KRT: Scan routing tables separetely on linux to avoid congestion

Remove compile-time sysdep option CONFIG_ALL_TABLES_AT_ONCE, replace it
with runtime ability to run either separate table scans or shared scan.

On Linux, use separate table scans by default when the netlink socket
option NETLINK_GET_STRICT_CHK is available, but retreat to shared scan
when it fails.

Running separate table scans has advantages where some routing tables are
managed independently, e.g. when multiple routing daemons are running on
the same machine, as kernel routing table modification performance is
significantly reduced when the table is modified while it is being
scanned.

Thanks Daniel Gröber for the original patch and Toke Høiland-Jørgensen
for suggestions.

5 files changed, 78 insertions, 49 deletions

diff --git a/sysdep/cf/README b/sysdep/cf/README
index 9a7a4afa..68078bbe 100644
--- a/sysdep/cf/README
+++ b/sysdep/cf/README
@@ -4,7 +4,6 @@ Available configuration variables:
 CONFIG_AUTO_ROUTES	Device routes are added automagically by the kernel
 CONFIG_SELF_CONSCIOUS	We're able to recognize whether route was installed by us
 CONFIG_MULTIPLE_TABLES	The kernel supports multiple routing tables
-CONFIG_ALL_TABLES_AT_ONCE	Kernel scanner wants to process all tables at once
 CONFIG_SINGLE_ROUTE	There is only one route per network
 
 CONFIG_MC_PROPER_SRC	Multicast packets have source address according to socket saddr field
diff --git a/sysdep/cf/linux.h b/sysdep/cf/linux.h
index 047d3764..c640bef4 100644
--- a/sysdep/cf/linux.h
+++ b/sysdep/cf/linux.h
@@ -9,7 +9,6 @@
 #define CONFIG_AUTO_ROUTES
 #define CONFIG_SELF_CONSCIOUS
 #define CONFIG_MULTIPLE_TABLES
-#define CONFIG_ALL_TABLES_AT_ONCE
 #define CONFIG_IP6_SADR_KERNEL
 
 #define CONFIG_MC_PROPER_SRC
diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c
index 29b744cb..bc65e0d2 100644
--- a/sysdep/linux/netlink.c
+++ b/sysdep/linux/netlink.c
@@ -161,16 +161,13 @@ nl_open_sock(struct nl_sock *nl)
     }
 }
 
-static void
+static int
 nl_set_strict_dump(struct nl_sock *nl UNUSED, int strict UNUSED)
 {
-  /*
-   * Strict checking is not necessary, it improves behavior on newer kernels.
-   * If it is not available (missing SOL_NETLINK compile-time, or ENOPROTOOPT
-   * run-time), we can just ignore it.
-   */
 #ifdef SOL_NETLINK
-  setsockopt(nl->fd, SOL_NETLINK, NETLINK_GET_STRICT_CHK, &strict, sizeof(strict));
+  return setsockopt(nl->fd, SOL_NETLINK, NETLINK_GET_STRICT_CHK, &strict, sizeof(strict));
+#else
+  return -1;
 #endif
 }
 
@@ -198,10 +195,17 @@ nl_cfg_rx_buffer_size(struct config *cfg)
 static void
 nl_open(void)
 {
+  if ((nl_scan.fd >= 0) && (nl_req.fd >= 0))
+    return;
+
   nl_open_sock(&nl_scan);
   nl_open_sock(&nl_req);
 
-  nl_set_strict_dump(&nl_scan, 1);
+  if (nl_set_strict_dump(&nl_scan, 1) < 0)
+  {
+    log(L_WARN "KRT: Netlink strict checking failed, will scan all tables at once");
+    krt_use_shared_scan();
+  }
 }
 
 static void
@@ -256,11 +260,13 @@ nl_request_dump_addr(int af)
 }
 
 static void
-nl_request_dump_route(int af)
+nl_request_dump_route(int af, int table_id)
 {
   struct {
     struct nlmsghdr nh;
     struct rtmsg rtm;
+    struct rtattr rta;
+    u32 table_id;
   } req = {
     .nh.nlmsg_type = RTM_GETROUTE,
     .nh.nlmsg_len = NLMSG_LENGTH(sizeof(struct rtmsg)),
@@ -269,7 +275,17 @@ nl_request_dump_route(int af)
     .rtm.rtm_family = af,
   };
 
-  send(nl_scan.fd, &req, sizeof(req), 0);
+  if (table_id < 256)
+    req.rtm.rtm_table = table_id;
+  else
+  {
+    req.rta.rta_type = RTA_TABLE;
+    req.rta.rta_len = RTA_LENGTH(4);
+    req.table_id = table_id;
+    req.nh.nlmsg_len = NLMSG_ALIGN(req.nh.nlmsg_len) + req.rta.rta_len;
+  }
+
+  send(nl_scan.fd, &req, req.nh.nlmsg_len, 0);
   nl_scan.last_hdr = NULL;
 }
 
@@ -1976,18 +1992,27 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h)
 }
 
 void
-krt_do_scan(struct krt_proto *p UNUSED)	/* CONFIG_ALL_TABLES_AT_ONCE => p is NULL */
+krt_do_scan(struct krt_proto *p)
 {
   struct nlmsghdr *h;
   struct nl_parse_state s;
 
   nl_parse_begin(&s, 1);
-  nl_request_dump_route(AF_UNSPEC);
+
+  /* Table-specific scan or shared scan */
+  if (p)
+    nl_request_dump_route(p->af, krt_table_id(p));
+  else
+    nl_request_dump_route(AF_UNSPEC, 0);
+
   while (h = nl_get_scan())
+  {
     if (h->nlmsg_type == RTM_NEWROUTE || h->nlmsg_type == RTM_DELROUTE)
       nl_parse_route(&s, h);
     else
       log(L_DEBUG "nl_scan_fire: Unknown packet received (type=%d)", h->nlmsg_type);
+  }
+
   nl_parse_end(&s);
 }
 
diff --git a/sysdep/unix/krt.c b/sysdep/unix/krt.c
index a02cf977..342adee5 100644
--- a/sysdep/unix/krt.c
+++ b/sysdep/unix/krt.c
@@ -783,18 +783,17 @@ krt_got_route_async(struct krt_proto *p, rte *e, int new)
   rte_free(e);
 }
 
+
 /*
  *	Periodic scanning
  */
 
-
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
-
-static timer *krt_scan_timer;
-static int krt_scan_count;
+static timer *krt_scan_all_timer;
+static int krt_scan_all_count;
+static _Bool krt_scan_all_tables;
 
 static void
-krt_scan(timer *t UNUSED)
+krt_scan_all(timer *t UNUSED)
 {
   struct krt_proto *p;
   node *n;
@@ -815,35 +814,42 @@ krt_scan(timer *t UNUSED)
 }
 
 static void
-krt_scan_timer_start(struct krt_proto *p)
+krt_scan_all_timer_start(struct krt_proto *p)
 {
-  if (!krt_scan_count)
-    krt_scan_timer = tm_new_init(krt_pool, krt_scan, NULL, KRT_CF->scan_time, 0);
+  if (!krt_scan_all_count)
+    krt_scan_all_timer = tm_new_init(krt_pool, krt_scan_all, NULL, KRT_CF->scan_time, 0);
 
-  krt_scan_count++;
+  krt_scan_all_count++;
 
-  tm_start(krt_scan_timer, 1 S);
+  tm_start(krt_scan_all_timer, 1 S);
 }
 
 static void
-krt_scan_timer_stop(struct krt_proto *p UNUSED)
+krt_scan_all_timer_stop(void)
 {
-  krt_scan_count--;
+  ASSERT(krt_scan_all_count > 0);
+
+  krt_scan_all_count--;
 
-  if (!krt_scan_count)
+  if (!krt_scan_all_count)
   {
-    rfree(krt_scan_timer);
-    krt_scan_timer = NULL;
+    rfree(krt_scan_all_timer);
+    krt_scan_all_timer = NULL;
   }
 }
 
 static void
-krt_scan_timer_kick(struct krt_proto *p UNUSED)
+krt_scan_all_timer_kick(void)
 {
-  tm_start(krt_scan_timer, 0);
+  tm_start(krt_scan_all_timer, 0);
+}
+
+void
+krt_use_shared_scan(void)
+{
+  krt_scan_all_tables = 1;
 }
 
-#else
 
 static void
 krt_scan(timer *t)
@@ -861,26 +867,33 @@ krt_scan(timer *t)
 static void
 krt_scan_timer_start(struct krt_proto *p)
 {
-  p->scan_timer = tm_new_init(p->p.pool, krt_scan, p, KRT_CF->scan_time, 0);
-  tm_start(p->scan_timer, 1 S);
+  if (krt_scan_all_tables)
+    krt_scan_all_timer_start(p);
+  else
+  {
+    p->scan_timer = tm_new_init(p->p.pool, krt_scan, p, KRT_CF->scan_time, 0);
+    tm_start(p->scan_timer, 1 S);
+  }
 }
 
 static void
 krt_scan_timer_stop(struct krt_proto *p)
 {
-  tm_stop(p->scan_timer);
+  if (krt_scan_all_tables)
+    krt_scan_all_timer_stop();
+  else
+    tm_stop(p->scan_timer);
 }
 
 static void
 krt_scan_timer_kick(struct krt_proto *p)
 {
-  tm_start(p->scan_timer, 0);
+  if (krt_scan_all_tables)
+    krt_scan_all_timer_kick();
+  else
+    tm_start(p->scan_timer, 0);
 }
 
-#endif
-
-
-
 
 /*
  *	Updates
@@ -1016,11 +1029,6 @@ krt_postconfig(struct proto_config *CF)
   if (! proto_cf_main_channel(CF))
     cf_error("Channel not specified");
 
-#ifdef CONFIG_ALL_TABLES_AT_ONCE
-  if (krt_cf->scan_time != cf->scan_time)
-    cf_error("All kernel syncers must use the same table scan interval");
-#endif
-
   struct channel_config *cc = proto_cf_main_channel(CF);
   struct rtable_config *tab = cc->table;
   if (tab->krt_attached)
diff --git a/sysdep/unix/krt.h b/sysdep/unix/krt.h
index 62228f08..1536e502 100644
--- a/sysdep/unix/krt.h
+++ b/sysdep/unix/krt.h
@@ -52,10 +52,7 @@ struct krt_proto {
   struct rtable *krt_table;	/* Internal table of inherited routes */
 #endif
 
-#ifndef CONFIG_ALL_TABLES_AT_ONCE
   timer *scan_timer;
-#endif
-
   struct bmap sync_map;		/* Keeps track which exported routes were successfully written to kernel */
   struct bmap seen_map;		/* Routes seen during last periodic scan */
   node krt_node;		/* Node in krt_proto_list */
@@ -76,6 +73,7 @@ extern pool *krt_pool;
 
 struct proto_config * kif_init_config(int class);
 void kif_request_scan(void);
+void krt_use_shared_scan(void);
 void krt_got_route(struct krt_proto *p, struct rte *e);
 void krt_got_route_async(struct krt_proto *p, struct rte *e, int new);