Dual loop decrement TTL.

Outer fragmentation.
ICMP PMTUD.
Add DF ignore knob.

Change-Id: Icfd7b5c5d9629db3b8130ba15dc6c9a5e709d23b
Signed-off-by: Ole Troan <ot@cisco.com>

3 files changed, 124 insertions, 33 deletions

diff --git a/vnet/vnet/map/ip4_map.c b/vnet/vnet/map/ip4_map.c
index 343b57d210f..7b9b3ed9d89 100644
--- a/vnet/vnet/map/ip4_map.c
+++ b/vnet/vnet/map/ip4_map.c
@@ -27,8 +27,10 @@ enum ip4_map_next_e {
 #ifdef MAP_SKIP_IP6_LOOKUP
   IP4_MAP_NEXT_IP6_REWRITE,
 #endif
-  IP4_MAP_NEXT_FRAGMENT,
+  IP4_MAP_NEXT_IP4_FRAGMENT,
+  IP4_MAP_NEXT_IP6_FRAGMENT,
   IP4_MAP_NEXT_REASS,
+  IP4_MAP_NEXT_ICMP_ERROR,
   IP4_MAP_NEXT_DROP,
   IP4_MAP_N_NEXT,
 };
@@ -177,6 +179,27 @@ ip4_map_decrement_ttl (ip4_header_t *ip, u8 *error)
   ASSERT (ip->checksum == ip4_header_checksum(ip));
 }
 
+static u32
+ip4_map_fragment (vlib_buffer_t *b, u16 mtu, bool df, u8 *error)
+{
+  map_main_t *mm = &map_main;
+
+  if (mm->frag_inner) {
+    ip_frag_set_vnet_buffer(b, sizeof(ip6_header_t), mtu, IP4_FRAG_NEXT_IP6_LOOKUP, IP_FRAG_FLAG_IP6_HEADER);
+    return (IP4_MAP_NEXT_IP4_FRAGMENT);
+  } else {
+    if (df && !mm->frag_ignore_df) {
+      icmp4_error_set_vnet_buffer(b, ICMP4_destination_unreachable,
+				  ICMP4_destination_unreachable_fragmentation_needed_and_dont_fragment_set, mtu);
+      vlib_buffer_advance(b, sizeof(ip6_header_t));
+      *error = MAP_ERROR_DF_SET;
+      return (IP4_MAP_NEXT_ICMP_ERROR);
+    }
+    ip_frag_set_vnet_buffer(b, 0, mtu, IP6_FRAG_NEXT_IP6_LOOKUP, IP_FRAG_FLAG_IP6_HEADER);
+    return (IP4_MAP_NEXT_IP6_FRAGMENT);
+  }
+}
+
 /*
  * ip4_map
  */
@@ -247,6 +270,12 @@ ip4_map (vlib_main_t *vm,
       port0 = ip4_map_port_and_security_check(d0, ip40, &next0, &error0);
       port1 = ip4_map_port_and_security_check(d1, ip41, &next1, &error1);
 
+      /* Decrement IPv4 TTL */
+      ip4_map_decrement_ttl(ip40, &error0);
+      ip4_map_decrement_ttl(ip41, &error1);
+      bool df0 = ip40->flags_and_fragment_offset & clib_host_to_net_u16(IP4_HEADER_FLAG_DONT_FRAGMENT);
+      bool df1 = ip41->flags_and_fragment_offset & clib_host_to_net_u16(IP4_HEADER_FLAG_DONT_FRAGMENT);
+
       /* MAP calc */
       u32 da40 = clib_net_to_host_u32(ip40->dst_address.as_u32);
       u32 da41 = clib_net_to_host_u32(ip41->dst_address.as_u32);
@@ -288,11 +317,7 @@ ip4_map (vlib_main_t *vm,
        */
       if (PREDICT_TRUE(error0 == MAP_ERROR_NONE)) {
 	if (PREDICT_FALSE(d0->mtu && (clib_net_to_host_u16(ip6h0->payload_length) + sizeof(*ip6h0) > d0->mtu))) {
-	  vnet_buffer(p0)->ip_frag.header_offset = sizeof(*ip6h0);
-	  vnet_buffer(p0)->ip_frag.next_index = IP4_FRAG_NEXT_IP6_LOOKUP;
-	  vnet_buffer(p0)->ip_frag.mtu = d0->mtu;
-	  vnet_buffer(p0)->ip_frag.flags = IP_FRAG_FLAG_IP6_HEADER;
-	  next0 = IP4_MAP_NEXT_FRAGMENT;
+	  next0 = ip4_map_fragment(p0, d0->mtu, df0, &error0);
 	} else {
 	  next0 = ip4_map_ip6_lookup_bypass(p0, ip40) ? IP4_MAP_NEXT_IP6_REWRITE : next0;
 	  vlib_increment_combined_counter(cm + MAP_DOMAIN_COUNTER_TX, cpu_index, map_domain_index0, 1,
@@ -308,11 +333,7 @@ ip4_map (vlib_main_t *vm,
        */
       if (PREDICT_TRUE(error1 == MAP_ERROR_NONE)) {
 	if (PREDICT_FALSE(d1->mtu && (clib_net_to_host_u16(ip6h1->payload_length) + sizeof(*ip6h1) > d1->mtu))) {
-	  vnet_buffer(p1)->ip_frag.header_offset = sizeof(*ip6h1);
-	  vnet_buffer(p1)->ip_frag.next_index = IP4_FRAG_NEXT_IP6_LOOKUP;
-	  vnet_buffer(p1)->ip_frag.mtu = d1->mtu;
-	  vnet_buffer(p1)->ip_frag.flags = IP_FRAG_FLAG_IP6_HEADER;
-	  next1 = IP4_MAP_NEXT_FRAGMENT;
+	  next1 = ip4_map_fragment(p1, d1->mtu, df1, &error1);
 	} else {
 	  next1 = ip4_map_ip6_lookup_bypass(p1, ip41) ? IP4_MAP_NEXT_IP6_REWRITE : next1;
 	  vlib_increment_combined_counter(cm + MAP_DOMAIN_COUNTER_TX, cpu_index, map_domain_index1, 1,
@@ -369,6 +390,7 @@ ip4_map (vlib_main_t *vm,
 
       /* Decrement IPv4 TTL */
       ip4_map_decrement_ttl(ip40, &error0);
+      bool df0 = ip40->flags_and_fragment_offset & clib_host_to_net_u16(IP4_HEADER_FLAG_DONT_FRAGMENT);
 
       /* MAP calc */
       u32 da40 = clib_net_to_host_u32(ip40->dst_address.as_u32);
@@ -396,11 +418,7 @@ ip4_map (vlib_main_t *vm,
        */
       if (PREDICT_TRUE(error0 == MAP_ERROR_NONE)) {
 	if (PREDICT_FALSE(d0->mtu && (clib_net_to_host_u16(ip6h0->payload_length) + sizeof(*ip6h0) > d0->mtu))) {
-	  vnet_buffer(p0)->ip_frag.header_offset = sizeof(*ip6h0);
-	  vnet_buffer(p0)->ip_frag.next_index = IP4_FRAG_NEXT_IP6_LOOKUP;
-	  vnet_buffer(p0)->ip_frag.mtu = d0->mtu;
-	  vnet_buffer(p0)->ip_frag.flags = IP_FRAG_FLAG_IP6_HEADER;
-	  next0 = IP4_MAP_NEXT_FRAGMENT;
+	  next0 = ip4_map_fragment(p0, d0->mtu, df0, &error0);
 	} else {
 	  next0 = ip4_map_ip6_lookup_bypass(p0, ip40) ? IP4_MAP_NEXT_IP6_REWRITE : next0;
 	  vlib_increment_combined_counter(cm + MAP_DOMAIN_COUNTER_TX, cpu_index, map_domain_index0, 1,
@@ -591,8 +609,10 @@ VLIB_REGISTER_NODE(ip4_map_node) = {
 #ifdef MAP_SKIP_IP6_LOOKUP
     [IP4_MAP_NEXT_IP6_REWRITE] = "ip6-rewrite",
 #endif
-    [IP4_MAP_NEXT_FRAGMENT] = "ip4-frag",
+    [IP4_MAP_NEXT_IP4_FRAGMENT] = "ip4-frag",
+    [IP4_MAP_NEXT_IP6_FRAGMENT] = "ip6-frag",
     [IP4_MAP_NEXT_REASS] = "ip4-map-reass",
+    [IP4_MAP_NEXT_ICMP_ERROR] = "ip4-icmp-error",
     [IP4_MAP_NEXT_DROP] = "error-drop",
   },
 };
diff --git a/vnet/vnet/map/map.c b/vnet/vnet/map/map.c
index eb5496b8829..a63122b58f3 100644
--- a/vnet/vnet/map/map.c
+++ b/vnet/vnet/map/map.c
@@ -653,6 +653,58 @@ map_icmp_unreachables_command_fn (vlib_main_t *vm,
 }
 
 static clib_error_t *
+map_fragment_command_fn (vlib_main_t *vm,
+			 unformat_input_t *input,
+			 vlib_cli_command_t *cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  map_main_t *mm = &map_main;
+
+  /* Get a line of input. */
+  if (!unformat_user(input, unformat_line_input, line_input))
+    return 0;
+ 
+  while (unformat_check_input(line_input) != UNFORMAT_END_OF_INPUT) {
+    if (unformat(line_input, "inner"))
+      mm->frag_inner = true;
+    else if (unformat(line_input, "outer"))
+      mm->frag_inner = false;
+    else
+      return clib_error_return(0, "unknown input `%U'",
+                               format_unformat_error, input);
+  }
+  unformat_free(line_input);
+
+  return 0;
+}
+
+static clib_error_t *
+map_fragment_df_command_fn (vlib_main_t *vm,
+			    unformat_input_t *input,
+			    vlib_cli_command_t *cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  map_main_t *mm = &map_main;
+
+  /* Get a line of input. */
+  if (!unformat_user(input, unformat_line_input, line_input))
+    return 0;
+ 
+  while (unformat_check_input(line_input) != UNFORMAT_END_OF_INPUT) {
+    if (unformat(line_input, "on"))
+      mm->frag_ignore_df = true;
+    else if (unformat(line_input, "off"))
+      mm->frag_ignore_df = false;
+    else
+      return clib_error_return(0, "unknown input `%U'",
+                               format_unformat_error, input);
+  }
+  unformat_free(line_input);
+
+  return 0;
+}
+
+static clib_error_t *
 map_traffic_class_command_fn (vlib_main_t *vm,
 			      unformat_input_t *input,
 			      vlib_cli_command_t *cmd)
@@ -869,6 +921,8 @@ show_map_stats_command_fn (vlib_main_t *vm, unformat_input_t *input, vlib_cli_co
 
   vlib_cli_output(vm, "ICMP-relay IPv4 source address: %U\n", format_ip4_address, &mm->icmp4_src_address);
   vlib_cli_output(vm, "ICMP6 unreachables sent for unmatched packets: %s\n", mm->icmp6_enabled ? "enabled" : "disabled");
+  vlib_cli_output(vm, "Inner fragmentation: %s\n", mm->frag_inner ? "enabled" : "disabled");
+  vlib_cli_output(vm, "Fragment packets regardless of DF flag: %s\n", mm->frag_ignore_df ? "enabled" : "disabled");
 
   /*
    * Counters
@@ -1563,11 +1617,23 @@ VLIB_CLI_COMMAND(map_icmp_relay_source_address_command, static) = {
 };
 
 VLIB_CLI_COMMAND(map_icmp_unreachables_command, static) = {
-  .path = "map params icmp unreachables",
+  .path = "map params icmp6 unreachables",
   .short_help = "unreachables {on|off}",
   .function = map_icmp_unreachables_command_fn,
 };
 
+VLIB_CLI_COMMAND(map_fragment_command, static) = {
+  .path = "map params fragment",
+  .short_help = "[inner|outer] [ignore-df [on|off]]",
+  .function = map_fragment_command_fn,
+};
+
+VLIB_CLI_COMMAND(map_fragment_df_command, static) = {
+  .path = "map params fragment ignore-df",
+  .short_help = "on|off",
+  .function = map_fragment_df_command_fn,
+};
+
 VLIB_CLI_COMMAND(map_security_check_frag_command, static) = {
   .path = "map params security-check fragments",
   .short_help = 
@@ -1639,6 +1705,10 @@ clib_error_t *map_init (vlib_main_t *vm)
   /* ICMP6 Type 1, Code 5 for security check failure */
   mm->icmp6_enabled = false;
 
+  /* Inner or outer fragmentation */
+  mm->frag_inner = false;
+  mm->frag_ignore_df = false;
+
   vec_validate(mm->domain_counters, MAP_N_DOMAIN_COUNTER - 1);
   mm->domain_counters[MAP_DOMAIN_COUNTER_RX].name = "rx";
   mm->domain_counters[MAP_DOMAIN_COUNTER_TX].name = "tx";
diff --git a/vnet/vnet/map/map.h b/vnet/vnet/map/map.h
index 6d12b71be2b..d38d7f47241 100644
--- a/vnet/vnet/map/map.h
+++ b/vnet/vnet/map/map.h
@@ -190,9 +190,6 @@ typedef struct {
   vlib_combined_counter_main_t *domain_counters;
   volatile u32 *counter_lock;
 
-  /* Global counters */
-  vlib_simple_counter_main_t icmp_relayed;
-
 #ifdef MAP_SKIP_IP6_LOOKUP
   /* pre-presolve */
   u32 adj6_index, adj4_index;
@@ -203,12 +200,14 @@ typedef struct {
   /* Traffic class: zero, copy (~0) or fixed value */
   u8 tc;
   bool tc_copy;
-  bool sec_check;
-  bool sec_check_frag;
-  bool icmp6_enabled;
+
+  bool sec_check;		/* Inbound security check */
+  bool sec_check_frag;		/* Inbound security check for (subsequent) fragments */
+  bool icmp6_enabled;		/* Send destination unreachable for security check failure */
 
   /* ICMPv6 -> ICMPv4 relay parameters */
   ip4_address_t icmp4_src_address;
+  vlib_simple_counter_main_t icmp_relayed;
 
   /* convenience */
   vlib_main_t *vlib_main;
@@ -217,13 +216,13 @@ typedef struct {
   /*
    * IPv4 encap and decap reassembly
    */
-  //Conf
+  /* Configuration */
   f32 ip4_reass_conf_ht_ratio; //Size of ht is 2^ceil(log2(ratio*pool_size))
   u16 ip4_reass_conf_pool_size; //Max number of allocated reass structures
   u16 ip4_reass_conf_lifetime_ms; //Time a reassembly struct is considered valid in ms
   u32 ip4_reass_conf_buffers; //Maximum number of buffers used by ip4 reassembly
 
-  //Runtime
+  /* Runtime */
   map_ip4_reass_t *ip4_reass_pool;
   u8 ip4_reass_ht_log2len; //Hash table size is 2^log2len
   u16 ip4_reass_allocated;
@@ -231,19 +230,22 @@ typedef struct {
   u16 ip4_reass_fifo_last;
   volatile u32 *ip4_reass_lock;
 
-  //Counters
+  /* Counters */
   u32 ip4_reass_buffered_counter;
 
+  bool frag_inner;		/* Inner or outer fragmentation */
+  bool frag_ignore_df;		/* Fragment (outer) packet even if DF is set */
+
   /*
    * IPv6 decap reassembly
    */
-  //Conf
+  /* Configuration */
   f32 ip6_reass_conf_ht_ratio; //Size of ht is 2^ceil(log2(ratio*pool_size))
   u16 ip6_reass_conf_pool_size; //Max number of allocated reass structures
   u16 ip6_reass_conf_lifetime_ms; //Time a reassembly struct is considered valid in ms
   u32 ip6_reass_conf_buffers; //Maximum number of buffers used by ip6 reassembly
 
-  //Runtime
+  /* Runtime */
   map_ip6_reass_t *ip6_reass_pool;
   u8 ip6_reass_ht_log2len; //Hash table size is 2^log2len
   u16 ip6_reass_allocated;
@@ -251,19 +253,18 @@ typedef struct {
   u16 ip6_reass_fifo_last;
   volatile u32 *ip6_reass_lock;
 
-  //Counters
+  /* Counters */
   u32 ip6_reass_buffered_counter;
 
 } map_main_t;
 
 /*
- * TODO: Remove SEC_CHECK / TRANSLATED_4TO6 / TRANSLATED_6TO4
+ * MAP Error counters/messages
  */
 #define foreach_map_error				\
   /* Must be first. */					\
  _(NONE, "valid MAP packets")				\
  _(BAD_PROTOCOL, "bad protocol")			\
- _(WRONG_ICMP_TYPE, "wrong icmp type")			\
  _(SEC_CHECK, "security check failed")			\
  _(ENCAP_SEC_CHECK, "encap security check failed")	\
  _(DECAP_SEC_CHECK, "decap security check failed")	\
@@ -277,7 +278,7 @@ typedef struct {
  _(FRAGMENT_MALFORMED, "fragment has unexpected format")\
  _(FRAGMENT_DROPPED, "dropped cached fragment")         \
  _(MALFORMED, "malformed packet")			\
- _(IP4_ERROR_TIME_EXPIRED, "time expired")
+ _(DF_SET, "can't fragment, DF set")
 
 typedef enum {
 #define _(sym,str) MAP_ERROR_##sym,