15c88420449c5bb789a4475fe1eb2312895d3b65
[exim.git] / src / src / lookups / cdb.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /*
6 * Exim - CDB database lookup module
7 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
8 *
9 * Copyright (c) 1998 Nigel Metheringham, Planet Online Ltd
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version 2
14 * of the License, or (at your option) any later version.
15 *
16 * --------------------------------------------------------------
17 * Modified by PH for Exim 4:
18 * Changed over to using unsigned chars
19 * Makes use of lf_check_file() for file checking
20 * --------------------------------------------------------------
21 * Modified by The Exim Maintainers 2015:
22 * const propagation
23 * --------------------------------------------------------------
24 *
25 * This program is distributed in the hope that it will be useful,
26 * but WITHOUT ANY WARRANTY; without even the implied warranty of
27 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
28 * GNU General Public License for more details.
29 *
30 * You should have received a copy of the GNU General Public License
31 * along with this program; if not, write to the Free Software
32 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
33 * 02111-1307, USA.
34 *
35 *
36 * This code implements Dan Bernstein's Constant DataBase (cdb) spec.
37 * Information, the spec and sample code for cdb can be obtained from
38 * http://www.pobox.com/~djb/cdb.html
39 *
40 * This implementation borrows some code from Dan Bernstein's
41 * implementation (which has no license restrictions applied to it).
42 * This (read-only) implementation is completely contained within
43 * cdb.[ch] it does *not* link against an external cdb library.
44 *
45 *
46 * There are 2 variants included within this code. One uses MMAP and
47 * should give better performance especially for multiple lookups on a
48 * modern machine. The other is the default implementation which is
49 * used in the case where the MMAP fails or if MMAP was not compiled
50 * in. this implementation is the same as the original reference cdb
51 * implementation. The MMAP version is compiled in if the HAVE_MMAP
52 * preprocessor define is defined - this should be set in the system
53 * specific os.h file.
54 *
55 */
56
57
58 #include "../exim.h"
59 #include "lf_functions.h"
60
61 #ifdef HAVE_MMAP
62 # include <sys/mman.h>
63 /* Not all implementations declare MAP_FAILED */
64 # ifndef MAP_FAILED
65 # define MAP_FAILED ((void *) -1)
66 # endif /* MAP_FAILED */
67 #endif /* HAVE_MMAP */
68
69
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)
74
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....
80 */
81
82 struct cdb_state {
83 int fileno;
84 off_t filelen;
85 uschar *cdb_map;
86 uschar *cdb_offsets;
87 };
88
89 /* 32 bit unsigned type - this is an int on all modern machines */
90 typedef unsigned int uint32;
91
92 /*
93 * cdb_hash()
94 * Internal function to make hash value */
95
96 static uint32
97 cdb_hash(const uschar *buf, unsigned int len)
98 {
99 uint32 h;
100
101 h = 5381;
102 while (len) {
103 --len;
104 h += (h << 5);
105 h ^= (uint32) *buf++;
106 }
107 return h;
108 }
109
110 /*
111 * cdb_bread()
112 * Internal function to read len bytes from disk, coping with oddities */
113
114 static int
115 cdb_bread(int fd,
116 uschar *buf,
117 int len)
118 {
119 int r;
120 while (len > 0) {
121 do
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; }
126 buf += r;
127 len -= r;
128 }
129 return 0;
130 }
131
132 /*
133 * cdb_bread()
134 * Internal function to parse 4 byte number (endian independent) */
135
136 static uint32
137 cdb_unpack(uschar *buf)
138 {
139 uint32 num;
140 num = buf[3]; num <<= 8;
141 num += buf[2]; num <<= 8;
142 num += buf[1]; num <<= 8;
143 num += buf[0];
144 return num;
145 }
146
147 static void cdb_close(void *handle);
148
149 static void *
150 cdb_open(const uschar * filename, uschar ** errmsg)
151 {
152 int fileno;
153 struct cdb_state *cdbp;
154 struct stat statbuf;
155 void * mapbuf;
156
157 if ((fileno = Uopen(filename, O_RDONLY, 0)) < 0)
158 {
159 int save_errno = errno;
160 *errmsg = string_open_failed(errno, "%s for cdb lookup", filename);
161 errno = save_errno;
162 return NULL;
163 }
164
165 if (fstat(fileno, &statbuf) != 0)
166 {
167 int save_errno = errno;
168 *errmsg = string_open_failed(errno,
169 "fstat(%s) failed - cannot do cdb lookup",
170 filename);
171 errno = save_errno;
172 return NULL;
173 }
174
175 /* If this is a valid file, then it *must* be at least
176 CDB_HASH_TABLE bytes long */
177
178 if (statbuf.st_size < CDB_HASH_TABLE)
179 {
180 int save_errno = errno;
181 *errmsg = string_open_failed(errno,
182 "%s too short for cdb lookup",
183 filename);
184 errno = save_errno;
185 return NULL;
186 }
187
188 /* Having got a file open we need the structure to put things in */
189 cdbp = store_get(sizeof(struct cdb_state), FALSE);
190 /* store_get() does not return if memory was not available... */
191 /* preload the structure.... */
192 cdbp->fileno = fileno;
193 cdbp->filelen = statbuf.st_size;
194 cdbp->cdb_map = NULL;
195 cdbp->cdb_offsets = NULL;
196
197 /* if we are allowed to we use mmap here.... */
198 #ifdef HAVE_MMAP
199 if ((mapbuf = mmap(NULL, statbuf.st_size, PROT_READ, MAP_SHARED, fileno, 0))
200 != MAP_FAILED)
201 {
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
206 */
207 cdbp->cdb_offsets = mapbuf;
208
209 /* Now return the state struct */
210 return(cdbp);
211 }
212
213 /* If we got here the map failed. Basically we can ignore this since we fall
214 back to slower methods.... However lets debug log it... */
215
216 DEBUG(D_lookup) debug_printf_indent("cdb mmap failed - %d\n", errno);
217 #endif /* HAVE_MMAP */
218
219 /* In this case we have either not got MMAP allowed, or it failed */
220
221 /* get a buffer to stash the basic offsets in - this should speed
222 things up a lot - especially on multiple lookups */
223
224 cdbp->cdb_offsets = store_get(CDB_HASH_TABLE, FALSE);
225
226 /* now fill the buffer up... */
227
228 if (cdb_bread(fileno, cdbp->cdb_offsets, CDB_HASH_TABLE) == -1)
229 {
230 /* read of hash table failed, oh dear, oh..... time to give up I think....
231 call the close routine (deallocs the memory), and return NULL */
232
233 *errmsg = string_open_failed(errno,
234 "cannot read header from %s for cdb lookup",
235 filename);
236 cdb_close(cdbp);
237 return NULL;
238 }
239
240 /* Everything else done - return the cache structure */
241 return cdbp;
242 }
243
244
245
246 /*************************************************
247 * Check entry point *
248 *************************************************/
249
250 static BOOL
251 cdb_check(void * handle, const uschar * filename, int modemask,
252 uid_t * owners, gid_t * owngroups, uschar ** errmsg)
253 {
254 struct cdb_state * cdbp = handle;
255 return lf_check_file(cdbp->fileno, filename, S_IFREG, modemask,
256 owners, owngroups, "cdb", errmsg) == 0;
257 }
258
259
260
261 /*************************************************
262 * Find entry point *
263 *************************************************/
264
265 static int
266 cdb_find(void * handle, const uschar * filename, const uschar * keystring,
267 int key_len, uschar ** result, uschar ** errmsg, uint * do_cache,
268 const uschar * opts)
269 {
270 struct cdb_state * cdbp = handle;
271 uint32 item_key_len,
272 item_dat_len,
273 key_hash,
274 item_hash,
275 item_posn,
276 cur_offset,
277 end_offset,
278 hash_offset_entry,
279 hash_offset,
280 hash_offlen,
281 hash_slotnm;
282
283 /* Keep picky compilers happy */
284 do_cache = do_cache;
285
286 key_hash = cdb_hash(keystring, key_len);
287
288 hash_offset_entry = CDB_HASH_ENTRY * (key_hash & CDB_HASH_MASK);
289 hash_offset = cdb_unpack(cdbp->cdb_offsets + hash_offset_entry);
290 hash_offlen = cdb_unpack(cdbp->cdb_offsets + hash_offset_entry + 4);
291
292 /* If the offset length is zero this key cannot be in the file */
293
294 if (hash_offlen == 0)
295 return FAIL;
296
297 hash_slotnm = (key_hash >> 8) % hash_offlen;
298
299 /* check to ensure that the file is not corrupt
300 * if the hash_offset + (hash_offlen * CDB_HASH_ENTRY) is longer
301 * than the file, then we have problems.... */
302
303 if ((hash_offset + (hash_offlen * CDB_HASH_ENTRY)) > cdbp->filelen)
304 {
305 *errmsg = string_sprintf("cdb: corrupt cdb file %s (too short)",
306 filename);
307 DEBUG(D_lookup) debug_printf_indent("%s\n", *errmsg);
308 return DEFER;
309 }
310
311 cur_offset = hash_offset + (hash_slotnm * CDB_HASH_ENTRY);
312 end_offset = hash_offset + (hash_offlen * CDB_HASH_ENTRY);
313
314 /* if we are allowed to we use mmap here.... */
315
316 #ifdef HAVE_MMAP
317 /* make sure the mmap was OK */
318 if (cdbp->cdb_map != NULL)
319 {
320 uschar * cur_pos = cur_offset + cdbp->cdb_map;
321 uschar * end_pos = end_offset + cdbp->cdb_map;
322
323 for (int loop = 0; (loop < hash_offlen); ++loop)
324 {
325 item_hash = cdb_unpack(cur_pos);
326 cur_pos += 4;
327 item_posn = cdb_unpack(cur_pos);
328 cur_pos += 4;
329
330 /* if the position is zero then we have a definite miss */
331
332 if (item_posn == 0)
333 return FAIL;
334
335 if (item_hash == key_hash)
336 { /* matching hash value */
337 uschar * item_ptr = cdbp->cdb_map + item_posn;
338
339 item_key_len = cdb_unpack(item_ptr);
340 item_ptr += 4;
341 item_dat_len = cdb_unpack(item_ptr);
342 item_ptr += 4;
343
344 /* check key length matches */
345
346 if (item_key_len == key_len)
347 {
348 /* finally check if key matches */
349 if (Ustrncmp(keystring, item_ptr, key_len) == 0)
350 {
351 /* we have a match.... * make item_ptr point to data */
352
353 item_ptr += item_key_len;
354
355 /* ... and the returned result. Assume it is not
356 tainted, lacking any way of telling. */
357
358 *result = store_get(item_dat_len + 1, FALSE);
359 memcpy(*result, item_ptr, item_dat_len);
360 (*result)[item_dat_len] = 0;
361 return OK;
362 }
363 }
364 }
365 /* handle warp round of table */
366 if (cur_pos == end_pos)
367 cur_pos = cdbp->cdb_map + hash_offset;
368 }
369 /* looks like we failed... */
370 return FAIL;
371 }
372
373 #endif /* HAVE_MMAP */
374
375 for (int loop = 0; (loop < hash_offlen); ++loop)
376 {
377 uschar packbuf[8];
378
379 if (lseek(cdbp->fileno, (off_t) cur_offset, SEEK_SET) == -1) return DEFER;
380 if (cdb_bread(cdbp->fileno, packbuf, 8) == -1) return DEFER;
381
382 item_hash = cdb_unpack(packbuf);
383 item_posn = cdb_unpack(packbuf + 4);
384
385 /* if the position is zero then we have a definite miss */
386
387 if (item_posn == 0)
388 return FAIL;
389
390 if (item_hash == key_hash)
391 { /* matching hash value */
392 if (lseek(cdbp->fileno, (off_t) item_posn, SEEK_SET) == -1) return DEFER;
393 if (cdb_bread(cdbp->fileno, packbuf, 8) == -1) return DEFER;
394
395 item_key_len = cdb_unpack(packbuf);
396
397 /* check key length matches */
398
399 if (item_key_len == key_len)
400 { /* finally check if key matches */
401 rmark reset_point = store_mark();
402 uschar * item_key = store_get(key_len, TRUE); /* keys liable to be tainted */
403
404 if (cdb_bread(cdbp->fileno, item_key, key_len) == -1) return DEFER;
405 if (Ustrncmp(keystring, item_key, key_len) == 0)
406 {
407 /* Reclaim some store */
408 store_reset(reset_point);
409
410 /* matches - get data length */
411 item_dat_len = cdb_unpack(packbuf + 4);
412
413 /* then we build a new result string. We know we have enough
414 memory so disable Coverity errors about the tainted item_dat_ken */
415
416 *result = store_get(item_dat_len + 1, FALSE);
417 /* coverity[tainted_data] */
418 if (cdb_bread(cdbp->fileno, *result, item_dat_len) == -1)
419 return DEFER;
420
421 /* coverity[tainted_data] */
422 (*result)[item_dat_len] = 0;
423 return OK;
424 }
425 /* Reclaim some store */
426 store_reset(reset_point);
427 }
428 }
429 cur_offset += 8;
430
431 /* handle warp round of table */
432 if (cur_offset == end_offset)
433 cur_offset = hash_offset;
434 }
435 return FAIL;
436 }
437
438
439
440 /*************************************************
441 * Close entry point *
442 *************************************************/
443
444 /* See local README for interface description */
445
446 static void
447 cdb_close(void *handle)
448 {
449 struct cdb_state * cdbp = handle;
450
451 #ifdef HAVE_MMAP
452 if (cdbp->cdb_map)
453 {
454 munmap(CS cdbp->cdb_map, cdbp->filelen);
455 if (cdbp->cdb_map == cdbp->cdb_offsets)
456 cdbp->cdb_offsets = NULL;
457 }
458 #endif /* HAVE_MMAP */
459
460 (void)close(cdbp->fileno);
461 }
462
463
464
465 /*************************************************
466 * Version reporting entry point *
467 *************************************************/
468
469 /* See local README for interface description. */
470
471 #include "../version.h"
472
473 void
474 cdb_version_report(FILE *f)
475 {
476 #ifdef DYNLOOKUP
477 fprintf(f, "Library version: CDB: Exim version %s\n", EXIM_VERSION_STR);
478 #endif
479 }
480
481
482 lookup_info cdb_lookup_info = {
483 .name = US"cdb", /* lookup name */
484 .type = lookup_absfile, /* absolute file name */
485 .open = cdb_open, /* open function */
486 .check = cdb_check, /* check function */
487 .find = cdb_find, /* find function */
488 .close = cdb_close, /* close function */
489 .tidy = NULL, /* no tidy function */
490 .quote = NULL, /* no quoting function */
491 .version_report = cdb_version_report /* version reporting */
492 };
493
494 #ifdef DYNLOOKUP
495 #define cdb_lookup_module_info _lookup_module_info
496 #endif
497
498 static lookup_info *_lookup_list[] = { &cdb_lookup_info };
499 lookup_module_info cdb_lookup_module_info = { LOOKUP_MODULE_INFO_MAGIC, _lookup_list, 1 };
500
501 /* End of lookups/cdb.c */