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) tidyup_subtree(t->left);
244 if (t->right) tidyup_subtree(t->right);
245 if (c && c->handle && lookup_list[c->search_type]->close)
246 lookup_list[c->search_type]->close(c->handle);
250 /* The external entry point
259 int old_pool = store_pool;
261 DEBUG(D_lookup) debug_printf_indent("search_tidyup called\n");
263 /* Close individually each cached open file. */
265 store_pool = POOL_SEARCH;
268 tidyup_subtree(search_tree);
271 open_top = open_bot = NULL;
274 /* Call the general tidyup entry for any drivers that have one. */
276 for (int i = 0; i < lookup_list_count; i++) if (lookup_list[i]->tidy)
277 (lookup_list[i]->tidy)();
279 if (search_reset_point) search_reset_point = store_reset(search_reset_point);
280 store_pool = old_pool;
286 /*************************************************
287 * Open search database *
288 *************************************************/
290 /* A mode, and lists of owners and groups, are passed over for checking in
291 the cases where the database is one or more files. Return NULL, with a message
292 pointed to by message, in cases of error.
294 For search types that use a file or files, check up on the mode after
295 opening. It is tempting to do a stat before opening the file, and use it as
296 an existence check. However, doing that opens a small security loophole in
297 that the status could be changed before the file is opened. Can't quite see
298 what problems this might lead to, but you can't be too careful where security
299 is concerned. Fstat() on an open file can normally be expected to succeed,
300 but there are some NFS states where it does not.
302 There are two styles of query: (1) in the "single-key+file" style, a single
303 key string and a file name are given, for example, for linear searches, DBM
304 files, or for NIS. (2) In the "query" style, no "filename" is given; instead
305 just a single query string is passed. This applies to multiple-key lookup
308 Before opening, scan the tree of cached files to see if this file is already
309 open for the correct search type. If so, return the saved handle. If not, put
310 the handle in the tree for possible subsequent use. See search_tidyup above for
311 closing all the cached files.
313 A count of open databases which use real files is maintained, and if this
314 gets too large, we have to close a cached file. Its entry remains in the tree,
315 but is marked closed.
318 filename the name of the file for single-key+file style lookups,
319 NULL for query-style lookups
320 search_type the type of search required
321 modemask if a real single file is used, this specifies mode bits that
322 must not be set; otherwise it is ignored
323 owners if a real single file is used, this specifies the possible
324 owners of the file; otherwise it is ignored
325 owngroups if a real single file is used, this specifies the possible
326 group owners of the file; otherwise it is ignored
328 Returns: an identifying handle for the open database;
329 this is the pointer to the tree block in the
330 cache of open files; return NULL on open failure, with
331 a message in search_error_message
335 search_open(const uschar * filename, int search_type, int modemask,
336 uid_t * owners, gid_t * owngroups)
341 lookup_info *lk = lookup_list[search_type];
342 uschar keybuffer[256];
343 int old_pool = store_pool;
345 if (filename && is_tainted(filename))
347 log_write(0, LOG_MAIN|LOG_PANIC,
348 "Tainted filename for search: '%s'", filename);
352 /* Change to the search store pool and remember our reset point */
354 store_pool = POOL_SEARCH;
355 if (!search_reset_point) search_reset_point = store_mark();
357 DEBUG(D_lookup) debug_printf_indent("search_open: %s \"%s\"\n", lk->name,
358 filename ? filename : US"NULL");
360 /* See if we already have this open for this type of search, and if so,
361 pass back the tree block as the handle. The key for the tree node is the search
362 type plus '0' concatenated with the file name. There may be entries in the tree
363 with closed files if a lot of files have been opened. */
365 sprintf(CS keybuffer, "%c%.254s", search_type + '0',
366 filename ? filename : US"");
368 if ((t = tree_search(search_tree, keybuffer)))
370 if ((c = (search_cache *)t->data.ptr)->handle)
372 DEBUG(D_lookup) debug_printf_indent(" cached open\n");
373 store_pool = old_pool;
376 DEBUG(D_lookup) debug_printf_indent(" cached closed\n");
379 /* Otherwise, we need to open the file or database - each search type has its
380 own code, which is now split off into separately compiled modules. Before doing
381 this, if the search type is one that uses real files, check on the number that
382 we are holding open in the cache. If the limit is reached, close the least
383 recently used one. */
385 if (lk->type == lookup_absfile && open_filecount >= lookup_open_max)
387 log_write(0, LOG_MAIN|LOG_PANIC, "too many lookups open, but can't find "
391 search_cache *c = (search_cache *)(open_bot->data.ptr);
392 DEBUG(D_lookup) debug_printf_indent("Too many lookup files open\n closing %s\n",
394 if ((open_bot = c->up))
395 ((search_cache *)(open_bot->data.ptr))->down = NULL;
398 ((lookup_list[c->search_type])->close)(c->handle);
403 /* If opening is successful, call the file-checking function if there is one,
404 and if all is still well, enter the open database into the tree. */
406 if (!(handle = (lk->open)(filename, &search_error_message)))
408 store_pool = old_pool;
413 && !lk->check(handle, filename, modemask, owners, owngroups,
414 &search_error_message))
417 store_pool = old_pool;
421 /* If this is a search type that uses real files, keep count. */
423 if (lk->type == lookup_absfile) open_filecount++;
425 /* If we found a previously opened entry in the tree, re-use it; otherwise
426 insert a new entry. On re-use, leave any cached lookup data and the lookup
431 t = store_get(sizeof(tree_node) + Ustrlen(keybuffer), FALSE);
432 t->data.ptr = c = store_get(sizeof(search_cache), FALSE);
433 c->item_cache = NULL;
434 Ustrcpy(t->name, keybuffer);
435 tree_insertnode(&search_tree, t);
437 else c = t->data.ptr;
440 c->search_type = search_type;
441 c->up = c->down = NULL;
443 store_pool = old_pool;
451 /*************************************************
452 * Internal function: Find one item in database *
453 *************************************************/
455 /* The answer is always put into dynamic store. The last lookup for each handle
459 handle the handle from search_open; points to tree node
460 filename the filename that was handed to search_open, or
461 NULL for query-style searches
462 keystring the keystring for single-key+file lookups, or
463 the querystring for query-style lookups
464 opts type-specific options
466 Returns: a pointer to a dynamic string containing the answer,
467 or NULL if the query failed or was deferred; in the
468 latter case, search_find_defer is set TRUE; after an unusual
469 failure, there may be a message in search_error_message.
473 internal_search_find(void * handle, const uschar * filename, uschar * keystring,
476 tree_node * t = (tree_node *)handle;
477 search_cache * c = (search_cache *)(t->data.ptr);
478 expiring_data * e = NULL; /* compiler quietening */
479 uschar * data = NULL;
480 int search_type = t->name[0] - '0';
481 int old_pool = store_pool;
483 /* Lookups that return DEFER may not always set an error message. So that
484 the callers don't have to test for NULL, set an empty string. */
486 search_error_message = US"";
487 f.search_find_defer = FALSE;
489 DEBUG(D_lookup) debug_printf_indent("internal_search_find: file=\"%s\"\n "
490 "type=%s key=\"%s\" opts=%s%s%s\n", filename,
491 lookup_list[search_type]->name, keystring,
492 opts ? "\"" : "", opts, opts ? "\"" : "");
494 /* Insurance. If the keystring is empty, just fail. */
496 if (keystring[0] == 0) return NULL;
498 /* Use the special store pool for search data */
500 store_pool = POOL_SEARCH;
502 /* Look up the data for the key, unless it is already in the cache for this
503 file. No need to check c->item_cache for NULL, tree_search will do so. */
505 if ( (t = tree_search(c->item_cache, keystring))
506 && (!(e = t->data.ptr)->expiry || e->expiry > time(NULL))
507 && (!opts && !e->opts || opts && e->opts && Ustrcmp(opts, e->opts) == 0)
509 { /* Data was in the cache already; set the pointer from the tree node */
511 DEBUG(D_lookup) debug_printf_indent("cached data used for lookup of %s%s%s\n",
513 filename ? US"\n in " : US"", filename ? filename : US"");
517 uint do_cache = UINT_MAX;
518 int keylength = Ustrlen(keystring);
523 debug_printf_indent("cached data found but %s; ",
524 e->expiry && e->expiry <= time(NULL) ? "out-of-date" : "wrong opts");
525 debug_printf_indent("%s lookup required for %s%s%s\n",
526 filename ? US"file" : US"database",
528 filename ? US"\n in " : US"", filename ? filename : US"");
531 /* Call the code for the different kinds of search. DEFER is handled
532 like FAIL, except that search_find_defer is set so the caller can
533 distinguish if necessary. */
535 if (lookup_list[search_type]->find(c->handle, filename, keystring, keylength,
536 &data, &search_error_message, &do_cache, opts) == DEFER)
537 f.search_find_defer = TRUE;
539 /* A record that has been found is now in data, which is either NULL
540 or points to a bit of dynamic store. Cache the result of the lookup if
541 caching is permitted. Lookups can disable caching, when they did something
542 that changes their data. The mysql and pgsql lookups do this when an
543 UPDATE/INSERT query was executed. */
547 if (!t) /* No existing entry. Create new one. */
549 int len = keylength + 1;
550 e = store_get(sizeof(expiring_data) + sizeof(tree_node) + len, is_tainted(keystring));
551 t = (tree_node *)(e+1);
552 memcpy(t->name, keystring, len);
554 tree_insertnode(&c->item_cache, t);
556 /* Else previous, out-of-date cache entry. Update with the */
557 /* new result and forget the old one */
558 e->expiry = do_cache == UINT_MAX ? 0 : time(NULL)+do_cache;
559 e->opts = opts ? string_copy(opts) : NULL;
563 /* If caching was disabled, empty the cache tree. We just set the cache
564 pointer to NULL here, because we cannot release the store at this stage. */
568 DEBUG(D_lookup) debug_printf_indent("lookup forced cache cleanup\n");
569 c->item_cache = NULL; /* forget all lookups on this connection */
576 debug_printf_indent("lookup yielded: %s\n", data);
577 else if (f.search_find_defer)
578 debug_printf_indent("lookup deferred: %s\n", search_error_message);
579 else debug_printf_indent("lookup failed\n");
582 /* Return it in new dynamic store in the regular pool */
584 store_pool = old_pool;
585 return data ? string_copy(data) : NULL;
591 /*************************************************
592 * Find one item in database, possibly wildcarded *
593 *************************************************/
595 /* This function calls the internal function above; once only if there
596 is no partial matching, but repeatedly when partial matching is requested.
599 handle the handle from search_open
600 filename the filename that was handed to search_open, or
601 NULL for query-style searches
602 keystring the keystring for single-key+file lookups, or
603 the querystring for query-style lookups
604 partial -1 means no partial matching;
605 otherwise it's the minimum number of components;
606 affix the affix string for partial matching
607 affixlen the length of the affix string
608 starflags SEARCH_STAR and SEARCH_STARAT flags
609 expand_setup pointer to offset for setting up expansion strings;
611 opts type-specific options
613 Returns: a pointer to a dynamic string containing the answer,
614 or NULL if the query failed or was deferred; in the
615 latter case, search_find_defer is set TRUE
619 search_find(void * handle, const uschar * filename, uschar * keystring,
620 int partial, const uschar * affix, int affixlen, int starflags,
621 int * expand_setup, const uschar * opts)
623 tree_node *t = (tree_node *)handle;
624 BOOL set_null_wild = FALSE;
629 if (partial < 0) affixlen = 99; /* So that "NULL" prints */
630 debug_printf_indent("search_find: file=\"%s\"\n key=\"%s\" "
631 "partial=%d affix=%.*s starflags=%x opts=%s%s%s\n",
632 filename ? filename : US"NULL",
633 keystring, partial, affixlen, affix, starflags,
634 opts ? "\"" : "", opts, opts ? "\"" : "");
638 /* Arrange to put this database at the top of the LRU chain if it is a type
639 that opens real files. */
641 if ( open_top != (tree_node *)handle
642 && lookup_list[t->name[0]-'0']->type == lookup_absfile)
644 search_cache *c = (search_cache *)(t->data.ptr);
645 tree_node *up = c->up;
646 tree_node *down = c->down;
648 /* Cut it out of the list. A newly opened file will a NULL up pointer.
649 Otherwise there will be a non-NULL up pointer, since we checked above that
650 this block isn't already at the top of the list. */
654 ((search_cache *)(up->data.ptr))->down = down;
656 ((search_cache *)(down->data.ptr))->up = up;
661 /* Now put it at the head of the list. */
665 if (!open_top) open_bot = t;
666 else ((search_cache *)(open_top->data.ptr))->up = t;
672 debug_printf_indent("LRU list:\n");
673 for (tree_node *t = open_top; t; )
675 search_cache *c = (search_cache *)(t->data.ptr);
676 debug_printf_indent(" %s\n", t->name);
677 if (t == open_bot) debug_printf_indent(" End\n");
682 /* First of all, try to match the key string verbatim. If matched a complete
683 entry but could have been partial, flag to set up variables. */
685 yield = internal_search_find(handle, filename, keystring, opts);
686 if (f.search_find_defer) return NULL;
688 if (yield) { if (partial >= 0) set_null_wild = TRUE; }
690 /* Not matched a complete entry; handle partial lookups, but only if the full
691 search didn't defer. Don't use string_sprintf() to construct the initial key,
692 just in case the original key is too long for the string_sprintf() buffer (it
693 *has* happened!). The case of a zero-length affix has to be treated specially.
696 else if (partial >= 0)
698 int len = Ustrlen(keystring);
701 /* Try with the affix on the front, except for a zero-length affix */
703 if (affixlen == 0) keystring2 = keystring; else
705 keystring2 = store_get(len + affixlen + 1,
706 is_tainted(keystring) || is_tainted(affix));
707 Ustrncpy(keystring2, affix, affixlen);
708 Ustrcpy(keystring2 + affixlen, keystring);
709 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring2);
710 yield = internal_search_find(handle, filename, keystring2, opts);
711 if (f.search_find_defer) return NULL;
714 /* The key in its entirety did not match a wild entry; try chopping off
715 leading components. */
720 uschar *keystring3 = keystring2 + affixlen;
721 uschar *s = keystring3;
722 while (*s != 0) if (*s++ == '.') dotcount++;
724 while (dotcount-- >= partial)
726 while (*keystring3 != 0 && *keystring3 != '.') keystring3++;
728 /* If we get right to the end of the string (which will be the last time
729 through this loop), we've failed if the affix is null. Otherwise do one
730 last lookup for the affix itself, but if it is longer than 1 character,
731 remove the last character if it is ".". */
733 if (*keystring3 == 0)
735 if (affixlen < 1) break;
736 if (affixlen > 1 && affix[affixlen-1] == '.') affixlen--;
737 Ustrncpy(keystring2, affix, affixlen);
738 keystring2[affixlen] = 0;
739 keystring3 = keystring2;
743 keystring3 -= affixlen - 1;
744 if (affixlen > 0) Ustrncpy(keystring3, affix, affixlen);
747 DEBUG(D_lookup) debug_printf_indent("trying partial match %s\n", keystring3);
748 yield = internal_search_find(handle, filename, keystring3, opts);
749 if (f.search_find_defer) return NULL;
752 /* First variable is the wild part; second is the fixed part. Take care
753 to get it right when keystring3 is just "*". */
755 if (expand_setup && *expand_setup >= 0)
757 int fixedlength = Ustrlen(keystring3) - affixlen;
758 int wildlength = Ustrlen(keystring) - fixedlength - 1;
760 expand_nstring[*expand_setup] = keystring;
761 expand_nlength[*expand_setup] = wildlength;
763 expand_nstring[*expand_setup] = keystring + wildlength + 1;
764 expand_nlength[*expand_setup] = (fixedlength < 0)? 0 : fixedlength;
768 keystring3 += affixlen;
772 else set_null_wild = TRUE; /* Matched a wild entry without any wild part */
775 /* If nothing has been matched, but the option to look for "*@" is set, try
776 replacing everything to the left of @ by *. After a match, the wild part
777 is set to the string to the left of the @. */
779 if (!yield && starflags & SEARCH_STARAT)
781 uschar *atat = Ustrrchr(keystring, '@');
782 if (atat != NULL && atat > keystring)
785 savechar = *(--atat);
788 DEBUG(D_lookup) debug_printf_indent("trying default match %s\n", atat);
789 yield = internal_search_find(handle, filename, atat, opts);
791 if (f.search_find_defer) return NULL;
793 if (yield && expand_setup && *expand_setup >= 0)
796 expand_nstring[*expand_setup] = keystring;
797 expand_nlength[*expand_setup] = atat - keystring + 1;
799 expand_nstring[*expand_setup] = keystring;
800 expand_nlength[*expand_setup] = 0;
805 /* If we still haven't matched anything, and the option to look for "*" is set,
806 try that. If we do match, the first variable (the wild part) is the whole key,
807 and the second is empty. */
809 if (!yield && starflags & (SEARCH_STAR|SEARCH_STARAT))
811 DEBUG(D_lookup) debug_printf_indent("trying to match *\n");
812 yield = internal_search_find(handle, filename, US"*", opts);
813 if (yield && expand_setup && *expand_setup >= 0)
816 expand_nstring[*expand_setup] = keystring;
817 expand_nlength[*expand_setup] = Ustrlen(keystring);
819 expand_nstring[*expand_setup] = keystring;
820 expand_nlength[*expand_setup] = 0;
824 /* If this was a potentially partial lookup, and we matched either a
825 complete non-wild domain entry, or we matched a wild-carded entry without
826 chopping off any of the domain components, set up the expansion variables
827 (if required) so that the first one is empty, and the second one is the
828 fixed part of the domain. The set_null_wild flag is set only when yield is not
831 if (set_null_wild && expand_setup && *expand_setup >= 0)
834 expand_nstring[*expand_setup] = keystring;
835 expand_nlength[*expand_setup] = 0;
837 expand_nstring[*expand_setup] = keystring;
838 expand_nlength[*expand_setup] = Ustrlen(keystring);
844 /* End of search.c */