Rewrite of buggy AS path matching.

Old AS path maching supposes thath AS number appears
only once in AS path, but that is not true. It also
contains some bugs related to AS path sets.

New code does not use any assumptions about semantic
structure of AS path. It is asymptotically slower than
the old code, but on real paths it is not significant.

It also allows '?' for matching one arbitrary AS number.

doc/bird.sgml

@@ -549,7 +549,8 @@ incompatible with each other (that is to prevent you from shooting in the foot).
 	  (using <cf>path &tilde; [= 2 3 5 * =]</cf> syntax). The masks
 	  resemble wildcard patterns as used by UNIX shells. Autonomous
 	  system numbers match themselves, <cf/*/ matches any (even empty)
-	  sequence of arbitrary AS numbers. For example, if <cf>bgp_path</cf> is 4 3 2 1, then:
+	  sequence of arbitrary AS numbers and <cf/?/ matches one arbitrary AS number.
+	  For example, if <cf>bgp_path</cf> is 4 3 2 1, then:
 	  <tt>bgp_path &tilde; [= * 4 3 * =]</tt> is true, but 
 	  <tt>bgp_path &tilde; [= * 4 5 * =]</tt> is false.
 	  There is also old syntax that uses / .. / instead of [= .. =] and ? instead of *.

filter/config.Y

@@ -284,14 +284,15 @@ bgp_path:
  ;
 
 bgp_path_tail1:
-   NUM bgp_path_tail1 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2;   $$->val  = $1; $$->any = 0; }
+   NUM bgp_path_tail1 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2; $$->kind = PM_ASN;      $$->val = $1; }
- | '*' bgp_path_tail1 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2;   $$->val  = 0;  $$->any = 1; }
+ | '*' bgp_path_tail1 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2; $$->kind = PM_ASTERISK; $$->val  = 0; }
+ | '?' bgp_path_tail1 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2; $$->kind = PM_QUESTION; $$->val  = 0; }
  |  		      { $$ = NULL; }
  ;
 
 bgp_path_tail2:
-   NUM bgp_path_tail2 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2;   $$->val  = $1; $$->any = 0; }
+   NUM bgp_path_tail2 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2; $$->kind = PM_ASN;      $$->val = $1; }
- | '?' bgp_path_tail2 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2;   $$->val  = 0;  $$->any = 1; }
+ | '?' bgp_path_tail2 { $$ = cfg_alloc(sizeof(struct f_path_mask)); $$->next = $2; $$->kind = PM_ASTERISK; $$->val  = 0; }
  | 		      { $$ = NULL; }
  ;
 

filter/filter.c

@@ -82,10 +82,10 @@ pm_format(struct f_path_mask *p, byte *buf, unsigned int size)
 	  return;
 	}
 
-      if (p->any)
+      if (p->kind == PM_ASN)
-	buf += bsprintf(buf, " *");
-      else
 	buf += bsprintf(buf, " %u", p->val);
+      else
+	buf += bsprintf(buf, (p->kind == PM_ASTERISK) ? " *" : " ?");
 
       p = p->next;
     }

nest/a-path.c

@@ -280,79 +280,171 @@ as_path_is_member(struct adata *path, u32 as)
 }
 
 
+struct pm_pos
+{
+  u8 set;
+  u8 mark;
+  union
+  {
+    char *sp;
+    u32 asn;
+  } val;
+};
 
-#define MASK_PLUS do { mask = mask->next; if (!mask) return next == q; \
+static int
-		       asterisk = mask->any; \
+parse_path(struct adata *path, struct pm_pos *pos)
-                       if (asterisk) { mask = mask->next; if (!mask) { return 1; } } \
+{
-		       } while(0)
+  int bs = bgp_as4_support ? 4 : 2;
+  u8 *p = path->data;
+  u8 *q = p + path->length;
+  struct pm_pos *opos = pos;
+  int i, j, len;
+
+
+  while (p < q)
+    switch (*p++)
+      {
+      case AS_PATH_SET:
+	pos->set = 1;
+	pos->mark = 0;
+	pos->val.sp = p;
+	len = *p;
+	p += 1 + bs * len;
+	pos++;
+	break;
+      
+      case AS_PATH_SEQUENCE:
+	len = *p++;
+	for (i = 0; i < len; i++)
+	  {
+	    pos->set = 0;
+	    pos->mark = 0;
+	    pos->val.asn = get_as(p);
+	    p += bs;
+	    pos++;
+	  }
+	break;
+
+      default:
+	bug("as_path_match: Invalid path component");
+      }
+  
+  return pos - opos;
+}
+
+
+static int
+pm_match(struct pm_pos *pos, u32 asn)
+{
+  if (! pos->set)
+    return pos->val.asn == asn;
+
+  int bs = bgp_as4_support ? 4 : 2;
+  u8 *p = pos->val.sp;
+  int len = *p++;
+  int i;
+
+  for (i = 0; i < len; i++)
+    if (get_as(p + i * bs) == asn)
+      return 1;
+
+  return 0;
+}
+
+static void
+pm_mark(struct pm_pos *pos, int i, int plen, int *nl, int *nh)
+{
+  int j;
+
+  if (pos[i].set)
+    pos[i].mark = 1;
+  
+  for (j = i + 1; (j < plen) && pos[j].set && (! pos[j].mark); j++)
+    pos[j].mark = 1;
+  pos[j].mark = 1;
+
+  /* We are going downwards, therefore every mark is
+     new low and just the first mark is new high */
+
+  *nl = i + (pos[i].set ? 0 : 1);
+
+  if (*nh < 0)
+    *nh = j;
+}
+
+/* AS path matching is nontrivial. Because AS path can
+ * contain sets, it is not a plain wildcard matching. A set 
+ * in an AS path is interpreted as it might represent any
+ * sequence of AS numbers from that set (possibly with
+ * repetitions). So it is also a kind of a pattern,
+ * more complicated than a path mask.
+ *
+ * The algorithm for AS path matching is a variant
+ * of nondeterministic finite state machine, where
+ * positions in AS path are states, and items in
+ * path mask are input for that finite state machine.
+ * During execution of the algorithm we maintain a set
+ * of marked states - a state is marked if it can be
+ * reached by any walk through NFSM with regard to
+ * currently processed part of input. When we process
+ * next part of mask, we advance each marked state.
+ * We start with marked first position, when we
+ * run out of marked positions, we reject. When
+ * we process the whole mask, we accept iff final position
+ * (auxiliary position after last real position in AS path)
+ * is marked.
+ */
 
 int
 as_path_match(struct adata *path, struct f_path_mask *mask)
 {
-  int bs = bgp_as4_support ? 4 : 2;
+  struct pm_pos pos[2048 + 1];
-  int i;
+  int plen = parse_path(path, pos);
-  int asterisk = 0;
+  int l, h, i, nh, nl;
-  u8 *p = path->data;
-  u8 *q = p+path->length;
-  int len;
-  u8 *next;
-  u32 as;
 
-  if (!mask)
+  /* l and h are bound of interval of positions where
-    return ! path->length;
+     are marked states */
 
-  asterisk = mask->any;
+  pos[plen].set = 0;
-  if (asterisk)
+  pos[plen].mark = 0;
-    { mask = mask->next; if (!mask) return 1; }
 
-  while (p<q) {
+  l = h = 0;
-    switch (*p++) {
+  pos[0].mark = 1;
-    case AS_PATH_SET:
-      len = *p++;
-      {
-	u8 *p_save = p;
-	next = p_save + bs * len;
-      retry:
-	p = p_save;
-	for (i=0; i<len; i++) {
-	  as = get_as(p);
-	  if (asterisk && (as == mask->val)) {
-	    MASK_PLUS;
-	    goto retry;
-	  }
-	  if (!asterisk && (as == mask->val)) {
-	    p = next;
-	    MASK_PLUS;
-	    goto okay;
-	  }
-	  p += bs;
-	}
-	if (!asterisk)
-	  return 0;
-      okay: ;
-      }
-      break;
   
-    case AS_PATH_SEQUENCE:
+  while (mask)
-      len = *p++;
+    {
-      for (i=0; i<len; i++) {
+      /* We remove this mark to not step after pos[plen] */
-	next = p + bs;
+      pos[plen].mark = 0;
-	as = get_as(p);
+
-	if (asterisk && (as == mask->val))
+      switch (mask->kind)
-	  MASK_PLUS;
+	{
-	else if (!asterisk) {
+	case PM_ASTERISK:
-	  if (as != mask->val)
+	  for (i = l; i <= plen; i++)
+	    pos[i].mark = 1;
+	  h = plen;
+	  break;
+
+	case PM_QUESTION:
+	case PM_ASN:
+	  nh = -1;
+	  for (i = h; i >= l; i--)
+	    if (pos[i].mark)
+	      {
+		pos[i].mark = 0;
+		if ((mask->kind == PM_QUESTION) || pm_match(pos + i, mask->val))
+		  pm_mark(pos, i, plen, &nl, &nh);
+	      }
+
+	  if (nh < 0)
 	    return 0;
-	  MASK_PLUS;
+
+	  h = nh;
+	  l = nl;
+	  break;
 	}
-	p += bs;
-      }
-      break;
 
-    default:
+      mask = mask->next;
-      bug("as_path_match: Invalid path component");
     }
-  }
-  return 0;
-}
 
+  return pos[plen].mark;
+}

nest/attrs.h

@@ -32,15 +32,16 @@ int as_path_get_first(struct adata *path, u32 *orig_as);
 int as_path_get_last(struct adata *path, u32 *last_as);
 int as_path_is_member(struct adata *path, u32 as);
 
+#define PM_ASN		0
+#define PM_QUESTION	1
+#define PM_ASTERISK	2
 
 struct f_path_mask {
   struct f_path_mask *next;
+  int kind;
   u32 val;
-  int any;
 };
 
-// #define PM_ANY -1
-
 int as_path_match(struct adata *path, struct f_path_mask *mask);
 
 /* a-set.c */