hash_page.c revision 190490
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 * 4. 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 33#if defined(LIBC_SCCS) && !defined(lint) 34static char sccsid[] = "@(#)hash_page.c 8.7 (Berkeley) 8/16/94"; 35#endif /* LIBC_SCCS and not lint */ 36#include <sys/cdefs.h> 37__FBSDID("$FreeBSD: head/lib/libc/db/hash/hash_page.c 190490 2009-03-28 06:25:33Z delphij $"); 38 39/* 40 * PACKAGE: hashing 41 * 42 * DESCRIPTION: 43 * Page manipulation for hashing package. 44 * 45 * ROUTINES: 46 * 47 * External 48 * __get_page 49 * __add_ovflpage 50 * Internal 51 * overflow_page 52 * open_temp 53 */ 54 55#include "namespace.h" 56#include <sys/param.h> 57 58#include <errno.h> 59#include <fcntl.h> 60#include <signal.h> 61#include <stdio.h> 62#include <stdlib.h> 63#include <string.h> 64#include <unistd.h> 65#ifdef DEBUG 66#include <assert.h> 67#endif 68#include "un-namespace.h" 69 70#include <db.h> 71#include "hash.h" 72#include "page.h" 73#include "extern.h" 74 75static u_int32_t *fetch_bitmap(HTAB *, int); 76static u_int32_t first_free(u_int32_t); 77static int open_temp(HTAB *); 78static u_int16_t overflow_page(HTAB *); 79static void putpair(char *, const DBT *, const DBT *); 80static void squeeze_key(u_int16_t *, const DBT *, const DBT *); 81static int ugly_split(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 */ 94static void 95putpair(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 */ 124int 125__delpair(HTAB *hashp, BUFHEAD *bufp, int ndx) 126{ 127 u_int16_t *bp, newoff, pairlen; 128 int n; 129 130 bp = (u_int16_t *)bufp->page; 131 n = bp[0]; 132 133 if (bp[ndx + 1] < REAL_KEY) 134 return (__big_delete(hashp, bufp)); 135 if (ndx != 1) 136 newoff = bp[ndx - 1]; 137 else 138 newoff = hashp->BSIZE; 139 pairlen = newoff - bp[ndx + 1]; 140 141 if (ndx != (n - 1)) { 142 /* Hard Case -- need to shuffle keys */ 143 int i; 144 char *src = bufp->page + (int)OFFSET(bp); 145 char *dst = src + (int)pairlen; 146 memmove(dst, src, bp[ndx + 1] - OFFSET(bp)); 147 148 /* Now adjust the pointers */ 149 for (i = ndx + 2; i <= n; i += 2) { 150 if (bp[i + 1] == OVFLPAGE) { 151 bp[i - 2] = bp[i]; 152 bp[i - 1] = bp[i + 1]; 153 } else { 154 bp[i - 2] = bp[i] + pairlen; 155 bp[i - 1] = bp[i + 1] + pairlen; 156 } 157 } 158 } 159 /* Finally adjust the page data */ 160 bp[n] = OFFSET(bp) + pairlen; 161 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t); 162 bp[0] = n - 2; 163 hashp->NKEYS--; 164 165 bufp->flags |= BUF_MOD; 166 return (0); 167} 168/* 169 * Returns: 170 * 0 ==> OK 171 * -1 ==> Error 172 */ 173int 174__split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket) 175{ 176 BUFHEAD *new_bufp, *old_bufp; 177 u_int16_t *ino; 178 char *np; 179 DBT key, val; 180 int n, ndx, retval; 181 u_int16_t copyto, diff, off, moved; 182 char *op; 183 184 copyto = (u_int16_t)hashp->BSIZE; 185 off = (u_int16_t)hashp->BSIZE; 186 old_bufp = __get_buf(hashp, obucket, NULL, 0); 187 if (old_bufp == NULL) 188 return (-1); 189 new_bufp = __get_buf(hashp, nbucket, NULL, 0); 190 if (new_bufp == NULL) 191 return (-1); 192 193 old_bufp->flags |= (BUF_MOD | BUF_PIN); 194 new_bufp->flags |= (BUF_MOD | BUF_PIN); 195 196 ino = (u_int16_t *)(op = old_bufp->page); 197 np = new_bufp->page; 198 199 moved = 0; 200 201 for (n = 1, ndx = 1; n < ino[0]; n += 2) { 202 if (ino[n + 1] < REAL_KEY) { 203 retval = ugly_split(hashp, obucket, old_bufp, new_bufp, 204 (int)copyto, (int)moved); 205 old_bufp->flags &= ~BUF_PIN; 206 new_bufp->flags &= ~BUF_PIN; 207 return (retval); 208 209 } 210 key.data = (u_char *)op + ino[n]; 211 key.size = off - ino[n]; 212 213 if (__call_hash(hashp, key.data, key.size) == obucket) { 214 /* Don't switch page */ 215 diff = copyto - off; 216 if (diff) { 217 copyto = ino[n + 1] + diff; 218 memmove(op + copyto, op + ino[n + 1], 219 off - ino[n + 1]); 220 ino[ndx] = copyto + ino[n] - ino[n + 1]; 221 ino[ndx + 1] = copyto; 222 } else 223 copyto = ino[n + 1]; 224 ndx += 2; 225 } else { 226 /* Switch page */ 227 val.data = (u_char *)op + ino[n + 1]; 228 val.size = ino[n] - ino[n + 1]; 229 putpair(np, &key, &val); 230 moved += 2; 231 } 232 233 off = ino[n + 1]; 234 } 235 236 /* Now clean up the page */ 237 ino[0] -= moved; 238 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3); 239 OFFSET(ino) = copyto; 240 241#ifdef DEBUG3 242 (void)fprintf(stderr, "split %d/%d\n", 243 ((u_int16_t *)np)[0] / 2, 244 ((u_int16_t *)op)[0] / 2); 245#endif 246 /* unpin both pages */ 247 old_bufp->flags &= ~BUF_PIN; 248 new_bufp->flags &= ~BUF_PIN; 249 return (0); 250} 251 252/* 253 * Called when we encounter an overflow or big key/data page during split 254 * handling. This is special cased since we have to begin checking whether 255 * the key/data pairs fit on their respective pages and because we may need 256 * overflow pages for both the old and new pages. 257 * 258 * The first page might be a page with regular key/data pairs in which case 259 * we have a regular overflow condition and just need to go on to the next 260 * page or it might be a big key/data pair in which case we need to fix the 261 * big key/data pair. 262 * 263 * Returns: 264 * 0 ==> success 265 * -1 ==> failure 266 */ 267static int 268ugly_split(HTAB *hashp, 269 u_int32_t obucket, /* Same as __split_page. */ 270 BUFHEAD *old_bufp, 271 BUFHEAD *new_bufp, 272 int copyto, /* First byte on page which contains key/data values. */ 273 int moved) /* Number of pairs moved to new page. */ 274{ 275 BUFHEAD *bufp; /* Buffer header for ino */ 276 u_int16_t *ino; /* Page keys come off of */ 277 u_int16_t *np; /* New page */ 278 u_int16_t *op; /* Page keys go on to if they aren't moving */ 279 280 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */ 281 DBT key, val; 282 SPLIT_RETURN ret; 283 u_int16_t n, off, ov_addr, scopyto; 284 char *cino; /* Character value of ino */ 285 286 bufp = old_bufp; 287 ino = (u_int16_t *)old_bufp->page; 288 np = (u_int16_t *)new_bufp->page; 289 op = (u_int16_t *)old_bufp->page; 290 last_bfp = NULL; 291 scopyto = (u_int16_t)copyto; /* ANSI */ 292 293 n = ino[0] - 1; 294 while (n < ino[0]) { 295 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) { 296 if (__big_split(hashp, old_bufp, 297 new_bufp, bufp, bufp->addr, obucket, &ret)) 298 return (-1); 299 old_bufp = ret.oldp; 300 if (!old_bufp) 301 return (-1); 302 op = (u_int16_t *)old_bufp->page; 303 new_bufp = ret.newp; 304 if (!new_bufp) 305 return (-1); 306 np = (u_int16_t *)new_bufp->page; 307 bufp = ret.nextp; 308 if (!bufp) 309 return (0); 310 cino = (char *)bufp->page; 311 ino = (u_int16_t *)cino; 312 last_bfp = ret.nextp; 313 } else if (ino[n + 1] == OVFLPAGE) { 314 ov_addr = ino[n]; 315 /* 316 * Fix up the old page -- the extra 2 are the fields 317 * which contained the overflow information. 318 */ 319 ino[0] -= (moved + 2); 320 FREESPACE(ino) = 321 scopyto - sizeof(u_int16_t) * (ino[0] + 3); 322 OFFSET(ino) = scopyto; 323 324 bufp = __get_buf(hashp, ov_addr, bufp, 0); 325 if (!bufp) 326 return (-1); 327 328 ino = (u_int16_t *)bufp->page; 329 n = 1; 330 scopyto = hashp->BSIZE; 331 moved = 0; 332 333 if (last_bfp) 334 __free_ovflpage(hashp, last_bfp); 335 last_bfp = bufp; 336 } 337 /* Move regular sized pairs of there are any */ 338 off = hashp->BSIZE; 339 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) { 340 cino = (char *)ino; 341 key.data = (u_char *)cino + ino[n]; 342 key.size = off - ino[n]; 343 val.data = (u_char *)cino + ino[n + 1]; 344 val.size = ino[n] - ino[n + 1]; 345 off = ino[n + 1]; 346 347 if (__call_hash(hashp, key.data, key.size) == obucket) { 348 /* Keep on old page */ 349 if (PAIRFITS(op, (&key), (&val))) 350 putpair((char *)op, &key, &val); 351 else { 352 old_bufp = 353 __add_ovflpage(hashp, old_bufp); 354 if (!old_bufp) 355 return (-1); 356 op = (u_int16_t *)old_bufp->page; 357 putpair((char *)op, &key, &val); 358 } 359 old_bufp->flags |= BUF_MOD; 360 } else { 361 /* Move to new page */ 362 if (PAIRFITS(np, (&key), (&val))) 363 putpair((char *)np, &key, &val); 364 else { 365 new_bufp = 366 __add_ovflpage(hashp, new_bufp); 367 if (!new_bufp) 368 return (-1); 369 np = (u_int16_t *)new_bufp->page; 370 putpair((char *)np, &key, &val); 371 } 372 new_bufp->flags |= BUF_MOD; 373 } 374 } 375 } 376 if (last_bfp) 377 __free_ovflpage(hashp, last_bfp); 378 return (0); 379} 380 381/* 382 * Add the given pair to the page 383 * 384 * Returns: 385 * 0 ==> OK 386 * 1 ==> failure 387 */ 388int 389__addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val) 390{ 391 u_int16_t *bp, *sop; 392 int do_expand; 393 394 bp = (u_int16_t *)bufp->page; 395 do_expand = 0; 396 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY)) 397 /* Exception case */ 398 if (bp[2] == FULL_KEY_DATA && bp[0] == 2) 399 /* This is the last page of a big key/data pair 400 and we need to add another page */ 401 break; 402 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) { 403 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 404 if (!bufp) 405 return (-1); 406 bp = (u_int16_t *)bufp->page; 407 } else if (bp[bp[0]] != OVFLPAGE) { 408 /* Short key/data pairs, no more pages */ 409 break; 410 } else { 411 /* Try to squeeze key on this page */ 412 if (bp[2] >= REAL_KEY && 413 FREESPACE(bp) >= PAIRSIZE(key, val)) { 414 squeeze_key(bp, key, val); 415 goto stats; 416 } else { 417 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); 418 if (!bufp) 419 return (-1); 420 bp = (u_int16_t *)bufp->page; 421 } 422 } 423 424 if (PAIRFITS(bp, key, val)) 425 putpair(bufp->page, key, val); 426 else { 427 do_expand = 1; 428 bufp = __add_ovflpage(hashp, bufp); 429 if (!bufp) 430 return (-1); 431 sop = (u_int16_t *)bufp->page; 432 433 if (PAIRFITS(sop, key, val)) 434 putpair((char *)sop, key, val); 435 else 436 if (__big_insert(hashp, bufp, key, val)) 437 return (-1); 438 } 439stats: 440 bufp->flags |= BUF_MOD; 441 /* 442 * If the average number of keys per bucket exceeds the fill factor, 443 * expand the table. 444 */ 445 hashp->NKEYS++; 446 if (do_expand || 447 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR)) 448 return (__expand_table(hashp)); 449 return (0); 450} 451 452/* 453 * 454 * Returns: 455 * pointer on success 456 * NULL on error 457 */ 458BUFHEAD * 459__add_ovflpage(HTAB *hashp, BUFHEAD *bufp) 460{ 461 u_int16_t *sp, ndx, ovfl_num; 462#ifdef DEBUG1 463 int tmp1, tmp2; 464#endif 465 sp = (u_int16_t *)bufp->page; 466 467 /* Check if we are dynamically determining the fill factor */ 468 if (hashp->FFACTOR == DEF_FFACTOR) { 469 hashp->FFACTOR = sp[0] >> 1; 470 if (hashp->FFACTOR < MIN_FFACTOR) 471 hashp->FFACTOR = MIN_FFACTOR; 472 } 473 bufp->flags |= BUF_MOD; 474 ovfl_num = overflow_page(hashp); 475#ifdef DEBUG1 476 tmp1 = bufp->addr; 477 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0; 478#endif 479 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1))) 480 return (NULL); 481 bufp->ovfl->flags |= BUF_MOD; 482#ifdef DEBUG1 483 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n", 484 tmp1, tmp2, bufp->ovfl->addr); 485#endif 486 ndx = sp[0]; 487 /* 488 * Since a pair is allocated on a page only if there's room to add 489 * an overflow page, we know that the OVFL information will fit on 490 * the page. 491 */ 492 sp[ndx + 4] = OFFSET(sp); 493 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE; 494 sp[ndx + 1] = ovfl_num; 495 sp[ndx + 2] = OVFLPAGE; 496 sp[0] = ndx + 2; 497#ifdef HASH_STATISTICS 498 hash_overflows++; 499#endif 500 return (bufp->ovfl); 501} 502 503/* 504 * Returns: 505 * 0 indicates SUCCESS 506 * -1 indicates FAILURE 507 */ 508int 509__get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk, 510 int is_bitmap) 511{ 512 int fd, page, size, rsize; 513 u_int16_t *bp; 514 515 fd = hashp->fp; 516 size = hashp->BSIZE; 517 518 if ((fd == -1) || !is_disk) { 519 PAGE_INIT(p); 520 return (0); 521 } 522 if (is_bucket) 523 page = BUCKET_TO_PAGE(bucket); 524 else 525 page = OADDR_TO_PAGE(bucket); 526 if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1) 527 return (-1); 528 bp = (u_int16_t *)p; 529 if (!rsize) 530 bp[0] = 0; /* We hit the EOF, so initialize a new page */ 531 else 532 if (rsize != size) { 533 errno = EFTYPE; 534 return (-1); 535 } 536 if (!is_bitmap && !bp[0]) { 537 PAGE_INIT(p); 538 } else 539 if (hashp->LORDER != BYTE_ORDER) { 540 int i, max; 541 542 if (is_bitmap) { 543 max = hashp->BSIZE >> 2; /* divide by 4 */ 544 for (i = 0; i < max; i++) 545 M_32_SWAP(((int *)p)[i]); 546 } else { 547 M_16_SWAP(bp[0]); 548 max = bp[0] + 2; 549 for (i = 1; i <= max; i++) 550 M_16_SWAP(bp[i]); 551 } 552 } 553 return (0); 554} 555 556/* 557 * Write page p to disk 558 * 559 * Returns: 560 * 0 ==> OK 561 * -1 ==>failure 562 */ 563int 564__put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap) 565{ 566 int fd, page, size, wsize; 567 568 size = hashp->BSIZE; 569 if ((hashp->fp == -1) && open_temp(hashp)) 570 return (-1); 571 fd = hashp->fp; 572 573 if (hashp->LORDER != BYTE_ORDER) { 574 int i, 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 ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1) 591 /* Errno is set */ 592 return (-1); 593 if (wsize != size) { 594 errno = EFTYPE; 595 return (-1); 596 } 597 return (0); 598} 599 600#define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1) 601/* 602 * Initialize a new bitmap page. Bitmap pages are left in memory 603 * once they are read in. 604 */ 605int 606__ibitmap(HTAB *hashp, int pnum, int nbits, int ndx) 607{ 608 u_int32_t *ip; 609 int clearbytes, clearints; 610 611 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 612 return (1); 613 hashp->nmaps++; 614 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1; 615 clearbytes = clearints << INT_TO_BYTE; 616 (void)memset((char *)ip, 0, clearbytes); 617 (void)memset(((char *)ip) + clearbytes, 0xFF, 618 hashp->BSIZE - clearbytes); 619 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK); 620 SETBIT(ip, 0); 621 hashp->BITMAPS[ndx] = (u_int16_t)pnum; 622 hashp->mapp[ndx] = ip; 623 return (0); 624} 625 626static u_int32_t 627first_free(u_int32_t map) 628{ 629 u_int32_t i, mask; 630 631 mask = 0x1; 632 for (i = 0; i < BITS_PER_MAP; i++) { 633 if (!(mask & map)) 634 return (i); 635 mask = mask << 1; 636 } 637 return (i); 638} 639 640static u_int16_t 641overflow_page(HTAB *hashp) 642{ 643 u_int32_t *freep; 644 int max_free, offset, splitnum; 645 u_int16_t addr; 646 int bit, first_page, free_bit, free_page, i, in_use_bits, j; 647#ifdef DEBUG2 648 int tmp1, tmp2; 649#endif 650 splitnum = hashp->OVFL_POINT; 651 max_free = hashp->SPARES[splitnum]; 652 653 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT); 654 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1); 655 656 /* Look through all the free maps to find the first free block */ 657 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT); 658 for ( i = first_page; i <= free_page; i++ ) { 659 if (!(freep = (u_int32_t *)hashp->mapp[i]) && 660 !(freep = fetch_bitmap(hashp, i))) 661 return (0); 662 if (i == free_page) 663 in_use_bits = free_bit; 664 else 665 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1; 666 667 if (i == first_page) { 668 bit = hashp->LAST_FREED & 669 ((hashp->BSIZE << BYTE_SHIFT) - 1); 670 j = bit / BITS_PER_MAP; 671 bit = bit & ~(BITS_PER_MAP - 1); 672 } else { 673 bit = 0; 674 j = 0; 675 } 676 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP) 677 if (freep[j] != ALL_SET) 678 goto found; 679 } 680 681 /* No Free Page Found */ 682 hashp->LAST_FREED = hashp->SPARES[splitnum]; 683 hashp->SPARES[splitnum]++; 684 offset = hashp->SPARES[splitnum] - 685 (splitnum ? hashp->SPARES[splitnum - 1] : 0); 686 687#define OVMSG "HASH: Out of overflow pages. Increase page size\n" 688 if (offset > SPLITMASK) { 689 if (++splitnum >= NCACHED) { 690 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 691 errno = EFBIG; 692 return (0); 693 } 694 hashp->OVFL_POINT = splitnum; 695 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 696 hashp->SPARES[splitnum-1]--; 697 offset = 1; 698 } 699 700 /* Check if we need to allocate a new bitmap page */ 701 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) { 702 free_page++; 703 if (free_page >= NCACHED) { 704 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 705 errno = EFBIG; 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 (void)_write(STDERR_FILENO, OVMSG, 730 sizeof(OVMSG) - 1); 731 errno = EFBIG; 732 return (0); 733 } 734 hashp->OVFL_POINT = splitnum; 735 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; 736 hashp->SPARES[splitnum-1]--; 737 offset = 0; 738 } 739 } else { 740 /* 741 * Free_bit addresses the last used bit. Bump it to address 742 * the first available bit. 743 */ 744 free_bit++; 745 SETBIT(freep, free_bit); 746 } 747 748 /* Calculate address of the new overflow page */ 749 addr = OADDR_OF(splitnum, offset); 750#ifdef DEBUG2 751 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 752 addr, free_bit, free_page); 753#endif 754 return (addr); 755 756found: 757 bit = bit + first_free(freep[j]); 758 SETBIT(freep, bit); 759#ifdef DEBUG2 760 tmp1 = bit; 761 tmp2 = i; 762#endif 763 /* 764 * Bits are addressed starting with 0, but overflow pages are addressed 765 * beginning at 1. Bit is a bit addressnumber, so we need to increment 766 * it to convert it to a page number. 767 */ 768 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT)); 769 if (bit >= hashp->LAST_FREED) 770 hashp->LAST_FREED = bit - 1; 771 772 /* Calculate the split number for this page */ 773 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++); 774 offset = (i ? bit - hashp->SPARES[i - 1] : bit); 775 if (offset >= SPLITMASK) { 776 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); 777 errno = EFBIG; 778 return (0); /* Out of overflow pages */ 779 } 780 addr = OADDR_OF(i, offset); 781#ifdef DEBUG2 782 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", 783 addr, tmp1, tmp2); 784#endif 785 786 /* Allocate and return the overflow page */ 787 return (addr); 788} 789 790/* 791 * Mark this overflow page as free. 792 */ 793void 794__free_ovflpage(HTAB *hashp, BUFHEAD *obufp) 795{ 796 u_int16_t addr; 797 u_int32_t *freep; 798 int bit_address, free_page, free_bit; 799 u_int16_t ndx; 800 801 addr = obufp->addr; 802#ifdef DEBUG1 803 (void)fprintf(stderr, "Freeing %d\n", addr); 804#endif 805 ndx = (((u_int16_t)addr) >> SPLITSHIFT); 806 bit_address = 807 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1; 808 if (bit_address < hashp->LAST_FREED) 809 hashp->LAST_FREED = bit_address; 810 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT)); 811 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1); 812 813 if (!(freep = hashp->mapp[free_page])) 814 freep = fetch_bitmap(hashp, free_page); 815#ifdef DEBUG 816 /* 817 * This had better never happen. It means we tried to read a bitmap 818 * that has already had overflow pages allocated off it, and we 819 * failed to read it from the file. 820 */ 821 if (!freep) 822 assert(0); 823#endif 824 CLRBIT(freep, free_bit); 825#ifdef DEBUG2 826 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n", 827 obufp->addr, free_bit, free_page); 828#endif 829 __reclaim_buf(hashp, obufp); 830} 831 832/* 833 * Returns: 834 * 0 success 835 * -1 failure 836 */ 837static int 838open_temp(HTAB *hashp) 839{ 840 sigset_t set, oset; 841 int len; 842 char *envtmp = NULL; 843 char path[MAXPATHLEN]; 844 845 if (issetugid() == 0) 846 envtmp = getenv("TMPDIR"); 847 len = snprintf(path, 848 sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp"); 849 if (len < 0 || len >= sizeof(path)) { 850 errno = ENAMETOOLONG; 851 return (-1); 852 } 853 854 /* Block signals; make sure file goes away at process exit. */ 855 (void)sigfillset(&set); 856 (void)_sigprocmask(SIG_BLOCK, &set, &oset); 857 if ((hashp->fp = mkstemp(path)) != -1) { 858 (void)unlink(path); 859 (void)_fcntl(hashp->fp, F_SETFD, 1); 860 } 861 (void)_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL); 862 return (hashp->fp != -1 ? 0 : -1); 863} 864 865/* 866 * We have to know that the key will fit, but the last entry on the page is 867 * an overflow pair, so we need to shift things. 868 */ 869static void 870squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val) 871{ 872 char *p; 873 u_int16_t free_space, n, off, pageno; 874 875 p = (char *)sp; 876 n = sp[0]; 877 free_space = FREESPACE(sp); 878 off = OFFSET(sp); 879 880 pageno = sp[n - 1]; 881 off -= key->size; 882 sp[n - 1] = off; 883 memmove(p + off, key->data, key->size); 884 off -= val->size; 885 sp[n] = off; 886 memmove(p + off, val->data, val->size); 887 sp[0] = n + 2; 888 sp[n + 1] = pageno; 889 sp[n + 2] = OVFLPAGE; 890 FREESPACE(sp) = free_space - PAIRSIZE(key, val); 891 OFFSET(sp) = off; 892} 893 894static u_int32_t * 895fetch_bitmap(HTAB *hashp, int ndx) 896{ 897 if (ndx >= hashp->nmaps) 898 return (NULL); 899 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) 900 return (NULL); 901 if (__get_page(hashp, 902 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) { 903 free(hashp->mapp[ndx]); 904 return (NULL); 905 } 906 return (hashp->mapp[ndx]); 907} 908 909#ifdef DEBUG4 910int 911print_chain(int addr) 912{ 913 BUFHEAD *bufp; 914 short *bp, oaddr; 915 916 (void)fprintf(stderr, "%d ", addr); 917 bufp = __get_buf(hashp, addr, NULL, 0); 918 bp = (short *)bufp->page; 919 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) || 920 ((bp[0] > 2) && bp[2] < REAL_KEY))) { 921 oaddr = bp[bp[0] - 1]; 922 (void)fprintf(stderr, "%d ", (int)oaddr); 923 bufp = __get_buf(hashp, (int)oaddr, bufp, 0); 924 bp = (short *)bufp->page; 925 } 926 (void)fprintf(stderr, "\n"); 927} 928#endif 929