1 /*-
2 * See the file LICENSE for redistribution information.
3 *
4 * Copyright (c) 1996, 1997, 1998
5 * Sleepycat Software. All rights reserved.
6 */
7
8 #include "config.h"
9
10 #ifndef lint
11 static const char sccsid[] = "@(#)db_dup.c 10.35 (Sleepycat) 12/2/98";
12 #endif /* not lint */
13
14 #ifndef NO_SYSTEM_INCLUDES
15 #include <sys/types.h>
16
17 #include <errno.h>
18 #include <string.h>
19 #endif
20
21 #include "db_int.h"
22 #include "db_page.h"
23 #include "btree.h"
24 #include "db_am.h"
25
26 static int __db_addpage __P((DBC *,
27 PAGE **, db_indx_t *, int (*)(DBC *, u_int32_t, PAGE **)));
28 static int __db_dsplit __P((DBC *,
29 PAGE **, db_indx_t *, u_int32_t, int (*)(DBC *, u_int32_t, PAGE **)));
30
31 /*
32 * __db_dput --
33 * Put a duplicate item onto a duplicate page at the given index.
34 *
35 * PUBLIC: int __db_dput __P((DBC *, DBT *,
36 * PUBLIC: PAGE **, db_indx_t *, int (*)(DBC *, u_int32_t, PAGE **)));
37 */
38 int
__db_dput(dbc,dbt,pp,indxp,newfunc)39 __db_dput(dbc, dbt, pp, indxp, newfunc)
40 DBC *dbc;
41 DBT *dbt;
42 PAGE **pp;
43 db_indx_t *indxp;
44 int (*newfunc) __P((DBC *, u_int32_t, PAGE **));
45 {
46 BOVERFLOW bo;
47 DBT *data_dbtp, hdr_dbt, *hdr_dbtp;
48 PAGE *pagep;
49 db_indx_t size, isize;
50 db_pgno_t pgno;
51 int ret;
52
53 /*
54 * We need some access method independent threshold for when we put
55 * a duplicate item onto an overflow page.
56 */
57 if (dbt->size > 0.25 * dbc->dbp->pgsize) {
58 if ((ret = __db_poff(dbc, dbt, &pgno, newfunc)) != 0)
59 return (ret);
60 UMRW(bo.unused1);
61 B_TSET(bo.type, B_OVERFLOW, 0);
62 UMRW(bo.unused2);
63 bo.tlen = dbt->size;
64 bo.pgno = pgno;
65 hdr_dbt.data = &bo;
66 hdr_dbt.size = isize = BOVERFLOW_SIZE;
67 hdr_dbtp = &hdr_dbt;
68 size = BOVERFLOW_PSIZE;
69 data_dbtp = NULL;
70 } else {
71 size = BKEYDATA_PSIZE(dbt->size);
72 isize = BKEYDATA_SIZE(dbt->size);
73 hdr_dbtp = NULL;
74 data_dbtp = dbt;
75 }
76
77 pagep = *pp;
78 if (size > P_FREESPACE(pagep)) {
79 if (*indxp == NUM_ENT(*pp) && NEXT_PGNO(*pp) == PGNO_INVALID)
80 ret = __db_addpage(dbc, pp, indxp, newfunc);
81 else
82 ret = __db_dsplit(dbc, pp, indxp, isize, newfunc);
83 if (ret != 0)
84 /*
85 * XXX
86 * Pages not returned to free list.
87 */
88 return (ret);
89 pagep = *pp;
90 }
91
92 /*
93 * Now, pagep references the page on which to insert and indx is the
94 * the location to insert.
95 */
96 if ((ret = __db_pitem(dbc,
97 pagep, (u_int32_t)*indxp, isize, hdr_dbtp, data_dbtp)) != 0)
98 return (ret);
99
100 (void)memp_fset(dbc->dbp->mpf, pagep, DB_MPOOL_DIRTY);
101 return (0);
102 }
103
104 /*
105 * __db_drem --
106 * Remove a duplicate at the given index on the given page.
107 *
108 * PUBLIC: int __db_drem __P((DBC *,
109 * PUBLIC: PAGE **, u_int32_t, int (*)(DBC *, PAGE *)));
110 */
111 int
__db_drem(dbc,pp,indx,freefunc)112 __db_drem(dbc, pp, indx, freefunc)
113 DBC *dbc;
114 PAGE **pp;
115 u_int32_t indx;
116 int (*freefunc) __P((DBC *, PAGE *));
117 {
118 PAGE *pagep;
119 int ret;
120
121 pagep = *pp;
122
123 /* Check if we are freeing a big item. */
124 if (B_TYPE(GET_BKEYDATA(pagep, indx)->type) == B_OVERFLOW) {
125 if ((ret = __db_doff(dbc,
126 GET_BOVERFLOW(pagep, indx)->pgno, freefunc)) != 0)
127 return (ret);
128 ret = __db_ditem(dbc, pagep, indx, BOVERFLOW_SIZE);
129 } else
130 ret = __db_ditem(dbc, pagep, indx,
131 BKEYDATA_SIZE(GET_BKEYDATA(pagep, indx)->len));
132 if (ret != 0)
133 return (ret);
134
135 if (NUM_ENT(pagep) == 0) {
136 /*
137 * If the page is emptied, then the page is freed and the pp
138 * parameter is set to reference the next, locked page in the
139 * duplicate chain, if one exists. If there was no such page,
140 * then it is set to NULL.
141 *
142 * !!!
143 * __db_relink will set the dirty bit for us.
144 */
145 if ((ret = __db_relink(dbc, DB_REM_PAGE, pagep, pp, 0)) != 0)
146 return (ret);
147 if ((ret = freefunc(dbc, pagep)) != 0)
148 return (ret);
149 } else
150 (void)memp_fset(dbc->dbp->mpf, pagep, DB_MPOOL_DIRTY);
151
152 return (0);
153 }
154
155 /*
156 * __db_dend --
157 * Find the last page in a set of offpage duplicates.
158 *
159 * PUBLIC: int __db_dend __P((DBC *, db_pgno_t, PAGE **));
160 */
161 int
__db_dend(dbc,pgno,pp)162 __db_dend(dbc, pgno, pp)
163 DBC *dbc;
164 db_pgno_t pgno;
165 PAGE **pp;
166 {
167 DB *dbp;
168 PAGE *h;
169 int ret;
170
171 dbp = dbc->dbp;
172
173 /*
174 * This implements DB_KEYLAST. The last page is returned in pp; pgno
175 * should be the page number of the first page of the duplicate chain.
176 *
177 * *pp may be non-NULL -- if given a valid page use it.
178 */
179 if (*pp != NULL)
180 goto started;
181 for (;;) {
182 if ((ret = memp_fget(dbp->mpf, &pgno, 0, pp)) != 0) {
183 (void)__db_pgerr(dbp, pgno);
184 return (ret);
185 }
186 started: h = *pp;
187
188 if ((pgno = NEXT_PGNO(h)) == PGNO_INVALID)
189 break;
190
191 if ((ret = memp_fput(dbp->mpf, h, 0)) != 0)
192 return (ret);
193 }
194 return (0);
195 }
196
197 /*
198 * __db_dsplit --
199 * Split a page of duplicates, calculating the split point based
200 * on an element of size "size" being added at "*indxp".
201 * On entry hp contains a pointer to the page-pointer of the original
202 * page. On exit, it returns a pointer to the page containing "*indxp"
203 * and "indxp" has been modified to reflect the index on the new page
204 * where the element should be added. The function returns with
205 * the page on which the insert should happen, not yet put.
206 */
207 static int
__db_dsplit(dbc,hp,indxp,size,newfunc)208 __db_dsplit(dbc, hp, indxp, size, newfunc)
209 DBC *dbc;
210 PAGE **hp;
211 db_indx_t *indxp;
212 u_int32_t size;
213 int (*newfunc) __P((DBC *, u_int32_t, PAGE **));
214 {
215 PAGE *h, *np, *tp;
216 BKEYDATA *bk;
217 DBT page_dbt;
218 DB *dbp;
219 size_t pgsize;
220 db_indx_t halfbytes, i, indx, lastsum, nindex, oindex, s, sum;
221 int did_indx, ret, t_ret;
222
223 h = *hp;
224 indx = *indxp;
225 ret = 0;
226 dbp = dbc->dbp;
227 pgsize = dbp->pgsize;
228
229 /* Create a temporary page to do compaction onto. */
230 if ((ret = __os_malloc(pgsize, NULL, &tp)) != 0)
231 return (ret);
232
233 /* Create new page for the split. */
234 if ((ret = newfunc(dbc, P_DUPLICATE, &np)) != 0) {
235 __os_free(tp, pgsize);
236 return (ret);
237 }
238
239 P_INIT(np, pgsize, PGNO(np), PGNO(h), NEXT_PGNO(h), 0,
240 P_DUPLICATE);
241 P_INIT(tp, pgsize, PGNO(h), PREV_PGNO(h), PGNO(np), 0,
242 P_DUPLICATE);
243
244 /* Figure out the split point */
245 halfbytes = (pgsize - HOFFSET(h)) / 2;
246 did_indx = 0;
247 for (sum = 0, lastsum = 0, i = 0; i < NUM_ENT(h); i++) {
248 if (i == indx) {
249 sum += size;
250 did_indx = 1;
251 if (lastsum < halfbytes && sum >= halfbytes) {
252 /* We've crossed the halfway point. */
253 if ((db_indx_t)(halfbytes - lastsum) <
254 (db_indx_t)(sum - halfbytes)) {
255 *hp = np;
256 *indxp = 0;
257 } else
258 *indxp = i;
259 break;
260 }
261 *indxp = i;
262 lastsum = sum;
263 }
264 if (B_TYPE(GET_BKEYDATA(h, i)->type) == B_KEYDATA)
265 sum += BKEYDATA_SIZE(GET_BKEYDATA(h, i)->len);
266 else
267 sum += BOVERFLOW_SIZE;
268
269 if (lastsum < halfbytes && sum >= halfbytes) {
270 /* We've crossed the halfway point. */
271 if ((db_indx_t)(sum - halfbytes) <
272 (db_indx_t)(halfbytes - lastsum))
273 i++;
274 break;
275 }
276 }
277 /*
278 * Check if we have set the return values of the index pointer and
279 * page pointer.
280 */
281 if (!did_indx) {
282 *hp = np;
283 *indxp = indx - i;
284 }
285
286 if (DB_LOGGING(dbc)) {
287 page_dbt.size = dbp->pgsize;
288 page_dbt.data = h;
289 if ((ret = __db_split_log(dbp->dbenv->lg_info,
290 dbc->txn, &LSN(h), 0, DB_SPLITOLD, dbp->log_fileid,
291 PGNO(h), &page_dbt, &LSN(h))) != 0) {
292 __os_free(tp, pgsize);
293 return (ret);
294 }
295 LSN(tp) = LSN(h);
296 }
297
298 /*
299 * If it's a btree, adjust the cursors.
300 *
301 * i is the index of the first element to move onto the new page.
302 */
303 if (dbp->type == DB_BTREE)
304 __bam_ca_split(dbp, PGNO(h), PGNO(h), PGNO(np), i, 0);
305
306 for (nindex = 0, oindex = i; oindex < NUM_ENT(h); oindex++) {
307 bk = GET_BKEYDATA(h, oindex);
308 if (B_TYPE(bk->type) == B_KEYDATA)
309 s = BKEYDATA_SIZE(bk->len);
310 else
311 s = BOVERFLOW_SIZE;
312
313 np->inp[nindex++] = HOFFSET(np) -= s;
314 memcpy((u_int8_t *)np + HOFFSET(np), bk, s);
315 NUM_ENT(np)++;
316 }
317
318 /*
319 * Now do data compaction by copying the remaining stuff onto the
320 * temporary page and then copying it back to the real page.
321 */
322 for (nindex = 0, oindex = 0; oindex < i; oindex++) {
323 bk = GET_BKEYDATA(h, oindex);
324 if (B_TYPE(bk->type) == B_KEYDATA)
325 s = BKEYDATA_SIZE(bk->len);
326 else
327 s = BOVERFLOW_SIZE;
328
329 tp->inp[nindex++] = HOFFSET(tp) -= s;
330 memcpy((u_int8_t *)tp + HOFFSET(tp), bk, s);
331 NUM_ENT(tp)++;
332 }
333
334 /*
335 * This page (the temporary) should be only half full, so we do two
336 * memcpy's, one for the top of the page and one for the bottom of
337 * the page. This way we avoid copying the middle which should be
338 * about half a page.
339 */
340 memcpy(h, tp, LOFFSET(tp));
341 memcpy((u_int8_t *)h + HOFFSET(tp),
342 (u_int8_t *)tp + HOFFSET(tp), pgsize - HOFFSET(tp));
343 __os_free(tp, pgsize);
344
345 if (DB_LOGGING(dbc)) {
346 /*
347 * XXX
348 * If either of these fails, are we leaving pages pinned?
349 * Yes, but it seems like this happens in error case.
350 */
351 page_dbt.size = pgsize;
352 page_dbt.data = h;
353 if ((ret = __db_split_log(dbp->dbenv->lg_info,
354 dbc->txn, &LSN(h), 0, DB_SPLITNEW, dbp->log_fileid,
355 PGNO(h), &page_dbt, &LSN(h))) != 0)
356 return (ret);
357
358 page_dbt.size = pgsize;
359 page_dbt.data = np;
360 if ((ret = __db_split_log(dbp->dbenv->lg_info,
361 dbc->txn, &LSN(np), 0, DB_SPLITNEW, dbp->log_fileid,
362 PGNO(np), &page_dbt, &LSN(np))) != 0)
363 return (ret);
364 }
365
366 /*
367 * Finally, if there was a next page after the page being
368 * split, fix its prev pointer.
369 */
370 if (np->next_pgno != PGNO_INVALID)
371 ret = __db_relink(dbc, DB_ADD_PAGE, np, NULL, 1);
372
373 /*
374 * Figure out if the location we're interested in is on the new
375 * page, and if so, reset the callers' pointer. Push the other
376 * page back to the store.
377 */
378 if (*hp == h)
379 t_ret = memp_fput(dbp->mpf, np, DB_MPOOL_DIRTY);
380 else
381 t_ret = memp_fput(dbp->mpf, h, DB_MPOOL_DIRTY);
382
383 return (ret != 0 ? ret : t_ret);
384 }
385
386 /*
387 * __db_ditem --
388 * Remove an item from a page.
389 *
390 * PUBLIC: int __db_ditem __P((DBC *, PAGE *, u_int32_t, u_int32_t));
391 */
392 int
__db_ditem(dbc,pagep,indx,nbytes)393 __db_ditem(dbc, pagep, indx, nbytes)
394 DBC *dbc;
395 PAGE *pagep;
396 u_int32_t indx, nbytes;
397 {
398 DB *dbp;
399 DBT ldbt;
400 db_indx_t cnt, offset;
401 int ret;
402 u_int8_t *from;
403
404 dbp = dbc->dbp;
405 if (DB_LOGGING(dbc)) {
406 ldbt.data = P_ENTRY(pagep, indx);
407 ldbt.size = nbytes;
408 if ((ret = __db_addrem_log(dbp->dbenv->lg_info, dbc->txn,
409 &LSN(pagep), 0, DB_REM_DUP, dbp->log_fileid, PGNO(pagep),
410 (u_int32_t)indx, nbytes, &ldbt, NULL, &LSN(pagep))) != 0)
411 return (ret);
412 }
413
414 /*
415 * If there's only a single item on the page, we don't have to
416 * work hard.
417 */
418 if (NUM_ENT(pagep) == 1) {
419 NUM_ENT(pagep) = 0;
420 HOFFSET(pagep) = dbp->pgsize;
421 return (0);
422 }
423
424 /*
425 * Pack the remaining key/data items at the end of the page. Use
426 * memmove(3), the regions may overlap.
427 */
428 from = (u_int8_t *)pagep + HOFFSET(pagep);
429 memmove(from + nbytes, from, pagep->inp[indx] - HOFFSET(pagep));
430 HOFFSET(pagep) += nbytes;
431
432 /* Adjust the indices' offsets. */
433 offset = pagep->inp[indx];
434 for (cnt = 0; cnt < NUM_ENT(pagep); ++cnt)
435 if (pagep->inp[cnt] < offset)
436 pagep->inp[cnt] += nbytes;
437
438 /* Shift the indices down. */
439 --NUM_ENT(pagep);
440 if (indx != NUM_ENT(pagep))
441 memmove(&pagep->inp[indx], &pagep->inp[indx + 1],
442 sizeof(db_indx_t) * (NUM_ENT(pagep) - indx));
443
444 /* If it's a btree, adjust the cursors. */
445 if (dbp->type == DB_BTREE)
446 __bam_ca_di(dbp, PGNO(pagep), indx, -1);
447
448 return (0);
449 }
450
451 /*
452 * __db_pitem --
453 * Put an item on a page.
454 *
455 * PUBLIC: int __db_pitem
456 * PUBLIC: __P((DBC *, PAGE *, u_int32_t, u_int32_t, DBT *, DBT *));
457 */
458 int
__db_pitem(dbc,pagep,indx,nbytes,hdr,data)459 __db_pitem(dbc, pagep, indx, nbytes, hdr, data)
460 DBC *dbc;
461 PAGE *pagep;
462 u_int32_t indx;
463 u_int32_t nbytes;
464 DBT *hdr, *data;
465 {
466 DB *dbp;
467 BKEYDATA bk;
468 DBT thdr;
469 int ret;
470 u_int8_t *p;
471
472 /*
473 * Put a single item onto a page. The logic figuring out where to
474 * insert and whether it fits is handled in the caller. All we do
475 * here is manage the page shuffling. We cheat a little bit in that
476 * we don't want to copy the dbt on a normal put twice. If hdr is
477 * NULL, we create a BKEYDATA structure on the page, otherwise, just
478 * copy the caller's information onto the page.
479 *
480 * This routine is also used to put entries onto the page where the
481 * entry is pre-built, e.g., during recovery. In this case, the hdr
482 * will point to the entry, and the data argument will be NULL.
483 *
484 * !!!
485 * There's a tremendous potential for off-by-one errors here, since
486 * the passed in header sizes must be adjusted for the structure's
487 * placeholder for the trailing variable-length data field.
488 */
489 dbp = dbc->dbp;
490 if (DB_LOGGING(dbc))
491 if ((ret = __db_addrem_log(dbp->dbenv->lg_info, dbc->txn,
492 &LSN(pagep), 0, DB_ADD_DUP, dbp->log_fileid, PGNO(pagep),
493 (u_int32_t)indx, nbytes, hdr, data, &LSN(pagep))) != 0)
494 return (ret);
495
496 if (hdr == NULL) {
497 B_TSET(bk.type, B_KEYDATA, 0);
498 bk.len = data == NULL ? 0 : data->size;
499
500 thdr.data = &bk;
501 thdr.size = SSZA(BKEYDATA, data);
502 hdr = &thdr;
503 }
504
505 /* Adjust the index table, then put the item on the page. */
506 if (indx != NUM_ENT(pagep))
507 memmove(&pagep->inp[indx + 1], &pagep->inp[indx],
508 sizeof(db_indx_t) * (NUM_ENT(pagep) - indx));
509 HOFFSET(pagep) -= nbytes;
510 pagep->inp[indx] = HOFFSET(pagep);
511 ++NUM_ENT(pagep);
512
513 p = P_ENTRY(pagep, indx);
514 memcpy(p, hdr->data, hdr->size);
515 if (data != NULL)
516 memcpy(p + hdr->size, data->data, data->size);
517
518 /* If it's a btree, adjust the cursors. */
519 if (dbp->type == DB_BTREE)
520 __bam_ca_di(dbp, PGNO(pagep), indx, 1);
521
522 return (0);
523 }
524
525 /*
526 * __db_relink --
527 * Relink around a deleted page.
528 *
529 * PUBLIC: int __db_relink __P((DBC *, u_int32_t, PAGE *, PAGE **, int));
530 */
531 int
__db_relink(dbc,add_rem,pagep,new_next,needlock)532 __db_relink(dbc, add_rem, pagep, new_next, needlock)
533 DBC *dbc;
534 u_int32_t add_rem;
535 PAGE *pagep, **new_next;
536 int needlock;
537 {
538 DB *dbp;
539 PAGE *np, *pp;
540 DB_LOCK npl, ppl;
541 DB_LSN *nlsnp, *plsnp;
542 int ret;
543
544 ret = 0;
545 np = pp = NULL;
546 npl = ppl = LOCK_INVALID;
547 nlsnp = plsnp = NULL;
548 dbp = dbc->dbp;
549
550 /*
551 * Retrieve and lock the one/two pages. For a remove, we may need
552 * two pages (the before and after). For an add, we only need one
553 * because, the split took care of the prev.
554 */
555 if (pagep->next_pgno != PGNO_INVALID) {
556 if (needlock && (ret = __bam_lget(dbc,
557 0, pagep->next_pgno, DB_LOCK_WRITE, &npl)) != 0)
558 goto err;
559 if ((ret = memp_fget(dbp->mpf,
560 &pagep->next_pgno, 0, &np)) != 0) {
561 (void)__db_pgerr(dbp, pagep->next_pgno);
562 goto err;
563 }
564 nlsnp = &np->lsn;
565 }
566 if (add_rem == DB_REM_PAGE && pagep->prev_pgno != PGNO_INVALID) {
567 if (needlock && (ret = __bam_lget(dbc,
568 0, pagep->prev_pgno, DB_LOCK_WRITE, &ppl)) != 0)
569 goto err;
570 if ((ret = memp_fget(dbp->mpf,
571 &pagep->prev_pgno, 0, &pp)) != 0) {
572 (void)__db_pgerr(dbp, pagep->next_pgno);
573 goto err;
574 }
575 plsnp = &pp->lsn;
576 }
577
578 /* Log the change. */
579 if (DB_LOGGING(dbc)) {
580 if ((ret = __db_relink_log(dbp->dbenv->lg_info, dbc->txn,
581 &pagep->lsn, 0, add_rem, dbp->log_fileid,
582 pagep->pgno, &pagep->lsn,
583 pagep->prev_pgno, plsnp, pagep->next_pgno, nlsnp)) != 0)
584 goto err;
585 if (np != NULL)
586 np->lsn = pagep->lsn;
587 if (pp != NULL)
588 pp->lsn = pagep->lsn;
589 }
590
591 /*
592 * Modify and release the two pages.
593 *
594 * !!!
595 * The parameter new_next gets set to the page following the page we
596 * are removing. If there is no following page, then new_next gets
597 * set to NULL.
598 */
599 if (np != NULL) {
600 if (add_rem == DB_ADD_PAGE)
601 np->prev_pgno = pagep->pgno;
602 else
603 np->prev_pgno = pagep->prev_pgno;
604 if (new_next == NULL)
605 ret = memp_fput(dbp->mpf, np, DB_MPOOL_DIRTY);
606 else {
607 *new_next = np;
608 ret = memp_fset(dbp->mpf, np, DB_MPOOL_DIRTY);
609 }
610 if (ret != 0)
611 goto err;
612 if (needlock)
613 (void)__bam_lput(dbc, npl);
614 } else if (new_next != NULL)
615 *new_next = NULL;
616
617 if (pp != NULL) {
618 pp->next_pgno = pagep->next_pgno;
619 if ((ret = memp_fput(dbp->mpf, pp, DB_MPOOL_DIRTY)) != 0)
620 goto err;
621 if (needlock)
622 (void)__bam_lput(dbc, ppl);
623 }
624 return (0);
625
626 err: if (np != NULL)
627 (void)memp_fput(dbp->mpf, np, 0);
628 if (needlock && npl != LOCK_INVALID)
629 (void)__bam_lput(dbc, npl);
630 if (pp != NULL)
631 (void)memp_fput(dbp->mpf, pp, 0);
632 if (needlock && ppl != LOCK_INVALID)
633 (void)__bam_lput(dbc, ppl);
634 return (ret);
635 }
636
637 /*
638 * __db_ddup --
639 * Delete an offpage chain of duplicates.
640 *
641 * PUBLIC: int __db_ddup __P((DBC *, db_pgno_t, int (*)(DBC *, PAGE *)));
642 */
643 int
__db_ddup(dbc,pgno,freefunc)644 __db_ddup(dbc, pgno, freefunc)
645 DBC *dbc;
646 db_pgno_t pgno;
647 int (*freefunc) __P((DBC *, PAGE *));
648 {
649 DB *dbp;
650 PAGE *pagep;
651 DBT tmp_dbt;
652 int ret;
653
654 dbp = dbc->dbp;
655 do {
656 if ((ret = memp_fget(dbp->mpf, &pgno, 0, &pagep)) != 0) {
657 (void)__db_pgerr(dbp, pgno);
658 return (ret);
659 }
660
661 if (DB_LOGGING(dbc)) {
662 tmp_dbt.data = pagep;
663 tmp_dbt.size = dbp->pgsize;
664 if ((ret = __db_split_log(dbp->dbenv->lg_info,
665 dbc->txn, &LSN(pagep), 0, DB_SPLITOLD,
666 dbp->log_fileid, PGNO(pagep), &tmp_dbt,
667 &LSN(pagep))) != 0)
668 return (ret);
669 }
670 pgno = pagep->next_pgno;
671 if ((ret = freefunc(dbc, pagep)) != 0)
672 return (ret);
673 } while (pgno != PGNO_INVALID);
674
675 return (0);
676 }
677
678 /*
679 * __db_addpage --
680 * Create a new page and link it onto the next_pgno field of the
681 * current page.
682 */
683 static int
__db_addpage(dbc,hp,indxp,newfunc)684 __db_addpage(dbc, hp, indxp, newfunc)
685 DBC *dbc;
686 PAGE **hp;
687 db_indx_t *indxp;
688 int (*newfunc) __P((DBC *, u_int32_t, PAGE **));
689 {
690 DB *dbp;
691 PAGE *newpage;
692 int ret;
693
694 dbp = dbc->dbp;
695 if ((ret = newfunc(dbc, P_DUPLICATE, &newpage)) != 0)
696 return (ret);
697
698 if (DB_LOGGING(dbc)) {
699 if ((ret = __db_addpage_log(dbp->dbenv->lg_info,
700 dbc->txn, &LSN(*hp), 0, dbp->log_fileid,
701 PGNO(*hp), &LSN(*hp), PGNO(newpage), &LSN(newpage))) != 0) {
702 return (ret);
703 }
704 LSN(newpage) = LSN(*hp);
705 }
706
707 PREV_PGNO(newpage) = PGNO(*hp);
708 NEXT_PGNO(*hp) = PGNO(newpage);
709
710 if ((ret = memp_fput(dbp->mpf, *hp, DB_MPOOL_DIRTY)) != 0)
711 return (ret);
712 *hp = newpage;
713 *indxp = 0;
714 return (0);
715 }
716
717 /*
718 * __db_dsearch --
719 * Search a set of duplicates for the proper position for a new duplicate.
720 *
721 * + pgno is the page number of the page on which to begin searching.
722 * Since we can continue duplicate searches, it might not be the first
723 * page.
724 *
725 * + If we are continuing a search, then *pp may be non-NULL in which
726 * case we do not have to retrieve the page.
727 *
728 * + If we are continuing a search, then *indxp contains the first
729 * on pgno of where we should begin the search.
730 *
731 * NOTE: if there is no comparison function, then continuing is
732 * meaningless, and *pp should always be NULL and *indxp will be
733 * ignored.
734 *
735 * 3 return values::
736 *
737 * + pp is the returned page pointer of where this element should go.
738 * + indxp is the returned index on that page
739 * + cmpp is the returned final comparison result.
740 *
741 * PUBLIC: int __db_dsearch __P((DBC *,
742 * PUBLIC: int, DBT *, db_pgno_t, db_indx_t *, PAGE **, int *));
743 */
744 int
__db_dsearch(dbc,is_insert,dbt,pgno,indxp,pp,cmpp)745 __db_dsearch(dbc, is_insert, dbt, pgno, indxp, pp, cmpp)
746 DBC *dbc;
747 int is_insert, *cmpp;
748 DBT *dbt;
749 db_pgno_t pgno;
750 db_indx_t *indxp;
751 PAGE **pp;
752 {
753 DB *dbp;
754 PAGE *h;
755 db_indx_t base, indx, lim, save_indx;
756 db_pgno_t save_pgno;
757 int ret;
758
759 dbp = dbc->dbp;
760
761 if (dbp->dup_compare == NULL) {
762 /*
763 * We may have been given a valid page, but we may not be
764 * able to use it. The problem is that the application is
765 * doing a join and we're trying to continue the search,
766 * but since the items aren't sorted, we can't. Discard
767 * the page if it's not the one we're going to start with
768 * anyway.
769 */
770 if (*pp != NULL && (*pp)->pgno != pgno) {
771 if ((ret = memp_fput(dbp->mpf, *pp, 0)) != 0)
772 return (ret);
773 *pp = NULL;
774 }
775
776 /*
777 * If no duplicate function is specified, just go to the end
778 * of the duplicate set.
779 */
780 if (is_insert) {
781 if ((ret = __db_dend(dbc, pgno, pp)) != 0)
782 return (ret);
783 *indxp = NUM_ENT(*pp);
784 return (0);
785 }
786
787 /*
788 * We are looking for a specific duplicate, so do a linear
789 * search.
790 */
791 if (*pp != NULL)
792 goto nocmp_started;
793 for (;;) {
794 if ((ret = memp_fget(dbp->mpf, &pgno, 0, pp)) != 0)
795 goto pg_err;
796 nocmp_started: h = *pp;
797
798 for (*indxp = 0; *indxp < NUM_ENT(h); ++*indxp) {
799 if ((*cmpp = __bam_cmp(dbp,
800 dbt, h, *indxp, __bam_defcmp)) != 0)
801 continue;
802 /*
803 * The duplicate may have already been deleted,
804 * if it's a btree page, in which case we skip
805 * it.
806 */
807 if (dbp->type == DB_BTREE &&
808 B_DISSET(GET_BKEYDATA(h, *indxp)->type))
809 continue;
810
811 return (0);
812 }
813
814 if ((pgno = h->next_pgno) == PGNO_INVALID)
815 break;
816
817 if ((ret = memp_fput(dbp->mpf, h, 0)) != 0)
818 return (ret);
819 }
820 *cmpp = 1; /* We didn't succeed... */
821 return (0);
822 }
823
824 /*
825 * We have a comparison routine, i.e., the duplicates are sorted.
826 * Walk through the chain of duplicates, checking the last entry
827 * on each page to decide if it's the page we want to search.
828 *
829 * *pp may be non-NULL -- if we were given a valid page (e.g., are
830 * in mid-search), then use the provided page.
831 */
832 if (*pp != NULL)
833 goto cmp_started;
834 for (;;) {
835 if ((ret = memp_fget(dbp->mpf, &pgno, 0, pp)) != 0)
836 goto pg_err;
837 cmp_started: h = *pp;
838
839 if ((pgno = h->next_pgno) == PGNO_INVALID || __bam_cmp(dbp,
840 dbt, h, h->entries - 1, dbp->dup_compare) <= 0)
841 break;
842 /*
843 * Even when continuing a search, make sure we don't skip
844 * entries on a new page
845 */
846 *indxp = 0;
847
848 if ((ret = memp_fput(dbp->mpf, h, 0)) != 0)
849 return (ret);
850 }
851
852 /* Next, do a binary search on the page. */
853 base = F_ISSET(dbc, DBC_CONTINUE) ? *indxp : 0;
854 for (lim = NUM_ENT(h) - base; lim != 0; lim >>= 1) {
855 indx = base + (lim >> 1);
856 if ((*cmpp = __bam_cmp(dbp,
857 dbt, h, indx, dbp->dup_compare)) == 0) {
858 *indxp = indx;
859
860 if (dbp->type != DB_BTREE ||
861 !B_DISSET(GET_BKEYDATA(h, *indxp)->type))
862 return (0);
863 goto check_delete;
864 }
865 if (*cmpp > 0) {
866 base = indx + 1;
867 lim--;
868 }
869 }
870
871 /*
872 * Base references the smallest index larger than the supplied DBT's
873 * data item, potentially both 0 and NUM_ENT.
874 */
875 *indxp = base;
876 return (0);
877
878 check_delete:
879 /*
880 * The duplicate may have already been deleted, if it's a btree page,
881 * in which case we wander around, hoping to find an entry that hasn't
882 * been deleted. First, wander in a forwardly direction.
883 */
884 save_pgno = (*pp)->pgno;
885 save_indx = *indxp;
886 for (++*indxp;;) {
887 for (; *indxp < NUM_ENT(h); ++*indxp) {
888 if ((*cmpp = __bam_cmp(dbp,
889 dbt, h, *indxp, dbp->dup_compare)) != 0)
890 goto check_delete_rev;
891
892 if (!B_DISSET(GET_BKEYDATA(h, *indxp)->type))
893 return (0);
894 }
895 if ((pgno = h->next_pgno) == PGNO_INVALID)
896 break;
897
898 if ((ret = memp_fput(dbp->mpf, h, 0)) != 0)
899 return (ret);
900
901 if ((ret = memp_fget(dbp->mpf, &pgno, 0, pp)) != 0)
902 goto pg_err;
903 h = *pp;
904
905 *indxp = 0;
906 }
907
908 check_delete_rev:
909 /* Go back to where we started, and wander in a backwardly direction. */
910 if (h->pgno != save_pgno) {
911 if ((ret = memp_fput(dbp->mpf, h, 0)) != 0)
912 return (ret);
913 if ((ret = memp_fget(dbp->mpf, &save_pgno, 0, pp)) != 0)
914 goto pg_err;
915 h = *pp;
916 }
917
918 for (;;) {
919 while (*indxp > 0) {
920 --*indxp;
921 if ((*cmpp = __bam_cmp(dbp,
922 dbt, h, *indxp, dbp->dup_compare)) != 0)
923 goto check_delete_fail;
924
925 if (!B_DISSET(GET_BKEYDATA(h, *indxp)->type))
926 return (0);
927 }
928 if ((pgno = h->prev_pgno) == PGNO_INVALID)
929 break;
930
931 if ((ret = memp_fput(dbp->mpf, h, 0)) != 0)
932 return (ret);
933
934 if ((ret = memp_fget(dbp->mpf, &pgno, 0, pp)) != 0)
935 goto pg_err;
936 h = *pp;
937
938 *indxp = NUM_ENT(h);
939 }
940
941 check_delete_fail:
942 *cmpp = 1; /* We didn't succeed... */
943 return (0);
944
945 pg_err: __db_pgerr(dbp, pgno);
946 return (ret);
947 }
948