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 1.8 2005/11/19 20:46:32 swildner 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(char *p, const DBT *key, const DBT *val) 96 { 97 u_int16_t *bp, n, off; 98 99 bp = (u_int16_t *)p; 100 101 /* Enter the key first. */ 102 n = bp[0]; 103 104 off = OFFSET(bp) - key->size; 105 memmove(p + off, key->data, key->size); 106 bp[++n] = off; 107 108 /* Now the data. */ 109 off -= val->size; 110 memmove(p + off, val->data, val->size); 111 bp[++n] = off; 112 113 /* Adjust page info. */ 114 bp[0] = n; 115 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t)); 116 bp[n + 2] = off; 117 } 118 119 /* 120 * Returns: 121 * 0 OK 122 * -1 error 123 */ 124 extern int 125 __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx) 126 { 127 u_int16_t *bp, newoff; 128 int n; 129 u_int16_t pairlen; 130 131 bp = (u_int16_t *)bufp->page; 132 n = bp[0]; 133 134 if (bp[ndx + 1] < REAL_KEY) 135 return (__big_delete(hashp, bufp)); 136 if (ndx != 1) 137 newoff = bp[ndx - 1]; 138 else 139 newoff = hashp->BSIZE; 140 pairlen = newoff - bp[ndx + 1]; 141 142 if (ndx != (n - 1)) { 143 /* Hard Case -- need to shuffle keys */ 144 int i; 145 char *src = bufp->page + (int)OFFSET(bp); 146 char *dst = src + (int)pairlen; 147 memmove(dst, src, bp[ndx + 1] - OFFSET(bp)); 148 149 /* Now adjust the pointers */ 150 for (i = ndx + 2; i <= n; i += 2) { 151 if (bp[i + 1] == OVFLPAGE) { 152 bp[i - 2] = bp[i]; 153 bp[i - 1] = bp[i + 1]; 154 } else { 155 bp[i - 2] = bp[i] + pairlen; 156 bp[i - 1] = bp[i + 1] + pairlen; 157 } 158 } 159 } 160 /* Finally adjust the page data */ 161 bp[n] = OFFSET(bp) + pairlen; 162 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t); 163 bp[0] = n - 2; 164 hashp->NKEYS--; 165 166 bufp->flags |= BUF_MOD; 167 return (0); 168 } 169 /* 170 * Returns: 171 * 0 ==> OK 172 * -1 ==> Error 173 */ 174 extern int 175 __split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket) 176 { 177 BUFHEAD *new_bufp, *old_bufp; 178 u_int16_t *ino; 179 char *np; 180 DBT key, val; 181 int n, ndx, retval; 182 u_int16_t copyto, diff, off, moved; 183 char *op; 184 185 copyto = (u_int16_t)hashp->BSIZE; 186 off = (u_int16_t)hashp->BSIZE; 187 old_bufp = __get_buf(hashp, obucket, NULL, 0); 188 if (old_bufp == NULL) 189 return (-1); 190 new_bufp = __get_buf(hashp, nbucket, NULL, 0); 191 if (new_bufp == NULL) 192 return (-1); 193 194 old_bufp->flags |= (BUF_MOD | BUF_PIN); 195 new_bufp->flags |= (BUF_MOD | BUF_PIN); 196 197 ino = (u_int16_t *)(op = old_bufp->page); 198 np = new_bufp->page; 199 200 moved = 0; 201 202 for (n = 1, ndx = 1; n < ino[0]; n += 2) { 203 if (ino[n + 1] < REAL_KEY) { 204 retval = ugly_split(hashp, obucket, old_bufp, new_bufp, 205 (int)copyto, (int)moved); 206 old_bufp->flags &= ~BUF_PIN; 207 new_bufp->flags &= ~BUF_PIN; 208 return (retval); 209 210 } 211 key.data = (u_char *)op + ino[n]; 212 key.size = off - ino[n]; 213 214 if (__call_hash(hashp, key.data, key.size) == obucket) { 215 /* Don't switch page */ 216 diff = copyto - off; 217 if (diff) { 218 copyto = ino[n + 1] + diff; 219 memmove(op + copyto, op + ino[n + 1], 220 off - ino[n + 1]); 221 ino[ndx] = copyto + ino[n] - ino[n + 1]; 222 ino[ndx + 1] = copyto; 223 } else 224 copyto = ino[n + 1]; 225 ndx += 2; 226 } else { 227 /* Switch page */ 228 val.data = (u_char *)op + ino[n + 1]; 229 val.size = ino[n] - ino[n + 1]; 230 putpair(np, &key, &val); 231 moved += 2; 232 } 233 234 off = ino[n + 1]; 235 } 236 237 /* Now clean up the page */ 238 ino[0] -= moved; 239 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3); 240 OFFSET(ino) = copyto; 241 242 #ifdef DEBUG3 243 fprintf(stderr, "split %d/%d\n", 244 ((u_int16_t *)np)[0] / 2, 245 ((u_int16_t *)op)[0] / 2); 246 #endif 247 /* unpin both pages */ 248 old_bufp->flags &= ~BUF_PIN; 249 new_bufp->flags &= ~BUF_PIN; 250 return (0); 251 } 252 253 /* 254 * Called when we encounter an overflow or big key/data page during split 255 * handling. This is special cased since we have to begin checking whether 256 * the key/data pairs fit on their respective pages and because we may need 257 * overflow pages for both the old and new pages. 258 * 259 * The first page might be a page with regular key/data pairs in which case 260 * we have a regular overflow condition and just need to go on to the next 261 * page or it might be a big key/data pair in which case we need to fix the 262 * big key/data pair. 263 * 264 * Parameters: 265 * obucket - Same as __split_page 266 * copyto - First byte on page which contains key/data values 267 * moved - Number of pairs moved to new page 268 * Returns: 269 * 0 ==> success 270 * -1 ==> failure 271 */ 272 static int 273 ugly_split(HTAB *hashp, u_int32_t obucket, BUFHEAD *old_bufp, 274 BUFHEAD *new_bufp, int copyto, int moved) 275 { 276 BUFHEAD *bufp; /* Buffer header for ino */ 277 u_int16_t *ino; /* Page keys come off of */ 278 u_int16_t *np; /* New page */ 279 u_int16_t *op; /* Page keys go on to if they aren't moving */ 280 281 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */ 282 DBT key, val; 283 SPLIT_RETURN ret; 284 u_int16_t n, off, ov_addr, scopyto; 285 char *cino; /* Character value of ino */ 286 287 bufp = old_bufp; 288 ino = (u_int16_t *)old_bufp->page; 289 np = (u_int16_t *)new_bufp->page; 290 op = (u_int16_t *)old_bufp->page; 291 last_bfp = NULL; 292 scopyto = (u_int16_t)copyto; /* ANSI */ 293 294 n = ino[0] - 1; 295 while (n < ino[0]) { 296 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) { 297 if (__big_split(hashp, old_bufp, 298 new_bufp, bufp, bufp->addr, obucket, &ret)) 299 return (-1); 300 old_bufp = ret.oldp; 301 if (!old_bufp) 302 return (-1); 303 op = (u_int16_t *)old_bufp->page; 304 new_bufp = ret.newp; 305 if (!new_bufp) 306 return (-1); 307 np = (u_int16_t *)new_bufp->page; 308 bufp = ret.nextp; 309 if (!bufp) 310 return (0); 311 cino = (char *)bufp->page; 312 ino = (u_int16_t *)cino; 313 last_bfp = ret.nextp; 314 } else if (ino[n + 1] == OVFLPAGE) { 315 ov_addr = ino[n]; 316 /* 317 * Fix up the old page -- the extra 2 are the fields 318 * which contained the overflow information. 319 */ 320 ino[0] -= (moved + 2); 321 FREESPACE(ino) = 322 scopyto - sizeof(u_int16_t) * (ino[0] + 3); 323 OFFSET(ino) = scopyto; 324 325 bufp = __get_buf(hashp, ov_addr, bufp, 0); 326 if (!bufp) 327 return (-1); 328 329 ino = (u_int16_t *)bufp->page; 330 n = 1; 331 scopyto = hashp->BSIZE; 332 moved = 0; 333 334 if (last_bfp) 335 __free_ovflpage(hashp, last_bfp); 336 last_bfp = bufp; 337 } 338 /* Move regular sized pairs of there are any */ 339 off = hashp->BSIZE; 340 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) { 341 cino = (char *)ino; 342 key.data = (u_char *)cino + ino[n]; 343 key.size = off - ino[n]; 344 val.data = (u_char *)cino + ino[n + 1]; 345 val.size = ino[n] - ino[n + 1]; 346 off = ino[n + 1]; 347 348 if (__call_hash(hashp, key.data, key.size) == obucket) { 349 /* Keep on old page */ 350 if (PAIRFITS(op, (&key), (&val))) 351 putpair((char *)op, &key, &val); 352 else { 353 old_bufp = 354 __add_ovflpage(hashp, old_bufp); 355 if (!old_bufp) 356 return (-1); 357 op = (u_int16_t *)old_bufp->page; 358 putpair((char *)op, &key, &val); 359 } 360 old_bufp->flags |= BUF_MOD; 361 } else { 362 /* Move to new page */ 363 if (PAIRFITS(np, (&key), (&val))) 364 putpair((char *)np, &key, &val); 365 else { 366 new_bufp = 367 __add_ovflpage(hashp, new_bufp); 368 if (!new_bufp) 369 return (-1); 370 np = (u_int16_t *)new_bufp->page; 371 putpair((char *)np, &key, &val); 372 } 373 new_bufp->flags |= BUF_MOD; 374 } 375 } 376 } 377 if (last_bfp) 378 __free_ovflpage(hashp, last_bfp); 379 return (0); 380 } 381 382 /* 383 * Add the given pair to the page 384 * 385 * Returns: 386 * 0 ==> OK 387 * 1 ==> failure 388 */ 389 extern int 390 __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val) 391 { 392 u_int16_t *bp, *sop; 393 int do_expand; 394 395 bp = (u_int16_t *)bufp->page; 396 do_expand = 0; 397 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY)) 398 /* Exception case */ 399 if (bp[2] == FULL_KEY_DATA && bp[0] == 2) 400 /* This is the last page of a big key/data pair 401 and we need to add another page */ 402 break; 403 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) { 404 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 405 if (!bufp) 406 return (-1); 407 bp = (u_int16_t *)bufp->page; 408 } else 409 /* Try to squeeze key on this page */ 410 if (FREESPACE(bp) > PAIRSIZE(key, val)) { 411 squeeze_key(bp, key, val); 412 return (0); 413 } else { 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 } 419 420 if (PAIRFITS(bp, key, val)) 421 putpair(bufp->page, key, val); 422 else { 423 do_expand = 1; 424 bufp = __add_ovflpage(hashp, bufp); 425 if (!bufp) 426 return (-1); 427 sop = (u_int16_t *)bufp->page; 428 429 if (PAIRFITS(sop, key, val)) 430 putpair((char *)sop, key, val); 431 else 432 if (__big_insert(hashp, bufp, key, val)) 433 return (-1); 434 } 435 bufp->flags |= BUF_MOD; 436 /* 437 * If the average number of keys per bucket exceeds the fill factor, 438 * expand the table. 439 */ 440 hashp->NKEYS++; 441 if (do_expand || 442 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR)) 443 return (__expand_table(hashp)); 444 return (0); 445 } 446 447 /* 448 * 449 * Returns: 450 * pointer on success 451 * NULL on error 452 */ 453 extern BUFHEAD * 454 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp) 455 { 456 u_int16_t *sp; 457 u_int16_t ndx, ovfl_num; 458 #ifdef DEBUG1 459 int tmp1, tmp2; 460 #endif 461 sp = (u_int16_t *)bufp->page; 462 463 /* Check if we are dynamically determining the fill factor */ 464 if (hashp->FFACTOR == DEF_FFACTOR) { 465 hashp->FFACTOR = sp[0] >> 1; 466 if (hashp->FFACTOR < MIN_FFACTOR) 467 hashp->FFACTOR = MIN_FFACTOR; 468 } 469 bufp->flags |= BUF_MOD; 470 ovfl_num = overflow_page(hashp); 471 #ifdef DEBUG1 472 tmp1 = bufp->addr; 473 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0; 474 #endif 475 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1))) 476 return (NULL); 477 bufp->ovfl->flags |= BUF_MOD; 478 #ifdef DEBUG1 479 fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n", 480 tmp1, tmp2, bufp->ovfl->addr); 481 #endif 482 ndx = sp[0]; 483 /* 484 * Since a pair is allocated on a page only if there's room to add 485 * an overflow page, we know that the OVFL information will fit on 486 * the page. 487 */ 488 sp[ndx + 4] = OFFSET(sp); 489 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE; 490 sp[ndx + 1] = ovfl_num; 491 sp[ndx + 2] = OVFLPAGE; 492 sp[0] = ndx + 2; 493 #ifdef HASH_STATISTICS 494 hash_overflows++; 495 #endif 496 return (bufp->ovfl); 497 } 498 499 /* 500 * Returns: 501 * 0 indicates SUCCESS 502 * -1 indicates FAILURE 503 */ 504 extern int 505 __get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk, 506 int is_bitmap) 507 { 508 int fd, page, size; 509 int rsize; 510 u_int16_t *bp; 511 512 fd = hashp->fp; 513 size = hashp->BSIZE; 514 515 if ((fd == -1) || !is_disk) { 516 PAGE_INIT(p); 517 return (0); 518 } 519 if (is_bucket) 520 page = BUCKET_TO_PAGE(bucket); 521 else 522 page = OADDR_TO_PAGE(bucket); 523 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || 524 ((rsize = _read(fd, p, size)) == -1)) 525 return (-1); 526 bp = (u_int16_t *)p; 527 if (!rsize) 528 bp[0] = 0; /* We hit the EOF, so initialize a new page */ 529 else 530 if (rsize != size) { 531 errno = EFTYPE; 532 return (-1); 533 } 534 if (!is_bitmap && !bp[0]) { 535 PAGE_INIT(p); 536 } else 537 if (hashp->LORDER != BYTE_ORDER) { 538 int i, max; 539 540 if (is_bitmap) { 541 max = hashp->BSIZE >> 2; /* divide by 4 */ 542 for (i = 0; i < max; i++) 543 M_32_SWAP(((int *)p)[i]); 544 } else { 545 M_16_SWAP(bp[0]); 546 max = bp[0] + 2; 547 for (i = 1; i <= max; i++) 548 M_16_SWAP(bp[i]); 549 } 550 } 551 return (0); 552 } 553 554 /* 555 * Write page p to disk 556 * 557 * Returns: 558 * 0 ==> OK 559 * -1 ==>failure 560 */ 561 extern int 562 __put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap) 563 { 564 int fd, page, size; 565 int wsize; 566 567 size = hashp->BSIZE; 568 if ((hashp->fp == -1) && open_temp(hashp)) 569 return (-1); 570 fd = hashp->fp; 571 572 if (hashp->LORDER != BYTE_ORDER) { 573 int i; 574 int max; 575 576 if (is_bitmap) { 577 max = hashp->BSIZE >> 2; /* divide by 4 */ 578 for (i = 0; i < max; i++) 579 M_32_SWAP(((int *)p)[i]); 580 } else { 581 max = ((u_int16_t *)p)[0] + 2; 582 for (i = 0; i <= max; i++) 583 M_16_SWAP(((u_int16_t *)p)[i]); 584 } 585 } 586 if (is_bucket) 587 page = BUCKET_TO_PAGE(bucket); 588 else 589 page = OADDR_TO_PAGE(bucket); 590 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || 591 ((wsize = _write(fd, p, size)) == -1)) 592 /* Errno is set */ 593 return (-1); 594 if (wsize != size) { 595 errno = EFTYPE; 596 return (-1); 597 } 598 return (0); 599 } 600 601 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1) 602 /* 603 * Initialize a new bitmap page. Bitmap pages are left in memory 604 * once they are read in. 605 */ 606 extern int 607 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx) 608 { 609 u_int32_t *ip; 610 int clearbytes, clearints; 611 612 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 613 return (1); 614 hashp->nmaps++; 615 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1; 616 clearbytes = clearints << INT_TO_BYTE; 617 memset((char *)ip, 0, clearbytes); 618 memset(((char *)ip) + clearbytes, 0xFF, 619 hashp->BSIZE - clearbytes); 620 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK); 621 SETBIT(ip, 0); 622 hashp->BITMAPS[ndx] = (u_int16_t)pnum; 623 hashp->mapp[ndx] = ip; 624 return (0); 625 } 626 627 static u_int32_t 628 first_free(u_int32_t map) 629 { 630 u_int32_t i, mask; 631 632 mask = 0x1; 633 for (i = 0; i < BITS_PER_MAP; i++) { 634 if (!(mask & map)) 635 return (i); 636 mask = mask << 1; 637 } 638 return (i); 639 } 640 641 static u_int16_t 642 overflow_page(HTAB *hashp) 643 { 644 u_int32_t *freep; 645 int max_free, offset, splitnum; 646 u_int16_t addr; 647 int bit, first_page, free_bit, free_page, i, in_use_bits, j; 648 #ifdef DEBUG2 649 int tmp1, tmp2; 650 #endif 651 freep = NULL; 652 splitnum = hashp->OVFL_POINT; 653 max_free = hashp->SPARES[splitnum]; 654 655 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT); 656 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1); 657 658 /* Look through all the free maps to find the first free block */ 659 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT); 660 for ( i = first_page; i <= free_page; i++ ) { 661 if (!(freep = (u_int32_t *)hashp->mapp[i]) && 662 !(freep = fetch_bitmap(hashp, i))) 663 return (0); 664 if (i == free_page) 665 in_use_bits = free_bit; 666 else 667 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1; 668 669 if (i == first_page) { 670 bit = hashp->LAST_FREED & 671 ((hashp->BSIZE << BYTE_SHIFT) - 1); 672 j = bit / BITS_PER_MAP; 673 bit = bit & ~(BITS_PER_MAP - 1); 674 } else { 675 bit = 0; 676 j = 0; 677 } 678 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP) 679 if (freep[j] != ALL_SET) 680 goto found; 681 } 682 683 /* No Free Page Found */ 684 hashp->LAST_FREED = hashp->SPARES[splitnum]; 685 hashp->SPARES[splitnum]++; 686 offset = hashp->SPARES[splitnum] - 687 (splitnum ? hashp->SPARES[splitnum - 1] : 0); 688 689 #define OVMSG "HASH: Out of overflow pages. Increase page size\n" 690 if (offset > SPLITMASK) { 691 if (++splitnum >= NCACHED) { 692 _write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 693 return (0); 694 } 695 hashp->OVFL_POINT = splitnum; 696 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 697 hashp->SPARES[splitnum-1]--; 698 offset = 1; 699 } 700 701 /* Check if we need to allocate a new bitmap page */ 702 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) { 703 free_page++; 704 if (free_page >= NCACHED) { 705 _write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 706 return (0); 707 } 708 /* 709 * This is tricky. The 1 indicates that you want the new page 710 * allocated with 1 clear bit. Actually, you are going to 711 * allocate 2 pages from this map. The first is going to be 712 * the map page, the second is the overflow page we were 713 * looking for. The init_bitmap routine automatically, sets 714 * the first bit of itself to indicate that the bitmap itself 715 * is in use. We would explicitly set the second bit, but 716 * don't have to if we tell init_bitmap not to leave it clear 717 * in the first place. 718 */ 719 if (__ibitmap(hashp, 720 (int)OADDR_OF(splitnum, offset), 1, free_page)) 721 return (0); 722 hashp->SPARES[splitnum]++; 723 #ifdef DEBUG2 724 free_bit = 2; 725 #endif 726 offset++; 727 if (offset > SPLITMASK) { 728 if (++splitnum >= NCACHED) { 729 _write(STDERR_FILENO, OVMSG, 730 sizeof(OVMSG) - 1); 731 return (0); 732 } 733 hashp->OVFL_POINT = splitnum; 734 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 735 hashp->SPARES[splitnum-1]--; 736 offset = 0; 737 } 738 } else { 739 /* 740 * Free_bit addresses the last used bit. Bump it to address 741 * the first available bit. 742 */ 743 free_bit++; 744 SETBIT(freep, free_bit); 745 } 746 747 /* Calculate address of the new overflow page */ 748 addr = OADDR_OF(splitnum, offset); 749 #ifdef DEBUG2 750 fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 751 addr, free_bit, free_page); 752 #endif 753 return (addr); 754 755 found: 756 bit = bit + first_free(freep[j]); 757 SETBIT(freep, bit); 758 #ifdef DEBUG2 759 tmp1 = bit; 760 tmp2 = i; 761 #endif 762 /* 763 * Bits are addressed starting with 0, but overflow pages are addressed 764 * beginning at 1. Bit is a bit addressnumber, so we need to increment 765 * it to convert it to a page number. 766 */ 767 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT)); 768 if (bit >= hashp->LAST_FREED) 769 hashp->LAST_FREED = bit - 1; 770 771 /* Calculate the split number for this page */ 772 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++); 773 offset = (i ? bit - hashp->SPARES[i - 1] : bit); 774 if (offset >= SPLITMASK) 775 return (0); /* Out of overflow pages */ 776 addr = OADDR_OF(i, offset); 777 #ifdef DEBUG2 778 fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 779 addr, tmp1, tmp2); 780 #endif 781 782 /* Allocate and return the overflow page */ 783 return (addr); 784 } 785 786 /* 787 * Mark this overflow page as free. 788 */ 789 extern void 790 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp) 791 { 792 u_int16_t addr; 793 u_int32_t *freep; 794 int bit_address, free_page, free_bit; 795 u_int16_t ndx; 796 797 addr = obufp->addr; 798 #ifdef DEBUG1 799 fprintf(stderr, "Freeing %d\n", addr); 800 #endif 801 ndx = (((u_int16_t)addr) >> SPLITSHIFT); 802 bit_address = 803 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1; 804 if (bit_address < hashp->LAST_FREED) 805 hashp->LAST_FREED = bit_address; 806 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT)); 807 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1); 808 809 if (!(freep = hashp->mapp[free_page])) 810 freep = fetch_bitmap(hashp, free_page); 811 #ifdef DEBUG 812 /* 813 * This had better never happen. It means we tried to read a bitmap 814 * that has already had overflow pages allocated off it, and we 815 * failed to read it from the file. 816 */ 817 if (!freep) 818 assert(0); 819 #endif 820 CLRBIT(freep, free_bit); 821 #ifdef DEBUG2 822 fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n", 823 obufp->addr, free_bit, free_page); 824 #endif 825 __reclaim_buf(hashp, obufp); 826 } 827 828 /* 829 * Returns: 830 * 0 success 831 * -1 failure 832 */ 833 static int 834 open_temp(HTAB *hashp) 835 { 836 sigset_t set, oset; 837 static char namestr[] = "_hashXXXXXX"; 838 839 /* Block signals; make sure file goes away at process exit. */ 840 sigfillset(&set); 841 _sigprocmask(SIG_BLOCK, &set, &oset); 842 if ((hashp->fp = mkstemp(namestr)) != -1) { 843 unlink(namestr); 844 _fcntl(hashp->fp, F_SETFD, 1); 845 } 846 _sigprocmask(SIG_SETMASK, &oset, NULL); 847 return (hashp->fp != -1 ? 0 : -1); 848 } 849 850 /* 851 * We have to know that the key will fit, but the last entry on the page is 852 * an overflow pair, so we need to shift things. 853 */ 854 static void 855 squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val) 856 { 857 char *p; 858 u_int16_t free_space, n, off, pageno; 859 860 p = (char *)sp; 861 n = sp[0]; 862 free_space = FREESPACE(sp); 863 off = OFFSET(sp); 864 865 pageno = sp[n - 1]; 866 off -= key->size; 867 sp[n - 1] = off; 868 memmove(p + off, key->data, key->size); 869 off -= val->size; 870 sp[n] = off; 871 memmove(p + off, val->data, val->size); 872 sp[0] = n + 2; 873 sp[n + 1] = pageno; 874 sp[n + 2] = OVFLPAGE; 875 FREESPACE(sp) = free_space - PAIRSIZE(key, val); 876 OFFSET(sp) = off; 877 } 878 879 static u_int32_t * 880 fetch_bitmap(HTAB *hashp, int ndx) 881 { 882 if (ndx >= hashp->nmaps) 883 return (NULL); 884 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 885 return (NULL); 886 if (__get_page(hashp, 887 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) { 888 free(hashp->mapp[ndx]); 889 return (NULL); 890 } 891 return (hashp->mapp[ndx]); 892 } 893 894 #ifdef DEBUG4 895 int 896 print_chain(int addr) 897 { 898 BUFHEAD *bufp; 899 short *bp, oaddr; 900 901 fprintf(stderr, "%d ", addr); 902 bufp = __get_buf(hashp, addr, NULL, 0); 903 bp = (short *)bufp->page; 904 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) || 905 ((bp[0] > 2) && bp[2] < REAL_KEY))) { 906 oaddr = bp[bp[0] - 1]; 907 fprintf(stderr, "%d ", (int)oaddr); 908 bufp = __get_buf(hashp, (int)oaddr, bufp, 0); 909 bp = (short *)bufp->page; 910 } 911 fprintf(stderr, "\n"); 912 } 913 #endif 914