1 |
|
|
/* $OpenBSD: bt_split.c,v 1.13 2005/08/05 13:03:00 espie Exp $ */ |
2 |
|
|
|
3 |
|
|
/*- |
4 |
|
|
* Copyright (c) 1990, 1993, 1994 |
5 |
|
|
* The Regents of the University of California. All rights reserved. |
6 |
|
|
* |
7 |
|
|
* This code is derived from software contributed to Berkeley by |
8 |
|
|
* Mike Olson. |
9 |
|
|
* |
10 |
|
|
* Redistribution and use in source and binary forms, with or without |
11 |
|
|
* modification, are permitted provided that the following conditions |
12 |
|
|
* are met: |
13 |
|
|
* 1. Redistributions of source code must retain the above copyright |
14 |
|
|
* notice, this list of conditions and the following disclaimer. |
15 |
|
|
* 2. Redistributions in binary form must reproduce the above copyright |
16 |
|
|
* notice, this list of conditions and the following disclaimer in the |
17 |
|
|
* documentation and/or other materials provided with the distribution. |
18 |
|
|
* 3. Neither the name of the University nor the names of its contributors |
19 |
|
|
* may be used to endorse or promote products derived from this software |
20 |
|
|
* without specific prior written permission. |
21 |
|
|
* |
22 |
|
|
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND |
23 |
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
24 |
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
25 |
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE |
26 |
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
27 |
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
28 |
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
29 |
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
30 |
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
31 |
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
32 |
|
|
* SUCH DAMAGE. |
33 |
|
|
*/ |
34 |
|
|
|
35 |
|
|
#include <sys/types.h> |
36 |
|
|
|
37 |
|
|
#include <limits.h> |
38 |
|
|
#include <stdio.h> |
39 |
|
|
#include <stdlib.h> |
40 |
|
|
#include <string.h> |
41 |
|
|
|
42 |
|
|
#include <db.h> |
43 |
|
|
#include "btree.h" |
44 |
|
|
|
45 |
|
|
static int bt_broot(BTREE *, PAGE *, PAGE *, PAGE *); |
46 |
|
|
static PAGE *bt_page(BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t); |
47 |
|
|
static int bt_preserve(BTREE *, pgno_t); |
48 |
|
|
static PAGE *bt_psplit(BTREE *, PAGE *, PAGE *, PAGE *, indx_t *, size_t); |
49 |
|
|
static PAGE *bt_root(BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t); |
50 |
|
|
static int bt_rroot(BTREE *, PAGE *, PAGE *, PAGE *); |
51 |
|
|
static recno_t rec_total(PAGE *); |
52 |
|
|
|
53 |
|
|
#ifdef STATISTICS |
54 |
|
|
u_long bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved; |
55 |
|
|
#endif |
56 |
|
|
|
57 |
|
|
/* |
58 |
|
|
* __BT_SPLIT -- Split the tree. |
59 |
|
|
* |
60 |
|
|
* Parameters: |
61 |
|
|
* t: tree |
62 |
|
|
* sp: page to split |
63 |
|
|
* key: key to insert |
64 |
|
|
* data: data to insert |
65 |
|
|
* flags: BIGKEY/BIGDATA flags |
66 |
|
|
* ilen: insert length |
67 |
|
|
* skip: index to leave open |
68 |
|
|
* |
69 |
|
|
* Returns: |
70 |
|
|
* RET_ERROR, RET_SUCCESS |
71 |
|
|
*/ |
72 |
|
|
int |
73 |
|
|
__bt_split(BTREE *t, PAGE *sp, const DBT *key, const DBT *data, int flags, |
74 |
|
|
size_t ilen, u_int32_t argskip) |
75 |
|
|
{ |
76 |
|
|
BINTERNAL *bi; |
77 |
|
|
BLEAF *bl, *tbl; |
78 |
|
|
DBT a, b; |
79 |
|
|
EPGNO *parent; |
80 |
|
|
PAGE *h, *l, *r, *lchild, *rchild; |
81 |
|
|
indx_t nxtindex; |
82 |
|
|
u_int16_t skip; |
83 |
|
|
u_int32_t n, nbytes, nksize; |
84 |
|
|
int parentsplit; |
85 |
|
|
char *dest; |
86 |
|
|
|
87 |
|
|
/* |
88 |
|
|
* Split the page into two pages, l and r. The split routines return |
89 |
|
|
* a pointer to the page into which the key should be inserted and with |
90 |
|
|
* skip set to the offset which should be used. Additionally, l and r |
91 |
|
|
* are pinned. |
92 |
|
|
*/ |
93 |
|
|
skip = argskip; |
94 |
|
|
h = sp->pgno == P_ROOT ? |
95 |
|
|
bt_root(t, sp, &l, &r, &skip, ilen) : |
96 |
|
|
bt_page(t, sp, &l, &r, &skip, ilen); |
97 |
|
|
if (h == NULL) |
98 |
|
|
return (RET_ERROR); |
99 |
|
|
|
100 |
|
|
/* |
101 |
|
|
* Insert the new key/data pair into the leaf page. (Key inserts |
102 |
|
|
* always cause a leaf page to split first.) |
103 |
|
|
*/ |
104 |
|
|
h->linp[skip] = h->upper -= ilen; |
105 |
|
|
dest = (char *)h + h->upper; |
106 |
|
|
if (F_ISSET(t, R_RECNO)) |
107 |
|
|
WR_RLEAF(dest, data, flags) |
108 |
|
|
else |
109 |
|
|
WR_BLEAF(dest, key, data, flags) |
110 |
|
|
|
111 |
|
|
/* If the root page was split, make it look right. */ |
112 |
|
|
if (sp->pgno == P_ROOT && |
113 |
|
|
(F_ISSET(t, R_RECNO) ? |
114 |
|
|
bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR) |
115 |
|
|
goto err2; |
116 |
|
|
|
117 |
|
|
/* |
118 |
|
|
* Now we walk the parent page stack -- a LIFO stack of the pages that |
119 |
|
|
* were traversed when we searched for the page that split. Each stack |
120 |
|
|
* entry is a page number and a page index offset. The offset is for |
121 |
|
|
* the page traversed on the search. We've just split a page, so we |
122 |
|
|
* have to insert a new key into the parent page. |
123 |
|
|
* |
124 |
|
|
* If the insert into the parent page causes it to split, may have to |
125 |
|
|
* continue splitting all the way up the tree. We stop if the root |
126 |
|
|
* splits or the page inserted into didn't have to split to hold the |
127 |
|
|
* new key. Some algorithms replace the key for the old page as well |
128 |
|
|
* as the new page. We don't, as there's no reason to believe that the |
129 |
|
|
* first key on the old page is any better than the key we have, and, |
130 |
|
|
* in the case of a key being placed at index 0 causing the split, the |
131 |
|
|
* key is unavailable. |
132 |
|
|
* |
133 |
|
|
* There are a maximum of 5 pages pinned at any time. We keep the left |
134 |
|
|
* and right pages pinned while working on the parent. The 5 are the |
135 |
|
|
* two children, left parent and right parent (when the parent splits) |
136 |
|
|
* and the root page or the overflow key page when calling bt_preserve. |
137 |
|
|
* This code must make sure that all pins are released other than the |
138 |
|
|
* root page or overflow page which is unlocked elsewhere. |
139 |
|
|
*/ |
140 |
|
|
while ((parent = BT_POP(t)) != NULL) { |
141 |
|
|
lchild = l; |
142 |
|
|
rchild = r; |
143 |
|
|
|
144 |
|
|
/* Get the parent page. */ |
145 |
|
|
if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL) |
146 |
|
|
goto err2; |
147 |
|
|
|
148 |
|
|
/* |
149 |
|
|
* The new key goes ONE AFTER the index, because the split |
150 |
|
|
* was to the right. |
151 |
|
|
*/ |
152 |
|
|
skip = parent->index + 1; |
153 |
|
|
|
154 |
|
|
/* |
155 |
|
|
* Calculate the space needed on the parent page. |
156 |
|
|
* |
157 |
|
|
* Prefix trees: space hack when inserting into BINTERNAL |
158 |
|
|
* pages. Retain only what's needed to distinguish between |
159 |
|
|
* the new entry and the LAST entry on the page to its left. |
160 |
|
|
* If the keys compare equal, retain the entire key. Note, |
161 |
|
|
* we don't touch overflow keys, and the entire key must be |
162 |
|
|
* retained for the next-to-left most key on the leftmost |
163 |
|
|
* page of each level, or the search will fail. Applicable |
164 |
|
|
* ONLY to internal pages that have leaf pages as children. |
165 |
|
|
* Further reduction of the key between pairs of internal |
166 |
|
|
* pages loses too much information. |
167 |
|
|
*/ |
168 |
|
|
switch (rchild->flags & P_TYPE) { |
169 |
|
|
case P_BINTERNAL: |
170 |
|
|
bi = GETBINTERNAL(rchild, 0); |
171 |
|
|
nbytes = NBINTERNAL(bi->ksize); |
172 |
|
|
break; |
173 |
|
|
case P_BLEAF: |
174 |
|
|
bl = GETBLEAF(rchild, 0); |
175 |
|
|
nbytes = NBINTERNAL(bl->ksize); |
176 |
|
|
if (t->bt_pfx && !(bl->flags & P_BIGKEY) && |
177 |
|
|
(h->prevpg != P_INVALID || skip > 1)) { |
178 |
|
|
tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1); |
179 |
|
|
a.size = tbl->ksize; |
180 |
|
|
a.data = tbl->bytes; |
181 |
|
|
b.size = bl->ksize; |
182 |
|
|
b.data = bl->bytes; |
183 |
|
|
nksize = t->bt_pfx(&a, &b); |
184 |
|
|
n = NBINTERNAL(nksize); |
185 |
|
|
if (n < nbytes) { |
186 |
|
|
#ifdef STATISTICS |
187 |
|
|
bt_pfxsaved += nbytes - n; |
188 |
|
|
#endif |
189 |
|
|
nbytes = n; |
190 |
|
|
} else |
191 |
|
|
nksize = 0; |
192 |
|
|
} else |
193 |
|
|
nksize = 0; |
194 |
|
|
break; |
195 |
|
|
case P_RINTERNAL: |
196 |
|
|
case P_RLEAF: |
197 |
|
|
nbytes = NRINTERNAL; |
198 |
|
|
break; |
199 |
|
|
default: |
200 |
|
|
abort(); |
201 |
|
|
} |
202 |
|
|
|
203 |
|
|
/* Split the parent page if necessary or shift the indices. */ |
204 |
|
|
if (h->upper - h->lower < nbytes + sizeof(indx_t)) { |
205 |
|
|
sp = h; |
206 |
|
|
h = h->pgno == P_ROOT ? |
207 |
|
|
bt_root(t, h, &l, &r, &skip, nbytes) : |
208 |
|
|
bt_page(t, h, &l, &r, &skip, nbytes); |
209 |
|
|
if (h == NULL) |
210 |
|
|
goto err1; |
211 |
|
|
parentsplit = 1; |
212 |
|
|
} else { |
213 |
|
|
if (skip < (nxtindex = NEXTINDEX(h))) |
214 |
|
|
memmove(h->linp + skip + 1, h->linp + skip, |
215 |
|
|
(nxtindex - skip) * sizeof(indx_t)); |
216 |
|
|
h->lower += sizeof(indx_t); |
217 |
|
|
parentsplit = 0; |
218 |
|
|
} |
219 |
|
|
|
220 |
|
|
/* Insert the key into the parent page. */ |
221 |
|
|
switch (rchild->flags & P_TYPE) { |
222 |
|
|
case P_BINTERNAL: |
223 |
|
|
h->linp[skip] = h->upper -= nbytes; |
224 |
|
|
dest = (char *)h + h->linp[skip]; |
225 |
|
|
memmove(dest, bi, nbytes); |
226 |
|
|
((BINTERNAL *)dest)->pgno = rchild->pgno; |
227 |
|
|
break; |
228 |
|
|
case P_BLEAF: |
229 |
|
|
h->linp[skip] = h->upper -= nbytes; |
230 |
|
|
dest = (char *)h + h->linp[skip]; |
231 |
|
|
WR_BINTERNAL(dest, nksize ? nksize : bl->ksize, |
232 |
|
|
rchild->pgno, bl->flags & P_BIGKEY); |
233 |
|
|
memmove(dest, bl->bytes, nksize ? nksize : bl->ksize); |
234 |
|
|
if (bl->flags & P_BIGKEY && |
235 |
|
|
bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR) |
236 |
|
|
goto err1; |
237 |
|
|
break; |
238 |
|
|
case P_RINTERNAL: |
239 |
|
|
/* |
240 |
|
|
* Update the left page count. If split |
241 |
|
|
* added at index 0, fix the correct page. |
242 |
|
|
*/ |
243 |
|
|
if (skip > 0) |
244 |
|
|
dest = (char *)h + h->linp[skip - 1]; |
245 |
|
|
else |
246 |
|
|
dest = (char *)l + l->linp[NEXTINDEX(l) - 1]; |
247 |
|
|
((RINTERNAL *)dest)->nrecs = rec_total(lchild); |
248 |
|
|
((RINTERNAL *)dest)->pgno = lchild->pgno; |
249 |
|
|
|
250 |
|
|
/* Update the right page count. */ |
251 |
|
|
h->linp[skip] = h->upper -= nbytes; |
252 |
|
|
dest = (char *)h + h->linp[skip]; |
253 |
|
|
((RINTERNAL *)dest)->nrecs = rec_total(rchild); |
254 |
|
|
((RINTERNAL *)dest)->pgno = rchild->pgno; |
255 |
|
|
break; |
256 |
|
|
case P_RLEAF: |
257 |
|
|
/* |
258 |
|
|
* Update the left page count. If split |
259 |
|
|
* added at index 0, fix the correct page. |
260 |
|
|
*/ |
261 |
|
|
if (skip > 0) |
262 |
|
|
dest = (char *)h + h->linp[skip - 1]; |
263 |
|
|
else |
264 |
|
|
dest = (char *)l + l->linp[NEXTINDEX(l) - 1]; |
265 |
|
|
((RINTERNAL *)dest)->nrecs = NEXTINDEX(lchild); |
266 |
|
|
((RINTERNAL *)dest)->pgno = lchild->pgno; |
267 |
|
|
|
268 |
|
|
/* Update the right page count. */ |
269 |
|
|
h->linp[skip] = h->upper -= nbytes; |
270 |
|
|
dest = (char *)h + h->linp[skip]; |
271 |
|
|
((RINTERNAL *)dest)->nrecs = NEXTINDEX(rchild); |
272 |
|
|
((RINTERNAL *)dest)->pgno = rchild->pgno; |
273 |
|
|
break; |
274 |
|
|
default: |
275 |
|
|
abort(); |
276 |
|
|
} |
277 |
|
|
|
278 |
|
|
/* Unpin the held pages. */ |
279 |
|
|
if (!parentsplit) { |
280 |
|
|
mpool_put(t->bt_mp, h, MPOOL_DIRTY); |
281 |
|
|
break; |
282 |
|
|
} |
283 |
|
|
|
284 |
|
|
/* If the root page was split, make it look right. */ |
285 |
|
|
if (sp->pgno == P_ROOT && |
286 |
|
|
(F_ISSET(t, R_RECNO) ? |
287 |
|
|
bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR) |
288 |
|
|
goto err1; |
289 |
|
|
|
290 |
|
|
mpool_put(t->bt_mp, lchild, MPOOL_DIRTY); |
291 |
|
|
mpool_put(t->bt_mp, rchild, MPOOL_DIRTY); |
292 |
|
|
} |
293 |
|
|
|
294 |
|
|
/* Unpin the held pages. */ |
295 |
|
|
mpool_put(t->bt_mp, l, MPOOL_DIRTY); |
296 |
|
|
mpool_put(t->bt_mp, r, MPOOL_DIRTY); |
297 |
|
|
|
298 |
|
|
/* Clear any pages left on the stack. */ |
299 |
|
|
return (RET_SUCCESS); |
300 |
|
|
|
301 |
|
|
/* |
302 |
|
|
* If something fails in the above loop we were already walking back |
303 |
|
|
* up the tree and the tree is now inconsistent. Nothing much we can |
304 |
|
|
* do about it but release any memory we're holding. |
305 |
|
|
*/ |
306 |
|
|
err1: mpool_put(t->bt_mp, lchild, MPOOL_DIRTY); |
307 |
|
|
mpool_put(t->bt_mp, rchild, MPOOL_DIRTY); |
308 |
|
|
|
309 |
|
|
err2: mpool_put(t->bt_mp, l, 0); |
310 |
|
|
mpool_put(t->bt_mp, r, 0); |
311 |
|
|
__dbpanic(t->bt_dbp); |
312 |
|
|
return (RET_ERROR); |
313 |
|
|
} |
314 |
|
|
|
315 |
|
|
/* |
316 |
|
|
* BT_PAGE -- Split a non-root page of a btree. |
317 |
|
|
* |
318 |
|
|
* Parameters: |
319 |
|
|
* t: tree |
320 |
|
|
* h: root page |
321 |
|
|
* lp: pointer to left page pointer |
322 |
|
|
* rp: pointer to right page pointer |
323 |
|
|
* skip: pointer to index to leave open |
324 |
|
|
* ilen: insert length |
325 |
|
|
* |
326 |
|
|
* Returns: |
327 |
|
|
* Pointer to page in which to insert or NULL on error. |
328 |
|
|
*/ |
329 |
|
|
static PAGE * |
330 |
|
|
bt_page(BTREE *t, PAGE *h, PAGE **lp, PAGE **rp, indx_t *skip, size_t ilen) |
331 |
|
|
{ |
332 |
|
|
PAGE *l, *r, *tp; |
333 |
|
|
pgno_t npg; |
334 |
|
|
|
335 |
|
|
#ifdef STATISTICS |
336 |
|
|
++bt_split; |
337 |
|
|
#endif |
338 |
|
|
/* Put the new right page for the split into place. */ |
339 |
|
|
if ((r = __bt_new(t, &npg)) == NULL) |
340 |
|
|
return (NULL); |
341 |
|
|
r->pgno = npg; |
342 |
|
|
r->lower = BTDATAOFF; |
343 |
|
|
r->upper = t->bt_psize; |
344 |
|
|
r->nextpg = h->nextpg; |
345 |
|
|
r->prevpg = h->pgno; |
346 |
|
|
r->flags = h->flags & P_TYPE; |
347 |
|
|
|
348 |
|
|
/* |
349 |
|
|
* If we're splitting the last page on a level because we're appending |
350 |
|
|
* a key to it (skip is NEXTINDEX()), it's likely that the data is |
351 |
|
|
* sorted. Adding an empty page on the side of the level is less work |
352 |
|
|
* and can push the fill factor much higher than normal. If we're |
353 |
|
|
* wrong it's no big deal, we'll just do the split the right way next |
354 |
|
|
* time. It may look like it's equally easy to do a similar hack for |
355 |
|
|
* reverse sorted data, that is, split the tree left, but it's not. |
356 |
|
|
* Don't even try. |
357 |
|
|
*/ |
358 |
|
|
if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) { |
359 |
|
|
#ifdef STATISTICS |
360 |
|
|
++bt_sortsplit; |
361 |
|
|
#endif |
362 |
|
|
h->nextpg = r->pgno; |
363 |
|
|
r->lower = BTDATAOFF + sizeof(indx_t); |
364 |
|
|
*skip = 0; |
365 |
|
|
*lp = h; |
366 |
|
|
*rp = r; |
367 |
|
|
return (r); |
368 |
|
|
} |
369 |
|
|
|
370 |
|
|
/* Put the new left page for the split into place. */ |
371 |
|
|
if ((l = (PAGE *)malloc(t->bt_psize)) == NULL) { |
372 |
|
|
mpool_put(t->bt_mp, r, 0); |
373 |
|
|
return (NULL); |
374 |
|
|
} |
375 |
|
|
memset(l, 0xff, t->bt_psize); |
376 |
|
|
l->pgno = h->pgno; |
377 |
|
|
l->nextpg = r->pgno; |
378 |
|
|
l->prevpg = h->prevpg; |
379 |
|
|
l->lower = BTDATAOFF; |
380 |
|
|
l->upper = t->bt_psize; |
381 |
|
|
l->flags = h->flags & P_TYPE; |
382 |
|
|
|
383 |
|
|
/* Fix up the previous pointer of the page after the split page. */ |
384 |
|
|
if (h->nextpg != P_INVALID) { |
385 |
|
|
if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) { |
386 |
|
|
free(l); |
387 |
|
|
/* XXX mpool_free(t->bt_mp, r->pgno); */ |
388 |
|
|
return (NULL); |
389 |
|
|
} |
390 |
|
|
tp->prevpg = r->pgno; |
391 |
|
|
mpool_put(t->bt_mp, tp, MPOOL_DIRTY); |
392 |
|
|
} |
393 |
|
|
|
394 |
|
|
/* |
395 |
|
|
* Split right. The key/data pairs aren't sorted in the btree page so |
396 |
|
|
* it's simpler to copy the data from the split page onto two new pages |
397 |
|
|
* instead of copying half the data to the right page and compacting |
398 |
|
|
* the left page in place. Since the left page can't change, we have |
399 |
|
|
* to swap the original and the allocated left page after the split. |
400 |
|
|
*/ |
401 |
|
|
tp = bt_psplit(t, h, l, r, skip, ilen); |
402 |
|
|
|
403 |
|
|
/* Move the new left page onto the old left page. */ |
404 |
|
|
memmove(h, l, t->bt_psize); |
405 |
|
|
if (tp == l) |
406 |
|
|
tp = h; |
407 |
|
|
free(l); |
408 |
|
|
|
409 |
|
|
*lp = h; |
410 |
|
|
*rp = r; |
411 |
|
|
return (tp); |
412 |
|
|
} |
413 |
|
|
|
414 |
|
|
/* |
415 |
|
|
* BT_ROOT -- Split the root page of a btree. |
416 |
|
|
* |
417 |
|
|
* Parameters: |
418 |
|
|
* t: tree |
419 |
|
|
* h: root page |
420 |
|
|
* lp: pointer to left page pointer |
421 |
|
|
* rp: pointer to right page pointer |
422 |
|
|
* skip: pointer to index to leave open |
423 |
|
|
* ilen: insert length |
424 |
|
|
* |
425 |
|
|
* Returns: |
426 |
|
|
* Pointer to page in which to insert or NULL on error. |
427 |
|
|
*/ |
428 |
|
|
static PAGE * |
429 |
|
|
bt_root(BTREE *t, PAGE *h, PAGE **lp, PAGE **rp, indx_t *skip, size_t ilen) |
430 |
|
|
{ |
431 |
|
|
PAGE *l, *r, *tp; |
432 |
|
|
pgno_t lnpg, rnpg; |
433 |
|
|
|
434 |
|
|
#ifdef STATISTICS |
435 |
|
|
++bt_split; |
436 |
|
|
++bt_rootsplit; |
437 |
|
|
#endif |
438 |
|
|
/* Put the new left and right pages for the split into place. */ |
439 |
|
|
if ((l = __bt_new(t, &lnpg)) == NULL || |
440 |
|
|
(r = __bt_new(t, &rnpg)) == NULL) |
441 |
|
|
return (NULL); |
442 |
|
|
l->pgno = lnpg; |
443 |
|
|
r->pgno = rnpg; |
444 |
|
|
l->nextpg = r->pgno; |
445 |
|
|
r->prevpg = l->pgno; |
446 |
|
|
l->prevpg = r->nextpg = P_INVALID; |
447 |
|
|
l->lower = r->lower = BTDATAOFF; |
448 |
|
|
l->upper = r->upper = t->bt_psize; |
449 |
|
|
l->flags = r->flags = h->flags & P_TYPE; |
450 |
|
|
|
451 |
|
|
/* Split the root page. */ |
452 |
|
|
tp = bt_psplit(t, h, l, r, skip, ilen); |
453 |
|
|
|
454 |
|
|
*lp = l; |
455 |
|
|
*rp = r; |
456 |
|
|
return (tp); |
457 |
|
|
} |
458 |
|
|
|
459 |
|
|
/* |
460 |
|
|
* BT_RROOT -- Fix up the recno root page after it has been split. |
461 |
|
|
* |
462 |
|
|
* Parameters: |
463 |
|
|
* t: tree |
464 |
|
|
* h: root page |
465 |
|
|
* l: left page |
466 |
|
|
* r: right page |
467 |
|
|
* |
468 |
|
|
* Returns: |
469 |
|
|
* RET_ERROR, RET_SUCCESS |
470 |
|
|
*/ |
471 |
|
|
static int |
472 |
|
|
bt_rroot(BTREE *t, PAGE *h, PAGE *l, PAGE *r) |
473 |
|
|
{ |
474 |
|
|
char *dest; |
475 |
|
|
|
476 |
|
|
/* Insert the left and right keys, set the header information. */ |
477 |
|
|
h->linp[0] = h->upper = t->bt_psize - NRINTERNAL; |
478 |
|
|
dest = (char *)h + h->upper; |
479 |
|
|
WR_RINTERNAL(dest, |
480 |
|
|
l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno); |
481 |
|
|
|
482 |
|
|
h->linp[1] = h->upper -= NRINTERNAL; |
483 |
|
|
dest = (char *)h + h->upper; |
484 |
|
|
WR_RINTERNAL(dest, |
485 |
|
|
r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno); |
486 |
|
|
|
487 |
|
|
h->lower = BTDATAOFF + 2 * sizeof(indx_t); |
488 |
|
|
|
489 |
|
|
/* Unpin the root page, set to recno internal page. */ |
490 |
|
|
h->flags &= ~P_TYPE; |
491 |
|
|
h->flags |= P_RINTERNAL; |
492 |
|
|
mpool_put(t->bt_mp, h, MPOOL_DIRTY); |
493 |
|
|
|
494 |
|
|
return (RET_SUCCESS); |
495 |
|
|
} |
496 |
|
|
|
497 |
|
|
/* |
498 |
|
|
* BT_BROOT -- Fix up the btree root page after it has been split. |
499 |
|
|
* |
500 |
|
|
* Parameters: |
501 |
|
|
* t: tree |
502 |
|
|
* h: root page |
503 |
|
|
* l: left page |
504 |
|
|
* r: right page |
505 |
|
|
* |
506 |
|
|
* Returns: |
507 |
|
|
* RET_ERROR, RET_SUCCESS |
508 |
|
|
*/ |
509 |
|
|
static int |
510 |
|
|
bt_broot(BTREE *t, PAGE *h, PAGE *l, PAGE *r) |
511 |
|
|
{ |
512 |
|
|
BINTERNAL *bi; |
513 |
|
|
BLEAF *bl; |
514 |
|
|
u_int32_t nbytes; |
515 |
|
|
char *dest; |
516 |
|
|
|
517 |
|
|
/* |
518 |
|
|
* If the root page was a leaf page, change it into an internal page. |
519 |
|
|
* We copy the key we split on (but not the key's data, in the case of |
520 |
|
|
* a leaf page) to the new root page. |
521 |
|
|
* |
522 |
|
|
* The btree comparison code guarantees that the left-most key on any |
523 |
|
|
* level of the tree is never used, so it doesn't need to be filled in. |
524 |
|
|
*/ |
525 |
|
|
nbytes = NBINTERNAL(0); |
526 |
|
|
h->linp[0] = h->upper = t->bt_psize - nbytes; |
527 |
|
|
dest = (char *)h + h->upper; |
528 |
|
|
WR_BINTERNAL(dest, 0, l->pgno, 0); |
529 |
|
|
|
530 |
|
|
switch (h->flags & P_TYPE) { |
531 |
|
|
case P_BLEAF: |
532 |
|
|
bl = GETBLEAF(r, 0); |
533 |
|
|
nbytes = NBINTERNAL(bl->ksize); |
534 |
|
|
h->linp[1] = h->upper -= nbytes; |
535 |
|
|
dest = (char *)h + h->upper; |
536 |
|
|
WR_BINTERNAL(dest, bl->ksize, r->pgno, 0); |
537 |
|
|
memmove(dest, bl->bytes, bl->ksize); |
538 |
|
|
|
539 |
|
|
/* |
540 |
|
|
* If the key is on an overflow page, mark the overflow chain |
541 |
|
|
* so it isn't deleted when the leaf copy of the key is deleted. |
542 |
|
|
*/ |
543 |
|
|
if (bl->flags & P_BIGKEY && |
544 |
|
|
bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR) |
545 |
|
|
return (RET_ERROR); |
546 |
|
|
break; |
547 |
|
|
case P_BINTERNAL: |
548 |
|
|
bi = GETBINTERNAL(r, 0); |
549 |
|
|
nbytes = NBINTERNAL(bi->ksize); |
550 |
|
|
h->linp[1] = h->upper -= nbytes; |
551 |
|
|
dest = (char *)h + h->upper; |
552 |
|
|
memmove(dest, bi, nbytes); |
553 |
|
|
((BINTERNAL *)dest)->pgno = r->pgno; |
554 |
|
|
break; |
555 |
|
|
default: |
556 |
|
|
abort(); |
557 |
|
|
} |
558 |
|
|
|
559 |
|
|
/* There are two keys on the page. */ |
560 |
|
|
h->lower = BTDATAOFF + 2 * sizeof(indx_t); |
561 |
|
|
|
562 |
|
|
/* Unpin the root page, set to btree internal page. */ |
563 |
|
|
h->flags &= ~P_TYPE; |
564 |
|
|
h->flags |= P_BINTERNAL; |
565 |
|
|
mpool_put(t->bt_mp, h, MPOOL_DIRTY); |
566 |
|
|
|
567 |
|
|
return (RET_SUCCESS); |
568 |
|
|
} |
569 |
|
|
|
570 |
|
|
/* |
571 |
|
|
* BT_PSPLIT -- Do the real work of splitting the page. |
572 |
|
|
* |
573 |
|
|
* Parameters: |
574 |
|
|
* t: tree |
575 |
|
|
* h: page to be split |
576 |
|
|
* l: page to put lower half of data |
577 |
|
|
* r: page to put upper half of data |
578 |
|
|
* pskip: pointer to index to leave open |
579 |
|
|
* ilen: insert length |
580 |
|
|
* |
581 |
|
|
* Returns: |
582 |
|
|
* Pointer to page in which to insert. |
583 |
|
|
*/ |
584 |
|
|
static PAGE * |
585 |
|
|
bt_psplit(BTREE *t, PAGE *h, PAGE *l, PAGE *r, indx_t *pskip, size_t ilen) |
586 |
|
|
{ |
587 |
|
|
BINTERNAL *bi; |
588 |
|
|
BLEAF *bl; |
589 |
|
|
CURSOR *c; |
590 |
|
|
RLEAF *rl; |
591 |
|
|
PAGE *rval; |
592 |
|
|
void *src; |
593 |
|
|
indx_t full, half, nxt, off, skip, top, used; |
594 |
|
|
u_int32_t nbytes; |
595 |
|
|
int bigkeycnt, isbigkey; |
596 |
|
|
|
597 |
|
|
/* |
598 |
|
|
* Split the data to the left and right pages. Leave the skip index |
599 |
|
|
* open. Additionally, make some effort not to split on an overflow |
600 |
|
|
* key. This makes internal page processing faster and can save |
601 |
|
|
* space as overflow keys used by internal pages are never deleted. |
602 |
|
|
*/ |
603 |
|
|
bigkeycnt = 0; |
604 |
|
|
skip = *pskip; |
605 |
|
|
full = t->bt_psize - BTDATAOFF; |
606 |
|
|
half = full / 2; |
607 |
|
|
used = 0; |
608 |
|
|
for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) { |
609 |
|
|
if (skip == off) { |
610 |
|
|
nbytes = ilen; |
611 |
|
|
isbigkey = 0; /* XXX: not really known. */ |
612 |
|
|
} else |
613 |
|
|
switch (h->flags & P_TYPE) { |
614 |
|
|
case P_BINTERNAL: |
615 |
|
|
src = bi = GETBINTERNAL(h, nxt); |
616 |
|
|
nbytes = NBINTERNAL(bi->ksize); |
617 |
|
|
isbigkey = bi->flags & P_BIGKEY; |
618 |
|
|
break; |
619 |
|
|
case P_BLEAF: |
620 |
|
|
src = bl = GETBLEAF(h, nxt); |
621 |
|
|
nbytes = NBLEAF(bl); |
622 |
|
|
isbigkey = bl->flags & P_BIGKEY; |
623 |
|
|
break; |
624 |
|
|
case P_RINTERNAL: |
625 |
|
|
src = GETRINTERNAL(h, nxt); |
626 |
|
|
nbytes = NRINTERNAL; |
627 |
|
|
isbigkey = 0; |
628 |
|
|
break; |
629 |
|
|
case P_RLEAF: |
630 |
|
|
src = rl = GETRLEAF(h, nxt); |
631 |
|
|
nbytes = NRLEAF(rl); |
632 |
|
|
isbigkey = 0; |
633 |
|
|
break; |
634 |
|
|
default: |
635 |
|
|
abort(); |
636 |
|
|
} |
637 |
|
|
|
638 |
|
|
/* |
639 |
|
|
* If the key/data pairs are substantial fractions of the max |
640 |
|
|
* possible size for the page, it's possible to get situations |
641 |
|
|
* where we decide to try and copy too much onto the left page. |
642 |
|
|
* Make sure that doesn't happen. |
643 |
|
|
*/ |
644 |
|
|
if ((skip <= off && used + nbytes + sizeof(indx_t) >= full) || |
645 |
|
|
nxt == top - 1) { |
646 |
|
|
--off; |
647 |
|
|
break; |
648 |
|
|
} |
649 |
|
|
|
650 |
|
|
/* Copy the key/data pair, if not the skipped index. */ |
651 |
|
|
if (skip != off) { |
652 |
|
|
++nxt; |
653 |
|
|
|
654 |
|
|
l->linp[off] = l->upper -= nbytes; |
655 |
|
|
memmove((char *)l + l->upper, src, nbytes); |
656 |
|
|
} |
657 |
|
|
|
658 |
|
|
used += nbytes + sizeof(indx_t); |
659 |
|
|
if (used >= half) { |
660 |
|
|
if (!isbigkey || bigkeycnt == 3) |
661 |
|
|
break; |
662 |
|
|
else |
663 |
|
|
++bigkeycnt; |
664 |
|
|
} |
665 |
|
|
} |
666 |
|
|
|
667 |
|
|
/* |
668 |
|
|
* Off is the last offset that's valid for the left page. |
669 |
|
|
* Nxt is the first offset to be placed on the right page. |
670 |
|
|
*/ |
671 |
|
|
l->lower += (off + 1) * sizeof(indx_t); |
672 |
|
|
|
673 |
|
|
/* |
674 |
|
|
* If splitting the page that the cursor was on, the cursor has to be |
675 |
|
|
* adjusted to point to the same record as before the split. If the |
676 |
|
|
* cursor is at or past the skipped slot, the cursor is incremented by |
677 |
|
|
* one. If the cursor is on the right page, it is decremented by the |
678 |
|
|
* number of records split to the left page. |
679 |
|
|
*/ |
680 |
|
|
c = &t->bt_cursor; |
681 |
|
|
if (F_ISSET(c, CURS_INIT) && c->pg.pgno == h->pgno) { |
682 |
|
|
if (c->pg.index >= skip) |
683 |
|
|
++c->pg.index; |
684 |
|
|
if (c->pg.index < nxt) /* Left page. */ |
685 |
|
|
c->pg.pgno = l->pgno; |
686 |
|
|
else { /* Right page. */ |
687 |
|
|
c->pg.pgno = r->pgno; |
688 |
|
|
c->pg.index -= nxt; |
689 |
|
|
} |
690 |
|
|
} |
691 |
|
|
|
692 |
|
|
/* |
693 |
|
|
* If the skipped index was on the left page, just return that page. |
694 |
|
|
* Otherwise, adjust the skip index to reflect the new position on |
695 |
|
|
* the right page. |
696 |
|
|
*/ |
697 |
|
|
if (skip <= off) { |
698 |
|
|
skip = MAX_PAGE_OFFSET; |
699 |
|
|
rval = l; |
700 |
|
|
} else { |
701 |
|
|
rval = r; |
702 |
|
|
*pskip -= nxt; |
703 |
|
|
} |
704 |
|
|
|
705 |
|
|
for (off = 0; nxt < top; ++off) { |
706 |
|
|
if (skip == nxt) { |
707 |
|
|
++off; |
708 |
|
|
skip = MAX_PAGE_OFFSET; |
709 |
|
|
} |
710 |
|
|
switch (h->flags & P_TYPE) { |
711 |
|
|
case P_BINTERNAL: |
712 |
|
|
src = bi = GETBINTERNAL(h, nxt); |
713 |
|
|
nbytes = NBINTERNAL(bi->ksize); |
714 |
|
|
break; |
715 |
|
|
case P_BLEAF: |
716 |
|
|
src = bl = GETBLEAF(h, nxt); |
717 |
|
|
nbytes = NBLEAF(bl); |
718 |
|
|
break; |
719 |
|
|
case P_RINTERNAL: |
720 |
|
|
src = GETRINTERNAL(h, nxt); |
721 |
|
|
nbytes = NRINTERNAL; |
722 |
|
|
break; |
723 |
|
|
case P_RLEAF: |
724 |
|
|
src = rl = GETRLEAF(h, nxt); |
725 |
|
|
nbytes = NRLEAF(rl); |
726 |
|
|
break; |
727 |
|
|
default: |
728 |
|
|
abort(); |
729 |
|
|
} |
730 |
|
|
++nxt; |
731 |
|
|
r->linp[off] = r->upper -= nbytes; |
732 |
|
|
memmove((char *)r + r->upper, src, nbytes); |
733 |
|
|
} |
734 |
|
|
r->lower += off * sizeof(indx_t); |
735 |
|
|
|
736 |
|
|
/* If the key is being appended to the page, adjust the index. */ |
737 |
|
|
if (skip == top) |
738 |
|
|
r->lower += sizeof(indx_t); |
739 |
|
|
|
740 |
|
|
return (rval); |
741 |
|
|
} |
742 |
|
|
|
743 |
|
|
/* |
744 |
|
|
* BT_PRESERVE -- Mark a chain of pages as used by an internal node. |
745 |
|
|
* |
746 |
|
|
* Chains of indirect blocks pointed to by leaf nodes get reclaimed when the |
747 |
|
|
* record that references them gets deleted. Chains pointed to by internal |
748 |
|
|
* pages never get deleted. This routine marks a chain as pointed to by an |
749 |
|
|
* internal page. |
750 |
|
|
* |
751 |
|
|
* Parameters: |
752 |
|
|
* t: tree |
753 |
|
|
* pg: page number of first page in the chain. |
754 |
|
|
* |
755 |
|
|
* Returns: |
756 |
|
|
* RET_SUCCESS, RET_ERROR. |
757 |
|
|
*/ |
758 |
|
|
static int |
759 |
|
|
bt_preserve(BTREE *t, pgno_t pg) |
760 |
|
|
{ |
761 |
|
|
PAGE *h; |
762 |
|
|
|
763 |
|
|
if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) |
764 |
|
|
return (RET_ERROR); |
765 |
|
|
h->flags |= P_PRESERVE; |
766 |
|
|
mpool_put(t->bt_mp, h, MPOOL_DIRTY); |
767 |
|
|
return (RET_SUCCESS); |
768 |
|
|
} |
769 |
|
|
|
770 |
|
|
/* |
771 |
|
|
* REC_TOTAL -- Return the number of recno entries below a page. |
772 |
|
|
* |
773 |
|
|
* Parameters: |
774 |
|
|
* h: page |
775 |
|
|
* |
776 |
|
|
* Returns: |
777 |
|
|
* The number of recno entries below a page. |
778 |
|
|
* |
779 |
|
|
* XXX |
780 |
|
|
* These values could be set by the bt_psplit routine. The problem is that the |
781 |
|
|
* entry has to be popped off of the stack etc. or the values have to be passed |
782 |
|
|
* all the way back to bt_split/bt_rroot and it's not very clean. |
783 |
|
|
*/ |
784 |
|
|
static recno_t |
785 |
|
|
rec_total(PAGE *h) |
786 |
|
|
{ |
787 |
|
|
recno_t recs; |
788 |
|
|
indx_t nxt, top; |
789 |
|
|
|
790 |
|
|
for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt) |
791 |
|
|
recs += GETRINTERNAL(h, nxt)->nrecs; |
792 |
|
|
return (recs); |
793 |
|
|
} |