Thu, 22 Jan 2015 13:21:57 +0100
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
michael@0 | 1 | /*- |
michael@0 | 2 | * Copyright (c) 1990, 1993, 1994 |
michael@0 | 3 | * The Regents of the University of California. All rights reserved. |
michael@0 | 4 | * |
michael@0 | 5 | * This code is derived from software contributed to Berkeley by |
michael@0 | 6 | * Margo Seltzer. |
michael@0 | 7 | * |
michael@0 | 8 | * Redistribution and use in source and binary forms, with or without |
michael@0 | 9 | * modification, are permitted provided that the following conditions |
michael@0 | 10 | * are met: |
michael@0 | 11 | * 1. Redistributions of source code must retain the above copyright |
michael@0 | 12 | * notice, this list of conditions and the following disclaimer. |
michael@0 | 13 | * 2. Redistributions in binary form must reproduce the above copyright |
michael@0 | 14 | * notice, this list of conditions and the following disclaimer in the |
michael@0 | 15 | * documentation and/or other materials provided with the distribution. |
michael@0 | 16 | * 3. ***REMOVED*** - see |
michael@0 | 17 | * ftp://ftp.cs.berkeley.edu/pub/4bsd/README.Impt.License.Change |
michael@0 | 18 | * 4. Neither the name of the University nor the names of its contributors |
michael@0 | 19 | * may be used to endorse or promote products derived from this software |
michael@0 | 20 | * without specific prior written permission. |
michael@0 | 21 | * |
michael@0 | 22 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
michael@0 | 23 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
michael@0 | 24 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
michael@0 | 25 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
michael@0 | 26 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
michael@0 | 27 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
michael@0 | 28 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
michael@0 | 29 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
michael@0 | 30 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
michael@0 | 31 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
michael@0 | 32 | * SUCH DAMAGE. |
michael@0 | 33 | */ |
michael@0 | 34 | |
michael@0 | 35 | #if defined(LIBC_SCCS) && !defined(lint) |
michael@0 | 36 | static char sccsid[] = "@(#)hash_bigkey.c 8.3 (Berkeley) 5/31/94"; |
michael@0 | 37 | #endif /* LIBC_SCCS and not lint */ |
michael@0 | 38 | |
michael@0 | 39 | /* |
michael@0 | 40 | * PACKAGE: hash |
michael@0 | 41 | * DESCRIPTION: |
michael@0 | 42 | * Big key/data handling for the hashing package. |
michael@0 | 43 | * |
michael@0 | 44 | * ROUTINES: |
michael@0 | 45 | * External |
michael@0 | 46 | * __big_keydata |
michael@0 | 47 | * __big_split |
michael@0 | 48 | * __big_insert |
michael@0 | 49 | * __big_return |
michael@0 | 50 | * __big_delete |
michael@0 | 51 | * __find_last_page |
michael@0 | 52 | * Internal |
michael@0 | 53 | * collect_key |
michael@0 | 54 | * collect_data |
michael@0 | 55 | */ |
michael@0 | 56 | |
michael@0 | 57 | #if !defined(_WIN32) && !defined(_WINDOWS) && !defined(macintosh) |
michael@0 | 58 | #include <sys/param.h> |
michael@0 | 59 | #endif |
michael@0 | 60 | |
michael@0 | 61 | #include <errno.h> |
michael@0 | 62 | #include <stdio.h> |
michael@0 | 63 | #include <stdlib.h> |
michael@0 | 64 | #include <string.h> |
michael@0 | 65 | |
michael@0 | 66 | #ifdef DEBUG |
michael@0 | 67 | #include <assert.h> |
michael@0 | 68 | #endif |
michael@0 | 69 | |
michael@0 | 70 | #include "mcom_db.h" |
michael@0 | 71 | #include "hash.h" |
michael@0 | 72 | #include "page.h" |
michael@0 | 73 | /* #include "extern.h" */ |
michael@0 | 74 | |
michael@0 | 75 | static int collect_key __P((HTAB *, BUFHEAD *, int, DBT *, int)); |
michael@0 | 76 | static int collect_data __P((HTAB *, BUFHEAD *, int, int)); |
michael@0 | 77 | |
michael@0 | 78 | /* |
michael@0 | 79 | * Big_insert |
michael@0 | 80 | * |
michael@0 | 81 | * You need to do an insert and the key/data pair is too big |
michael@0 | 82 | * |
michael@0 | 83 | * Returns: |
michael@0 | 84 | * 0 ==> OK |
michael@0 | 85 | *-1 ==> ERROR |
michael@0 | 86 | */ |
michael@0 | 87 | extern int |
michael@0 | 88 | __big_insert(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val) |
michael@0 | 89 | { |
michael@0 | 90 | register uint16 *p; |
michael@0 | 91 | uint key_size, n, val_size; |
michael@0 | 92 | uint16 space, move_bytes, off; |
michael@0 | 93 | char *cp, *key_data, *val_data; |
michael@0 | 94 | |
michael@0 | 95 | cp = bufp->page; /* Character pointer of p. */ |
michael@0 | 96 | p = (uint16 *)cp; |
michael@0 | 97 | |
michael@0 | 98 | key_data = (char *)key->data; |
michael@0 | 99 | key_size = key->size; |
michael@0 | 100 | val_data = (char *)val->data; |
michael@0 | 101 | val_size = val->size; |
michael@0 | 102 | |
michael@0 | 103 | /* First move the Key */ |
michael@0 | 104 | for (space = FREESPACE(p) - BIGOVERHEAD; key_size; |
michael@0 | 105 | space = FREESPACE(p) - BIGOVERHEAD) { |
michael@0 | 106 | move_bytes = PR_MIN(space, key_size); |
michael@0 | 107 | off = OFFSET(p) - move_bytes; |
michael@0 | 108 | memmove(cp + off, key_data, move_bytes); |
michael@0 | 109 | key_size -= move_bytes; |
michael@0 | 110 | key_data += move_bytes; |
michael@0 | 111 | n = p[0]; |
michael@0 | 112 | p[++n] = off; |
michael@0 | 113 | p[0] = ++n; |
michael@0 | 114 | FREESPACE(p) = off - PAGE_META(n); |
michael@0 | 115 | OFFSET(p) = off; |
michael@0 | 116 | p[n] = PARTIAL_KEY; |
michael@0 | 117 | bufp = __add_ovflpage(hashp, bufp); |
michael@0 | 118 | if (!bufp) |
michael@0 | 119 | return (-1); |
michael@0 | 120 | n = p[0]; |
michael@0 | 121 | if (!key_size) { |
michael@0 | 122 | if (FREESPACE(p)) { |
michael@0 | 123 | move_bytes = PR_MIN(FREESPACE(p), val_size); |
michael@0 | 124 | off = OFFSET(p) - move_bytes; |
michael@0 | 125 | p[n] = off; |
michael@0 | 126 | memmove(cp + off, val_data, move_bytes); |
michael@0 | 127 | val_data += move_bytes; |
michael@0 | 128 | val_size -= move_bytes; |
michael@0 | 129 | p[n - 2] = FULL_KEY_DATA; |
michael@0 | 130 | FREESPACE(p) = FREESPACE(p) - move_bytes; |
michael@0 | 131 | OFFSET(p) = off; |
michael@0 | 132 | } else |
michael@0 | 133 | p[n - 2] = FULL_KEY; |
michael@0 | 134 | } |
michael@0 | 135 | p = (uint16 *)bufp->page; |
michael@0 | 136 | cp = bufp->page; |
michael@0 | 137 | bufp->flags |= BUF_MOD; |
michael@0 | 138 | } |
michael@0 | 139 | |
michael@0 | 140 | /* Now move the data */ |
michael@0 | 141 | for (space = FREESPACE(p) - BIGOVERHEAD; val_size; |
michael@0 | 142 | space = FREESPACE(p) - BIGOVERHEAD) { |
michael@0 | 143 | move_bytes = PR_MIN(space, val_size); |
michael@0 | 144 | /* |
michael@0 | 145 | * Here's the hack to make sure that if the data ends on the |
michael@0 | 146 | * same page as the key ends, FREESPACE is at least one. |
michael@0 | 147 | */ |
michael@0 | 148 | if (space == val_size && val_size == val->size) |
michael@0 | 149 | move_bytes--; |
michael@0 | 150 | off = OFFSET(p) - move_bytes; |
michael@0 | 151 | memmove(cp + off, val_data, move_bytes); |
michael@0 | 152 | val_size -= move_bytes; |
michael@0 | 153 | val_data += move_bytes; |
michael@0 | 154 | n = p[0]; |
michael@0 | 155 | p[++n] = off; |
michael@0 | 156 | p[0] = ++n; |
michael@0 | 157 | FREESPACE(p) = off - PAGE_META(n); |
michael@0 | 158 | OFFSET(p) = off; |
michael@0 | 159 | if (val_size) { |
michael@0 | 160 | p[n] = FULL_KEY; |
michael@0 | 161 | bufp = __add_ovflpage(hashp, bufp); |
michael@0 | 162 | if (!bufp) |
michael@0 | 163 | return (-1); |
michael@0 | 164 | cp = bufp->page; |
michael@0 | 165 | p = (uint16 *)cp; |
michael@0 | 166 | } else |
michael@0 | 167 | p[n] = FULL_KEY_DATA; |
michael@0 | 168 | bufp->flags |= BUF_MOD; |
michael@0 | 169 | } |
michael@0 | 170 | return (0); |
michael@0 | 171 | } |
michael@0 | 172 | |
michael@0 | 173 | /* |
michael@0 | 174 | * Called when bufp's page contains a partial key (index should be 1) |
michael@0 | 175 | * |
michael@0 | 176 | * All pages in the big key/data pair except bufp are freed. We cannot |
michael@0 | 177 | * free bufp because the page pointing to it is lost and we can't get rid |
michael@0 | 178 | * of its pointer. |
michael@0 | 179 | * |
michael@0 | 180 | * Returns: |
michael@0 | 181 | * 0 => OK |
michael@0 | 182 | *-1 => ERROR |
michael@0 | 183 | */ |
michael@0 | 184 | extern int |
michael@0 | 185 | __big_delete(HTAB *hashp, BUFHEAD *bufp) |
michael@0 | 186 | { |
michael@0 | 187 | register BUFHEAD *last_bfp, *rbufp; |
michael@0 | 188 | uint16 *bp, pageno; |
michael@0 | 189 | int key_done, n; |
michael@0 | 190 | |
michael@0 | 191 | rbufp = bufp; |
michael@0 | 192 | last_bfp = NULL; |
michael@0 | 193 | bp = (uint16 *)bufp->page; |
michael@0 | 194 | pageno = 0; |
michael@0 | 195 | key_done = 0; |
michael@0 | 196 | |
michael@0 | 197 | while (!key_done || (bp[2] != FULL_KEY_DATA)) { |
michael@0 | 198 | if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) |
michael@0 | 199 | key_done = 1; |
michael@0 | 200 | |
michael@0 | 201 | /* |
michael@0 | 202 | * If there is freespace left on a FULL_KEY_DATA page, then |
michael@0 | 203 | * the data is short and fits entirely on this page, and this |
michael@0 | 204 | * is the last page. |
michael@0 | 205 | */ |
michael@0 | 206 | if (bp[2] == FULL_KEY_DATA && FREESPACE(bp)) |
michael@0 | 207 | break; |
michael@0 | 208 | pageno = bp[bp[0] - 1]; |
michael@0 | 209 | rbufp->flags |= BUF_MOD; |
michael@0 | 210 | rbufp = __get_buf(hashp, pageno, rbufp, 0); |
michael@0 | 211 | if (last_bfp) |
michael@0 | 212 | __free_ovflpage(hashp, last_bfp); |
michael@0 | 213 | last_bfp = rbufp; |
michael@0 | 214 | if (!rbufp) |
michael@0 | 215 | return (-1); /* Error. */ |
michael@0 | 216 | bp = (uint16 *)rbufp->page; |
michael@0 | 217 | } |
michael@0 | 218 | |
michael@0 | 219 | /* |
michael@0 | 220 | * If we get here then rbufp points to the last page of the big |
michael@0 | 221 | * key/data pair. Bufp points to the first one -- it should now be |
michael@0 | 222 | * empty pointing to the next page after this pair. Can't free it |
michael@0 | 223 | * because we don't have the page pointing to it. |
michael@0 | 224 | */ |
michael@0 | 225 | |
michael@0 | 226 | /* This is information from the last page of the pair. */ |
michael@0 | 227 | n = bp[0]; |
michael@0 | 228 | pageno = bp[n - 1]; |
michael@0 | 229 | |
michael@0 | 230 | /* Now, bp is the first page of the pair. */ |
michael@0 | 231 | bp = (uint16 *)bufp->page; |
michael@0 | 232 | if (n > 2) { |
michael@0 | 233 | /* There is an overflow page. */ |
michael@0 | 234 | bp[1] = pageno; |
michael@0 | 235 | bp[2] = OVFLPAGE; |
michael@0 | 236 | bufp->ovfl = rbufp->ovfl; |
michael@0 | 237 | } else |
michael@0 | 238 | /* This is the last page. */ |
michael@0 | 239 | bufp->ovfl = NULL; |
michael@0 | 240 | n -= 2; |
michael@0 | 241 | bp[0] = n; |
michael@0 | 242 | FREESPACE(bp) = hashp->BSIZE - PAGE_META(n); |
michael@0 | 243 | OFFSET(bp) = hashp->BSIZE - 1; |
michael@0 | 244 | |
michael@0 | 245 | bufp->flags |= BUF_MOD; |
michael@0 | 246 | if (rbufp) |
michael@0 | 247 | __free_ovflpage(hashp, rbufp); |
michael@0 | 248 | if (last_bfp != rbufp) |
michael@0 | 249 | __free_ovflpage(hashp, last_bfp); |
michael@0 | 250 | |
michael@0 | 251 | hashp->NKEYS--; |
michael@0 | 252 | return (0); |
michael@0 | 253 | } |
michael@0 | 254 | /* |
michael@0 | 255 | * Returns: |
michael@0 | 256 | * 0 = key not found |
michael@0 | 257 | * -1 = get next overflow page |
michael@0 | 258 | * -2 means key not found and this is big key/data |
michael@0 | 259 | * -3 error |
michael@0 | 260 | */ |
michael@0 | 261 | extern int |
michael@0 | 262 | __find_bigpair(HTAB *hashp, BUFHEAD *bufp, int ndx, char *key, int size) |
michael@0 | 263 | { |
michael@0 | 264 | register uint16 *bp; |
michael@0 | 265 | register char *p; |
michael@0 | 266 | int ksize; |
michael@0 | 267 | uint16 bytes; |
michael@0 | 268 | char *kkey; |
michael@0 | 269 | |
michael@0 | 270 | bp = (uint16 *)bufp->page; |
michael@0 | 271 | p = bufp->page; |
michael@0 | 272 | ksize = size; |
michael@0 | 273 | kkey = key; |
michael@0 | 274 | |
michael@0 | 275 | for (bytes = hashp->BSIZE - bp[ndx]; |
michael@0 | 276 | bytes <= size && bp[ndx + 1] == PARTIAL_KEY; |
michael@0 | 277 | bytes = hashp->BSIZE - bp[ndx]) { |
michael@0 | 278 | if (memcmp(p + bp[ndx], kkey, bytes)) |
michael@0 | 279 | return (-2); |
michael@0 | 280 | kkey += bytes; |
michael@0 | 281 | ksize -= bytes; |
michael@0 | 282 | bufp = __get_buf(hashp, bp[ndx + 2], bufp, 0); |
michael@0 | 283 | if (!bufp) |
michael@0 | 284 | return (-3); |
michael@0 | 285 | p = bufp->page; |
michael@0 | 286 | bp = (uint16 *)p; |
michael@0 | 287 | ndx = 1; |
michael@0 | 288 | } |
michael@0 | 289 | |
michael@0 | 290 | if (bytes != ksize || memcmp(p + bp[ndx], kkey, bytes)) { |
michael@0 | 291 | #ifdef HASH_STATISTICS |
michael@0 | 292 | ++hash_collisions; |
michael@0 | 293 | #endif |
michael@0 | 294 | return (-2); |
michael@0 | 295 | } else |
michael@0 | 296 | return (ndx); |
michael@0 | 297 | } |
michael@0 | 298 | |
michael@0 | 299 | /* |
michael@0 | 300 | * Given the buffer pointer of the first overflow page of a big pair, |
michael@0 | 301 | * find the end of the big pair |
michael@0 | 302 | * |
michael@0 | 303 | * This will set bpp to the buffer header of the last page of the big pair. |
michael@0 | 304 | * It will return the pageno of the overflow page following the last page |
michael@0 | 305 | * of the pair; 0 if there isn't any (i.e. big pair is the last key in the |
michael@0 | 306 | * bucket) |
michael@0 | 307 | */ |
michael@0 | 308 | extern uint16 |
michael@0 | 309 | __find_last_page(HTAB *hashp, BUFHEAD **bpp) |
michael@0 | 310 | { |
michael@0 | 311 | BUFHEAD *bufp; |
michael@0 | 312 | uint16 *bp, pageno; |
michael@0 | 313 | uint n; |
michael@0 | 314 | |
michael@0 | 315 | bufp = *bpp; |
michael@0 | 316 | bp = (uint16 *)bufp->page; |
michael@0 | 317 | for (;;) { |
michael@0 | 318 | n = bp[0]; |
michael@0 | 319 | |
michael@0 | 320 | /* |
michael@0 | 321 | * This is the last page if: the tag is FULL_KEY_DATA and |
michael@0 | 322 | * either only 2 entries OVFLPAGE marker is explicit there |
michael@0 | 323 | * is freespace on the page. |
michael@0 | 324 | */ |
michael@0 | 325 | if (bp[2] == FULL_KEY_DATA && |
michael@0 | 326 | ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp)))) |
michael@0 | 327 | break; |
michael@0 | 328 | |
michael@0 | 329 | /* LJM bound the size of n to reasonable limits |
michael@0 | 330 | */ |
michael@0 | 331 | if(n > hashp->BSIZE/sizeof(uint16)) |
michael@0 | 332 | return(0); |
michael@0 | 333 | |
michael@0 | 334 | pageno = bp[n - 1]; |
michael@0 | 335 | bufp = __get_buf(hashp, pageno, bufp, 0); |
michael@0 | 336 | if (!bufp) |
michael@0 | 337 | return (0); /* Need to indicate an error! */ |
michael@0 | 338 | bp = (uint16 *)bufp->page; |
michael@0 | 339 | } |
michael@0 | 340 | |
michael@0 | 341 | *bpp = bufp; |
michael@0 | 342 | if (bp[0] > 2) |
michael@0 | 343 | return (bp[3]); |
michael@0 | 344 | else |
michael@0 | 345 | return (0); |
michael@0 | 346 | } |
michael@0 | 347 | |
michael@0 | 348 | /* |
michael@0 | 349 | * Return the data for the key/data pair that begins on this page at this |
michael@0 | 350 | * index (index should always be 1). |
michael@0 | 351 | */ |
michael@0 | 352 | extern int |
michael@0 | 353 | __big_return( |
michael@0 | 354 | HTAB *hashp, |
michael@0 | 355 | BUFHEAD *bufp, |
michael@0 | 356 | int ndx, |
michael@0 | 357 | DBT *val, |
michael@0 | 358 | int set_current) |
michael@0 | 359 | { |
michael@0 | 360 | BUFHEAD *save_p; |
michael@0 | 361 | uint16 *bp, len, off, save_addr; |
michael@0 | 362 | char *tp; |
michael@0 | 363 | int save_flags; |
michael@0 | 364 | |
michael@0 | 365 | bp = (uint16 *)bufp->page; |
michael@0 | 366 | while (bp[ndx + 1] == PARTIAL_KEY) { |
michael@0 | 367 | bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); |
michael@0 | 368 | if (!bufp) |
michael@0 | 369 | return (-1); |
michael@0 | 370 | bp = (uint16 *)bufp->page; |
michael@0 | 371 | ndx = 1; |
michael@0 | 372 | } |
michael@0 | 373 | |
michael@0 | 374 | if (bp[ndx + 1] == FULL_KEY) { |
michael@0 | 375 | bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); |
michael@0 | 376 | if (!bufp) |
michael@0 | 377 | return (-1); |
michael@0 | 378 | bp = (uint16 *)bufp->page; |
michael@0 | 379 | save_p = bufp; |
michael@0 | 380 | save_addr = save_p->addr; |
michael@0 | 381 | off = bp[1]; |
michael@0 | 382 | len = 0; |
michael@0 | 383 | } else |
michael@0 | 384 | if (!FREESPACE(bp)) { |
michael@0 | 385 | /* |
michael@0 | 386 | * This is a hack. We can't distinguish between |
michael@0 | 387 | * FULL_KEY_DATA that contains complete data or |
michael@0 | 388 | * incomplete data, so we require that if the data |
michael@0 | 389 | * is complete, there is at least 1 byte of free |
michael@0 | 390 | * space left. |
michael@0 | 391 | */ |
michael@0 | 392 | off = bp[bp[0]]; |
michael@0 | 393 | len = bp[1] - off; |
michael@0 | 394 | save_p = bufp; |
michael@0 | 395 | save_addr = bufp->addr; |
michael@0 | 396 | bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); |
michael@0 | 397 | if (!bufp) |
michael@0 | 398 | return (-1); |
michael@0 | 399 | bp = (uint16 *)bufp->page; |
michael@0 | 400 | } else { |
michael@0 | 401 | /* The data is all on one page. */ |
michael@0 | 402 | tp = (char *)bp; |
michael@0 | 403 | off = bp[bp[0]]; |
michael@0 | 404 | val->data = (uint8 *)tp + off; |
michael@0 | 405 | val->size = bp[1] - off; |
michael@0 | 406 | if (set_current) { |
michael@0 | 407 | if (bp[0] == 2) { /* No more buckets in |
michael@0 | 408 | * chain */ |
michael@0 | 409 | hashp->cpage = NULL; |
michael@0 | 410 | hashp->cbucket++; |
michael@0 | 411 | hashp->cndx = 1; |
michael@0 | 412 | } else { |
michael@0 | 413 | hashp->cpage = __get_buf(hashp, |
michael@0 | 414 | bp[bp[0] - 1], bufp, 0); |
michael@0 | 415 | if (!hashp->cpage) |
michael@0 | 416 | return (-1); |
michael@0 | 417 | hashp->cndx = 1; |
michael@0 | 418 | if (!((uint16 *) |
michael@0 | 419 | hashp->cpage->page)[0]) { |
michael@0 | 420 | hashp->cbucket++; |
michael@0 | 421 | hashp->cpage = NULL; |
michael@0 | 422 | } |
michael@0 | 423 | } |
michael@0 | 424 | } |
michael@0 | 425 | return (0); |
michael@0 | 426 | } |
michael@0 | 427 | |
michael@0 | 428 | /* pin our saved buf so that we don't lose if |
michael@0 | 429 | * we run out of buffers */ |
michael@0 | 430 | save_flags = save_p->flags; |
michael@0 | 431 | save_p->flags |= BUF_PIN; |
michael@0 | 432 | val->size = collect_data(hashp, bufp, (int)len, set_current); |
michael@0 | 433 | save_p->flags = save_flags; |
michael@0 | 434 | if (val->size == (size_t)-1) |
michael@0 | 435 | return (-1); |
michael@0 | 436 | if (save_p->addr != save_addr) { |
michael@0 | 437 | /* We are pretty short on buffers. */ |
michael@0 | 438 | errno = EINVAL; /* OUT OF BUFFERS */ |
michael@0 | 439 | return (-1); |
michael@0 | 440 | } |
michael@0 | 441 | memmove(hashp->tmp_buf, (save_p->page) + off, len); |
michael@0 | 442 | val->data = (uint8 *)hashp->tmp_buf; |
michael@0 | 443 | return (0); |
michael@0 | 444 | } |
michael@0 | 445 | |
michael@0 | 446 | |
michael@0 | 447 | /* |
michael@0 | 448 | * Count how big the total datasize is by looping through the pages. Then |
michael@0 | 449 | * allocate a buffer and copy the data in the second loop. NOTE: Our caller |
michael@0 | 450 | * may already have a bp which it is holding onto. The caller is |
michael@0 | 451 | * responsible for copying that bp into our temp buffer. 'len' is how much |
michael@0 | 452 | * space to reserve for that buffer. |
michael@0 | 453 | */ |
michael@0 | 454 | static int |
michael@0 | 455 | collect_data( |
michael@0 | 456 | HTAB *hashp, |
michael@0 | 457 | BUFHEAD *bufp, |
michael@0 | 458 | int len, int set) |
michael@0 | 459 | { |
michael@0 | 460 | register uint16 *bp; |
michael@0 | 461 | BUFHEAD *save_bufp; |
michael@0 | 462 | int save_flags; |
michael@0 | 463 | int mylen, totlen; |
michael@0 | 464 | |
michael@0 | 465 | /* |
michael@0 | 466 | * save the input buf head because we need to walk the list twice. |
michael@0 | 467 | * pin it to make sure it doesn't leave the buffer pool. |
michael@0 | 468 | * This has the effect of growing the buffer pool if necessary. |
michael@0 | 469 | */ |
michael@0 | 470 | save_bufp = bufp; |
michael@0 | 471 | save_flags = save_bufp->flags; |
michael@0 | 472 | save_bufp->flags |= BUF_PIN; |
michael@0 | 473 | |
michael@0 | 474 | /* read the length of the buffer */ |
michael@0 | 475 | for (totlen = len; bufp ; bufp = __get_buf(hashp, bp[bp[0]-1], bufp, 0)) { |
michael@0 | 476 | bp = (uint16 *)bufp->page; |
michael@0 | 477 | mylen = hashp->BSIZE - bp[1]; |
michael@0 | 478 | |
michael@0 | 479 | /* if mylen ever goes negative it means that the |
michael@0 | 480 | * page is screwed up. |
michael@0 | 481 | */ |
michael@0 | 482 | if (mylen < 0) { |
michael@0 | 483 | save_bufp->flags = save_flags; |
michael@0 | 484 | return (-1); |
michael@0 | 485 | } |
michael@0 | 486 | totlen += mylen; |
michael@0 | 487 | if (bp[2] == FULL_KEY_DATA) { /* End of Data */ |
michael@0 | 488 | break; |
michael@0 | 489 | } |
michael@0 | 490 | } |
michael@0 | 491 | |
michael@0 | 492 | if (!bufp) { |
michael@0 | 493 | save_bufp->flags = save_flags; |
michael@0 | 494 | return (-1); |
michael@0 | 495 | } |
michael@0 | 496 | |
michael@0 | 497 | /* allocate a temp buf */ |
michael@0 | 498 | if (hashp->tmp_buf) |
michael@0 | 499 | free(hashp->tmp_buf); |
michael@0 | 500 | if ((hashp->tmp_buf = (char *)malloc((size_t)totlen)) == NULL) { |
michael@0 | 501 | save_bufp->flags = save_flags; |
michael@0 | 502 | return (-1); |
michael@0 | 503 | } |
michael@0 | 504 | |
michael@0 | 505 | /* copy the buffers back into temp buf */ |
michael@0 | 506 | for (bufp = save_bufp; bufp ; |
michael@0 | 507 | bufp = __get_buf(hashp, bp[bp[0]-1], bufp, 0)) { |
michael@0 | 508 | bp = (uint16 *)bufp->page; |
michael@0 | 509 | mylen = hashp->BSIZE - bp[1]; |
michael@0 | 510 | memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], (size_t)mylen); |
michael@0 | 511 | len += mylen; |
michael@0 | 512 | if (bp[2] == FULL_KEY_DATA) { |
michael@0 | 513 | break; |
michael@0 | 514 | } |
michael@0 | 515 | } |
michael@0 | 516 | |
michael@0 | 517 | /* 'clear' the pin flags */ |
michael@0 | 518 | save_bufp->flags = save_flags; |
michael@0 | 519 | |
michael@0 | 520 | /* update the database cursor */ |
michael@0 | 521 | if (set) { |
michael@0 | 522 | hashp->cndx = 1; |
michael@0 | 523 | if (bp[0] == 2) { /* No more buckets in chain */ |
michael@0 | 524 | hashp->cpage = NULL; |
michael@0 | 525 | hashp->cbucket++; |
michael@0 | 526 | } else { |
michael@0 | 527 | hashp->cpage = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); |
michael@0 | 528 | if (!hashp->cpage) |
michael@0 | 529 | return (-1); |
michael@0 | 530 | else if (!((uint16 *)hashp->cpage->page)[0]) { |
michael@0 | 531 | hashp->cbucket++; |
michael@0 | 532 | hashp->cpage = NULL; |
michael@0 | 533 | } |
michael@0 | 534 | } |
michael@0 | 535 | } |
michael@0 | 536 | return (totlen); |
michael@0 | 537 | } |
michael@0 | 538 | |
michael@0 | 539 | /* |
michael@0 | 540 | * Fill in the key and data for this big pair. |
michael@0 | 541 | */ |
michael@0 | 542 | extern int |
michael@0 | 543 | __big_keydata( |
michael@0 | 544 | HTAB *hashp, |
michael@0 | 545 | BUFHEAD *bufp, |
michael@0 | 546 | DBT *key, DBT *val, |
michael@0 | 547 | int set) |
michael@0 | 548 | { |
michael@0 | 549 | key->size = collect_key(hashp, bufp, 0, val, set); |
michael@0 | 550 | if (key->size == (size_t)-1) |
michael@0 | 551 | return (-1); |
michael@0 | 552 | key->data = (uint8 *)hashp->tmp_key; |
michael@0 | 553 | return (0); |
michael@0 | 554 | } |
michael@0 | 555 | |
michael@0 | 556 | /* |
michael@0 | 557 | * Count how big the total key size is by recursing through the pages. Then |
michael@0 | 558 | * collect the data, allocate a buffer and copy the key as you recurse up. |
michael@0 | 559 | */ |
michael@0 | 560 | static int |
michael@0 | 561 | collect_key( |
michael@0 | 562 | HTAB *hashp, |
michael@0 | 563 | BUFHEAD *bufp, |
michael@0 | 564 | int len, |
michael@0 | 565 | DBT *val, |
michael@0 | 566 | int set) |
michael@0 | 567 | { |
michael@0 | 568 | BUFHEAD *xbp; |
michael@0 | 569 | char *p; |
michael@0 | 570 | int mylen, totlen; |
michael@0 | 571 | uint16 *bp, save_addr; |
michael@0 | 572 | |
michael@0 | 573 | p = bufp->page; |
michael@0 | 574 | bp = (uint16 *)p; |
michael@0 | 575 | mylen = hashp->BSIZE - bp[1]; |
michael@0 | 576 | |
michael@0 | 577 | save_addr = bufp->addr; |
michael@0 | 578 | totlen = len + mylen; |
michael@0 | 579 | if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) { /* End of Key. */ |
michael@0 | 580 | if (hashp->tmp_key != NULL) |
michael@0 | 581 | free(hashp->tmp_key); |
michael@0 | 582 | if ((hashp->tmp_key = (char *)malloc((size_t)totlen)) == NULL) |
michael@0 | 583 | return (-1); |
michael@0 | 584 | if (__big_return(hashp, bufp, 1, val, set)) |
michael@0 | 585 | return (-1); |
michael@0 | 586 | } else { |
michael@0 | 587 | xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); |
michael@0 | 588 | if (!xbp || ((totlen = |
michael@0 | 589 | collect_key(hashp, xbp, totlen, val, set)) < 1)) |
michael@0 | 590 | return (-1); |
michael@0 | 591 | } |
michael@0 | 592 | if (bufp->addr != save_addr) { |
michael@0 | 593 | errno = EINVAL; /* MIS -- OUT OF BUFFERS */ |
michael@0 | 594 | return (-1); |
michael@0 | 595 | } |
michael@0 | 596 | memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], (size_t)mylen); |
michael@0 | 597 | return (totlen); |
michael@0 | 598 | } |
michael@0 | 599 | |
michael@0 | 600 | /* |
michael@0 | 601 | * Returns: |
michael@0 | 602 | * 0 => OK |
michael@0 | 603 | * -1 => error |
michael@0 | 604 | */ |
michael@0 | 605 | extern int |
michael@0 | 606 | __big_split( |
michael@0 | 607 | HTAB *hashp, |
michael@0 | 608 | BUFHEAD *op, /* Pointer to where to put keys that go in old bucket */ |
michael@0 | 609 | BUFHEAD *np, /* Pointer to new bucket page */ |
michael@0 | 610 | /* Pointer to first page containing the big key/data */ |
michael@0 | 611 | BUFHEAD *big_keyp, |
michael@0 | 612 | uint32 addr, /* Address of big_keyp */ |
michael@0 | 613 | uint32 obucket,/* Old Bucket */ |
michael@0 | 614 | SPLIT_RETURN *ret) |
michael@0 | 615 | { |
michael@0 | 616 | register BUFHEAD *tmpp; |
michael@0 | 617 | register uint16 *tp; |
michael@0 | 618 | BUFHEAD *bp; |
michael@0 | 619 | DBT key, val; |
michael@0 | 620 | uint32 change; |
michael@0 | 621 | uint16 free_space, n, off; |
michael@0 | 622 | |
michael@0 | 623 | bp = big_keyp; |
michael@0 | 624 | |
michael@0 | 625 | /* Now figure out where the big key/data goes */ |
michael@0 | 626 | if (__big_keydata(hashp, big_keyp, &key, &val, 0)) |
michael@0 | 627 | return (-1); |
michael@0 | 628 | change = (__call_hash(hashp,(char*) key.data, key.size) != obucket); |
michael@0 | 629 | |
michael@0 | 630 | if ((ret->next_addr = __find_last_page(hashp, &big_keyp))) { |
michael@0 | 631 | if (!(ret->nextp = |
michael@0 | 632 | __get_buf(hashp, ret->next_addr, big_keyp, 0))) |
michael@0 | 633 | return (-1);; |
michael@0 | 634 | } else |
michael@0 | 635 | ret->nextp = NULL; |
michael@0 | 636 | |
michael@0 | 637 | /* Now make one of np/op point to the big key/data pair */ |
michael@0 | 638 | #ifdef DEBUG |
michael@0 | 639 | assert(np->ovfl == NULL); |
michael@0 | 640 | #endif |
michael@0 | 641 | if (change) |
michael@0 | 642 | tmpp = np; |
michael@0 | 643 | else |
michael@0 | 644 | tmpp = op; |
michael@0 | 645 | |
michael@0 | 646 | tmpp->flags |= BUF_MOD; |
michael@0 | 647 | #ifdef DEBUG1 |
michael@0 | 648 | (void)fprintf(stderr, |
michael@0 | 649 | "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr, |
michael@0 | 650 | (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0)); |
michael@0 | 651 | #endif |
michael@0 | 652 | tmpp->ovfl = bp; /* one of op/np point to big_keyp */ |
michael@0 | 653 | tp = (uint16 *)tmpp->page; |
michael@0 | 654 | |
michael@0 | 655 | |
michael@0 | 656 | #if 0 /* this get's tripped on database corrupted error */ |
michael@0 | 657 | assert(FREESPACE(tp) >= OVFLSIZE); |
michael@0 | 658 | #endif |
michael@0 | 659 | if(FREESPACE(tp) < OVFLSIZE) |
michael@0 | 660 | return(DATABASE_CORRUPTED_ERROR); |
michael@0 | 661 | |
michael@0 | 662 | n = tp[0]; |
michael@0 | 663 | off = OFFSET(tp); |
michael@0 | 664 | free_space = FREESPACE(tp); |
michael@0 | 665 | tp[++n] = (uint16)addr; |
michael@0 | 666 | tp[++n] = OVFLPAGE; |
michael@0 | 667 | tp[0] = n; |
michael@0 | 668 | OFFSET(tp) = off; |
michael@0 | 669 | FREESPACE(tp) = free_space - OVFLSIZE; |
michael@0 | 670 | |
michael@0 | 671 | /* |
michael@0 | 672 | * Finally, set the new and old return values. BIG_KEYP contains a |
michael@0 | 673 | * pointer to the last page of the big key_data pair. Make sure that |
michael@0 | 674 | * big_keyp has no following page (2 elements) or create an empty |
michael@0 | 675 | * following page. |
michael@0 | 676 | */ |
michael@0 | 677 | |
michael@0 | 678 | ret->newp = np; |
michael@0 | 679 | ret->oldp = op; |
michael@0 | 680 | |
michael@0 | 681 | tp = (uint16 *)big_keyp->page; |
michael@0 | 682 | big_keyp->flags |= BUF_MOD; |
michael@0 | 683 | if (tp[0] > 2) { |
michael@0 | 684 | /* |
michael@0 | 685 | * There may be either one or two offsets on this page. If |
michael@0 | 686 | * there is one, then the overflow page is linked on normally |
michael@0 | 687 | * and tp[4] is OVFLPAGE. If there are two, tp[4] contains |
michael@0 | 688 | * the second offset and needs to get stuffed in after the |
michael@0 | 689 | * next overflow page is added. |
michael@0 | 690 | */ |
michael@0 | 691 | n = tp[4]; |
michael@0 | 692 | free_space = FREESPACE(tp); |
michael@0 | 693 | off = OFFSET(tp); |
michael@0 | 694 | tp[0] -= 2; |
michael@0 | 695 | FREESPACE(tp) = free_space + OVFLSIZE; |
michael@0 | 696 | OFFSET(tp) = off; |
michael@0 | 697 | tmpp = __add_ovflpage(hashp, big_keyp); |
michael@0 | 698 | if (!tmpp) |
michael@0 | 699 | return (-1); |
michael@0 | 700 | tp[4] = n; |
michael@0 | 701 | } else |
michael@0 | 702 | tmpp = big_keyp; |
michael@0 | 703 | |
michael@0 | 704 | if (change) |
michael@0 | 705 | ret->newp = tmpp; |
michael@0 | 706 | else |
michael@0 | 707 | ret->oldp = tmpp; |
michael@0 | 708 | return (0); |
michael@0 | 709 | } |