1 /*
2 * sdbm - ndbm work-alike hashed database library
3 * based on Per-Aake Larson's Dynamic Hashing algorithms. BIT 18 (1978).
4 * author: oz@nexus.yorku.ca
5 * status: public domain.
6 *
7 * core routines
8 */
9
10 #include "INTERN.h"
11 #include "config.h"
12 #ifdef WIN32
13 #include "io.h"
14 #endif
15 #include "sdbm.h"
16 #include "tune.h"
17 #include "pair.h"
18
19 #ifdef I_FCNTL
20 # include <fcntl.h>
21 #endif
22 #ifdef I_SYS_FILE
23 # include <sys/file.h>
24 #endif
25
26 #ifdef I_STRING
27 # ifndef __ultrix__
28 # include <string.h>
29 # endif
30 #else
31 # include <strings.h>
32 #endif
33
34 /*
35 * externals
36 */
37 #ifndef WIN32
38 #ifndef sun
39 extern int errno;
40 #endif
41 #endif
42
43 extern Malloc_t malloc proto((MEM_SIZE));
44 extern Free_t free proto((Malloc_t));
45
46 /*
47 * forward
48 */
49 static int getdbit proto((DBM *, long));
50 static int setdbit proto((DBM *, long));
51 static int getpage proto((DBM *, long));
52 static datum getnext proto((DBM *));
53 static int makroom proto((DBM *, long, int));
54
55 /*
56 * useful macros
57 */
58 #define bad(x) ((x).dptr == NULL || (x).dsize < 0)
59 #define exhash(item) sdbm_hash((item).dptr, (item).dsize)
60 #define ioerr(db) ((db)->flags |= DBM_IOERR)
61
62 #define OFF_PAG(off) (long) (off) * PBLKSIZ
63 #define OFF_DIR(off) (long) (off) * DBLKSIZ
64
65 static long masks[] = {
66 000000000000, 000000000001, 000000000003, 000000000007,
67 000000000017, 000000000037, 000000000077, 000000000177,
68 000000000377, 000000000777, 000000001777, 000000003777,
69 000000007777, 000000017777, 000000037777, 000000077777,
70 000000177777, 000000377777, 000000777777, 000001777777,
71 000003777777, 000007777777, 000017777777, 000037777777,
72 000077777777, 000177777777, 000377777777, 000777777777,
73 001777777777, 003777777777, 007777777777, 017777777777
74 };
75
76 DBM *
sdbm_open(register char * file,register int flags,register int mode)77 sdbm_open(register char *file, register int flags, register int mode)
78 {
79 register DBM *db;
80 register char *dirname;
81 register char *pagname;
82 register int n;
83
84 if (file == NULL || !*file)
85 return errno = EINVAL, (DBM *) NULL;
86 /*
87 * need space for two seperate filenames
88 */
89 n = strlen(file) * 2 + strlen(DIRFEXT) + strlen(PAGFEXT) + 2;
90
91 if ((dirname = (char *) malloc((unsigned) n)) == NULL)
92 return errno = ENOMEM, (DBM *) NULL;
93 /*
94 * build the file names
95 */
96 dirname = strcat(strcpy(dirname, file), DIRFEXT);
97 pagname = strcpy(dirname + strlen(dirname) + 1, file);
98 pagname = strcat(pagname, PAGFEXT);
99
100 db = sdbm_prep(dirname, pagname, flags, mode);
101 free((char *) dirname);
102 return db;
103 }
104
105 DBM *
sdbm_prep(char * dirname,char * pagname,int flags,int mode)106 sdbm_prep(char *dirname, char *pagname, int flags, int mode)
107 {
108 register DBM *db;
109 struct stat dstat;
110
111 if ((db = (DBM *) malloc(sizeof(DBM))) == NULL)
112 return errno = ENOMEM, (DBM *) NULL;
113
114 db->flags = 0;
115 db->hmask = 0;
116 db->blkptr = 0;
117 db->keyptr = 0;
118 /*
119 * adjust user flags so that WRONLY becomes RDWR,
120 * as required by this package. Also set our internal
121 * flag for RDONLY if needed.
122 */
123 if (flags & O_WRONLY)
124 flags = (flags & ~O_WRONLY) | O_RDWR;
125
126 else if ((flags & 03) == O_RDONLY)
127 db->flags = DBM_RDONLY;
128 /*
129 * open the files in sequence, and stat the dirfile.
130 * If we fail anywhere, undo everything, return NULL.
131 */
132 #if defined(OS2) || defined(MSDOS) || defined(WIN32) || defined(__CYGWIN__)
133 flags |= O_BINARY;
134 # endif
135 if ((db->pagf = open(pagname, flags, mode)) > -1) {
136 if ((db->dirf = open(dirname, flags, mode)) > -1) {
137 /*
138 * need the dirfile size to establish max bit number.
139 */
140 if (fstat(db->dirf, &dstat) == 0) {
141 /*
142 * zero size: either a fresh database, or one with a single,
143 * unsplit data page: dirpage is all zeros.
144 */
145 db->dirbno = (!dstat.st_size) ? 0 : -1;
146 db->pagbno = -1;
147 db->maxbno = dstat.st_size * BYTESIZ;
148
149 (void) memset(db->pagbuf, 0, PBLKSIZ);
150 (void) memset(db->dirbuf, 0, DBLKSIZ);
151 /*
152 * success
153 */
154 return db;
155 }
156 (void) close(db->dirf);
157 }
158 (void) close(db->pagf);
159 }
160 free((char *) db);
161 return (DBM *) NULL;
162 }
163
164 void
sdbm_close(register DBM * db)165 sdbm_close(register DBM *db)
166 {
167 if (db == NULL)
168 errno = EINVAL;
169 else {
170 (void) close(db->dirf);
171 (void) close(db->pagf);
172 free((char *) db);
173 }
174 }
175
176 datum
sdbm_fetch(register DBM * db,datum key)177 sdbm_fetch(register DBM *db, datum key)
178 {
179 if (db == NULL || bad(key))
180 return errno = EINVAL, nullitem;
181
182 if (getpage(db, exhash(key)))
183 return getpair(db->pagbuf, key);
184
185 return ioerr(db), nullitem;
186 }
187
188 int
sdbm_exists(register DBM * db,datum key)189 sdbm_exists(register DBM *db, datum key)
190 {
191 if (db == NULL || bad(key))
192 return errno = EINVAL, -1;
193
194 if (getpage(db, exhash(key)))
195 return exipair(db->pagbuf, key);
196
197 return ioerr(db), -1;
198 }
199
200 int
sdbm_delete(register DBM * db,datum key)201 sdbm_delete(register DBM *db, datum key)
202 {
203 if (db == NULL || bad(key))
204 return errno = EINVAL, -1;
205 if (sdbm_rdonly(db))
206 return errno = EPERM, -1;
207
208 if (getpage(db, exhash(key))) {
209 if (!delpair(db->pagbuf, key))
210 return -1;
211 /*
212 * update the page file
213 */
214 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
215 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
216 return ioerr(db), -1;
217
218 return 0;
219 }
220
221 return ioerr(db), -1;
222 }
223
224 int
sdbm_store(register DBM * db,datum key,datum val,int flags)225 sdbm_store(register DBM *db, datum key, datum val, int flags)
226 {
227 int need;
228 register long hash;
229
230 if (db == NULL || bad(key))
231 return errno = EINVAL, -1;
232 if (sdbm_rdonly(db))
233 return errno = EPERM, -1;
234
235 need = key.dsize + val.dsize;
236 /*
237 * is the pair too big (or too small) for this database ??
238 */
239 if (need < 0 || need > PAIRMAX)
240 return errno = EINVAL, -1;
241
242 if (getpage(db, (hash = exhash(key)))) {
243 /*
244 * if we need to replace, delete the key/data pair
245 * first. If it is not there, ignore.
246 */
247 if (flags == DBM_REPLACE)
248 (void) delpair(db->pagbuf, key);
249 #ifdef SEEDUPS
250 else if (duppair(db->pagbuf, key))
251 return 1;
252 #endif
253 /*
254 * if we do not have enough room, we have to split.
255 */
256 if (!fitpair(db->pagbuf, need))
257 if (!makroom(db, hash, need))
258 return ioerr(db), -1;
259 /*
260 * we have enough room or split is successful. insert the key,
261 * and update the page file.
262 */
263 (void) putpair(db->pagbuf, key, val);
264
265 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
266 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
267 return ioerr(db), -1;
268 /*
269 * success
270 */
271 return 0;
272 }
273
274 return ioerr(db), -1;
275 }
276
277 /*
278 * makroom - make room by splitting the overfull page
279 * this routine will attempt to make room for SPLTMAX times before
280 * giving up.
281 */
282 static int
makroom(register DBM * db,long int hash,int need)283 makroom(register DBM *db, long int hash, int need)
284 {
285 long newp;
286 char twin[PBLKSIZ];
287 #if defined(DOSISH) || defined(WIN32)
288 char zer[PBLKSIZ];
289 long oldtail;
290 #endif
291 char *pag = db->pagbuf;
292 char *New = twin;
293 register int smax = SPLTMAX;
294
295 do {
296 /*
297 * split the current page
298 */
299 (void) splpage(pag, New, db->hmask + 1);
300 /*
301 * address of the new page
302 */
303 newp = (hash & db->hmask) | (db->hmask + 1);
304
305 /*
306 * write delay, read avoidence/cache shuffle:
307 * select the page for incoming pair: if key is to go to the new page,
308 * write out the previous one, and copy the new one over, thus making
309 * it the current page. If not, simply write the new page, and we are
310 * still looking at the page of interest. current page is not updated
311 * here, as sdbm_store will do so, after it inserts the incoming pair.
312 */
313
314 #if defined(DOSISH) || defined(WIN32)
315 /*
316 * Fill hole with 0 if made it.
317 * (hole is NOT read as 0)
318 */
319 oldtail = lseek(db->pagf, 0L, SEEK_END);
320 memset(zer, 0, PBLKSIZ);
321 while (OFF_PAG(newp) > oldtail) {
322 if (lseek(db->pagf, 0L, SEEK_END) < 0 ||
323 write(db->pagf, zer, PBLKSIZ) < 0) {
324
325 return 0;
326 }
327 oldtail += PBLKSIZ;
328 }
329 #endif
330 if (hash & (db->hmask + 1)) {
331 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
332 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
333 return 0;
334 db->pagbno = newp;
335 (void) memcpy(pag, New, PBLKSIZ);
336 }
337 else if (lseek(db->pagf, OFF_PAG(newp), SEEK_SET) < 0
338 || write(db->pagf, New, PBLKSIZ) < 0)
339 return 0;
340
341 if (!setdbit(db, db->curbit))
342 return 0;
343 /*
344 * see if we have enough room now
345 */
346 if (fitpair(pag, need))
347 return 1;
348 /*
349 * try again... update curbit and hmask as getpage would have
350 * done. because of our update of the current page, we do not
351 * need to read in anything. BUT we have to write the current
352 * [deferred] page out, as the window of failure is too great.
353 */
354 db->curbit = 2 * db->curbit +
355 ((hash & (db->hmask + 1)) ? 2 : 1);
356 db->hmask |= db->hmask + 1;
357
358 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
359 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
360 return 0;
361
362 } while (--smax);
363 /*
364 * if we are here, this is real bad news. After SPLTMAX splits,
365 * we still cannot fit the key. say goodnight.
366 */
367 #ifdef BADMESS
368 (void) write(2, "sdbm: cannot insert after SPLTMAX attempts.\n", 44);
369 #endif
370 return 0;
371
372 }
373
374 /*
375 * the following two routines will break if
376 * deletions aren't taken into account. (ndbm bug)
377 */
378 datum
sdbm_firstkey(register DBM * db)379 sdbm_firstkey(register DBM *db)
380 {
381 if (db == NULL)
382 return errno = EINVAL, nullitem;
383 /*
384 * start at page 0
385 */
386 if (lseek(db->pagf, OFF_PAG(0), SEEK_SET) < 0
387 || read(db->pagf, db->pagbuf, PBLKSIZ) < 0)
388 return ioerr(db), nullitem;
389 db->pagbno = 0;
390 db->blkptr = 0;
391 db->keyptr = 0;
392
393 return getnext(db);
394 }
395
396 datum
sdbm_nextkey(register DBM * db)397 sdbm_nextkey(register DBM *db)
398 {
399 if (db == NULL)
400 return errno = EINVAL, nullitem;
401 return getnext(db);
402 }
403
404 /*
405 * all important binary trie traversal
406 */
407 static int
getpage(register DBM * db,register long int hash)408 getpage(register DBM *db, register long int hash)
409 {
410 register int hbit;
411 register long dbit;
412 register long pagb;
413
414 dbit = 0;
415 hbit = 0;
416 while (dbit < db->maxbno && getdbit(db, dbit))
417 dbit = 2 * dbit + ((hash & (1 << hbit++)) ? 2 : 1);
418
419 debug(("dbit: %d...", dbit));
420
421 db->curbit = dbit;
422 db->hmask = masks[hbit];
423
424 pagb = hash & db->hmask;
425 /*
426 * see if the block we need is already in memory.
427 * note: this lookaside cache has about 10% hit rate.
428 */
429 if (pagb != db->pagbno) {
430 /*
431 * note: here, we assume a "hole" is read as 0s.
432 * if not, must zero pagbuf first.
433 */
434 if (lseek(db->pagf, OFF_PAG(pagb), SEEK_SET) < 0
435 || read(db->pagf, db->pagbuf, PBLKSIZ) < 0)
436 return 0;
437 if (!chkpage(db->pagbuf))
438 return 0;
439 db->pagbno = pagb;
440
441 debug(("pag read: %d\n", pagb));
442 }
443 return 1;
444 }
445
446 static int
getdbit(register DBM * db,register long int dbit)447 getdbit(register DBM *db, register long int dbit)
448 {
449 register long c;
450 register long dirb;
451
452 c = dbit / BYTESIZ;
453 dirb = c / DBLKSIZ;
454
455 if (dirb != db->dirbno) {
456 int got;
457 if (lseek(db->dirf, OFF_DIR(dirb), SEEK_SET) < 0
458 || (got=read(db->dirf, db->dirbuf, DBLKSIZ)) < 0)
459 return 0;
460 if (got==0)
461 memset(db->dirbuf,0,DBLKSIZ);
462 db->dirbno = dirb;
463
464 debug(("dir read: %d\n", dirb));
465 }
466
467 return db->dirbuf[c % DBLKSIZ] & (1 << dbit % BYTESIZ);
468 }
469
470 static int
setdbit(register DBM * db,register long int dbit)471 setdbit(register DBM *db, register long int dbit)
472 {
473 register long c;
474 register long dirb;
475
476 c = dbit / BYTESIZ;
477 dirb = c / DBLKSIZ;
478
479 if (dirb != db->dirbno) {
480 int got;
481 if (lseek(db->dirf, OFF_DIR(dirb), SEEK_SET) < 0
482 || (got=read(db->dirf, db->dirbuf, DBLKSIZ)) < 0)
483 return 0;
484 if (got==0)
485 memset(db->dirbuf,0,DBLKSIZ);
486 db->dirbno = dirb;
487
488 debug(("dir read: %d\n", dirb));
489 }
490
491 db->dirbuf[c % DBLKSIZ] |= (1 << dbit % BYTESIZ);
492
493 #if 0
494 if (dbit >= db->maxbno)
495 db->maxbno += DBLKSIZ * BYTESIZ;
496 #else
497 if (OFF_DIR((dirb+1))*BYTESIZ > db->maxbno)
498 db->maxbno=OFF_DIR((dirb+1))*BYTESIZ;
499 #endif
500
501 if (lseek(db->dirf, OFF_DIR(dirb), SEEK_SET) < 0
502 || write(db->dirf, db->dirbuf, DBLKSIZ) < 0)
503 return 0;
504
505 return 1;
506 }
507
508 /*
509 * getnext - get the next key in the page, and if done with
510 * the page, try the next page in sequence
511 */
512 static datum
getnext(register DBM * db)513 getnext(register DBM *db)
514 {
515 datum key;
516
517 for (;;) {
518 db->keyptr++;
519 key = getnkey(db->pagbuf, db->keyptr);
520 if (key.dptr != NULL)
521 return key;
522 /*
523 * we either run out, or there is nothing on this page..
524 * try the next one... If we lost our position on the
525 * file, we will have to seek.
526 */
527 db->keyptr = 0;
528 if (db->pagbno != db->blkptr++)
529 if (lseek(db->pagf, OFF_PAG(db->blkptr), SEEK_SET) < 0)
530 break;
531 db->pagbno = db->blkptr;
532 if (read(db->pagf, db->pagbuf, PBLKSIZ) <= 0)
533 break;
534 if (!chkpage(db->pagbuf))
535 break;
536 }
537
538 return ioerr(db), nullitem;
539 }
540
541