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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: src/lib/libc/db/hash/hash_page.c,v 1.5 2000/01/27 23:06:08 jasone Exp $ 33 * $DragonFly: src/lib/libc/db/hash/hash_page.c,v 1.6 2005/09/19 09:20:37 asmodai Exp $ 34 * 35 * @(#)hash_page.c 8.7 (Berkeley) 8/16/94 36 */ 37 38 /* 39 * PACKAGE: hashing 40 * 41 * DESCRIPTION: 42 * Page manipulation for hashing package. 43 * 44 * ROUTINES: 45 * 46 * External 47 * __get_page 48 * __add_ovflpage 49 * Internal 50 * overflow_page 51 * open_temp 52 */ 53 54 #include "namespace.h" 55 #include <sys/types.h> 56 57 #include <errno.h> 58 #include <fcntl.h> 59 #include <signal.h> 60 #include <stdio.h> 61 #include <stdlib.h> 62 #include <string.h> 63 #include <unistd.h> 64 #ifdef DEBUG 65 #include <assert.h> 66 #endif 67 #include "un-namespace.h" 68 69 #include <db.h> 70 #include "hash.h" 71 #include "page.h" 72 #include "extern.h" 73 74 static u_int32_t *fetch_bitmap (HTAB *, int); 75 static u_int32_t first_free (u_int32_t); 76 static int open_temp (HTAB *); 77 static u_int16_t overflow_page (HTAB *); 78 static void putpair (char *, const DBT *, const DBT *); 79 static void squeeze_key (u_int16_t *, const DBT *, const DBT *); 80 static int ugly_split 81 (HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int); 82 83 #define PAGE_INIT(P) { \ 84 ((u_int16_t *)(P))[0] = 0; \ 85 ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \ 86 ((u_int16_t *)(P))[2] = hashp->BSIZE; \ 87 } 88 89 /* 90 * This is called AFTER we have verified that there is room on the page for 91 * the pair (PAIRFITS has returned true) so we go right ahead and start moving 92 * stuff on. 93 */ 94 static void 95 putpair(p, key, val) 96 char *p; 97 const DBT *key, *val; 98 { 99 u_int16_t *bp, n, off; 100 101 bp = (u_int16_t *)p; 102 103 /* Enter the key first. */ 104 n = bp[0]; 105 106 off = OFFSET(bp) - key->size; 107 memmove(p + off, key->data, key->size); 108 bp[++n] = off; 109 110 /* Now the data. */ 111 off -= val->size; 112 memmove(p + off, val->data, val->size); 113 bp[++n] = off; 114 115 /* Adjust page info. */ 116 bp[0] = n; 117 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t)); 118 bp[n + 2] = off; 119 } 120 121 /* 122 * Returns: 123 * 0 OK 124 * -1 error 125 */ 126 extern int 127 __delpair(hashp, bufp, ndx) 128 HTAB *hashp; 129 BUFHEAD *bufp; 130 int ndx; 131 { 132 u_int16_t *bp, newoff; 133 int n; 134 u_int16_t pairlen; 135 136 bp = (u_int16_t *)bufp->page; 137 n = bp[0]; 138 139 if (bp[ndx + 1] < REAL_KEY) 140 return (__big_delete(hashp, bufp)); 141 if (ndx != 1) 142 newoff = bp[ndx - 1]; 143 else 144 newoff = hashp->BSIZE; 145 pairlen = newoff - bp[ndx + 1]; 146 147 if (ndx != (n - 1)) { 148 /* Hard Case -- need to shuffle keys */ 149 int i; 150 char *src = bufp->page + (int)OFFSET(bp); 151 char *dst = src + (int)pairlen; 152 memmove(dst, src, bp[ndx + 1] - OFFSET(bp)); 153 154 /* Now adjust the pointers */ 155 for (i = ndx + 2; i <= n; i += 2) { 156 if (bp[i + 1] == OVFLPAGE) { 157 bp[i - 2] = bp[i]; 158 bp[i - 1] = bp[i + 1]; 159 } else { 160 bp[i - 2] = bp[i] + pairlen; 161 bp[i - 1] = bp[i + 1] + pairlen; 162 } 163 } 164 } 165 /* Finally adjust the page data */ 166 bp[n] = OFFSET(bp) + pairlen; 167 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t); 168 bp[0] = n - 2; 169 hashp->NKEYS--; 170 171 bufp->flags |= BUF_MOD; 172 return (0); 173 } 174 /* 175 * Returns: 176 * 0 ==> OK 177 * -1 ==> Error 178 */ 179 extern int 180 __split_page(hashp, obucket, nbucket) 181 HTAB *hashp; 182 u_int32_t obucket, nbucket; 183 { 184 BUFHEAD *new_bufp, *old_bufp; 185 u_int16_t *ino; 186 char *np; 187 DBT key, val; 188 int n, ndx, retval; 189 u_int16_t copyto, diff, off, moved; 190 char *op; 191 192 copyto = (u_int16_t)hashp->BSIZE; 193 off = (u_int16_t)hashp->BSIZE; 194 old_bufp = __get_buf(hashp, obucket, NULL, 0); 195 if (old_bufp == NULL) 196 return (-1); 197 new_bufp = __get_buf(hashp, nbucket, NULL, 0); 198 if (new_bufp == NULL) 199 return (-1); 200 201 old_bufp->flags |= (BUF_MOD | BUF_PIN); 202 new_bufp->flags |= (BUF_MOD | BUF_PIN); 203 204 ino = (u_int16_t *)(op = old_bufp->page); 205 np = new_bufp->page; 206 207 moved = 0; 208 209 for (n = 1, ndx = 1; n < ino[0]; n += 2) { 210 if (ino[n + 1] < REAL_KEY) { 211 retval = ugly_split(hashp, obucket, old_bufp, new_bufp, 212 (int)copyto, (int)moved); 213 old_bufp->flags &= ~BUF_PIN; 214 new_bufp->flags &= ~BUF_PIN; 215 return (retval); 216 217 } 218 key.data = (u_char *)op + ino[n]; 219 key.size = off - ino[n]; 220 221 if (__call_hash(hashp, key.data, key.size) == obucket) { 222 /* Don't switch page */ 223 diff = copyto - off; 224 if (diff) { 225 copyto = ino[n + 1] + diff; 226 memmove(op + copyto, op + ino[n + 1], 227 off - ino[n + 1]); 228 ino[ndx] = copyto + ino[n] - ino[n + 1]; 229 ino[ndx + 1] = copyto; 230 } else 231 copyto = ino[n + 1]; 232 ndx += 2; 233 } else { 234 /* Switch page */ 235 val.data = (u_char *)op + ino[n + 1]; 236 val.size = ino[n] - ino[n + 1]; 237 putpair(np, &key, &val); 238 moved += 2; 239 } 240 241 off = ino[n + 1]; 242 } 243 244 /* Now clean up the page */ 245 ino[0] -= moved; 246 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3); 247 OFFSET(ino) = copyto; 248 249 #ifdef DEBUG3 250 (void)fprintf(stderr, "split %d/%d\n", 251 ((u_int16_t *)np)[0] / 2, 252 ((u_int16_t *)op)[0] / 2); 253 #endif 254 /* unpin both pages */ 255 old_bufp->flags &= ~BUF_PIN; 256 new_bufp->flags &= ~BUF_PIN; 257 return (0); 258 } 259 260 /* 261 * Called when we encounter an overflow or big key/data page during split 262 * handling. This is special cased since we have to begin checking whether 263 * the key/data pairs fit on their respective pages and because we may need 264 * overflow pages for both the old and new pages. 265 * 266 * The first page might be a page with regular key/data pairs in which case 267 * we have a regular overflow condition and just need to go on to the next 268 * page or it might be a big key/data pair in which case we need to fix the 269 * big key/data pair. 270 * 271 * Returns: 272 * 0 ==> success 273 * -1 ==> failure 274 */ 275 static int 276 ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved) 277 HTAB *hashp; 278 u_int32_t obucket; /* Same as __split_page. */ 279 BUFHEAD *old_bufp, *new_bufp; 280 int copyto; /* First byte on page which contains key/data values. */ 281 int moved; /* Number of pairs moved to new page. */ 282 { 283 BUFHEAD *bufp; /* Buffer header for ino */ 284 u_int16_t *ino; /* Page keys come off of */ 285 u_int16_t *np; /* New page */ 286 u_int16_t *op; /* Page keys go on to if they aren't moving */ 287 288 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */ 289 DBT key, val; 290 SPLIT_RETURN ret; 291 u_int16_t n, off, ov_addr, scopyto; 292 char *cino; /* Character value of ino */ 293 294 bufp = old_bufp; 295 ino = (u_int16_t *)old_bufp->page; 296 np = (u_int16_t *)new_bufp->page; 297 op = (u_int16_t *)old_bufp->page; 298 last_bfp = NULL; 299 scopyto = (u_int16_t)copyto; /* ANSI */ 300 301 n = ino[0] - 1; 302 while (n < ino[0]) { 303 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) { 304 if (__big_split(hashp, old_bufp, 305 new_bufp, bufp, bufp->addr, obucket, &ret)) 306 return (-1); 307 old_bufp = ret.oldp; 308 if (!old_bufp) 309 return (-1); 310 op = (u_int16_t *)old_bufp->page; 311 new_bufp = ret.newp; 312 if (!new_bufp) 313 return (-1); 314 np = (u_int16_t *)new_bufp->page; 315 bufp = ret.nextp; 316 if (!bufp) 317 return (0); 318 cino = (char *)bufp->page; 319 ino = (u_int16_t *)cino; 320 last_bfp = ret.nextp; 321 } else if (ino[n + 1] == OVFLPAGE) { 322 ov_addr = ino[n]; 323 /* 324 * Fix up the old page -- the extra 2 are the fields 325 * which contained the overflow information. 326 */ 327 ino[0] -= (moved + 2); 328 FREESPACE(ino) = 329 scopyto - sizeof(u_int16_t) * (ino[0] + 3); 330 OFFSET(ino) = scopyto; 331 332 bufp = __get_buf(hashp, ov_addr, bufp, 0); 333 if (!bufp) 334 return (-1); 335 336 ino = (u_int16_t *)bufp->page; 337 n = 1; 338 scopyto = hashp->BSIZE; 339 moved = 0; 340 341 if (last_bfp) 342 __free_ovflpage(hashp, last_bfp); 343 last_bfp = bufp; 344 } 345 /* Move regular sized pairs of there are any */ 346 off = hashp->BSIZE; 347 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) { 348 cino = (char *)ino; 349 key.data = (u_char *)cino + ino[n]; 350 key.size = off - ino[n]; 351 val.data = (u_char *)cino + ino[n + 1]; 352 val.size = ino[n] - ino[n + 1]; 353 off = ino[n + 1]; 354 355 if (__call_hash(hashp, key.data, key.size) == obucket) { 356 /* Keep on old page */ 357 if (PAIRFITS(op, (&key), (&val))) 358 putpair((char *)op, &key, &val); 359 else { 360 old_bufp = 361 __add_ovflpage(hashp, old_bufp); 362 if (!old_bufp) 363 return (-1); 364 op = (u_int16_t *)old_bufp->page; 365 putpair((char *)op, &key, &val); 366 } 367 old_bufp->flags |= BUF_MOD; 368 } else { 369 /* Move to new page */ 370 if (PAIRFITS(np, (&key), (&val))) 371 putpair((char *)np, &key, &val); 372 else { 373 new_bufp = 374 __add_ovflpage(hashp, new_bufp); 375 if (!new_bufp) 376 return (-1); 377 np = (u_int16_t *)new_bufp->page; 378 putpair((char *)np, &key, &val); 379 } 380 new_bufp->flags |= BUF_MOD; 381 } 382 } 383 } 384 if (last_bfp) 385 __free_ovflpage(hashp, last_bfp); 386 return (0); 387 } 388 389 /* 390 * Add the given pair to the page 391 * 392 * Returns: 393 * 0 ==> OK 394 * 1 ==> failure 395 */ 396 extern int 397 __addel(hashp, bufp, key, val) 398 HTAB *hashp; 399 BUFHEAD *bufp; 400 const DBT *key, *val; 401 { 402 u_int16_t *bp, *sop; 403 int do_expand; 404 405 bp = (u_int16_t *)bufp->page; 406 do_expand = 0; 407 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY)) 408 /* Exception case */ 409 if (bp[2] == FULL_KEY_DATA && bp[0] == 2) 410 /* This is the last page of a big key/data pair 411 and we need to add another page */ 412 break; 413 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) { 414 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 415 if (!bufp) 416 return (-1); 417 bp = (u_int16_t *)bufp->page; 418 } else 419 /* Try to squeeze key on this page */ 420 if (FREESPACE(bp) > PAIRSIZE(key, val)) { 421 squeeze_key(bp, key, val); 422 return (0); 423 } else { 424 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 425 if (!bufp) 426 return (-1); 427 bp = (u_int16_t *)bufp->page; 428 } 429 430 if (PAIRFITS(bp, key, val)) 431 putpair(bufp->page, key, val); 432 else { 433 do_expand = 1; 434 bufp = __add_ovflpage(hashp, bufp); 435 if (!bufp) 436 return (-1); 437 sop = (u_int16_t *)bufp->page; 438 439 if (PAIRFITS(sop, key, val)) 440 putpair((char *)sop, key, val); 441 else 442 if (__big_insert(hashp, bufp, key, val)) 443 return (-1); 444 } 445 bufp->flags |= BUF_MOD; 446 /* 447 * If the average number of keys per bucket exceeds the fill factor, 448 * expand the table. 449 */ 450 hashp->NKEYS++; 451 if (do_expand || 452 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR)) 453 return (__expand_table(hashp)); 454 return (0); 455 } 456 457 /* 458 * 459 * Returns: 460 * pointer on success 461 * NULL on error 462 */ 463 extern BUFHEAD * 464 __add_ovflpage(hashp, bufp) 465 HTAB *hashp; 466 BUFHEAD *bufp; 467 { 468 u_int16_t *sp; 469 u_int16_t ndx, ovfl_num; 470 #ifdef DEBUG1 471 int tmp1, tmp2; 472 #endif 473 sp = (u_int16_t *)bufp->page; 474 475 /* Check if we are dynamically determining the fill factor */ 476 if (hashp->FFACTOR == DEF_FFACTOR) { 477 hashp->FFACTOR = sp[0] >> 1; 478 if (hashp->FFACTOR < MIN_FFACTOR) 479 hashp->FFACTOR = MIN_FFACTOR; 480 } 481 bufp->flags |= BUF_MOD; 482 ovfl_num = overflow_page(hashp); 483 #ifdef DEBUG1 484 tmp1 = bufp->addr; 485 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0; 486 #endif 487 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1))) 488 return (NULL); 489 bufp->ovfl->flags |= BUF_MOD; 490 #ifdef DEBUG1 491 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n", 492 tmp1, tmp2, bufp->ovfl->addr); 493 #endif 494 ndx = sp[0]; 495 /* 496 * Since a pair is allocated on a page only if there's room to add 497 * an overflow page, we know that the OVFL information will fit on 498 * the page. 499 */ 500 sp[ndx + 4] = OFFSET(sp); 501 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE; 502 sp[ndx + 1] = ovfl_num; 503 sp[ndx + 2] = OVFLPAGE; 504 sp[0] = ndx + 2; 505 #ifdef HASH_STATISTICS 506 hash_overflows++; 507 #endif 508 return (bufp->ovfl); 509 } 510 511 /* 512 * Returns: 513 * 0 indicates SUCCESS 514 * -1 indicates FAILURE 515 */ 516 extern int 517 __get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap) 518 HTAB *hashp; 519 char *p; 520 u_int32_t bucket; 521 int is_bucket, is_disk, is_bitmap; 522 { 523 int fd, page, size; 524 int rsize; 525 u_int16_t *bp; 526 527 fd = hashp->fp; 528 size = hashp->BSIZE; 529 530 if ((fd == -1) || !is_disk) { 531 PAGE_INIT(p); 532 return (0); 533 } 534 if (is_bucket) 535 page = BUCKET_TO_PAGE(bucket); 536 else 537 page = OADDR_TO_PAGE(bucket); 538 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || 539 ((rsize = _read(fd, p, size)) == -1)) 540 return (-1); 541 bp = (u_int16_t *)p; 542 if (!rsize) 543 bp[0] = 0; /* We hit the EOF, so initialize a new page */ 544 else 545 if (rsize != size) { 546 errno = EFTYPE; 547 return (-1); 548 } 549 if (!is_bitmap && !bp[0]) { 550 PAGE_INIT(p); 551 } else 552 if (hashp->LORDER != BYTE_ORDER) { 553 int i, max; 554 555 if (is_bitmap) { 556 max = hashp->BSIZE >> 2; /* divide by 4 */ 557 for (i = 0; i < max; i++) 558 M_32_SWAP(((int *)p)[i]); 559 } else { 560 M_16_SWAP(bp[0]); 561 max = bp[0] + 2; 562 for (i = 1; i <= max; i++) 563 M_16_SWAP(bp[i]); 564 } 565 } 566 return (0); 567 } 568 569 /* 570 * Write page p to disk 571 * 572 * Returns: 573 * 0 ==> OK 574 * -1 ==>failure 575 */ 576 extern int 577 __put_page(hashp, p, bucket, is_bucket, is_bitmap) 578 HTAB *hashp; 579 char *p; 580 u_int32_t bucket; 581 int is_bucket, is_bitmap; 582 { 583 int fd, page, size; 584 int wsize; 585 586 size = hashp->BSIZE; 587 if ((hashp->fp == -1) && open_temp(hashp)) 588 return (-1); 589 fd = hashp->fp; 590 591 if (hashp->LORDER != BYTE_ORDER) { 592 int i; 593 int max; 594 595 if (is_bitmap) { 596 max = hashp->BSIZE >> 2; /* divide by 4 */ 597 for (i = 0; i < max; i++) 598 M_32_SWAP(((int *)p)[i]); 599 } else { 600 max = ((u_int16_t *)p)[0] + 2; 601 for (i = 0; i <= max; i++) 602 M_16_SWAP(((u_int16_t *)p)[i]); 603 } 604 } 605 if (is_bucket) 606 page = BUCKET_TO_PAGE(bucket); 607 else 608 page = OADDR_TO_PAGE(bucket); 609 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || 610 ((wsize = _write(fd, p, size)) == -1)) 611 /* Errno is set */ 612 return (-1); 613 if (wsize != size) { 614 errno = EFTYPE; 615 return (-1); 616 } 617 return (0); 618 } 619 620 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1) 621 /* 622 * Initialize a new bitmap page. Bitmap pages are left in memory 623 * once they are read in. 624 */ 625 extern int 626 __ibitmap(hashp, pnum, nbits, ndx) 627 HTAB *hashp; 628 int pnum, nbits, ndx; 629 { 630 u_int32_t *ip; 631 int clearbytes, clearints; 632 633 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 634 return (1); 635 hashp->nmaps++; 636 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1; 637 clearbytes = clearints << INT_TO_BYTE; 638 (void)memset((char *)ip, 0, clearbytes); 639 (void)memset(((char *)ip) + clearbytes, 0xFF, 640 hashp->BSIZE - clearbytes); 641 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK); 642 SETBIT(ip, 0); 643 hashp->BITMAPS[ndx] = (u_int16_t)pnum; 644 hashp->mapp[ndx] = ip; 645 return (0); 646 } 647 648 static u_int32_t 649 first_free(map) 650 u_int32_t map; 651 { 652 u_int32_t i, mask; 653 654 mask = 0x1; 655 for (i = 0; i < BITS_PER_MAP; i++) { 656 if (!(mask & map)) 657 return (i); 658 mask = mask << 1; 659 } 660 return (i); 661 } 662 663 static u_int16_t 664 overflow_page(hashp) 665 HTAB *hashp; 666 { 667 u_int32_t *freep; 668 int max_free, offset, splitnum; 669 u_int16_t addr; 670 int bit, first_page, free_bit, free_page, i, in_use_bits, j; 671 #ifdef DEBUG2 672 int tmp1, tmp2; 673 #endif 674 splitnum = hashp->OVFL_POINT; 675 max_free = hashp->SPARES[splitnum]; 676 677 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT); 678 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1); 679 680 /* Look through all the free maps to find the first free block */ 681 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT); 682 for ( i = first_page; i <= free_page; i++ ) { 683 if (!(freep = (u_int32_t *)hashp->mapp[i]) && 684 !(freep = fetch_bitmap(hashp, i))) 685 return (0); 686 if (i == free_page) 687 in_use_bits = free_bit; 688 else 689 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1; 690 691 if (i == first_page) { 692 bit = hashp->LAST_FREED & 693 ((hashp->BSIZE << BYTE_SHIFT) - 1); 694 j = bit / BITS_PER_MAP; 695 bit = bit & ~(BITS_PER_MAP - 1); 696 } else { 697 bit = 0; 698 j = 0; 699 } 700 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP) 701 if (freep[j] != ALL_SET) 702 goto found; 703 } 704 705 /* No Free Page Found */ 706 hashp->LAST_FREED = hashp->SPARES[splitnum]; 707 hashp->SPARES[splitnum]++; 708 offset = hashp->SPARES[splitnum] - 709 (splitnum ? hashp->SPARES[splitnum - 1] : 0); 710 711 #define OVMSG "HASH: Out of overflow pages. Increase page size\n" 712 if (offset > SPLITMASK) { 713 if (++splitnum >= NCACHED) { 714 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 715 return (0); 716 } 717 hashp->OVFL_POINT = splitnum; 718 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 719 hashp->SPARES[splitnum-1]--; 720 offset = 1; 721 } 722 723 /* Check if we need to allocate a new bitmap page */ 724 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) { 725 free_page++; 726 if (free_page >= NCACHED) { 727 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 728 return (0); 729 } 730 /* 731 * This is tricky. The 1 indicates that you want the new page 732 * allocated with 1 clear bit. Actually, you are going to 733 * allocate 2 pages from this map. The first is going to be 734 * the map page, the second is the overflow page we were 735 * looking for. The init_bitmap routine automatically, sets 736 * the first bit of itself to indicate that the bitmap itself 737 * is in use. We would explicitly set the second bit, but 738 * don't have to if we tell init_bitmap not to leave it clear 739 * in the first place. 740 */ 741 if (__ibitmap(hashp, 742 (int)OADDR_OF(splitnum, offset), 1, free_page)) 743 return (0); 744 hashp->SPARES[splitnum]++; 745 #ifdef DEBUG2 746 free_bit = 2; 747 #endif 748 offset++; 749 if (offset > SPLITMASK) { 750 if (++splitnum >= NCACHED) { 751 (void)_write(STDERR_FILENO, OVMSG, 752 sizeof(OVMSG) - 1); 753 return (0); 754 } 755 hashp->OVFL_POINT = splitnum; 756 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 757 hashp->SPARES[splitnum-1]--; 758 offset = 0; 759 } 760 } else { 761 /* 762 * Free_bit addresses the last used bit. Bump it to address 763 * the first available bit. 764 */ 765 free_bit++; 766 SETBIT(freep, free_bit); 767 } 768 769 /* Calculate address of the new overflow page */ 770 addr = OADDR_OF(splitnum, offset); 771 #ifdef DEBUG2 772 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 773 addr, free_bit, free_page); 774 #endif 775 return (addr); 776 777 found: 778 bit = bit + first_free(freep[j]); 779 SETBIT(freep, bit); 780 #ifdef DEBUG2 781 tmp1 = bit; 782 tmp2 = i; 783 #endif 784 /* 785 * Bits are addressed starting with 0, but overflow pages are addressed 786 * beginning at 1. Bit is a bit addressnumber, so we need to increment 787 * it to convert it to a page number. 788 */ 789 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT)); 790 if (bit >= hashp->LAST_FREED) 791 hashp->LAST_FREED = bit - 1; 792 793 /* Calculate the split number for this page */ 794 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++); 795 offset = (i ? bit - hashp->SPARES[i - 1] : bit); 796 if (offset >= SPLITMASK) 797 return (0); /* Out of overflow pages */ 798 addr = OADDR_OF(i, offset); 799 #ifdef DEBUG2 800 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 801 addr, tmp1, tmp2); 802 #endif 803 804 /* Allocate and return the overflow page */ 805 return (addr); 806 } 807 808 /* 809 * Mark this overflow page as free. 810 */ 811 extern void 812 __free_ovflpage(hashp, obufp) 813 HTAB *hashp; 814 BUFHEAD *obufp; 815 { 816 u_int16_t addr; 817 u_int32_t *freep; 818 int bit_address, free_page, free_bit; 819 u_int16_t ndx; 820 821 addr = obufp->addr; 822 #ifdef DEBUG1 823 (void)fprintf(stderr, "Freeing %d\n", addr); 824 #endif 825 ndx = (((u_int16_t)addr) >> SPLITSHIFT); 826 bit_address = 827 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1; 828 if (bit_address < hashp->LAST_FREED) 829 hashp->LAST_FREED = bit_address; 830 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT)); 831 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1); 832 833 if (!(freep = hashp->mapp[free_page])) 834 freep = fetch_bitmap(hashp, free_page); 835 #ifdef DEBUG 836 /* 837 * This had better never happen. It means we tried to read a bitmap 838 * that has already had overflow pages allocated off it, and we 839 * failed to read it from the file. 840 */ 841 if (!freep) 842 assert(0); 843 #endif 844 CLRBIT(freep, free_bit); 845 #ifdef DEBUG2 846 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n", 847 obufp->addr, free_bit, free_page); 848 #endif 849 __reclaim_buf(hashp, obufp); 850 } 851 852 /* 853 * Returns: 854 * 0 success 855 * -1 failure 856 */ 857 static int 858 open_temp(hashp) 859 HTAB *hashp; 860 { 861 sigset_t set, oset; 862 static char namestr[] = "_hashXXXXXX"; 863 864 /* Block signals; make sure file goes away at process exit. */ 865 (void)sigfillset(&set); 866 (void)_sigprocmask(SIG_BLOCK, &set, &oset); 867 if ((hashp->fp = mkstemp(namestr)) != -1) { 868 (void)unlink(namestr); 869 (void)_fcntl(hashp->fp, F_SETFD, 1); 870 } 871 (void)_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL); 872 return (hashp->fp != -1 ? 0 : -1); 873 } 874 875 /* 876 * We have to know that the key will fit, but the last entry on the page is 877 * an overflow pair, so we need to shift things. 878 */ 879 static void 880 squeeze_key(sp, key, val) 881 u_int16_t *sp; 882 const DBT *key, *val; 883 { 884 char *p; 885 u_int16_t free_space, n, off, pageno; 886 887 p = (char *)sp; 888 n = sp[0]; 889 free_space = FREESPACE(sp); 890 off = OFFSET(sp); 891 892 pageno = sp[n - 1]; 893 off -= key->size; 894 sp[n - 1] = off; 895 memmove(p + off, key->data, key->size); 896 off -= val->size; 897 sp[n] = off; 898 memmove(p + off, val->data, val->size); 899 sp[0] = n + 2; 900 sp[n + 1] = pageno; 901 sp[n + 2] = OVFLPAGE; 902 FREESPACE(sp) = free_space - PAIRSIZE(key, val); 903 OFFSET(sp) = off; 904 } 905 906 static u_int32_t * 907 fetch_bitmap(hashp, ndx) 908 HTAB *hashp; 909 int ndx; 910 { 911 if (ndx >= hashp->nmaps) 912 return (NULL); 913 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 914 return (NULL); 915 if (__get_page(hashp, 916 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) { 917 free(hashp->mapp[ndx]); 918 return (NULL); 919 } 920 return (hashp->mapp[ndx]); 921 } 922 923 #ifdef DEBUG4 924 int 925 print_chain(addr) 926 int addr; 927 { 928 BUFHEAD *bufp; 929 short *bp, oaddr; 930 931 (void)fprintf(stderr, "%d ", addr); 932 bufp = __get_buf(hashp, addr, NULL, 0); 933 bp = (short *)bufp->page; 934 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) || 935 ((bp[0] > 2) && bp[2] < REAL_KEY))) { 936 oaddr = bp[bp[0] - 1]; 937 (void)fprintf(stderr, "%d ", (int)oaddr); 938 bufp = __get_buf(hashp, (int)oaddr, bufp, 0); 939 bp = (short *)bufp->page; 940 } 941 (void)fprintf(stderr, "\n"); 942 } 943 #endif 944