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 rmark 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
252 int old_pool = store_pool;
254 DEBUG(D_lookup) debug_printf_indent("search_tidyup called\n");
256 /* Close individually each cached open file. */
258 store_pool = POOL_SEARCH;
261 tidyup_subtree(search_tree);
264 open_top = open_bot = NULL;
267 /* Call the general tidyup entry for any drivers that have one. */
269 for (int i = 0; i < lookup_list_count; i++) if (lookup_list[i]->tidy)
270 (lookup_list[i]->tidy)();
272 if (search_reset_point) search_reset_point = store_reset(search_reset_point);
273 store_pool = old_pool;
279 /*************************************************
280 * Open search database *
281 *************************************************/
283 /* A mode, and lists of owners and groups, are passed over for checking in
284 the cases where the database is one or more files. Return NULL, with a message
285 pointed to by message, in cases of error.
287 For search types that use a file or files, check up on the mode after
288 opening. It is tempting to do a stat before opening the file, and use it as
289 an existence check. However, doing that opens a small security loophole in
290 that the status could be changed before the file is opened. Can't quite see
291 what problems this might lead to, but you can't be too careful where security
292 is concerned. Fstat() on an open file can normally be expected to succeed,
293 but there are some NFS states where it does not.
295 There are two styles of query: (1) in the "single-key+file" style, a single
296 key string and a file name are given, for example, for linear searches, DBM
297 files, or for NIS. (2) In the "query" style, no "filename" is given; instead
298 just a single query string is passed. This applies to multiple-key lookup
301 Before opening, scan the tree of cached files to see if this file is already
302 open for the correct search type. If so, return the saved handle. If not, put
303 the handle in the tree for possible subsequent use. See search_tidyup above for
304 closing all the cached files.
306 A count of open databases which use real files is maintained, and if this
307 gets too large, we have to close a cached file. Its entry remains in the tree,
308 but is marked closed.
311 filename the name of the file for single-key+file style lookups,
312 NULL for query-style lookups
313 search_type the type of search required
314 modemask if a real single file is used, this specifies mode bits that
315 must not be set; otherwise it is ignored
316 owners if a real single file is used, this specifies the possible
317 owners of the file; otherwise it is ignored
318 owngroups if a real single file is used, this specifies the possible
319 group owners of the file; otherwise it is ignored
321 Returns: an identifying handle for the open database;
322 this is the pointer to the tree block in the
323 cache of open files; return NULL on open failure, with
324 a message in search_error_message
328 search_open(uschar *filename, int search_type, int modemask, uid_t *owners,
334 lookup_info *lk = lookup_list[search_type];
335 uschar keybuffer[256];
336 int old_pool = store_pool;
338 /* Change to the search store pool and remember our reset point */
340 store_pool = POOL_SEARCH;
341 if (!search_reset_point) search_reset_point = store_mark();
343 DEBUG(D_lookup) debug_printf_indent("search_open: %s \"%s\"\n", lk->name,
344 filename ? filename : US"NULL");
346 /* See if we already have this open for this type of search, and if so,
347 pass back the tree block as the handle. The key for the tree node is the search
348 type plus '0' concatenated with the file name. There may be entries in the tree
349 with closed files if a lot of files have been opened. */
351 sprintf(CS keybuffer, "%c%.254s", search_type + '0',
352 filename ? filename : US"");
354 if ((t = tree_search(search_tree, keybuffer)))
356 c = (search_cache *)(t->data.ptr);
359 DEBUG(D_lookup) debug_printf_indent(" cached open\n");
360 store_pool = old_pool;
363 DEBUG(D_lookup) debug_printf_indent(" cached closed\n");
366 /* Otherwise, we need to open the file or database - each search type has its
367 own code, which is now split off into separately compiled modules. Before doing
368 this, if the search type is one that uses real files, check on the number that
369 we are holding open in the cache. If the limit is reached, close the least
370 recently used one. */
372 if (lk->type == lookup_absfile && open_filecount >= lookup_open_max)
375 log_write(0, LOG_MAIN|LOG_PANIC, "too many lookups open, but can't find "
379 search_cache *c = (search_cache *)(open_bot->data.ptr);
380 DEBUG(D_lookup) debug_printf_indent("Too many lookup files open\n closing %s\n",
382 if ((open_bot = c->up))
383 ((search_cache *)(open_bot->data.ptr))->down = NULL;
386 ((lookup_list[c->search_type])->close)(c->handle);
392 /* If opening is successful, call the file-checking function if there is one,
393 and if all is still well, enter the open database into the tree. */
395 if (!(handle = (lk->open)(filename, &search_error_message)))
397 store_pool = old_pool;
402 && !lk->check(handle, filename, modemask, owners, owngroups,
403 &search_error_message))
406 store_pool = old_pool;
410 /* If this is a search type that uses real files, keep count. */
412 if (lk->type == lookup_absfile) open_filecount++;
414 /* If we found a previously opened entry in the tree, re-use it; otherwise
415 insert a new entry. On re-use, leave any cached lookup data and the lookup
420 t = store_get(sizeof(tree_node) + Ustrlen(keybuffer), FALSE);
421 t->data.ptr = c = store_get(sizeof(search_cache), FALSE);
422 c->item_cache = NULL;
423 Ustrcpy(t->name, keybuffer);
424 tree_insertnode(&search_tree, t);
426 else c = t->data.ptr;
429 c->search_type = search_type;
430 c->up = c->down = NULL;
432 store_pool = old_pool;
440 /*************************************************
441 * Internal function: Find one item in database *
442 *************************************************/
444 /* The answer is always put into dynamic store. The last lookup for each handle
448 handle the handle from search_open; points to tree node
449 filename the filename that was handed to search_open, or
450 NULL for query-style searches
451 keystring the keystring for single-key+file lookups, or
452 the querystring for query-style lookups
454 Returns: a pointer to a dynamic string containing the answer,
455 or NULL if the query failed or was deferred; in the
456 latter case, search_find_defer is set TRUE; after an unusual
457 failure, there may be a message in search_error_message.
461 internal_search_find(void *handle, uschar *filename, uschar *keystring)
463 tree_node * t = (tree_node *)handle;
464 search_cache * c = (search_cache *)(t->data.ptr);
465 expiring_data * e = NULL; /* compiler quietening */
466 uschar * data = NULL;
467 int search_type = t->name[0] - '0';
468 int old_pool = store_pool;
470 /* Lookups that return DEFER may not always set an error message. So that
471 the callers don't have to test for NULL, set an empty string. */
473 search_error_message = US"";
474 f.search_find_defer = FALSE;
476 DEBUG(D_lookup) debug_printf_indent("internal_search_find: file=\"%s\"\n "
477 "type=%s key=\"%s\"\n", filename,
478 lookup_list[search_type]->name, keystring);
480 /* Insurance. If the keystring is empty, just fail. */
482 if (keystring[0] == 0) return NULL;
484 /* Use the special store pool for search data */
486 store_pool = POOL_SEARCH;
488 /* Look up the data for the key, unless it is already in the cache for this
489 file. No need to check c->item_cache for NULL, tree_search will do so. */
491 if ( (t = tree_search(c->item_cache, keystring))
492 && (!(e = t->data.ptr)->expiry || e->expiry > time(NULL))
494 { /* Data was in the cache already; set the pointer from the tree node */
496 DEBUG(D_lookup) debug_printf_indent("cached data used for lookup of %s%s%s\n",
498 filename ? US"\n in " : US"", filename ? filename : US"");
502 uint do_cache = UINT_MAX;
503 int keylength = Ustrlen(keystring);
507 if (t) debug_printf_indent("cached data found but past valid time; ");
508 debug_printf_indent("%s lookup required for %s%s%s\n",
509 filename ? US"file" : US"database",
511 filename ? US"\n in " : US"", filename ? filename : US"");
514 /* Call the code for the different kinds of search. DEFER is handled
515 like FAIL, except that search_find_defer is set so the caller can
516 distinguish if necessary. */
518 if (lookup_list[search_type]->find(c->handle, filename, keystring, keylength,
519 &data, &search_error_message, &do_cache) == DEFER)
520 f.search_find_defer = TRUE;
522 /* A record that has been found is now in data, which is either NULL
523 or points to a bit of dynamic store. Cache the result of the lookup if
524 caching is permitted. Lookups can disable caching, when they did something
525 that changes their data. The mysql and pgsql lookups do this when an
526 UPDATE/INSERT query was executed. */
530 int len = keylength + 1;
532 if (t) /* Previous, out-of-date cache entry. Update with the */
533 { /* new result and forget the old one */
534 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
539 e = store_get(sizeof(expiring_data) + sizeof(tree_node) + len, is_tainted(keystring));
540 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
542 t = (tree_node *)(e+1);
543 memcpy(t->name, keystring, len);
545 tree_insertnode(&c->item_cache, t);
549 /* If caching was disabled, empty the cache tree. We just set the cache
550 pointer to NULL here, because we cannot release the store at this stage. */
554 DEBUG(D_lookup) debug_printf_indent("lookup forced cache cleanup\n");
555 c->item_cache = NULL;
562 debug_printf_indent("lookup yielded: %s\n", data);
563 else if (f.search_find_defer)
564 debug_printf_indent("lookup deferred: %s\n", search_error_message);
565 else debug_printf_indent("lookup failed\n");
568 /* Return it in new dynamic store in the regular pool */
570 store_pool = old_pool;
571 return data ? string_copy(data) : NULL;
577 /*************************************************
578 * Find one item in database, possibly wildcarded *
579 *************************************************/
581 /* This function calls the internal function above; once only if there
582 is no partial matching, but repeatedly when partial matching is requested.
585 handle the handle from search_open
586 filename the filename that was handed to search_open, or
587 NULL for query-style searches
588 keystring the keystring for single-key+file lookups, or
589 the querystring for query-style lookups
590 partial -1 means no partial matching;
591 otherwise it's the minimum number of components;
592 affix the affix string for partial matching
593 affixlen the length of the affix string
594 starflags SEARCH_STAR and SEARCH_STARAT flags
595 expand_setup pointer to offset for setting up expansion strings;
598 Returns: a pointer to a dynamic string containing the answer,
599 or NULL if the query failed or was deferred; in the
600 latter case, search_find_defer is set TRUE
604 search_find(void *handle, uschar *filename, uschar *keystring, int partial,
605 const uschar *affix, int affixlen, int starflags, int *expand_setup)
607 tree_node *t = (tree_node *)handle;
608 BOOL set_null_wild = FALSE;
613 if (partial < 0) affixlen = 99; /* So that "NULL" prints */
614 debug_printf_indent("search_find: file=\"%s\"\n key=\"%s\" "
615 "partial=%d affix=%.*s starflags=%x\n",
616 (filename == NULL)? US"NULL" : filename,
617 keystring, partial, affixlen, affix, starflags);
620 /* Arrange to put this database at the top of the LRU chain if it is a type
621 that opens real files. */
623 if ( open_top != (tree_node *)handle
624 && lookup_list[t->name[0]-'0']->type == lookup_absfile)
626 search_cache *c = (search_cache *)(t->data.ptr);
627 tree_node *up = c->up;
628 tree_node *down = c->down;
630 /* Cut it out of the list. A newly opened file will a NULL up pointer.
631 Otherwise there will be a non-NULL up pointer, since we checked above that
632 this block isn't already at the top of the list. */
636 ((search_cache *)(up->data.ptr))->down = down;
638 ((search_cache *)(down->data.ptr))->up = up;
643 /* Now put it at the head of the list. */
647 if (!open_top) open_bot = t;
648 else ((search_cache *)(open_top->data.ptr))->up = t;
654 debug_printf_indent("LRU list:\n");
655 for (tree_node *t = open_top; t; )
657 search_cache *c = (search_cache *)(t->data.ptr);
658 debug_printf_indent(" %s\n", t->name);
659 if (t == open_bot) debug_printf_indent(" End\n");
664 /* First of all, try to match the key string verbatim. If matched a complete
665 entry but could have been partial, flag to set up variables. */
667 yield = internal_search_find(handle, filename, keystring);
668 if (f.search_find_defer) return NULL;
670 if (yield) { 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 is_tainted(keystring) || is_tainted(affix));
689 Ustrncpy(keystring2, affix, affixlen);
690 Ustrcpy(keystring2 + affixlen, keystring);
691 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring2);
692 yield = internal_search_find(handle, filename, keystring2);
693 if (f.search_find_defer) return NULL;
696 /* The key in its entirety did not match a wild entry; try chopping off
697 leading components. */
702 uschar *keystring3 = keystring2 + affixlen;
703 uschar *s = keystring3;
704 while (*s != 0) if (*s++ == '.') dotcount++;
706 while (dotcount-- >= partial)
708 while (*keystring3 != 0 && *keystring3 != '.') keystring3++;
710 /* If we get right to the end of the string (which will be the last time
711 through this loop), we've failed if the affix is null. Otherwise do one
712 last lookup for the affix itself, but if it is longer than 1 character,
713 remove the last character if it is ".". */
715 if (*keystring3 == 0)
717 if (affixlen < 1) break;
718 if (affixlen > 1 && affix[affixlen-1] == '.') affixlen--;
719 Ustrncpy(keystring2, affix, affixlen);
720 keystring2[affixlen] = 0;
721 keystring3 = keystring2;
725 keystring3 -= affixlen - 1;
726 if (affixlen > 0) Ustrncpy(keystring3, affix, affixlen);
729 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring3);
730 yield = internal_search_find(handle, filename, keystring3);
731 if (f.search_find_defer) return NULL;
734 /* First variable is the wild part; second is the fixed part. Take care
735 to get it right when keystring3 is just "*". */
737 if (expand_setup && *expand_setup >= 0)
739 int fixedlength = Ustrlen(keystring3) - affixlen;
740 int wildlength = Ustrlen(keystring) - fixedlength - 1;
742 expand_nstring[*expand_setup] = keystring;
743 expand_nlength[*expand_setup] = wildlength;
745 expand_nstring[*expand_setup] = keystring + wildlength + 1;
746 expand_nlength[*expand_setup] = (fixedlength < 0)? 0 : fixedlength;
750 keystring3 += affixlen;
754 else set_null_wild = TRUE; /* Matched a wild entry without any wild part */
757 /* If nothing has been matched, but the option to look for "*@" is set, try
758 replacing everything to the left of @ by *. After a match, the wild part
759 is set to the string to the left of the @. */
761 if (!yield && starflags & SEARCH_STARAT)
763 uschar *atat = Ustrrchr(keystring, '@');
764 if (atat != NULL && atat > keystring)
767 savechar = *(--atat);
770 DEBUG(D_lookup) debug_printf_indent("trying default match %s\n", atat);
771 yield = internal_search_find(handle, filename, atat);
773 if (f.search_find_defer) return NULL;
775 if (yield && expand_setup && *expand_setup >= 0)
778 expand_nstring[*expand_setup] = keystring;
779 expand_nlength[*expand_setup] = atat - keystring + 1;
781 expand_nstring[*expand_setup] = keystring;
782 expand_nlength[*expand_setup] = 0;
787 /* If we still haven't matched anything, and the option to look for "*" is set,
788 try that. If we do match, the first variable (the wild part) is the whole key,
789 and the second is empty. */
791 if (!yield && starflags & (SEARCH_STAR|SEARCH_STARAT))
793 DEBUG(D_lookup) debug_printf_indent("trying to match *\n");
794 yield = internal_search_find(handle, filename, US"*");
795 if (yield && expand_setup && *expand_setup >= 0)
798 expand_nstring[*expand_setup] = keystring;
799 expand_nlength[*expand_setup] = Ustrlen(keystring);
801 expand_nstring[*expand_setup] = keystring;
802 expand_nlength[*expand_setup] = 0;
806 /* If this was a potentially partial lookup, and we matched either a
807 complete non-wild domain entry, or we matched a wild-carded entry without
808 chopping off any of the domain components, set up the expansion variables
809 (if required) so that the first one is empty, and the second one is the
810 fixed part of the domain. The set_null_wild flag is set only when yield is not
813 if (set_null_wild && expand_setup && *expand_setup >= 0)
816 expand_nstring[*expand_setup] = keystring;
817 expand_nlength[*expand_setup] = 0;
819 expand_nstring[*expand_setup] = keystring;
820 expand_nlength[*expand_setup] = Ustrlen(keystring);
826 /* End of search.c */