1 /*- 2 * Copyright (c) 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Margo Seltzer. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * $FreeBSD: src/lib/libc/db/hash/hash_page.c,v 1.5 2000/01/27 23:06:08 jasone Exp $ 37 * $DragonFly: src/lib/libc/db/hash/hash_page.c,v 1.5 2005/01/31 22:29:09 dillon Exp $ 38 * 39 * @(#)hash_page.c 8.7 (Berkeley) 8/16/94 40 */ 41 42 /* 43 * PACKAGE: hashing 44 * 45 * DESCRIPTION: 46 * Page manipulation for hashing package. 47 * 48 * ROUTINES: 49 * 50 * External 51 * __get_page 52 * __add_ovflpage 53 * Internal 54 * overflow_page 55 * open_temp 56 */ 57 58 #include "namespace.h" 59 #include <sys/types.h> 60 61 #include <errno.h> 62 #include <fcntl.h> 63 #include <signal.h> 64 #include <stdio.h> 65 #include <stdlib.h> 66 #include <string.h> 67 #include <unistd.h> 68 #ifdef DEBUG 69 #include <assert.h> 70 #endif 71 #include "un-namespace.h" 72 73 #include <db.h> 74 #include "hash.h" 75 #include "page.h" 76 #include "extern.h" 77 78 static u_int32_t *fetch_bitmap (HTAB *, int); 79 static u_int32_t first_free (u_int32_t); 80 static int open_temp (HTAB *); 81 static u_int16_t overflow_page (HTAB *); 82 static void putpair (char *, const DBT *, const DBT *); 83 static void squeeze_key (u_int16_t *, const DBT *, const DBT *); 84 static int ugly_split 85 (HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int); 86 87 #define PAGE_INIT(P) { \ 88 ((u_int16_t *)(P))[0] = 0; \ 89 ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \ 90 ((u_int16_t *)(P))[2] = hashp->BSIZE; \ 91 } 92 93 /* 94 * This is called AFTER we have verified that there is room on the page for 95 * the pair (PAIRFITS has returned true) so we go right ahead and start moving 96 * stuff on. 97 */ 98 static void 99 putpair(p, key, val) 100 char *p; 101 const DBT *key, *val; 102 { 103 u_int16_t *bp, n, off; 104 105 bp = (u_int16_t *)p; 106 107 /* Enter the key first. */ 108 n = bp[0]; 109 110 off = OFFSET(bp) - key->size; 111 memmove(p + off, key->data, key->size); 112 bp[++n] = off; 113 114 /* Now the data. */ 115 off -= val->size; 116 memmove(p + off, val->data, val->size); 117 bp[++n] = off; 118 119 /* Adjust page info. */ 120 bp[0] = n; 121 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t)); 122 bp[n + 2] = off; 123 } 124 125 /* 126 * Returns: 127 * 0 OK 128 * -1 error 129 */ 130 extern int 131 __delpair(hashp, bufp, ndx) 132 HTAB *hashp; 133 BUFHEAD *bufp; 134 int ndx; 135 { 136 u_int16_t *bp, newoff; 137 int n; 138 u_int16_t pairlen; 139 140 bp = (u_int16_t *)bufp->page; 141 n = bp[0]; 142 143 if (bp[ndx + 1] < REAL_KEY) 144 return (__big_delete(hashp, bufp)); 145 if (ndx != 1) 146 newoff = bp[ndx - 1]; 147 else 148 newoff = hashp->BSIZE; 149 pairlen = newoff - bp[ndx + 1]; 150 151 if (ndx != (n - 1)) { 152 /* Hard Case -- need to shuffle keys */ 153 int i; 154 char *src = bufp->page + (int)OFFSET(bp); 155 char *dst = src + (int)pairlen; 156 memmove(dst, src, bp[ndx + 1] - OFFSET(bp)); 157 158 /* Now adjust the pointers */ 159 for (i = ndx + 2; i <= n; i += 2) { 160 if (bp[i + 1] == OVFLPAGE) { 161 bp[i - 2] = bp[i]; 162 bp[i - 1] = bp[i + 1]; 163 } else { 164 bp[i - 2] = bp[i] + pairlen; 165 bp[i - 1] = bp[i + 1] + pairlen; 166 } 167 } 168 } 169 /* Finally adjust the page data */ 170 bp[n] = OFFSET(bp) + pairlen; 171 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t); 172 bp[0] = n - 2; 173 hashp->NKEYS--; 174 175 bufp->flags |= BUF_MOD; 176 return (0); 177 } 178 /* 179 * Returns: 180 * 0 ==> OK 181 * -1 ==> Error 182 */ 183 extern int 184 __split_page(hashp, obucket, nbucket) 185 HTAB *hashp; 186 u_int32_t obucket, nbucket; 187 { 188 BUFHEAD *new_bufp, *old_bufp; 189 u_int16_t *ino; 190 char *np; 191 DBT key, val; 192 int n, ndx, retval; 193 u_int16_t copyto, diff, off, moved; 194 char *op; 195 196 copyto = (u_int16_t)hashp->BSIZE; 197 off = (u_int16_t)hashp->BSIZE; 198 old_bufp = __get_buf(hashp, obucket, NULL, 0); 199 if (old_bufp == NULL) 200 return (-1); 201 new_bufp = __get_buf(hashp, nbucket, NULL, 0); 202 if (new_bufp == NULL) 203 return (-1); 204 205 old_bufp->flags |= (BUF_MOD | BUF_PIN); 206 new_bufp->flags |= (BUF_MOD | BUF_PIN); 207 208 ino = (u_int16_t *)(op = old_bufp->page); 209 np = new_bufp->page; 210 211 moved = 0; 212 213 for (n = 1, ndx = 1; n < ino[0]; n += 2) { 214 if (ino[n + 1] < REAL_KEY) { 215 retval = ugly_split(hashp, obucket, old_bufp, new_bufp, 216 (int)copyto, (int)moved); 217 old_bufp->flags &= ~BUF_PIN; 218 new_bufp->flags &= ~BUF_PIN; 219 return (retval); 220 221 } 222 key.data = (u_char *)op + ino[n]; 223 key.size = off - ino[n]; 224 225 if (__call_hash(hashp, key.data, key.size) == obucket) { 226 /* Don't switch page */ 227 diff = copyto - off; 228 if (diff) { 229 copyto = ino[n + 1] + diff; 230 memmove(op + copyto, op + ino[n + 1], 231 off - ino[n + 1]); 232 ino[ndx] = copyto + ino[n] - ino[n + 1]; 233 ino[ndx + 1] = copyto; 234 } else 235 copyto = ino[n + 1]; 236 ndx += 2; 237 } else { 238 /* Switch page */ 239 val.data = (u_char *)op + ino[n + 1]; 240 val.size = ino[n] - ino[n + 1]; 241 putpair(np, &key, &val); 242 moved += 2; 243 } 244 245 off = ino[n + 1]; 246 } 247 248 /* Now clean up the page */ 249 ino[0] -= moved; 250 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3); 251 OFFSET(ino) = copyto; 252 253 #ifdef DEBUG3 254 (void)fprintf(stderr, "split %d/%d\n", 255 ((u_int16_t *)np)[0] / 2, 256 ((u_int16_t *)op)[0] / 2); 257 #endif 258 /* unpin both pages */ 259 old_bufp->flags &= ~BUF_PIN; 260 new_bufp->flags &= ~BUF_PIN; 261 return (0); 262 } 263 264 /* 265 * Called when we encounter an overflow or big key/data page during split 266 * handling. This is special cased since we have to begin checking whether 267 * the key/data pairs fit on their respective pages and because we may need 268 * overflow pages for both the old and new pages. 269 * 270 * The first page might be a page with regular key/data pairs in which case 271 * we have a regular overflow condition and just need to go on to the next 272 * page or it might be a big key/data pair in which case we need to fix the 273 * big key/data pair. 274 * 275 * Returns: 276 * 0 ==> success 277 * -1 ==> failure 278 */ 279 static int 280 ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved) 281 HTAB *hashp; 282 u_int32_t obucket; /* Same as __split_page. */ 283 BUFHEAD *old_bufp, *new_bufp; 284 int copyto; /* First byte on page which contains key/data values. */ 285 int moved; /* Number of pairs moved to new page. */ 286 { 287 BUFHEAD *bufp; /* Buffer header for ino */ 288 u_int16_t *ino; /* Page keys come off of */ 289 u_int16_t *np; /* New page */ 290 u_int16_t *op; /* Page keys go on to if they aren't moving */ 291 292 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */ 293 DBT key, val; 294 SPLIT_RETURN ret; 295 u_int16_t n, off, ov_addr, scopyto; 296 char *cino; /* Character value of ino */ 297 298 bufp = old_bufp; 299 ino = (u_int16_t *)old_bufp->page; 300 np = (u_int16_t *)new_bufp->page; 301 op = (u_int16_t *)old_bufp->page; 302 last_bfp = NULL; 303 scopyto = (u_int16_t)copyto; /* ANSI */ 304 305 n = ino[0] - 1; 306 while (n < ino[0]) { 307 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) { 308 if (__big_split(hashp, old_bufp, 309 new_bufp, bufp, bufp->addr, obucket, &ret)) 310 return (-1); 311 old_bufp = ret.oldp; 312 if (!old_bufp) 313 return (-1); 314 op = (u_int16_t *)old_bufp->page; 315 new_bufp = ret.newp; 316 if (!new_bufp) 317 return (-1); 318 np = (u_int16_t *)new_bufp->page; 319 bufp = ret.nextp; 320 if (!bufp) 321 return (0); 322 cino = (char *)bufp->page; 323 ino = (u_int16_t *)cino; 324 last_bfp = ret.nextp; 325 } else if (ino[n + 1] == OVFLPAGE) { 326 ov_addr = ino[n]; 327 /* 328 * Fix up the old page -- the extra 2 are the fields 329 * which contained the overflow information. 330 */ 331 ino[0] -= (moved + 2); 332 FREESPACE(ino) = 333 scopyto - sizeof(u_int16_t) * (ino[0] + 3); 334 OFFSET(ino) = scopyto; 335 336 bufp = __get_buf(hashp, ov_addr, bufp, 0); 337 if (!bufp) 338 return (-1); 339 340 ino = (u_int16_t *)bufp->page; 341 n = 1; 342 scopyto = hashp->BSIZE; 343 moved = 0; 344 345 if (last_bfp) 346 __free_ovflpage(hashp, last_bfp); 347 last_bfp = bufp; 348 } 349 /* Move regular sized pairs of there are any */ 350 off = hashp->BSIZE; 351 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) { 352 cino = (char *)ino; 353 key.data = (u_char *)cino + ino[n]; 354 key.size = off - ino[n]; 355 val.data = (u_char *)cino + ino[n + 1]; 356 val.size = ino[n] - ino[n + 1]; 357 off = ino[n + 1]; 358 359 if (__call_hash(hashp, key.data, key.size) == obucket) { 360 /* Keep on old page */ 361 if (PAIRFITS(op, (&key), (&val))) 362 putpair((char *)op, &key, &val); 363 else { 364 old_bufp = 365 __add_ovflpage(hashp, old_bufp); 366 if (!old_bufp) 367 return (-1); 368 op = (u_int16_t *)old_bufp->page; 369 putpair((char *)op, &key, &val); 370 } 371 old_bufp->flags |= BUF_MOD; 372 } else { 373 /* Move to new page */ 374 if (PAIRFITS(np, (&key), (&val))) 375 putpair((char *)np, &key, &val); 376 else { 377 new_bufp = 378 __add_ovflpage(hashp, new_bufp); 379 if (!new_bufp) 380 return (-1); 381 np = (u_int16_t *)new_bufp->page; 382 putpair((char *)np, &key, &val); 383 } 384 new_bufp->flags |= BUF_MOD; 385 } 386 } 387 } 388 if (last_bfp) 389 __free_ovflpage(hashp, last_bfp); 390 return (0); 391 } 392 393 /* 394 * Add the given pair to the page 395 * 396 * Returns: 397 * 0 ==> OK 398 * 1 ==> failure 399 */ 400 extern int 401 __addel(hashp, bufp, key, val) 402 HTAB *hashp; 403 BUFHEAD *bufp; 404 const DBT *key, *val; 405 { 406 u_int16_t *bp, *sop; 407 int do_expand; 408 409 bp = (u_int16_t *)bufp->page; 410 do_expand = 0; 411 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY)) 412 /* Exception case */ 413 if (bp[2] == FULL_KEY_DATA && bp[0] == 2) 414 /* This is the last page of a big key/data pair 415 and we need to add another page */ 416 break; 417 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) { 418 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 419 if (!bufp) 420 return (-1); 421 bp = (u_int16_t *)bufp->page; 422 } else 423 /* Try to squeeze key on this page */ 424 if (FREESPACE(bp) > PAIRSIZE(key, val)) { 425 squeeze_key(bp, key, val); 426 return (0); 427 } else { 428 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 429 if (!bufp) 430 return (-1); 431 bp = (u_int16_t *)bufp->page; 432 } 433 434 if (PAIRFITS(bp, key, val)) 435 putpair(bufp->page, key, val); 436 else { 437 do_expand = 1; 438 bufp = __add_ovflpage(hashp, bufp); 439 if (!bufp) 440 return (-1); 441 sop = (u_int16_t *)bufp->page; 442 443 if (PAIRFITS(sop, key, val)) 444 putpair((char *)sop, key, val); 445 else 446 if (__big_insert(hashp, bufp, key, val)) 447 return (-1); 448 } 449 bufp->flags |= BUF_MOD; 450 /* 451 * If the average number of keys per bucket exceeds the fill factor, 452 * expand the table. 453 */ 454 hashp->NKEYS++; 455 if (do_expand || 456 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR)) 457 return (__expand_table(hashp)); 458 return (0); 459 } 460 461 /* 462 * 463 * Returns: 464 * pointer on success 465 * NULL on error 466 */ 467 extern BUFHEAD * 468 __add_ovflpage(hashp, bufp) 469 HTAB *hashp; 470 BUFHEAD *bufp; 471 { 472 u_int16_t *sp; 473 u_int16_t ndx, ovfl_num; 474 #ifdef DEBUG1 475 int tmp1, tmp2; 476 #endif 477 sp = (u_int16_t *)bufp->page; 478 479 /* Check if we are dynamically determining the fill factor */ 480 if (hashp->FFACTOR == DEF_FFACTOR) { 481 hashp->FFACTOR = sp[0] >> 1; 482 if (hashp->FFACTOR < MIN_FFACTOR) 483 hashp->FFACTOR = MIN_FFACTOR; 484 } 485 bufp->flags |= BUF_MOD; 486 ovfl_num = overflow_page(hashp); 487 #ifdef DEBUG1 488 tmp1 = bufp->addr; 489 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0; 490 #endif 491 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1))) 492 return (NULL); 493 bufp->ovfl->flags |= BUF_MOD; 494 #ifdef DEBUG1 495 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n", 496 tmp1, tmp2, bufp->ovfl->addr); 497 #endif 498 ndx = sp[0]; 499 /* 500 * Since a pair is allocated on a page only if there's room to add 501 * an overflow page, we know that the OVFL information will fit on 502 * the page. 503 */ 504 sp[ndx + 4] = OFFSET(sp); 505 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE; 506 sp[ndx + 1] = ovfl_num; 507 sp[ndx + 2] = OVFLPAGE; 508 sp[0] = ndx + 2; 509 #ifdef HASH_STATISTICS 510 hash_overflows++; 511 #endif 512 return (bufp->ovfl); 513 } 514 515 /* 516 * Returns: 517 * 0 indicates SUCCESS 518 * -1 indicates FAILURE 519 */ 520 extern int 521 __get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap) 522 HTAB *hashp; 523 char *p; 524 u_int32_t bucket; 525 int is_bucket, is_disk, is_bitmap; 526 { 527 int fd, page, size; 528 int rsize; 529 u_int16_t *bp; 530 531 fd = hashp->fp; 532 size = hashp->BSIZE; 533 534 if ((fd == -1) || !is_disk) { 535 PAGE_INIT(p); 536 return (0); 537 } 538 if (is_bucket) 539 page = BUCKET_TO_PAGE(bucket); 540 else 541 page = OADDR_TO_PAGE(bucket); 542 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || 543 ((rsize = _read(fd, p, size)) == -1)) 544 return (-1); 545 bp = (u_int16_t *)p; 546 if (!rsize) 547 bp[0] = 0; /* We hit the EOF, so initialize a new page */ 548 else 549 if (rsize != size) { 550 errno = EFTYPE; 551 return (-1); 552 } 553 if (!is_bitmap && !bp[0]) { 554 PAGE_INIT(p); 555 } else 556 if (hashp->LORDER != BYTE_ORDER) { 557 int i, max; 558 559 if (is_bitmap) { 560 max = hashp->BSIZE >> 2; /* divide by 4 */ 561 for (i = 0; i < max; i++) 562 M_32_SWAP(((int *)p)[i]); 563 } else { 564 M_16_SWAP(bp[0]); 565 max = bp[0] + 2; 566 for (i = 1; i <= max; i++) 567 M_16_SWAP(bp[i]); 568 } 569 } 570 return (0); 571 } 572 573 /* 574 * Write page p to disk 575 * 576 * Returns: 577 * 0 ==> OK 578 * -1 ==>failure 579 */ 580 extern int 581 __put_page(hashp, p, bucket, is_bucket, is_bitmap) 582 HTAB *hashp; 583 char *p; 584 u_int32_t bucket; 585 int is_bucket, is_bitmap; 586 { 587 int fd, page, size; 588 int wsize; 589 590 size = hashp->BSIZE; 591 if ((hashp->fp == -1) && open_temp(hashp)) 592 return (-1); 593 fd = hashp->fp; 594 595 if (hashp->LORDER != BYTE_ORDER) { 596 int i; 597 int max; 598 599 if (is_bitmap) { 600 max = hashp->BSIZE >> 2; /* divide by 4 */ 601 for (i = 0; i < max; i++) 602 M_32_SWAP(((int *)p)[i]); 603 } else { 604 max = ((u_int16_t *)p)[0] + 2; 605 for (i = 0; i <= max; i++) 606 M_16_SWAP(((u_int16_t *)p)[i]); 607 } 608 } 609 if (is_bucket) 610 page = BUCKET_TO_PAGE(bucket); 611 else 612 page = OADDR_TO_PAGE(bucket); 613 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || 614 ((wsize = _write(fd, p, size)) == -1)) 615 /* Errno is set */ 616 return (-1); 617 if (wsize != size) { 618 errno = EFTYPE; 619 return (-1); 620 } 621 return (0); 622 } 623 624 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1) 625 /* 626 * Initialize a new bitmap page. Bitmap pages are left in memory 627 * once they are read in. 628 */ 629 extern int 630 __ibitmap(hashp, pnum, nbits, ndx) 631 HTAB *hashp; 632 int pnum, nbits, ndx; 633 { 634 u_int32_t *ip; 635 int clearbytes, clearints; 636 637 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 638 return (1); 639 hashp->nmaps++; 640 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1; 641 clearbytes = clearints << INT_TO_BYTE; 642 (void)memset((char *)ip, 0, clearbytes); 643 (void)memset(((char *)ip) + clearbytes, 0xFF, 644 hashp->BSIZE - clearbytes); 645 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK); 646 SETBIT(ip, 0); 647 hashp->BITMAPS[ndx] = (u_int16_t)pnum; 648 hashp->mapp[ndx] = ip; 649 return (0); 650 } 651 652 static u_int32_t 653 first_free(map) 654 u_int32_t map; 655 { 656 u_int32_t i, mask; 657 658 mask = 0x1; 659 for (i = 0; i < BITS_PER_MAP; i++) { 660 if (!(mask & map)) 661 return (i); 662 mask = mask << 1; 663 } 664 return (i); 665 } 666 667 static u_int16_t 668 overflow_page(hashp) 669 HTAB *hashp; 670 { 671 u_int32_t *freep; 672 int max_free, offset, splitnum; 673 u_int16_t addr; 674 int bit, first_page, free_bit, free_page, i, in_use_bits, j; 675 #ifdef DEBUG2 676 int tmp1, tmp2; 677 #endif 678 splitnum = hashp->OVFL_POINT; 679 max_free = hashp->SPARES[splitnum]; 680 681 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT); 682 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1); 683 684 /* Look through all the free maps to find the first free block */ 685 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT); 686 for ( i = first_page; i <= free_page; i++ ) { 687 if (!(freep = (u_int32_t *)hashp->mapp[i]) && 688 !(freep = fetch_bitmap(hashp, i))) 689 return (0); 690 if (i == free_page) 691 in_use_bits = free_bit; 692 else 693 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1; 694 695 if (i == first_page) { 696 bit = hashp->LAST_FREED & 697 ((hashp->BSIZE << BYTE_SHIFT) - 1); 698 j = bit / BITS_PER_MAP; 699 bit = bit & ~(BITS_PER_MAP - 1); 700 } else { 701 bit = 0; 702 j = 0; 703 } 704 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP) 705 if (freep[j] != ALL_SET) 706 goto found; 707 } 708 709 /* No Free Page Found */ 710 hashp->LAST_FREED = hashp->SPARES[splitnum]; 711 hashp->SPARES[splitnum]++; 712 offset = hashp->SPARES[splitnum] - 713 (splitnum ? hashp->SPARES[splitnum - 1] : 0); 714 715 #define OVMSG "HASH: Out of overflow pages. Increase page size\n" 716 if (offset > SPLITMASK) { 717 if (++splitnum >= NCACHED) { 718 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 719 return (0); 720 } 721 hashp->OVFL_POINT = splitnum; 722 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 723 hashp->SPARES[splitnum-1]--; 724 offset = 1; 725 } 726 727 /* Check if we need to allocate a new bitmap page */ 728 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) { 729 free_page++; 730 if (free_page >= NCACHED) { 731 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 732 return (0); 733 } 734 /* 735 * This is tricky. The 1 indicates that you want the new page 736 * allocated with 1 clear bit. Actually, you are going to 737 * allocate 2 pages from this map. The first is going to be 738 * the map page, the second is the overflow page we were 739 * looking for. The init_bitmap routine automatically, sets 740 * the first bit of itself to indicate that the bitmap itself 741 * is in use. We would explicitly set the second bit, but 742 * don't have to if we tell init_bitmap not to leave it clear 743 * in the first place. 744 */ 745 if (__ibitmap(hashp, 746 (int)OADDR_OF(splitnum, offset), 1, free_page)) 747 return (0); 748 hashp->SPARES[splitnum]++; 749 #ifdef DEBUG2 750 free_bit = 2; 751 #endif 752 offset++; 753 if (offset > SPLITMASK) { 754 if (++splitnum >= NCACHED) { 755 (void)_write(STDERR_FILENO, OVMSG, 756 sizeof(OVMSG) - 1); 757 return (0); 758 } 759 hashp->OVFL_POINT = splitnum; 760 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 761 hashp->SPARES[splitnum-1]--; 762 offset = 0; 763 } 764 } else { 765 /* 766 * Free_bit addresses the last used bit. Bump it to address 767 * the first available bit. 768 */ 769 free_bit++; 770 SETBIT(freep, free_bit); 771 } 772 773 /* Calculate address of the new overflow page */ 774 addr = OADDR_OF(splitnum, offset); 775 #ifdef DEBUG2 776 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 777 addr, free_bit, free_page); 778 #endif 779 return (addr); 780 781 found: 782 bit = bit + first_free(freep[j]); 783 SETBIT(freep, bit); 784 #ifdef DEBUG2 785 tmp1 = bit; 786 tmp2 = i; 787 #endif 788 /* 789 * Bits are addressed starting with 0, but overflow pages are addressed 790 * beginning at 1. Bit is a bit addressnumber, so we need to increment 791 * it to convert it to a page number. 792 */ 793 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT)); 794 if (bit >= hashp->LAST_FREED) 795 hashp->LAST_FREED = bit - 1; 796 797 /* Calculate the split number for this page */ 798 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++); 799 offset = (i ? bit - hashp->SPARES[i - 1] : bit); 800 if (offset >= SPLITMASK) 801 return (0); /* Out of overflow pages */ 802 addr = OADDR_OF(i, offset); 803 #ifdef DEBUG2 804 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 805 addr, tmp1, tmp2); 806 #endif 807 808 /* Allocate and return the overflow page */ 809 return (addr); 810 } 811 812 /* 813 * Mark this overflow page as free. 814 */ 815 extern void 816 __free_ovflpage(hashp, obufp) 817 HTAB *hashp; 818 BUFHEAD *obufp; 819 { 820 u_int16_t addr; 821 u_int32_t *freep; 822 int bit_address, free_page, free_bit; 823 u_int16_t ndx; 824 825 addr = obufp->addr; 826 #ifdef DEBUG1 827 (void)fprintf(stderr, "Freeing %d\n", addr); 828 #endif 829 ndx = (((u_int16_t)addr) >> SPLITSHIFT); 830 bit_address = 831 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1; 832 if (bit_address < hashp->LAST_FREED) 833 hashp->LAST_FREED = bit_address; 834 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT)); 835 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1); 836 837 if (!(freep = hashp->mapp[free_page])) 838 freep = fetch_bitmap(hashp, free_page); 839 #ifdef DEBUG 840 /* 841 * This had better never happen. It means we tried to read a bitmap 842 * that has already had overflow pages allocated off it, and we 843 * failed to read it from the file. 844 */ 845 if (!freep) 846 assert(0); 847 #endif 848 CLRBIT(freep, free_bit); 849 #ifdef DEBUG2 850 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n", 851 obufp->addr, free_bit, free_page); 852 #endif 853 __reclaim_buf(hashp, obufp); 854 } 855 856 /* 857 * Returns: 858 * 0 success 859 * -1 failure 860 */ 861 static int 862 open_temp(hashp) 863 HTAB *hashp; 864 { 865 sigset_t set, oset; 866 static char namestr[] = "_hashXXXXXX"; 867 868 /* Block signals; make sure file goes away at process exit. */ 869 (void)sigfillset(&set); 870 (void)_sigprocmask(SIG_BLOCK, &set, &oset); 871 if ((hashp->fp = mkstemp(namestr)) != -1) { 872 (void)unlink(namestr); 873 (void)_fcntl(hashp->fp, F_SETFD, 1); 874 } 875 (void)_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL); 876 return (hashp->fp != -1 ? 0 : -1); 877 } 878 879 /* 880 * We have to know that the key will fit, but the last entry on the page is 881 * an overflow pair, so we need to shift things. 882 */ 883 static void 884 squeeze_key(sp, key, val) 885 u_int16_t *sp; 886 const DBT *key, *val; 887 { 888 char *p; 889 u_int16_t free_space, n, off, pageno; 890 891 p = (char *)sp; 892 n = sp[0]; 893 free_space = FREESPACE(sp); 894 off = OFFSET(sp); 895 896 pageno = sp[n - 1]; 897 off -= key->size; 898 sp[n - 1] = off; 899 memmove(p + off, key->data, key->size); 900 off -= val->size; 901 sp[n] = off; 902 memmove(p + off, val->data, val->size); 903 sp[0] = n + 2; 904 sp[n + 1] = pageno; 905 sp[n + 2] = OVFLPAGE; 906 FREESPACE(sp) = free_space - PAIRSIZE(key, val); 907 OFFSET(sp) = off; 908 } 909 910 static u_int32_t * 911 fetch_bitmap(hashp, ndx) 912 HTAB *hashp; 913 int ndx; 914 { 915 if (ndx >= hashp->nmaps) 916 return (NULL); 917 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 918 return (NULL); 919 if (__get_page(hashp, 920 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) { 921 free(hashp->mapp[ndx]); 922 return (NULL); 923 } 924 return (hashp->mapp[ndx]); 925 } 926 927 #ifdef DEBUG4 928 int 929 print_chain(addr) 930 int addr; 931 { 932 BUFHEAD *bufp; 933 short *bp, oaddr; 934 935 (void)fprintf(stderr, "%d ", addr); 936 bufp = __get_buf(hashp, addr, NULL, 0); 937 bp = (short *)bufp->page; 938 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) || 939 ((bp[0] > 2) && bp[2] < REAL_KEY))) { 940 oaddr = bp[bp[0] - 1]; 941 (void)fprintf(stderr, "%d ", (int)oaddr); 942 bufp = __get_buf(hashp, (int)oaddr, bufp, 0); 943 bp = (short *)bufp->page; 944 } 945 (void)fprintf(stderr, "\n"); 946 } 947 #endif 948