1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
|
/*-
* See the file LICENSE for redistribution information.
*
* Copyright (c) 2008-2011 WiredTiger, Inc.
* All rights reserved.
*
* $Id$
*/
#include "wt_internal.h"
/*
* WT_STACK --
* We maintain a stack of parent pages as we build the tree, encapsulated
* in this structure.
*/
typedef struct {
WT_PAGE *page; /* page header */
uint8_t *first_free; /* page's first free byte */
uint32_t space_avail; /* page's space available */
DBT *tmp; /* page-in-a-buffer */
void *data; /* last on-page WT_COL/WT_ROW */
} WT_STACK_ELEM;
typedef struct {
WT_STACK_ELEM *elem; /* stack */
u_int size; /* stack size */
} WT_STACK;
static int __wt_bulk_dbt_copy(ENV *, DBT *, DBT *);
static int __wt_bulk_dup_offpage(WT_TOC *, DBT **, DBT **, DBT *, WT_ITEM *,
uint32_t, uint32_t, WT_OFF *, int (*)(DB *, DBT **, DBT **));
static int __wt_bulk_fix(WT_TOC *, void (*)(const char *,
uint64_t), int (*)(DB *, DBT **, DBT **));
static int __wt_bulk_ovfl_copy(WT_TOC *, WT_OVFL *, WT_OVFL *);
static int __wt_bulk_ovfl_write(WT_TOC *, DBT *, WT_OVFL *);
static int __wt_bulk_promote(
WT_TOC *, WT_PAGE *, uint64_t, WT_STACK *, u_int, uint32_t *);
static int __wt_bulk_scratch_page(
WT_TOC *, uint32_t, uint32_t, uint32_t, WT_PAGE **, DBT **);
static int __wt_bulk_stack_put(WT_TOC *, WT_STACK *);
static int __wt_bulk_var(WT_TOC *, uint32_t, void (*)(const char *,
uint64_t), int (*)(DB *, DBT **, DBT **));
static int __wt_item_build_key(WT_TOC *, DBT *, WT_ITEM *, WT_OVFL *);
/*
* __wt_db_bulk_load --
* Db.bulk_load method.
*/
int
__wt_db_bulk_load(WT_TOC *toc, uint32_t flags,
void (*f)(const char *, uint64_t), int (*cb)(DB *, DBT **, DBT **))
{
DB *db;
IDB *idb;
uint32_t addr;
db = toc->db;
idb = db->idb;
/*
* XXX
* Write out the description record -- this goes away when we figure
* out how the table schema is going to work, but for now, we use the
* first sector, and this file extend makes sure we don't allocate it
* as a table page.
*/
WT_RET(__wt_file_alloc(toc, &addr, 512));
if (F_ISSET(idb, WT_COLUMN))
WT_DB_FCHK(db, "DB.bulk_load", flags, 0);
/*
* There are two styles of bulk-load: variable length pages or
* fixed-length pages.
*/
if (F_ISSET(idb, WT_COLUMN) && db->fixed_len != 0)
WT_RET(__wt_bulk_fix(toc, f, cb));
else
WT_RET(__wt_bulk_var(toc, flags, f, cb));
/* Get a permanent root page reference. */
return (__wt_root_pin(toc));
}
/*
* __wt_bulk_fix
* Db.bulk_load method for column-store, fixed-length database pages.
*/
static int
__wt_bulk_fix(WT_TOC *toc,
void (*f)(const char *, uint64_t), int (*cb)(DB *, DBT **, DBT **))
{
DB *db;
DBT *key, *data, *tmp;
IDB *idb;
WT_PAGE *page;
WT_PAGE_DISK *dsk;
WT_STACK stack;
uint64_t insert_cnt;
uint32_t len, space_avail;
uint16_t *last_repeat;
uint8_t *first_free, *last_data;
int rle, ret;
db = toc->db;
tmp = NULL;
idb = db->idb;
insert_cnt = 0;
WT_CLEAR(stack);
rle = F_ISSET(idb, WT_RLE) ? 1 : 0;
/* Figure out how large is the chunk we're storing on the page. */
len = db->fixed_len;
if (rle)
len += sizeof(uint16_t);
/* Get a scratch buffer and make it look like our work page. */
WT_ERR(__wt_bulk_scratch_page(toc, db->leafmin,
rle ? WT_PAGE_COL_RLE : WT_PAGE_COL_FIX, WT_LLEAF, &page, &tmp));
dsk = page->dsk;
dsk->start_recno = 1;
__wt_set_ff_and_sa_from_offset(
page, WT_PAGE_BYTE(page), &first_free, &space_avail);
while ((ret = cb(db, &key, &data)) == 0) {
if (key != NULL) {
__wt_api_db_errx(db,
"column database keys are implied and so should "
"not be set by the bulk load input routine");
ret = WT_ERROR;
goto err;
}
if (data->size != db->fixed_len)
WT_ERR(__wt_database_wrong_fixed_size(toc, data->size));
/*
* We use the high bit of the data field as a "deleted" value,
* make sure the user's data doesn't set it.
*/
if (WT_FIX_DELETE_ISSET(data->data)) {
__wt_api_db_errx(db,
"the first bit may not be stored in fixed-length "
"column-store database items");
ret = WT_ERROR;
goto err;
}
/* Report on progress every 100 inserts. */
if (f != NULL && ++insert_cnt % 100 == 0)
f(toc->name, insert_cnt);
WT_STAT_INCR(idb->stats, ITEMS_INSERTED);
/*
* If doing run-length encoding, check to see if this record
* matches the last data inserted. If there's a match try
* and increment that item's repeat count instead of entering
* new data.
*/
if (rle && dsk->u.entries != 0)
if (*last_repeat < UINT16_MAX &&
memcmp(last_data, data->data, data->size) == 0) {
++*last_repeat;
++page->records;
WT_STAT_INCR(idb->stats, REPEAT_COUNT);
continue;
}
/*
* We now have the data item to store on the page. If there
* is insufficient space on the current page, allocate a new
* one.
*/
if (len > space_avail) {
/*
* We've finished with the page: promote its first key
* to its parent and discard it, then switch to the new
* page.
*/
WT_ERR(__wt_bulk_promote(
toc, page, page->records, &stack, 0, NULL));
WT_ERR(__wt_page_write(toc, page));
dsk->u.entries = 0;
page->records = 0;
dsk->start_recno = insert_cnt;
WT_ERR(
__wt_file_alloc(toc, &page->addr, db->leafmin));
__wt_set_ff_and_sa_from_offset(page,
WT_PAGE_BYTE(page), &first_free, &space_avail);
}
++dsk->u.entries;
++page->records;
/*
* Copy the data item onto the page -- if doing run-length
* encoding, track the location of the item for comparison.
*/
if (rle) {
last_repeat = (uint16_t *)first_free;
*last_repeat = 1;
first_free += sizeof(uint16_t);
space_avail -= sizeof(uint16_t);
last_data = first_free;
}
memcpy(first_free, data->data, data->size);
first_free += data->size;
space_avail -= data->size;
}
/* A ret of 1 just means we've reached the end of the input. */
if (ret != 1)
goto err;
ret = 0;
/* Promote a key from any partially-filled page and write it. */
if (dsk->u.entries != 0) {
ret = __wt_bulk_promote(
toc, page, page->records, &stack, 0, NULL);
WT_ERR(__wt_page_write(toc, page));
}
/* Wrap up reporting. */
if (f != NULL)
f(toc->name, insert_cnt);
err: WT_TRET(__wt_bulk_stack_put(toc, &stack));
if (tmp != NULL)
__wt_scr_release(&tmp);
return (ret);
}
/*
* __wt_bulk_var --
* Db.bulk_load method for row or column-store variable-length database
* pages.
*/
static int
__wt_bulk_var(WT_TOC *toc, uint32_t flags,
void (*f)(const char *, uint64_t), int (*cb)(DB *, DBT **, DBT **))
{
DB *db;
DBT *key, *data, key_copy, data_copy;
DBT *lastkey, *lastkey_copy, lastkey_std;
DBT *tmp1, *tmp2;
ENV *env;
IDB *idb;
WT_ITEM key_item, data_item, *dup_key, *dup_data;
WT_OFF off;
WT_OVFL key_ovfl, data_ovfl;
WT_PAGE *page, *next;
WT_STACK stack;
uint64_t insert_cnt;
uint32_t dup_count, dup_space, len, next_space_avail, space_avail;
uint8_t *first_free, *next_first_free, *p, type;
int ret;
db = toc->db;
tmp1 = tmp2 = NULL;
env = toc->env;
idb = db->idb;
ret = 0;
WT_CLEAR(stack);
dup_space = dup_count = 0;
insert_cnt = 0;
type = F_ISSET(idb, WT_COLUMN) ? WT_PAGE_COL_VAR : WT_PAGE_ROW_LEAF;
lastkey = &lastkey_std;
WT_CLEAR(data_copy);
WT_CLEAR(key_copy);
WT_CLEAR(key_item);
WT_CLEAR(lastkey_std);
WT_ERR(__wt_scr_alloc(toc, 0, &lastkey_copy));
/* Get a scratch buffer and make it look like our work page. */
WT_ERR(__wt_bulk_scratch_page(
toc, db->leafmin, type, WT_LLEAF, &page, &tmp1));
__wt_set_ff_and_sa_from_offset(
page, WT_PAGE_BYTE(page), &first_free, &space_avail);
if (type == WT_PAGE_COL_VAR)
page->dsk->start_recno = 1;
while ((ret = cb(db, &key, &data)) == 0) {
if (F_ISSET(idb, WT_COLUMN) ) {
if (key != NULL) {
__wt_api_db_errx(db,
"column database keys are implied and "
"so should not be returned by the bulk "
"load input routine");
ret = WT_ERROR;
goto err;
}
} else {
if (key == NULL && !LF_ISSET(WT_DUPLICATES)) {
__wt_api_db_errx(db,
"keys must be specified unless duplicates "
"are configured");
ret = WT_ERROR;
goto err;
}
if (key != NULL && key->size == 0) {
__wt_api_db_errx(db,
"zero-length keys are not supported");
ret = WT_ERROR;
goto err;
}
}
/* Report on progress every 100 inserts. */
if (f != NULL && ++insert_cnt % 100 == 0)
f(toc->name, insert_cnt);
WT_STAT_INCR(idb->stats, ITEMS_INSERTED);
/*
* We don't have a key to store on the page if we're building a
* column-store, and we don't store the key on the page in the
* case of a row-store duplicate data item. The check from here
* on is if "key == NULL" for both cases, that is, there's no
* key to store.
*/
skip_read: /*
* We pushed a set of duplicates off-page, and that routine
* returned an ending key/data pair to us.
*/
/*
* Copy the caller's DBTs, we don't want to modify them. But,
* copy them carefully, all we want is a pointer and a length.
*/
if (key != NULL) {
key_copy.data = key->data;
key_copy.size = key->size;
key = &key_copy;
}
data_copy.data = data->data;
data_copy.size = data->size;
data = &data_copy;
/* Build the data item we're going to store on the page. */
WT_ERR(__wt_item_build_data(
toc, data, &data_item, &data_ovfl, 0));
/*
* Check for duplicate keys; we don't store the key on the page
* in the case of a duplicate.
*
* !!!
* Do a fast check of the old and new sizes -- note checking
* lastkey->size is safe -- it's initialized to 0, and we do
* not allow zero-length keys.
*/
if (LF_ISSET(WT_DUPLICATES) &&
(key == NULL ||
(lastkey->size == key->size &&
db->btree_compare(db, lastkey, key) == 0))) {
/*
* The first duplicate in the set is already on the
* page, but with an item type set to WT_ITEM_DATA or
* WT_ITEM_DATA_OVFL. Correct the type and dup_count.
*/
if (++dup_count == 1) {
dup_count = 2;
WT_ITEM_SET_TYPE(dup_data,
WT_ITEM_TYPE(dup_data) == WT_ITEM_DATA ?
WT_ITEM_DATA_DUP : WT_ITEM_DATA_DUP_OVFL);
}
/* Reset the type of the current item to a duplicate. */
WT_ITEM_SET_TYPE(&data_item,
WT_ITEM_TYPE(&data_item) == WT_ITEM_DATA ?
WT_ITEM_DATA_DUP : WT_ITEM_DATA_DUP_OVFL);
WT_STAT_INCR(idb->stats, DUPLICATE_ITEMS_INSERTED);
key = NULL;
} else {
/*
* It's a new key, but if duplicates are possible we'll
* need a copy of the key for comparison with the next
* key. If the key is Huffman encoded or an overflow
* object, we can't use the on-page version, we have to
* save a copy.
*/
if (LF_ISSET(WT_DUPLICATES) &&
(key->size > db->leafitemsize ||
idb->huffman_key != NULL)) {
WT_ERR(
__wt_bulk_dbt_copy(env, key, lastkey_copy));
lastkey = lastkey_copy;
} else
lastkey = NULL;
dup_count = 0;
}
/* Build the key item we're going to store on the page. */
if (key != NULL)
WT_ERR(__wt_item_build_key(
toc, key, &key_item, &key_ovfl));
/*
* We now have the key/data items to store on the page. If
* there is insufficient space on the current page, allocate
* a new one.
*/
if ((key == NULL ? 0 : WT_ITEM_SPACE_REQ(key->size)) +
WT_ITEM_SPACE_REQ(data->size) > space_avail) {
WT_ERR(__wt_bulk_scratch_page(toc,
db->leafmin, type, WT_LLEAF, &next, &tmp2));
__wt_set_ff_and_sa_from_offset(next,
WT_PAGE_BYTE(next),
&next_first_free, &next_space_avail);
if (type == WT_PAGE_COL_VAR)
next->dsk->start_recno = insert_cnt;
/*
* If in the middle of loading a set of duplicates, but
* the set hasn't yet reached the boundary where we'd
* push them offpage, we can't split them across the two
* pages. Move the entire set to the new page. This
* can waste up to 25% of the old page, but it would be
* difficult and messy to move them and then go back
* and fix things up if and when they moved offpage.
*
* We use a check of dup_count instead of checking the
* WT_DUPLICATES flag, since we have to check it anyway.
*/
if (dup_count != 0) {
/*
* Reset the page entry and record counts -- we
* are moving a single key plus the duplicate
* set.
*
* Since dup_count was already incremented to
* reflect the data item we're loading now, it
* is the right number of elements to move, that
* is, move (dup_count - 1) + 1 for the key.
*/
page->dsk->u.entries -= dup_count;
page->records -= dup_count - 1;
next->dsk->u.entries += dup_count;
next->records += dup_count - 1;
/*
* Move the duplicate set and adjust the page
* information for "next" -- we don't have to
* fix up "page", we're never going to use it
* again.
*/
len =
(uint32_t)(first_free - (uint8_t *)dup_key);
memcpy(next_first_free, dup_key, len);
next_first_free += len;
next_space_avail -= len;
/*
* We'll never have to move this dup set to
* another primary page -- if the dup set
* continues to grow, it will be moved
* off-page. We still need to know where
* the dup set starts, though, for the
* possible move off-page: it's the second
* entry on the page, where the first entry
* is the dup set's key.
*/
dup_key = (WT_ITEM *)WT_PAGE_BYTE(next);
dup_data = (WT_ITEM *)((uint8_t *)dup_key +
WT_ITEM_SPACE_REQ(WT_ITEM_LEN(dup_key)));
/*
* The "lastkey" value just moved to a new page.
* If it's an overflow item, we have a copy; if
* it's not, then we need to reset it.
*/
if (lastkey == &lastkey_std) {
lastkey_std.data =
WT_ITEM_BYTE(dup_key);
lastkey_std.size = WT_ITEM_LEN(dup_key);
}
}
/*
* We've finished with the page: promote its first key
* to its parent and discard it, then switch to the new
* page.
*/
WT_ERR(__wt_bulk_promote(
toc, page, page->records, &stack, 0, NULL));
WT_ERR(__wt_page_write(toc, page));
__wt_scr_release(&tmp1);
/*
* Discard the last page, and switch to the next page.
*
* XXX
* The obvious speed-up here is to re-initialize page
* instead of discarding it and acquiring it again as
* as soon as the just-allocated page fills up. I am
* not doing that deliberately: eventually we'll use
* asynchronous I/O in bulk load, which means the page
* won't be reusable until the I/O completes.
*/
page = next;
first_free = next_first_free;
space_avail = next_space_avail;
next = NULL;
next_first_free = NULL;
next_space_avail = 0;
tmp1 = tmp2;
tmp2 = NULL;
}
++page->records;
/* Copy the key item onto the page. */
if (key != NULL) {
++page->dsk->u.entries;
memcpy(first_free, &key_item, sizeof(key_item));
memcpy(first_free +
sizeof(key_item), key->data, key->size);
space_avail -= WT_ITEM_SPACE_REQ(key->size);
/*
* If processing duplicates we'll need a copy of the key
* for comparison with the next key. If the key was an
* overflow or Huffman encoded item, we already have a
* copy -- otherwise, use the copy we just put on the
* page.
*
* We also save the location for the key of any current
* duplicate set in case we have to move the set to a
* different page (the case where a duplicate set isn't
* large enough to move offpage, but doesn't entirely
* fit on this page).
*/
if (LF_ISSET(WT_DUPLICATES)) {
if (lastkey == NULL) {
lastkey = &lastkey_std;
lastkey_std.data =
WT_ITEM_BYTE(first_free);
lastkey_std.size = key->size;
}
dup_key = (WT_ITEM *)first_free;
}
first_free += WT_ITEM_SPACE_REQ(key->size);
}
/* Copy the data item onto the page. */
++page->dsk->u.entries;
memcpy(first_free, &data_item, sizeof(data_item));
memcpy(first_free + sizeof(data_item), data->data, data->size);
space_avail -= WT_ITEM_SPACE_REQ(data->size);
/*
* If duplicates: if this isn't a duplicate data item, save
* the item location, since it's potentially the first of a
* duplicate data set, and we need to know where duplicate
* data sets start. Additionally, reset the counter and
* space calculation.
*/
if (LF_ISSET(WT_DUPLICATES) && dup_count == 0) {
dup_space = data->size;
dup_data = (WT_ITEM *)first_free;
}
first_free += WT_ITEM_SPACE_REQ(data->size);
/*
* If duplicates: check to see if the duplicate set crosses
* the (roughly) 25% of the page space boundary. If it does,
* move it offpage.
*/
if (LF_ISSET(WT_DUPLICATES) && dup_count != 0) {
dup_space += data->size;
if (dup_space < db->leafmin / db->btree_dup_offpage)
continue;
/*
* Move the duplicate set off our page, and read in the
* rest of the off-page duplicate set.
*/
WT_ERR(__wt_bulk_dup_offpage(toc, &key, &data, lastkey,
dup_data,
(uint32_t)(first_free - (uint8_t *)dup_data),
dup_count, &off, cb));
/* Reset the page entry and record counts. */
page->dsk->u.entries -= (dup_count - 1);
page->records -= dup_count;
page->records += WT_RECORDS(&off);
/*
* Replace the duplicate set with a WT_OFF structure,
* that is, we've replaced dup_count entries with a
* single entry.
*/
WT_ITEM_SET(&data_item, WT_ITEM_OFF, sizeof(WT_OFF));
p = (uint8_t *)dup_data;
memcpy(p, &data_item, sizeof(data_item));
memcpy(p + sizeof(data_item), &off, sizeof(WT_OFF));
__wt_set_ff_and_sa_from_offset(page,
(uint8_t *)p + WT_ITEM_SPACE_REQ(sizeof(WT_OFF)),
&first_free, &space_avail);
/* Reset local counters. */
dup_count = dup_space = 0;
goto skip_read;
}
}
/* A ret of 1 just means we've reached the end of the input. */
if (ret != 1)
goto err;
ret = 0;
/* Promote a key from any partially-filled page and write it. */
if (page->dsk->u.entries != 0) {
WT_ERR(__wt_bulk_promote(
toc, page, page->records, &stack, 0, NULL));
WT_ERR(__wt_page_write(toc, page));
}
/* Wrap up reporting. */
if (f != NULL)
f(toc->name, insert_cnt);
err: WT_TRET(__wt_bulk_stack_put(toc, &stack));
if (lastkey_copy != NULL)
__wt_scr_release(&lastkey_copy);
if (tmp1 != NULL)
__wt_scr_release(&tmp1);
if (tmp2 != NULL)
__wt_scr_release(&tmp2);
return (ret);
}
/*
* __wt_bulk_dup_offpage --
* Move the last set of duplicates on the page to a page of their own,
* then load the rest of the duplicate set.
*/
static int
__wt_bulk_dup_offpage(WT_TOC *toc, DBT **keyp, DBT **datap, DBT *lastkey,
WT_ITEM *dup_data, uint32_t dup_len, uint32_t dup_count, WT_OFF *off,
int (*cb)(DB *, DBT **, DBT **))
{
DB *db;
DBT *key, *data, *tmp;
IDB *idb;
WT_ITEM data_item;
WT_OVFL data_ovfl;
WT_PAGE *page;
WT_STACK stack;
uint32_t root_addr, space_avail;
uint8_t *first_free;
int ret, success_return;
db = toc->db;
idb = db->idb;
success_return = 0;
/*
* This routine is the same as the bulk load routine, except it loads
* only data items into off-page duplicate trees. It's passed a lot
* of state from the bulk load routine, and updates that state as a
* side-effect.
*
* In summary, the bulk load routine stops loading a primary btree leaf
* page, calls us to load a set of duplicate data items into a separate
* btree, and then continues on with its primary leaf page when we
* return. The arguments are complex enough that it's worth describing
* them:
*
* keyp/datap --
* The key and data pairs the application is filling in -- we
* get them passed to us because we get additional key/data
* pairs returned to us, and the last one we get is likely to
* be consumed by our caller.
* lastkey --
* The last key pushed onto the caller's page -- we use this to
* compare against future keys we read.
* dup_data --
* On-page reference to the first duplicate data item in the set.
* dup_count --
* Count of duplicates in the set.
* off --
* Callers WT_OFF structure, which we have to fill in.
* cb --
* User's callback function.
*/
WT_CLEAR(data_item);
WT_CLEAR(stack);
ret = 0;
/* Get a scratch buffer and make it look like our work page. */
WT_ERR(__wt_bulk_scratch_page(toc,
db->leafmin, WT_PAGE_DUP_LEAF, WT_LLEAF, &page, &tmp));
__wt_set_ff_and_sa_from_offset(
page, WT_PAGE_BYTE(page), &first_free, &space_avail);
/* Move the duplicates onto the newly allocated page. */
page->records = dup_count;
page->dsk->u.entries = dup_count;
memcpy(first_free, dup_data, (size_t)dup_len);
first_free += dup_len;
space_avail -= dup_len;
/*
* Unless we have enough duplicates to split this page, it will be the
* "root" of the offpage duplicates.
*/
root_addr = page->addr;
/* Read in new duplicate records until the key changes. */
while ((ret = cb(db, &key, &data)) == 0) {
if (key->size == 0) {
__wt_api_db_errx(
db, "zero-length keys are not supported");
return (WT_ERROR);
}
WT_STAT_INCR(idb->stats, ITEMS_INSERTED);
WT_STAT_INCR(idb->stats, DUPLICATE_ITEMS_INSERTED);
/* Loading duplicates, so a key change means we're done. */
if (lastkey->size != key->size ||
db->btree_compare_dup(db, lastkey, key) != 0) {
*keyp = key;
*datap = data;
break;
}
/* Build the data item we're going to store on the page. */
WT_ERR(__wt_item_build_data(
toc, data, &data_item, &data_ovfl, WT_IS_DUP));
/*
* If there's insufficient space available, allocate a new
* page.
*/
if (WT_ITEM_SPACE_REQ(data->size) > space_avail) {
/*
* We've finished with the page: promote its first key
* to its parent and discard it, then switch to the new
* page.
*
* If we promoted a key, we might have split, and so
* there may be a new offpage duplicates root page.
*/
WT_RET(__wt_bulk_promote(toc,
page, page->records, &stack, 0, &root_addr));
WT_ERR(__wt_page_write(toc, page));
page->records = 0;
page->dsk->u.entries = 0;
__wt_set_ff_and_sa_from_offset(page,
WT_PAGE_BYTE(page), &first_free, &space_avail);
}
++dup_count; /* Total duplicate count */
++page->records; /* On-page key/data count */
++page->dsk->u.entries; /* On-page entry count */
/* Copy the data item onto the page. */
WT_ITEM_SET_LEN(&data_item, data->size);
memcpy(first_free, &data_item, sizeof(data_item));
memcpy(first_free + sizeof(data_item), data->data, data->size);
space_avail -= WT_ITEM_SPACE_REQ(data->size);
first_free += WT_ITEM_SPACE_REQ(data->size);
}
/*
* Ret values of 1 and 0 are both "OK", the ret value of 1 means we
* reached the end of the bulk input. Save the successful return
* for our final return value.
*/
if (ret != 0 && ret != 1)
goto err;
success_return = ret;
/* Promote a key from the partially-filled page and write it. */
WT_ERR(
__wt_bulk_promote(toc, page, page->records, &stack, 0, &root_addr));
WT_ERR(__wt_page_write(toc, page));
/* Fill in the caller's WT_OFF structure. */
WT_RECORDS(off) = dup_count;
off->addr = root_addr;
off->size = db->intlmin;
err: WT_TRET(__wt_bulk_stack_put(toc, &stack));
if (tmp != NULL)
__wt_scr_release(&tmp);
return (ret == 0 ? success_return : ret);
}
/*
* __wt_bulk_promote --
* Promote the first entry on a page to its parent.
*/
static int
__wt_bulk_promote(WT_TOC *toc, WT_PAGE *page, uint64_t incr,
WT_STACK *stack, u_int level, uint32_t *dup_root_addrp)
{
DB *db;
DBT *key, key_build, *next_tmp;
ENV *env;
WT_ITEM *key_item, item;
WT_OFF off;
WT_OVFL tmp_ovfl;
WT_PAGE *next, *parent;
WT_PAGE_DISK *dsk;
WT_STACK_ELEM *elem;
uint32_t next_space_avail;
uint8_t *next_first_free;
u_int type;
int need_promotion, ret;
void *parent_data;
db = toc->db;
env = toc->env;
dsk = page->dsk;
WT_CLEAR(item);
next_tmp = NULL;
next = parent = NULL;
ret = 0;
/*
* If it's a row-store, get a copy of the first item on the page -- it
* might be an overflow item, in which case we need to make a copy for
* the database. Most versions of Berkeley DB tried to reference count
* overflow items if they were promoted to internal pages. That turned
* out to be hard to get right, so I'm not doing it again.
*
* If it's a column-store page, we don't promote a key at all.
*/
switch (dsk->type) {
case WT_PAGE_DUP_INT:
case WT_PAGE_DUP_LEAF:
case WT_PAGE_ROW_INT:
case WT_PAGE_ROW_LEAF:
key = &key_build;
WT_CLEAR(key_build);
key_item = (WT_ITEM *)WT_PAGE_BYTE(page);
switch (WT_ITEM_TYPE(key_item)) {
case WT_ITEM_KEY:
case WT_ITEM_DATA_DUP:
key->data = WT_ITEM_BYTE(key_item);
key->size = WT_ITEM_LEN(key_item);
switch (dsk->type) {
case WT_PAGE_ROW_INT:
case WT_PAGE_ROW_LEAF:
WT_ITEM_SET(&item, WT_ITEM_KEY, key->size);
break;
case WT_PAGE_DUP_INT:
case WT_PAGE_DUP_LEAF:
WT_ITEM_SET(&item, WT_ITEM_KEY_DUP, key->size);
break;
default: /* Not possible */
break;
}
break;
case WT_ITEM_KEY_OVFL:
case WT_ITEM_DATA_DUP_OVFL:
/*
* Assume overflow keys remain overflow keys when they
* are promoted; not necessarily true if internal nodes
* are larger than leaf nodes), but that's unlikely.
*/
WT_CLEAR(tmp_ovfl);
WT_RET(__wt_bulk_ovfl_copy(toc,
WT_ITEM_BYTE_OVFL(key_item), &tmp_ovfl));
key->data = &tmp_ovfl;
key->size = sizeof(tmp_ovfl);
switch (dsk->type) {
case WT_PAGE_ROW_INT:
case WT_PAGE_ROW_LEAF:
WT_ITEM_SET(&item,
WT_ITEM_KEY_OVFL, sizeof(WT_OVFL));
break;
case WT_PAGE_DUP_INT:
case WT_PAGE_DUP_LEAF:
WT_ITEM_SET(&item,
WT_ITEM_KEY_DUP_OVFL, sizeof(WT_OVFL));
break;
default: /* Not possible */
break;
}
break;
WT_ILLEGAL_FORMAT(db);
}
break;
case WT_PAGE_COL_FIX:
case WT_PAGE_COL_INT:
case WT_PAGE_COL_RLE:
case WT_PAGE_COL_VAR:
key = NULL;
break;
WT_ILLEGAL_FORMAT(db);
}
/*
* There are two paths into this code based on whether the page already
* has a parent.
*
* If we have a page with no parent page, create the parent page. In
* this path, there's not much to do -- allocate a parent page, copy
* reference information from the page to the parent, and we're done.
* This is a modified root-split: we're putting a single key on an
* internal page, which is illegal, but we know another page on this
* page's level will be created, and it will be promoted to the parent
* at some point. This is case #1.
*
* The second path into this code is if we have a page and its parent,
* but the page's reference information doesn't fit on the parent and
* we have to split the parent. This path has two different cases,
* based on whether the page's parent itself has a parent.
*
* Here's a diagram of case #2, where the parent also has a parent:
*
* P2 -> P1 -> L (case #2)
*
* The promoted key from leaf L won't fit onto P1, and so we split P1:
*
* P2 -> P1
* -> P3 -> L
*
* In case #2, allocate P3 and copy reference information from the leaf
* page to it, then recursively call the promote code to promote the
* first entry from P3 to P2.
*
* Here's a diagram of case #3, where the parent does not have a parent,
* in other words, a root split:
*
* P1 -> L (case #3)
*
* The promoted key from leaf L won't fit onto P1, and so we split P1:
*
* P1 ->
* P2 -> L
*
* In case #3, we allocate P2, copy reference information from the page
* to it, and then recursively call the promote code twice: first to
* promote the first entry from P1 to a new page, and again to promote
* the first entry from P2 to a new page, creating a new root level of
* the tree:
*
* P3 -> P1
* -> P2 -> L
*/
/*
* To simplify the rest of the code, check to see if there's room for
* another entry in our stack structure. Allocate the stack in groups
* of 20, which is probably big enough for any tree we'll ever see in
* the field, we'll never test the realloc code unless we work at it.
*/
#ifdef HAVE_DIAGNOSTIC
#define WT_STACK_ALLOC_INCR 2
#else
#define WT_STACK_ALLOC_INCR 20
#endif
if (stack->size == 0 || level == stack->size - 1) {
uint32_t bytes_allocated = stack->size * sizeof(WT_STACK_ELEM);
WT_RET(__wt_realloc(env, &bytes_allocated,
(stack->size + WT_STACK_ALLOC_INCR) * sizeof(WT_STACK_ELEM),
&stack->elem));
stack->size += WT_STACK_ALLOC_INCR;
/*
* Note, the stack structure may be entirely uninitialized here,
* that is, everything set to 0 bytes. That's OK: the level of
* the stack starts out at 0, that is, the 0th element of the
* stack is the 1st level of internal/parent pages in the tree.
*/
}
elem = &stack->elem[level];
parent = elem->page;
if (parent == NULL) {
split: switch (dsk->type) {
case WT_PAGE_COL_FIX:
case WT_PAGE_COL_INT:
case WT_PAGE_COL_RLE:
case WT_PAGE_COL_VAR:
type = WT_PAGE_COL_INT;
break;
case WT_PAGE_DUP_INT:
case WT_PAGE_DUP_LEAF:
type = WT_PAGE_DUP_INT;
break;
case WT_PAGE_ROW_INT:
case WT_PAGE_ROW_LEAF:
type = WT_PAGE_ROW_INT;
break;
WT_ILLEGAL_FORMAT(db);
}
WT_ERR(__wt_bulk_scratch_page(
toc, db->intlmin, type, dsk->level + 1, &next, &next_tmp));
__wt_set_ff_and_sa_from_offset(next,
WT_PAGE_BYTE(next), &next_first_free, &next_space_avail);
/*
* Column stores set the starting record number to the starting
* record number of the promoted leaf -- the new leaf is always
* the first record in the new parent's page. Ignore the type
* of the database, it's simpler ot just promote 0 up the tree
* in row store databases.
*/
next->dsk->start_recno = page->dsk->start_recno;
/*
* If we don't have a parent page, it's case #1 -- allocate the
* parent page immediately.
*/
if (parent == NULL) {
/*
* Case #1 -- there's no parent, it's a root split. No
* additional work in the main tree. In an off-page
* duplicates tree, return the new root of the off-page
* tree.
*/
if (type == WT_PAGE_DUP_INT)
*dup_root_addrp = next->addr;
need_promotion = 0;
} else {
/*
* Case #2 and #3.
*
* Case #3: a root split, so we have to promote a key
* from both of the parent pages: promote the key from
* the existing parent page.
*/
if (stack->elem[level + 1].page == NULL)
WT_ERR(__wt_bulk_promote(toc, parent,
incr, stack, level + 1, dup_root_addrp));
need_promotion = 1;
/* Write the last parent page, we have a new one. */
WT_ERR(__wt_page_write(toc, parent));
__wt_scr_release(&stack->elem[level].tmp);
}
/* There's a new parent page, reset the stack. */
elem = &stack->elem[level];
elem->page = parent = next;
elem->first_free = next_first_free;
elem->space_avail = next_space_avail;
elem->tmp = next_tmp;
next = NULL;
next_first_free = NULL;
next_space_avail = 0;
next_tmp = NULL;
} else
need_promotion = 0;
/*
* See if the promoted data will fit (if they don't, we have to split).
* We don't need to check for overflow keys: if the key was an overflow,
* we already created a smaller, on-page version of it.
*
* If there's room, copy the promoted data onto the parent's page.
*/
switch (parent->dsk->type) {
case WT_PAGE_COL_INT:
if (elem->space_avail < sizeof(WT_OFF))
goto split;
/* Create the WT_OFF reference. */
WT_RECORDS(&off) = page->records;
off.addr = page->addr;
off.size = dsk->level == WT_LLEAF ? db->leafmin : db->intlmin;
/* Store the data item. */
++parent->dsk->u.entries;
parent_data = elem->first_free;
memcpy(elem->first_free, &off, sizeof(off));
elem->first_free += sizeof(WT_OFF);
elem->space_avail -= sizeof(WT_OFF);
/* Track the last entry on the page for record count updates. */
stack->elem[level].data = parent_data;
break;
case WT_PAGE_ROW_INT:
case WT_PAGE_DUP_INT:
if (elem->space_avail <
WT_ITEM_SPACE_REQ(sizeof(WT_OFF)) +
WT_ITEM_SPACE_REQ(key->size))
goto split;
/* Store the key. */
++parent->dsk->u.entries;
memcpy(elem->first_free, &item, sizeof(item));
memcpy(elem->first_free + sizeof(item), key->data, key->size);
elem->first_free += WT_ITEM_SPACE_REQ(key->size);
elem->space_avail -= WT_ITEM_SPACE_REQ(key->size);
/* Create the WT_ITEM(WT_OFF) reference. */
WT_ITEM_SET(&item, WT_ITEM_OFF, sizeof(WT_OFF));
WT_RECORDS(&off) = page->records;
off.addr = page->addr;
off.size = dsk->level == WT_LLEAF ? db->leafmin : db->intlmin;
/* Store the data item. */
++parent->dsk->u.entries;
parent_data = elem->first_free;
memcpy(elem->first_free, &item, sizeof(item));
memcpy(elem->first_free + sizeof(item), &off, sizeof(off));
elem->first_free += WT_ITEM_SPACE_REQ(sizeof(WT_OFF));
elem->space_avail -= WT_ITEM_SPACE_REQ(sizeof(WT_OFF));
/* Track the last entry on the page for record count updates. */
stack->elem[level].data = parent_data;
break;
WT_ILLEGAL_FORMAT(db);
}
parent->records += page->records;
/*
* The promotion for case #2 and the second part of case #3 -- promote
* the key from the newly allocated internal page to its parent.
*/
if (need_promotion)
WT_RET(__wt_bulk_promote(
toc, parent, incr, stack, level + 1, dup_root_addrp));
else {
/*
* We've finished promoting the new page's key into the tree.
* What remains is to push the new record counts all the way
* to the root. We've already corrected our current "parent"
* page, so proceed from there to the root.
*/
for (elem =
&stack->elem[level + 1]; elem->page != NULL; ++elem) {
switch (elem->page->dsk->type) {
case WT_PAGE_COL_INT:
WT_RECORDS((WT_OFF *)elem->data) += incr;
break;
case WT_PAGE_ROW_INT:
case WT_PAGE_DUP_INT:
WT_RECORDS(
(WT_OFF *)WT_ITEM_BYTE(elem->data)) += incr;
break;
WT_ILLEGAL_FORMAT(db);
}
elem->page->records += incr;
}
}
err: if (next_tmp != NULL)
__wt_scr_release(&next_tmp);
return (ret);
}
/*
* __wt_item_build_key --
* Process an inserted key item and return an WT_ITEM structure and byte
* string to be stored on the page.
*/
static int
__wt_item_build_key(WT_TOC *toc, DBT *dbt, WT_ITEM *item, WT_OVFL *ovfl)
{
DB *db;
IDB *idb;
WT_STATS *stats;
db = toc->db;
idb = db->idb;
stats = idb->stats;
/*
* We're called with a DBT that references a data/size pair. We can
* re-point that DBT's data and size fields to other memory, but we
* cannot allocate memory in that DBT -- all we can do is re-point it.
*
* For Huffman-encoded key/data items, we need a chunk of new space;
* use the WT_TOC key/data return memory: this routine is called during
* bulk insert and reconciliation, we aren't returning key/data pairs.
*/
/* Optionally compress the data using the Huffman engine. */
if (idb->huffman_key != NULL) {
WT_RET(__wt_huffman_encode(
idb->huffman_key, dbt->data, dbt->size,
&toc->key.data, &toc->key.mem_size, &toc->key.size));
if (toc->key.size > dbt->size)
WT_STAT_INCRV(stats,
HUFFMAN_KEY, toc->key.size - dbt->size);
dbt->data = toc->key.data;
dbt->size = toc->key.size;
}
/* Create an overflow object if the data won't fit. */
if (dbt->size > db->leafitemsize) {
WT_STAT_INCR(stats, OVERFLOW_KEY);
WT_RET(__wt_bulk_ovfl_write(toc, dbt, ovfl));
dbt->data = ovfl;
dbt->size = sizeof(*ovfl);
WT_ITEM_SET(item, WT_ITEM_KEY_OVFL, dbt->size);
} else
WT_ITEM_SET(item, WT_ITEM_KEY, dbt->size);
return (0);
}
/*
* __wt_item_build_data --
* Process an inserted data item and return an WT_ITEM structure and byte
* string to be stored on the page.
*/
int
__wt_item_build_data(
WT_TOC *toc, DBT *dbt, WT_ITEM *item, WT_OVFL *ovfl, u_int flags)
{
DB *db;
IDB *idb;
WT_STATS *stats;
WT_ENV_FCHK(toc->env,
"__wt_item_build_data", flags, WT_APIMASK_BT_BUILD_DATA_ITEM);
db = toc->db;
idb = db->idb;
stats = idb->stats;
/*
* We're called with a DBT that references a data/size pair. We can
* re-point that DBT's data and size fields to other memory, but we
* cannot allocate memory in that DBT -- all we can do is re-point it.
*
* For Huffman-encoded key/data items, we need a chunk of new space;
* use the WT_TOC key/data return memory: this routine is called during
* bulk insert and reconciliation, we aren't returning key/data pairs.
*/
WT_CLEAR(*item);
WT_ITEM_SET_TYPE(
item, LF_ISSET(WT_IS_DUP) ? WT_ITEM_DATA_DUP : WT_ITEM_DATA);
/*
* Handle zero-length items quickly -- this is a common value, it's
* a deleted column-store variable length item.
*/
if (dbt->size == 0) {
WT_ITEM_SET_LEN(item, 0);
return (0);
}
/* Optionally compress the data using the Huffman engine. */
if (idb->huffman_data != NULL) {
WT_RET(__wt_huffman_encode(
idb->huffman_data, dbt->data, dbt->size,
&toc->data.data, &toc->data.mem_size, &toc->data.size));
if (toc->data.size > dbt->size)
WT_STAT_INCRV(stats,
HUFFMAN_DATA, toc->data.size - dbt->size);
dbt->data = toc->data.data;
dbt->size = toc->data.size;
}
/* Create an overflow object if the data won't fit. */
if (dbt->size > db->leafitemsize) {
WT_RET(__wt_bulk_ovfl_write(toc, dbt, ovfl));
dbt->data = ovfl;
dbt->size = sizeof(*ovfl);
WT_ITEM_SET_TYPE(item, LF_ISSET(WT_IS_DUP) ?
WT_ITEM_DATA_DUP_OVFL : WT_ITEM_DATA_OVFL);
WT_STAT_INCR(stats, OVERFLOW_DATA);
}
WT_ITEM_SET_LEN(item, dbt->size);
return (0);
}
/*
* __wt_bulk_ovfl_copy --
* Copy bulk-loaded overflow items in the database, returning the WT_OVFL
* structure, filled in.
*/
static int
__wt_bulk_ovfl_copy(WT_TOC *toc, WT_OVFL *from, WT_OVFL *to)
{
DB *db;
DBT *tmp;
WT_PAGE *page;
uint32_t size;
int ret;
db = toc->db;
tmp = NULL;
/* Get a scratch buffer and make it look like an overflow page. */
size = WT_ALIGN(sizeof(WT_PAGE_DISK) + from->size, db->allocsize);
WT_RET(__wt_bulk_scratch_page(
toc, size, WT_PAGE_OVFL, WT_LLEAF, &page, &tmp));
page->dsk->u.datalen = from->size;
/* Fill in the return information. */
to->addr = page->addr;
to->size = from->size;
/*
* Read the page into our scratch buffer, then write it out to the
* new location.
*/
if ((ret =
__wt_page_disk_read(toc, page->dsk, from->addr, from->size)) == 0)
ret =
__wt_page_disk_write(toc, page->dsk, to->addr, from->size);
__wt_scr_release(&tmp);
return (ret);
}
/*
* __wt_bulk_ovfl_write --
* Store bulk-loaded overflow items in the database, returning the page
* addr.
*/
static int
__wt_bulk_ovfl_write(WT_TOC *toc, DBT *dbt, WT_OVFL *to)
{
DB *db;
DBT *tmp;
WT_PAGE *page;
WT_PAGE_DISK *dsk;
uint32_t size;
int ret;
db = toc->db;
tmp = NULL;
/* Get a scratch buffer and make it look like our work page. */
size = WT_ALIGN(sizeof(WT_PAGE_DISK) + dbt->size, db->allocsize);
WT_ERR(__wt_bulk_scratch_page(
toc, size, WT_PAGE_OVFL, WT_LLEAF, &page, &tmp));
/* Fill in the return information. */
to->addr = page->addr;
to->size = dbt->size;
/* Initialize the page header and copy the record into place. */
dsk = page->dsk;
dsk->u.datalen = dbt->size;
memcpy((uint8_t *)dsk + sizeof(WT_PAGE_DISK), dbt->data, dbt->size);
ret = __wt_page_write(toc, page);
err: if (tmp != NULL)
__wt_scr_release(&tmp);
return (ret);
}
/*
* __wt_bulk_scratch_page --
* Allocate a scratch buffer and make it look like a database page.
*/
static int
__wt_bulk_scratch_page(WT_TOC *toc, uint32_t page_size,
uint32_t page_type, uint32_t page_level, WT_PAGE **page_ret, DBT **tmp_ret)
{
DBT *tmp;
WT_PAGE *page;
WT_PAGE_DISK *dsk;
uint32_t size;
int ret;
ret = 0;
/*
* Allocate a scratch buffer and make sure it's big enough to hold a
* WT_PAGE structure plus the page itself, and clear the memory so
* it's never random bytes.
*/
size = page_size + sizeof(WT_PAGE);
WT_ERR(__wt_scr_alloc(toc, size, &tmp));
memset(tmp->data, 0, size);
/*
* Set up the page and allocate a file address.
*
* We don't run the leaf pages through the cache -- that means passing
* a lot of messages we don't want to bother with. We're the only user
* of the file, which means we can grab file space whenever we want.
*/
page = tmp->data;
page->dsk = dsk =
(WT_PAGE_DISK *)((uint8_t *)tmp->data + sizeof(WT_PAGE));
WT_ERR(__wt_file_alloc(toc, &page->addr, page_size));
page->size = page_size;
dsk->type = (uint8_t)page_type;
dsk->level = (uint8_t)page_level;
*page_ret = page;
*tmp_ret = tmp;
return (0);
err: if (tmp != NULL)
__wt_scr_release(&tmp);
return (ret);
}
/*
* __wt_bulk_stack_put --
* Push out the tree's stack of pages.
*/
static int
__wt_bulk_stack_put(WT_TOC *toc, WT_STACK *stack)
{
ENV *env;
IDB *idb;
WT_STACK_ELEM *elem;
int ret;
env = toc->env;
idb = toc->db->idb;
ret = 0;
for (elem = stack->elem; elem->page != NULL; ++elem) {
WT_TRET(__wt_page_write(toc, elem->page));
/*
* If we've reached the last element in the stack, it's the
* root page of the tree. Update the in-memory root address
* and the descriptor record.
*/
if ((elem + 1)->page == NULL) {
idb->root_off.addr = elem->page->addr;
idb->root_off.size = elem->page->size;
WT_RECORDS(&idb->root_off) = elem->page->records;
WT_TRET(__wt_desc_write(toc));
}
__wt_scr_release(&elem->tmp);
}
__wt_free(env, stack->elem, stack->size * sizeof(WT_STACK_ELEM));
return (0);
}
/*
* __wt_bulk_dbt_copy --
* Get a copy of DBT referenced object.
*/
static int
__wt_bulk_dbt_copy(ENV *env, DBT *orig, DBT *copy)
{
if (copy->mem_size < orig->size)
WT_RET(__wt_realloc(
env, ©->mem_size, orig->size, ©->data));
memcpy(copy->data, orig->data, orig->size);
copy->size = orig->size;
return (0);
}
|