ipsec: GCM, Anti-replay and ESN fixess

Type: fix

Several Fixes:
 1 - Anti-replay did not work with GCM becuase it overwrote the sequence
number in the ESP header. To fix i added the seq num to the per-packet
data so it is preserved
 2 - The high sequence number was not byte swapped during ESP encrypt.
 3 - openssl engine was the only one to return FAIL_DECRYPT for bad GCM
the others return BAD_HMAC. removed the former
 4 - improved tracing to show the low and high seq numbers
 5 - documented the anti-replay window checks
 6 - fixed scapy patch for ESN support for GCM
 7 - tests for anti-reply (w/ and w/o ESN) for each crypto algo

Change-Id: Id65d96b6d1d4dd821b2ab557e87468fff6d70e5b
Signed-off-by: Neale Ranns <nranns@cisco.com>

4 files changed, 129 insertions, 42 deletions

diff --git a/src/vnet/ipsec/ah_decrypt.c b/src/vnet/ipsec/ah_decrypt.c
index 741fa91b95c..bc6b5c4ec9d 100644
--- a/src/vnet/ipsec/ah_decrypt.c
+++ b/src/vnet/ipsec/ah_decrypt.c
@@ -203,7 +203,7 @@ ah_decrypt_inline (vlib_main_t * vm,
       pd->seq = clib_host_to_net_u32 (ah0->seq_no);
 
       /* anti-replay check */
-      if (ipsec_sa_anti_replay_check (sa0, &ah0->seq_no))
+      if (ipsec_sa_anti_replay_check (sa0, pd->seq))
 	{
 	  b[0]->error = node->errors[AH_DECRYPT_ERROR_REPLAY];
 	  next[0] = AH_DECRYPT_NEXT_DROP;
@@ -303,7 +303,7 @@ ah_decrypt_inline (vlib_main_t * vm,
 
       if (PREDICT_TRUE (sa0->integ_alg != IPSEC_INTEG_ALG_NONE))
 	{
-	  ipsec_sa_anti_replay_advance (sa0, clib_host_to_net_u32 (pd->seq));
+	  ipsec_sa_anti_replay_advance (sa0, pd->seq);
 	}
 
       u16 ah_hdr_len = sizeof (ah_header_t) + pd->icv_size
diff --git a/src/vnet/ipsec/esp_decrypt.c b/src/vnet/ipsec/esp_decrypt.c
index 48f08f42e33..c2b9bf4dc0c 100644
--- a/src/vnet/ipsec/esp_decrypt.c
+++ b/src/vnet/ipsec/esp_decrypt.c
@@ -67,6 +67,8 @@ static char *esp_decrypt_error_strings[] = {
 typedef struct
 {
   u32 seq;
+  u32 sa_seq;
+  u32 sa_seq_hi;
   ipsec_crypto_alg_t crypto_alg;
   ipsec_integ_alg_t integ_alg;
 } esp_decrypt_trace_t;
@@ -79,9 +81,11 @@ format_esp_decrypt_trace (u8 * s, va_list * args)
   CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *);
   esp_decrypt_trace_t *t = va_arg (*args, esp_decrypt_trace_t *);
 
-  s = format (s, "esp: crypto %U integrity %U seq %u",
-	      format_ipsec_crypto_alg, t->crypto_alg,
-	      format_ipsec_integ_alg, t->integ_alg, t->seq);
+  s =
+    format (s,
+	    "esp: crypto %U integrity %U pkt-seq %d sa-seq %u sa-seq-hi %u",
+	    format_ipsec_crypto_alg, t->crypto_alg, format_ipsec_integ_alg,
+	    t->integ_alg, t->seq, t->sa_seq, t->sa_seq_hi);
   return s;
 }
 
@@ -99,12 +103,13 @@ typedef struct
     u64 sa_data;
   };
 
+  u32 seq;
   i16 current_data;
   i16 current_length;
   u16 hdr_sz;
 } esp_decrypt_packet_data_t;
 
-STATIC_ASSERT_SIZEOF (esp_decrypt_packet_data_t, 2 * sizeof (u64));
+STATIC_ASSERT_SIZEOF (esp_decrypt_packet_data_t, 3 * sizeof (u64));
 
 #define ESP_ENCRYPT_PD_F_FD_TRANSPORT (1 << 2)
 
@@ -177,6 +182,7 @@ esp_decrypt_inline (vlib_main_t * vm,
       pd->current_length = b[0]->current_length;
       pd->hdr_sz = pd->current_data - vnet_buffer (b[0])->l3_hdr_offset;
       payload = b[0]->data + pd->current_data;
+      pd->seq = clib_host_to_net_u32 (((esp_header_t *) payload)->seq);
 
       /* we need 4 extra bytes for HMAC calculation when ESN are used */
       if (ipsec_sa_is_set_USE_ESN (sa0) && pd->icv_sz &&
@@ -188,7 +194,7 @@ esp_decrypt_inline (vlib_main_t * vm,
 	}
 
       /* anti-reply check */
-      if (ipsec_sa_anti_replay_check (sa0, &((esp_header_t *) payload)->seq))
+      if (ipsec_sa_anti_replay_check (sa0, pd->seq))
 	{
 	  b[0]->error = node->errors[ESP_DECRYPT_ERROR_REPLAY];
 	  next[0] = ESP_DECRYPT_NEXT_DROP;
@@ -221,10 +227,11 @@ esp_decrypt_inline (vlib_main_t * vm,
 	  op->len = len;
 	  if (ipsec_sa_is_set_USE_ESN (sa0))
 	    {
-	      /* shift ICV for 4 bytes to insert ESN */
+	      /* shift ICV by 4 bytes to insert ESN */
+	      u32 seq_hi = clib_host_to_net_u32 (sa0->seq_hi);
 	      u8 tmp[ESP_MAX_ICV_SIZE], sz = sizeof (sa0->seq_hi);
 	      clib_memcpy_fast (tmp, payload + len, ESP_MAX_ICV_SIZE);
-	      clib_memcpy_fast (payload + len, &sa0->seq_hi, sz);
+	      clib_memcpy_fast (payload + len, &seq_hi, sz);
 	      clib_memcpy_fast (payload + len + sz, tmp, ESP_MAX_ICV_SIZE);
 	      op->len += sz;
 	      op->digest += sz;
@@ -368,9 +375,8 @@ esp_decrypt_inline (vlib_main_t * vm,
 	goto trace;
 
       sa0 = vec_elt_at_index (im->sad, pd->sa_index);
-      u8 *payload = b[0]->data + pd->current_data;
 
-      ipsec_sa_anti_replay_advance (sa0, ((esp_header_t *) payload)->seq);
+      ipsec_sa_anti_replay_advance (sa0, pd->seq);
 
       esp_footer_t *f = (esp_footer_t *) (b[0]->data + pd->current_data +
 					  pd->current_length - sizeof (*f) -
@@ -485,13 +491,14 @@ esp_decrypt_inline (vlib_main_t * vm,
       if (PREDICT_FALSE (b[0]->flags & VLIB_BUFFER_IS_TRACED))
 	{
 	  esp_decrypt_trace_t *tr;
-	  u8 *payload = b[0]->data + pd->current_data;
 	  tr = vlib_add_trace (vm, node, b[0], sizeof (*tr));
 	  sa0 = pool_elt_at_index (im->sad,
 				   vnet_buffer (b[0])->ipsec.sad_index);
 	  tr->crypto_alg = sa0->crypto_alg;
 	  tr->integ_alg = sa0->integ_alg;
-	  tr->seq = clib_host_to_net_u32 (((esp_header_t *) payload)->seq);
+	  tr->seq = pd->seq;
+	  tr->sa_seq = sa0->last_seq;
+	  tr->sa_seq_hi = sa0->seq_hi;
 	}
 
       /* next */
diff --git a/src/vnet/ipsec/esp_encrypt.c b/src/vnet/ipsec/esp_encrypt.c
index 041b268975d..47c079d95d2 100644
--- a/src/vnet/ipsec/esp_encrypt.c
+++ b/src/vnet/ipsec/esp_encrypt.c
@@ -65,6 +65,7 @@ typedef struct
   u32 sa_index;
   u32 spi;
   u32 seq;
+  u32 sa_seq_hi;
   u8 udp_encap;
   ipsec_crypto_alg_t crypto_alg;
   ipsec_integ_alg_t integ_alg;
@@ -80,8 +81,9 @@ format_esp_encrypt_trace (u8 * s, va_list * args)
 
   s =
     format (s,
-	    "esp: sa-index %d spi %u (0x%08x) seq %u crypto %U integrity %U%s",
-	    t->sa_index, t->spi, t->spi, t->seq, format_ipsec_crypto_alg,
+	    "esp: sa-index %d spi %u (0x%08x) seq %u sa-seq-hi %u crypto %U integrity %U%s",
+	    t->sa_index, t->spi, t->spi, t->seq, t->sa_seq_hi,
+	    format_ipsec_crypto_alg,
 	    t->crypto_alg, format_ipsec_integ_alg, t->integ_alg,
 	    t->udp_encap ? " udp-encap-enabled" : "");
   return s;
@@ -521,7 +523,8 @@ esp_encrypt_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
 						    sizeof (*tr));
 	  tr->sa_index = sa_index0;
 	  tr->spi = sa0->spi;
-	  tr->seq = sa0->seq - 1;
+	  tr->seq = sa0->seq;
+	  tr->sa_seq_hi = sa0->seq_hi;
 	  tr->udp_encap = ipsec_sa_is_set_UDP_ENCAP (sa0);
 	  tr->crypto_alg = sa0->crypto_alg;
 	  tr->integ_alg = sa0->integ_alg;
diff --git a/src/vnet/ipsec/ipsec_sa.h b/src/vnet/ipsec/ipsec_sa.h
index 811f4cabcf0..86d76b3ac0a 100644
--- a/src/vnet/ipsec/ipsec_sa.h
+++ b/src/vnet/ipsec/ipsec_sa.h
@@ -19,8 +19,6 @@
 #include <vnet/ip/ip.h>
 #include <vnet/fib/fib_node.h>
 
-#define IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE (64)
-
 #define foreach_ipsec_crypto_alg    \
   _ (0, NONE, "none")               \
   _ (1, AES_CBC_128, "aes-cbc-128") \
@@ -232,18 +230,34 @@ extern uword unformat_ipsec_integ_alg (unformat_input_t * input,
 				       va_list * args);
 extern uword unformat_ipsec_key (unformat_input_t * input, va_list * args);
 
+/*
+ * Anti Replay definitions
+ */
+
+#define IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE (64)
+#define IPSEC_SA_ANTI_REPLAY_WINDOW_MAX_INDEX (IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE-1)
+
+/*
+ * sequence number less than the lower bound are outside of the window
+ * From RFC4303 Appendix A:
+ *  Bl = Tl - W + 1
+ */
+#define IPSEC_SA_ANTI_REPLAY_WINDOW_LOWER_BOUND(_tl) (_tl - IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE + 1)
+
+/*
+ * Anti replay check.
+ *  inputs need to be in host byte order.
+ */
 always_inline int
-ipsec_sa_anti_replay_check (ipsec_sa_t * sa, u32 * seqp)
+ipsec_sa_anti_replay_check (ipsec_sa_t * sa, u32 seq)
 {
-  u32 seq, diff, tl, th;
+  u32 diff, tl, th;
+
   if ((sa->flags & IPSEC_SA_FLAG_USE_ANTI_REPLAY) == 0)
     return 0;
 
-  seq = clib_net_to_host_u32 (*seqp);
-
-  if ((sa->flags & IPSEC_SA_FLAG_USE_ESN) == 0)
+  if (!ipsec_sa_is_set_USE_ESN (sa))
     {
-
       if (PREDICT_TRUE (seq > sa->last_seq))
 	return 0;
 
@@ -261,50 +275,113 @@ ipsec_sa_anti_replay_check (ipsec_sa_t * sa, u32 * seqp)
   th = sa->last_seq_hi;
   diff = tl - seq;
 
-  if (PREDICT_TRUE (tl >= (IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE - 1)))
+  if (PREDICT_TRUE (tl >= (IPSEC_SA_ANTI_REPLAY_WINDOW_MAX_INDEX)))
     {
-      if (seq >= (tl - IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE + 1))
+      /*
+       * the last sequence number VPP recieved is more than one
+       * window size greater than zero.
+       * Case A from RFC4303 Appendix A.
+       */
+      if (seq < IPSEC_SA_ANTI_REPLAY_WINDOW_LOWER_BOUND (tl))
+	{
+	  /*
+	   * the received sequence number is lower than the lower bound
+	   * of the window, this could mean either a replay packet or that
+	   * the high sequence number has wrapped. if it decrypts corrently
+	   * then it's the latter.
+	   */
+	  sa->seq_hi = th + 1;
+	  return 0;
+	}
+      else
 	{
+	  /*
+	   * the recieved sequence number greater than the low
+	   * end of the window.
+	   */
 	  sa->seq_hi = th;
 	  if (seq <= tl)
+	    /*
+	     * The recieved seq number is within bounds of the window
+	     * check if it's a duplicate
+	     */
 	    return (sa->replay_window & (1ULL << diff)) ? 1 : 0;
 	  else
+	    /*
+	     * The received sequence number is greater than the window
+	     * upper bound. this packet will move the window along, assuming
+	     * it decrypts correctly.
+	     */
 	    return 0;
 	}
-      else
-	{
-	  sa->seq_hi = th + 1;
-	  return 0;
-	}
     }
   else
     {
-      if (seq >= (tl - IPSEC_SA_ANTI_REPLAY_WINDOW_SIZE + 1))
+      /*
+       * the last sequence number VPP recieved is within one window
+       * size of zero, i.e. 0 < TL < WINDOW_SIZE, the lower bound is thus a
+       * large sequence number.
+       * Note that the check below uses unsiged integer arthimetic, so the
+       * RHS will be a larger number.
+       * Case B from RFC4303 Appendix A.
+       */
+      if (seq < IPSEC_SA_ANTI_REPLAY_WINDOW_LOWER_BOUND (tl))
 	{
-	  sa->seq_hi = th - 1;
-	  return (sa->replay_window & (1ULL << diff)) ? 1 : 0;
+	  /*
+	   * the sequence number is less than the lower bound.
+	   */
+	  if (seq <= tl)
+	    {
+	      /*
+	       * the packet is within the window upper bound.
+	       * check for duplicates.
+	       */
+	      sa->seq_hi = th;
+	      return (sa->replay_window & (1ULL << diff)) ? 1 : 0;
+	    }
+	  else
+	    {
+	      /*
+	       * the packet is less the window lower bound or greater than
+	       * the higher bound, depending on how you look at it...
+	       * We're assuming, given that the last sequence number received,
+	       * TL < WINDOW_SIZE, that a largeer seq num is more likely to be
+	       * a packet that moves the window forward, than a packet that has
+	       * wrapped the high sequence again. If it were the latter then
+	       * we've lost close to 2^32 packets.
+	       */
+	      sa->seq_hi = th;
+	      return 0;
+	    }
 	}
       else
 	{
-	  sa->seq_hi = th;
-	  if (seq <= tl)
-	    return (sa->replay_window & (1ULL << diff)) ? 1 : 0;
-	  else
-	    return 0;
+	  /*
+	   * the packet seq number is between the lower bound (a large nubmer)
+	   * and MAX_SEQ_NUM. This is in the window since the window upper bound
+	   * tl > 0.
+	   * However, since TL is the other side of 0 to the received
+	   * packet, the SA has moved on to a higher sequence number.
+	   */
+	  sa->seq_hi = th - 1;
+	  return (sa->replay_window & (1ULL << diff)) ? 1 : 0;
 	}
     }
 
   return 0;
 }
 
+/*
+ * Anti replay window advance
+ *  inputs need to be in host byte order.
+ */
 always_inline void
-ipsec_sa_anti_replay_advance (ipsec_sa_t * sa, u32 seqp)
+ipsec_sa_anti_replay_advance (ipsec_sa_t * sa, u32 seq)
 {
-  u32 pos, seq;
+  u32 pos;
   if (PREDICT_TRUE (sa->flags & IPSEC_SA_FLAG_USE_ANTI_REPLAY) == 0)
     return;
 
-  seq = clib_host_to_net_u32 (seqp);
   if (PREDICT_TRUE (sa->flags & IPSEC_SA_FLAG_USE_ESN))
     {
       int wrap = sa->seq_hi - sa->last_seq_hi;