filter/trie.c

   1 /*
   2  *      Filters: Trie for prefix sets
   3  *
   4  *      Copyright 2009 Ondrej Zajicek <santiago@crfreenet.org>
   5  *
   6  *      Can be freely distributed and used under the terms of the GNU GPL.
   7  */
   8
   9 /**
  10  * DOC: Trie for prefix sets
  11  *
  12  * We use a (compressed) trie to represent prefix sets. Every node
  13  * in the trie represents one prefix (&addr/&plen) and &plen also
  14  * indicates the index of the bit in the address that is used to
  15  * branch at the node. If we need to represent just a set of
  16  * prefixes, it would be simple, but we have to represent a
  17  * set of prefix patterns. Each prefix pattern consists of
  18  * &ppaddr/&pplen and two integers: &low and &high, and a prefix
  19  * &paddr/&plen matches that pattern if the first MIN(&plen, &pplen)
  20  * bits of &paddr and &ppaddr are the same and &low <= &plen <= &high.
  21  *
  22  * We use a bitmask (&accept) to represent accepted prefix lengths
  23  * at a node. As there are 33 prefix lengths (0..32 for IPv4), but
  24  * there is just one prefix of zero length in the whole trie so we
  25  * have &zero flag in &f_trie (indicating whether the trie accepts
  26  * prefix 0.0.0.0/0) as a special case, and &accept bitmask
  27  * represents accepted prefix lengths from 1 to 32.
  28  *
  29  * There are two cases in prefix matching - a match when the length
  30  * of the prefix is smaller that the length of the prefix pattern,
  31  * (&plen < &pplen) and otherwise. The second case is simple - we
  32  * just walk through the trie and look at every visited node
  33  * whether that prefix accepts our prefix length (&plen). The
  34  * first case is tricky - we don't want to examine every descendant
  35  * of a final node, so (when we create the trie) we have to propagate
  36  * that information from nodes to their ascendants.
  37  *
  38  * Suppose that we have two masks (M1 and M2) for a node. Mask M1
  39  * represents accepted prefix lengths by just the node and mask M2
  40  * represents accepted prefix lengths by the node or any of its
  41  * descendants. Therefore M2 is a bitwise or of M1 and children's
  42  * M2 and this is a maintained invariant during trie building.
  43  * Basically, when we want to match a prefix, we walk through the trie,
  44  * check mask M1 for our prefix length and when we came to
  45  * final node, we check mask M2.
  46  *
  47  * There are two differences in the real implementation. First,
  48  * we use a compressed trie so there is a case that we skip our
  49  * final node (if it is not in the trie) and we came to node that
  50  * is either extension of our prefix, or completely out of path
  51  * In the first case, we also have to check M2.
  52  *
  53  * Second, we really need not to maintain two separate bitmasks.
  54  * Checks for mask M1 are always larger than &applen and we need
  55  * just the first &pplen bits of mask M2 (if trie compression
  56  * hadn't been used it would suffice to know just $applen-th bit),
  57  * so we have to store them together in &accept mask - the first
  58  * &pplen bits of mask M2 and then mask M1.
  59  *
  60  * There are four cases when we walk through a trie:
  61  *
  62  * - we are in NULL
  63  * - we are out of path (prefixes are inconsistent)
  64  * - we are in the wanted (final) node (node length == &plen)
  65  * - we are beyond the end of path (node length > &plen)
  66  * - we are still on path and keep walking (node length < &plen)
  67  *
  68  * The walking code in trie_match_prefix() is structured according to
  69  * these cases.
  70  */
  71
  72 #include "nest/bird.h"
  73 #include "lib/string.h"
  74 #include "conf/conf.h"
  75 #include "filter/filter.h"
  76
  77 /**
  78  * f_new_trie
  79  *
  80  * Allocates and returns a new empty trie.
  81  */
  82 struct f_trie *
  83 f_new_trie(linpool *lp)
  84 {
  85   struct f_trie * ret;
  86   ret = lp_allocz(lp, sizeof(struct f_trie));
  87   ret->lp = lp;
  88   return ret;
  89 }
  90
  91 static inline struct f_trie_node *
  92 new_node(struct f_trie *t, int plen, ip_addr paddr, ip_addr pmask, ip_addr amask)
  93 {
  94   struct f_trie_node *n = lp_allocz(t->lp, sizeof(struct f_trie_node));
  95   n->plen = plen;
  96   n->addr = paddr;
  97   n->mask = pmask;
  98   n->accept = amask;
  99   return n;
 100 }
 101
 102 static inline void
 103 attach_node(struct f_trie_node *parent, struct f_trie_node *child)
 104 {
 105   parent->c[ipa_getbit(child->addr, parent->plen) ? 1 : 0] = child;
 106 }
 107
 108 /**
 109  * trie_add_prefix
 110  * @t: trie to add to
 111  * @px: prefix address
 112  * @plen: prefix length
 113  * @l: prefix lower bound
 114  * @h: prefix upper bound
 115  *
 116  * Adds prefix (prefix pattern) @px/@plen to trie @t.  @l and @h are lower
 117  * and upper bounds on accepted prefix lengths, both inclusive.
 118  * 0 <= l, h <= 32 (128 for IPv6).
 119  */
 120
 121 void
 122 trie_add_prefix(struct f_trie *t, ip_addr px, int plen, int l, int h)
 123 {
 124   if (l == 0)
 125     t->zero = 1;
 126   else
 127     l--;
 128
 129   if (h < plen)
 130     plen = h;
 131
 132   ip_addr amask = ipa_xor(ipa_mkmask(l), ipa_mkmask(h));
 133   ip_addr pmask = ipa_mkmask(plen);
 134   ip_addr paddr = ipa_and(px, pmask);
 135   struct f_trie_node *o = NULL;
 136   struct f_trie_node *n = &t->root;
 137
 138   while(n)
 139     {
 140       ip_addr cmask = ipa_and(n->mask, pmask);
 141
 142       if (ipa_compare(ipa_and(paddr, cmask), ipa_and(n->addr, cmask)))
 143         {
 144           /* We are out of path - we have to add branching node 'b'
 145              between node 'o' and node 'n', and attach new node 'a'
 146              as the other child of 'b'. */
 147           int blen = ipa_pxlen(paddr, n->addr);
 148           ip_addr bmask = ipa_mkmask(blen);
 149           ip_addr baddr = ipa_and(px, bmask);
 150
 151           /* Merge accept masks from children to get accept mask for node 'b' */
 152           ip_addr baccm = ipa_and(ipa_or(amask, n->accept), bmask);
 153
 154           struct f_trie_node *a = new_node(t, plen, paddr, pmask, amask);
 155           struct f_trie_node *b = new_node(t, blen, baddr, bmask, baccm);
 156           attach_node(o, b);
 157           attach_node(b, n);
 158           attach_node(b, a);
 159           return;
 160         }
 161
 162       if (plen < n->plen)
 163         {
 164           /* We add new node 'a' between node 'o' and node 'n' */
 165           amask = ipa_or(amask, ipa_and(n->accept, pmask));
 166           struct f_trie_node *a = new_node(t, plen, paddr, pmask, amask);
 167           attach_node(o, a);
 168           attach_node(a, n);
 169           return;
 170         }
 171
 172       if (plen == n->plen)
 173         {
 174           /* We already found added node in trie. Just update accept mask */
 175           n->accept = ipa_or(n->accept, amask);
 176           return;
 177         }
 178
 179       /* Update accept mask part M2 and go deeper */
 180       n->accept = ipa_or(n->accept, ipa_and(amask, n->mask));
 181
 182       /* n->plen < plen and plen <= 32 (128) */
 183       o = n;
 184       n = n->c[ipa_getbit(paddr, n->plen) ? 1 : 0];
 185     }
 186
 187   /* We add new tail node 'a' after node 'o' */
 188   struct f_trie_node *a = new_node(t, plen, paddr, pmask, amask);
 189   attach_node(o, a);
 190 }
 191
 192 /**
 193  * trie_match
 194  * @t: trie
 195  * @px: prefix address
 196  * @plen: prefix length
 197  *
 198  * Tries to find a matching prefix pattern in the trie such that
 199  * prefix @px/@plen matches that prefix pattern. Returns 1 if there
 200  * is such prefix pattern in the trie.
 201  */
 202 int
 203 trie_match_prefix(struct f_trie *t, ip_addr px, int plen)
 204 {
 205   ip_addr pmask = ipa_mkmask(plen);
 206   ip_addr paddr = ipa_and(px, pmask);
 207
 208   if (plen == 0)
 209     return t->zero;
 210
 211   int plentest = plen - 1;
 212   struct f_trie_node *n = &t->root;
 213
 214   while(n)
 215     {
 216       ip_addr cmask = ipa_and(n->mask, pmask);
 217
 218       /* We are out of path */
 219       if (ipa_compare(ipa_and(paddr, cmask), ipa_and(n->addr, cmask)))
 220         return 0;
 221
 222       /* Check accept mask */
 223       if (ipa_getbit(n->accept, plentest))
 224         return 1;
 225
 226       /* We finished trie walk and still no match */
 227       if (plen <= n->plen)
 228         return 0;
 229
 230       /* Choose children */
 231       n =  n->c[(ipa_getbit(paddr, n->plen)) ? 1 : 0];
 232     }
 233
 234   return 0;
 235 }
 236
 237 static int
 238 trie_node_same(struct f_trie_node *t1, struct f_trie_node *t2)
 239 {
 240   if ((t1 == NULL) && (t2 == NULL))
 241     return 1;
 242
 243   if ((t1 == NULL) || (t2 == NULL))
 244     return 0;
 245
 246   if ((t1->plen != t2->plen) ||
 247       (! ipa_equal(t1->addr, t2->addr)) ||
 248       (! ipa_equal(t1->accept, t2->accept)))
 249     return 0;
 250
 251   return trie_node_same(t1->c[0], t2->c[0]) && trie_node_same(t1->c[1], t2->c[1]);
 252 }
 253
 254 /**
 255  * trie_same
 256  * @t1: first trie to be compared
 257  * @t2: second one
 258  *
 259  * Compares two tries and returns 1 if they are same
 260  */
 261 int
 262 trie_same(struct f_trie *t1, struct f_trie *t2)
 263 {
 264   return (t1->zero == t2->zero) && trie_node_same(&t1->root, &t2->root);
 265 }
 266
 267 static void
 268 trie_node_print(struct f_trie_node *t, char **sep)
 269 {
 270   if (t == NULL)
 271     return;
 272
 273   if (ipa_nonzero(t->accept))
 274     {
 275       logn("%s%I/%d{%I}", *sep, t->addr, t->plen, t->accept);
 276       *sep = ", ";
 277     }
 278
 279   trie_node_print(t->c[0], sep);
 280   trie_node_print(t->c[1], sep);
 281 }
 282
 283 /**
 284  * trie_print
 285  * @t: trie to be printed
 286  *
 287  * Prints the trie to the log buffer.
 288  */
 289 void
 290 trie_print(struct f_trie *t)
 291 {
 292   char *sep = "";
 293   logn("[");
 294   if (t->zero)
 295     {
 296       logn("0.0.0.0/0");
 297       sep = ", ";
 298     }
 299   trie_node_print(&t->root, &sep);
 300   logn("]");
 301 }