1 /* $Cambridge: exim/src/src/lookups/cdb.c,v 1.1 2004/10/07 13:10:01 ph10 Exp $ */
3 /*************************************************
4 * Exim - an Internet mail transport agent *
5 *************************************************/
8 * Exim - CDB database lookup module
9 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
11 * Copyright (c) 1998 Nigel Metheringham, Planet Online Ltd
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version 2
16 * of the License, or (at your option) any later version.
18 * --------------------------------------------------------------
19 * Modified by PH for Exim 4:
20 * Changed over to using unsigned chars
21 * Makes use of lf_check_file() for file checking
22 * --------------------------------------------------------------
24 * This program is distributed in the hope that it will be useful,
25 * but WITHOUT ANY WARRANTY; without even the implied warranty of
26 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27 * GNU General Public License for more details.
29 * You should have received a copy of the GNU General Public License
30 * along with this program; if not, write to the Free Software
31 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
35 * This code implements Dan Bernstein's Constant DataBase (cdb) spec.
36 * Information, the spec and sample code for cdb can be obtained from
37 * http://www.pobox.com/~djb/cdb.html
39 * This implementation borrows some code from Dan Bernstein's
40 * implementation (which has no license restrictions applied to it).
41 * This (read-only) implementation is completely contained within
42 * cdb.[ch] it does *not* link against an external cdb library.
45 * There are 2 varients included within this code. One uses MMAP and
46 * should give better performance especially for multiple lookups on a
47 * modern machine. The other is the default implementation which is
48 * used in the case where the MMAP fails or if MMAP was not compiled
49 * in. this implementation is the same as the original reference cdb
50 * implementation. The MMAP version is compiled in if the HAVE_MMAP
51 * preprocessor define is defined - this should be set in the system
58 #include "lf_functions.h"
62 # include <sys/mman.h>
63 /* Not all implementations declare MAP_FAILED */
65 # define MAP_FAILED ((void *) -1)
66 # endif /* MAP_FAILED */
67 #endif /* HAVE_MMAP */
70 #define CDB_HASH_SPLIT 256 /* num pieces the hash table is split into */
71 #define CDB_HASH_MASK 255 /* mask to and off split value */
72 #define CDB_HASH_ENTRY 8 /* how big each offset it */
73 #define CDB_HASH_TABLE (CDB_HASH_SPLIT * CDB_HASH_ENTRY)
75 /* State information for cdb databases that are open NB while the db
76 * is open its contents will not change (cdb dbs are normally updated
77 * atomically by renaming). However the lifetime of one of these
78 * state structures should be limited - ie a long running daemon
79 * that opens one may hit problems....
89 /* 32 bit unsigned type - this is an int on all modern machines */
90 typedef unsigned int uint32
;
94 * Internal function to make hash value */
97 cdb_hash(uschar
*buf
, unsigned int len
)
105 h
^= (uint32
) *buf
++;
112 * Internal function to read len bytes from disk, coping with oddities */
122 r
= Uread(fd
,buf
,len
);
123 while ((r
== -1) && (errno
== EINTR
));
124 if (r
== -1) return -1;
125 if (r
== 0) { errno
= EIO
; return -1; }
134 * Internal function to parse 4 byte number (endian independant) */
137 cdb_unpack(uschar
*buf
)
140 num
= buf
[3]; num
<<= 8;
141 num
+= buf
[2]; num
<<= 8;
142 num
+= buf
[1]; num
<<= 8;
148 cdb_open(uschar
*filename
,
152 struct cdb_state
*cdbp
;
156 fileno
= Uopen(filename
, O_RDONLY
, 0);
158 int save_errno
= errno
;
159 *errmsg
= string_open_failed(errno
, "%s for cdb lookup", filename
);
164 if (fstat(fileno
, &statbuf
) == 0) {
165 /* If this is a valid file, then it *must* be at least
166 * CDB_HASH_TABLE bytes long */
167 if (statbuf
.st_size
< CDB_HASH_TABLE
) {
168 int save_errno
= errno
;
169 *errmsg
= string_open_failed(errno
,
170 "%s to short for cdb lookup",
176 int save_errno
= errno
;
177 *errmsg
= string_open_failed(errno
,
178 "fstat(%s) failed - cannot do cdb lookup",
184 /* Having got a file open we need the structure to put things in */
185 cdbp
= store_get(sizeof(struct cdb_state
));
186 /* store_get() does not return if memory was not available... */
187 /* preload the structure.... */
188 cdbp
->fileno
= fileno
;
189 cdbp
->filelen
= statbuf
.st_size
;
190 cdbp
->cdb_map
= NULL
;
191 cdbp
->cdb_offsets
= NULL
;
193 /* if we are allowed to we use mmap here.... */
201 if (mapbuf
!= MAP_FAILED
) {
202 /* We have an mmap-ed section. Now we can just use it */
203 cdbp
->cdb_map
= mapbuf
;
204 /* The offsets can be set to the same value since they should
205 * effectively be cached as well
207 cdbp
->cdb_offsets
= mapbuf
;
209 /* Now return the state struct */
212 /* If we got here the map failed. Basically we can ignore
213 * this since we fall back to slower methods....
214 * However lets debug log it...
216 DEBUG(D_lookup
) debug_printf("cdb mmap failed - %d\n", errno
);
218 #endif /* HAVE_MMAP */
220 /* In this case we have either not got MMAP allowed, or it failed */
222 /* get a buffer to stash the basic offsets in - this should speed
223 * things up a lot - especially on multiple lookups */
224 cdbp
->cdb_offsets
= store_get(CDB_HASH_TABLE
);
226 /* now fill the buffer up... */
227 if (cdb_bread(fileno
, cdbp
->cdb_offsets
, CDB_HASH_TABLE
) == -1) {
228 /* read of hash table failed, oh dear, oh.....
229 * time to give up I think....
230 * call the close routine (deallocs the memory), and return NULL */
231 *errmsg
= string_open_failed(errno
,
232 "cannot read header from %s for cdb lookup",
238 /* Everything else done - return the cache structure */
244 /*************************************************
245 * Check entry point *
246 *************************************************/
249 cdb_check(void *handle
,
256 struct cdb_state
* cdbp
= handle
;
257 return lf_check_file(cdbp
->fileno
,
269 /*************************************************
271 *************************************************/
274 cdb_find(void *handle
,
282 struct cdb_state
* cdbp
= handle
;
296 /* Keep picky compilers happy */
299 key_hash
= cdb_hash((uschar
*)keystring
, key_len
);
301 hash_offset_entry
= CDB_HASH_ENTRY
* (key_hash
& CDB_HASH_MASK
);
302 hash_offset
= cdb_unpack(cdbp
->cdb_offsets
+ hash_offset_entry
);
303 hash_offlen
= cdb_unpack(cdbp
->cdb_offsets
+ hash_offset_entry
+ 4);
305 /* If the offset length is zero this key cannot be in the file */
306 if (hash_offlen
== 0) {
309 hash_slotnm
= (key_hash
>> 8) % hash_offlen
;
311 /* check to ensure that the file is not corrupt
312 * if the hash_offset + (hash_offlen * CDB_HASH_ENTRY) is longer
313 * than the file, then we have problems.... */
314 if ((hash_offset
+ (hash_offlen
* CDB_HASH_ENTRY
)) > cdbp
->filelen
) {
315 *errmsg
= string_sprintf("cdb: corrupt cdb file %s (too short)",
317 DEBUG(D_lookup
) debug_printf("%s\n", *errmsg
);
321 cur_offset
= hash_offset
+ (hash_slotnm
* CDB_HASH_ENTRY
);
322 end_offset
= hash_offset
+ (hash_offlen
* CDB_HASH_ENTRY
);
323 /* if we are allowed to we use mmap here.... */
325 /* make sure the mmap was OK */
326 if (cdbp
->cdb_map
!= NULL
) {
327 uschar
* cur_pos
= cur_offset
+ cdbp
->cdb_map
;
328 uschar
* end_pos
= end_offset
+ cdbp
->cdb_map
;
329 for (loop
= 0; (loop
< hash_offlen
); ++loop
) {
330 item_hash
= cdb_unpack(cur_pos
);
332 item_posn
= cdb_unpack(cur_pos
);
334 /* if the position is zero then we have a definite miss */
338 if (item_hash
== key_hash
) {
339 /* matching hash value */
340 uschar
* item_ptr
= cdbp
->cdb_map
+ item_posn
;
341 item_key_len
= cdb_unpack(item_ptr
);
343 item_dat_len
= cdb_unpack(item_ptr
);
345 /* check key length matches */
346 if (item_key_len
== key_len
) {
347 /* finally check if key matches */
348 if (Ustrncmp(keystring
, item_ptr
, key_len
) == 0) {
349 /* we have a match....
350 * make item_ptr point to data */
351 item_ptr
+= item_key_len
;
352 /* ... and the returned result */
353 *result
= store_get(item_dat_len
+ 1);
354 memcpy(*result
, item_ptr
, item_dat_len
);
355 (*result
)[item_dat_len
] = 0;
360 /* handle warp round of table */
361 if (cur_pos
== end_pos
)
362 cur_pos
= cdbp
->cdb_map
+ hash_offset
;
364 /* looks like we failed... */
367 #endif /* HAVE_MMAP */
368 for (loop
= 0; (loop
< hash_offlen
); ++loop
) {
370 if (lseek(cdbp
->fileno
, (off_t
) cur_offset
,SEEK_SET
) == -1) return DEFER
;
371 if (cdb_bread(cdbp
->fileno
, packbuf
,8) == -1) return DEFER
;
372 item_hash
= cdb_unpack(packbuf
);
373 item_posn
= cdb_unpack(packbuf
+ 4);
374 /* if the position is zero then we have a definite miss */
378 if (item_hash
== key_hash
) {
379 /* matching hash value */
380 if (lseek(cdbp
->fileno
, (off_t
) item_posn
, SEEK_SET
) == -1) return DEFER
;
381 if (cdb_bread(cdbp
->fileno
, packbuf
, 8) == -1) return DEFER
;
382 item_key_len
= cdb_unpack(packbuf
);
383 /* check key length matches */
384 if (item_key_len
== key_len
) {
385 /* finally check if key matches */
386 uschar
* item_key
= store_get(key_len
);
387 if (cdb_bread(cdbp
->fileno
, item_key
, key_len
) == -1) return DEFER
;
388 if (Ustrncmp(keystring
, item_key
, key_len
) == 0) {
389 /* Reclaim some store */
390 store_reset(item_key
);
391 /* matches - get data length */
392 item_dat_len
= cdb_unpack(packbuf
+ 4);
393 /* then we build a new result string */
394 *result
= store_get(item_dat_len
+ 1);
395 if (cdb_bread(cdbp
->fileno
, *result
, item_dat_len
) == -1)
397 (*result
)[item_dat_len
] = 0;
400 /* Reclaim some store */
401 store_reset(item_key
);
406 /* handle warp round of table */
407 if (cur_offset
== end_offset
)
408 cur_offset
= hash_offset
;
415 /*************************************************
416 * Close entry point *
417 *************************************************/
419 /* See local README for interface description */
422 cdb_close(void *handle
)
424 struct cdb_state
* cdbp
= handle
;
428 munmap(CS cdbp
->cdb_map
, cdbp
->filelen
);
429 if (cdbp
->cdb_map
== cdbp
->cdb_offsets
)
430 cdbp
->cdb_offsets
= NULL
;
432 #endif /* HAVE_MMAP */
437 /* End of lookups/cdb.c */