tab = c->out_table;
+ struct rte_src *src;
+ net *net;
+
+ if (new)
+ {
+ net = net_get(tab, n);
+ src = new->src;
+ }
+ else
+ {
+ net = net_find(tab, n);
+ src = old0->src;
+
+ if (!net)
+ goto drop;
+ }
+
+ /* Find the old rte */
+ struct rte_storage **pos = (c->ra_mode == RA_ANY) ? rte_find(net, src) : &net->routes;
+ struct rte_storage *old = NULL;
+
+ if (old = *pos)
+ {
+ if (new && rte_same(&(*pos)->rte, new))
+ goto drop;
+
+ /* Remove the old rte */
+ *pos = old->next;
+ *old_exported = old;
+ tab->rt_count--;
+ }
+
+ if (!new)
+ {
+ if (!old)
+ goto drop;
+
+ if (!net->routes)
+ fib_delete(&tab->fib, net);
+
+ return 1;
+ }
+
+ /* Insert the new rte */
+ struct rte_storage *e = rte_store(new, net, tab);
+ e->rte.lastmod = current_time();
+ e->next = *pos;
+ *pos = e;
+ tab->rt_count++;
+ return 1;
+
+drop:
+ return 0;
+}
+
+void
+rt_refeed_channel(struct channel *c)
+{
+ 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
*/
diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c
index 892b26e3..90490b4f 100644
--- a/proto/bgp/attrs.c
+++ b/proto/bgp/attrs.c
@@ -2267,44 +2267,30 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best)
return !old_suppressed;
}
-void
-bgp_rte_modify_stale(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe UNUSED, rte **feed, uint count)
+rte *
+bgp_rte_modify_stale(struct rte *r, struct linpool *pool)
{
- struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req);
-
- do {
- rte *r = feed[--count];
- if (r->sender != c->c.in_req.hook)
- continue;
-
- /* A new route, do not mark as stale */
- if (r->stale_cycle == c->c.in_req.hook->stale_set)
- continue;
-
- eattr *ea = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY));
- const struct adata *ad = ea ? ea->u.ptr : NULL;
- uint flags = ea ? ea->flags : BAF_PARTIAL;
+ eattr *ea = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY));
+ const struct adata *ad = ea ? ea->u.ptr : NULL;
+ uint flags = ea ? ea->flags : BAF_PARTIAL;
- rte e0 = *r;
- e0.flags |= REF_USE_STALE;
-
- if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR))
- rte_import(&c->c.in_req, n, NULL, r->src);
+ if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR))
+ return NULL;
- else if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE))
- rte_import(&c->c.in_req, n, &e0, r->src);
+ if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE))
+ return r;
- else {
- rta *a = e0.attrs = rta_do_cow(r->attrs, bgp_linpool);
+ rta *a = rta_do_cow(r->attrs, pool);
+
+ _Thread_local static rte e0;
+ e0 = *r;
+ e0.attrs = a;
- bgp_set_attr_ptr(&(a->eattrs), bgp_linpool, BA_COMMUNITY, flags,
- int_set_add(bgp_linpool, ad, BGP_COMM_LLGR_STALE));
- e0.pflags |= BGP_REF_STALE;
+ bgp_set_attr_ptr(&(a->eattrs), pool, BA_COMMUNITY, flags,
+ int_set_add(pool, ad, BGP_COMM_LLGR_STALE));
+ e0.pflags |= BGP_REF_STALE;
- rte_import(&c->c.in_req, n, &e0, r->src);
- lp_flush(bgp_linpool);
- }
- } while (count);
+ return &e0;
}
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 35e9ea59..78c36bc7 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -140,15 +140,6 @@ static void bgp_update_bfd(struct bgp_proto *p, const struct bfd_options *bfd);
static int bgp_incoming_connection(sock *sk, uint dummy UNUSED);
static void bgp_listen_sock_err(sock *sk UNUSED, int err);
-static void bgp_graceful_restart_feed(struct bgp_channel *c);
-static inline void channel_refresh_end_reload(struct channel *c)
-{
- channel_refresh_end(c);
-
- if (c->in_table)
- channel_request_reload(c);
-}
-
/**
* bgp_open - open a BGP instance
* @p: BGP instance
@@ -784,25 +775,25 @@ bgp_handle_graceful_restart(struct bgp_proto *p)
{
case BGP_GRS_NONE:
c->gr_active = BGP_GRS_ACTIVE;
- channel_refresh_begin(&c->c);
+ rt_refresh_begin(c->c.table, &c->c.in_req);
break;
case BGP_GRS_ACTIVE:
- channel_refresh_end(&c->c);
- channel_refresh_begin(&c->c);
+ rt_refresh_end(c->c.table, &c->c.in_req);
+ rt_refresh_begin(c->c.table, &c->c.in_req);
break;
case BGP_GRS_LLGR:
- channel_refresh_begin(&c->c);
- bgp_graceful_restart_feed(c);
+ rt_refresh_begin(c->c.table, &c->c.in_req);
+ rt_modify_stale(c->c.table, &c->c.in_req);
break;
}
}
else
{
/* Just flush the routes */
- channel_refresh_begin(&c->c);
- channel_refresh_end(&c->c);
+ rt_refresh_begin(c->c.table, &c->c.in_req);
+ rt_refresh_end(c->c.table, &c->c.in_req);
}
/* Reset bucket and prefix tables */
@@ -820,50 +811,6 @@ bgp_handle_graceful_restart(struct bgp_proto *p)
tm_start(p->gr_timer, p->conn->remote_caps->gr_time S);
}
-static void
-bgp_graceful_restart_feed_done(struct rt_export_request *req)
-{
- req->hook = NULL;
-}
-
-static void
-bgp_graceful_restart_feed_dump_req(struct rt_export_request *req)
-{
- struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req);
- debug(" BGP-GR %s.%s export request %p\n", c->c.proto->name, c->c.name, req);
-}
-
-static void
-bgp_graceful_restart_feed_log_state_change(struct rt_export_request *req, u8 state)
-{
- struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req);
- struct bgp_proto *p = (void *) c->c.proto;
- BGP_TRACE(D_EVENTS, "Long-lived graceful restart export state changed to %s", rt_export_state_name(state));
-
- if (state == TES_READY)
- rt_stop_export(req, bgp_graceful_restart_feed_done);
-}
-
-static void
-bgp_graceful_restart_drop_export(struct rt_export_request *req UNUSED, const net_addr *n UNUSED, struct rt_pending_export *rpe UNUSED)
-{ /* Nothing to do */ }
-
-static void
-bgp_graceful_restart_feed(struct bgp_channel *c)
-{
- c->stale_feed = (struct rt_export_request) {
- .name = "BGP-GR",
- .trace_routes = c->c.debug | c->c.proto->debug,
- .dump_req = bgp_graceful_restart_feed_dump_req,
- .log_state_change = bgp_graceful_restart_feed_log_state_change,
- .export_bulk = bgp_rte_modify_stale,
- .export_one = bgp_graceful_restart_drop_export,
- };
-
- rt_request_export(c->c.table, &c->stale_feed);
-}
-
-
/**
* bgp_graceful_restart_done - finish active BGP graceful restart
* @c: BGP channel
@@ -886,11 +833,8 @@ bgp_graceful_restart_done(struct bgp_channel *c)
if (!p->gr_active_num)
BGP_TRACE(D_EVENTS, "Neighbor graceful restart done");
- if (c->stale_feed.hook)
- rt_stop_export(&c->stale_feed, bgp_graceful_restart_feed_done);
-
tm_stop(c->stale_timer);
- channel_refresh_end_reload(&c->c);
+ rt_refresh_end(c->c.table, &c->c.in_req);
}
/**
@@ -932,7 +876,7 @@ bgp_graceful_restart_timeout(timer *t)
/* Channel is in GR, and supports LLGR -> start LLGR */
c->gr_active = BGP_GRS_LLGR;
tm_start(c->stale_timer, c->stale_time S);
- bgp_graceful_restart_feed(c);
+ rt_modify_stale(c->c.table, &c->c.in_req);
}
}
else
@@ -970,7 +914,10 @@ bgp_refresh_begin(struct bgp_channel *c)
{ log(L_WARN "%s: BEGIN-OF-RR received before END-OF-RIB, ignoring", p->p.name); return; }
c->load_state = BFS_REFRESHING;
- channel_refresh_begin(&c->c);
+ rt_refresh_begin(c->c.table, &c->c.in_req);
+
+ if (c->c.in_table)
+ rt_refresh_begin(c->c.in_table, &c->c.in_req);
}
/**
@@ -991,7 +938,10 @@ bgp_refresh_end(struct bgp_channel *c)
{ log(L_WARN "%s: END-OF-RR received without prior BEGIN-OF-RR, ignoring", p->p.name); return; }
c->load_state = BFS_NONE;
- channel_refresh_end_reload(&c->c);
+ rt_refresh_end(c->c.table, &c->c.in_req);
+
+ if (c->c.in_table)
+ rt_prune_sync(c->c.in_table, 0);
}
@@ -1458,9 +1408,12 @@ bgp_reload_routes(struct channel *C)
struct bgp_proto *p = (void *) C->proto;
struct bgp_channel *c = (void *) C;
- ASSERT(p->conn && (p->route_refresh));
+ ASSERT(p->conn && (p->route_refresh || c->c.in_table));
- bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH);
+ if (c->c.in_table)
+ channel_schedule_reload(C);
+ else
+ bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH);
}
static void
@@ -1740,6 +1693,7 @@ bgp_init(struct proto_config *CF)
P->rte_better = bgp_rte_better;
P->rte_mergable = bgp_rte_mergable;
P->rte_recalculate = cf->deterministic_med ? bgp_rte_recalculate : NULL;
+ P->rte_modify = bgp_rte_modify_stale;
P->rte_igp_metric = bgp_rte_igp_metric;
p->cf = cf;
@@ -1802,7 +1756,7 @@ bgp_channel_start(struct channel *C)
bgp_init_prefix_table(c);
if (c->cf->import_table)
- channel_setup_in_table(C, 0);
+ channel_setup_in_table(C);
if (c->cf->export_table)
channel_setup_out_table(C);
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index 342dc023..c79dd1b2 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -366,7 +366,6 @@ struct bgp_channel {
timer *stale_timer; /* Long-lived stale timer for LLGR */
u32 stale_time; /* Stored LLGR stale time from last session */
- struct rt_export_request stale_feed; /* Feeder request for stale route modification */
u8 add_path_rx; /* Session expects receive of ADD-PATH extended NLRI */
u8 add_path_tx; /* Session expects transmit of ADD-PATH extended NLRI */
@@ -586,7 +585,7 @@ void bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *bp);
int bgp_rte_better(struct rte *, struct rte *);
int bgp_rte_mergable(rte *pri, rte *sec);
int bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best);
-void bgp_rte_modify_stale(struct rt_export_request *, const net_addr *, struct rt_pending_export *, rte **, uint);
+struct rte *bgp_rte_modify_stale(struct rte *r, struct linpool *pool);
u32 bgp_rte_igp_metric(struct rte *);
void bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old);
int bgp_preexport(struct channel *, struct rte *);
diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c
index 647551e5..f1e6d7d2 100644
--- a/proto/bgp/packets.c
+++ b/proto/bgp/packets.c
@@ -2695,7 +2695,7 @@ bgp_rx_route_refresh(struct bgp_conn *conn, byte *pkt, uint len)
{
case BGP_RR_REQUEST:
BGP_TRACE(D_PACKETS, "Got ROUTE-REFRESH");
- channel_request_feeding(&c->c);
+ rt_refeed_channel(&c->c);
break;
case BGP_RR_BEGIN:
diff --git a/proto/rpki/packets.c b/proto/rpki/packets.c
index 897edc09..943485d7 100644
--- a/proto/rpki/packets.c
+++ b/proto/rpki/packets.c
@@ -661,9 +661,9 @@ rpki_handle_cache_response_pdu(struct rpki_cache *cache, const struct pdu_cache_
* a refresh cycle.
*/
if (cache->p->roa4_channel)
- rt_refresh_begin(&cache->p->roa4_channel->in_req);
+ rt_refresh_begin(cache->p->roa4_channel->table, &cache->p->roa4_channel->in_req);
if (cache->p->roa6_channel)
- rt_refresh_begin(&cache->p->roa6_channel->in_req);
+ rt_refresh_begin(cache->p->roa6_channel->table, &cache->p->roa6_channel->in_req);
cache->p->refresh_channels = 1;
}
@@ -819,9 +819,9 @@ rpki_handle_end_of_data_pdu(struct rpki_cache *cache, const struct pdu_end_of_da
{
cache->p->refresh_channels = 0;
if (cache->p->roa4_channel)
- rt_refresh_end(&cache->p->roa4_channel->in_req);
+ rt_refresh_end(cache->p->roa4_channel->table, &cache->p->roa4_channel->in_req);
if (cache->p->roa6_channel)
- rt_refresh_end(&cache->p->roa6_channel->in_req);
+ rt_refresh_end(cache->p->roa6_channel->table, &cache->p->roa6_channel->in_req);
}
cache->last_update = current_time();
diff --git a/sysdep/unix/krt.c b/sysdep/unix/krt.c
index 609ee921..40a58442 100644
--- a/sysdep/unix/krt.c
+++ b/sysdep/unix/krt.c
@@ -285,24 +285,249 @@ krt_metric(rte *a)
}
static inline int
-krt_rte_better(rte *a, rte *b)
+krt_same_key(rte *a, rte *b)
{
- return (krt_metric(a) > krt_metric(b));
+ return (krt_metric(a) == krt_metric(b));
+}
+
+static inline int
+krt_uptodate(rte *a, rte *b)
+{
+ return (a->attrs == b->attrs);
+}
+
+static void
+krt_learn_announce_update(struct krt_proto *p, rte *e)
+{
+ rte e0 = {
+ .attrs = rta_clone(e->attrs),
+ .src = p->p.main_source,
+ };
+
+ rte_update(p->p.main_channel, e->net, &e0, p->p.main_source);
+}
+
+static void
+krt_learn_announce_delete(struct krt_proto *p, net_addr *n)
+{
+ rte_update(p->p.main_channel, n, NULL, p->p.main_source);
}
/* Called when alien route is discovered during scan */
static void
-krt_learn_rte(struct krt_proto *p, rte *e)
+krt_learn_scan(struct krt_proto *p, rte *e)
+{
+ net *n = net_get(p->krt_table, e->net);
+ struct rte_storage *m, **mm;
+
+ struct rte_storage *ee = rte_store(e, n, p->krt_table);
+
+ for(mm = &n->routes; m = *mm; mm = &m->next)
+ if (krt_same_key(&m->rte, e))
+ break;
+ if (m)
+ {
+ if (krt_uptodate(&m->rte, e))
+ {
+ krt_trace_in_rl(&rl_alien, p, e, "[alien] seen");
+ rte_free(ee, p->krt_table);
+ m->rte.pflags |= KRT_REF_SEEN;
+ }
+ else
+ {
+ krt_trace_in(p, e, "[alien] updated");
+ *mm = m->next;
+ rte_free(m, p->krt_table);
+ m = NULL;
+ }
+ }
+ else
+ krt_trace_in(p, e, "[alien] created");
+
+ if (!m)
+ {
+ ee->next = n->routes;
+ n->routes = ee;
+ ee->rte.pflags |= KRT_REF_SEEN;
+ }
+}
+
+static void
+krt_learn_prune(struct krt_proto *p)
{
- e->src = rt_get_source(&p->p, krt_metric(e));
- rte_update(p->p.main_channel, e->net, e, e->src);
+ struct fib *fib = &p->krt_table->fib;
+ struct fib_iterator fit;
+
+ KRT_TRACE(p, D_EVENTS, "Pruning inherited routes");
+
+ FIB_ITERATE_INIT(&fit, fib);
+again:
+ FIB_ITERATE_START(fib, &fit, net, n)
+ {
+ struct rte_storage *e, **ee, *best, **pbest, *old_best;
+
+ /*
+ * Note that old_best may be NULL even if there was an old best route in
+ * the previous step, because it might be replaced in krt_learn_scan().
+ * But in that case there is a new valid best route.
+ */
+
+ old_best = NULL;
+ best = NULL;
+ pbest = NULL;
+ ee = &n->routes;
+ while (e = *ee)
+ {
+ if (e->rte.pflags & KRT_REF_BEST)
+ old_best = e;
+
+ if (!(e->rte.pflags & KRT_REF_SEEN))
+ {
+ *ee = e->next;
+ rte_free(e, p->krt_table);
+ continue;
+ }
+
+ if (!best || krt_metric(&best->rte) > krt_metric(&e->rte))
+ {
+ best = e;
+ pbest = ee;
+ }
+
+ e->rte.pflags &= ~(KRT_REF_SEEN | KRT_REF_BEST);
+ ee = &e->next;
+ }
+ if (!n->routes)
+ {
+ DBG("%I/%d: deleting\n", n->n.prefix, n->n.pxlen);
+ if (old_best)
+ krt_learn_announce_delete(p, n->n.addr);
+
+ FIB_ITERATE_PUT(&fit);
+ fib_delete(fib, n);
+ goto again;
+ }
+
+ best->rte.pflags |= KRT_REF_BEST;
+ *pbest = best->next;
+ best->next = n->routes;
+ n->routes = best;
+
+ if ((best != old_best) || p->reload)
+ {
+ DBG("%I/%d: announcing (metric=%d)\n", n->n.prefix, n->n.pxlen, krt_metric(&best->rte));
+ krt_learn_announce_update(p, &best->rte);
+ }
+ else
+ DBG("%I/%d: uptodate (metric=%d)\n", n->n.prefix, n->n.pxlen, krt_metric(&best->rte));
+ }
+ FIB_ITERATE_END;
+
+ p->reload = 0;
+}
+
+static void
+krt_learn_async(struct krt_proto *p, rte *e, int new)
+{
+ net *n = net_get(p->krt_table, e->net);
+ struct rte_storage *g, **gg, *best, **bestp, *old_best;
+
+ ASSERT(!e->attrs->cached);
+ e->attrs->pref = p->p.main_channel->preference;
+
+ struct rte_storage *ee = rte_store(e, n, p->krt_table);
+
+ old_best = n->routes;
+ for(gg=&n->routes; g = *gg; gg = &g->next)
+ if (krt_same_key(&g->rte, e))
+ break;
+ if (new)
+ {
+ if (g)
+ {
+ if (krt_uptodate(&g->rte, e))
+ {
+ krt_trace_in(p, e, "[alien async] same");
+ rte_free(ee, p->krt_table);
+ return;
+ }
+ krt_trace_in(p, e, "[alien async] updated");
+ *gg = g->next;
+ rte_free(g, p->krt_table);
+ }
+ else
+ krt_trace_in(p, e, "[alien async] created");
+
+ ee->next = n->routes;
+ n->routes = ee;
+ }
+ else if (!g)
+ {
+ krt_trace_in(p, e, "[alien async] delete failed");
+ rte_free(ee, p->krt_table);
+ return;
+ }
+ else
+ {
+ krt_trace_in(p, e, "[alien async] removed");
+ *gg = g->next;
+ rte_free(ee, p->krt_table);
+ rte_free(g, p->krt_table);
+ }
+ best = n->routes;
+ bestp = &n->routes;
+ for(gg=&n->routes; g=*gg; gg=&g->next)
+ {
+ if (krt_metric(&best->rte) > krt_metric(&g->rte))
+ {
+ best = g;
+ bestp = gg;
+ }
+
+ g->rte.pflags &= ~KRT_REF_BEST;
+ }
+
+ if (best)
+ {
+ best->rte.pflags |= KRT_REF_BEST;
+ *bestp = best->next;
+ best->next = n->routes;
+ n->routes = best;
+ }
+
+ if (best != old_best)
+ {
+ DBG("krt_learn_async: distributing change\n");
+ if (best)
+ krt_learn_announce_update(p, &best->rte);
+ else
+ krt_learn_announce_delete(p, n->n.addr);
+ }
}
static void
krt_learn_init(struct krt_proto *p)
{
if (KRT_CF->learn)
- channel_setup_in_table(p->p.main_channel, 1);
+ {
+ struct rtable_config *cf = mb_allocz(p->p.pool, sizeof(struct rtable_config));
+ cf->name = "Inherited";
+ cf->addr_type = p->p.net_type;
+ cf->internal = 1;
+
+ p->krt_table = rt_setup(p->p.pool, cf);
+ }
+}
+
+static void
+krt_dump(struct proto *P)
+{
+ struct krt_proto *p = (struct krt_proto *) P;
+
+ if (!KRT_CF->learn)
+ return;
+ debug("KRT: Table of inheritable routes\n");
+ rt_dump(p->krt_table);
}
#endif
@@ -322,7 +547,7 @@ rte_feed_count(net *n)
{
uint count = 0;
for (struct rte_storage *e = n->routes; e; e = e->next)
- if (rte_is_valid(RTES_OR_NULL(e)))
+ if (rte_is_valid(RTE_OR_NULL(e)))
count++;
return count;
}
@@ -332,7 +557,7 @@ 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(RTES_OR_NULL(e)))
+ if (rte_is_valid(RTE_OR_NULL(e)))
{
ASSERT_DIE(i < count);
feed[i++] = &e->rte;
@@ -418,7 +643,7 @@ krt_got_route(struct krt_proto *p, rte *e, s8 src)
case KRT_SRC_ALIEN:
if (KRT_CF->learn)
- krt_learn_rte(p, e);
+ krt_learn_scan(p, e);
else
krt_trace_in_rl(&rl_alien, p, e, "[alien] ignored");
return;
@@ -487,11 +712,6 @@ static void
krt_init_scan(struct krt_proto *p)
{
bmap_reset(&p->seen_map, 1024);
-
-#ifdef KRT_ALLOW_LEARN
- if (KRT_CF->learn)
- channel_refresh_begin(p->p.main_channel);
-#endif
}
static void
@@ -519,7 +739,7 @@ krt_prune(struct krt_proto *p)
#ifdef KRT_ALLOW_LEARN
if (KRT_CF->learn)
- channel_refresh_end(p->p.main_channel);
+ krt_learn_prune(p);
#endif
if (p->ready)
@@ -561,7 +781,7 @@ krt_got_route_async(struct krt_proto *p, rte *e, int new, s8 src)
case KRT_SRC_ALIEN:
if (KRT_CF->learn)
{
- krt_learn_rte(p, e);
+ krt_learn_async(p, e, new);
return;
}
#endif
@@ -807,7 +1027,6 @@ krt_init(struct proto_config *CF)
p->p.if_notify = krt_if_notify;
p->p.reload_routes = krt_reload_routes;
p->p.feed_end = krt_feed_end;
- p->p.rte_better = krt_rte_better;
krt_sys_init(p);
return &p->p;
@@ -963,4 +1182,7 @@ struct protocol proto_unix_kernel = {
.reconfigure = krt_reconfigure,
.copy_config = krt_copy_config,
.get_attr = krt_get_attr,
+#ifdef KRT_ALLOW_LEARN
+ .dump = krt_dump,
+#endif
};
diff --git a/sysdep/unix/krt.h b/sysdep/unix/krt.h
index 968c5b16..f6ad6fde 100644
--- a/sysdep/unix/krt.h
+++ b/sysdep/unix/krt.h
@@ -51,6 +51,10 @@ struct krt_proto {
struct proto p;
struct krt_state sys; /* Sysdep state */
+#ifdef KRT_ALLOW_LEARN
+ struct rtable *krt_table; /* Internal table of inherited routes */
+#endif
+
#ifndef CONFIG_ALL_TABLES_AT_ONCE
timer *scan_timer;
#endif