1 /* $NetBSD: hash_page.c,v 1.18 2006/03/26 02:00:37 rtr Exp $ */ 2 3 /*- 4 * Copyright (c) 1990, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Margo Seltzer. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 #if defined(LIBC_SCCS) && !defined(lint) 37 #if 0 38 static char sccsid[] = "@(#)hash_page.c 8.7 (Berkeley) 8/16/94"; 39 #else 40 __RCSID("$NetBSD: hash_page.c,v 1.18 2006/03/26 02:00:37 rtr Exp $"); 41 #endif 42 #endif /* LIBC_SCCS and not lint */ 43 44 /* 45 * PACKAGE: hashing 46 * 47 * DESCRIPTION: 48 * Page manipulation for hashing package. 49 * 50 * ROUTINES: 51 * 52 * External 53 * __get_page 54 * __add_ovflpage 55 * Internal 56 * overflow_page 57 * open_temp 58 */ 59 60 #include "namespace.h" 61 62 #include <sys/types.h> 63 64 #include <errno.h> 65 #include <fcntl.h> 66 #include <signal.h> 67 #include <stdio.h> 68 #include <stdlib.h> 69 #include <string.h> 70 #include <unistd.h> 71 #include <paths.h> 72 #ifdef DEBUG 73 #include <assert.h> 74 #endif 75 76 #include <db.h> 77 #include "hash.h" 78 #include "page.h" 79 #include "extern.h" 80 81 static u_int32_t *fetch_bitmap __P((HTAB *, int)); 82 static u_int32_t first_free __P((u_int32_t)); 83 static int open_temp __P((HTAB *)); 84 static u_int16_t overflow_page __P((HTAB *)); 85 static void putpair __P((char *, const DBT *, const DBT *)); 86 static void squeeze_key __P((u_int16_t *, const DBT *, const DBT *)); 87 static int ugly_split 88 __P((HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int)); 89 90 #define PAGE_INIT(P) { \ 91 ((u_int16_t *)(void *)(P))[0] = 0; \ 92 ((u_int16_t *)(void *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \ 93 ((u_int16_t *)(void *)(P))[2] = hashp->BSIZE; \ 94 } 95 96 /* 97 * This is called AFTER we have verified that there is room on the page for 98 * the pair (PAIRFITS has returned true) so we go right ahead and start moving 99 * stuff on. 100 */ 101 static void 102 putpair(p, key, val) 103 char *p; 104 const DBT *key, *val; 105 { 106 register u_int16_t *bp, n, off; 107 108 bp = (u_int16_t *)(void *)p; 109 110 /* Enter the key first. */ 111 n = bp[0]; 112 113 off = OFFSET(bp) - key->size; 114 memmove(p + off, key->data, key->size); 115 bp[++n] = off; 116 117 /* Now the data. */ 118 off -= val->size; 119 memmove(p + off, val->data, val->size); 120 bp[++n] = off; 121 122 /* Adjust page info. */ 123 bp[0] = n; 124 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t)); 125 bp[n + 2] = off; 126 } 127 128 /* 129 * Returns: 130 * 0 OK 131 * -1 error 132 */ 133 extern int 134 __delpair(hashp, bufp, ndx) 135 HTAB *hashp; 136 BUFHEAD *bufp; 137 register int ndx; 138 { 139 register u_int16_t *bp, newoff; 140 register int n; 141 u_int16_t pairlen; 142 143 bp = (u_int16_t *)(void *)bufp->page; 144 n = bp[0]; 145 146 if (bp[ndx + 1] < REAL_KEY) 147 return (__big_delete(hashp, bufp)); 148 if (ndx != 1) 149 newoff = bp[ndx - 1]; 150 else 151 newoff = hashp->BSIZE; 152 pairlen = newoff - bp[ndx + 1]; 153 154 if (ndx != (n - 1)) { 155 /* Hard Case -- need to shuffle keys */ 156 register int i; 157 register char *src = bufp->page + (int)OFFSET(bp); 158 register char *dst = src + (int)pairlen; 159 memmove(dst, src, (size_t)(bp[ndx + 1] - OFFSET(bp))); 160 161 /* Now adjust the pointers */ 162 for (i = ndx + 2; i <= n; i += 2) { 163 if (bp[i + 1] == OVFLPAGE) { 164 bp[i - 2] = bp[i]; 165 bp[i - 1] = bp[i + 1]; 166 } else { 167 bp[i - 2] = bp[i] + pairlen; 168 bp[i - 1] = bp[i + 1] + pairlen; 169 } 170 } 171 } 172 /* Finally adjust the page data */ 173 bp[n] = OFFSET(bp) + pairlen; 174 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t); 175 bp[0] = n - 2; 176 hashp->NKEYS--; 177 178 bufp->flags |= BUF_MOD; 179 return (0); 180 } 181 /* 182 * Returns: 183 * 0 ==> OK 184 * -1 ==> Error 185 */ 186 extern int 187 __split_page(hashp, obucket, nbucket) 188 HTAB *hashp; 189 u_int32_t obucket, nbucket; 190 { 191 register BUFHEAD *new_bufp, *old_bufp; 192 register u_int16_t *ino; 193 register char *np; 194 DBT key, val; 195 int n, ndx, retval; 196 u_int16_t copyto, diff, off, moved; 197 char *op; 198 199 copyto = (u_int16_t)hashp->BSIZE; 200 off = (u_int16_t)hashp->BSIZE; 201 old_bufp = __get_buf(hashp, obucket, NULL, 0); 202 if (old_bufp == NULL) 203 return (-1); 204 new_bufp = __get_buf(hashp, nbucket, NULL, 0); 205 if (new_bufp == NULL) 206 return (-1); 207 208 old_bufp->flags |= (BUF_MOD | BUF_PIN); 209 new_bufp->flags |= (BUF_MOD | BUF_PIN); 210 211 ino = (u_int16_t *)(void *)(op = old_bufp->page); 212 np = new_bufp->page; 213 214 moved = 0; 215 216 for (n = 1, ndx = 1; n < ino[0]; n += 2) { 217 if (ino[n + 1] < REAL_KEY) { 218 retval = ugly_split(hashp, obucket, old_bufp, new_bufp, 219 (int)copyto, (int)moved); 220 old_bufp->flags &= ~BUF_PIN; 221 new_bufp->flags &= ~BUF_PIN; 222 return (retval); 223 224 } 225 key.data = (u_char *)op + ino[n]; 226 key.size = off - ino[n]; 227 228 if (__call_hash(hashp, key.data, (int)key.size) == obucket) { 229 /* Don't switch page */ 230 diff = copyto - off; 231 if (diff) { 232 copyto = ino[n + 1] + diff; 233 memmove(op + copyto, op + ino[n + 1], 234 (size_t)(off - ino[n + 1])); 235 ino[ndx] = copyto + ino[n] - ino[n + 1]; 236 ino[ndx + 1] = copyto; 237 } else 238 copyto = ino[n + 1]; 239 ndx += 2; 240 } else { 241 /* Switch page */ 242 val.data = (u_char *)op + ino[n + 1]; 243 val.size = ino[n] - ino[n + 1]; 244 putpair(np, &key, &val); 245 moved += 2; 246 } 247 248 off = ino[n + 1]; 249 } 250 251 /* Now clean up the page */ 252 ino[0] -= moved; 253 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3); 254 OFFSET(ino) = copyto; 255 256 #ifdef DEBUG3 257 (void)fprintf(stderr, "split %d/%d\n", 258 ((u_int16_t *)np)[0] / 2, 259 ((u_int16_t *)op)[0] / 2); 260 #endif 261 /* unpin both pages */ 262 old_bufp->flags &= ~BUF_PIN; 263 new_bufp->flags &= ~BUF_PIN; 264 return (0); 265 } 266 267 /* 268 * Called when we encounter an overflow or big key/data page during split 269 * handling. This is special cased since we have to begin checking whether 270 * the key/data pairs fit on their respective pages and because we may need 271 * overflow pages for both the old and new pages. 272 * 273 * The first page might be a page with regular key/data pairs in which case 274 * we have a regular overflow condition and just need to go on to the next 275 * page or it might be a big key/data pair in which case we need to fix the 276 * big key/data pair. 277 * 278 * Returns: 279 * 0 ==> success 280 * -1 ==> failure 281 */ 282 static int 283 ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved) 284 HTAB *hashp; 285 u_int32_t obucket; /* Same as __split_page. */ 286 BUFHEAD *old_bufp, *new_bufp; 287 int copyto; /* First byte on page which contains key/data values. */ 288 int moved; /* Number of pairs moved to new page. */ 289 { 290 register BUFHEAD *bufp; /* Buffer header for ino */ 291 register u_int16_t *ino; /* Page keys come off of */ 292 register u_int16_t *np; /* New page */ 293 register u_int16_t *op; /* Page keys go on to if they aren't moving */ 294 295 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */ 296 DBT key, val; 297 SPLIT_RETURN ret; 298 u_int16_t n, off, ov_addr, scopyto; 299 char *cino; /* Character value of ino */ 300 301 bufp = old_bufp; 302 ino = (u_int16_t *)(void *)old_bufp->page; 303 np = (u_int16_t *)(void *)new_bufp->page; 304 op = (u_int16_t *)(void *)old_bufp->page; 305 last_bfp = NULL; 306 scopyto = (u_int16_t)copyto; /* ANSI */ 307 308 n = ino[0] - 1; 309 while (n < ino[0]) { 310 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) { 311 if (__big_split(hashp, old_bufp, 312 new_bufp, bufp, (int)bufp->addr, obucket, &ret)) 313 return (-1); 314 old_bufp = ret.oldp; 315 if (!old_bufp) 316 return (-1); 317 op = (u_int16_t *)(void *)old_bufp->page; 318 new_bufp = ret.newp; 319 if (!new_bufp) 320 return (-1); 321 np = (u_int16_t *)(void *)new_bufp->page; 322 bufp = ret.nextp; 323 if (!bufp) 324 return (0); 325 cino = (char *)bufp->page; 326 ino = (u_int16_t *)(void *)cino; 327 last_bfp = ret.nextp; 328 } else if (ino[n + 1] == OVFLPAGE) { 329 ov_addr = ino[n]; 330 /* 331 * Fix up the old page -- the extra 2 are the fields 332 * which contained the overflow information. 333 */ 334 ino[0] -= (moved + 2); 335 FREESPACE(ino) = 336 scopyto - sizeof(u_int16_t) * (ino[0] + 3); 337 OFFSET(ino) = scopyto; 338 339 bufp = __get_buf(hashp, (u_int32_t)ov_addr, bufp, 0); 340 if (!bufp) 341 return (-1); 342 343 ino = (u_int16_t *)(void *)bufp->page; 344 n = 1; 345 scopyto = hashp->BSIZE; 346 moved = 0; 347 348 if (last_bfp) 349 __free_ovflpage(hashp, last_bfp); 350 last_bfp = bufp; 351 } 352 /* Move regular sized pairs of there are any */ 353 off = hashp->BSIZE; 354 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) { 355 cino = (char *)(void *)ino; 356 key.data = (u_char *)cino + ino[n]; 357 key.size = off - ino[n]; 358 val.data = (u_char *)cino + ino[n + 1]; 359 val.size = ino[n] - ino[n + 1]; 360 off = ino[n + 1]; 361 362 if (__call_hash(hashp, key.data, (int)key.size) == obucket) { 363 /* Keep on old page */ 364 if (PAIRFITS(op, (&key), (&val))) 365 putpair((char *)(void *)op, &key, &val); 366 else { 367 old_bufp = 368 __add_ovflpage(hashp, old_bufp); 369 if (!old_bufp) 370 return (-1); 371 op = (u_int16_t *)(void *)old_bufp->page; 372 putpair((char *)(void *)op, &key, &val); 373 } 374 old_bufp->flags |= BUF_MOD; 375 } else { 376 /* Move to new page */ 377 if (PAIRFITS(np, (&key), (&val))) 378 putpair((char *)(void *)np, &key, &val); 379 else { 380 new_bufp = 381 __add_ovflpage(hashp, new_bufp); 382 if (!new_bufp) 383 return (-1); 384 np = (u_int16_t *)(void *)new_bufp->page; 385 putpair((char *)(void *)np, &key, &val); 386 } 387 new_bufp->flags |= BUF_MOD; 388 } 389 } 390 } 391 if (last_bfp) 392 __free_ovflpage(hashp, last_bfp); 393 return (0); 394 } 395 396 /* 397 * Add the given pair to the page 398 * 399 * Returns: 400 * 0 ==> OK 401 * 1 ==> failure 402 */ 403 extern int 404 __addel(hashp, bufp, key, val) 405 HTAB *hashp; 406 BUFHEAD *bufp; 407 const DBT *key, *val; 408 { 409 register u_int16_t *bp, *sop; 410 int do_expand; 411 412 bp = (u_int16_t *)(void *)bufp->page; 413 do_expand = 0; 414 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY)) 415 /* Exception case */ 416 if (bp[2] == FULL_KEY_DATA && bp[0] == 2) 417 /* This is the last page of a big key/data pair 418 and we need to add another page */ 419 break; 420 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) { 421 bufp = __get_buf(hashp, (u_int32_t)bp[bp[0] - 1], bufp, 422 0); 423 if (!bufp) 424 return (-1); 425 bp = (u_int16_t *)(void *)bufp->page; 426 } else if (bp[bp[0]] != OVFLPAGE) { 427 /* Short key/data pairs, no more pages */ 428 break; 429 } else { 430 /* Try to squeeze key on this page */ 431 if (bp[2] >= REAL_KEY && 432 FREESPACE(bp) >= PAIRSIZE(key, val)) { 433 squeeze_key(bp, key, val); 434 goto stats; 435 } else { 436 bufp = __get_buf(hashp, 437 (u_int32_t)bp[bp[0] - 1], bufp, 0); 438 if (!bufp) 439 return (-1); 440 bp = (u_int16_t *)(void *)bufp->page; 441 } 442 } 443 444 if (PAIRFITS(bp, key, val)) 445 putpair(bufp->page, key, val); 446 else { 447 do_expand = 1; 448 bufp = __add_ovflpage(hashp, bufp); 449 if (!bufp) 450 return (-1); 451 sop = (u_int16_t *)(void *)bufp->page; 452 453 if (PAIRFITS(sop, key, val)) 454 putpair((char *)(void *)sop, key, val); 455 else 456 if (__big_insert(hashp, bufp, key, val)) 457 return (-1); 458 } 459 stats: 460 bufp->flags |= BUF_MOD; 461 /* 462 * If the average number of keys per bucket exceeds the fill factor, 463 * expand the table. 464 */ 465 hashp->NKEYS++; 466 if (do_expand || 467 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR)) 468 return (__expand_table(hashp)); 469 return (0); 470 } 471 472 /* 473 * 474 * Returns: 475 * pointer on success 476 * NULL on error 477 */ 478 extern BUFHEAD * 479 __add_ovflpage(hashp, bufp) 480 HTAB *hashp; 481 BUFHEAD *bufp; 482 { 483 register u_int16_t *sp; 484 u_int16_t ndx, ovfl_num; 485 #ifdef DEBUG1 486 int tmp1, tmp2; 487 #endif 488 sp = (u_int16_t *)(void *)bufp->page; 489 490 /* Check if we are dynamically determining the fill factor */ 491 if (hashp->FFACTOR == DEF_FFACTOR) { 492 hashp->FFACTOR = (u_int32_t)sp[0] >> 1; 493 if (hashp->FFACTOR < MIN_FFACTOR) 494 hashp->FFACTOR = MIN_FFACTOR; 495 } 496 bufp->flags |= BUF_MOD; 497 ovfl_num = overflow_page(hashp); 498 #ifdef DEBUG1 499 tmp1 = bufp->addr; 500 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0; 501 #endif 502 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, (u_int32_t)ovfl_num, 503 bufp, 1))) 504 return (NULL); 505 bufp->ovfl->flags |= BUF_MOD; 506 #ifdef DEBUG1 507 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n", 508 tmp1, tmp2, bufp->ovfl->addr); 509 #endif 510 ndx = sp[0]; 511 /* 512 * Since a pair is allocated on a page only if there's room to add 513 * an overflow page, we know that the OVFL information will fit on 514 * the page. 515 */ 516 sp[ndx + 4] = OFFSET(sp); 517 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE; 518 sp[ndx + 1] = ovfl_num; 519 sp[ndx + 2] = OVFLPAGE; 520 sp[0] = ndx + 2; 521 #ifdef HASH_STATISTICS 522 hash_overflows++; 523 #endif 524 return (bufp->ovfl); 525 } 526 527 /* 528 * Returns: 529 * 0 indicates SUCCESS 530 * -1 indicates FAILURE 531 */ 532 extern int 533 __get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap) 534 HTAB *hashp; 535 char *p; 536 u_int32_t bucket; 537 int is_bucket, is_disk, is_bitmap; 538 { 539 register int fd, page, size; 540 int rsize; 541 u_int16_t *bp; 542 543 fd = hashp->fp; 544 size = hashp->BSIZE; 545 546 if ((fd == -1) || !is_disk) { 547 PAGE_INIT(p); 548 return (0); 549 } 550 if (is_bucket) 551 page = BUCKET_TO_PAGE(bucket); 552 else 553 page = OADDR_TO_PAGE(bucket); 554 if ((rsize = pread(fd, p, (size_t)size, (off_t)page << hashp->BSHIFT)) == -1) 555 return (-1); 556 bp = (u_int16_t *)(void *)p; 557 if (!rsize) 558 bp[0] = 0; /* We hit the EOF, so initialize a new page */ 559 else 560 if (rsize != size) { 561 errno = EFTYPE; 562 return (-1); 563 } 564 if (!is_bitmap && !bp[0]) { 565 PAGE_INIT(p); 566 } else 567 if (hashp->LORDER != BYTE_ORDER) { 568 register int i, max; 569 570 if (is_bitmap) { 571 max = (u_int32_t)hashp->BSIZE >> 2; /* divide by 4 */ 572 for (i = 0; i < max; i++) 573 M_32_SWAP(((int *)(void *)p)[i]); 574 } else { 575 M_16_SWAP(bp[0]); 576 max = bp[0] + 2; 577 for (i = 1; i <= max; i++) 578 M_16_SWAP(bp[i]); 579 } 580 } 581 return (0); 582 } 583 584 /* 585 * Write page p to disk 586 * 587 * Returns: 588 * 0 ==> OK 589 * -1 ==>failure 590 */ 591 extern int 592 __put_page(hashp, p, bucket, is_bucket, is_bitmap) 593 HTAB *hashp; 594 char *p; 595 u_int32_t bucket; 596 int is_bucket, is_bitmap; 597 { 598 register int fd, page, size; 599 int wsize; 600 601 size = hashp->BSIZE; 602 if ((hashp->fp == -1) && open_temp(hashp)) 603 return (-1); 604 fd = hashp->fp; 605 606 if (hashp->LORDER != BYTE_ORDER) { 607 register int i; 608 register int max; 609 610 if (is_bitmap) { 611 max = (u_int32_t)hashp->BSIZE >> 2; /* divide by 4 */ 612 for (i = 0; i < max; i++) 613 M_32_SWAP(((int *)(void *)p)[i]); 614 } else { 615 max = ((u_int16_t *)(void *)p)[0] + 2; 616 for (i = 0; i <= max; i++) 617 M_16_SWAP(((u_int16_t *)(void *)p)[i]); 618 } 619 } 620 if (is_bucket) 621 page = BUCKET_TO_PAGE(bucket); 622 else 623 page = OADDR_TO_PAGE(bucket); 624 if ((wsize = pwrite(fd, p, (size_t)size, (off_t)page << hashp->BSHIFT)) == -1) 625 /* Errno is set */ 626 return (-1); 627 if (wsize != size) { 628 errno = EFTYPE; 629 return (-1); 630 } 631 return (0); 632 } 633 634 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1) 635 /* 636 * Initialize a new bitmap page. Bitmap pages are left in memory 637 * once they are read in. 638 */ 639 extern int 640 __ibitmap(hashp, pnum, nbits, ndx) 641 HTAB *hashp; 642 int pnum, nbits, ndx; 643 { 644 u_int32_t *ip; 645 int clearbytes, clearints; 646 647 if ((ip = (u_int32_t *)malloc((size_t)hashp->BSIZE)) == NULL) 648 return (1); 649 hashp->nmaps++; 650 clearints = ((u_int32_t)(nbits - 1) >> INT_BYTE_SHIFT) + 1; 651 clearbytes = clearints << INT_TO_BYTE; 652 (void)memset(ip, 0, (size_t)clearbytes); 653 (void)memset(((char *)(void *)ip) + clearbytes, 0xFF, 654 (size_t)(hashp->BSIZE - clearbytes)); 655 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK); 656 SETBIT(ip, 0); 657 hashp->BITMAPS[ndx] = (u_int16_t)pnum; 658 hashp->mapp[ndx] = ip; 659 return (0); 660 } 661 662 static u_int32_t 663 first_free(map) 664 u_int32_t map; 665 { 666 register u_int32_t i, mask; 667 668 mask = 0x1; 669 for (i = 0; i < BITS_PER_MAP; i++) { 670 if (!(mask & map)) 671 return (i); 672 mask = mask << 1; 673 } 674 return (i); 675 } 676 677 static u_int16_t 678 overflow_page(hashp) 679 HTAB *hashp; 680 { 681 register u_int32_t *freep = NULL; 682 register int max_free, offset, splitnum; 683 u_int16_t addr; 684 int bit, first_page, free_bit, free_page, i, in_use_bits, j; 685 #ifdef DEBUG2 686 int tmp1, tmp2; 687 #endif 688 splitnum = hashp->OVFL_POINT; 689 max_free = hashp->SPARES[splitnum]; 690 691 free_page = (u_int32_t)(max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT); 692 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1); 693 694 /* Look through all the free maps to find the first free block */ 695 first_page = (u_int32_t)hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT); 696 for ( i = first_page; i <= free_page; i++ ) { 697 if (!(freep = (u_int32_t *)hashp->mapp[i]) && 698 !(freep = fetch_bitmap(hashp, i))) 699 return (0); 700 if (i == free_page) 701 in_use_bits = free_bit; 702 else 703 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1; 704 705 if (i == first_page) { 706 bit = hashp->LAST_FREED & 707 ((hashp->BSIZE << BYTE_SHIFT) - 1); 708 j = bit / BITS_PER_MAP; 709 bit = bit & ~(BITS_PER_MAP - 1); 710 } else { 711 bit = 0; 712 j = 0; 713 } 714 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP) 715 if (freep[j] != ALL_SET) 716 goto found; 717 } 718 719 /* No Free Page Found */ 720 hashp->LAST_FREED = hashp->SPARES[splitnum]; 721 hashp->SPARES[splitnum]++; 722 offset = hashp->SPARES[splitnum] - 723 (splitnum ? hashp->SPARES[splitnum - 1] : 0); 724 725 #define OVMSG "HASH: Out of overflow pages. Increase page size\n" 726 if (offset > SPLITMASK) { 727 if (++splitnum >= NCACHED) { 728 (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 729 errno = EFBIG; 730 return (0); 731 } 732 hashp->OVFL_POINT = splitnum; 733 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 734 hashp->SPARES[splitnum-1]--; 735 offset = 1; 736 } 737 738 /* Check if we need to allocate a new bitmap page */ 739 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) { 740 free_page++; 741 if (free_page >= NCACHED) { 742 (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 743 errno = EFBIG; 744 return (0); 745 } 746 /* 747 * This is tricky. The 1 indicates that you want the new page 748 * allocated with 1 clear bit. Actually, you are going to 749 * allocate 2 pages from this map. The first is going to be 750 * the map page, the second is the overflow page we were 751 * looking for. The init_bitmap routine automatically, sets 752 * the first bit of itself to indicate that the bitmap itself 753 * is in use. We would explicitly set the second bit, but 754 * don't have to if we tell init_bitmap not to leave it clear 755 * in the first place. 756 */ 757 if (__ibitmap(hashp, 758 (int)OADDR_OF(splitnum, offset), 1, free_page)) 759 return (0); 760 hashp->SPARES[splitnum]++; 761 #ifdef DEBUG2 762 free_bit = 2; 763 #endif 764 offset++; 765 if (offset > SPLITMASK) { 766 if (++splitnum >= NCACHED) { 767 (void)write(STDERR_FILENO, OVMSG, 768 sizeof(OVMSG) - 1); 769 errno = EFBIG; 770 return (0); 771 } 772 hashp->OVFL_POINT = splitnum; 773 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 774 hashp->SPARES[splitnum-1]--; 775 offset = 0; 776 } 777 } else { 778 /* 779 * Free_bit addresses the last used bit. Bump it to address 780 * the first available bit. 781 */ 782 free_bit++; 783 SETBIT(freep, free_bit); 784 } 785 786 /* Calculate address of the new overflow page */ 787 addr = OADDR_OF(splitnum, offset); 788 #ifdef DEBUG2 789 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 790 addr, free_bit, free_page); 791 #endif 792 return (addr); 793 794 found: 795 bit = bit + first_free(freep[j]); 796 SETBIT(freep, bit); 797 #ifdef DEBUG2 798 tmp1 = bit; 799 tmp2 = i; 800 #endif 801 /* 802 * Bits are addressed starting with 0, but overflow pages are addressed 803 * beginning at 1. Bit is a bit addressnumber, so we need to increment 804 * it to convert it to a page number. 805 */ 806 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT)); 807 if (bit >= hashp->LAST_FREED) 808 hashp->LAST_FREED = bit - 1; 809 810 /* Calculate the split number for this page */ 811 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++); 812 offset = (i ? bit - hashp->SPARES[i - 1] : bit); 813 if (offset >= SPLITMASK) { 814 (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 815 errno = EFBIG; 816 return (0); /* Out of overflow pages */ 817 } 818 addr = OADDR_OF(i, offset); 819 #ifdef DEBUG2 820 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 821 addr, tmp1, tmp2); 822 #endif 823 824 /* Allocate and return the overflow page */ 825 return (addr); 826 } 827 828 /* 829 * Mark this overflow page as free. 830 */ 831 extern void 832 __free_ovflpage(hashp, obufp) 833 HTAB *hashp; 834 BUFHEAD *obufp; 835 { 836 register u_int16_t addr; 837 u_int32_t *freep; 838 int bit_address, free_page, free_bit; 839 u_int16_t ndx; 840 841 addr = obufp->addr; 842 #ifdef DEBUG1 843 (void)fprintf(stderr, "Freeing %d\n", addr); 844 #endif 845 ndx = (((u_int32_t)addr) >> SPLITSHIFT); 846 bit_address = 847 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1; 848 if (bit_address < hashp->LAST_FREED) 849 hashp->LAST_FREED = bit_address; 850 free_page = ((u_int32_t)bit_address >> (hashp->BSHIFT + BYTE_SHIFT)); 851 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1); 852 853 if (!(freep = hashp->mapp[free_page])) 854 freep = fetch_bitmap(hashp, free_page); 855 #ifdef DEBUG 856 /* 857 * This had better never happen. It means we tried to read a bitmap 858 * that has already had overflow pages allocated off it, and we 859 * failed to read it from the file. 860 */ 861 assert(freep != NULL); 862 #endif 863 CLRBIT(freep, free_bit); 864 #ifdef DEBUG2 865 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n", 866 obufp->addr, free_bit, free_page); 867 #endif 868 __reclaim_buf(hashp, obufp); 869 } 870 871 /* 872 * Returns: 873 * 0 success 874 * -1 failure 875 */ 876 static int 877 open_temp(hashp) 878 HTAB *hashp; 879 { 880 sigset_t set, oset; 881 char *envtmp; 882 char namestr[PATH_MAX]; 883 884 if (issetugid()) 885 envtmp = NULL; 886 else 887 envtmp = getenv("TMPDIR"); 888 889 if (-1 == snprintf(namestr, sizeof(namestr), "%s/_hashXXXXXX", 890 envtmp ? envtmp : _PATH_TMP)) 891 return -1; 892 893 /* Block signals; make sure file goes away at process exit. */ 894 (void)sigfillset(&set); 895 (void)sigprocmask(SIG_BLOCK, &set, &oset); 896 if ((hashp->fp = mkstemp(namestr)) != -1) { 897 (void)unlink(namestr); 898 (void)fcntl(hashp->fp, F_SETFD, 1); 899 } 900 (void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL); 901 return (hashp->fp != -1 ? 0 : -1); 902 } 903 904 /* 905 * We have to know that the key will fit, but the last entry on the page is 906 * an overflow pair, so we need to shift things. 907 */ 908 static void 909 squeeze_key(sp, key, val) 910 u_int16_t *sp; 911 const DBT *key, *val; 912 { 913 register char *p; 914 u_int16_t free_space, n, off, pageno; 915 916 p = (char *)(void *)sp; 917 n = sp[0]; 918 free_space = FREESPACE(sp); 919 off = OFFSET(sp); 920 921 pageno = sp[n - 1]; 922 off -= key->size; 923 sp[n - 1] = off; 924 memmove(p + off, key->data, key->size); 925 off -= val->size; 926 sp[n] = off; 927 memmove(p + off, val->data, val->size); 928 sp[0] = n + 2; 929 sp[n + 1] = pageno; 930 sp[n + 2] = OVFLPAGE; 931 FREESPACE(sp) = free_space - PAIRSIZE(key, val); 932 OFFSET(sp) = off; 933 } 934 935 static u_int32_t * 936 fetch_bitmap(hashp, ndx) 937 HTAB *hashp; 938 int ndx; 939 { 940 if (ndx >= hashp->nmaps) 941 return (NULL); 942 if ((hashp->mapp[ndx] = (u_int32_t *)malloc((size_t)hashp->BSIZE)) == NULL) 943 return (NULL); 944 if (__get_page(hashp, 945 (char *)(void *)hashp->mapp[ndx], (u_int32_t)hashp->BITMAPS[ndx], 0, 1, 1)) { 946 free(hashp->mapp[ndx]); 947 return (NULL); 948 } 949 return (hashp->mapp[ndx]); 950 } 951 952 #ifdef DEBUG4 953 int 954 print_chain(addr) 955 int addr; 956 { 957 BUFHEAD *bufp; 958 short *bp, oaddr; 959 960 (void)fprintf(stderr, "%d ", addr); 961 bufp = __get_buf(hashp, addr, NULL, 0); 962 bp = (short *)bufp->page; 963 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) || 964 ((bp[0] > 2) && bp[2] < REAL_KEY))) { 965 oaddr = bp[bp[0] - 1]; 966 (void)fprintf(stderr, "%d ", (int)oaddr); 967 bufp = __get_buf(hashp, (int)oaddr, bufp, 0); 968 bp = (short *)bufp->page; 969 } 970 (void)fprintf(stderr, "\n"); 971 } 972 #endif 973