modules/rp/rp_search.c
/* [<][>][^][v][top][bottom][index][help] */
FUNCTIONS
This source file includes following functions.
- rp_exclude_datlink
- rp_preflist_search
- rp_find_smallest_span
- rp_leaf_occ_inc
- rp_exclude_exact_match
- rp_find_longest_prefix
- rp_asc_process_datlist
- rp_asc_append_datref
- rp_srch_copyresults
- rp_begend_preselection
- RP_asc_search
1 /***************************************
2 $Revision: 1.12 $
3
4 Radix payload (rp) - user level functions for storing data in radix trees
5
6 rp_search = search the loaded radix trees using an ascii key
7
8 Motto: "And all that for inetnums..."
9
10 Status: NOT REVIEWED, TESTED
11
12 Design and implementation by: Marek Bukowy
13
14 ******************/ /******************
15 Copyright (c) 1999 RIPE NCC
16
17 All Rights Reserved
18
19 Permission to use, copy, modify, and distribute this software and its
20 documentation for any purpose and without fee is hereby granted,
21 provided that the above copyright notice appear in all copies and that
22 both that copyright notice and this permission notice appear in
23 supporting documentation, and that the name of the author not be
24 used in advertising or publicity pertaining to distribution of the
25 software without specific, written prior permission.
26
27 THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING
28 ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS; IN NO EVENT SHALL
29 AUTHOR BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY
30 DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
31 AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
32 OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
33 ***************************************/
34
35
36 #include <rp.h>
37
38 static
39 void
40 rp_exclude_datlink(GList **datlist, GList *element)
/* [<][>][^][v][top][bottom][index][help] */
41 {
42 /* remove element from list(becomes a self-consistent list) */
43 *datlist = g_list_remove_link(*datlist, element);
44
45 /* free it and the payload */
46 wr_clear_list( &element );
47 }
48
49
50 /**************************************************************************/
51 /*+++++++++++
52 helper:
53 this routine goes through the list of prefixes and performs a bin_search
54 on each of them; attaches the results to datlist.
55 +++++++++++*/
56 static
57 er_ret_t
58 rp_preflist_search (
/* [<][>][^][v][top][bottom][index][help] */
59 rx_srch_mt search_mode,
60 int par_a,
61 int par_b,
62 rx_tree_t *mytree,
63 GList **preflist,
64 GList **datlist
65 )
66
67 {
68 char prefstr[IP_PREFSTR_MAX];
69 GList *qitem;
70 ip_prefix_t *querypref;
71 er_ret_t err;
72
73 for( qitem = g_list_first(*preflist);
74 qitem != NULL;
75 qitem = g_list_next(qitem)) {
76
77 querypref = qitem->data;
78
79 if( IP_pref_b2a( querypref, prefstr, IP_PREFSTR_MAX) != IP_OK ) {
80 die;
81 }
82 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
83 "rx_preflist_search: mode %d (%s) (par %d) for %s",
84 search_mode, RX_text_srch_mode(search_mode), par_a, prefstr);
85
86 if (mytree->num_nodes > 0) {
87 err = RX_bin_search( search_mode, par_a, par_b, mytree, querypref,
88 datlist, RX_ANS_ALL);
89 if( err != RX_OK ) {
90 return err;
91 }
92 }
93 }
94
95 return RX_OK;
96 }
97
98 /*++++
99 this is a helper: goes through a datlist and returns the smallest
100 size of a range
101
102 works for IPv4 only
103 +++*/
104 static
105 ip_rangesize_t
106 rp_find_smallest_span( GList *datlist ) {
/* [<][>][^][v][top][bottom][index][help] */
107 ip_rangesize_t min_span, span;
108 GList *ditem;
109
110 min_span = 0xffffffff; /* IPv4 only!!!!*/
111
112 /* go through the list and find the shortest range. */
113 for(ditem = g_list_first(datlist);
114 ditem != NULL;
115 ditem = g_list_next(ditem)) {
116 rx_datref_t *refptr = (rx_datref_t *) (ditem->data);
117
118 span = IP_rang_span( & refptr->leafptr->iprange);
119
120 if( span < min_span ) {
121 min_span = span;
122 }
123 }
124 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
125 "rp_find_smallest_span: minimal span is %d", min_span);
126
127 return min_span;
128 }
129
130
131
132 /* helper for the inetnum/exless search - for this one a hash of pairs
133 (leafptr,occurences) must be maintained.
134
135 This routine increments the counter for a leafptr, creating a new
136 pair if this leafptr was not referenced before.
137
138 */
139 static
140 int rp_leaf_occ_inc(GHashTable *hash, rx_dataleaf_t *leafptr)
/* [<][>][^][v][top][bottom][index][help] */
141 {
142 /* one little trick: store the number of occurences
143 as cast (void *) */
144 int val;
145
146 val = (int) g_hash_table_lookup(hash, leafptr);
147 /* 0 if it's not known yet. anyway: put it in the hash (value==key) */
148
149 g_hash_table_insert(hash, leafptr, (void *) ++val);
150
151 return val;
152 }
153
154 /* exclude exact match - not to be merged with preselction :-( */
155 static void
156 rp_exclude_exact_match( GList **datlist, ip_range_t *testrang)
/* [<][>][^][v][top][bottom][index][help] */
157 {
158 GList *ditem, *newitem;
159
160 ditem = g_list_first(*datlist);
161
162 while( ditem != NULL ) {
163 rx_datref_t *refptr = (rx_datref_t *) (ditem->data);
164
165 newitem = g_list_next(ditem);
166
167 if( memcmp( & refptr->leafptr->iprange,
168 testrang, sizeof(ip_range_t)) == 0 ) {
169 rp_exclude_datlink(datlist, ditem);
170 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
171 "process_datlist: discarded an exact match");
172 }
173 ditem = newitem;
174 } /* while */
175 }
176
177 static int
178 rp_find_longest_prefix(GList **datlist)
/* [<][>][^][v][top][bottom][index][help] */
179 {
180 GList *ditem;
181 int max_pref=0;
182
183 for(ditem = g_list_first(*datlist);
184 ditem != NULL;
185 ditem = g_list_next(ditem)) {
186 rx_datref_t *refptr = (rx_datref_t *) (ditem->data);
187
188 if( refptr->leafptr->preflen > max_pref ) {
189 max_pref = refptr->leafptr->preflen;
190 }
191 }
192
193 return max_pref;
194 }
195
196
197 /*+ rp_asc_process_datlist() - helper for RP_asc_search()
198
199 fetches the copies of objects from the radix tree into datlist
200
201 ASSUMES LOCKED TREE
202
203 the behaviour for a default inetnum (range) query is:
204 do an exact match;
205 if it fails, do an exless match on the encompassing prefix
206 for routes(prefixes):
207 do an exless match
208
209 So if it's the default search mode on an inetnum tree,
210 and the key is a range,
211 then an exact search is performed on one of the composing prefixes.
212
213 Then the resulting data leaves are checked for exact matching with
214 the range queried for.
215 Any dataleaves that do not match are discarded, and if none are left,
216 the procedure falls back to searching for the encompassing prefix.
217 (calculated in the smart_conv routine).
218 Add the dataleaf copies to the list of answers,
219 taking span into account
220 +*/
221 static
222 er_ret_t
223 rp_asc_process_datlist(
/* [<][>][^][v][top][bottom][index][help] */
224 rx_srch_mt search_mode,
225 int par_a,
226 rx_fam_t fam_id,
227 int prefnumber,
228 GList **datlist,
229 ip_range_t *testrang,
230 int *hits
231 )
232 {
233 ip_rangesize_t min_span=0, span;
234 int max_pref = -1;
235 GList *ditem, *newitem;
236 GHashTable *lohash = g_hash_table_new(NULL, NULL);
237
238 /* in MORE and LESS(1) search exact match must not be displayed */
239 if ( search_mode == RX_SRCH_MORE
240 || ( search_mode == RX_SRCH_LESS && par_a == 1 ) ) {
241 rp_exclude_exact_match(datlist, testrang);
242 }
243
244 /* Preselection moved to processing, only span calculation done here *
245 *
246
247 EXLESS and LESS(1) search: the smallest span must be found,
248 but if the less spec node is not the same for all composing prefixes,
249 it means it's not really this one.
250
251 we check that by the number of references to this node is less than
252 the number of composing prefixes
253
254 We do the same for the less specific search - a node must be less
255 specific to all prefixes.
256
257 if the number of references is not enough, then return no hits,
258 another try will be made, this time with one, encompassing prefix.
259 */
260
261 if ( (search_mode == RX_SRCH_EXLESS )
262 || ( search_mode == RX_SRCH_LESS && par_a == 1 ) ) {
263 /* span works only for IP_V4. We use it only for inetnums,
264 although RT/v4 would work too */
265 if( testrang->begin.space == IP_V4 &&
266 fam_id == RX_FAM_IN ) {
267 min_span = rp_find_smallest_span(*datlist);
268 }
269 else {
270 /* in IPv6 and RT trees in general, we can obtain the same
271 result by selecting the longest prefix */
272 max_pref = rp_find_longest_prefix(datlist);
273 }
274 }
275
276 /* Process the dataleaf copies and add to the list of answers. */
277 ditem = g_list_first(*datlist);
278 while(ditem != NULL) {
279 rx_datref_t *refptr = (rx_datref_t *) (ditem->data);
280 int exclude = 0;
281
282 if(search_mode == RX_SRCH_EXLESS || search_mode == RX_SRCH_LESS ) {
283
284 /* min_span defined <=> EXLESS or LESS(1) search of INETNUMS:
285 the smallest span must be returned */
286 if( !exclude && min_span != 0
287 && (span = IP_rang_span( &refptr->leafptr->iprange))!=min_span) {
288 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
289 "process_datlist: (EX)LESS: discarded object with span %d", span);
290 exclude = 1;
291 }
292 /* max_pref defined <=> EXLESS search of INETNUMS or LESS(1) of RT:
293 */
294 if( !exclude && max_pref >= 0
295 && refptr->leafptr->preflen < max_pref ) {
296 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
297 "process_datlist: (EX)LESS: discarded object with preflen %d",
298 refptr->leafptr->preflen);
299 exclude = 1;
300 }
301
302 /* number of occurences */
303 /* XXX this will go when the old algorithm goes */
304 if( !exclude
305 && prefnumber > 1 ) { /* do not check if all will be approved */
306
307 if( rp_leaf_occ_inc(lohash, refptr->leafptr) < prefnumber ) {
308 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
309 "process_datlist: (EX)LESS: leafptr %x not enough",refptr->leafptr);
310 exclude = 1;
311 }
312 else {
313 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
314 "process_datlist: (EX)LESS: leafptr %x GOOD enough",refptr->leafptr);
315 }
316 }
317 }
318 else if( search_mode == RX_SRCH_EXACT ) {
319 /* EXACT search - discard if the range does not match */
320 if( memcmp( & refptr->leafptr->iprange,
321 testrang, sizeof(ip_range_t)) != 0) {
322
323 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
324 "process_datlist: EXACT; discarded a mismatch");
325 exclude = 1;
326 } /* EXACT match */
327 }
328 else if( search_mode == RX_SRCH_MORE ) {
329 /* MORE: exclude if not fully contained in the search term */
330 if( ! (IP_addr_in_rang(&refptr->leafptr->iprange.begin, testrang )
331 && IP_addr_in_rang(&refptr->leafptr->iprange.end, testrang ))) {
332 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
333 "process_datlist: MORE; discarded a not-fully contained one");
334 exclude = 1;
335 }
336 }
337
338
339 /* get next item now, before the current gets deleted */
340 newitem = g_list_next(ditem);
341 if( exclude ) {
342 /* get rid of it */
343 rp_exclude_datlink(datlist, ditem);
344 }
345 else {
346 /* OK, so we ACCEPT these results*/
347 /* uniqueness ensured in copy_results */
348 (*hits)++;
349 }
350 ditem = newitem;
351 } /* while ditem */
352
353 /* wr_clear_list(&lolist); */
354 g_hash_table_destroy(lohash);
355 return RX_OK;
356 }
357
358 /**************************************************************************/
359
360 /*+ appends the element pointed to by datref to finallist +*/
361 static
362 er_ret_t
363 rp_asc_append_datref(rx_datref_t *refptr, GList **finallist)
/* [<][>][^][v][top][bottom][index][help] */
364 {
365 er_ret_t err;
366 rx_datcpy_t *datcpy;
367 void *dataptr;
368
369 /* OK, so we ACCEPT this result. Copy it.*/
370
371 if( (err=wr_calloc( (void **)& datcpy, 1, sizeof(rx_datcpy_t))) != UT_OK) {
372 return err; /* die;*/
373 }
374
375 datcpy->leafcpy = *(refptr->leafptr);
376
377 /* copy the immediate data too. Set the ptr.*/
378
379 if( (err=wr_calloc( (void **) & dataptr, 1, refptr->leafptr->data_len))
380 != UT_OK) {
381 return err; /* die;*/
382 }
383 memcpy(dataptr, refptr->leafptr->data_ptr, refptr->leafptr->data_len);
384
385 datcpy->leafcpy.data_ptr = dataptr;
386
387 *finallist = g_list_prepend(*finallist, datcpy);
388
389 /* XXX this wouldn't work in access_control */
390 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DATA,
391 "rp_asc_append 'ed: %s", dataptr);
392
393 return RX_OK;
394 }
395
396 /*+ goes through datlist (list of references "datref") and add copies of
397 leaves referenced to the finallist
398
399 maintains its own uniqhash which holds pointers to copied dataleaves.
400
401 modifies: finallist
402
403 returns: error from wr_malloc
404
405 +*/
406 static
407 er_ret_t
408 rp_srch_copyresults(GList *datlist,
/* [<][>][^][v][top][bottom][index][help] */
409 GList **finallist,
410 int maxcount)
411 {
412 er_ret_t err;
413 GList *ditem;
414 GHashTable *uniqhash = g_hash_table_new(NULL, NULL); /* defaults */
415 int count = 0;
416
417 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET, "srch_copyresults");
418
419 /* copy dataleaves pointed to by entries from the datlist
420 only once (check uniqueness in the hash table) */
421 for(ditem = g_list_first(datlist);
422 ditem != NULL;
423 ditem = g_list_next(ditem)) {
424 rx_datref_t *refptr = (rx_datref_t *) (ditem->data);
425 rx_dataleaf_t *ansptr = refptr->leafptr;
426
427 /* search for every ansptr (dataleaf pointer) in uniqhash */
428 if( g_hash_table_lookup(uniqhash, ansptr) == NULL ) {
429
430 /* it's not known yet. OK: put it in the hash (value==key) */
431 g_hash_table_insert(uniqhash, ansptr, ansptr);
432
433 /* and copy the dataleaf */
434 if( !NOERR(err = rp_asc_append_datref(refptr, finallist)) ) {
435 return err;
436 }
437 }
438
439 /* check the limit on number of objects if defined ( >0) */
440 count++;
441 if( maxcount > 0 && count > maxcount ) {
442 break;
443 }
444
445 } /* foreach (datlist) */
446
447 g_hash_table_destroy(uniqhash); /* elements are still linked to through datlist */
448
449 return RP_OK;
450 }
451
452 static
453 void
454 rp_begend_preselection(GList **datlist, rx_fam_t fam_id, ip_range_t *testrang)
/* [<][>][^][v][top][bottom][index][help] */
455 {
456 GList *ditem, *newitem;
457
458 ditem = g_list_first(*datlist);
459
460 while( ditem != NULL ) {
461 rx_datref_t *refptr = (rx_datref_t *) (ditem->data);
462 newitem = g_list_next(ditem);
463
464 /* the test is indentical for route & inetnum trees */
465 if( IP_addr_in_rang(&testrang->end, &refptr->leafptr->iprange) == 0 ) {
466
467 rp_exclude_datlink(datlist, ditem);
468 ER_dbg_va(FAC_RP, ASP_RP_SRCH_DET,
469 "process_datlist: discarded an uncovering leafptr %x",
470 refptr->leafptr);
471
472 }
473 ditem = newitem;
474 } /* while */
475 }
476
477 /*+++++++++++++++
478 search.
479
480 2 approaches:
481
482 1. (most modes): look up all less specifics of beginning and end of range,
483 compare/select/etc.
484
485 2. More spec mode: break up the query range into prefixes, [erform a search
486 for each of them. Add all results together.
487
488 translates a query into a binary prefix (or prefixes, if range).
489 for registry+space (or if they are zero, for all
490 registries/spaces)
491 finds tree
492 calls RX_bin_search (returning node copies).
493 will not put duplicate entries (composed inetnums).
494 returns some sort of error code :-)
495
496 Cuts the number of answers from RX_bin_search down to max_count,
497 but since some of the answers may have been "normalized" in the
498 underlying functions (multiple occurences removed),
499 the result is _at_most_ max_count.
500
501 appends to a given list of data blocks (not nodes!)
502
503 The EXLESS search on inetnum tree should return the shortest range
504 that was found, by means of comparing span (size) of the range.
505 If there are more of size equal to the smallest one, they are also
506 returned.
507
508 returns RX_OK or a code from an underlying function
509 ++++++++++++*/
510 er_ret_t
511 RP_asc_search (
/* [<][>][^][v][top][bottom][index][help] */
512 rx_srch_mt search_mode,
513 int par_a,
514 int par_b,
515 char *key, /*+ search term: (string) prefix/range/IP +*/
516 rp_regid_t reg_id,
517 rp_attr_t attr, /*+ extra tree id (within the same reg/spc/fam +*/
518 GList **finallist, /*+ answers go here, please +*/
519 int max_count /*+ max # of answers. RX_ALLANS == unlimited +*/
520 )
521 {
522 GList *preflist = NULL;
523 GList *datlist = NULL;
524 er_ret_t err;
525 ip_range_t testrang;
526 int locked = 0;
527 ip_keytype_t key_type;
528 ip_space_t spc_id;
529 rx_fam_t fam_id = RP_attr2fam( attr );
530 rx_tree_t *mytree;
531 int hits=0;
532 ip_prefix_t beginpref;
533
534
535 /* abort on error (but unlock the tree) */
536 ER_dbg_va(FAC_RP, ASP_RP_SRCH_GEN,
537 "RP_NEW_asc_search: query %s : mode %d (%s) (par %d) for %s",
538 DF_get_attribute_name(attr),
539 search_mode, RX_text_srch_mode(search_mode), par_a, key);
540
541
542 /* parse the key into a prefix list */
543 if( ( err = IP_smart_conv(key, 0, 0,
544 &preflist, IP_EXPN, &key_type)) != IP_OK ) {
545 /* operational trouble (UT_*) or invalid key (IP_INVARG)*/
546 return err;
547 }
548
549 /* set the test values */
550 IP_smart_range(key, &testrang, IP_EXPN, &key_type);
551
552 /* find the tree */
553 if( NOERR(err) ) {
554 spc_id = IP_pref_b2_space( g_list_first(preflist)->data );
555 if( ! NOERR(err = RP_tree_get( &mytree, reg_id, spc_id, attr ))) {
556 return err;
557 }
558 }
559 /* the point of no return: now we lock the tree. From here, even if errors
560 occur, we still go through all procedure to unlock the tree at the end */
561
562 /* lock the tree */
563 TH_acquire_read_lock( &(mytree->rwlock) );
564 locked = 1;
565
566 /* Collection: this procedure is used for some search_modes only */
567 if( search_mode == RX_SRCH_EXLESS
568 || search_mode == RX_SRCH_LESS
569 || search_mode == RX_SRCH_EXACT ) {
570
571 /* 1. compose a /32(/128) prefix for beginning of range */
572 beginpref.ip = testrang.begin;
573 beginpref.bits = IP_sizebits(spc_id);
574
575 /* 2. dataleaves collection: look up the beginning prefix in LESS(255) mode */
576 if( NOERR(err) ) {
577 err = RX_bin_search( RX_SRCH_LESS, 255, 0, mytree, &beginpref,
578 &datlist, RX_ANS_ALL);
579 }
580
581 /* 3. preselection: exclude those that do not include end of range
582 */
583 if( NOERR(err) ) {
584 rp_begend_preselection(&datlist, fam_id, &testrang);
585 }
586
587 } /* if exless|less|exact */
588 else {
589 /* MORE */
590
591 /* standard collection using the traditional method:
592 repeat the search for all prefixes and join results */
593
594 if( NOERR(err) ) {
595 err = rp_preflist_search ( search_mode, par_a, par_b,
596 mytree, &preflist, &datlist);
597 }
598 } /* collection */
599
600 ER_dbg_va(FAC_RP, ASP_RP_SRCH_GEN,
601 "RP_NEW_asc_search: collected %d references ",
602 g_list_length(datlist));
603
604
605 /* 5. processing - using the same processing function */
606 if( NOERR(err) ) {
607 err = rp_asc_process_datlist( search_mode, par_a, fam_id,
608 1, /* one occurence is enough */
609 &datlist,
610 &testrang, &hits );
611 }
612
613 /* 6. copy results */
614 if( NOERR(err) ) {
615 err = rp_srch_copyresults(datlist, finallist, max_count); /* and uniq */
616 }
617
618 if( locked ) {
619 /* 100. unlock the tree */
620 TH_release_read_lock( &(mytree->rwlock) );
621 }
622
623 /* clean up */
624 wr_clear_list( &preflist );
625 wr_clear_list( &datlist );
626
627 /* NOTE if error occured, finallist may be partly filled in. */
628 return err;
629 }
630