FIB Interpose Source

The interpose source allows the source/provider to insert/interpose
a DPO in the forwarding chain of the FIB entry ahead of the forwarding
provided by the next best source. For example if the API source (i.e
the 'control plane') has provided an adjacency for forwarding, then
an interpose source (e.g. a monitoring service) couold interpose a
replicatte DPO to copy the traffic to another location AND forward
using the API's adjacency.
To use the interose feature an existing source (i.e FIB_SOURCE_PLUGIN_HI)
cn specifiy as a flag FIB_ENTRY_FLAG_INTERPOSE and provide a DPO to
interpose. One might also consider using interpose in conjunction with
FIB_ENTRY_FLAG_COVER_INHERIT to ensure the interpose object affects
all prefixes in the sub-tree.

Change-Id: I8b2737b985f8f7c08123406d0491881def347b52
Signed-off-by: Neale Ranns <nranns@cisco.com>

5 files changed, 114 insertions, 26 deletions

diff --git a/src/vnet/dpo/dpo.c b/src/vnet/dpo/dpo.c
index 85f2c5d3622..19e3714c536 100644
--- a/src/vnet/dpo/dpo.c
+++ b/src/vnet/dpo/dpo.c
@@ -153,20 +153,22 @@ format_dpo_id (u8 * s, va_list * args)
 
     if (NULL != dpo_vfts[dpo->dpoi_type].dv_format)
     {
-        return (format(s, "%U",
-                       dpo_vfts[dpo->dpoi_type].dv_format,
-                       dpo->dpoi_index,
-                       indent));
+        s = format(s, "%U",
+                   dpo_vfts[dpo->dpoi_type].dv_format,
+                   dpo->dpoi_index,
+                   indent);
     }
-
-    switch (dpo->dpoi_type)
+    else
     {
-    case DPO_FIRST:
-	s = format(s, "unset");
-	break;
-    default:
-	s = format(s, "unknown");
-	break;
+        switch (dpo->dpoi_type)
+        {
+        case DPO_FIRST:
+            s = format(s, "unset");
+            break;
+        default:
+            s = format(s, "unknown");
+            break;
+        }
     }
     return (s);
 }
@@ -303,6 +305,18 @@ dpo_default_get_next_node (const dpo_id_t *dpo)
     return (node_indices);
 }
 
+/**
+ * A default variant of the make interpose function that just returns
+ * the original
+ */
+static void
+dpo_default_mk_interpose (const dpo_id_t *original,
+                          const dpo_id_t *parent,
+                          dpo_id_t *clone)
+{
+    dpo_copy(clone, original);
+}
+
 void
 dpo_register (dpo_type_t type,
 	      const dpo_vft_t *vft,
@@ -314,6 +328,10 @@ dpo_register (dpo_type_t type,
     {
         dpo_vfts[type].dv_get_next_node = dpo_default_get_next_node;
     }
+    if (NULL == dpo_vfts[type].dv_mk_interpose)
+    {
+        dpo_vfts[type].dv_mk_interpose = dpo_default_mk_interpose;
+    }
 
     vec_validate(dpo_nodes, type);
     dpo_nodes[type] = nodes;
@@ -331,6 +349,17 @@ dpo_register_new_type (const dpo_vft_t *vft,
 }
 
 void
+dpo_mk_interpose (const dpo_id_t *original,
+                  const dpo_id_t *parent,
+                  dpo_id_t *clone)
+{
+    if (!dpo_id_is_valid(original))
+	return;
+
+    dpo_vfts[original->dpoi_type].dv_mk_interpose(original, parent, clone);
+}
+
+void
 dpo_lock (dpo_id_t *dpo)
 {
     if (!dpo_id_is_valid(dpo))
diff --git a/src/vnet/dpo/dpo.h b/src/vnet/dpo/dpo.h
index 21a2ae2a33d..0eeca67b74b 100644
--- a/src/vnet/dpo/dpo.h
+++ b/src/vnet/dpo/dpo.h
@@ -225,6 +225,14 @@ extern void dpo_lock(dpo_id_t *dpo);
 extern void dpo_unlock(dpo_id_t *dpo);
 
 /**
+ * @brief
+ *  Make an interpose DPO from an original
+ */
+extern void dpo_mk_interpose(const dpo_id_t *original,
+                             const dpo_id_t *parent,
+                             dpo_id_t *clone);
+
+/**
  * @brief Set/create a DPO ID
  * The DPO will be locked.
  *
@@ -374,6 +382,18 @@ typedef u32* (*dpo_get_next_node_t)(const dpo_id_t *dpo);
 typedef u32 (*dpo_get_urpf_t)(const dpo_id_t *dpo);
 
 /**
+ * @brief Called during FIB interposition when the originally
+ * registered DPO is used to 'clone' an instance for interposition
+ * at a particular location in the FIB graph.
+ * The parent is the next DPO in the chain that the clone will
+ * be used instead of. The clone may then choose to stack itself
+ * on the parent.
+ */
+typedef void (*dpo_mk_interpose_t)(const dpo_id_t *original,
+                                   const dpo_id_t *parent,
+                                   dpo_id_t *clone);
+
+/**
  * @brief A virtual function table regisitered for a DPO type
  */
 typedef struct dpo_vft_t_
@@ -405,6 +425,10 @@ typedef struct dpo_vft_t_
      * Get uRPF interface
      */
     dpo_get_urpf_t dv_get_urpf;
+    /**
+     * Signal on an interposed child that the parent has changed
+     */
+    dpo_mk_interpose_t dv_mk_interpose;
 } dpo_vft_t;
 
 
diff --git a/src/vnet/dpo/load_balance.c b/src/vnet/dpo/load_balance.c
index db0ebcdceb8..bb38233a55e 100644
--- a/src/vnet/dpo/load_balance.c
+++ b/src/vnet/dpo/load_balance.c
@@ -198,16 +198,6 @@ load_balance_create (u32 n_buckets,
     return (load_balance_get_index(load_balance_create_i(n_buckets, lb_proto, fhc)));
 }
 
-u16
-load_balance_n_buckets (index_t lbi)
-{
-    load_balance_t *lb;
-
-    lb = load_balance_get(lbi);
-
-    return (lb->lb_n_buckets);
-}
-
 static inline void
 load_balance_set_bucket_i (load_balance_t *lb,
                            u32 bucket,
@@ -250,6 +240,16 @@ load_balance_is_drop (const dpo_id_t *dpo)
     return (0);
 }
 
+u16
+load_balance_n_buckets (index_t lbi)
+{
+    load_balance_t *lb;
+
+    lb = load_balance_get(lbi);
+
+    return (lb->lb_n_buckets);
+}
+
 void
 load_balance_set_fib_entry_flags (index_t lbi,
                                   fib_entry_flag_t flags)
diff --git a/src/vnet/dpo/mpls_label_dpo.c b/src/vnet/dpo/mpls_label_dpo.c
index ebbbbec9b64..0a7063cc99a 100644
--- a/src/vnet/dpo/mpls_label_dpo.c
+++ b/src/vnet/dpo/mpls_label_dpo.c
@@ -1204,11 +1204,44 @@ mpls_label_dpo_mem_show (void)
 			  sizeof(mpls_label_dpo_t));
 }
 
+/**
+ * Interpose a label DPO. used in the FIB unit tests
+ */
+static void
+mpls_label_interpose (const dpo_id_t *original,
+                      const dpo_id_t *parent,
+                      dpo_id_t *clone)
+{
+    mpls_label_dpo_t *mld, *mld_clone;
+
+    mld_clone = mpls_label_dpo_alloc();
+    mld = mpls_label_dpo_get(original->dpoi_index);
+
+    mld_clone->mld_locks = 0;
+    clib_memcpy(&mld_clone->mld_hdr,
+                &mld->mld_hdr,
+                sizeof(mld_clone->mld_hdr));
+    mld_clone->mld_payload_proto = mld->mld_payload_proto;
+    mld_clone->mld_n_labels = mld->mld_n_labels;
+    mld_clone->mld_n_hdr_bytes = mld->mld_n_hdr_bytes;
+
+    dpo_stack(mpls_label_dpo_types[MPLS_LABEL_DPO_FLAG_NONE],
+              mld_clone->mld_payload_proto,
+              &mld_clone->mld_dpo,
+              parent);
+
+    dpo_set(clone,
+            mpls_label_dpo_types[MPLS_LABEL_DPO_FLAG_NONE],
+            mld_clone->mld_payload_proto,
+            mpls_label_dpo_get_index(mld_clone));
+}
+
 const static dpo_vft_t mld_vft = {
     .dv_lock = mpls_label_dpo_lock,
     .dv_unlock = mpls_label_dpo_unlock,
     .dv_format = format_mpls_label_dpo,
     .dv_mem_show = mpls_label_dpo_mem_show,
+    .dv_mk_interpose = mpls_label_interpose,
 };
 
 const static char* const mpls_label_imp_pipe_ip4_nodes[] =
diff --git a/src/vnet/dpo/replicate_dpo.h b/src/vnet/dpo/replicate_dpo.h
index 7383184a2ea..ccb25630e16 100644
--- a/src/vnet/dpo/replicate_dpo.h
+++ b/src/vnet/dpo/replicate_dpo.h
@@ -53,7 +53,7 @@ extern replicate_main_t replicate_main;
  */
 typedef struct replicate_t_ {
     /**
-     * number of buckets in the load-balance. always a power of 2.
+     * number of buckets in the replicate.
      */
     u16 rep_n_buckets;
 
@@ -104,15 +104,17 @@ extern void replicate_multipath_update(
     load_balance_path_t *next_hops);
 
 extern void replicate_set_bucket(index_t repi,
-				    u32 bucket,
-				    const dpo_id_t *next);
+                                 u32 bucket,
+                                 const dpo_id_t *next);
 
 extern u8* format_replicate(u8 * s, va_list * args);
 
 extern const dpo_id_t *replicate_get_bucket(index_t repi,
-					       u32 bucket);
+                                            u32 bucket);
 extern int replicate_is_drop(const dpo_id_t *dpo);
 
+extern u16 replicate_n_buckets(index_t repi);
+
 /**
  * The encapsulation breakages are for fast DP access
  */