Fix SR multicast post mfib commit

 1 - use the SR policy to construct the replicate DPO. Each bucket therein is a SR tunnel.
 2 - install a special mfib entry that links via this replicate
 3 - forwarding is now mfib-lookup -> replicate -> sr_rewrite (per-tunnel)
        no need for a separate sr_replicate node.
 4 - Stack the sr tunnel on the forwarding DPO of the first-hop FIB entry.
        no need for a second lookup post SR encap.
 5 - fix some path-list lock leaks in the MFIB entry.

Change-Id: I20de96ea4c4be4fae252625bde159d9c435c8315
Signed-off-by: Neale Ranns <nranns@cisco.com>

6 files changed, 210 insertions, 22 deletions

diff --git a/src/vnet/mfib/mfib_entry.c b/src/vnet/mfib/mfib_entry.c
index 479ce5f1..5170080c 100644
--- a/src/vnet/mfib/mfib_entry.c
+++ b/src/vnet/mfib/mfib_entry.c
@@ -292,6 +292,7 @@ mfib_entry_src_flush (mfib_entry_src_t *msrc)
     ({
         mfib_itf_delete(mfib_itf_get(mfii));
     }));
+    fib_path_list_unlock(msrc->mfes_pl);
 }
 
 static void
@@ -474,37 +475,60 @@ mfib_entry_src_collect_forwarding (fib_node_index_t pl_index,
 static void
 mfib_entry_stack (mfib_entry_t *mfib_entry)
 {
-    mfib_entry_collect_forwarding_ctx_t ctx = {
-        .next_hops = NULL,
-        .fct = mfib_entry_get_default_chain_type(mfib_entry),
-    };
     dpo_proto_t dp;
 
     dp = fib_proto_to_dpo(mfib_entry_get_proto(mfib_entry));
 
     if (FIB_NODE_INDEX_INVALID != mfib_entry->mfe_parent)
     {
+        mfib_entry_collect_forwarding_ctx_t ctx = {
+            .next_hops = NULL,
+            .fct = mfib_entry_get_default_chain_type(mfib_entry),
+        };
+
         fib_path_list_walk(mfib_entry->mfe_parent,
                            mfib_entry_src_collect_forwarding,
                            &ctx);
 
-        if (!dpo_id_is_valid(&mfib_entry->mfe_rep) ||
-            dpo_is_drop(&mfib_entry->mfe_rep))
+        if (!(MFIB_ENTRY_FLAG_EXCLUSIVE & mfib_entry->mfe_flags))
         {
-            dpo_id_t tmp_dpo = DPO_INVALID;
+            /*
+             * each path contirbutes a next-hop. form a replicate
+             * from those choices.
+             */
+            if (!dpo_id_is_valid(&mfib_entry->mfe_rep) ||
+                dpo_is_drop(&mfib_entry->mfe_rep))
+            {
+                dpo_id_t tmp_dpo = DPO_INVALID;
 
-            dpo_set(&tmp_dpo,
-                    DPO_REPLICATE, dp,
-                    replicate_create(0, dp));
+                dpo_set(&tmp_dpo,
+                        DPO_REPLICATE, dp,
+                        replicate_create(0, dp));
+
+                dpo_stack(DPO_MFIB_ENTRY, dp,
+                          &mfib_entry->mfe_rep,
+                          &tmp_dpo);
+
+                dpo_reset(&tmp_dpo);
+            }
+            replicate_multipath_update(&mfib_entry->mfe_rep,
+                                       ctx.next_hops);
+        }
+        else
+        {
+            /*
+             * for exclusive routes the source provided a replicate DPO
+             * we we stashed inthe special path list with one path
+             * so we can stack directly on that.
+             */
+            ASSERT(1 == vec_len(ctx.next_hops));
 
             dpo_stack(DPO_MFIB_ENTRY, dp,
                       &mfib_entry->mfe_rep,
-                      &tmp_dpo);
-
-            dpo_reset(&tmp_dpo);
+                      &ctx.next_hops[0].path_dpo);
+            dpo_reset(&ctx.next_hops[0].path_dpo);
+            vec_free(ctx.next_hops);
         }
-        replicate_multipath_update(&mfib_entry->mfe_rep,
-                                   ctx.next_hops);
     }
     else
     {
@@ -521,6 +545,8 @@ mfib_entry_forwarding_path_add (mfib_entry_src_t *msrc,
     fib_node_index_t old_pl_index;
     fib_route_path_t *rpaths;
 
+    ASSERT(!(MFIB_ENTRY_FLAG_EXCLUSIVE & msrc->mfes_flags));
+
     /*
      * path-lists require a vector of paths
      */
@@ -555,6 +581,8 @@ mfib_entry_forwarding_path_remove (mfib_entry_src_t *msrc,
     fib_node_index_t old_pl_index;
     fib_route_path_t *rpaths;
 
+    ASSERT(!(MFIB_ENTRY_FLAG_EXCLUSIVE & msrc->mfes_flags));
+
     /*
      * path-lists require a vector of paths
      */
@@ -650,7 +678,8 @@ mfib_entry_src_ok_for_delete (const mfib_entry_src_t *msrc)
 int
 mfib_entry_update (fib_node_index_t mfib_entry_index,
                    mfib_source_t source,
-                   mfib_entry_flags_t entry_flags)
+                   mfib_entry_flags_t entry_flags,
+                   index_t repi)
 {
     mfib_entry_t *mfib_entry;
     mfib_entry_src_t *msrc;
@@ -659,6 +688,35 @@ mfib_entry_update (fib_node_index_t mfib_entry_index,
     msrc = mfib_entry_src_find_or_create(mfib_entry, source);
     msrc->mfes_flags = entry_flags;
 
+    if (INDEX_INVALID != repi)
+    {
+        /*
+         * The source is providing its own replicate DPO.
+         * Create a sepcial path-list to manage it, that way
+         * this entry and the source are equivalent to a normal
+         * entry
+         */
+        fib_node_index_t old_pl_index;
+        fib_protocol_t fp;
+        dpo_id_t dpo = DPO_INVALID;
+
+        fp = mfib_entry_get_proto(mfib_entry);
+        old_pl_index = msrc->mfes_pl;
+
+        dpo_set(&dpo, DPO_REPLICATE,
+                fib_proto_to_dpo(fp),
+                repi);
+
+        msrc->mfes_pl =
+            fib_path_list_create_special(fp,
+                                         FIB_PATH_LIST_FLAG_EXCLUSIVE,
+                                         &dpo);
+
+        dpo_reset(&dpo);
+        fib_path_list_lock(msrc->mfes_pl);
+        fib_path_list_unlock(old_pl_index);
+    }
+
     if (mfib_entry_src_ok_for_delete(msrc))
     {
         /*
diff --git a/src/vnet/mfib/mfib_entry.h b/src/vnet/mfib/mfib_entry.h
index cc5d5326..36fc73e1 100644
--- a/src/vnet/mfib/mfib_entry.h
+++ b/src/vnet/mfib/mfib_entry.h
@@ -65,7 +65,7 @@ typedef struct mfib_entry_t_ {
     CLIB_CACHE_LINE_ALIGN_MARK(cacheline1);
 
     /**
-     * The Replicate used for forwarding.
+     * The Replicate DPO used for forwarding.
      */
     dpo_id_t mfe_rep;
 
@@ -94,7 +94,8 @@ extern fib_node_index_t mfib_entry_create(u32 fib_index,
 
 extern int mfib_entry_update(fib_node_index_t fib_entry_index,
                              mfib_source_t source,
-                             mfib_entry_flags_t entry_flags);
+                             mfib_entry_flags_t entry_flags,
+                             index_t rep_dpo);
 
 extern void mfib_entry_path_update(fib_node_index_t fib_entry_index,
                                    mfib_source_t source,
diff --git a/src/vnet/mfib/mfib_table.c b/src/vnet/mfib/mfib_table.c
index e4c0936d..b4e855ff 100644
--- a/src/vnet/mfib/mfib_table.c
+++ b/src/vnet/mfib/mfib_table.c
@@ -195,7 +195,10 @@ mfib_table_entry_update (u32 fib_index,
     {
         mfib_entry_lock(mfib_entry_index);
 
-        if (mfib_entry_update(mfib_entry_index, source, entry_flags))
+        if (mfib_entry_update(mfib_entry_index,
+                              source,
+                              entry_flags,
+                              INDEX_INVALID))
         {
             /*
              * this update means we can now remove the entry.
@@ -283,6 +286,36 @@ mfib_table_entry_path_remove (u32 fib_index,
     }
 }
 
+fib_node_index_t
+mfib_table_entry_special_add (u32 fib_index,
+                              const mfib_prefix_t *prefix,
+                              mfib_source_t source,
+                              mfib_entry_flags_t entry_flags,
+                              index_t rep_dpo)
+{
+    fib_node_index_t mfib_entry_index;
+    mfib_table_t *mfib_table;
+
+    mfib_table = mfib_table_get(fib_index, prefix->fp_proto);
+    mfib_entry_index = mfib_table_lookup_exact_match_i(mfib_table, prefix);
+
+    if (FIB_NODE_INDEX_INVALID == mfib_entry_index)
+    {
+        mfib_entry_index = mfib_entry_create(fib_index,
+                                             source,
+                                             prefix,
+                                             MFIB_ENTRY_FLAG_NONE);
+
+        mfib_table_entry_insert(mfib_table, prefix, mfib_entry_index);
+    }
+
+    mfib_entry_update(mfib_entry_index, source,
+                      (MFIB_ENTRY_FLAG_EXCLUSIVE | entry_flags),
+                      rep_dpo);
+
+    return (mfib_entry_index);
+}
+
 static void
 mfib_table_entry_delete_i (u32 fib_index,
                            fib_node_index_t mfib_entry_index,
diff --git a/src/vnet/mfib/mfib_table.h b/src/vnet/mfib/mfib_table.h
index 4faa69ee..4c51b70f 100644
--- a/src/vnet/mfib/mfib_table.h
+++ b/src/vnet/mfib/mfib_table.h
@@ -18,6 +18,7 @@
 
 #include <vnet/ip/ip.h>
 #include <vnet/adj/adj.h>
+#include <vnet/dpo/replicate_dpo.h>
 
 #include <vnet/mfib/mfib_types.h>
 
@@ -214,6 +215,37 @@ extern void mfib_table_entry_delete_index(fib_node_index_t entry_index,
 
 /**
  * @brief
+ *  Add a 'special' entry to the mFIB that links to the DPO passed
+ *  A special entry is an entry that the FIB is not expect to resolve
+ *  via the usual mechanisms (i.e. recurisve or neighbour adj DB lookup).
+ *  Instead the client/source provides the index of a replicate DPO to link to.
+ *
+  * @param fib_index
+ *  The index of the FIB
+ *
+ * @param prefix
+ *  The prefix to add
+ *
+ * @param source
+ *  The ID of the client/source adding the entry.
+ *
+ * @param flags
+ *  Flags for the entry.
+ *
+ * @param rep_dpo
+ *  The replicate DPO index to link to.
+ *
+ * @return
+ *  the index of the fib_entry_t that is created (or existed already).
+ */
+extern fib_node_index_t mfib_table_entry_special_add(u32 fib_index,
+                                                     const mfib_prefix_t *prefix,
+                                                     mfib_source_t source,
+                                                     mfib_entry_flags_t flags,
+                                                     index_t rep_dpo);
+
+/**
+ * @brief
  *  Flush all entries from a table for the source
  *
  * @param fib_index
diff --git a/src/vnet/mfib/mfib_test.c b/src/vnet/mfib/mfib_test.c
index 8082a6bb..36a303e8 100644
--- a/src/vnet/mfib/mfib_test.c
+++ b/src/vnet/mfib/mfib_test.c
@@ -19,6 +19,7 @@
 #include <vnet/mfib/mfib_entry.h>
 #include <vnet/mfib/mfib_signal.h>
 #include <vnet/mfib/ip6_mfib.h>
+#include <vnet/fib/fib_path_list.h>
 
 #include <vnet/dpo/replicate_dpo.h>
 #include <vnet/adj/adj_mcast.h>
@@ -337,7 +338,7 @@ mfib_test_i (fib_protocol_t PROTO,
              const mfib_prefix_t *pfx_star_g_slash_m)
 {
     fib_node_index_t mfei, mfei_dflt, mfei_no_f, mfei_s_g, mfei_g_1, mfei_g_2, mfei_g_3, mfei_g_m;
-    u32 fib_index, n_entries, n_itfs, n_reps;
+    u32 fib_index, n_entries, n_itfs, n_reps, n_pls;
     fib_node_index_t ai_1, ai_2, ai_3;
     test_main_t *tm;
 
@@ -347,6 +348,7 @@ mfib_test_i (fib_protocol_t PROTO,
     n_entries = pool_elts(mfib_entry_pool);
     n_itfs = pool_elts(mfib_itf_pool);
     n_reps = pool_elts(replicate_pool);
+    n_pls = fib_path_list_pool_size();
     tm = &test_main;
 
     ai_1 = adj_mcast_add_or_lock(PROTO,
@@ -1024,6 +1026,54 @@ mfib_test_i (fib_protocol_t PROTO,
               format_mfib_prefix, pfx_star_g_slash_m);
 
     /*
+     * Add a prefix as a special/exclusive route
+     */
+    dpo_id_t td = DPO_INVALID;
+    index_t repi = replicate_create(1, fib_proto_to_dpo(PROTO));
+
+    dpo_set(&td, DPO_ADJACENCY_MCAST, fib_proto_to_dpo(PROTO), ai_2);
+    replicate_set_bucket(repi, 0, &td);
+
+    mfei = mfib_table_entry_special_add(fib_index,
+                                        pfx_star_g_3,
+                                        MFIB_SOURCE_SRv6,
+                                        MFIB_ENTRY_FLAG_ACCEPT_ALL_ITF,
+                                        repi);
+    MFIB_TEST(mfib_test_entry(mfei,
+                              (MFIB_ENTRY_FLAG_ACCEPT_ALL_ITF |
+                               MFIB_ENTRY_FLAG_EXCLUSIVE),
+                              1,
+                              DPO_ADJACENCY_MCAST, ai_2),
+              "%U exclusive replicate OK",
+              format_mfib_prefix, pfx_star_g_3);
+
+    /*
+     * update a special/exclusive route
+     */
+    index_t repi2 = replicate_create(1, fib_proto_to_dpo(PROTO));
+
+    dpo_set(&td, DPO_ADJACENCY_MCAST, fib_proto_to_dpo(PROTO), ai_1);
+    replicate_set_bucket(repi2, 0, &td);
+
+    mfei = mfib_table_entry_special_add(fib_index,
+                                        pfx_star_g_3,
+                                        MFIB_SOURCE_SRv6,
+                                        MFIB_ENTRY_FLAG_ACCEPT_ALL_ITF,
+                                        repi2);
+    MFIB_TEST(mfib_test_entry(mfei,
+                              (MFIB_ENTRY_FLAG_ACCEPT_ALL_ITF |
+                               MFIB_ENTRY_FLAG_EXCLUSIVE),
+                              1,
+                              DPO_ADJACENCY_MCAST, ai_1),
+              "%U exclusive update replicate OK",
+              format_mfib_prefix, pfx_star_g_3);
+
+    mfib_table_entry_delete(fib_index,
+                            pfx_star_g_3,
+                            MFIB_SOURCE_SRv6);
+    dpo_reset(&td);
+
+    /*
      * Unlock the table - it's the last lock so should be gone thereafter
      */
     mfib_table_unlock(fib_index, PROTO);
@@ -1040,6 +1090,8 @@ mfib_test_i (fib_protocol_t PROTO,
      * test we've leaked no resources
      */
     MFIB_TEST(0 == adj_mcast_db_size(), "%d MCAST adjs", adj_mcast_db_size());
+    MFIB_TEST(n_pls == fib_path_list_pool_size(), "%d=%d path-lists",
+              n_pls, fib_path_list_pool_size());
     MFIB_TEST(n_reps == pool_elts(replicate_pool), "%d=%d replicates",
               n_reps, pool_elts(replicate_pool));
     MFIB_TEST(n_entries == pool_elts(mfib_entry_pool),
@@ -1214,7 +1266,7 @@ mfib_test (vlib_main_t * vm,
 
 VLIB_CLI_COMMAND (test_fib_command, static) = {
     .path = "test mfib",
-    .short_help = "fib unit tests - DO NOT RUN ON A LIVE SYSTEM",
+    .short_help = "mfib unit tests - DO NOT RUN ON A LIVE SYSTEM",
     .function = mfib_test,
 };
 
diff --git a/src/vnet/mfib/mfib_types.h b/src/vnet/mfib/mfib_types.h
index 37898a07..fe53aa68 100644
--- a/src/vnet/mfib/mfib_types.h
+++ b/src/vnet/mfib/mfib_types.h
@@ -68,6 +68,13 @@ typedef enum mfib_entry_attribute_t_
      *        Use with extreme caution
      */
     MFIB_ENTRY_ACCEPT_ALL_ITF,
+    /**
+     * Exclusive - like its unicast counterpart. the source has provided
+     * the forwarding DPO directly. The entry therefore does not resolve
+     * paths via a path-list
+     */
+    MFIB_ENTRY_EXCLUSIVE,
+
     MFIB_ENTRY_INHERIT_ACCEPT,
     MFIB_ENTRY_ATTRIBUTE_LAST = MFIB_ENTRY_INHERIT_ACCEPT,
 } mfib_entry_attribute_t;
@@ -83,6 +90,7 @@ typedef enum mfib_entry_attribute_t_
     [MFIB_ENTRY_DROP]           = "D",     \
     [MFIB_ENTRY_ACCEPT_ALL_ITF] = "AA",    \
     [MFIB_ENTRY_INHERIT_ACCEPT] = "IA",    \
+    [MFIB_ENTRY_EXCLUSIVE]      = "E",     \
 }
 
 #define MFIB_ENTRY_NAMES_LONG  {                    \
@@ -91,6 +99,7 @@ typedef enum mfib_entry_attribute_t_
     [MFIB_ENTRY_DROP]           = "Drop",           \
     [MFIB_ENTRY_ACCEPT_ALL_ITF] = "Accept-all-itf", \
     [MFIB_ENTRY_INHERIT_ACCEPT] = "Inherit-Accept", \
+    [MFIB_ENTRY_EXCLUSIVE]      = "Exclusive",      \
 }
 
 typedef enum mfib_entry_flags_t_
@@ -99,8 +108,9 @@ typedef enum mfib_entry_flags_t_
     MFIB_ENTRY_FLAG_SIGNAL = (1 << MFIB_ENTRY_SIGNAL),
     MFIB_ENTRY_FLAG_DROP = (1 << MFIB_ENTRY_DROP),
     MFIB_ENTRY_FLAG_CONNECTED = (1 << MFIB_ENTRY_CONNECTED),
-    MFIB_ENTRY_FLAG_INHERIT_ACCEPT = (1 << MFIB_ENTRY_INHERIT_ACCEPT),
     MFIB_ENTRY_FLAG_ACCEPT_ALL_ITF = (1 << MFIB_ENTRY_ACCEPT_ALL_ITF),
+    MFIB_ENTRY_FLAG_EXCLUSIVE = (1 << MFIB_ENTRY_EXCLUSIVE),
+    MFIB_ENTRY_FLAG_INHERIT_ACCEPT = (1 << MFIB_ENTRY_INHERIT_ACCEPT),
 } mfib_entry_flags_t;
 
 typedef enum mfib_itf_attribute_t_
@@ -155,6 +165,7 @@ typedef enum mfib_source_t_
     MFIB_SOURCE_CLI,
     MFIB_SOURCE_VXLAN,
     MFIB_SOURCE_DHCP,
+    MFIB_SOURCE_SRv6,
     MFIB_SOURCE_DEFAULT_ROUTE,
 } mfib_source_t;
 
@@ -164,6 +175,7 @@ typedef enum mfib_source_t_
     [MFIB_SOURCE_CLI] = "CLI",                     \
     [MFIB_SOURCE_DHCP] = "DHCP",                   \
     [MFIB_SOURCE_VXLAN] = "VXLAN",                 \
+    [MFIB_SOURCE_SRv6] = "SRv6",                   \
     [MFIB_SOURCE_DEFAULT_ROUTE] = "Default Route", \
 }