1/* Licensed to the Apache Software Foundation (ASF) under one or more 2 * contributor license agreements. See the NOTICE file distributed with 3 * this work for additional information regarding copyright ownership. 4 * The ASF licenses this file to You under the Apache License, Version 2.0 5 * (the "License"); you may not use this file except in compliance with 6 * the License. You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 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. 15 */ 16 17/* 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 22 * core routines 23 */ 24 25#include "apr.h" 26#include "apr_file_io.h" 27#include "apr_strings.h" 28#include "apr_errno.h" 29#include "apr_sdbm.h" 30 31#include "sdbm_tune.h" 32#include "sdbm_pair.h" 33#include "sdbm_private.h" 34 35#include <string.h> /* for memset() */ 36#include <stdlib.h> /* for malloc() and free() */ 37 38/* 39 * forward 40 */ 41static int getdbit (apr_sdbm_t *, long); 42static apr_status_t setdbit(apr_sdbm_t *, long); 43static apr_status_t getpage(apr_sdbm_t *db, long, int, int); 44static apr_status_t getnext(apr_sdbm_datum_t *key, apr_sdbm_t *db); 45static apr_status_t makroom(apr_sdbm_t *, long, int); 46 47/* 48 * useful macros 49 */ 50#define bad(x) ((x).dptr == NULL || (x).dsize <= 0) 51#define exhash(item) sdbm_hash((item).dptr, (item).dsize) 52 53#define OFF_PAG(off) (apr_off_t) (off) * PBLKSIZ 54#define OFF_DIR(off) (apr_off_t) (off) * DBLKSIZ 55 56static const long masks[] = { 57 000000000000, 000000000001, 000000000003, 000000000007, 58 000000000017, 000000000037, 000000000077, 000000000177, 59 000000000377, 000000000777, 000000001777, 000000003777, 60 000000007777, 000000017777, 000000037777, 000000077777, 61 000000177777, 000000377777, 000000777777, 000001777777, 62 000003777777, 000007777777, 000017777777, 000037777777, 63 000077777777, 000177777777, 000377777777, 000777777777, 64 001777777777, 003777777777, 007777777777, 017777777777 65}; 66 67const apr_sdbm_datum_t sdbm_nullitem = { NULL, 0 }; 68 69static apr_status_t database_cleanup(void *data) 70{ 71 apr_sdbm_t *db = data; 72 73 /* 74 * Can't rely on apr_sdbm_unlock, since it will merely 75 * decrement the refcnt if several locks are held. 76 */ 77 if (db->flags & (SDBM_SHARED_LOCK | SDBM_EXCLUSIVE_LOCK)) 78 (void) apr_file_unlock(db->dirf); 79 (void) apr_file_close(db->dirf); 80 (void) apr_file_close(db->pagf); 81 free(db); 82 83 return APR_SUCCESS; 84} 85 86static apr_status_t prep(apr_sdbm_t **pdb, const char *dirname, const char *pagname, 87 apr_int32_t flags, apr_fileperms_t perms, apr_pool_t *p) 88{ 89 apr_sdbm_t *db; 90 apr_status_t status; 91 92 *pdb = NULL; 93 94 db = malloc(sizeof(*db)); 95 memset(db, 0, sizeof(*db)); 96 db->pagbno = -1L; 97 98 db->pool = p; 99 100 /* 101 * adjust user flags so that WRONLY becomes RDWR, 102 * as required by this package. Also set our internal 103 * flag for RDONLY if needed. 104 */ 105 if (!(flags & APR_FOPEN_WRITE)) { 106 db->flags |= SDBM_RDONLY; 107 } 108 109 /* 110 * adjust the file open flags so that we handle locking 111 * on our own (don't rely on any locking behavior within 112 * an apr_file_t, in case it's ever introduced, and set 113 * our own flag. 114 */ 115 if (flags & APR_FOPEN_SHARELOCK) { 116 db->flags |= SDBM_SHARED; 117 flags &= ~APR_FOPEN_SHARELOCK; 118 } 119 120 flags |= APR_FOPEN_BINARY | APR_FOPEN_READ; 121 122 /* 123 * open the files in sequence, and stat the dirfile. 124 * If we fail anywhere, undo everything, return NULL. 125 */ 126 127 if ((status = apr_file_open(&db->dirf, dirname, flags, perms, p)) 128 != APR_SUCCESS) 129 goto error; 130 131 if ((status = apr_file_open(&db->pagf, pagname, flags, perms, p)) 132 != APR_SUCCESS) 133 goto error; 134 135 if ((status = apr_sdbm_lock(db, (db->flags & SDBM_RDONLY) 136 ? APR_FLOCK_SHARED 137 : APR_FLOCK_EXCLUSIVE)) 138 != APR_SUCCESS) 139 goto error; 140 141 /* apr_pcalloc zeroed the buffers 142 * apr_sdbm_lock stated the dirf->size and invalidated the cache 143 */ 144 145 /* 146 * if we are opened in SHARED mode, unlock ourself 147 */ 148 if (db->flags & SDBM_SHARED) 149 if ((status = apr_sdbm_unlock(db)) != APR_SUCCESS) 150 goto error; 151 152 /* make sure that we close the database at some point */ 153 apr_pool_cleanup_register(p, db, database_cleanup, apr_pool_cleanup_null); 154 155 /* Done! */ 156 *pdb = db; 157 return APR_SUCCESS; 158 159error: 160 if (db->dirf && db->pagf) 161 (void) apr_sdbm_unlock(db); 162 if (db->dirf != NULL) 163 (void) apr_file_close(db->dirf); 164 if (db->pagf != NULL) { 165 (void) apr_file_close(db->pagf); 166 } 167 free(db); 168 return status; 169} 170 171APU_DECLARE(apr_status_t) apr_sdbm_open(apr_sdbm_t **db, const char *file, 172 apr_int32_t flags, 173 apr_fileperms_t perms, apr_pool_t *p) 174{ 175 char *dirname = apr_pstrcat(p, file, APR_SDBM_DIRFEXT, NULL); 176 char *pagname = apr_pstrcat(p, file, APR_SDBM_PAGFEXT, NULL); 177 178 return prep(db, dirname, pagname, flags, perms, p); 179} 180 181APU_DECLARE(apr_status_t) apr_sdbm_close(apr_sdbm_t *db) 182{ 183 return apr_pool_cleanup_run(db->pool, db, database_cleanup); 184} 185 186APU_DECLARE(apr_status_t) apr_sdbm_fetch(apr_sdbm_t *db, apr_sdbm_datum_t *val, 187 apr_sdbm_datum_t key) 188{ 189 apr_status_t status; 190 191 if (db == NULL || bad(key)) 192 return APR_EINVAL; 193 194 if ((status = apr_sdbm_lock(db, APR_FLOCK_SHARED)) != APR_SUCCESS) 195 return status; 196 197 if ((status = getpage(db, exhash(key), 0, 1)) == APR_SUCCESS) { 198 *val = getpair(db->pagbuf, key); 199 /* ### do we want a not-found result? */ 200 } 201 202 (void) apr_sdbm_unlock(db); 203 204 return status; 205} 206 207static apr_status_t write_page(apr_sdbm_t *db, const char *buf, long pagno) 208{ 209 apr_status_t status; 210 apr_off_t off = OFF_PAG(pagno); 211 212 if ((status = apr_file_seek(db->pagf, APR_SET, &off)) == APR_SUCCESS) 213 status = apr_file_write_full(db->pagf, buf, PBLKSIZ, NULL); 214 215 return status; 216} 217 218APU_DECLARE(apr_status_t) apr_sdbm_delete(apr_sdbm_t *db, 219 const apr_sdbm_datum_t key) 220{ 221 apr_status_t status; 222 223 if (db == NULL || bad(key)) 224 return APR_EINVAL; 225 if (apr_sdbm_rdonly(db)) 226 return APR_EINVAL; 227 228 if ((status = apr_sdbm_lock(db, APR_FLOCK_EXCLUSIVE)) != APR_SUCCESS) 229 return status; 230 231 if ((status = getpage(db, exhash(key), 0, 1)) == APR_SUCCESS) { 232 if (!delpair(db->pagbuf, key)) 233 /* ### should we define some APRUTIL codes? */ 234 status = APR_EGENERAL; 235 else 236 status = write_page(db, db->pagbuf, db->pagbno); 237 } 238 239 (void) apr_sdbm_unlock(db); 240 241 return status; 242} 243 244APU_DECLARE(apr_status_t) apr_sdbm_store(apr_sdbm_t *db, apr_sdbm_datum_t key, 245 apr_sdbm_datum_t val, int flags) 246{ 247 int need; 248 register long hash; 249 apr_status_t status; 250 251 if (db == NULL || bad(key)) 252 return APR_EINVAL; 253 if (apr_sdbm_rdonly(db)) 254 return APR_EINVAL; 255 need = key.dsize + val.dsize; 256 /* 257 * is the pair too big (or too small) for this database ?? 258 */ 259 if (need < 0 || need > PAIRMAX) 260 return APR_EINVAL; 261 262 if ((status = apr_sdbm_lock(db, APR_FLOCK_EXCLUSIVE)) != APR_SUCCESS) 263 return status; 264 265 if ((status = getpage(db, (hash = exhash(key)), 0, 1)) == APR_SUCCESS) { 266 267 /* 268 * if we need to replace, delete the key/data pair 269 * first. If it is not there, ignore. 270 */ 271 if (flags == APR_SDBM_REPLACE) 272 (void) delpair(db->pagbuf, key); 273 else if (!(flags & APR_SDBM_INSERTDUP) && duppair(db->pagbuf, key)) { 274 status = APR_EEXIST; 275 goto error; 276 } 277 /* 278 * if we do not have enough room, we have to split. 279 */ 280 if (!fitpair(db->pagbuf, need)) 281 if ((status = makroom(db, hash, need)) != APR_SUCCESS) 282 goto error; 283 /* 284 * we have enough room or split is successful. insert the key, 285 * and update the page file. 286 */ 287 (void) putpair(db->pagbuf, key, val); 288 289 status = write_page(db, db->pagbuf, db->pagbno); 290 } 291 292error: 293 (void) apr_sdbm_unlock(db); 294 295 return status; 296} 297 298/* 299 * makroom - make room by splitting the overfull page 300 * this routine will attempt to make room for SPLTMAX times before 301 * giving up. 302 */ 303static apr_status_t makroom(apr_sdbm_t *db, long hash, int need) 304{ 305 long newp; 306 char twin[PBLKSIZ]; 307 char *pag = db->pagbuf; 308 char *new = twin; 309 register int smax = SPLTMAX; 310 apr_status_t status; 311 312 do { 313 /* 314 * split the current page 315 */ 316 (void) splpage(pag, new, db->hmask + 1); 317 /* 318 * address of the new page 319 */ 320 newp = (hash & db->hmask) | (db->hmask + 1); 321 322 /* 323 * write delay, read avoidence/cache shuffle: 324 * select the page for incoming pair: if key is to go to the new page, 325 * write out the previous one, and copy the new one over, thus making 326 * it the current page. If not, simply write the new page, and we are 327 * still looking at the page of interest. current page is not updated 328 * here, as sdbm_store will do so, after it inserts the incoming pair. 329 */ 330 if (hash & (db->hmask + 1)) { 331 if ((status = write_page(db, db->pagbuf, db->pagbno)) 332 != APR_SUCCESS) 333 return status; 334 335 db->pagbno = newp; 336 (void) memcpy(pag, new, PBLKSIZ); 337 } 338 else { 339 if ((status = write_page(db, new, newp)) != APR_SUCCESS) 340 return status; 341 } 342 343 if ((status = setdbit(db, db->curbit)) != APR_SUCCESS) 344 return status; 345 /* 346 * see if we have enough room now 347 */ 348 if (fitpair(pag, need)) 349 return APR_SUCCESS; 350 /* 351 * try again... update curbit and hmask as getpage would have 352 * done. because of our update of the current page, we do not 353 * need to read in anything. BUT we have to write the current 354 * [deferred] page out, as the window of failure is too great. 355 */ 356 db->curbit = 2 * db->curbit 357 + ((hash & (db->hmask + 1)) ? 2 : 1); 358 db->hmask |= db->hmask + 1; 359 360 if ((status = write_page(db, db->pagbuf, db->pagbno)) 361 != APR_SUCCESS) 362 return status; 363 364 } while (--smax); 365 366 /* 367 * if we are here, this is real bad news. After SPLTMAX splits, 368 * we still cannot fit the key. say goodnight. 369 */ 370#if 0 371 (void) write(2, "sdbm: cannot insert after SPLTMAX attempts.\n", 44); 372#endif 373 /* ### ENOSPC not really appropriate but better than nothing */ 374 return APR_ENOSPC; 375 376} 377 378/* Reads 'len' bytes from file 'f' at offset 'off' into buf. 379 * 'off' is given relative to the start of the file. 380 * If 'create' is asked and EOF is returned while reading, this is taken 381 * as success (i.e. a cleared buffer is returned). 382 */ 383static apr_status_t read_from(apr_file_t *f, void *buf, 384 apr_off_t off, apr_size_t len, 385 int create) 386{ 387 apr_status_t status; 388 389 if ((status = apr_file_seek(f, APR_SET, &off)) != APR_SUCCESS || 390 ((status = apr_file_read_full(f, buf, len, NULL)) != APR_SUCCESS)) { 391 /* if EOF is reached, pretend we read all zero's */ 392 if (status == APR_EOF && create) { 393 memset(buf, 0, len); 394 status = APR_SUCCESS; 395 } 396 } 397 398 return status; 399} 400 401/* 402 * the following two routines will break if 403 * deletions aren't taken into account. (ndbm bug) 404 */ 405APU_DECLARE(apr_status_t) apr_sdbm_firstkey(apr_sdbm_t *db, 406 apr_sdbm_datum_t *key) 407{ 408 apr_status_t status; 409 410 if ((status = apr_sdbm_lock(db, APR_FLOCK_SHARED)) != APR_SUCCESS) 411 return status; 412 413 /* 414 * start at page 0 415 */ 416 if ((status = getpage(db, 0, 1, 1)) == APR_SUCCESS) { 417 db->blkptr = 0; 418 db->keyptr = 0; 419 status = getnext(key, db); 420 } 421 422 (void) apr_sdbm_unlock(db); 423 424 return status; 425} 426 427APU_DECLARE(apr_status_t) apr_sdbm_nextkey(apr_sdbm_t *db, 428 apr_sdbm_datum_t *key) 429{ 430 apr_status_t status; 431 432 if ((status = apr_sdbm_lock(db, APR_FLOCK_SHARED)) != APR_SUCCESS) 433 return status; 434 435 status = getnext(key, db); 436 437 (void) apr_sdbm_unlock(db); 438 439 return status; 440} 441 442/* 443 * all important binary tree traversal 444 */ 445static apr_status_t getpage(apr_sdbm_t *db, long hash, int by_num, int create) 446{ 447 apr_status_t status; 448 register long pagb; 449 450 if (by_num) { 451 pagb = hash; 452 } 453 else { 454 register int hbit = 0; 455 register long dbit = 0; 456 457 while (dbit < db->maxbno && getdbit(db, dbit)) 458 dbit = 2 * dbit + ((hash & (1 << hbit++)) ? 2 : 1); 459 debug(("dbit: %d...", dbit)); 460 461 db->curbit = dbit; 462 db->hmask = masks[hbit]; 463 464 pagb = hash & db->hmask; 465 } 466 467 /* 468 * see if the block we need is already in memory. 469 * note: this lookaside cache has about 10% hit rate. 470 */ 471 if (pagb != db->pagbno) { 472 /* 473 * note: here, we assume a "hole" is read as 0s. 474 * if not, must zero pagbuf first. 475 * ### joe: this assumption was surely never correct? but 476 * ### we make it so in read_from anyway. 477 */ 478 if ((status = read_from(db->pagf, db->pagbuf, 479 OFF_PAG(pagb), PBLKSIZ, 480 create)) != APR_SUCCESS) 481 return status; 482 483 if (!chkpage(db->pagbuf)) 484 return APR_ENOSPC; /* ### better error? */ 485 486 db->pagbno = pagb; 487 488 debug(("pag read: %d\n", pagb)); 489 } 490 return APR_SUCCESS; 491} 492 493static int getdbit(apr_sdbm_t *db, long dbit) 494{ 495 register long c; 496 register long dirb; 497 498 c = dbit / BYTESIZ; 499 dirb = c / DBLKSIZ; 500 501 if (dirb != db->dirbno) { 502 if (read_from(db->dirf, db->dirbuf, 503 OFF_DIR(dirb), DBLKSIZ, 504 1) != APR_SUCCESS) 505 return 0; 506 507 db->dirbno = dirb; 508 509 debug(("dir read: %d\n", dirb)); 510 } 511 512 return db->dirbuf[c % DBLKSIZ] & (1 << dbit % BYTESIZ); 513} 514 515static apr_status_t setdbit(apr_sdbm_t *db, long dbit) 516{ 517 register long c; 518 register long dirb; 519 apr_status_t status; 520 apr_off_t off; 521 522 c = dbit / BYTESIZ; 523 dirb = c / DBLKSIZ; 524 525 if (dirb != db->dirbno) { 526 if ((status = read_from(db->dirf, db->dirbuf, 527 OFF_DIR(dirb), DBLKSIZ, 528 1)) != APR_SUCCESS) 529 return status; 530 531 db->dirbno = dirb; 532 533 debug(("dir read: %d\n", dirb)); 534 } 535 536 db->dirbuf[c % DBLKSIZ] |= (1 << dbit % BYTESIZ); 537 538 if (dbit >= db->maxbno) 539 db->maxbno += DBLKSIZ * BYTESIZ; 540 541 off = OFF_DIR(dirb); 542 if ((status = apr_file_seek(db->dirf, APR_SET, &off)) == APR_SUCCESS) 543 status = apr_file_write_full(db->dirf, db->dirbuf, DBLKSIZ, NULL); 544 545 return status; 546} 547 548/* 549* getnext - get the next key in the page, and if done with 550* the page, try the next page in sequence 551*/ 552static apr_status_t getnext(apr_sdbm_datum_t *key, apr_sdbm_t *db) 553{ 554 apr_status_t status; 555 for (;;) { 556 db->keyptr++; 557 *key = getnkey(db->pagbuf, db->keyptr); 558 if (key->dptr != NULL) 559 return APR_SUCCESS; 560 /* 561 * we either run out, or there is nothing on this page.. 562 * try the next one... If we lost our position on the 563 * file, we will have to seek. 564 */ 565 db->blkptr++; 566 db->keyptr = 0; 567 568 /* ### EOF acceptable here too? */ 569 if ((status = getpage(db, db->blkptr, 1, 0)) != APR_SUCCESS) 570 return status; 571 } 572 573 /* NOTREACHED */ 574} 575 576 577APU_DECLARE(int) apr_sdbm_rdonly(apr_sdbm_t *db) 578{ 579 /* ### Should we return true if the first lock is a share lock, 580 * to reflect that apr_sdbm_store and apr_sdbm_delete will fail? 581 */ 582 return (db->flags & SDBM_RDONLY) != 0; 583} 584 585