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.
23 # include <sys/file.h>
38 #include <errno.h> /* See notes in perl.h about avoiding
41 extern Malloc_t malloc proto((MEM_SIZE));
42 extern Free_t free proto((Malloc_t));
47 static int getdbit proto((DBM *, long));
48 static int setdbit proto((DBM *, long));
49 static int getpage proto((DBM *, long));
50 static datum getnext proto((DBM *));
51 static int makroom proto((DBM *, long, int));
56 #define bad(x) ((x).dptr == NULL || (x).dsize < 0)
57 #define exhash(item) sdbm_hash((item).dptr, (item).dsize)
58 #define ioerr(db) ((db)->flags |= DBM_IOERR)
60 #define OFF_PAG(off) (long) (off) * PBLKSIZ
61 #define OFF_DIR(off) (long) (off) * DBLKSIZ
63 static const long masks[] = {
64 000000000000, 000000000001, 000000000003, 000000000007,
65 000000000017, 000000000037, 000000000077, 000000000177,
66 000000000377, 000000000777, 000000001777, 000000003777,
67 000000007777, 000000017777, 000000037777, 000000077777,
68 000000177777, 000000377777, 000000777777, 000001777777,
69 000003777777, 000007777777, 000017777777, 000037777777,
70 000077777777, 000177777777, 000377777777, 000777777777,
71 001777777777, 003777777777, 007777777777, 017777777777
75 sdbm_open(register char *file, register int flags, register int mode)
78 register char *dirname;
79 register char *pagname;
81 const size_t dirfext_len = sizeof(DIRFEXT "");
82 const size_t pagfext_len = sizeof(PAGFEXT "");
84 if (file == NULL || !*file)
85 return errno = EINVAL, (DBM *) NULL;
87 * need space for two seperate filenames
89 filelen = strlen(file);
91 if ((dirname = (char *) malloc(filelen + dirfext_len + 1
92 + filelen + pagfext_len + 1)) == NULL)
93 return errno = ENOMEM, (DBM *) NULL;
95 * build the file names
97 memcpy(dirname, file, filelen);
98 memcpy(dirname + filelen, DIRFEXT, dirfext_len + 1);
99 pagname = dirname + filelen + dirfext_len + 1;
100 memcpy(pagname, file, filelen);
101 memcpy(pagname + filelen, PAGFEXT, pagfext_len + 1);
103 db = sdbm_prep(dirname, pagname, flags, mode);
104 free((char *) dirname);
109 sdbm_prep(char *dirname, char *pagname, int flags, int mode)
114 if ((db = (DBM *) malloc(sizeof(DBM))) == NULL)
115 return errno = ENOMEM, (DBM *) NULL;
122 * adjust user flags so that WRONLY becomes RDWR,
123 * as required by this package. Also set our internal
124 * flag for RDONLY if needed.
126 if (flags & O_WRONLY)
127 flags = (flags & ~O_WRONLY) | O_RDWR;
129 else if ((flags & 03) == O_RDONLY)
130 db->flags = DBM_RDONLY;
132 * open the files in sequence, and stat the dirfile.
133 * If we fail anywhere, undo everything, return NULL.
135 #if defined(OS2) || defined(MSDOS) || defined(WIN32) || defined(__CYGWIN__)
138 if ((db->pagf = open(pagname, flags, mode)) > -1) {
139 if ((db->dirf = open(dirname, flags, mode)) > -1) {
141 * need the dirfile size to establish max bit number.
143 if (fstat(db->dirf, &dstat) == 0) {
145 * zero size: either a fresh database, or one with a single,
146 * unsplit data page: dirpage is all zeros.
148 db->dirbno = (!dstat.st_size) ? 0 : -1;
150 db->maxbno = dstat.st_size * BYTESIZ;
152 (void) memset(db->pagbuf, 0, PBLKSIZ);
153 (void) memset(db->dirbuf, 0, DBLKSIZ);
159 (void) close(db->dirf);
161 (void) close(db->pagf);
168 sdbm_close(register DBM *db)
173 (void) close(db->dirf);
174 (void) close(db->pagf);
180 sdbm_fetch(register DBM *db, datum key)
182 if (db == NULL || bad(key))
183 return errno = EINVAL, nullitem;
185 if (getpage(db, exhash(key)))
186 return getpair(db->pagbuf, key);
188 return ioerr(db), nullitem;
192 sdbm_exists(register DBM *db, datum key)
194 if (db == NULL || bad(key))
195 return errno = EINVAL, -1;
197 if (getpage(db, exhash(key)))
198 return exipair(db->pagbuf, key);
200 return ioerr(db), -1;
204 sdbm_delete(register DBM *db, datum key)
206 if (db == NULL || bad(key))
207 return errno = EINVAL, -1;
209 return errno = EPERM, -1;
211 if (getpage(db, exhash(key))) {
212 if (!delpair(db->pagbuf, key))
215 * update the page file
217 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
218 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
219 return ioerr(db), -1;
224 return ioerr(db), -1;
228 sdbm_store(register DBM *db, datum key, datum val, int flags)
233 if (db == NULL || bad(key))
234 return errno = EINVAL, -1;
236 return errno = EPERM, -1;
238 need = key.dsize + val.dsize;
240 * is the pair too big (or too small) for this database ??
242 if (need < 0 || need > PAIRMAX)
243 return errno = EINVAL, -1;
245 if (getpage(db, (hash = exhash(key)))) {
247 * if we need to replace, delete the key/data pair
248 * first. If it is not there, ignore.
250 if (flags == DBM_REPLACE)
251 (void) delpair(db->pagbuf, key);
253 else if (duppair(db->pagbuf, key))
257 * if we do not have enough room, we have to split.
259 if (!fitpair(db->pagbuf, need))
260 if (!makroom(db, hash, need))
261 return ioerr(db), -1;
263 * we have enough room or split is successful. insert the key,
264 * and update the page file.
266 (void) putpair(db->pagbuf, key, val);
268 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
269 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
270 return ioerr(db), -1;
277 return ioerr(db), -1;
281 * makroom - make room by splitting the overfull page
282 * this routine will attempt to make room for SPLTMAX times before
286 makroom(register DBM *db, long int hash, int need)
290 #if defined(DOSISH) || defined(WIN32)
294 char *pag = db->pagbuf;
296 register int smax = SPLTMAX;
300 * split the current page
302 (void) splpage(pag, New, db->hmask + 1);
304 * address of the new page
306 newp = (hash & db->hmask) | (db->hmask + 1);
309 * write delay, read avoidence/cache shuffle:
310 * select the page for incoming pair: if key is to go to the new page,
311 * write out the previous one, and copy the new one over, thus making
312 * it the current page. If not, simply write the new page, and we are
313 * still looking at the page of interest. current page is not updated
314 * here, as sdbm_store will do so, after it inserts the incoming pair.
317 #if defined(DOSISH) || defined(WIN32)
319 * Fill hole with 0 if made it.
320 * (hole is NOT read as 0)
322 oldtail = lseek(db->pagf, 0L, SEEK_END);
323 memset(zer, 0, PBLKSIZ);
324 while (OFF_PAG(newp) > oldtail) {
325 if (lseek(db->pagf, 0L, SEEK_END) < 0 ||
326 write(db->pagf, zer, PBLKSIZ) < 0) {
333 if (hash & (db->hmask + 1)) {
334 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
335 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
338 (void) memcpy(pag, New, PBLKSIZ);
340 else if (lseek(db->pagf, OFF_PAG(newp), SEEK_SET) < 0
341 || write(db->pagf, New, PBLKSIZ) < 0)
344 if (!setdbit(db, db->curbit))
347 * see if we have enough room now
349 if (fitpair(pag, need))
352 * try again... update curbit and hmask as getpage would have
353 * done. because of our update of the current page, we do not
354 * need to read in anything. BUT we have to write the current
355 * [deferred] page out, as the window of failure is too great.
357 db->curbit = 2 * db->curbit +
358 ((hash & (db->hmask + 1)) ? 2 : 1);
359 db->hmask |= db->hmask + 1;
361 if (lseek(db->pagf, OFF_PAG(db->pagbno), SEEK_SET) < 0
362 || write(db->pagf, db->pagbuf, PBLKSIZ) < 0)
367 * if we are here, this is real bad news. After SPLTMAX splits,
368 * we still cannot fit the key. say goodnight.
371 (void) write(2, "sdbm: cannot insert after SPLTMAX attempts.\n", 44);
378 * the following two routines will break if
379 * deletions aren't taken into account. (ndbm bug)
382 sdbm_firstkey(register DBM *db)
385 return errno = EINVAL, nullitem;
389 if (lseek(db->pagf, OFF_PAG(0), SEEK_SET) < 0
390 || read(db->pagf, db->pagbuf, PBLKSIZ) < 0)
391 return ioerr(db), nullitem;
400 sdbm_nextkey(register DBM *db)
403 return errno = EINVAL, nullitem;
408 * all important binary trie traversal
411 getpage(register DBM *db, register long int hash)
419 while (dbit < db->maxbno && getdbit(db, dbit))
420 dbit = 2 * dbit + ((hash & (1 << hbit++)) ? 2 : 1);
422 debug(("dbit: %d...", dbit));
425 db->hmask = masks[hbit];
427 pagb = hash & db->hmask;
429 * see if the block we need is already in memory.
430 * note: this lookaside cache has about 10% hit rate.
432 if (pagb != db->pagbno) {
434 * note: here, we assume a "hole" is read as 0s.
435 * if not, must zero pagbuf first.
437 if (lseek(db->pagf, OFF_PAG(pagb), SEEK_SET) < 0
438 || read(db->pagf, db->pagbuf, PBLKSIZ) < 0)
440 if (!chkpage(db->pagbuf))
444 debug(("pag read: %d\n", pagb));
450 getdbit(register DBM *db, register long int dbit)
458 if (dirb != db->dirbno) {
460 if (lseek(db->dirf, OFF_DIR(dirb), SEEK_SET) < 0
461 || (got=read(db->dirf, db->dirbuf, DBLKSIZ)) < 0)
464 memset(db->dirbuf,0,DBLKSIZ);
467 debug(("dir read: %d\n", dirb));
470 return db->dirbuf[c % DBLKSIZ] & (1 << dbit % BYTESIZ);
474 setdbit(register DBM *db, register long int dbit)
482 if (dirb != db->dirbno) {
484 if (lseek(db->dirf, OFF_DIR(dirb), SEEK_SET) < 0
485 || (got=read(db->dirf, db->dirbuf, DBLKSIZ)) < 0)
488 memset(db->dirbuf,0,DBLKSIZ);
491 debug(("dir read: %d\n", dirb));
494 db->dirbuf[c % DBLKSIZ] |= (1 << dbit % BYTESIZ);
497 if (dbit >= db->maxbno)
498 db->maxbno += DBLKSIZ * BYTESIZ;
500 if (OFF_DIR((dirb+1))*BYTESIZ > db->maxbno)
501 db->maxbno=OFF_DIR((dirb+1))*BYTESIZ;
504 if (lseek(db->dirf, OFF_DIR(dirb), SEEK_SET) < 0
505 || write(db->dirf, db->dirbuf, DBLKSIZ) < 0)
512 * getnext - get the next key in the page, and if done with
513 * the page, try the next page in sequence
516 getnext(register DBM *db)
522 key = getnkey(db->pagbuf, db->keyptr);
523 if (key.dptr != NULL)
526 * we either run out, or there is nothing on this page..
527 * try the next one... If we lost our position on the
528 * file, we will have to seek.
531 if (db->pagbno != db->blkptr++)
532 if (lseek(db->pagf, OFF_PAG(db->blkptr), SEEK_SET) < 0)
534 db->pagbno = db->blkptr;
535 if (read(db->pagf, db->pagbuf, PBLKSIZ) <= 0)
537 if (!chkpage(db->pagbuf))
541 return ioerr(db), nullitem;