1 /* $Cambridge: exim/src/src/search.c,v 1.5 2007/01/08 10:50:18 ph10 Exp $ */
3 /*************************************************
4 * Exim - an Internet mail transport agent *
5 *************************************************/
7 /* Copyright (c) University of Cambridge 1995 - 2007 */
8 /* See the file NOTICE for conditions of use and distribution. */
10 /* A set of functions to search databases in various formats. An open
11 database is represented by a void * value which is returned from a lookup-
12 specific "open" function. These are now all held in individual modules in the
13 lookups subdirectory and the functions here form a generic interface.
15 Caching is used to improve performance. Open files are cached until a tidyup
16 function is called, and for each file the result of the last lookup is cached.
17 However, if too many files are opened, some of those that are not in use have
18 to be closed. Those open items that use real files are kept on a LRU chain to
21 All the data is held in permanent store so as to be independent of the stacking
22 pool that is reset from time to time. In fact, we use malloc'd store so that it
23 can be freed when the caches are tidied up. It isn't actually clear whether
24 this is a benefit or not, to be honest. */
29 /* Tree in which to cache open files until tidyup called. */
31 static tree_node *search_tree = NULL;
33 /* Two-way chain of open databases that use real files. This is maintained in
34 recently-used order for the purposes of closing the least recently used when
35 too many files are open. */
37 static tree_node *open_top = NULL;
38 static tree_node *open_bot = NULL;
40 /* Count of open databases that use real files */
42 static int open_filecount = 0;
44 /* Allow us to reset store used for lookups and lookup caching */
46 static void *search_reset_point = NULL;
50 /*************************************************
51 * Validate a plain lookup type name *
52 *************************************************/
54 /* Only those names that are recognized and whose code is included in the
55 binary give an OK response. Use a binary chop search now that the list has got
59 name lookup type name - not necessarily zero terminated (e.g. dbm*)
60 len length of the name
62 Returns: +ve => valid lookup name; value is offset in lookup_list
63 -ve => invalid name; message in search_error_message.
67 search_findtype(uschar *name, int len)
70 int top = lookup_list_count;
73 int mid = (top + bot)/2;
74 int c = Ustrncmp(name, lookup_list[mid].name, len);
76 /* If c == 0 we have matched the incoming name with the start of the search
77 type name. However, some search types are substrings of others (e.g. nis and
78 nisplus) so we need to check that the lengths are the same. The length of the
79 type name cannot be shorter (else c would not be 0); if it is not equal it
80 must be longer, and in that case, the incoming name comes before the name we
81 are testing. By leaving c == 0 when the lengths are different, and doing a
82 > 0 test below, this all falls out correctly. */
84 if (c == 0 && Ustrlen(lookup_list[mid].name) == len)
86 if (lookup_list[mid].find != NULL) return mid;
87 search_error_message = string_sprintf("lookup type \"%.*s\" is not "
88 "available (not in the binary - check buildtime LOOKUP configuration)",
93 if (c > 0) bot = mid + 1; else top = mid;
96 search_error_message = string_sprintf("unknown lookup type \"%.*s\"",len,name);
102 /*************************************************
103 * Validate a full lookup type name *
104 *************************************************/
106 /* This function recognizes the "partial-" prefix and also terminating * and *@
110 name the full lookup type name
111 ptypeptr where to put the partial type
112 after subtraction of 1024 or 2048:
113 negative => no partial matching
114 non-negative => minimum number of non-wild components
115 ptypeaff where to put a pointer to the affix
116 the affix is within name if supplied therein
117 otherwise it's a literal string
118 afflen the length of the affix
119 starflags where to put the SEARCH_STAR and SEARCH_STARAT flags
121 Returns: +ve => valid lookup name; value is offset in lookup_list
122 -ve => invalid name; message in search_error_message.
126 search_findtype_partial(uschar *name, int *ptypeptr, uschar **ptypeaff,
127 int *afflen, int *starflags)
136 /* Check for a partial matching type. It must start with "partial", optionally
137 followed by a sequence of digits. If this is followed by "-", the affix is the
138 default "*." string. Otherwise we expect an affix in parentheses. Affixes are a
139 limited number of characters, not including parens. */
141 if (Ustrncmp(name, "partial", 7) == 0)
147 while (isdigit(*ss)) pv = pv*10 + *ss++ - '0';
149 else pv = 2; /* Default number of wild components */
154 while (ispunct(*ss) && *ss != ')') ss++;
155 if (*ss != ')') goto BAD_TYPE;
156 *afflen = ss++ - *ptypeaff;
158 else if (*ss++ == '-')
166 search_error_message = string_sprintf("format error in lookup type \"%s\"",
172 /* Now we are left with a lookup name, possibly followed by * or *@. */
175 if (len >= 2 && Ustrncmp(ss + len - 2, "*@", 2) == 0)
177 *starflags |= SEARCH_STARAT;
180 else if (len >= 1 && ss[len-1] == '*')
182 *starflags |= SEARCH_STAR;
186 /* Check for the individual search type. Only those that are actually in the
187 binary are valid. For query-style types, "partial" and default types are
190 stype = search_findtype(ss, len);
191 if (stype >= 0 && mac_islookup(stype, lookup_querystyle))
195 search_error_message = string_sprintf("\"partial\" is not permitted "
196 "for lookup type \"%s\"", ss);
199 if ((*starflags & (SEARCH_STAR|SEARCH_STARAT)) != 0)
201 search_error_message = string_sprintf("defaults using \"*\" or \"*@\" are "
202 "not permitted for lookup type \"%s\"", ss);
213 /*************************************************
214 * Release cached resources *
215 *************************************************/
217 /* When search_open is called it caches the "file" that it opens in
218 search_tree. The name of the tree node is a concatenation of the search type
219 with the file name. For query-style lookups, the file name is empty. Real files
220 are normally closed only when this tidyup routine is called, typically at the
221 end of sections of code where a number of lookups might occur. However, if too
222 many files are open simultaneously, some get closed beforehand. They can't be
223 removed from the tree. There is also a general tidyup function which is called
224 for the lookup driver, if it exists.
226 First, there is an internal, recursive subroutine.
228 Argument: a pointer to a search_openfile tree node
233 tidyup_subtree(tree_node *t)
235 search_cache *c = (search_cache *)(t->data.ptr);
236 if (t->left != NULL) tidyup_subtree(t->left);
237 if (t->right != NULL) tidyup_subtree(t->right);
240 lookup_list[c->search_type].close != NULL)
241 lookup_list[c->search_type].close(c->handle);
245 /* The external entry point
255 int old_pool = store_pool;
257 DEBUG(D_lookup) debug_printf("search_tidyup called\n");
259 /* Close individually each cached open file. */
261 store_pool = POOL_SEARCH;
262 if (search_tree != NULL)
264 tidyup_subtree(search_tree);
267 open_top = open_bot = NULL;
270 /* Call the general tidyup entry for any drivers that have one. */
272 for (i = 0; i < lookup_list_count; i++)
273 if (lookup_list[i].tidy != NULL) (lookup_list[i].tidy)();
275 if (search_reset_point != NULL) store_reset(search_reset_point);
276 search_reset_point = NULL;
277 store_pool = old_pool;
283 /*************************************************
284 * Open search database *
285 *************************************************/
287 /* A mode, and lists of owners and groups, are passed over for checking in
288 the cases where the database is one or more files. Return NULL, with a message
289 pointed to by message, in cases of error.
291 For search types that use a file or files, check up on the mode after
292 opening. It is tempting to do a stat before opening the file, and use it as
293 an existence check. However, doing that opens a small security loophole in
294 that the status could be changed before the file is opened. Can't quite see
295 what problems this might lead to, but you can't be too careful where security
296 is concerned. Fstat() on an open file can normally be expected to succeed,
297 but there are some NFS states where it does not.
299 There are two styles of query: (1) in the "single-key+file" style, a single
300 key string and a file name are given, for example, for linear searches, DBM
301 files, or for NIS. (2) In the "query" style, no "filename" is given; instead
302 just a single query string is passed. This applies to multiple-key lookup
305 Before opening, scan the tree of cached files to see if this file is already
306 open for the correct search type. If so, return the saved handle. If not, put
307 the handle in the tree for possible subsequent use. See search_tidyup above for
308 closing all the cached files.
310 A count of open databases which use real files is maintained, and if this
311 gets too large, we have to close a cached file. Its entry remains in the tree,
312 but is marked closed.
315 filename the name of the file for single-key+file style lookups,
316 NULL for query-style lookups
317 search_type the type of search required
318 modemask if a real single file is used, this specifies mode bits that
319 must not be set; otherwise it is ignored
320 owners if a real single file is used, this specifies the possible
321 owners of the file; otherwise it is ignored
322 owngroups if a real single file is used, this specifies the possible
323 group owners of the file; otherwise it is ignored
325 Returns: an identifying handle for the open database;
326 this is the pointer to the tree block in the
327 cache of open files; return NULL on open failure, with
328 a message in search_error_message
332 search_open(uschar *filename, int search_type, int modemask, uid_t *owners,
338 lookup_info *lk = lookup_list + search_type;
339 uschar keybuffer[256];
340 int old_pool = store_pool;
342 /* Change to the search store pool and remember our reset point */
344 store_pool = POOL_SEARCH;
345 if (search_reset_point == NULL) search_reset_point = store_get(0);
347 DEBUG(D_lookup) debug_printf("search_open: %s \"%s\"\n", lk->name,
348 (filename == NULL)? US"NULL" : filename);
350 /* See if we already have this open for this type of search, and if so,
351 pass back the tree block as the handle. The key for the tree node is the search
352 type plus '0' concatenated with the file name. There may be entries in the tree
353 with closed files if a lot of files have been opened. */
355 sprintf(CS keybuffer, "%c%.254s", search_type + '0',
356 (filename == NULL)? US"" : filename);
358 if ((t = tree_search(search_tree, keybuffer)) != NULL)
360 c = (search_cache *)(t->data.ptr);
361 if (c->handle != NULL)
363 DEBUG(D_lookup) debug_printf(" cached open\n");
364 store_pool = old_pool;
367 DEBUG(D_lookup) debug_printf(" cached closed\n");
370 /* Otherwise, we need to open the file or database - each search type has its
371 own code, which is now split off into separately compiled modules. Before doing
372 this, if the search type is one that uses real files, check on the number that
373 we are holding open in the cache. If the limit is reached, close the least
374 recently used one. */
376 if (lk->type == lookup_absfile && open_filecount >= lookup_open_max)
378 if (open_bot == NULL)
379 log_write(0, LOG_MAIN|LOG_PANIC, "too many lookups open, but can't find "
383 search_cache *c = (search_cache *)(open_bot->data.ptr);
384 DEBUG(D_lookup) debug_printf("Too many lookup files open\n closing %s\n",
387 if (open_bot != NULL)
388 ((search_cache *)(open_bot->data.ptr))->down = NULL;
391 ((lookup_list + c->search_type)->close)(c->handle);
397 /* If opening is successful, call the file-checking function if there is one,
398 and if all is still well, enter the open database into the tree. */
400 handle = lk->open(filename, &search_error_message);
403 store_pool = old_pool;
407 if (lk->check != NULL &&
408 !lk->check(handle, filename, modemask, owners, owngroups,
409 &search_error_message))
412 store_pool = old_pool;
416 /* If this is a search type that uses real files, keep count. */
418 if (lk->type == lookup_absfile) open_filecount++;
420 /* If we found a previously opened entry in the tree, re-use it; otherwise
421 insert a new entry. On re-use, leave any cached lookup data and the lookup
426 t = store_get(sizeof(tree_node) + Ustrlen(keybuffer));
427 t->data.ptr = c = store_get(sizeof(search_cache));
428 c->item_cache = NULL;
429 Ustrcpy(t->name, keybuffer);
430 tree_insertnode(&search_tree, t);
432 else c = t->data.ptr;
435 c->search_type = search_type;
436 c->up = c->down = NULL;
438 store_pool = old_pool;
446 /*************************************************
447 * Internal function: Find one item in database *
448 *************************************************/
450 /* The answer is always put into dynamic store. The last lookup for each handle
454 handle the handle from search_open; points to tree node
455 filename the filename that was handed to search_open, or
456 NULL for query-style searches
457 keystring the keystring for single-key+file lookups, or
458 the querystring for query-style lookups
460 Returns: a pointer to a dynamic string containing the answer,
461 or NULL if the query failed or was deferred; in the
462 latter case, search_find_defer is set TRUE; after an unusual
463 failure, there may be a message in search_error_message.
467 internal_search_find(void *handle, uschar *filename, uschar *keystring)
469 tree_node *t = (tree_node *)handle;
470 search_cache *c = (search_cache *)(t->data.ptr);
472 int search_type = t->name[0] - '0';
473 int old_pool = store_pool;
475 /* Lookups that return DEFER may not always set an error message. So that
476 the callers don't have to test for NULL, set an empty string. */
478 search_error_message = US"";
479 search_find_defer = FALSE;
481 DEBUG(D_lookup) debug_printf("internal_search_find: file=\"%s\"\n "
482 "type=%s key=\"%s\"\n", filename,
483 lookup_list[search_type].name, keystring);
485 /* Insurance. If the keystring is empty, just fail. */
487 if (keystring[0] == 0) return NULL;
489 /* Use the special store pool for search data */
491 store_pool = POOL_SEARCH;
493 /* Look up the data for the key, unless it is already in the cache for this
494 file. No need to check c->item_cache for NULL, tree_search will do so. */
496 if ((t = tree_search(c->item_cache, keystring)) == NULL)
498 BOOL do_cache = TRUE;
499 int keylength = Ustrlen(keystring);
503 if (filename != NULL)
504 debug_printf("file lookup required for %s\n in %s\n",
505 keystring, filename);
507 debug_printf("database lookup required for %s\n", keystring);
510 /* Call the code for the different kinds of search. DEFER is handled
511 like FAIL, except that search_find_defer is set so the caller can
512 distinguish if necessary. */
514 if (lookup_list[search_type].find(c->handle, filename, keystring, keylength,
515 &data, &search_error_message, &do_cache) == DEFER)
517 search_find_defer = TRUE;
520 /* A record that has been found is now in data, which is either NULL
521 or points to a bit of dynamic store. Cache the result of the lookup if
522 caching is permitted. Lookups can disable caching, when they did something
523 that changes their data. The mysql and pgsql lookups do this when an
524 UPDATE/INSERT query was executed. */
528 int len = keylength + 1;
529 t = store_get(sizeof(tree_node) + len);
530 memcpy(t->name, keystring, len);
532 tree_insertnode(&c->item_cache, t);
535 /* If caching was disabled, empty the cache tree. We just set the cache
536 pointer to NULL here, because we cannot release the store at this stage. */
540 DEBUG(D_lookup) debug_printf("lookup forced cache cleanup\n");
541 c->item_cache = NULL;
545 /* Data was in the cache already; set the pointer from the tree node */
549 data = US t->data.ptr;
550 DEBUG(D_lookup) debug_printf("cached data used for lookup of %s%s%s\n",
552 (filename == NULL)? US"" : US"\n in ",
553 (filename == NULL)? US"" : filename);
556 /* Debug: output the answer */
562 if (search_find_defer) debug_printf("lookup deferred: %s\n",
563 search_error_message);
564 else debug_printf("lookup failed\n");
566 else debug_printf("lookup yielded: %s\n", data);
569 /* Return it in new dynamic store in the regular pool */
571 store_pool = old_pool;
572 return (data == NULL)? NULL : string_copy(data);
578 /*************************************************
579 * Find one item in database, possibly wildcarded *
580 *************************************************/
582 /* This function calls the internal function above; once only if there
583 is no partial matching, but repeatedly when partial matching is requested.
586 handle the handle from search_open
587 filename the filename that was handed to search_open, or
588 NULL for query-style searches
589 keystring the keystring for single-key+file lookups, or
590 the querystring for query-style lookups
591 partial -1 means no partial matching;
592 otherwise it's the minimum number of components;
593 affix the affix string for partial matching
594 affixlen the length of the affix string
595 starflags SEARCH_STAR and SEARCH_STARAT flags
596 expand_setup pointer to offset for setting up expansion strings;
599 Returns: a pointer to a dynamic string containing the answer,
600 or NULL if the query failed or was deferred; in the
601 latter case, search_find_defer is set TRUE
605 search_find(void *handle, uschar *filename, uschar *keystring, int partial,
606 uschar *affix, int affixlen, int starflags, int *expand_setup)
608 tree_node *t = (tree_node *)handle;
609 BOOL set_null_wild = FALSE;
614 if (partial < 0) affixlen = 99; /* So that "NULL" prints */
615 debug_printf("search_find: file=\"%s\"\n key=\"%s\" "
616 "partial=%d affix=%.*s starflags=%x\n",
617 (filename == NULL)? US"NULL" : filename,
618 keystring, partial, affixlen, affix, starflags);
621 /* Arrange to put this database at the top of the LRU chain if it is a type
622 that opens real files. */
624 if (open_top != (tree_node *)handle &&
625 lookup_list[t->name[0]-'0'].type == lookup_absfile)
627 search_cache *c = (search_cache *)(t->data.ptr);
628 tree_node *up = c->up;
629 tree_node *down = c->down;
631 /* Cut it out of the list. A newly opened file will a NULL up pointer.
632 Otherwise there will be a non-NULL up pointer, since we checked above that
633 this block isn't already at the top of the list. */
637 ((search_cache *)(up->data.ptr))->down = down;
639 ((search_cache *)(down->data.ptr))->up = up;
643 /* Now put it at the head of the list. */
647 if (open_top == NULL) open_bot = t; else
648 ((search_cache *)(open_top->data.ptr))->up = t;
654 tree_node *t = open_top;
655 debug_printf("LRU list:\n");
658 search_cache *c = (search_cache *)(t->data.ptr);
659 debug_printf(" %s\n", t->name);
660 if (t == open_bot) debug_printf(" End\n");
665 /* First of all, try to match the key string verbatim. If matched a complete
666 entry but could have been partial, flag to set up variables. */
668 yield = internal_search_find(handle, filename, keystring);
669 if (search_find_defer) return NULL;
670 if (yield != NULL) { if (partial >= 0) set_null_wild = TRUE; }
672 /* Not matched a complete entry; handle partial lookups, but only if the full
673 search didn't defer. Don't use string_sprintf() to construct the initial key,
674 just in case the original key is too long for the string_sprintf() buffer (it
675 *has* happened!). The case of a zero-length affix has to be treated specially.
678 else if (partial >= 0)
680 int len = Ustrlen(keystring);
683 /* Try with the affix on the front, except for a zero-length affix */
685 if (affixlen == 0) keystring2 = keystring; else
687 keystring2 = store_get(len + affixlen + 1);
688 Ustrncpy(keystring2, affix, affixlen);
689 Ustrcpy(keystring2 + affixlen, keystring);
690 DEBUG(D_lookup) debug_printf("trying partial match %s\n", keystring2);
691 yield = internal_search_find(handle, filename, keystring2);
692 if (search_find_defer) return NULL;
695 /* The key in its entirety did not match a wild entry; try chopping off
696 leading components. */
701 uschar *keystring3 = keystring2 + affixlen;
702 uschar *s = keystring3;
703 while (*s != 0) if (*s++ == '.') dotcount++;
705 while (dotcount-- >= partial)
707 while (*keystring3 != 0 && *keystring3 != '.') keystring3++;
709 /* If we get right to the end of the string (which will be the last time
710 through this loop), we've failed if the affix is null. Otherwise do one
711 last lookup for the affix itself, but if it is longer than 1 character,
712 remove the last character if it is ".". */
714 if (*keystring3 == 0)
716 if (affixlen < 1) break;
717 if (affixlen > 1 && affix[affixlen-1] == '.') affixlen--;
718 Ustrncpy(keystring2, affix, affixlen);
719 keystring2[affixlen] = 0;
720 keystring3 = keystring2;
724 keystring3 -= affixlen - 1;
725 if (affixlen > 0) Ustrncpy(keystring3, affix, affixlen);
728 DEBUG(D_lookup) debug_printf("trying partial match %s\n", keystring3);
729 yield = internal_search_find(handle, filename, keystring3);
730 if (search_find_defer) return NULL;
733 /* First variable is the wild part; second is the fixed part. Take care
734 to get it right when keystring3 is just "*". */
736 if (expand_setup != NULL && *expand_setup >= 0)
738 int fixedlength = Ustrlen(keystring3) - affixlen;
739 int wildlength = Ustrlen(keystring) - fixedlength - 1;
741 expand_nstring[*expand_setup] = keystring;
742 expand_nlength[*expand_setup] = wildlength;
744 expand_nstring[*expand_setup] = keystring + wildlength + 1;
745 expand_nlength[*expand_setup] = (fixedlength < 0)? 0 : fixedlength;
749 keystring3 += affixlen;
753 else set_null_wild = TRUE; /* Matched a wild entry without any wild part */
756 /* If nothing has been matched, but the option to look for "*@" is set, try
757 replacing everthing to the left of @ by *. After a match, the wild part
758 is set to the string to the left of the @. */
760 if (yield == NULL && (starflags & SEARCH_STARAT) != 0)
762 uschar *atat = Ustrrchr(keystring, '@');
763 if (atat != NULL && atat > keystring)
766 savechar = *(--atat);
769 DEBUG(D_lookup) debug_printf("trying default match %s\n", atat);
770 yield = internal_search_find(handle, filename, atat);
772 if (search_find_defer) return NULL;
774 if (yield != NULL && expand_setup != NULL && *expand_setup >= 0)
777 expand_nstring[*expand_setup] = keystring;
778 expand_nlength[*expand_setup] = atat - keystring + 1;
780 expand_nstring[*expand_setup] = keystring;
781 expand_nlength[*expand_setup] = 0;
786 /* If we still haven't matched anything, and the option to look for "*" is set,
787 try that. If we do match, the first variable (the wild part) is the whole key,
788 and the second is empty. */
790 if (yield == NULL && (starflags & (SEARCH_STAR|SEARCH_STARAT)) != 0)
792 DEBUG(D_lookup) debug_printf("trying to match *\n");
793 yield = internal_search_find(handle, filename, US"*");
794 if (yield != NULL && expand_setup != NULL && *expand_setup >= 0)
797 expand_nstring[*expand_setup] = keystring;
798 expand_nlength[*expand_setup] = Ustrlen(keystring);
800 expand_nstring[*expand_setup] = keystring;
801 expand_nlength[*expand_setup] = 0;
805 /* If this was a potentially partial lookup, and we matched either a
806 complete non-wild domain entry, or we matched a wild-carded entry without
807 chopping off any of the domain components, set up the expansion variables
808 (if required) so that the first one is empty, and the second one is the
809 fixed part of the domain. The set_null_wild flag is set only when yield is not
812 if (set_null_wild && expand_setup != NULL && *expand_setup >= 0)
815 expand_nstring[*expand_setup] = keystring;
816 expand_nlength[*expand_setup] = 0;
818 expand_nstring[*expand_setup] = keystring;
819 expand_nlength[*expand_setup] = Ustrlen(keystring);
825 /* End of search.c */