1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
5 /* Copyright (c) University of Cambridge 1995 - 2015 */
6 /* See the file NOTICE for conditions of use and distribution. */
8 /* A set of functions to search databases in various formats. An open
9 database is represented by a void * value which is returned from a lookup-
10 specific "open" function. These are now all held in individual modules in the
11 lookups subdirectory and the functions here form a generic interface.
13 Caching is used to improve performance. Open files are cached until a tidyup
14 function is called, and for each file the result of the last lookup is cached.
15 However, if too many files are opened, some of those that are not in use have
16 to be closed. Those open items that use real files are kept on a LRU chain to
19 All the data is held in permanent store so as to be independent of the stacking
20 pool that is reset from time to time. In fact, we use malloc'd store so that it
21 can be freed when the caches are tidied up. It isn't actually clear whether
22 this is a benefit or not, to be honest. */
27 /* Tree in which to cache open files until tidyup called. */
29 static tree_node *search_tree = NULL;
31 /* Two-way chain of open databases that use real files. This is maintained in
32 recently-used order for the purposes of closing the least recently used when
33 too many files are open. */
35 static tree_node *open_top = NULL;
36 static tree_node *open_bot = NULL;
38 /* Count of open databases that use real files */
40 static int open_filecount = 0;
42 /* Allow us to reset store used for lookups and lookup caching */
44 static void *search_reset_point = NULL;
48 /*************************************************
49 * Validate a plain lookup type name *
50 *************************************************/
52 /* Only those names that are recognized and whose code is included in the
53 binary give an OK response. Use a binary chop search now that the list has got
57 name lookup type name - not necessarily zero terminated (e.g. dbm*)
58 len length of the name
60 Returns: +ve => valid lookup name; value is offset in lookup_list
61 -ve => invalid name; message in search_error_message.
65 search_findtype(const uschar *name, int len)
68 int top = lookup_list_count;
71 int mid = (top + bot)/2;
72 int c = Ustrncmp(name, lookup_list[mid]->name, len);
74 /* If c == 0 we have matched the incoming name with the start of the search
75 type name. However, some search types are substrings of others (e.g. nis and
76 nisplus) so we need to check that the lengths are the same. The length of the
77 type name cannot be shorter (else c would not be 0); if it is not equal it
78 must be longer, and in that case, the incoming name comes before the name we
79 are testing. By leaving c == 0 when the lengths are different, and doing a
80 > 0 test below, this all falls out correctly. */
82 if (c == 0 && Ustrlen(lookup_list[mid]->name) == len)
84 if (lookup_list[mid]->find != NULL) return mid;
85 search_error_message = string_sprintf("lookup type \"%.*s\" is not "
86 "available (not in the binary - check buildtime LOOKUP configuration)",
91 if (c > 0) bot = mid + 1; else top = mid;
94 search_error_message = string_sprintf("unknown lookup type \"%.*s\"",len,name);
100 /*************************************************
101 * Validate a full lookup type name *
102 *************************************************/
104 /* This function recognizes the "partial-" prefix and also terminating * and *@
108 name the full lookup type name
109 ptypeptr where to put the partial type
110 after subtraction of 1024 or 2048:
111 negative => no partial matching
112 non-negative => minimum number of non-wild components
113 ptypeaff where to put a pointer to the affix
114 the affix is within name if supplied therein
115 otherwise it's a literal string
116 afflen the length of the affix
117 starflags where to put the SEARCH_STAR and SEARCH_STARAT flags
119 Returns: +ve => valid lookup name; value is offset in lookup_list
120 -ve => invalid name; message in search_error_message.
124 search_findtype_partial(const uschar *name, int *ptypeptr, const uschar **ptypeaff,
125 int *afflen, int *starflags)
129 const uschar *ss = name;
134 /* Check for a partial matching type. It must start with "partial", optionally
135 followed by a sequence of digits. If this is followed by "-", the affix is the
136 default "*." string. Otherwise we expect an affix in parentheses. Affixes are a
137 limited number of characters, not including parens. */
139 if (Ustrncmp(name, "partial", 7) == 0)
145 while (isdigit(*ss)) pv = pv*10 + *ss++ - '0';
147 else pv = 2; /* Default number of wild components */
152 while (ispunct(*ss) && *ss != ')') ss++;
153 if (*ss != ')') goto BAD_TYPE;
154 *afflen = ss++ - *ptypeaff;
156 else if (*ss++ == '-')
164 search_error_message = string_sprintf("format error in lookup type \"%s\"",
170 /* Now we are left with a lookup name, possibly followed by * or *@. */
173 if (len >= 2 && Ustrncmp(ss + len - 2, "*@", 2) == 0)
175 *starflags |= SEARCH_STARAT;
178 else if (len >= 1 && ss[len-1] == '*')
180 *starflags |= SEARCH_STAR;
184 /* Check for the individual search type. Only those that are actually in the
185 binary are valid. For query-style types, "partial" and default types are
188 stype = search_findtype(ss, len);
189 if (stype >= 0 && mac_islookup(stype, lookup_querystyle))
193 search_error_message = string_sprintf("\"partial\" is not permitted "
194 "for lookup type \"%s\"", ss);
197 if ((*starflags & (SEARCH_STAR|SEARCH_STARAT)) != 0)
199 search_error_message = string_sprintf("defaults using \"*\" or \"*@\" are "
200 "not permitted for lookup type \"%s\"", ss);
211 /*************************************************
212 * Release cached resources *
213 *************************************************/
215 /* When search_open is called it caches the "file" that it opens in
216 search_tree. The name of the tree node is a concatenation of the search type
217 with the file name. For query-style lookups, the file name is empty. Real files
218 are normally closed only when this tidyup routine is called, typically at the
219 end of sections of code where a number of lookups might occur. However, if too
220 many files are open simultaneously, some get closed beforehand. They can't be
221 removed from the tree. There is also a general tidyup function which is called
222 for the lookup driver, if it exists.
224 First, there is an internal, recursive subroutine.
226 Argument: a pointer to a search_openfile tree node
231 tidyup_subtree(tree_node *t)
233 search_cache *c = (search_cache *)(t->data.ptr);
234 if (t->left != NULL) tidyup_subtree(t->left);
235 if (t->right != NULL) tidyup_subtree(t->right);
238 lookup_list[c->search_type]->close != NULL)
239 lookup_list[c->search_type]->close(c->handle);
243 /* The external entry point
253 int old_pool = store_pool;
255 DEBUG(D_lookup) debug_printf("search_tidyup called\n");
257 /* Close individually each cached open file. */
259 store_pool = POOL_SEARCH;
260 if (search_tree != NULL)
262 tidyup_subtree(search_tree);
265 open_top = open_bot = NULL;
268 /* Call the general tidyup entry for any drivers that have one. */
270 for (i = 0; i < lookup_list_count; i++)
271 if (lookup_list[i]->tidy != NULL) (lookup_list[i]->tidy)();
273 if (search_reset_point != NULL) store_reset(search_reset_point);
274 search_reset_point = NULL;
275 store_pool = old_pool;
281 /*************************************************
282 * Open search database *
283 *************************************************/
285 /* A mode, and lists of owners and groups, are passed over for checking in
286 the cases where the database is one or more files. Return NULL, with a message
287 pointed to by message, in cases of error.
289 For search types that use a file or files, check up on the mode after
290 opening. It is tempting to do a stat before opening the file, and use it as
291 an existence check. However, doing that opens a small security loophole in
292 that the status could be changed before the file is opened. Can't quite see
293 what problems this might lead to, but you can't be too careful where security
294 is concerned. Fstat() on an open file can normally be expected to succeed,
295 but there are some NFS states where it does not.
297 There are two styles of query: (1) in the "single-key+file" style, a single
298 key string and a file name are given, for example, for linear searches, DBM
299 files, or for NIS. (2) In the "query" style, no "filename" is given; instead
300 just a single query string is passed. This applies to multiple-key lookup
303 Before opening, scan the tree of cached files to see if this file is already
304 open for the correct search type. If so, return the saved handle. If not, put
305 the handle in the tree for possible subsequent use. See search_tidyup above for
306 closing all the cached files.
308 A count of open databases which use real files is maintained, and if this
309 gets too large, we have to close a cached file. Its entry remains in the tree,
310 but is marked closed.
313 filename the name of the file for single-key+file style lookups,
314 NULL for query-style lookups
315 search_type the type of search required
316 modemask if a real single file is used, this specifies mode bits that
317 must not be set; otherwise it is ignored
318 owners if a real single file is used, this specifies the possible
319 owners of the file; otherwise it is ignored
320 owngroups if a real single file is used, this specifies the possible
321 group owners of the file; otherwise it is ignored
323 Returns: an identifying handle for the open database;
324 this is the pointer to the tree block in the
325 cache of open files; return NULL on open failure, with
326 a message in search_error_message
330 search_open(uschar *filename, int search_type, int modemask, uid_t *owners,
336 lookup_info *lk = lookup_list[search_type];
337 uschar keybuffer[256];
338 int old_pool = store_pool;
340 /* Change to the search store pool and remember our reset point */
342 store_pool = POOL_SEARCH;
343 if (search_reset_point == NULL) search_reset_point = store_get(0);
345 DEBUG(D_lookup) debug_printf("search_open: %s \"%s\"\n", lk->name,
346 (filename == NULL)? US"NULL" : filename);
348 /* See if we already have this open for this type of search, and if so,
349 pass back the tree block as the handle. The key for the tree node is the search
350 type plus '0' concatenated with the file name. There may be entries in the tree
351 with closed files if a lot of files have been opened. */
353 sprintf(CS keybuffer, "%c%.254s", search_type + '0',
354 (filename == NULL)? US"" : filename);
356 if ((t = tree_search(search_tree, keybuffer)) != NULL)
358 c = (search_cache *)(t->data.ptr);
359 if (c->handle != NULL)
361 DEBUG(D_lookup) debug_printf(" cached open\n");
362 store_pool = old_pool;
365 DEBUG(D_lookup) debug_printf(" cached closed\n");
368 /* Otherwise, we need to open the file or database - each search type has its
369 own code, which is now split off into separately compiled modules. Before doing
370 this, if the search type is one that uses real files, check on the number that
371 we are holding open in the cache. If the limit is reached, close the least
372 recently used one. */
374 if (lk->type == lookup_absfile && open_filecount >= lookup_open_max)
376 if (open_bot == NULL)
377 log_write(0, LOG_MAIN|LOG_PANIC, "too many lookups open, but can't find "
381 search_cache *c = (search_cache *)(open_bot->data.ptr);
382 DEBUG(D_lookup) debug_printf("Too many lookup files open\n closing %s\n",
385 if (open_bot != NULL)
386 ((search_cache *)(open_bot->data.ptr))->down = NULL;
389 ((lookup_list[c->search_type])->close)(c->handle);
395 /* If opening is successful, call the file-checking function if there is one,
396 and if all is still well, enter the open database into the tree. */
398 handle = (lk->open)(filename, &search_error_message);
401 store_pool = old_pool;
405 if (lk->check != NULL &&
406 !lk->check(handle, filename, modemask, owners, owngroups,
407 &search_error_message))
410 store_pool = old_pool;
414 /* If this is a search type that uses real files, keep count. */
416 if (lk->type == lookup_absfile) open_filecount++;
418 /* If we found a previously opened entry in the tree, re-use it; otherwise
419 insert a new entry. On re-use, leave any cached lookup data and the lookup
424 t = store_get(sizeof(tree_node) + Ustrlen(keybuffer));
425 t->data.ptr = c = store_get(sizeof(search_cache));
426 c->item_cache = NULL;
427 Ustrcpy(t->name, keybuffer);
428 tree_insertnode(&search_tree, t);
430 else c = t->data.ptr;
433 c->search_type = search_type;
434 c->up = c->down = NULL;
436 store_pool = old_pool;
444 /*************************************************
445 * Internal function: Find one item in database *
446 *************************************************/
448 /* The answer is always put into dynamic store. The last lookup for each handle
452 handle the handle from search_open; points to tree node
453 filename the filename that was handed to search_open, or
454 NULL for query-style searches
455 keystring the keystring for single-key+file lookups, or
456 the querystring for query-style lookups
458 Returns: a pointer to a dynamic string containing the answer,
459 or NULL if the query failed or was deferred; in the
460 latter case, search_find_defer is set TRUE; after an unusual
461 failure, there may be a message in search_error_message.
465 internal_search_find(void *handle, uschar *filename, uschar *keystring)
467 tree_node *t = (tree_node *)handle;
468 search_cache *c = (search_cache *)(t->data.ptr);
471 int search_type = t->name[0] - '0';
472 int old_pool = store_pool;
474 /* Lookups that return DEFER may not always set an error message. So that
475 the callers don't have to test for NULL, set an empty string. */
477 search_error_message = US"";
478 search_find_defer = FALSE;
480 DEBUG(D_lookup) debug_printf("internal_search_find: file=\"%s\"\n "
481 "type=%s key=\"%s\"\n", filename,
482 lookup_list[search_type]->name, keystring);
484 /* Insurance. If the keystring is empty, just fail. */
486 if (keystring[0] == 0) return NULL;
488 /* Use the special store pool for search data */
490 store_pool = POOL_SEARCH;
492 /* Look up the data for the key, unless it is already in the cache for this
493 file. No need to check c->item_cache for NULL, tree_search will do so. */
495 if ( (t = tree_search(c->item_cache, keystring))
496 && (!(e = t->data.ptr)->expiry || e->expiry > time(NULL))
498 { /* Data was in the cache already; set the pointer from the tree node */
500 DEBUG(D_lookup) debug_printf("cached data used for lookup of %s%s%s\n",
502 filename ? US"\n in " : US"", filename ? filename : US"");
506 uint do_cache = UINT_MAX;
507 int keylength = Ustrlen(keystring);
511 if (t) debug_printf("cached data found but past valid time; ");
512 debug_printf("%s lookup required for %s%s%s\n",
513 filename ? US"file" : US"database",
515 filename ? US"\n in " : US"", filename ? filename : US"");
518 /* Call the code for the different kinds of search. DEFER is handled
519 like FAIL, except that search_find_defer is set so the caller can
520 distinguish if necessary. */
522 if (lookup_list[search_type]->find(c->handle, filename, keystring, keylength,
523 &data, &search_error_message, &do_cache) == DEFER)
524 search_find_defer = TRUE;
526 /* A record that has been found is now in data, which is either NULL
527 or points to a bit of dynamic store. Cache the result of the lookup if
528 caching is permitted. Lookups can disable caching, when they did something
529 that changes their data. The mysql and pgsql lookups do this when an
530 UPDATE/INSERT query was executed. */
534 int len = keylength + 1;
536 if (t) /* Previous, out-of-date cache entry. Update with the */
537 { /* new result and forget the old one */
538 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
543 e = store_get(sizeof(expiring_data) + sizeof(tree_node) + len);
544 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
546 t = (tree_node *)(e+1);
547 memcpy(t->name, keystring, len);
549 tree_insertnode(&c->item_cache, t);
553 /* If caching was disabled, empty the cache tree. We just set the cache
554 pointer to NULL here, because we cannot release the store at this stage. */
558 DEBUG(D_lookup) debug_printf("lookup forced cache cleanup\n");
559 c->item_cache = NULL;
566 debug_printf("lookup yielded: %s\n", data);
567 else if (search_find_defer)
568 debug_printf("lookup deferred: %s\n", search_error_message);
569 else debug_printf("lookup failed\n");
572 /* Return it in new dynamic store in the regular pool */
574 store_pool = old_pool;
575 return data ? string_copy(data) : NULL;
581 /*************************************************
582 * Find one item in database, possibly wildcarded *
583 *************************************************/
585 /* This function calls the internal function above; once only if there
586 is no partial matching, but repeatedly when partial matching is requested.
589 handle the handle from search_open
590 filename the filename that was handed to search_open, or
591 NULL for query-style searches
592 keystring the keystring for single-key+file lookups, or
593 the querystring for query-style lookups
594 partial -1 means no partial matching;
595 otherwise it's the minimum number of components;
596 affix the affix string for partial matching
597 affixlen the length of the affix string
598 starflags SEARCH_STAR and SEARCH_STARAT flags
599 expand_setup pointer to offset for setting up expansion strings;
602 Returns: a pointer to a dynamic string containing the answer,
603 or NULL if the query failed or was deferred; in the
604 latter case, search_find_defer is set TRUE
608 search_find(void *handle, uschar *filename, uschar *keystring, int partial,
609 const uschar *affix, int affixlen, int starflags, int *expand_setup)
611 tree_node *t = (tree_node *)handle;
612 BOOL set_null_wild = FALSE;
617 if (partial < 0) affixlen = 99; /* So that "NULL" prints */
618 debug_printf("search_find: file=\"%s\"\n key=\"%s\" "
619 "partial=%d affix=%.*s starflags=%x\n",
620 (filename == NULL)? US"NULL" : filename,
621 keystring, partial, affixlen, affix, starflags);
624 /* Arrange to put this database at the top of the LRU chain if it is a type
625 that opens real files. */
627 if (open_top != (tree_node *)handle &&
628 lookup_list[t->name[0]-'0']->type == lookup_absfile)
630 search_cache *c = (search_cache *)(t->data.ptr);
631 tree_node *up = c->up;
632 tree_node *down = c->down;
634 /* Cut it out of the list. A newly opened file will a NULL up pointer.
635 Otherwise there will be a non-NULL up pointer, since we checked above that
636 this block isn't already at the top of the list. */
640 ((search_cache *)(up->data.ptr))->down = down;
642 ((search_cache *)(down->data.ptr))->up = up;
646 /* Now put it at the head of the list. */
650 if (open_top == NULL) open_bot = t; else
651 ((search_cache *)(open_top->data.ptr))->up = t;
657 tree_node *t = open_top;
658 debug_printf("LRU list:\n");
661 search_cache *c = (search_cache *)(t->data.ptr);
662 debug_printf(" %s\n", t->name);
663 if (t == open_bot) debug_printf(" End\n");
668 /* First of all, try to match the key string verbatim. If matched a complete
669 entry but could have been partial, flag to set up variables. */
671 yield = internal_search_find(handle, filename, keystring);
672 if (search_find_defer) return NULL;
673 if (yield != NULL) { if (partial >= 0) set_null_wild = TRUE; }
675 /* Not matched a complete entry; handle partial lookups, but only if the full
676 search didn't defer. Don't use string_sprintf() to construct the initial key,
677 just in case the original key is too long for the string_sprintf() buffer (it
678 *has* happened!). The case of a zero-length affix has to be treated specially.
681 else if (partial >= 0)
683 int len = Ustrlen(keystring);
686 /* Try with the affix on the front, except for a zero-length affix */
688 if (affixlen == 0) keystring2 = keystring; else
690 keystring2 = store_get(len + affixlen + 1);
691 Ustrncpy(keystring2, affix, affixlen);
692 Ustrcpy(keystring2 + affixlen, keystring);
693 DEBUG(D_lookup) debug_printf("trying partial match %s\n", keystring2);
694 yield = internal_search_find(handle, filename, keystring2);
695 if (search_find_defer) return NULL;
698 /* The key in its entirety did not match a wild entry; try chopping off
699 leading components. */
704 uschar *keystring3 = keystring2 + affixlen;
705 uschar *s = keystring3;
706 while (*s != 0) if (*s++ == '.') dotcount++;
708 while (dotcount-- >= partial)
710 while (*keystring3 != 0 && *keystring3 != '.') keystring3++;
712 /* If we get right to the end of the string (which will be the last time
713 through this loop), we've failed if the affix is null. Otherwise do one
714 last lookup for the affix itself, but if it is longer than 1 character,
715 remove the last character if it is ".". */
717 if (*keystring3 == 0)
719 if (affixlen < 1) break;
720 if (affixlen > 1 && affix[affixlen-1] == '.') affixlen--;
721 Ustrncpy(keystring2, affix, affixlen);
722 keystring2[affixlen] = 0;
723 keystring3 = keystring2;
727 keystring3 -= affixlen - 1;
728 if (affixlen > 0) Ustrncpy(keystring3, affix, affixlen);
731 DEBUG(D_lookup) debug_printf("trying partial match %s\n", keystring3);
732 yield = internal_search_find(handle, filename, keystring3);
733 if (search_find_defer) return NULL;
736 /* First variable is the wild part; second is the fixed part. Take care
737 to get it right when keystring3 is just "*". */
739 if (expand_setup != NULL && *expand_setup >= 0)
741 int fixedlength = Ustrlen(keystring3) - affixlen;
742 int wildlength = Ustrlen(keystring) - fixedlength - 1;
744 expand_nstring[*expand_setup] = keystring;
745 expand_nlength[*expand_setup] = wildlength;
747 expand_nstring[*expand_setup] = keystring + wildlength + 1;
748 expand_nlength[*expand_setup] = (fixedlength < 0)? 0 : fixedlength;
752 keystring3 += affixlen;
756 else set_null_wild = TRUE; /* Matched a wild entry without any wild part */
759 /* If nothing has been matched, but the option to look for "*@" is set, try
760 replacing everthing to the left of @ by *. After a match, the wild part
761 is set to the string to the left of the @. */
763 if (yield == NULL && (starflags & SEARCH_STARAT) != 0)
765 uschar *atat = Ustrrchr(keystring, '@');
766 if (atat != NULL && atat > keystring)
769 savechar = *(--atat);
772 DEBUG(D_lookup) debug_printf("trying default match %s\n", atat);
773 yield = internal_search_find(handle, filename, atat);
775 if (search_find_defer) return NULL;
777 if (yield != NULL && expand_setup != NULL && *expand_setup >= 0)
780 expand_nstring[*expand_setup] = keystring;
781 expand_nlength[*expand_setup] = atat - keystring + 1;
783 expand_nstring[*expand_setup] = keystring;
784 expand_nlength[*expand_setup] = 0;
789 /* If we still haven't matched anything, and the option to look for "*" is set,
790 try that. If we do match, the first variable (the wild part) is the whole key,
791 and the second is empty. */
793 if (yield == NULL && (starflags & (SEARCH_STAR|SEARCH_STARAT)) != 0)
795 DEBUG(D_lookup) debug_printf("trying to match *\n");
796 yield = internal_search_find(handle, filename, US"*");
797 if (yield != NULL && expand_setup != NULL && *expand_setup >= 0)
800 expand_nstring[*expand_setup] = keystring;
801 expand_nlength[*expand_setup] = Ustrlen(keystring);
803 expand_nstring[*expand_setup] = keystring;
804 expand_nlength[*expand_setup] = 0;
808 /* If this was a potentially partial lookup, and we matched either a
809 complete non-wild domain entry, or we matched a wild-carded entry without
810 chopping off any of the domain components, set up the expansion variables
811 (if required) so that the first one is empty, and the second one is the
812 fixed part of the domain. The set_null_wild flag is set only when yield is not
815 if (set_null_wild && expand_setup != NULL && *expand_setup >= 0)
818 expand_nstring[*expand_setup] = keystring;
819 expand_nlength[*expand_setup] = 0;
821 expand_nstring[*expand_setup] = keystring;
822 expand_nlength[*expand_setup] = Ustrlen(keystring);
828 /* End of search.c */