ip: Path MTU

Type: feature

Support setting the MTU for a peer on an interface. The minimum value of
the path and interface MTU is used at forwarding time.

the path MTU is specified for a given peer, by address and table-ID.
In the forwarding plane the MTU is enfored either:
 1 - if the peer is attached, then the MTU is set on the peer's
adjacency
 2 - if the peer is not attached, it is remote, then a DPO is added to
the peer's FIB entry to perform the necessary fragmentation.

Signed-off-by: Neale Ranns <neale@graphiant.com>
Change-Id: I8b9ea6a07868b50e97e2561f18d9335407dea7ae

8 files changed, 63 insertions, 1 deletions

diff --git a/src/vnet/adj/adj.c b/src/vnet/adj/adj.c
index d3890223dc1..8808294f7a6 100644
--- a/src/vnet/adj/adj.c
+++ b/src/vnet/adj/adj.c
@@ -20,6 +20,7 @@
 #include <vnet/adj/adj_mcast.h>
 #include <vnet/adj/adj_delegate.h>
 #include <vnet/fib/fib_node_list.h>
+#include <vnet/fib/fib_walk.h>
 
 /* Adjacency packet/byte counters indexed by adjacency index. */
 vlib_combined_counter_main_t adjacency_counters = {
@@ -326,6 +327,16 @@ adj_dpo_get_urpf (const dpo_id_t *dpo)
     return (adj->rewrite_header.sw_if_index);
 }
 
+u16
+adj_dpo_get_mtu (const dpo_id_t *dpo)
+{
+    ip_adjacency_t *adj;
+
+    adj = adj_get(dpo->dpoi_index);
+
+    return (adj->rewrite_header.max_l3_packet_bytes);
+}
+
 void
 adj_lock (adj_index_t adj_index)
 {
@@ -465,6 +476,19 @@ adj_mtu_update_walk_cb (adj_index_t ai,
 
     vnet_rewrite_update_mtu (vnet_get_main(), adj->ia_link,
                              &adj->rewrite_header);
+    adj_delegate_adj_modified(adj);
+
+    /**
+     * Backwalk to all Path MTU trackers, casual like ..
+     */
+    {
+	fib_node_back_walk_ctx_t bw_ctx = {
+	    .fnbw_reason = FIB_NODE_BW_REASON_FLAG_ADJ_MTU,
+	};
+
+	fib_walk_async(FIB_NODE_TYPE_ADJ, ai,
+                       FIB_WALK_PRIORITY_LOW, &bw_ctx);
+    }
 
     return (ADJ_WALK_RC_CONTINUE);
 }
diff --git a/src/vnet/adj/adj.h b/src/vnet/adj/adj.h
index 44bb2bd981b..c1922c755ec 100644
--- a/src/vnet/adj/adj.h
+++ b/src/vnet/adj/adj.h
@@ -373,6 +373,9 @@ STATIC_ASSERT ((STRUCT_OFFSET_OF (ip_adjacency_t, cacheline3) ==
 /* An adj fits into 4 cachelines on your average machine */
 STATIC_ASSERT_SIZEOF (ip_adjacency_t, 4 * 64);
 #endif
+STATIC_ASSERT ((STRUCT_OFFSET_OF (ip_adjacency_t, sub_type.nbr.next_hop) ==
+                STRUCT_OFFSET_OF (ip_adjacency_t, sub_type.midchain.next_hop)),
+	       "IP adjacency nbr and midchain offsets don't match");
 
 /**
  * @brief
diff --git a/src/vnet/adj/adj_glean.c b/src/vnet/adj/adj_glean.c
index c52e3d09693..e956318a1ff 100644
--- a/src/vnet/adj/adj_glean.c
+++ b/src/vnet/adj/adj_glean.c
@@ -467,6 +467,7 @@ const static dpo_vft_t adj_glean_dpo_vft = {
     .dv_unlock = adj_dpo_unlock,
     .dv_format = format_adj_glean,
     .dv_get_urpf = adj_dpo_get_urpf,
+    .dv_get_mtu = adj_dpo_get_mtu,
 };
 
 /**
diff --git a/src/vnet/adj/adj_internal.h b/src/vnet/adj/adj_internal.h
index 6639d32267f..253c1e982c1 100644
--- a/src/vnet/adj/adj_internal.h
+++ b/src/vnet/adj/adj_internal.h
@@ -126,6 +126,7 @@ extern void adj_mcast_remove(fib_protocol_t proto,
 extern void adj_midchain_teardown(ip_adjacency_t *adj);
 
 extern u32 adj_dpo_get_urpf(const dpo_id_t *dpo);
+extern u16 adj_dpo_get_mtu(const dpo_id_t *dpo);
 
 /*
  * Adj BFD
diff --git a/src/vnet/adj/adj_mcast.c b/src/vnet/adj/adj_mcast.c
index 590652244e6..a20f61f6f6b 100644
--- a/src/vnet/adj/adj_mcast.c
+++ b/src/vnet/adj/adj_mcast.c
@@ -388,12 +388,14 @@ const static dpo_vft_t adj_mcast_dpo_vft = {
     .dv_unlock = adj_dpo_unlock,
     .dv_format = format_adj_mcast,
     .dv_get_urpf = adj_dpo_get_urpf,
+    .dv_get_mtu = adj_dpo_get_mtu,
 };
 const static dpo_vft_t adj_mcast_midchain_dpo_vft = {
     .dv_lock = adj_dpo_lock,
     .dv_unlock = adj_dpo_unlock,
     .dv_format = format_adj_mcast_midchain,
     .dv_get_urpf = adj_dpo_get_urpf,
+    .dv_get_mtu = adj_dpo_get_mtu,
 };
 
 /**
diff --git a/src/vnet/adj/adj_midchain.c b/src/vnet/adj/adj_midchain.c
index a21cd21ea25..3d879e9d7fc 100644
--- a/src/vnet/adj/adj_midchain.c
+++ b/src/vnet/adj/adj_midchain.c
@@ -744,6 +744,7 @@ const static dpo_vft_t adj_midchain_dpo_vft = {
     .dv_unlock = adj_dpo_unlock,
     .dv_format = format_adj_midchain,
     .dv_get_urpf = adj_dpo_get_urpf,
+    .dv_get_mtu = adj_dpo_get_mtu,
 };
 
 /**
diff --git a/src/vnet/adj/adj_nbr.c b/src/vnet/adj/adj_nbr.c
index 921588a7ef7..811d0b8faa2 100644
--- a/src/vnet/adj/adj_nbr.c
+++ b/src/vnet/adj/adj_nbr.c
@@ -222,6 +222,27 @@ adj_nbr_alloc (fib_protocol_t nh_proto,
     return (adj);
 }
 
+void
+adj_nbr_set_mtu (adj_index_t adj_index, u16 mtu)
+{
+    ip_adjacency_t *adj;
+
+    ASSERT(ADJ_INDEX_INVALID != adj_index);
+
+    adj = adj_get(adj_index);
+
+    if (0 == mtu)
+        vnet_rewrite_update_mtu(vnet_get_main(), adj->ia_link,
+                                &adj->rewrite_header);
+    else
+    {
+        vnet_rewrite_update_mtu(vnet_get_main(), adj->ia_link,
+                                &adj->rewrite_header);
+        adj->rewrite_header.max_l3_packet_bytes =
+            clib_min (adj->rewrite_header.max_l3_packet_bytes, mtu);
+    }
+}
+
 /*
  * adj_nbr_add_or_lock
  *
@@ -268,13 +289,13 @@ adj_nbr_add_or_lock (fib_protocol_t nh_proto,
 	 * So ask the interface to do it.
 	 */
 	vnet_update_adjacency_for_sw_interface(vnm, sw_if_index, adj_index);
+        adj_delegate_adj_created(adj_get(adj_index));
     }
     else
     {
 	adj_lock(adj_index);
     }
 
-    adj_delegate_adj_created(adj_get(adj_index));
     return (adj_index);
 }
 
@@ -1055,12 +1076,14 @@ const static dpo_vft_t adj_nbr_dpo_vft = {
     .dv_format = format_adj_nbr,
     .dv_mem_show = adj_mem_show,
     .dv_get_urpf = adj_dpo_get_urpf,
+    .dv_get_mtu = adj_dpo_get_mtu,
 };
 const static dpo_vft_t adj_nbr_incompl_dpo_vft = {
     .dv_lock = adj_dpo_lock,
     .dv_unlock = adj_dpo_unlock,
     .dv_format = format_adj_nbr_incomplete,
     .dv_get_urpf = adj_dpo_get_urpf,
+    .dv_get_mtu = adj_dpo_get_mtu,
 };
 
 /**
diff --git a/src/vnet/adj/adj_nbr.h b/src/vnet/adj/adj_nbr.h
index 3a89dc89a22..4874e73a45c 100644
--- a/src/vnet/adj/adj_nbr.h
+++ b/src/vnet/adj/adj_nbr.h
@@ -75,6 +75,13 @@ extern adj_index_t adj_nbr_add_or_lock_w_rewrite(fib_protocol_t nh_proto,
 						 const ip46_address_t *nh_addr,
 						 u32 sw_if_index,
 						 u8 *rewrite);
+
+/**
+ * Set the MTU on an adjacency
+ *
+ */
+extern void adj_nbr_set_mtu(adj_index_t ai, u16 mtu);
+
 /**
  * @brief When adding a rewrite to an adjacency these are flags that
  * apply to that rewrite