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
118 opts where to put the options
120 Returns: +ve => valid lookup name; value is offset in lookup_list
121 -ve => invalid name; message in search_error_message.
125 search_findtype_partial(const uschar *name, int *ptypeptr, const uschar **ptypeaff,
126 int *afflen, int *starflags, const uschar ** opts)
130 const uschar *ss = name;
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 *@,
173 and then by options starting with a "," */
176 if ((t = Ustrchr(ss, '*')))
179 *starflags |= (t[1] == '@' ? SEARCH_STARAT : SEARCH_STAR);
184 if ((t = Ustrchr(t, ',')))
187 if (l < len) len = l;
188 *opts = string_copy(t+1);
193 /* Check for the individual search type. Only those that are actually in the
194 binary are valid. For query-style types, "partial" and default types are
197 stype = search_findtype(ss, len);
198 if (stype >= 0 && mac_islookup(stype, lookup_querystyle))
202 search_error_message = string_sprintf("\"partial\" is not permitted "
203 "for lookup type \"%s\"", ss);
206 if ((*starflags & (SEARCH_STAR|SEARCH_STARAT)) != 0)
208 search_error_message = string_sprintf("defaults using \"*\" or \"*@\" are "
209 "not permitted for lookup type \"%s\"", ss);
220 /*************************************************
221 * Release cached resources *
222 *************************************************/
224 /* When search_open is called it caches the "file" that it opens in
225 search_tree. The name of the tree node is a concatenation of the search type
226 with the file name. For query-style lookups, the file name is empty. Real files
227 are normally closed only when this tidyup routine is called, typically at the
228 end of sections of code where a number of lookups might occur. However, if too
229 many files are open simultaneously, some get closed beforehand. They can't be
230 removed from the tree. There is also a general tidyup function which is called
231 for the lookup driver, if it exists.
233 First, there is an internal, recursive subroutine.
235 Argument: a pointer to a search_openfile tree node
240 tidyup_subtree(tree_node *t)
242 search_cache *c = (search_cache *)(t->data.ptr);
243 if (t->left != NULL) tidyup_subtree(t->left);
244 if (t->right != NULL) tidyup_subtree(t->right);
247 lookup_list[c->search_type]->close != NULL)
248 lookup_list[c->search_type]->close(c->handle);
252 /* The external entry point
261 int old_pool = store_pool;
263 DEBUG(D_lookup) debug_printf_indent("search_tidyup called\n");
265 /* Close individually each cached open file. */
267 store_pool = POOL_SEARCH;
270 tidyup_subtree(search_tree);
273 open_top = open_bot = NULL;
276 /* Call the general tidyup entry for any drivers that have one. */
278 for (int i = 0; i < lookup_list_count; i++) if (lookup_list[i]->tidy)
279 (lookup_list[i]->tidy)();
281 if (search_reset_point) search_reset_point = store_reset(search_reset_point);
282 store_pool = old_pool;
288 /*************************************************
289 * Open search database *
290 *************************************************/
292 /* A mode, and lists of owners and groups, are passed over for checking in
293 the cases where the database is one or more files. Return NULL, with a message
294 pointed to by message, in cases of error.
296 For search types that use a file or files, check up on the mode after
297 opening. It is tempting to do a stat before opening the file, and use it as
298 an existence check. However, doing that opens a small security loophole in
299 that the status could be changed before the file is opened. Can't quite see
300 what problems this might lead to, but you can't be too careful where security
301 is concerned. Fstat() on an open file can normally be expected to succeed,
302 but there are some NFS states where it does not.
304 There are two styles of query: (1) in the "single-key+file" style, a single
305 key string and a file name are given, for example, for linear searches, DBM
306 files, or for NIS. (2) In the "query" style, no "filename" is given; instead
307 just a single query string is passed. This applies to multiple-key lookup
310 Before opening, scan the tree of cached files to see if this file is already
311 open for the correct search type. If so, return the saved handle. If not, put
312 the handle in the tree for possible subsequent use. See search_tidyup above for
313 closing all the cached files.
315 A count of open databases which use real files is maintained, and if this
316 gets too large, we have to close a cached file. Its entry remains in the tree,
317 but is marked closed.
320 filename the name of the file for single-key+file style lookups,
321 NULL for query-style lookups
322 search_type the type of search required
323 modemask if a real single file is used, this specifies mode bits that
324 must not be set; otherwise it is ignored
325 owners if a real single file is used, this specifies the possible
326 owners of the file; otherwise it is ignored
327 owngroups if a real single file is used, this specifies the possible
328 group owners of the file; otherwise it is ignored
330 Returns: an identifying handle for the open database;
331 this is the pointer to the tree block in the
332 cache of open files; return NULL on open failure, with
333 a message in search_error_message
337 search_open(const uschar * filename, int search_type, int modemask,
338 uid_t * owners, gid_t * owngroups)
343 lookup_info *lk = lookup_list[search_type];
344 uschar keybuffer[256];
345 int old_pool = store_pool;
347 if (filename && is_tainted(filename))
349 log_write(0, LOG_MAIN|LOG_PANIC,
350 "Tainted filename for search: '%s'", filename);
354 /* Change to the search store pool and remember our reset point */
356 store_pool = POOL_SEARCH;
357 if (!search_reset_point) search_reset_point = store_mark();
359 DEBUG(D_lookup) debug_printf_indent("search_open: %s \"%s\"\n", lk->name,
360 filename ? filename : US"NULL");
362 /* See if we already have this open for this type of search, and if so,
363 pass back the tree block as the handle. The key for the tree node is the search
364 type plus '0' concatenated with the file name. There may be entries in the tree
365 with closed files if a lot of files have been opened. */
367 sprintf(CS keybuffer, "%c%.254s", search_type + '0',
368 filename ? filename : US"");
370 if ((t = tree_search(search_tree, keybuffer)))
372 if ((c = (search_cache *)t->data.ptr)->handle)
374 DEBUG(D_lookup) debug_printf_indent(" cached open\n");
375 store_pool = old_pool;
378 DEBUG(D_lookup) debug_printf_indent(" cached closed\n");
381 /* Otherwise, we need to open the file or database - each search type has its
382 own code, which is now split off into separately compiled modules. Before doing
383 this, if the search type is one that uses real files, check on the number that
384 we are holding open in the cache. If the limit is reached, close the least
385 recently used one. */
387 if (lk->type == lookup_absfile && open_filecount >= lookup_open_max)
389 log_write(0, LOG_MAIN|LOG_PANIC, "too many lookups open, but can't find "
393 search_cache *c = (search_cache *)(open_bot->data.ptr);
394 DEBUG(D_lookup) debug_printf_indent("Too many lookup files open\n closing %s\n",
396 if ((open_bot = c->up))
397 ((search_cache *)(open_bot->data.ptr))->down = NULL;
400 ((lookup_list[c->search_type])->close)(c->handle);
405 /* If opening is successful, call the file-checking function if there is one,
406 and if all is still well, enter the open database into the tree. */
408 if (!(handle = (lk->open)(filename, &search_error_message)))
410 store_pool = old_pool;
415 && !lk->check(handle, filename, modemask, owners, owngroups,
416 &search_error_message))
419 store_pool = old_pool;
423 /* If this is a search type that uses real files, keep count. */
425 if (lk->type == lookup_absfile) open_filecount++;
427 /* If we found a previously opened entry in the tree, re-use it; otherwise
428 insert a new entry. On re-use, leave any cached lookup data and the lookup
433 t = store_get(sizeof(tree_node) + Ustrlen(keybuffer), FALSE);
434 t->data.ptr = c = store_get(sizeof(search_cache), FALSE);
435 c->item_cache = NULL;
436 Ustrcpy(t->name, keybuffer);
437 tree_insertnode(&search_tree, t);
439 else c = t->data.ptr;
442 c->search_type = search_type;
443 c->up = c->down = NULL;
445 store_pool = old_pool;
453 /*************************************************
454 * Internal function: Find one item in database *
455 *************************************************/
457 /* The answer is always put into dynamic store. The last lookup for each handle
461 handle the handle from search_open; points to tree node
462 filename the filename that was handed to search_open, or
463 NULL for query-style searches
464 keystring the keystring for single-key+file lookups, or
465 the querystring for query-style lookups
466 opts type-specific options
468 Returns: a pointer to a dynamic string containing the answer,
469 or NULL if the query failed or was deferred; in the
470 latter case, search_find_defer is set TRUE; after an unusual
471 failure, there may be a message in search_error_message.
475 internal_search_find(void * handle, const uschar * filename, uschar * keystring,
478 tree_node * t = (tree_node *)handle;
479 search_cache * c = (search_cache *)(t->data.ptr);
480 expiring_data * e = NULL; /* compiler quietening */
481 uschar * data = NULL;
482 int search_type = t->name[0] - '0';
483 int old_pool = store_pool;
485 /* Lookups that return DEFER may not always set an error message. So that
486 the callers don't have to test for NULL, set an empty string. */
488 search_error_message = US"";
489 f.search_find_defer = FALSE;
491 DEBUG(D_lookup) debug_printf_indent("internal_search_find: file=\"%s\"\n "
492 "type=%s key=\"%s\" opts=%s%s%s\n", filename,
493 lookup_list[search_type]->name, keystring,
494 opts ? "\"" : "", opts, opts ? "\"" : "");
496 /* Insurance. If the keystring is empty, just fail. */
498 if (keystring[0] == 0) return NULL;
500 /* Use the special store pool for search data */
502 store_pool = POOL_SEARCH;
504 /* Look up the data for the key, unless it is already in the cache for this
505 file. No need to check c->item_cache for NULL, tree_search will do so. */
507 if ( (t = tree_search(c->item_cache, keystring))
508 && (!(e = t->data.ptr)->expiry || e->expiry > time(NULL))
509 && (!opts && !e->opts || opts && e->opts && Ustrcmp(opts, e->opts) == 0)
511 { /* Data was in the cache already; set the pointer from the tree node */
513 DEBUG(D_lookup) debug_printf_indent("cached data used for lookup of %s%s%s\n",
515 filename ? US"\n in " : US"", filename ? filename : US"");
519 uint do_cache = UINT_MAX;
520 int keylength = Ustrlen(keystring);
525 debug_printf_indent("cached data found but either wrong opts or dated; ");
526 debug_printf_indent("%s lookup required for %s%s%s\n",
527 filename ? US"file" : US"database",
529 filename ? US"\n in " : US"", filename ? filename : US"");
532 /* Call the code for the different kinds of search. DEFER is handled
533 like FAIL, except that search_find_defer is set so the caller can
534 distinguish if necessary. */
536 if (lookup_list[search_type]->find(c->handle, filename, keystring, keylength,
537 &data, &search_error_message, &do_cache, opts) == DEFER)
538 f.search_find_defer = TRUE;
540 /* A record that has been found is now in data, which is either NULL
541 or points to a bit of dynamic store. Cache the result of the lookup if
542 caching is permitted. Lookups can disable caching, when they did something
543 that changes their data. The mysql and pgsql lookups do this when an
544 UPDATE/INSERT query was executed. */
548 int len = keylength + 1;
550 if (t) /* Previous, out-of-date cache entry. Update with the */
551 { /* new result and forget the old one */
552 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
558 e = store_get(sizeof(expiring_data) + sizeof(tree_node) + len, is_tainted(keystring));
559 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
562 t = (tree_node *)(e+1);
563 memcpy(t->name, keystring, len);
565 tree_insertnode(&c->item_cache, t);
569 /* If caching was disabled, empty the cache tree. We just set the cache
570 pointer to NULL here, because we cannot release the store at this stage. */
574 DEBUG(D_lookup) debug_printf_indent("lookup forced cache cleanup\n");
575 c->item_cache = NULL;
582 debug_printf_indent("lookup yielded: %s\n", data);
583 else if (f.search_find_defer)
584 debug_printf_indent("lookup deferred: %s\n", search_error_message);
585 else debug_printf_indent("lookup failed\n");
588 /* Return it in new dynamic store in the regular pool */
590 store_pool = old_pool;
591 return data ? string_copy(data) : NULL;
597 /*************************************************
598 * Find one item in database, possibly wildcarded *
599 *************************************************/
601 /* This function calls the internal function above; once only if there
602 is no partial matching, but repeatedly when partial matching is requested.
605 handle the handle from search_open
606 filename the filename that was handed to search_open, or
607 NULL for query-style searches
608 keystring the keystring for single-key+file lookups, or
609 the querystring for query-style lookups
610 partial -1 means no partial matching;
611 otherwise it's the minimum number of components;
612 affix the affix string for partial matching
613 affixlen the length of the affix string
614 starflags SEARCH_STAR and SEARCH_STARAT flags
615 expand_setup pointer to offset for setting up expansion strings;
617 opts type-specific options
619 Returns: a pointer to a dynamic string containing the answer,
620 or NULL if the query failed or was deferred; in the
621 latter case, search_find_defer is set TRUE
625 search_find(void * handle, const uschar * filename, uschar * keystring,
626 int partial, const uschar * affix, int affixlen, int starflags,
627 int * expand_setup, const uschar * opts)
629 tree_node *t = (tree_node *)handle;
630 BOOL set_null_wild = FALSE;
635 if (partial < 0) affixlen = 99; /* So that "NULL" prints */
636 debug_printf_indent("search_find: file=\"%s\"\n key=\"%s\" "
637 "partial=%d affix=%.*s starflags=%x opts=%s%s%s\n",
638 filename ? filename : US"NULL",
639 keystring, partial, affixlen, affix, starflags,
640 opts ? "\"" : "", opts, opts ? "\"" : "");
644 /* Arrange to put this database at the top of the LRU chain if it is a type
645 that opens real files. */
647 if ( open_top != (tree_node *)handle
648 && lookup_list[t->name[0]-'0']->type == lookup_absfile)
650 search_cache *c = (search_cache *)(t->data.ptr);
651 tree_node *up = c->up;
652 tree_node *down = c->down;
654 /* Cut it out of the list. A newly opened file will a NULL up pointer.
655 Otherwise there will be a non-NULL up pointer, since we checked above that
656 this block isn't already at the top of the list. */
660 ((search_cache *)(up->data.ptr))->down = down;
662 ((search_cache *)(down->data.ptr))->up = up;
667 /* Now put it at the head of the list. */
671 if (!open_top) open_bot = t;
672 else ((search_cache *)(open_top->data.ptr))->up = t;
678 debug_printf_indent("LRU list:\n");
679 for (tree_node *t = open_top; t; )
681 search_cache *c = (search_cache *)(t->data.ptr);
682 debug_printf_indent(" %s\n", t->name);
683 if (t == open_bot) debug_printf_indent(" End\n");
688 /* First of all, try to match the key string verbatim. If matched a complete
689 entry but could have been partial, flag to set up variables. */
691 yield = internal_search_find(handle, filename, keystring, opts);
692 if (f.search_find_defer) return NULL;
694 if (yield) { if (partial >= 0) set_null_wild = TRUE; }
696 /* Not matched a complete entry; handle partial lookups, but only if the full
697 search didn't defer. Don't use string_sprintf() to construct the initial key,
698 just in case the original key is too long for the string_sprintf() buffer (it
699 *has* happened!). The case of a zero-length affix has to be treated specially.
702 else if (partial >= 0)
704 int len = Ustrlen(keystring);
707 /* Try with the affix on the front, except for a zero-length affix */
709 if (affixlen == 0) keystring2 = keystring; else
711 keystring2 = store_get(len + affixlen + 1,
712 is_tainted(keystring) || is_tainted(affix));
713 Ustrncpy(keystring2, affix, affixlen);
714 Ustrcpy(keystring2 + affixlen, keystring);
715 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring2);
716 yield = internal_search_find(handle, filename, keystring2, opts);
717 if (f.search_find_defer) return NULL;
720 /* The key in its entirety did not match a wild entry; try chopping off
721 leading components. */
726 uschar *keystring3 = keystring2 + affixlen;
727 uschar *s = keystring3;
728 while (*s != 0) if (*s++ == '.') dotcount++;
730 while (dotcount-- >= partial)
732 while (*keystring3 != 0 && *keystring3 != '.') keystring3++;
734 /* If we get right to the end of the string (which will be the last time
735 through this loop), we've failed if the affix is null. Otherwise do one
736 last lookup for the affix itself, but if it is longer than 1 character,
737 remove the last character if it is ".". */
739 if (*keystring3 == 0)
741 if (affixlen < 1) break;
742 if (affixlen > 1 && affix[affixlen-1] == '.') affixlen--;
743 Ustrncpy(keystring2, affix, affixlen);
744 keystring2[affixlen] = 0;
745 keystring3 = keystring2;
749 keystring3 -= affixlen - 1;
750 if (affixlen > 0) Ustrncpy(keystring3, affix, affixlen);
753 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring3);
754 yield = internal_search_find(handle, filename, keystring3, opts);
755 if (f.search_find_defer) return NULL;
758 /* First variable is the wild part; second is the fixed part. Take care
759 to get it right when keystring3 is just "*". */
761 if (expand_setup && *expand_setup >= 0)
763 int fixedlength = Ustrlen(keystring3) - affixlen;
764 int wildlength = Ustrlen(keystring) - fixedlength - 1;
766 expand_nstring[*expand_setup] = keystring;
767 expand_nlength[*expand_setup] = wildlength;
769 expand_nstring[*expand_setup] = keystring + wildlength + 1;
770 expand_nlength[*expand_setup] = (fixedlength < 0)? 0 : fixedlength;
774 keystring3 += affixlen;
778 else set_null_wild = TRUE; /* Matched a wild entry without any wild part */
781 /* If nothing has been matched, but the option to look for "*@" is set, try
782 replacing everything to the left of @ by *. After a match, the wild part
783 is set to the string to the left of the @. */
785 if (!yield && starflags & SEARCH_STARAT)
787 uschar *atat = Ustrrchr(keystring, '@');
788 if (atat != NULL && atat > keystring)
791 savechar = *(--atat);
794 DEBUG(D_lookup) debug_printf_indent("trying default match %s\n", atat);
795 yield = internal_search_find(handle, filename, atat, opts);
797 if (f.search_find_defer) return NULL;
799 if (yield && expand_setup && *expand_setup >= 0)
802 expand_nstring[*expand_setup] = keystring;
803 expand_nlength[*expand_setup] = atat - keystring + 1;
805 expand_nstring[*expand_setup] = keystring;
806 expand_nlength[*expand_setup] = 0;
811 /* If we still haven't matched anything, and the option to look for "*" is set,
812 try that. If we do match, the first variable (the wild part) is the whole key,
813 and the second is empty. */
815 if (!yield && starflags & (SEARCH_STAR|SEARCH_STARAT))
817 DEBUG(D_lookup) debug_printf_indent("trying to match *\n");
818 yield = internal_search_find(handle, filename, US"*", opts);
819 if (yield && expand_setup && *expand_setup >= 0)
822 expand_nstring[*expand_setup] = keystring;
823 expand_nlength[*expand_setup] = Ustrlen(keystring);
825 expand_nstring[*expand_setup] = keystring;
826 expand_nlength[*expand_setup] = 0;
830 /* If this was a potentially partial lookup, and we matched either a
831 complete non-wild domain entry, or we matched a wild-carded entry without
832 chopping off any of the domain components, set up the expansion variables
833 (if required) so that the first one is empty, and the second one is the
834 fixed part of the domain. The set_null_wild flag is set only when yield is not
837 if (set_null_wild && expand_setup && *expand_setup >= 0)
840 expand_nstring[*expand_setup] = keystring;
841 expand_nlength[*expand_setup] = 0;
843 expand_nstring[*expand_setup] = keystring;
844 expand_nlength[*expand_setup] = Ustrlen(keystring);
850 /* End of search.c */