]> git.ipfire.org Git - thirdparty/bird.git/blob - filter/trie.c
Merge commit '0c59f7ff' into haugesund
[thirdparty/bird.git] / filter / trie.c
1 /*
2 * Filters: Trie for prefix sets
3 *
4 * (c) 2009--2021 Ondrej Zajicek <santiago@crfreenet.org>
5 * (c) 2009--2021 CZ.NIC z.s.p.o.
6 *
7 * Can be freely distributed and used under the terms of the GNU GPL.
8 */
9
10 /**
11 * DOC: Trie for prefix sets
12 *
13 * We use a (compressed) trie to represent prefix sets. Every node in the trie
14 * represents one prefix (&addr/&plen) and &plen also indicates the index of
15 * bits in the address that are used to branch at the node. Note that such
16 * prefix is not necessary a member of the prefix set, it is just a canonical
17 * prefix associated with a node. Prefix lengths of nodes are aligned to
18 * multiples of &TRIE_STEP (4) and there is 16-way branching in each
19 * node. Therefore, we say that a node is associated with a range of prefix
20 * lengths (&plen .. &plen + TRIE_STEP - 1).
21 *
22 * The prefix set is not just a set of prefixes, it is defined by a set of
23 * prefix patterns. Each prefix pattern consists of &ppaddr/&pplen and two
24 * integers: &low and &high. The tested prefix &paddr/&plen matches that pattern
25 * if the first MIN(&plen, &pplen) bits of &paddr and &ppaddr are the same and
26 * &low <= &plen <= &high.
27 *
28 * There are two ways to represent accepted prefixes for a node. First, there is
29 * a bitmask &local, which represents independently all 15 prefixes that extend
30 * the canonical prefix of the node and are within a range of prefix lengths
31 * associated with the node. E.g., for node 10.0.0.0/8 they are 10.0.0.0/8,
32 * 10.0.0.0/9, 10.128.0.0/9, .. 10.224.0.0/11. This order (first by length, then
33 * lexicographically) is used for indexing the bitmask &local, starting at
34 * position 1. I.e., index is 2^(plen - base) + offset within the same length,
35 * see function trie_local_mask6() for details.
36 *
37 * Second, we use a bitmask &accept to represent accepted prefix lengths at a
38 * node. The bit is set means that all prefixes of given length that are either
39 * subprefixes or superprefixes of the canonical prefix are accepted. As there
40 * are 33 prefix lengths (0..32 for IPv4), but there is just one prefix of zero
41 * length in the whole trie so we have &zero flag in &f_trie (indicating whether
42 * the trie accepts prefix 0.0.0.0/0) as a special case, and &accept bitmask
43 * represents accepted prefix lengths from 1 to 32.
44 *
45 * One complication is handling of prefix patterns with unaligned prefix length.
46 * When such pattern is to be added, we add a primary node above (with rounded
47 * down prefix length &nlen) and a set of secondary nodes below (with rounded up
48 * prefix lengths &slen). Accepted prefix lengths of the original prefix pattern
49 * are then represented in different places based on their lengths. For prefixes
50 * shorter than &nlen, it is &accept bitmask of the primary node, for prefixes
51 * between &nlen and &slen - 1 it is &local bitmask of the primary node, and for
52 * prefixes longer of equal &slen it is &accept bitmasks of secondary nodes.
53 *
54 * There are two cases in prefix matching - a match when the length of the
55 * prefix is smaller that the length of the prefix pattern, (&plen < &pplen) and
56 * otherwise. The second case is simple - we just walk through the trie and look
57 * at every visited node whether that prefix accepts our prefix length (&plen).
58 * The first case is tricky - we do not want to examine every descendant of a
59 * final node, so (when we create the trie) we have to propagate that
60 * information from nodes to their ascendants.
61 *
62 * There are two kinds of propagations - propagation from child's &accept
63 * bitmask to parent's &accept bitmask, and propagation from child's &accept
64 * bitmask to parent's &local bitmask. The first kind is simple - as all
65 * superprefixes of a parent are also all superprefixes of appropriate length of
66 * a child, then we can just add (by bitwise or) a child &accept mask masked by
67 * parent prefix length mask to the parent &accept mask. This handles prefixes
68 * shorter than node &plen.
69 *
70 * The second kind of propagation is necessary to handle superprefixes of a
71 * child that are represented by parent &local mask - that are in the range of
72 * prefix lengths associated with the parent. For each accepted (by child
73 * &accept mask) prefix length from that range, we need to set appropriate bit
74 * in &local mask. See function trie_amask_to_local() for details.
75 *
76 * There are four cases when we walk through a trie:
77 *
78 * - we are in NULL
79 * - we are out of path (prefixes are inconsistent)
80 * - we are in the wanted (final) node (node length == &plen)
81 * - we are beyond the end of path (node length > &plen)
82 * - we are still on path and keep walking (node length < &plen)
83 *
84 * The walking code in trie_match_net() is structured according to these cases.
85 *
86 * Iteration over prefixes in a trie can be done using TRIE_WALK() macro, or
87 * directly using trie_walk_init() and trie_walk_next() functions. The second
88 * approach allows suspending the iteration and continuing in it later.
89 * Prefixes are enumerated in the usual lexicographic order and may be
90 * restricted to a subset of the trie (all subnets of a specified prefix).
91 *
92 * Note that the trie walk does not reliably enumerate `implicit' prefixes
93 * defined by &low and &high fields in prefix patterns, it is supposed to be
94 * used on tries constructed from `explicit' prefixes (&low == &plen == &high
95 * in call to trie_add_prefix()).
96 *
97 * The trie walk has three basic state variables stored in the struct
98 * &f_trie_walk_state -- the current node in &stack[stack_pos], &accept_length
99 * for iteration over inter-node prefixes (non-branching prefixes on compressed
100 * path between the current node and its parent node, stored in the bitmap
101 * &accept of the current node) and &local_pos for iteration over intra-node
102 * prefixes (stored in the bitmap &local).
103 *
104 * The trie also supports longest-prefix-match query by trie_match_longest_ip4()
105 * and it can be extended to iteration over all covering prefixes for a given
106 * prefix (from longest to shortest) using TRIE_WALK_TO_ROOT_IP4() macro. There
107 * are also IPv6 versions (for practical reasons, these functions and macros are
108 * separate for IPv4 and IPv6). There is the same limitation to enumeration of
109 * `implicit' prefixes like with the previous TRIE_WALK() macro.
110 */
111
112 #include "nest/bird.h"
113 #include "lib/string.h"
114 #include "conf/conf.h"
115 #include "filter/filter.h"
116 #include "filter/data.h"
117
118
119 /*
120 * In the trie_add_prefix(), we use ip_addr (assuming that it is the same as
121 * ip6_addr) to handle both IPv4 and IPv6 prefixes. In contrast to rest of the
122 * BIRD, IPv4 addresses are just zero-padded from right. That is why we have
123 * ipt_from_ip4() and ipt_to_ip4() macros below.
124 */
125
126 #define ipa_mkmask(x) ip6_mkmask(x)
127 #define ipa_masklen(x) ip6_masklen(&x)
128 #define ipa_pxlen(x,y) ip6_pxlen(x,y)
129 #define ipa_getbit(a,p) ip6_getbit(a,p)
130 #define ipa_getbits(a,p,n) ip6_getbits(a,p,n)
131 #define ipa_setbits(a,p,n) ip6_setbits(a,p,n)
132 #define trie_local_mask(a,b,c) trie_local_mask6(a,b,c)
133
134 #define ipt_from_ip4(x) _MI6(_I(x), 0, 0, 0)
135 #define ipt_to_ip4(x) _MI4(_I0(x))
136
137
138 /**
139 * f_new_trie - allocates and returns a new empty trie
140 * @lp: linear pool to allocate items from
141 * @data_size: user data attached to node
142 */
143 struct f_trie *
144 f_new_trie(linpool *lp, uint data_size)
145 {
146 struct f_trie * ret;
147 ret = lp_allocz(lp, sizeof(struct f_trie) + data_size);
148 ret->lp = lp;
149 ret->ipv4 = -1;
150 ret->data_size = data_size;
151 return ret;
152 }
153
154 static inline struct f_trie_node4 *
155 new_node4(struct f_trie *t, uint plen, uint local, ip4_addr paddr, ip4_addr pmask, ip4_addr amask)
156 {
157 struct f_trie_node4 *n = lp_allocz(t->lp, sizeof(struct f_trie_node4) + t->data_size);
158 n->plen = plen;
159 n->local = local;
160 n->addr = paddr;
161 n->mask = pmask;
162 n->accept = amask;
163 return n;
164 }
165
166 static inline struct f_trie_node6 *
167 new_node6(struct f_trie *t, uint plen, uint local, ip6_addr paddr, ip6_addr pmask, ip6_addr amask)
168 {
169 struct f_trie_node6 *n = lp_allocz(t->lp, sizeof(struct f_trie_node6) + t->data_size);
170 n->plen = plen;
171 n->local = local;
172 n->addr = paddr;
173 n->mask = pmask;
174 n->accept = amask;
175 return n;
176 }
177
178 static inline struct f_trie_node *
179 new_node(struct f_trie *t, uint plen, uint local, ip_addr paddr, ip_addr pmask, ip_addr amask)
180 {
181 if (t->ipv4)
182 return (struct f_trie_node *) new_node4(t, plen, local, ipt_to_ip4(paddr), ipt_to_ip4(pmask), ipt_to_ip4(amask));
183 else
184 return (struct f_trie_node *) new_node6(t, plen, local, ipa_to_ip6(paddr), ipa_to_ip6(pmask), ipa_to_ip6(amask));
185 }
186
187 static inline void
188 attach_node4(struct f_trie_node4 *parent, struct f_trie_node4 *child)
189 {
190 parent->c[ip4_getbits(child->addr, parent->plen, TRIE_STEP)] = child;
191 }
192
193 static inline void
194 attach_node6(struct f_trie_node6 *parent, struct f_trie_node6 *child)
195 {
196 parent->c[ip6_getbits(child->addr, parent->plen, TRIE_STEP)] = child;
197 }
198
199 static inline void
200 attach_node(struct f_trie_node *parent, struct f_trie_node *child, int v4)
201 {
202 if (v4)
203 attach_node4(&parent->v4, &child->v4);
204 else
205 attach_node6(&parent->v6, &child->v6);
206 }
207
208
209 /*
210 * Internal prefixes of a node a represented by the local bitmask, each bit for
211 * one prefix. Bit 0 is unused, Bit 1 is for the main prefix of the node,
212 * remaining bits correspond to subprefixes by this pattern:
213 *
214 * 1
215 * 2 3
216 * 4 5 6 7
217 * 8 9 A B C D E F
218 *
219 * E.g. for 10.0.0.0/8 node, the 10.64.0.0/10 would be position 5.
220 */
221
222 /*
223 * Compute appropriate mask representing prefix px/plen in local bitmask of node
224 * with prefix length nlen. Assuming that nlen <= plen < (nlen + TRIE_STEP).
225 */
226 static inline uint
227 trie_local_mask4(ip4_addr px, uint plen, uint nlen)
228 {
229 uint step = plen - nlen;
230 uint pos = (1u << step) + ip4_getbits(px, nlen, step);
231 return 1u << pos;
232 }
233
234 static inline uint
235 trie_local_mask6(ip6_addr px, uint plen, uint nlen)
236 {
237 uint step = plen - nlen;
238 uint pos = (1u << step) + ip6_getbits(px, nlen, step);
239 return 1u << pos;
240 }
241
242 /*
243 * Compute an appropriate local mask (for a node with prefix length nlen)
244 * representing prefixes of px that are accepted by amask and fall within the
245 * range associated with that node. Used for propagation of child accept mask
246 * to parent local mask.
247 */
248 static inline uint
249 trie_amask_to_local(ip_addr px, ip_addr amask, uint nlen)
250 {
251 uint local = 0;
252
253 for (uint plen = MAX(nlen, 1); plen < (nlen + TRIE_STEP); plen++)
254 if (ipa_getbit(amask, plen - 1))
255 local |= trie_local_mask(px, plen, nlen);
256
257 return local;
258 }
259
260 /*
261 * Compute a bitmask representing a level of subprefixes (of the same length),
262 * using specified position as a root. E.g., level 2 from root position 3 would
263 * be bit positions C-F, returned as bitmask 0xf000.
264 */
265 static inline uint
266 trie_level_mask(uint pos, uint level)
267 {
268 return ((1u << (1u << level)) - 1) << (pos << level);
269 }
270
271
272 #define GET_ADDR(N,F,X) ((X) ? ipt_from_ip4((N)->v4.F) : ipa_from_ip6((N)->v6.F))
273 #define SET_ADDR(N,F,X,V) ({ if (X) (N)->v4.F =ipt_to_ip4(V); else (N)->v6.F =ipa_to_ip6(V); })
274
275 #define GET_LOCAL(N,X) ((X) ? (N)->v4.local : (N)->v6.local)
276 #define ADD_LOCAL(N,X,V) ({ uint v_ = (V); if (X) (N)->v4.local |= v_; else (N)->v6.local |= v_; })
277
278 #define GET_CHILD(N,X,I) ((X) ? (struct f_trie_node *) (N)->v4.c[I] : (struct f_trie_node *) (N)->v6.c[I])
279
280
281 static void *
282 trie_add_node(struct f_trie *t, uint plen, ip_addr px, uint local, uint l, uint h)
283 {
284 uint l_ = l ? (l - 1) : 0;
285 ip_addr amask = (l_ < h) ? ipa_xor(ipa_mkmask(l_), ipa_mkmask(h)) : IPA_NONE;
286 ip_addr pmask = ipa_mkmask(plen);
287 ip_addr paddr = ipa_and(px, pmask);
288 struct f_trie_node *o = NULL;
289 struct f_trie_node *n = &t->root;
290 int v4 = t->ipv4;
291
292 /* Add all bits for each active level (0x0002 0x000c 0x00f0 0xff00) */
293 for (uint i = 0; i < TRIE_STEP; i++)
294 if ((l <= (plen + i)) && ((plen + i) <= h))
295 local |= trie_level_mask(1, i);
296
297 DBG("Insert node %I/%u (%I %x)\n", paddr, plen, amask, local);
298 while (n)
299 {
300 ip_addr naddr = GET_ADDR(n, addr, v4);
301 ip_addr nmask = GET_ADDR(n, mask, v4);
302 ip_addr accept = GET_ADDR(n, accept, v4);
303 ip_addr cmask = ipa_and(nmask, pmask);
304 uint nlen = v4 ? n->v4.plen : n->v6.plen;
305
306 DBG("Found node %I/%u (%I %x)\n",
307 naddr, nlen, accept, v4 ? n->v4.local : n->v6.local);
308
309 if (ipa_compare(ipa_and(paddr, cmask), ipa_and(naddr, cmask)))
310 {
311 /* We are out of path - we have to add branching node 'b'
312 between node 'o' and node 'n', and attach new node 'a'
313 as the other child of 'b'. */
314 int blen = ROUND_DOWN_POW2(ipa_pxlen(paddr, naddr), TRIE_STEP);
315 ip_addr bmask = ipa_mkmask(blen);
316 ip_addr baddr = ipa_and(px, bmask);
317
318 /* Merge accept masks from children to get accept mask for node 'b' */
319 ip_addr baccm = ipa_and(ipa_or(amask, accept), bmask);
320 uint bloc = trie_amask_to_local(naddr, accept, blen) |
321 trie_amask_to_local(paddr, amask, blen);
322
323 struct f_trie_node *a = new_node(t, plen, local, paddr, pmask, amask);
324 struct f_trie_node *b = new_node(t, blen, bloc, baddr, bmask, baccm);
325 attach_node(o, b, v4);
326 attach_node(b, n, v4);
327 attach_node(b, a, v4);
328 t->prefix_count++;
329
330 DBG("Case 1\n");
331 return a;
332 }
333
334 if (plen < nlen)
335 {
336 /* We add new node 'a' between node 'o' and node 'n' */
337 amask = ipa_or(amask, ipa_and(accept, pmask));
338 local |= trie_amask_to_local(naddr, accept, plen);
339 struct f_trie_node *a = new_node(t, plen, local, paddr, pmask, amask);
340 attach_node(o, a, v4);
341 attach_node(a, n, v4);
342 t->prefix_count++;
343
344 DBG("Case 2\n");
345 return a;
346 }
347
348 if (plen == nlen)
349 {
350 /* We already found added node in trie. Just update accept and local mask */
351 accept = ipa_or(accept, amask);
352 SET_ADDR(n, accept, v4, accept);
353
354 if ((GET_LOCAL(n, v4) & local) != local)
355 t->prefix_count++;
356
357 ADD_LOCAL(n, v4, local);
358
359 DBG("Case 3\n");
360 return n;
361 }
362
363 /* Update accept mask part M2 and go deeper */
364 accept = ipa_or(accept, ipa_and(amask, nmask));
365 SET_ADDR(n, accept, v4, accept);
366 ADD_LOCAL(n, v4, trie_amask_to_local(paddr, amask, nlen));
367
368 DBG("Step %u\n", ipa_getbits(paddr, nlen));
369
370 /* n->plen < plen and plen <= 32 (128) */
371 o = n;
372 n = GET_CHILD(n, v4, ipa_getbits(paddr, nlen, TRIE_STEP));
373 }
374
375 /* We add new tail node 'a' after node 'o' */
376 struct f_trie_node *a = new_node(t, plen, local, paddr, pmask, amask);
377 attach_node(o, a, v4);
378 t->prefix_count++;
379
380 DBG("Case 4\n");
381 return a;
382 }
383
384 /**
385 * trie_add_prefix
386 * @t: trie to add to
387 * @net: IP network prefix
388 * @l: prefix lower bound
389 * @h: prefix upper bound
390 *
391 * Adds prefix (prefix pattern) @n to trie @t. @l and @h are lower
392 * and upper bounds on accepted prefix lengths, both inclusive.
393 * 0 <= l, h <= 32 (128 for IPv6).
394 *
395 * Returns a pointer to the allocated node. The function can return a pointer to
396 * an existing node if @px and @plen are the same. If px/plen == 0/0 (or ::/0),
397 * a pointer to the root node is returned. Returns NULL when called with
398 * mismatched IPv4/IPv6 net type.
399 */
400 void *
401 trie_add_prefix(struct f_trie *t, const net_addr *net, uint l, uint h)
402 {
403 uint plen = net_pxlen(net);
404 ip_addr px;
405 int v4;
406
407 switch (net->type)
408 {
409 case NET_IP4:
410 case NET_VPN4:
411 case NET_ROA4:
412 px = ipt_from_ip4(net4_prefix(net));
413 v4 = 1;
414 break;
415
416 case NET_IP6:
417 case NET_VPN6:
418 case NET_ROA6:
419 case NET_IP6_SADR:
420 px = ipa_from_ip6(net6_prefix(net));
421 v4 = 0;
422 break;
423
424 default:
425 bug("invalid type");
426 }
427
428 if (t->ipv4 != v4)
429 {
430 if (t->ipv4 < 0)
431 t->ipv4 = v4;
432 else
433 return NULL;
434 }
435
436 DBG("\nInsert net %N (%u-%u)\n", net, l, h);
437
438 if (l == 0)
439 t->zero = 1;
440
441 if (h < plen)
442 plen = h;
443
444 /* Primary node length, plen rounded down */
445 uint nlen = ROUND_DOWN_POW2(plen, TRIE_STEP);
446
447 if (plen == nlen)
448 return trie_add_node(t, nlen, px, 0, l, h);
449
450 /* Secondary node length, plen rouned up */
451 uint slen = nlen + TRIE_STEP;
452 void *node = NULL;
453
454 /*
455 * For unaligned prefix lengths it is more complicated. We need to encode
456 * matching prefixes of lengths from l to h. There are three cases of lengths:
457 *
458 * 1) 0..nlen are encoded by the accept mask of the primary node
459 * 2) nlen..(slen-1) are encoded by the local mask of the primary node
460 * 3) slen..max are encoded in secondary nodes
461 */
462
463 if (l < slen)
464 {
465 uint local = 0;
466
467 /* Compute local bits for accepted nlen..(slen-1) prefixes */
468 for (uint i = 0; i < TRIE_STEP; i++)
469 if ((l <= (nlen + i)) && ((nlen + i) <= h))
470 {
471 uint pos = (1u << i) + ipa_getbits(px, nlen, i);
472 uint len = ((nlen + i) <= plen) ? 1 : (1u << (nlen + i - plen));
473
474 /* We need to fill 'len' bits starting at 'pos' position */
475 local |= ((1u << len) - 1) << pos;
476 }
477
478 /* Add the primary node */
479 node = trie_add_node(t, nlen, px, local, l, nlen);
480 }
481
482 if (slen <= h)
483 {
484 uint l2 = MAX(l, slen);
485 uint max = (1u << (slen - plen));
486
487 /* Add secondary nodes */
488 for (uint i = 0; i < max; i++)
489 node = trie_add_node(t, slen, ipa_setbits(px, slen - 1, i), 0, l2, h);
490 }
491
492 return node;
493 }
494
495
496 static int
497 trie_match_net4(const struct f_trie *t, ip4_addr px, uint plen)
498 {
499 if (plen == 0)
500 return t->zero;
501
502 int plentest = plen - 1;
503 uint nlen = ROUND_DOWN_POW2(plen, TRIE_STEP);
504 uint local = trie_local_mask4(px, plen, nlen);
505 const struct f_trie_node4 *n = &t->root.v4;
506
507 while (n)
508 {
509 /* We are out of path */
510 if (!ip4_prefix_equal(px, n->addr, MIN(plen, n->plen)))
511 return 0;
512
513 /* Check local mask */
514 if ((n->plen == nlen) && (n->local & local))
515 return 1;
516
517 /* Check accept mask */
518 if (ip4_getbit(n->accept, plentest))
519 return 1;
520
521 /* We finished trie walk and still no match */
522 if (nlen <= n->plen)
523 return 0;
524
525 /* Choose children */
526 n = n->c[ip4_getbits(px, n->plen, TRIE_STEP)];
527 }
528
529 return 0;
530 }
531
532 static int
533 trie_match_net6(const struct f_trie *t, ip6_addr px, uint plen)
534 {
535 if (plen == 0)
536 return t->zero;
537
538 int plentest = plen - 1;
539 uint nlen = ROUND_DOWN_POW2(plen, TRIE_STEP);
540 uint local = trie_local_mask6(px, plen, nlen);
541 const struct f_trie_node6 *n = &t->root.v6;
542
543 while (n)
544 {
545 /* We are out of path */
546 if (!ip6_prefix_equal(px, n->addr, MIN(plen, n->plen)))
547 return 0;
548
549 /* Check local mask */
550 if ((n->plen == nlen) && (n->local & local))
551 return 1;
552
553 /* Check accept mask */
554 if (ip6_getbit(n->accept, plentest))
555 return 1;
556
557 /* We finished trie walk and still no match */
558 if (nlen <= n->plen)
559 return 0;
560
561 /* Choose children */
562 n = n->c[ip6_getbits(px, n->plen, TRIE_STEP)];
563 }
564
565 return 0;
566 }
567
568 /**
569 * trie_match_net
570 * @t: trie
571 * @n: net address
572 *
573 * Tries to find a matching net in the trie such that
574 * prefix @n matches that prefix pattern. Returns 1 if there
575 * is such prefix pattern in the trie.
576 */
577 int
578 trie_match_net(const struct f_trie *t, const net_addr *n)
579 {
580 switch (n->type)
581 {
582 case NET_IP4:
583 case NET_VPN4:
584 case NET_ROA4:
585 return t->ipv4 ? trie_match_net4(t, net4_prefix(n), net_pxlen(n)) : 0;
586
587 case NET_IP6:
588 case NET_VPN6:
589 case NET_ROA6:
590 return !t->ipv4 ? trie_match_net6(t, net6_prefix(n), net_pxlen(n)) : 0;
591
592 default:
593 return 0;
594 }
595 }
596
597
598 /**
599 * trie_match_longest_ip4
600 * @t: trie
601 * @net: net address
602 * @dst: return value
603 * @found0: optional returned bitmask of found nodes
604 *
605 * Perform longest prefix match for the address @net and return the resulting
606 * prefix in the buffer @dst. The bitmask @found0 is used to report lengths of
607 * prefixes on the path from the root to the resulting prefix. E.g., if there is
608 * also a /20 shorter matching prefix, then 20-th bit is set in @found0. This
609 * can be used to enumerate all matching prefixes for the network @net using
610 * function trie_match_next_longest_ip4() or macro TRIE_WALK_TO_ROOT_IP4().
611 *
612 * This function assumes IPv4 trie, there is also an IPv6 variant. The @net
613 * argument is typed as net_addr_ip4, but would accept any IPv4-based net_addr,
614 * like net4_prefix(). Anyway, returned @dst is always net_addr_ip4.
615 *
616 * Result: 1 if a matching prefix was found, 0 if not.
617 */
618 int
619 trie_match_longest_ip4(const struct f_trie *t, const net_addr_ip4 *net, net_addr_ip4 *dst, ip4_addr *found0)
620 {
621 ASSERT(t->ipv4);
622
623 const ip4_addr prefix = net->prefix;
624 const int pxlen = net->pxlen;
625
626 const struct f_trie_node4 *n = &t->root.v4;
627 int len = 0;
628
629 ip4_addr found = IP4_NONE;
630 int last = -1;
631
632 while (n)
633 {
634 /* We are out of path */
635 if (!ip4_prefix_equal(prefix, n->addr, MIN(pxlen, n->plen)))
636 goto done;
637
638 /* Check accept mask */
639 for (; len < n->plen; len++)
640 {
641 if (len > pxlen)
642 goto done;
643
644 if (ip4_getbit(n->accept, len - 1))
645 {
646 /* len is always < 32 due to len < n->plen */
647 ip4_setbit(&found, len);
648 last = len;
649 }
650 }
651
652 /* Special case for max length, there is only one valid local position */
653 if (len == IP4_MAX_PREFIX_LENGTH)
654 {
655 if (n->local & (1u << 1))
656 last = len;
657
658 goto done;
659 }
660
661 /* Check local mask */
662 for (int pos = 1; pos < (1 << TRIE_STEP); pos = 2 * pos + ip4_getbit(prefix, len), len++)
663 {
664 if (len > pxlen)
665 goto done;
666
667 if (n->local & (1u << pos))
668 {
669 /* len is always < 32 due to special case above */
670 ip4_setbit(&found, len);
671 last = len;
672 }
673 }
674
675 /* Choose child */
676 n = n->c[ip4_getbits(prefix, n->plen, TRIE_STEP)];
677 }
678
679 done:
680 if (last < 0)
681 return 0;
682
683 *dst = NET_ADDR_IP4(ip4_and(prefix, ip4_mkmask(last)), last);
684
685 if (found0)
686 *found0 = found;
687
688 return 1;
689 }
690
691
692 /**
693 * trie_match_longest_ip6
694 * @t: trie
695 * @net: net address
696 * @dst: return value
697 * @found0: optional returned bitmask of found nodes
698 *
699 * Perform longest prefix match for the address @net and return the resulting
700 * prefix in the buffer @dst. The bitmask @found0 is used to report lengths of
701 * prefixes on the path from the root to the resulting prefix. E.g., if there is
702 * also a /20 shorter matching prefix, then 20-th bit is set in @found0. This
703 * can be used to enumerate all matching prefixes for the network @net using
704 * function trie_match_next_longest_ip6() or macro TRIE_WALK_TO_ROOT_IP6().
705 *
706 * This function assumes IPv6 trie, there is also an IPv4 variant. The @net
707 * argument is typed as net_addr_ip6, but would accept any IPv6-based net_addr,
708 * like net6_prefix(). Anyway, returned @dst is always net_addr_ip6.
709 *
710 * Result: 1 if a matching prefix was found, 0 if not.
711 */
712 int
713 trie_match_longest_ip6(const struct f_trie *t, const net_addr_ip6 *net, net_addr_ip6 *dst, ip6_addr *found0)
714 {
715 ASSERT(!t->ipv4);
716
717 const ip6_addr prefix = net->prefix;
718 const int pxlen = net->pxlen;
719
720 const struct f_trie_node6 *n = &t->root.v6;
721 int len = 0;
722
723 ip6_addr found = IP6_NONE;
724 int last = -1;
725
726 while (n)
727 {
728 /* We are out of path */
729 if (!ip6_prefix_equal(prefix, n->addr, MIN(pxlen, n->plen)))
730 goto done;
731
732 /* Check accept mask */
733 for (; len < n->plen; len++)
734 {
735 if (len > pxlen)
736 goto done;
737
738 if (ip6_getbit(n->accept, len - 1))
739 {
740 /* len is always < 128 due to len < n->plen */
741 ip6_setbit(&found, len);
742 last = len;
743 }
744 }
745
746 /* Special case for max length, there is only one valid local position */
747 if (len == IP6_MAX_PREFIX_LENGTH)
748 {
749 if (n->local & (1u << 1))
750 last = len;
751
752 goto done;
753 }
754
755 /* Check local mask */
756 for (int pos = 1; pos < (1 << TRIE_STEP); pos = 2 * pos + ip6_getbit(prefix, len), len++)
757 {
758 if (len > pxlen)
759 goto done;
760
761 if (n->local & (1u << pos))
762 {
763 /* len is always < 128 due to special case above */
764 ip6_setbit(&found, len);
765 last = len;
766 }
767 }
768
769 /* Choose child */
770 n = n->c[ip6_getbits(prefix, n->plen, TRIE_STEP)];
771 }
772
773 done:
774 if (last < 0)
775 return 0;
776
777 *dst = NET_ADDR_IP6(ip6_and(prefix, ip6_mkmask(last)), last);
778
779 if (found0)
780 *found0 = found;
781
782 return 1;
783 }
784
785 #define SAME_PREFIX(A,B,X,L) ((X) ? ip4_prefix_equal((A)->v4.addr, net4_prefix(B), (L)) : ip6_prefix_equal((A)->v6.addr, net6_prefix(B), (L)))
786 #define GET_NET_BITS(N,X,A,B) ((X) ? ip4_getbits(net4_prefix(N), (A), (B)) : ip6_getbits(net6_prefix(N), (A), (B)))
787
788 /**
789 * trie_walk_init
790 * @s: walk state
791 * @t: trie
792 * @net: optional subnet for walk
793 *
794 * Initialize walk state for subsequent walk through nodes of the trie @t by
795 * trie_walk_next(). The argument @net allows to restrict walk to given subnet,
796 * otherwise full walk over all nodes is used. This is done by finding node at
797 * or below @net and starting position in it.
798 */
799 void
800 trie_walk_init(struct f_trie_walk_state *s, const struct f_trie *t, const net_addr *net)
801 {
802 *s = (struct f_trie_walk_state) {
803 .ipv4 = t->ipv4,
804 .accept_length = 0,
805 .start_pos = 1,
806 .local_pos = 1,
807 .stack_pos = 0,
808 .stack[0] = &t->root
809 };
810
811 if (!net)
812 return;
813
814 /* We want to find node of level at least plen */
815 int plen = ROUND_DOWN_POW2(net->pxlen, TRIE_STEP);
816 const struct f_trie_node *n = &t->root;
817 const int v4 = t->ipv4;
818
819 while (n)
820 {
821 int nlen = v4 ? n->v4.plen : n->v6.plen;
822
823 /* We are out of path */
824 if (!SAME_PREFIX(n, net, v4, MIN(net->pxlen, nlen)))
825 break;
826
827 /* We found final node */
828 if (nlen >= plen)
829 {
830 if (nlen == plen)
831 {
832 /* Find proper local_pos, while accept_length is not used */
833 int step = net->pxlen - plen;
834 s->start_pos = s->local_pos = (1u << step) + GET_NET_BITS(net, v4, plen, step);
835 s->accept_length = plen;
836 }
837 else
838 {
839 /* Start from pos 1 in local node, but first try accept mask */
840 s->accept_length = net->pxlen;
841 }
842
843 s->stack[0] = n;
844 return;
845 }
846
847 /* Choose child */
848 n = GET_CHILD(n, v4, GET_NET_BITS(net, v4, nlen, TRIE_STEP));
849 }
850
851 s->stack[0] = NULL;
852 return;
853 }
854
855 #define GET_ACCEPT_BIT(N,X,B) ((X) ? ip4_getbit((N)->v4.accept, (B)) : ip6_getbit((N)->v6.accept, (B)))
856 #define GET_LOCAL_BIT(N,X,B) (((X) ? (N)->v4.local : (N)->v6.local) & (1u << (B)))
857
858 /**
859 * trie_walk_next
860 * @s: walk state
861 * @net: return value
862 *
863 * Find the next prefix in the trie walk and return it in the buffer @net.
864 * Prefixes are walked in the usual lexicographic order and may be restricted
865 * to a subset of the trie during walk setup by trie_walk_init(). Note that the
866 * trie walk does not iterate reliably over 'implicit' prefixes defined by &low
867 * and &high fields in prefix patterns, it is supposed to be used on tries
868 * constructed from 'explicit' prefixes (&low == &plen == &high in call to
869 * trie_add_prefix()).
870 *
871 * Result: 1 if the next prefix was found, 0 for the end of walk.
872 */
873 int
874 trie_walk_next(struct f_trie_walk_state *s, net_addr *net)
875 {
876 const struct f_trie_node *n = s->stack[s->stack_pos];
877 int len = s->accept_length;
878 int pos = s->local_pos;
879 int v4 = s->ipv4;
880
881 /*
882 * The walk has three basic state variables -- n, len and pos. In each node n,
883 * we first walk superprefixes (by len in &accept bitmask), and then we walk
884 * internal positions (by pos in &local bitmask). These positions are:
885 *
886 * 1
887 * 2 3
888 * 4 5 6 7
889 * 8 9 A B C D E F
890 *
891 * We walk them depth-first, including virtual positions 10-1F that are
892 * equivalent of position 1 in child nodes 0-F.
893 */
894
895 if (!n)
896 {
897 memset(net, 0, v4 ? sizeof(net_addr_ip4) : sizeof(net_addr_ip6));
898 return 0;
899 }
900
901 next_node:;
902 /* Current node prefix length */
903 int nlen = v4 ? n->v4.plen : n->v6.plen;
904
905 /* First, check for accept prefix */
906 for (; len < nlen; len++)
907 if (GET_ACCEPT_BIT(n, v4, len - 1))
908 {
909 if (v4)
910 net_fill_ip4(net, ip4_and(n->v4.addr, ip4_mkmask(len)), len);
911 else
912 net_fill_ip6(net, ip6_and(n->v6.addr, ip6_mkmask(len)), len);
913
914 s->local_pos = pos;
915 s->accept_length = len + 1;
916 return 1;
917 }
918
919 next_pos:
920 /* Bottom of this node */
921 if (pos >= (1 << TRIE_STEP))
922 {
923 const struct f_trie_node *child = GET_CHILD(n, v4, pos - (1 << TRIE_STEP));
924 int dir = 0;
925
926 /* No child node */
927 if (!child)
928 {
929 /* Step up until return from left child (pos is even) */
930 do
931 {
932 /* Step up from start node */
933 if ((s->stack_pos == 0) && (pos == s->start_pos))
934 {
935 s->stack[0] = NULL;
936 memset(net, 0, v4 ? sizeof(net_addr_ip4) : sizeof(net_addr_ip6));
937 return 0;
938 }
939
940 /* Top of this node */
941 if (pos == 1)
942 {
943 ASSERT(s->stack_pos);
944 const struct f_trie_node *old = n;
945
946 /* Move to parent node */
947 s->stack_pos--;
948 n = s->stack[s->stack_pos];
949 nlen = v4 ? n->v4.plen : n->v6.plen;
950
951 pos = v4 ?
952 ip4_getbits(old->v4.addr, nlen, TRIE_STEP) :
953 ip6_getbits(old->v6.addr, nlen, TRIE_STEP);
954 pos += (1 << TRIE_STEP);
955 len = nlen;
956
957 ASSERT(GET_CHILD(n, v4, pos - (1 << TRIE_STEP)) == old);
958 }
959
960 /* Step up */
961 dir = pos % 2;
962 pos = pos / 2;
963 }
964 while (dir);
965
966 /* Continue with step down to the right child */
967 pos = 2 * pos + 1;
968 goto next_pos;
969 }
970
971 /* Move to child node */
972 pos = 1;
973 len = nlen + TRIE_STEP;
974
975 s->stack_pos++;
976 n = s->stack[s->stack_pos] = child;
977 goto next_node;
978 }
979
980 /* Check for local prefix */
981 if (GET_LOCAL_BIT(n, v4, pos))
982 {
983 /* Convert pos to address of local network */
984 int x = (pos >= 2) + (pos >= 4) + (pos >= 8);
985 int y = pos & ((1u << x) - 1);
986
987 if (v4)
988 net_fill_ip4(net, !x ? n->v4.addr : ip4_setbits(n->v4.addr, nlen + x - 1, y), nlen + x);
989 else
990 net_fill_ip6(net, !x ? n->v6.addr : ip6_setbits(n->v6.addr, nlen + x - 1, y), nlen + x);
991
992 s->local_pos = 2 * pos;
993 s->accept_length = len;
994 return 1;
995 }
996
997 /* Step down */
998 pos = 2 * pos;
999 goto next_pos;
1000 }
1001
1002
1003 static int
1004 trie_node_same4(const struct f_trie_node4 *t1, const struct f_trie_node4 *t2)
1005 {
1006 if ((t1 == NULL) && (t2 == NULL))
1007 return 1;
1008
1009 if ((t1 == NULL) || (t2 == NULL))
1010 return 0;
1011
1012 if ((t1->plen != t2->plen) ||
1013 (! ip4_equal(t1->addr, t2->addr)) ||
1014 (! ip4_equal(t1->accept, t2->accept)))
1015 return 0;
1016
1017 for (uint i = 0; i < (1 << TRIE_STEP); i++)
1018 if (! trie_node_same4(t1->c[i], t2->c[i]))
1019 return 0;
1020
1021 return 1;
1022 }
1023
1024 static int
1025 trie_node_same6(const struct f_trie_node6 *t1, const struct f_trie_node6 *t2)
1026 {
1027 if ((t1 == NULL) && (t2 == NULL))
1028 return 1;
1029
1030 if ((t1 == NULL) || (t2 == NULL))
1031 return 0;
1032
1033 if ((t1->plen != t2->plen) ||
1034 (! ip6_equal(t1->addr, t2->addr)) ||
1035 (! ip6_equal(t1->accept, t2->accept)))
1036 return 0;
1037
1038 for (uint i = 0; i < (1 << TRIE_STEP); i++)
1039 if (! trie_node_same6(t1->c[i], t2->c[i]))
1040 return 0;
1041
1042 return 1;
1043 }
1044
1045 /**
1046 * trie_same
1047 * @t1: first trie to be compared
1048 * @t2: second one
1049 *
1050 * Compares two tries and returns 1 if they are same
1051 */
1052 int
1053 trie_same(const struct f_trie *t1, const struct f_trie *t2)
1054 {
1055 if ((t1->zero != t2->zero) || (t1->ipv4 != t2->ipv4))
1056 return 0;
1057
1058 if (t1->ipv4)
1059 return trie_node_same4(&t1->root.v4, &t2->root.v4);
1060 else
1061 return trie_node_same6(&t1->root.v6, &t2->root.v6);
1062 }
1063
1064
1065 static const u8 log2[16] = {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3};
1066
1067 static void
1068 trie_node_format(const struct f_trie_node *n, buffer *buf, int v4)
1069 {
1070 if (n == NULL)
1071 return;
1072
1073 if (v4)
1074 {
1075 if (ip4_nonzero(n->v4.accept))
1076 buffer_print(buf, "%I4/%d{%I4}, ", n->v4.addr, n->v4.plen, n->v4.accept);
1077 }
1078 else
1079 {
1080 if (ip6_nonzero(n->v6.accept))
1081 buffer_print(buf, "%I6/%d{%I6}, ", n->v6.addr, n->v6.plen, n->v6.accept);
1082 }
1083
1084 int nlen = v4 ? n->v4.plen : n->v6.plen;
1085 uint local = v4 ? n->v4.local : n->v6.local;
1086
1087 for (int i = (nlen ? 0 : 1); i < TRIE_STEP; i++)
1088 if (GET_ACCEPT_BIT(n, v4, nlen + i - 1))
1089 local &= ~trie_level_mask(1, i);
1090
1091 for (int pos = 2; local && (pos < (1 << TRIE_STEP)); pos++)
1092 if (local & (1u << pos))
1093 {
1094 int lvl = log2[pos];
1095 int plen = nlen + lvl;
1096
1097 int i;
1098 for (i = 0; lvl + i < TRIE_STEP; i++)
1099 {
1100 uint lmask = trie_level_mask(pos, i);
1101
1102 if ((local & lmask) != lmask)
1103 break;
1104
1105 local &= ~lmask;
1106 }
1107
1108 uint addr_bits = pos & ((1u << lvl) - 1);
1109 uint accept_bits = (1u << i) - 1;
1110 int h = plen + i - 1;
1111
1112 if (v4)
1113 {
1114 ip4_addr addr = ip4_setbits(n->v4.addr, plen - 1, addr_bits);
1115 ip4_addr mask = ip4_setbits(IP4_NONE, h - 1, accept_bits);
1116 buffer_print(buf, "%I4/%d{%I4}, ", addr, plen, mask);
1117 }
1118 else
1119 {
1120 ip6_addr addr = ip6_setbits(n->v6.addr, plen - 1, addr_bits);
1121 ip6_addr mask = ip6_setbits(IP6_NONE, h - 1, accept_bits);
1122 buffer_print(buf, "%I6/%d{%I6}, ", addr, plen, mask);
1123 }
1124 }
1125
1126 for (int i = 0; i < (1 << TRIE_STEP); i++)
1127 trie_node_format(GET_CHILD(n, v4, i), buf, v4);
1128 }
1129
1130 /**
1131 * trie_format
1132 * @t: trie to be formatted
1133 * @buf: destination buffer
1134 *
1135 * Prints the trie to the supplied buffer.
1136 */
1137 void
1138 trie_format(const struct f_trie *t, buffer *buf)
1139 {
1140 buffer_puts(buf, "[");
1141
1142 if (t->zero)
1143 buffer_print(buf, "%I/%d, ", t->ipv4 ? IPA_NONE4 : IPA_NONE6, 0);
1144
1145 trie_node_format(&t->root, buf, t->ipv4);
1146
1147 if (buf->pos == buf->end)
1148 return;
1149
1150 /* Undo last separator */
1151 if (buf->pos[-1] != '[')
1152 buf->pos -= 2;
1153
1154 buffer_puts(buf, "]");
1155 }