1 /* Copyright 2000-2005 The Apache Software Foundation or its licensors, as
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 * sdbm - ndbm work-alike hashed database library
19 * based on Per-Aake Larson's Dynamic Hashing algorithms. BIT 18 (1978).
20 * author: oz@nexus.yorku.ca
21 * ex-public domain, ported to APR for Apache 2
26 #include "apr_file_io.h"
27 #include "apr_strings.h"
28 #include "apr_errno.h"
31 #include "sdbm_tune.h"
32 #include "sdbm_pair.h"
33 #include "sdbm_private.h"
35 #include <string.h> /* for memset() */
36 #include <stdlib.h> /* for malloc() and free() */
41 static int getdbit (apr_sdbm_t *, long);
42 static apr_status_t setdbit(apr_sdbm_t *, long);
43 static apr_status_t getpage(apr_sdbm_t *db, long);
44 static apr_status_t getnext(apr_sdbm_datum_t *key, apr_sdbm_t *db);
45 static apr_status_t makroom(apr_sdbm_t *, long, int);
50 #define bad(x) ((x).dptr == NULL || (x).dsize <= 0)
51 #define exhash(item) sdbm_hash((item).dptr, (item).dsize)
53 /* ### Does anything need these externally? */
54 #define sdbm_dirfno(db) ((db)->dirf)
55 #define sdbm_pagfno(db) ((db)->pagf)
57 #define OFF_PAG(off) (apr_off_t) (off) * PBLKSIZ
58 #define OFF_DIR(off) (apr_off_t) (off) * DBLKSIZ
60 static long masks[] = {
61 000000000000, 000000000001, 000000000003, 000000000007,
62 000000000017, 000000000037, 000000000077, 000000000177,
63 000000000377, 000000000777, 000000001777, 000000003777,
64 000000007777, 000000017777, 000000037777, 000000077777,
65 000000177777, 000000377777, 000000777777, 000001777777,
66 000003777777, 000007777777, 000017777777, 000037777777,
67 000077777777, 000177777777, 000377777777, 000777777777,
68 001777777777, 003777777777, 007777777777, 017777777777
71 const apr_sdbm_datum_t sdbm_nullitem = { NULL, 0 };
73 static apr_status_t database_cleanup(void *data)
75 apr_sdbm_t *db = data;
78 * Can't rely on apr_sdbm_unlock, since it will merely
79 * decrement the refcnt if several locks are held.
81 if (db->flags & (SDBM_SHARED_LOCK | SDBM_EXCLUSIVE_LOCK))
82 (void) apr_file_unlock(db->dirf);
83 (void) apr_file_close(db->dirf);
84 (void) apr_file_close(db->pagf);
90 static apr_status_t prep(apr_sdbm_t **pdb, const char *dirname, const char *pagname,
91 apr_int32_t flags, apr_fileperms_t perms, apr_pool_t *p)
98 db = malloc(sizeof(*db));
99 memset(db, 0, sizeof(*db));
104 * adjust user flags so that WRONLY becomes RDWR,
105 * as required by this package. Also set our internal
106 * flag for RDONLY if needed.
108 if (!(flags & APR_WRITE)) {
109 db->flags |= SDBM_RDONLY;
113 * adjust the file open flags so that we handle locking
114 * on our own (don't rely on any locking behavior within
115 * an apr_file_t, in case it's ever introduced, and set
118 if (flags & APR_SHARELOCK) {
119 db->flags |= SDBM_SHARED;
120 flags &= ~APR_SHARELOCK;
123 flags |= APR_BINARY | APR_READ;
126 * open the files in sequence, and stat the dirfile.
127 * If we fail anywhere, undo everything, return NULL.
130 if ((status = apr_file_open(&db->dirf, dirname, flags, perms, p))
134 if ((status = apr_file_open(&db->pagf, pagname, flags, perms, p))
138 if ((status = apr_sdbm_lock(db, (db->flags & SDBM_RDONLY)
140 : APR_FLOCK_EXCLUSIVE))
144 /* apr_pcalloc zeroed the buffers
145 * apr_sdbm_lock stated the dirf->size and invalidated the cache
149 * if we are opened in SHARED mode, unlock ourself
151 if (db->flags & SDBM_SHARED)
152 if ((status = apr_sdbm_unlock(db)) != APR_SUCCESS)
155 /* make sure that we close the database at some point */
156 apr_pool_cleanup_register(p, db, database_cleanup, apr_pool_cleanup_null);
163 if (db->dirf && db->pagf)
164 (void) apr_sdbm_unlock(db);
165 if (db->dirf != NULL)
166 (void) apr_file_close(db->dirf);
167 if (db->pagf != NULL) {
168 (void) apr_file_close(db->pagf);
174 APU_DECLARE(apr_status_t) apr_sdbm_open(apr_sdbm_t **db, const char *file,
176 apr_fileperms_t perms, apr_pool_t *p)
178 char *dirname = apr_pstrcat(p, file, APR_SDBM_DIRFEXT, NULL);
179 char *pagname = apr_pstrcat(p, file, APR_SDBM_PAGFEXT, NULL);
181 return prep(db, dirname, pagname, flags, perms, p);
184 APU_DECLARE(apr_status_t) apr_sdbm_close(apr_sdbm_t *db)
186 return apr_pool_cleanup_run(db->pool, db, database_cleanup);
189 APU_DECLARE(apr_status_t) apr_sdbm_fetch(apr_sdbm_t *db, apr_sdbm_datum_t *val,
190 apr_sdbm_datum_t key)
194 if (db == NULL || bad(key))
197 if ((status = apr_sdbm_lock(db, APR_FLOCK_SHARED)) != APR_SUCCESS)
200 if ((status = getpage(db, exhash(key))) == APR_SUCCESS) {
201 *val = getpair(db->pagbuf, key);
202 /* ### do we want a not-found result? */
205 (void) apr_sdbm_unlock(db);
210 static apr_status_t write_page(apr_sdbm_t *db, const char *buf, long pagno)
213 apr_off_t off = OFF_PAG(pagno);
215 if ((status = apr_file_seek(db->pagf, APR_SET, &off)) == APR_SUCCESS)
216 status = apr_file_write_full(db->pagf, buf, PBLKSIZ, NULL);
221 APU_DECLARE(apr_status_t) apr_sdbm_delete(apr_sdbm_t *db,
222 const apr_sdbm_datum_t key)
226 if (db == NULL || bad(key))
228 if (apr_sdbm_rdonly(db))
231 if ((status = apr_sdbm_lock(db, APR_FLOCK_EXCLUSIVE)) != APR_SUCCESS)
234 if ((status = getpage(db, exhash(key))) == APR_SUCCESS) {
235 if (!delpair(db->pagbuf, key))
236 /* ### should we define some APRUTIL codes? */
237 status = APR_EGENERAL;
239 status = write_page(db, db->pagbuf, db->pagbno);
242 (void) apr_sdbm_unlock(db);
247 APU_DECLARE(apr_status_t) apr_sdbm_store(apr_sdbm_t *db, apr_sdbm_datum_t key,
248 apr_sdbm_datum_t val, int flags)
254 if (db == NULL || bad(key))
256 if (apr_sdbm_rdonly(db))
258 need = key.dsize + val.dsize;
260 * is the pair too big (or too small) for this database ??
262 if (need < 0 || need > PAIRMAX)
265 if ((status = apr_sdbm_lock(db, APR_FLOCK_EXCLUSIVE)) != APR_SUCCESS)
268 if ((status = getpage(db, (hash = exhash(key)))) == APR_SUCCESS) {
271 * if we need to replace, delete the key/data pair
272 * first. If it is not there, ignore.
274 if (flags == APR_SDBM_REPLACE)
275 (void) delpair(db->pagbuf, key);
276 else if (!(flags & APR_SDBM_INSERTDUP) && duppair(db->pagbuf, key)) {
281 * if we do not have enough room, we have to split.
283 if (!fitpair(db->pagbuf, need))
284 if ((status = makroom(db, hash, need)) != APR_SUCCESS)
287 * we have enough room or split is successful. insert the key,
288 * and update the page file.
290 (void) putpair(db->pagbuf, key, val);
292 status = write_page(db, db->pagbuf, db->pagbno);
296 (void) apr_sdbm_unlock(db);
302 * makroom - make room by splitting the overfull page
303 * this routine will attempt to make room for SPLTMAX times before
306 static apr_status_t makroom(apr_sdbm_t *db, long hash, int need)
310 char *pag = db->pagbuf;
312 register int smax = SPLTMAX;
317 * split the current page
319 (void) splpage(pag, new, db->hmask + 1);
321 * address of the new page
323 newp = (hash & db->hmask) | (db->hmask + 1);
326 * write delay, read avoidence/cache shuffle:
327 * select the page for incoming pair: if key is to go to the new page,
328 * write out the previous one, and copy the new one over, thus making
329 * it the current page. If not, simply write the new page, and we are
330 * still looking at the page of interest. current page is not updated
331 * here, as sdbm_store will do so, after it inserts the incoming pair.
333 if (hash & (db->hmask + 1)) {
334 if ((status = write_page(db, db->pagbuf, db->pagbno))
339 (void) memcpy(pag, new, PBLKSIZ);
342 if ((status = write_page(db, new, newp)) != APR_SUCCESS)
346 if ((status = setdbit(db, db->curbit)) != APR_SUCCESS)
349 * see if we have enough room now
351 if (fitpair(pag, need))
354 * try again... update curbit and hmask as getpage would have
355 * done. because of our update of the current page, we do not
356 * need to read in anything. BUT we have to write the current
357 * [deferred] page out, as the window of failure is too great.
359 db->curbit = 2 * db->curbit
360 + ((hash & (db->hmask + 1)) ? 2 : 1);
361 db->hmask |= db->hmask + 1;
363 if ((status = write_page(db, db->pagbuf, db->pagbno))
370 * if we are here, this is real bad news. After SPLTMAX splits,
371 * we still cannot fit the key. say goodnight.
374 (void) write(2, "sdbm: cannot insert after SPLTMAX attempts.\n", 44);
376 /* ### ENOSPC not really appropriate but better than nothing */
381 /* Reads 'len' bytes from file 'f' at offset 'off' into buf.
382 * 'off' is given relative to the start of the file.
383 * If EOF is returned while reading, this is taken as success.
385 static apr_status_t read_from(apr_file_t *f, void *buf,
386 apr_off_t off, apr_size_t len)
390 if ((status = apr_file_seek(f, APR_SET, &off)) != APR_SUCCESS ||
391 ((status = apr_file_read_full(f, buf, len, NULL)) != APR_SUCCESS)) {
392 /* if EOF is reached, pretend we read all zero's */
393 if (status == APR_EOF) {
395 status = APR_SUCCESS;
403 * the following two routines will break if
404 * deletions aren't taken into account. (ndbm bug)
406 APU_DECLARE(apr_status_t) apr_sdbm_firstkey(apr_sdbm_t *db,
407 apr_sdbm_datum_t *key)
411 if ((status = apr_sdbm_lock(db, APR_FLOCK_SHARED)) != APR_SUCCESS)
417 if ((status = read_from(db->pagf, db->pagbuf, OFF_PAG(0), PBLKSIZ))
422 status = getnext(key, db);
425 (void) apr_sdbm_unlock(db);
430 APU_DECLARE(apr_status_t) apr_sdbm_nextkey(apr_sdbm_t *db,
431 apr_sdbm_datum_t *key)
435 if ((status = apr_sdbm_lock(db, APR_FLOCK_SHARED)) != APR_SUCCESS)
438 status = getnext(key, db);
440 (void) apr_sdbm_unlock(db);
446 * all important binary tree traversal
448 static apr_status_t getpage(apr_sdbm_t *db, long hash)
457 while (dbit < db->maxbno && getdbit(db, dbit))
458 dbit = 2 * dbit + ((hash & (1 << hbit++)) ? 2 : 1);
460 debug(("dbit: %d...", dbit));
463 db->hmask = masks[hbit];
465 pagb = hash & db->hmask;
467 * see if the block we need is already in memory.
468 * note: this lookaside cache has about 10% hit rate.
470 if (pagb != db->pagbno) {
472 * note: here, we assume a "hole" is read as 0s.
473 * if not, must zero pagbuf first.
474 * ### joe: this assumption was surely never correct? but
475 * ### we make it so in read_from anyway.
477 if ((status = read_from(db->pagf, db->pagbuf, OFF_PAG(pagb), PBLKSIZ))
481 if (!chkpage(db->pagbuf))
482 return APR_ENOSPC; /* ### better error? */
485 debug(("pag read: %d\n", pagb));
490 static int getdbit(apr_sdbm_t *db, long dbit)
498 if (dirb != db->dirbno) {
499 if (read_from(db->dirf, db->dirbuf, OFF_DIR(dirb), DBLKSIZ)
505 debug(("dir read: %d\n", dirb));
508 return db->dirbuf[c % DBLKSIZ] & (1 << dbit % BYTESIZ);
511 static apr_status_t setdbit(apr_sdbm_t *db, long dbit)
521 if (dirb != db->dirbno) {
522 if ((status = read_from(db->dirf, db->dirbuf, OFF_DIR(dirb), DBLKSIZ))
528 debug(("dir read: %d\n", dirb));
531 db->dirbuf[c % DBLKSIZ] |= (1 << dbit % BYTESIZ);
533 if (dbit >= db->maxbno)
534 db->maxbno += DBLKSIZ * BYTESIZ;
537 if ((status = apr_file_seek(db->dirf, APR_SET, &off)) == APR_SUCCESS)
538 status = apr_file_write_full(db->dirf, db->dirbuf, DBLKSIZ, NULL);
544 * getnext - get the next key in the page, and if done with
545 * the page, try the next page in sequence
547 static apr_status_t getnext(apr_sdbm_datum_t *key, apr_sdbm_t *db)
552 *key = getnkey(db->pagbuf, db->keyptr);
553 if (key->dptr != NULL)
556 * we either run out, or there is nothing on this page..
557 * try the next one... If we lost our position on the
558 * file, we will have to seek.
561 if (db->pagbno != db->blkptr++) {
562 apr_off_t off = OFF_PAG(db->blkptr);
563 if ((status = apr_file_seek(db->pagf, APR_SET, &off)
568 db->pagbno = db->blkptr;
569 /* ### EOF acceptable here too? */
570 if ((status = apr_file_read_full(db->pagf, db->pagbuf, PBLKSIZ, NULL))
573 if (!chkpage(db->pagbuf))
574 return APR_EGENERAL; /* ### need better error */
581 APU_DECLARE(int) apr_sdbm_rdonly(apr_sdbm_t *db)
583 /* ### Should we return true if the first lock is a share lock,
584 * to reflect that apr_sdbm_store and apr_sdbm_delete will fail?
586 return (db->flags & SDBM_RDONLY) != 0;