summaryrefslogtreecommitdiffstats
path: root/src/btree.c
blob: 0fdc680f41cfb9bfcbeacb6b3dabd67663a3e8e5 (plain)
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
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623
6624
6625
6626
6627
6628
6629
6630
6631
6632
6633
6634
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666
6667
6668
6669
6670
6671
6672
6673
6674
6675
6676
6677
6678
6679
6680
6681
6682
6683
6684
6685
6686
6687
6688
6689
6690
6691
6692
6693
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709
6710
6711
6712
6713
6714
6715
6716
6717
6718
6719
6720
6721
6722
6723
6724
6725
6726
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736
6737
6738
6739
6740
6741
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770
6771
6772
6773
6774
6775
6776
6777
6778
6779
6780
6781
6782
6783
6784
6785
6786
6787
6788
6789
6790
6791
6792
6793
6794
6795
6796
6797
6798
6799
6800
6801
6802
6803
6804
6805
6806
6807
6808
6809
6810
6811
6812
6813
6814
6815
6816
6817
6818
6819
6820
6821
6822
6823
6824
6825
6826
6827
6828
6829
6830
6831
6832
6833
6834
6835
6836
6837
6838
6839
6840
6841
6842
6843
6844
6845
6846
6847
6848
6849
6850
6851
6852
6853
6854
6855
6856
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869
6870
6871
6872
6873
6874
6875
6876
6877
6878
6879
6880
6881
6882
6883
6884
6885
6886
6887
6888
6889
6890
6891
6892
6893
6894
6895
6896
6897
6898
6899
6900
6901
6902
6903
6904
6905
6906
6907
6908
6909
6910
6911
6912
6913
6914
6915
6916
6917
6918
6919
6920
6921
6922
6923
6924
6925
6926
6927
6928
6929
6930
6931
6932
6933
6934
6935
6936
6937
6938
6939
6940
6941
6942
6943
6944
6945
6946
6947
6948
6949
6950
6951
6952
6953
6954
6955
6956
6957
6958
6959
6960
6961
6962
6963
6964
6965
6966
6967
6968
6969
6970
6971
6972
6973
6974
6975
6976
6977
6978
6979
6980
6981
6982
6983
6984
6985
6986
6987
6988
6989
6990
6991
6992
6993
6994
6995
6996
6997
6998
6999
7000
7001
7002
7003
7004
7005
7006
7007
7008
7009
7010
7011
7012
7013
7014
7015
7016
7017
7018
7019
7020
7021
7022
7023
7024
7025
7026
7027
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037
7038
7039
7040
7041
7042
7043
7044
7045
7046
7047
7048
7049
7050
7051
7052
7053
7054
7055
7056
7057
7058
7059
7060
7061
7062
7063
7064
7065
7066
7067
7068
7069
7070
7071
7072
7073
7074
7075
7076
7077
7078
7079
7080
7081
7082
7083
7084
7085
7086
7087
7088
7089
7090
7091
7092
7093
7094
7095
7096
7097
7098
7099
7100
7101
7102
7103
7104
7105
7106
7107
7108
7109
7110
7111
7112
7113
7114
7115
7116
7117
7118
7119
7120
7121
7122
7123
7124
7125
7126
7127
7128
7129
7130
7131
7132
7133
7134
7135
7136
7137
7138
7139
7140
7141
7142
7143
7144
7145
7146
7147
7148
7149
7150
7151
7152
7153
7154
7155
7156
7157
7158
7159
7160
7161
7162
7163
7164
7165
7166
7167
7168
7169
7170
7171
7172
7173
7174
7175
7176
7177
7178
7179
7180
7181
7182
7183
7184
7185
7186
7187
7188
7189
7190
7191
7192
7193
7194
7195
7196
7197
7198
7199
7200
7201
7202
7203
7204
7205
7206
7207
7208
7209
7210
7211
7212
7213
7214
7215
7216
7217
7218
7219
7220
7221
7222
7223
7224
7225
7226
7227
7228
7229
7230
7231
7232
7233
7234
7235
7236
7237
7238
7239
7240
7241
7242
7243
7244
7245
7246
7247
7248
7249
7250
7251
7252
7253
7254
7255
7256
7257
7258
7259
7260
7261
7262
7263
7264
7265
7266
7267
7268
7269
7270
7271
7272
7273
7274
7275
7276
7277
7278
7279
7280
7281
7282
7283
7284
7285
7286
7287
7288
7289
7290
7291
7292
7293
7294
7295
7296
7297
7298
7299
7300
7301
7302
7303
7304
7305
7306
7307
7308
7309
7310
7311
7312
7313
7314
7315
7316
7317
7318
7319
7320
7321
7322
7323
7324
7325
7326
7327
7328
7329
7330
7331
7332
7333
7334
7335
7336
7337
7338
7339
7340
7341
7342
7343
7344
7345
7346
7347
7348
7349
7350
7351
7352
7353
7354
7355
7356
7357
7358
7359
7360
7361
7362
7363
7364
7365
7366
7367
7368
7369
7370
7371
7372
7373
7374
7375
7376
7377
7378
7379
7380
7381
7382
7383
7384
7385
7386
7387
7388
7389
7390
7391
7392
7393
7394
7395
7396
7397
7398
7399
7400
7401
7402
7403
7404
7405
7406
7407
7408
7409
7410
7411
7412
7413
7414
7415
7416
7417
7418
7419
7420
7421
7422
7423
7424
7425
7426
7427
7428
7429
7430
7431
7432
7433
7434
7435
7436
7437
7438
7439
7440
7441
7442
7443
7444
7445
7446
7447
7448
7449
7450
7451
7452
7453
7454
7455
7456
7457
7458
7459
7460
7461
7462
7463
7464
7465
7466
7467
7468
7469
7470
7471
7472
7473
7474
7475
7476
7477
7478
7479
7480
7481
7482
7483
7484
7485
7486
7487
7488
7489
7490
7491
7492
7493
7494
7495
7496
7497
7498
7499
7500
7501
7502
7503
7504
7505
7506
7507
7508
7509
7510
7511
7512
7513
7514
7515
7516
7517
7518
7519
7520
7521
7522
7523
7524
7525
7526
7527
7528
7529
7530
7531
7532
7533
7534
7535
7536
7537
7538
7539
7540
7541
7542
7543
7544
7545
7546
7547
7548
7549
7550
7551
7552
7553
7554
7555
7556
7557
7558
7559
7560
7561
7562
7563
7564
7565
7566
7567
7568
7569
7570
7571
7572
7573
7574
7575
7576
7577
7578
7579
7580
7581
7582
7583
7584
7585
7586
7587
7588
7589
7590
7591
7592
7593
7594
7595
7596
7597
7598
7599
7600
7601
7602
7603
7604
7605
7606
7607
7608
7609
7610
7611
7612
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622
7623
7624
7625
7626
7627
7628
7629
7630
7631
7632
7633
7634
7635
7636
7637
7638
7639
7640
7641
7642
7643
7644
7645
7646
7647
7648
7649
7650
7651
7652
7653
7654
7655
7656
7657
7658
7659
7660
7661
7662
7663
7664
7665
7666
7667
7668
7669
7670
7671
7672
7673
7674
7675
7676
7677
7678
7679
7680
7681
7682
7683
7684
7685
7686
7687
7688
7689
7690
7691
7692
7693
7694
7695
7696
7697
7698
7699
7700
7701
7702
7703
7704
7705
7706
7707
7708
7709
7710
7711
7712
7713
7714
7715
7716
7717
7718
7719
7720
7721
7722
7723
7724
7725
7726
7727
7728
7729
7730
7731
7732
7733
7734
7735
7736
7737
7738
7739
7740
7741
7742
7743
7744
7745
7746
7747
7748
7749
7750
7751
7752
7753
7754
7755
7756
7757
7758
7759
7760
7761
7762
7763
7764
7765
7766
7767
7768
7769
7770
7771
7772
7773
7774
7775
7776
7777
7778
7779
7780
7781
7782
7783
7784
7785
7786
7787
7788
7789
7790
7791
7792
7793
7794
7795
7796
7797
7798
7799
7800
7801
7802
7803
7804
7805
7806
7807
7808
7809
7810
7811
7812
7813
7814
7815
7816
7817
7818
7819
7820
7821
7822
7823
7824
7825
7826
7827
7828
7829
7830
7831
7832
7833
7834
7835
7836
7837
7838
7839
7840
7841
7842
7843
7844
7845
7846
7847
7848
7849
7850
7851
7852
7853
7854
7855
7856
7857
7858
7859
7860
7861
7862
7863
7864
7865
7866
7867
7868
7869
7870
7871
7872
7873
7874
7875
7876
7877
7878
7879
7880
7881
7882
7883
7884
7885
7886
7887
7888
7889
7890
7891
7892
7893
7894
7895
7896
7897
7898
7899
7900
7901
7902
7903
7904
7905
7906
7907
7908
7909
7910
7911
7912
7913
7914
7915
7916
7917
7918
7919
7920
7921
7922
7923
7924
7925
7926
7927
7928
7929
7930
7931
7932
7933
7934
7935
7936
7937
7938
7939
7940
7941
7942
7943
7944
7945
7946
7947
7948
7949
7950
7951
7952
7953
7954
7955
7956
7957
7958
7959
7960
7961
7962
7963
7964
7965
7966
7967
7968
7969
7970
7971
7972
7973
7974
7975
7976
7977
7978
7979
7980
7981
7982
7983
7984
7985
7986
7987
7988
7989
7990
7991
7992
7993
7994
7995
7996
7997
7998
7999
8000
8001
8002
8003
8004
8005
8006
8007
8008
8009
8010
8011
8012
8013
8014
8015
8016
8017
8018
8019
8020
8021
8022
8023
8024
8025
8026
8027
8028
8029
8030
8031
8032
8033
8034
8035
8036
8037
8038
8039
8040
8041
8042
8043
8044
8045
8046
8047
8048
8049
8050
8051
8052
8053
8054
8055
8056
8057
8058
8059
8060
8061
8062
8063
8064
8065
8066
8067
8068
8069
8070
8071
8072
8073
8074
8075
8076
8077
8078
8079
8080
8081
8082
8083
8084
8085
8086
8087
8088
8089
8090
8091
8092
8093
8094
8095
8096
8097
8098
8099
8100
8101
8102
8103
8104
8105
8106
8107
8108
8109
8110
8111
8112
8113
8114
8115
8116
8117
8118
8119
8120
8121
8122
8123
8124
8125
8126
8127
8128
8129
8130
8131
8132
8133
8134
8135
8136
8137
8138
8139
8140
8141
8142
8143
8144
8145
8146
8147
8148
8149
8150
8151
8152
8153
8154
8155
8156
8157
8158
8159
8160
8161
8162
8163
8164
8165
8166
8167
8168
8169
8170
8171
8172
8173
8174
8175
8176
8177
8178
8179
8180
8181
8182
8183
8184
8185
8186
8187
8188
8189
8190
8191
8192
8193
8194
8195
8196
8197
8198
8199
8200
8201
8202
8203
8204
8205
8206
8207
8208
8209
8210
8211
8212
8213
8214
8215
8216
8217
8218
8219
8220
8221
8222
8223
8224
8225
8226
8227
8228
8229
8230
8231
8232
8233
8234
8235
8236
8237
8238
8239
8240
8241
8242
8243
8244
8245
8246
8247
8248
8249
8250
8251
8252
8253
8254
8255
8256
8257
8258
8259
8260
8261
8262
8263
8264
8265
8266
8267
8268
8269
8270
8271
8272
8273
8274
8275
8276
8277
8278
8279
8280
8281
8282
8283
8284
8285
8286
8287
8288
8289
8290
8291
8292
8293
8294
8295
8296
8297
8298
8299
8300
8301
8302
8303
8304
8305
8306
8307
8308
8309
8310
8311
8312
8313
8314
8315
8316
8317
8318
8319
8320
8321
8322
8323
8324
8325
8326
8327
8328
8329
8330
8331
8332
8333
8334
8335
8336
8337
8338
8339
8340
8341
8342
8343
8344
8345
8346
8347
8348
8349
8350
8351
8352
8353
8354
8355
8356
8357
8358
8359
8360
8361
8362
8363
8364
8365
8366
8367
8368
8369
8370
8371
8372
8373
8374
8375
8376
8377
8378
8379
8380
8381
8382
8383
8384
8385
8386
8387
8388
8389
8390
8391
8392
8393
8394
8395
8396
8397
8398
8399
8400
8401
8402
8403
8404
8405
8406
8407
8408
8409
8410
8411
8412
8413
8414
8415
8416
8417
8418
8419
8420
8421
8422
8423
8424
8425
8426
8427
8428
8429
8430
8431
8432
8433
8434
8435
8436
8437
8438
8439
8440
8441
8442
8443
8444
8445
8446
8447
8448
8449
8450
8451
8452
8453
8454
8455
8456
8457
8458
8459
8460
8461
8462
8463
8464
8465
8466
8467
8468
8469
8470
8471
8472
8473
8474
8475
8476
8477
8478
8479
8480
8481
8482
8483
8484
8485
8486
8487
8488
8489
8490
8491
8492
8493
8494
8495
8496
8497
8498
8499
8500
8501
8502
8503
8504
8505
8506
8507
8508
8509
8510
8511
8512
8513
8514
8515
8516
8517
8518
8519
8520
8521
8522
8523
8524
8525
8526
8527
8528
8529
8530
8531
8532
8533
8534
8535
8536
8537
8538
8539
8540
8541
8542
8543
8544
8545
8546
8547
8548
8549
8550
8551
8552
8553
8554
8555
8556
8557
8558
8559
8560
8561
8562
8563
8564
8565
8566
8567
8568
8569
8570
8571
8572
8573
8574
8575
8576
8577
8578
8579
8580
8581
8582
8583
8584
8585
8586
8587
8588
8589
8590
8591
8592
8593
8594
8595
8596
8597
8598
8599
8600
8601
8602
8603
8604
8605
8606
8607
8608
8609
8610
8611
8612
8613
8614
8615
8616
8617
8618
8619
8620
8621
8622
8623
8624
8625
8626
8627
8628
8629
8630
8631
8632
8633
8634
8635
8636
8637
8638
8639
8640
8641
8642
8643
8644
8645
8646
8647
8648
8649
8650
8651
8652
8653
8654
8655
8656
8657
8658
8659
8660
8661
8662
8663
8664
8665
8666
8667
8668
8669
8670
8671
8672
8673
8674
8675
8676
8677
8678
8679
8680
8681
8682
8683
8684
8685
8686
8687
8688
8689
8690
8691
8692
8693
8694
8695
8696
8697
8698
8699
8700
8701
8702
8703
8704
8705
8706
8707
8708
8709
8710
8711
8712
8713
8714
8715
8716
8717
8718
8719
8720
8721
8722
8723
8724
8725
8726
8727
8728
8729
8730
8731
8732
8733
8734
8735
8736
8737
8738
8739
8740
8741
8742
8743
8744
8745
8746
8747
8748
8749
8750
8751
8752
8753
8754
8755
8756
8757
8758
8759
8760
8761
8762
8763
8764
8765
8766
8767
8768
8769
8770
8771
8772
8773
8774
8775
8776
8777
8778
8779
8780
8781
8782
8783
8784
8785
8786
8787
8788
8789
8790
8791
8792
8793
8794
8795
8796
8797
8798
8799
8800
8801
8802
8803
8804
8805
8806
8807
8808
8809
8810
8811
8812
8813
8814
8815
8816
8817
8818
8819
8820
8821
8822
8823
8824
8825
8826
8827
8828
8829
8830
8831
8832
8833
8834
8835
8836
8837
8838
8839
8840
8841
8842
8843
8844
8845
8846
8847
8848
8849
8850
8851
8852
8853
8854
8855
8856
8857
8858
8859
8860
8861
8862
8863
8864
8865
8866
8867
8868
8869
8870
8871
8872
8873
8874
8875
8876
8877
8878
8879
8880
8881
8882
8883
8884
8885
8886
8887
8888
8889
8890
8891
8892
8893
8894
8895
8896
8897
8898
8899
8900
8901
8902
8903
8904
8905
8906
8907
8908
8909
8910
8911
8912
8913
8914
8915
8916
8917
8918
8919
8920
8921
8922
8923
8924
8925
8926
8927
8928
8929
8930
8931
8932
8933
8934
8935
8936
8937
8938
8939
8940
8941
8942
8943
8944
8945
8946
8947
8948
8949
8950
8951
8952
8953
8954
8955
8956
8957
8958
8959
8960
8961
8962
8963
8964
8965
8966
8967
8968
8969
8970
8971
8972
8973
8974
8975
8976
8977
8978
8979
8980
8981
8982
8983
8984
8985
8986
8987
8988
8989
8990
8991
8992
8993
8994
8995
8996
8997
8998
8999
9000
9001
9002
9003
9004
9005
9006
9007
9008
9009
9010
9011
9012
9013
9014
9015
9016
9017
9018
9019
9020
9021
9022
9023
9024
9025
9026
9027
9028
9029
9030
9031
9032
9033
9034
9035
9036
9037
9038
9039
9040
9041
9042
9043
9044
9045
9046
9047
9048
9049
9050
9051
9052
9053
9054
9055
9056
9057
9058
9059
9060
9061
9062
9063
9064
9065
9066
9067
9068
9069
9070
9071
9072
9073
9074
9075
9076
9077
9078
9079
9080
9081
9082
9083
9084
9085
9086
9087
9088
9089
9090
9091
9092
9093
9094
9095
9096
9097
9098
9099
9100
9101
9102
9103
9104
9105
9106
9107
9108
9109
9110
9111
9112
9113
9114
9115
9116
9117
9118
9119
9120
9121
9122
9123
9124
9125
9126
9127
9128
9129
9130
9131
9132
9133
9134
9135
9136
9137
9138
9139
9140
9141
9142
9143
9144
9145
9146
9147
9148
9149
9150
9151
9152
9153
9154
9155
9156
9157
9158
9159
9160
9161
9162
9163
9164
9165
9166
9167
9168
9169
9170
9171
9172
9173
9174
9175
9176
9177
9178
9179
9180
9181
9182
9183
9184
9185
9186
9187
9188
9189
9190
9191
9192
9193
9194
9195
9196
9197
9198
9199
9200
9201
9202
9203
9204
9205
9206
9207
9208
9209
9210
9211
9212
9213
9214
9215
9216
9217
9218
9219
9220
9221
9222
9223
9224
9225
9226
9227
9228
9229
9230
9231
9232
9233
9234
9235
9236
9237
9238
9239
9240
9241
9242
9243
9244
9245
9246
9247
9248
9249
9250
9251
9252
9253
9254
9255
9256
9257
9258
9259
9260
9261
9262
9263
9264
9265
9266
9267
9268
9269
9270
9271
9272
9273
9274
9275
9276
9277
9278
9279
9280
9281
9282
9283
9284
9285
9286
9287
9288
9289
9290
9291
9292
9293
9294
9295
9296
9297
9298
9299
9300
9301
9302
9303
9304
9305
9306
9307
9308
9309
9310
9311
9312
9313
9314
9315
9316
9317
9318
9319
9320
9321
9322
9323
9324
9325
9326
9327
9328
9329
9330
9331
9332
9333
9334
9335
9336
9337
9338
9339
9340
9341
9342
9343
9344
9345
9346
9347
9348
9349
9350
9351
9352
9353
9354
9355
9356
9357
9358
9359
9360
9361
9362
9363
9364
9365
9366
9367
9368
9369
9370
9371
9372
9373
9374
9375
9376
9377
9378
9379
9380
9381
9382
9383
9384
9385
9386
9387
9388
9389
9390
9391
9392
9393
9394
9395
9396
9397
9398
9399
9400
9401
9402
9403
9404
9405
9406
9407
9408
9409
9410
9411
9412
9413
9414
9415
9416
9417
9418
9419
9420
9421
9422
9423
9424
9425
9426
9427
9428
9429
9430
9431
9432
9433
9434
9435
9436
9437
9438
9439
9440
9441
9442
9443
9444
9445
9446
9447
9448
9449
9450
9451
9452
9453
9454
9455
9456
9457
9458
9459
9460
9461
9462
9463
9464
9465
9466
9467
9468
9469
9470
9471
9472
9473
9474
9475
9476
9477
9478
9479
9480
9481
9482
9483
9484
9485
9486
9487
9488
9489
9490
9491
9492
9493
9494
9495
9496
9497
9498
9499
9500
9501
9502
9503
9504
9505
9506
9507
9508
9509
9510
9511
9512
9513
9514
9515
9516
9517
9518
9519
9520
9521
9522
9523
9524
9525
9526
9527
9528
9529
9530
9531
9532
9533
9534
9535
9536
9537
9538
9539
9540
9541
9542
9543
9544
9545
9546
9547
9548
9549
9550
9551
9552
9553
9554
9555
9556
9557
9558
9559
9560
9561
9562
9563
9564
9565
9566
9567
9568
9569
9570
9571
9572
9573
9574
9575
9576
9577
9578
9579
9580
9581
9582
9583
9584
9585
9586
9587
9588
9589
9590
9591
9592
9593
9594
9595
9596
9597
9598
9599
9600
9601
9602
9603
9604
9605
9606
9607
9608
9609
9610
9611
9612
9613
9614
9615
9616
9617
9618
9619
9620
9621
9622
9623
9624
9625
9626
9627
9628
9629
9630
9631
9632
9633
9634
9635
9636
9637
9638
9639
9640
9641
9642
9643
9644
9645
9646
9647
9648
9649
9650
9651
9652
9653
9654
9655
9656
9657
9658
9659
9660
9661
9662
9663
9664
9665
9666
9667
9668
9669
9670
9671
9672
9673
9674
9675
9676
9677
9678
9679
9680
9681
9682
9683
9684
9685
9686
9687
9688
9689
9690
9691
9692
9693
9694
9695
9696
9697
9698
9699
9700
9701
9702
9703
9704
9705
9706
9707
9708
9709
9710
9711
9712
9713
9714
9715
9716
9717
9718
9719
9720
9721
9722
9723
9724
9725
9726
9727
9728
9729
9730
9731
9732
9733
9734
9735
9736
9737
9738
9739
9740
9741
9742
9743
9744
9745
9746
9747
9748
9749
9750
9751
9752
9753
9754
9755
9756
9757
9758
9759
9760
9761
9762
9763
9764
9765
9766
9767
9768
9769
9770
9771
9772
9773
9774
9775
9776
9777
9778
9779
9780
9781
9782
9783
9784
9785
9786
9787
9788
9789
9790
9791
9792
9793
9794
9795
9796
9797
9798
9799
9800
9801
9802
9803
9804
9805
9806
9807
9808
9809
9810
9811
9812
9813
9814
9815
9816
9817
9818
9819
9820
9821
9822
9823
9824
9825
9826
9827
9828
9829
9830
9831
9832
9833
9834
9835
9836
9837
9838
9839
9840
9841
9842
9843
9844
9845
9846
9847
9848
9849
9850
9851
9852
9853
9854
9855
9856
9857
9858
9859
9860
9861
9862
9863
9864
9865
9866
9867
9868
9869
9870
9871
9872
9873
9874
9875
9876
9877
9878
9879
9880
9881
9882
9883
9884
9885
9886
9887
9888
9889
9890
9891
9892
9893
9894
9895
9896
9897
9898
9899
9900
9901
9902
9903
9904
9905
9906
9907
9908
9909
9910
9911
9912
9913
9914
9915
9916
9917
9918
9919
9920
9921
9922
9923
9924
9925
9926
9927
9928
9929
9930
9931
9932
9933
9934
9935
9936
9937
9938
9939
9940
9941
9942
9943
9944
9945
9946
9947
9948
9949
9950
9951
9952
9953
9954
9955
9956
9957
9958
9959
9960
9961
9962
9963
9964
9965
9966
9967
9968
9969
9970
9971
9972
9973
9974
9975
9976
9977
9978
9979
9980
9981
9982
9983
9984
9985
9986
9987
9988
9989
9990
9991
9992
9993
9994
9995
9996
9997
9998
9999
10000
10001
10002
10003
10004
10005
10006
10007
10008
10009
10010
10011
10012
10013
10014
10015
10016
10017
10018
10019
10020
10021
10022
10023
10024
10025
10026
10027
10028
10029
10030
10031
10032
10033
10034
10035
10036
10037
10038
10039
10040
10041
10042
10043
10044
10045
10046
10047
10048
10049
10050
10051
10052
10053
10054
10055
10056
10057
10058
10059
10060
10061
10062
10063
10064
10065
10066
10067
10068
10069
10070
10071
10072
10073
10074
10075
10076
10077
10078
10079
10080
10081
10082
10083
10084
10085
10086
10087
10088
10089
10090
10091
10092
10093
10094
10095
10096
10097
10098
10099
10100
10101
10102
10103
10104
10105
10106
10107
10108
10109
10110
10111
10112
10113
10114
10115
10116
10117
10118
10119
10120
10121
10122
10123
10124
10125
10126
10127
10128
10129
10130
10131
10132
10133
10134
10135
10136
10137
10138
10139
10140
10141
10142
10143
10144
10145
10146
10147
10148
10149
10150
10151
10152
10153
10154
10155
10156
10157
10158
10159
10160
10161
10162
10163
10164
10165
10166
10167
10168
10169
10170
10171
10172
10173
10174
10175
10176
10177
10178
10179
10180
10181
10182
10183
10184
10185
10186
10187
10188
10189
10190
10191
10192
10193
10194
10195
10196
10197
10198
10199
10200
10201
10202
10203
10204
10205
10206
10207
10208
10209
10210
10211
10212
10213
10214
10215
10216
10217
10218
10219
10220
10221
10222
10223
10224
10225
10226
10227
10228
10229
10230
10231
10232
10233
10234
10235
10236
10237
10238
10239
10240
10241
10242
10243
10244
10245
10246
10247
10248
10249
10250
10251
10252
10253
10254
10255
10256
10257
10258
10259
10260
10261
10262
10263
10264
10265
10266
10267
10268
10269
10270
10271
10272
10273
10274
10275
10276
10277
10278
10279
10280
10281
10282
10283
10284
10285
10286
10287
10288
10289
10290
10291
10292
10293
10294
10295
10296
10297
10298
10299
10300
10301
10302
10303
10304
10305
10306
10307
10308
10309
10310
10311
10312
10313
10314
10315
10316
10317
10318
10319
10320
10321
10322
10323
10324
10325
10326
10327
10328
10329
10330
10331
10332
10333
10334
10335
10336
10337
10338
10339
10340
10341
10342
10343
10344
10345
10346
10347
10348
10349
10350
10351
10352
10353
10354
10355
10356
10357
10358
10359
10360
10361
10362
10363
10364
10365
10366
10367
10368
10369
10370
10371
10372
10373
10374
10375
10376
10377
10378
10379
10380
10381
10382
10383
10384
10385
10386
10387
10388
10389
10390
10391
10392
10393
10394
10395
10396
10397
10398
10399
10400
10401
10402
10403
10404
10405
10406
10407
10408
10409
10410
10411
10412
10413
10414
10415
10416
10417
10418
10419
10420
10421
10422
10423
10424
10425
10426
10427
10428
10429
10430
10431
10432
10433
10434
10435
10436
10437
10438
10439
10440
10441
10442
10443
10444
10445
10446
10447
10448
10449
10450
10451
10452
10453
10454
10455
10456
10457
10458
10459
10460
10461
10462
10463
10464
10465
10466
10467
10468
10469
10470
10471
10472
10473
10474
10475
10476
10477
10478
10479
10480
10481
10482
10483
10484
10485
10486
10487
10488
10489
10490
10491
10492
10493
10494
10495
10496
10497
10498
10499
10500
10501
10502
10503
10504
10505
10506
10507
10508
10509
10510
10511
10512
10513
10514
10515
10516
10517
10518
10519
10520
10521
10522
10523
10524
10525
10526
10527
10528
10529
10530
10531
10532
10533
10534
10535
10536
10537
10538
10539
10540
10541
10542
10543
10544
10545
10546
10547
10548
10549
10550
10551
10552
10553
10554
10555
10556
10557
10558
10559
10560
10561
10562
10563
10564
10565
10566
10567
10568
10569
10570
10571
10572
10573
10574
10575
10576
10577
10578
10579
10580
10581
10582
10583
10584
10585
10586
10587
10588
10589
10590
10591
10592
10593
10594
10595
10596
10597
10598
10599
10600
10601
10602
10603
10604
10605
10606
10607
10608
10609
10610
10611
10612
10613
10614
10615
10616
10617
10618
10619
10620
10621
10622
10623
10624
10625
10626
10627
10628
10629
10630
10631
10632
10633
10634
10635
10636
10637
10638
10639
10640
10641
10642
10643
10644
10645
10646
10647
10648
10649
10650
10651
10652
10653
10654
10655
10656
10657
10658
10659
10660
10661
10662
10663
10664
10665
10666
10667
10668
10669
10670
10671
10672
10673
10674
10675
10676
10677
10678
10679
10680
10681
10682
10683
10684
10685
10686
10687
10688
10689
10690
10691
10692
10693
10694
10695
10696
10697
10698
10699
10700
10701
10702
10703
10704
10705
10706
10707
10708
10709
10710
10711
10712
10713
10714
10715
10716
10717
10718
10719
10720
10721
10722
10723
10724
10725
10726
10727
10728
10729
10730
10731
10732
10733
10734
10735
10736
10737
10738
10739
10740
10741
10742
10743
10744
10745
10746
10747
10748
10749
10750
10751
10752
10753
10754
10755
10756
10757
10758
10759
10760
10761
10762
10763
10764
10765
10766
10767
10768
10769
10770
10771
10772
10773
10774
10775
10776
10777
10778
10779
10780
10781
10782
10783
10784
10785
10786
10787
10788
10789
10790
10791
10792
10793
10794
10795
10796
10797
10798
10799
10800
10801
10802
10803
10804
10805
10806
10807
10808
10809
10810
10811
10812
10813
10814
10815
10816
10817
10818
10819
10820
10821
10822
10823
10824
10825
10826
10827
10828
10829
10830
10831
10832
10833
10834
10835
10836
10837
10838
10839
10840
10841
10842
10843
10844
10845
10846
10847
10848
10849
10850
10851
10852
10853
10854
10855
10856
10857
10858
10859
10860
10861
10862
10863
10864
10865
10866
10867
10868
10869
10870
10871
10872
10873
10874
10875
10876
10877
10878
10879
10880
10881
10882
10883
10884
10885
10886
10887
10888
10889
10890
10891
10892
10893
10894
10895
10896
10897
10898
10899
10900
10901
10902
10903
10904
10905
10906
10907
10908
10909
10910
10911
10912
10913
10914
10915
10916
10917
10918
10919
10920
10921
10922
10923
10924
10925
10926
10927
10928
10929
10930
10931
10932
10933
10934
10935
10936
10937
10938
10939
10940
10941
10942
10943
10944
10945
10946
10947
10948
10949
10950
10951
10952
10953
10954
10955
10956
10957
10958
10959
10960
10961
10962
10963
10964
10965
10966
10967
10968
10969
10970
10971
10972
10973
10974
10975
10976
10977
10978
10979
10980
10981
10982
10983
10984
10985
10986
10987
10988
10989
10990
10991
10992
10993
10994
10995
10996
10997
10998
10999
11000
11001
11002
11003
11004
11005
11006
11007
11008
11009
11010
11011
11012
11013
11014
11015
11016
11017
11018
11019
11020
11021
11022
11023
11024
11025
11026
11027
11028
11029
11030
11031
11032
11033
11034
11035
11036
11037
11038
11039
11040
11041
11042
11043
11044
11045
11046
11047
11048
11049
11050
11051
11052
11053
11054
11055
11056
11057
11058
11059
11060
11061
11062
11063
11064
11065
11066
11067
11068
11069
11070
11071
11072
11073
11074
11075
11076
11077
11078
11079
11080
11081
11082
11083
11084
11085
11086
11087
11088
11089
11090
11091
11092
11093
11094
11095
11096
11097
11098
11099
11100
11101
11102
11103
11104
11105
11106
11107
11108
11109
11110
11111
11112
11113
11114
11115
11116
11117
11118
11119
11120
11121
11122
11123
11124
11125
11126
11127
11128
11129
11130
11131
11132
11133
11134
11135
11136
11137
11138
11139
11140
11141
11142
11143
11144
11145
11146
11147
11148
11149
11150
11151
11152
11153
11154
11155
11156
11157
11158
11159
11160
11161
11162
11163
11164
11165
11166
11167
11168
11169
11170
11171
11172
11173
11174
11175
11176
11177
11178
11179
11180
11181
11182
11183
11184
11185
11186
11187
11188
11189
11190
11191
11192
11193
11194
11195
11196
11197
11198
11199
11200
11201
11202
11203
11204
11205
11206
11207
11208
11209
11210
11211
11212
11213
11214
11215
11216
11217
11218
11219
11220
11221
11222
11223
11224
11225
11226
11227
11228
11229
11230
11231
11232
11233
11234
11235
11236
11237
11238
11239
11240
11241
11242
11243
11244
11245
11246
11247
11248
11249
11250
11251
11252
11253
11254
11255
11256
11257
11258
11259
11260
11261
11262
11263
11264
11265
11266
11267
11268
11269
11270
11271
11272
11273
11274
11275
11276
11277
11278
11279
11280
11281
11282
11283
11284
11285
11286
11287
11288
11289
11290
11291
11292
11293
11294
11295
11296
11297
11298
11299
11300
11301
11302
11303
11304
11305
11306
11307
11308
11309
11310
11311
11312
11313
11314
11315
11316
11317
11318
11319
11320
11321
11322
11323
11324
11325
11326
11327
11328
11329
11330
11331
11332
11333
11334
11335
11336
11337
11338
11339
11340
11341
11342
11343
11344
11345
11346
11347
11348
11349
11350
11351
11352
11353
11354
11355
11356
11357
11358
11359
11360
11361
11362
11363
11364
11365
11366
11367
11368
11369
11370
11371
11372
11373
11374
11375
11376
11377
11378
11379
11380
11381
11382
11383
11384
11385
11386
11387
11388
11389
11390
11391
11392
/*
** 2004 April 6
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This file implements an external (disk-based) database using BTrees.
** See the header comment on "btreeInt.h" for additional information.
** Including a description of file format and an overview of operation.
*/
#include "btreeInt.h"

/*
** The header string that appears at the beginning of every
** SQLite database.
*/
static const char zMagicHeader[] = SQLITE_FILE_HEADER;

/*
** Set this global variable to 1 to enable tracing using the TRACE
** macro.
*/
#if 0
int sqlite3BtreeTrace=1;  /* True to enable tracing */
# define TRACE(X)  if(sqlite3BtreeTrace){printf X;fflush(stdout);}
#else
# define TRACE(X)
#endif

/*
** Extract a 2-byte big-endian integer from an array of unsigned bytes.
** But if the value is zero, make it 65536.
**
** This routine is used to extract the "offset to cell content area" value
** from the header of a btree page.  If the page size is 65536 and the page
** is empty, the offset should be 65536, but the 2-byte value stores zero.
** This routine makes the necessary adjustment to 65536.
*/
#define get2byteNotZero(X)  (((((int)get2byte(X))-1)&0xffff)+1)

/*
** Values passed as the 5th argument to allocateBtreePage()
*/
#define BTALLOC_ANY   0           /* Allocate any page */
#define BTALLOC_EXACT 1           /* Allocate exact page if possible */
#define BTALLOC_LE    2           /* Allocate any page <= the parameter */

/*
** Macro IfNotOmitAV(x) returns (x) if SQLITE_OMIT_AUTOVACUUM is not
** defined, or 0 if it is. For example:
**
**   bIncrVacuum = IfNotOmitAV(pBtShared->incrVacuum);
*/
#ifndef SQLITE_OMIT_AUTOVACUUM
#define IfNotOmitAV(expr) (expr)
#else
#define IfNotOmitAV(expr) 0
#endif

#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** A list of BtShared objects that are eligible for participation
** in shared cache.  This variable has file scope during normal builds,
** but the test harness needs to access it so we make it global for
** test builds.
**
** Access to this variable is protected by SQLITE_MUTEX_STATIC_MAIN.
*/
#ifdef SQLITE_TEST
BtShared *SQLITE_WSD sqlite3SharedCacheList = 0;
#else
static BtShared *SQLITE_WSD sqlite3SharedCacheList = 0;
#endif
#endif /* SQLITE_OMIT_SHARED_CACHE */

#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Enable or disable the shared pager and schema features.
**
** This routine has no effect on existing database connections.
** The shared cache setting effects only future calls to
** sqlite3_open(), sqlite3_open16(), or sqlite3_open_v2().
*/
int sqlite3_enable_shared_cache(int enable){
  sqlite3GlobalConfig.sharedCacheEnabled = enable;
  return SQLITE_OK;
}
#endif



#ifdef SQLITE_OMIT_SHARED_CACHE
  /*
  ** The functions querySharedCacheTableLock(), setSharedCacheTableLock(),
  ** and clearAllSharedCacheTableLocks()
  ** manipulate entries in the BtShared.pLock linked list used to store
  ** shared-cache table level locks. If the library is compiled with the
  ** shared-cache feature disabled, then there is only ever one user
  ** of each BtShared structure and so this locking is not necessary.
  ** So define the lock related functions as no-ops.
  */
  #define querySharedCacheTableLock(a,b,c) SQLITE_OK
  #define setSharedCacheTableLock(a,b,c) SQLITE_OK
  #define clearAllSharedCacheTableLocks(a)
  #define downgradeAllSharedCacheTableLocks(a)
  #define hasSharedCacheTableLock(a,b,c,d) 1
  #define hasReadConflicts(a, b) 0
#endif

#ifdef SQLITE_DEBUG
/*
** Return and reset the seek counter for a Btree object.
*/
sqlite3_uint64 sqlite3BtreeSeekCount(Btree *pBt){
  u64 n =  pBt->nSeek;
  pBt->nSeek = 0;
  return n;
}
#endif

/*
** Implementation of the SQLITE_CORRUPT_PAGE() macro. Takes a single
** (MemPage*) as an argument. The (MemPage*) must not be NULL.
**
** If SQLITE_DEBUG is not defined, then this macro is equivalent to
** SQLITE_CORRUPT_BKPT. Or, if SQLITE_DEBUG is set, then the log message
** normally produced as a side-effect of SQLITE_CORRUPT_BKPT is augmented
** with the page number and filename associated with the (MemPage*).
*/
#ifdef SQLITE_DEBUG
int corruptPageError(int lineno, MemPage *p){
  char *zMsg;
  sqlite3BeginBenignMalloc();
  zMsg = sqlite3_mprintf("database corruption page %u of %s",
             p->pgno, sqlite3PagerFilename(p->pBt->pPager, 0)
  );
  sqlite3EndBenignMalloc();
  if( zMsg ){
    sqlite3ReportError(SQLITE_CORRUPT, lineno, zMsg);
  }
  sqlite3_free(zMsg);
  return SQLITE_CORRUPT_BKPT;
}
# define SQLITE_CORRUPT_PAGE(pMemPage) corruptPageError(__LINE__, pMemPage)
#else
# define SQLITE_CORRUPT_PAGE(pMemPage) SQLITE_CORRUPT_PGNO(pMemPage->pgno)
#endif

#ifndef SQLITE_OMIT_SHARED_CACHE

#ifdef SQLITE_DEBUG
/*
**** This function is only used as part of an assert() statement. ***
**
** Check to see if pBtree holds the required locks to read or write to the
** table with root page iRoot.   Return 1 if it does and 0 if not.
**
** For example, when writing to a table with root-page iRoot via
** Btree connection pBtree:
**
**    assert( hasSharedCacheTableLock(pBtree, iRoot, 0, WRITE_LOCK) );
**
** When writing to an index that resides in a sharable database, the
** caller should have first obtained a lock specifying the root page of
** the corresponding table. This makes things a bit more complicated,
** as this module treats each table as a separate structure. To determine
** the table corresponding to the index being written, this
** function has to search through the database schema.
**
** Instead of a lock on the table/index rooted at page iRoot, the caller may
** hold a write-lock on the schema table (root page 1). This is also
** acceptable.
*/
static int hasSharedCacheTableLock(
  Btree *pBtree,         /* Handle that must hold lock */
  Pgno iRoot,            /* Root page of b-tree */
  int isIndex,           /* True if iRoot is the root of an index b-tree */
  int eLockType          /* Required lock type (READ_LOCK or WRITE_LOCK) */
){
  Schema *pSchema = (Schema *)pBtree->pBt->pSchema;
  Pgno iTab = 0;
  BtLock *pLock;

  /* If this database is not shareable, or if the client is reading
  ** and has the read-uncommitted flag set, then no lock is required.
  ** Return true immediately.
  */
  if( (pBtree->sharable==0)
   || (eLockType==READ_LOCK && (pBtree->db->flags & SQLITE_ReadUncommit))
  ){
    return 1;
  }

  /* If the client is reading  or writing an index and the schema is
  ** not loaded, then it is too difficult to actually check to see if
  ** the correct locks are held.  So do not bother - just return true.
  ** This case does not come up very often anyhow.
  */
  if( isIndex && (!pSchema || (pSchema->schemaFlags&DB_SchemaLoaded)==0) ){
    return 1;
  }

  /* Figure out the root-page that the lock should be held on. For table
  ** b-trees, this is just the root page of the b-tree being read or
  ** written. For index b-trees, it is the root page of the associated
  ** table.  */
  if( isIndex ){
    HashElem *p;
    int bSeen = 0;
    for(p=sqliteHashFirst(&pSchema->idxHash); p; p=sqliteHashNext(p)){
      Index *pIdx = (Index *)sqliteHashData(p);
      if( pIdx->tnum==iRoot ){
        if( bSeen ){
          /* Two or more indexes share the same root page.  There must
          ** be imposter tables.  So just return true.  The assert is not
          ** useful in that case. */
          return 1;
        }
        iTab = pIdx->pTable->tnum;
        bSeen = 1;
      }
    }
  }else{
    iTab = iRoot;
  }

  /* Search for the required lock. Either a write-lock on root-page iTab, a
  ** write-lock on the schema table, or (if the client is reading) a
  ** read-lock on iTab will suffice. Return 1 if any of these are found.  */
  for(pLock=pBtree->pBt->pLock; pLock; pLock=pLock->pNext){
    if( pLock->pBtree==pBtree
     && (pLock->iTable==iTab || (pLock->eLock==WRITE_LOCK && pLock->iTable==1))
     && pLock->eLock>=eLockType
    ){
      return 1;
    }
  }

  /* Failed to find the required lock. */
  return 0;
}
#endif /* SQLITE_DEBUG */

#ifdef SQLITE_DEBUG
/*
**** This function may be used as part of assert() statements only. ****
**
** Return true if it would be illegal for pBtree to write into the
** table or index rooted at iRoot because other shared connections are
** simultaneously reading that same table or index.
**
** It is illegal for pBtree to write if some other Btree object that
** shares the same BtShared object is currently reading or writing
** the iRoot table.  Except, if the other Btree object has the
** read-uncommitted flag set, then it is OK for the other object to
** have a read cursor.
**
** For example, before writing to any part of the table or index
** rooted at page iRoot, one should call:
**
**    assert( !hasReadConflicts(pBtree, iRoot) );
*/
static int hasReadConflicts(Btree *pBtree, Pgno iRoot){
  BtCursor *p;
  for(p=pBtree->pBt->pCursor; p; p=p->pNext){
    if( p->pgnoRoot==iRoot
     && p->pBtree!=pBtree
     && 0==(p->pBtree->db->flags & SQLITE_ReadUncommit)
    ){
      return 1;
    }
  }
  return 0;
}
#endif    /* #ifdef SQLITE_DEBUG */

/*
** Query to see if Btree handle p may obtain a lock of type eLock
** (READ_LOCK or WRITE_LOCK) on the table with root-page iTab. Return
** SQLITE_OK if the lock may be obtained (by calling
** setSharedCacheTableLock()), or SQLITE_LOCKED if not.
*/
static int querySharedCacheTableLock(Btree *p, Pgno iTab, u8 eLock){
  BtShared *pBt = p->pBt;
  BtLock *pIter;

  assert( sqlite3BtreeHoldsMutex(p) );
  assert( eLock==READ_LOCK || eLock==WRITE_LOCK );
  assert( p->db!=0 );
  assert( !(p->db->flags&SQLITE_ReadUncommit)||eLock==WRITE_LOCK||iTab==1 );
 
  /* If requesting a write-lock, then the Btree must have an open write
  ** transaction on this file. And, obviously, for this to be so there
  ** must be an open write transaction on the file itself.
  */
  assert( eLock==READ_LOCK || (p==pBt->pWriter && p->inTrans==TRANS_WRITE) );
  assert( eLock==READ_LOCK || pBt->inTransaction==TRANS_WRITE );
 
  /* This routine is a no-op if the shared-cache is not enabled */
  if( !p->sharable ){
    return SQLITE_OK;
  }

  /* If some other connection is holding an exclusive lock, the
  ** requested lock may not be obtained.
  */
  if( pBt->pWriter!=p && (pBt->btsFlags & BTS_EXCLUSIVE)!=0 ){
    sqlite3ConnectionBlocked(p->db, pBt->pWriter->db);
    return SQLITE_LOCKED_SHAREDCACHE;
  }

  for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
    /* The condition (pIter->eLock!=eLock) in the following if(...)
    ** statement is a simplification of:
    **
    **   (eLock==WRITE_LOCK || pIter->eLock==WRITE_LOCK)
    **
    ** since we know that if eLock==WRITE_LOCK, then no other connection
    ** may hold a WRITE_LOCK on any table in this file (since there can
    ** only be a single writer).
    */
    assert( pIter->eLock==READ_LOCK || pIter->eLock==WRITE_LOCK );
    assert( eLock==READ_LOCK || pIter->pBtree==p || pIter->eLock==READ_LOCK);
    if( pIter->pBtree!=p && pIter->iTable==iTab && pIter->eLock!=eLock ){
      sqlite3ConnectionBlocked(p->db, pIter->pBtree->db);
      if( eLock==WRITE_LOCK ){
        assert( p==pBt->pWriter );
        pBt->btsFlags |= BTS_PENDING;
      }
      return SQLITE_LOCKED_SHAREDCACHE;
    }
  }
  return SQLITE_OK;
}
#endif /* !SQLITE_OMIT_SHARED_CACHE */

#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Add a lock on the table with root-page iTable to the shared-btree used
** by Btree handle p. Parameter eLock must be either READ_LOCK or
** WRITE_LOCK.
**
** This function assumes the following:
**
**   (a) The specified Btree object p is connected to a sharable
**       database (one with the BtShared.sharable flag set), and
**
**   (b) No other Btree objects hold a lock that conflicts
**       with the requested lock (i.e. querySharedCacheTableLock() has
**       already been called and returned SQLITE_OK).
**
** SQLITE_OK is returned if the lock is added successfully. SQLITE_NOMEM
** is returned if a malloc attempt fails.
*/
static int setSharedCacheTableLock(Btree *p, Pgno iTable, u8 eLock){
  BtShared *pBt = p->pBt;
  BtLock *pLock = 0;
  BtLock *pIter;

  assert( sqlite3BtreeHoldsMutex(p) );
  assert( eLock==READ_LOCK || eLock==WRITE_LOCK );
  assert( p->db!=0 );

  /* A connection with the read-uncommitted flag set will never try to
  ** obtain a read-lock using this function. The only read-lock obtained
  ** by a connection in read-uncommitted mode is on the sqlite_schema
  ** table, and that lock is obtained in BtreeBeginTrans().  */
  assert( 0==(p->db->flags&SQLITE_ReadUncommit) || eLock==WRITE_LOCK );

  /* This function should only be called on a sharable b-tree after it
  ** has been determined that no other b-tree holds a conflicting lock.  */
  assert( p->sharable );
  assert( SQLITE_OK==querySharedCacheTableLock(p, iTable, eLock) );

  /* First search the list for an existing lock on this table. */
  for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
    if( pIter->iTable==iTable && pIter->pBtree==p ){
      pLock = pIter;
      break;
    }
  }

  /* If the above search did not find a BtLock struct associating Btree p
  ** with table iTable, allocate one and link it into the list.
  */
  if( !pLock ){
    pLock = (BtLock *)sqlite3MallocZero(sizeof(BtLock));
    if( !pLock ){
      return SQLITE_NOMEM_BKPT;
    }
    pLock->iTable = iTable;
    pLock->pBtree = p;
    pLock->pNext = pBt->pLock;
    pBt->pLock = pLock;
  }

  /* Set the BtLock.eLock variable to the maximum of the current lock
  ** and the requested lock. This means if a write-lock was already held
  ** and a read-lock requested, we don't incorrectly downgrade the lock.
  */
  assert( WRITE_LOCK>READ_LOCK );
  if( eLock>pLock->eLock ){
    pLock->eLock = eLock;
  }

  return SQLITE_OK;
}
#endif /* !SQLITE_OMIT_SHARED_CACHE */

#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Release all the table locks (locks obtained via calls to
** the setSharedCacheTableLock() procedure) held by Btree object p.
**
** This function assumes that Btree p has an open read or write
** transaction. If it does not, then the BTS_PENDING flag
** may be incorrectly cleared.
*/
static void clearAllSharedCacheTableLocks(Btree *p){
  BtShared *pBt = p->pBt;
  BtLock **ppIter = &pBt->pLock;

  assert( sqlite3BtreeHoldsMutex(p) );
  assert( p->sharable || 0==*ppIter );
  assert( p->inTrans>0 );

  while( *ppIter ){
    BtLock *pLock = *ppIter;
    assert( (pBt->btsFlags & BTS_EXCLUSIVE)==0 || pBt->pWriter==pLock->pBtree );
    assert( pLock->pBtree->inTrans>=pLock->eLock );
    if( pLock->pBtree==p ){
      *ppIter = pLock->pNext;
      assert( pLock->iTable!=1 || pLock==&p->lock );
      if( pLock->iTable!=1 ){
        sqlite3_free(pLock);
      }
    }else{
      ppIter = &pLock->pNext;
    }
  }

  assert( (pBt->btsFlags & BTS_PENDING)==0 || pBt->pWriter );
  if( pBt->pWriter==p ){
    pBt->pWriter = 0;
    pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING);
  }else if( pBt->nTransaction==2 ){
    /* This function is called when Btree p is concluding its
    ** transaction. If there currently exists a writer, and p is not
    ** that writer, then the number of locks held by connections other
    ** than the writer must be about to drop to zero. In this case
    ** set the BTS_PENDING flag to 0.
    **
    ** If there is not currently a writer, then BTS_PENDING must
    ** be zero already. So this next line is harmless in that case.
    */
    pBt->btsFlags &= ~BTS_PENDING;
  }
}

/*
** This function changes all write-locks held by Btree p into read-locks.
*/
static void downgradeAllSharedCacheTableLocks(Btree *p){
  BtShared *pBt = p->pBt;
  if( pBt->pWriter==p ){
    BtLock *pLock;
    pBt->pWriter = 0;
    pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING);
    for(pLock=pBt->pLock; pLock; pLock=pLock->pNext){
      assert( pLock->eLock==READ_LOCK || pLock->pBtree==p );
      pLock->eLock = READ_LOCK;
    }
  }
}

#endif /* SQLITE_OMIT_SHARED_CACHE */

static void releasePage(MemPage *pPage);         /* Forward reference */
static void releasePageOne(MemPage *pPage);      /* Forward reference */
static void releasePageNotNull(MemPage *pPage);  /* Forward reference */

/*
***** This routine is used inside of assert() only ****
**
** Verify that the cursor holds the mutex on its BtShared
*/
#ifdef SQLITE_DEBUG
static int cursorHoldsMutex(BtCursor *p){
  return sqlite3_mutex_held(p->pBt->mutex);
}

/* Verify that the cursor and the BtShared agree about what is the current
** database connetion. This is important in shared-cache mode. If the database
** connection pointers get out-of-sync, it is possible for routines like
** btreeInitPage() to reference an stale connection pointer that references a
** a connection that has already closed.  This routine is used inside assert()
** statements only and for the purpose of double-checking that the btree code
** does keep the database connection pointers up-to-date.
*/
static int cursorOwnsBtShared(BtCursor *p){
  assert( cursorHoldsMutex(p) );
  return (p->pBtree->db==p->pBt->db);
}
#endif

/*
** Invalidate the overflow cache of the cursor passed as the first argument.
** on the shared btree structure pBt.
*/
#define invalidateOverflowCache(pCur) (pCur->curFlags &= ~BTCF_ValidOvfl)

/*
** Invalidate the overflow page-list cache for all cursors opened
** on the shared btree structure pBt.
*/
static void invalidateAllOverflowCache(BtShared *pBt){
  BtCursor *p;
  assert( sqlite3_mutex_held(pBt->mutex) );
  for(p=pBt->pCursor; p; p=p->pNext){
    invalidateOverflowCache(p);
  }
}

#ifndef SQLITE_OMIT_INCRBLOB
/*
** This function is called before modifying the contents of a table
** to invalidate any incrblob cursors that are open on the
** row or one of the rows being modified.
**
** If argument isClearTable is true, then the entire contents of the
** table is about to be deleted. In this case invalidate all incrblob
** cursors open on any row within the table with root-page pgnoRoot.
**
** Otherwise, if argument isClearTable is false, then the row with
** rowid iRow is being replaced or deleted. In this case invalidate
** only those incrblob cursors open on that specific row.
*/
static void invalidateIncrblobCursors(
  Btree *pBtree,          /* The database file to check */
  Pgno pgnoRoot,          /* The table that might be changing */
  i64 iRow,               /* The rowid that might be changing */
  int isClearTable        /* True if all rows are being deleted */
){
  BtCursor *p;
  assert( pBtree->hasIncrblobCur );
  assert( sqlite3BtreeHoldsMutex(pBtree) );
  pBtree->hasIncrblobCur = 0;
  for(p=pBtree->pBt->pCursor; p; p=p->pNext){
    if( (p->curFlags & BTCF_Incrblob)!=0 ){
      pBtree->hasIncrblobCur = 1;
      if( p->pgnoRoot==pgnoRoot && (isClearTable || p->info.nKey==iRow) ){
        p->eState = CURSOR_INVALID;
      }
    }
  }
}

#else
  /* Stub function when INCRBLOB is omitted */
  #define invalidateIncrblobCursors(w,x,y,z)
#endif /* SQLITE_OMIT_INCRBLOB */

/*
** Set bit pgno of the BtShared.pHasContent bitvec. This is called
** when a page that previously contained data becomes a free-list leaf
** page.
**
** The BtShared.pHasContent bitvec exists to work around an obscure
** bug caused by the interaction of two useful IO optimizations surrounding
** free-list leaf pages:
**
**   1) When all data is deleted from a page and the page becomes
**      a free-list leaf page, the page is not written to the database
**      (as free-list leaf pages contain no meaningful data). Sometimes
**      such a page is not even journalled (as it will not be modified,
**      why bother journalling it?).
**
**   2) When a free-list leaf page is reused, its content is not read
**      from the database or written to the journal file (why should it
**      be, if it is not at all meaningful?).
**
** By themselves, these optimizations work fine and provide a handy
** performance boost to bulk delete or insert operations. However, if
** a page is moved to the free-list and then reused within the same
** transaction, a problem comes up. If the page is not journalled when
** it is moved to the free-list and it is also not journalled when it
** is extracted from the free-list and reused, then the original data
** may be lost. In the event of a rollback, it may not be possible
** to restore the database to its original configuration.
**
** The solution is the BtShared.pHasContent bitvec. Whenever a page is
** moved to become a free-list leaf page, the corresponding bit is
** set in the bitvec. Whenever a leaf page is extracted from the free-list,
** optimization 2 above is omitted if the corresponding bit is already
** set in BtShared.pHasContent. The contents of the bitvec are cleared
** at the end of every transaction.
*/
static int btreeSetHasContent(BtShared *pBt, Pgno pgno){
  int rc = SQLITE_OK;
  if( !pBt->pHasContent ){
    assert( pgno<=pBt->nPage );
    pBt->pHasContent = sqlite3BitvecCreate(pBt->nPage);
    if( !pBt->pHasContent ){
      rc = SQLITE_NOMEM_BKPT;
    }
  }
  if( rc==SQLITE_OK && pgno<=sqlite3BitvecSize(pBt->pHasContent) ){
    rc = sqlite3BitvecSet(pBt->pHasContent, pgno);
  }
  return rc;
}

/*
** Query the BtShared.pHasContent vector.
**
** This function is called when a free-list leaf page is removed from the
** free-list for reuse. It returns false if it is safe to retrieve the
** page from the pager layer with the 'no-content' flag set. True otherwise.
*/
static int btreeGetHasContent(BtShared *pBt, Pgno pgno){
  Bitvec *p = pBt->pHasContent;
  return p && (pgno>sqlite3BitvecSize(p) || sqlite3BitvecTestNotNull(p, pgno));
}

/*
** Clear (destroy) the BtShared.pHasContent bitvec. This should be
** invoked at the conclusion of each write-transaction.
*/
static void btreeClearHasContent(BtShared *pBt){
  sqlite3BitvecDestroy(pBt->pHasContent);
  pBt->pHasContent = 0;
}

/*
** Release all of the apPage[] pages for a cursor.
*/
static void btreeReleaseAllCursorPages(BtCursor *pCur){
  int i;
  if( pCur->iPage>=0 ){
    for(i=0; i<pCur->iPage; i++){
      releasePageNotNull(pCur->apPage[i]);
    }
    releasePageNotNull(pCur->pPage);
    pCur->iPage = -1;
  }
}

/*
** The cursor passed as the only argument must point to a valid entry
** when this function is called (i.e. have eState==CURSOR_VALID). This
** function saves the current cursor key in variables pCur->nKey and
** pCur->pKey. SQLITE_OK is returned if successful or an SQLite error
** code otherwise.
**
** If the cursor is open on an intkey table, then the integer key
** (the rowid) is stored in pCur->nKey and pCur->pKey is left set to
** NULL. If the cursor is open on a non-intkey table, then pCur->pKey is
** set to point to a malloced buffer pCur->nKey bytes in size containing
** the key.
*/
static int saveCursorKey(BtCursor *pCur){
  int rc = SQLITE_OK;
  assert( CURSOR_VALID==pCur->eState );
  assert( 0==pCur->pKey );
  assert( cursorHoldsMutex(pCur) );

  if( pCur->curIntKey ){
    /* Only the rowid is required for a table btree */
    pCur->nKey = sqlite3BtreeIntegerKey(pCur);
  }else{
    /* For an index btree, save the complete key content. It is possible
    ** that the current key is corrupt. In that case, it is possible that
    ** the sqlite3VdbeRecordUnpack() function may overread the buffer by
    ** up to the size of 1 varint plus 1 8-byte value when the cursor
    ** position is restored. Hence the 17 bytes of padding allocated
    ** below. */
    void *pKey;
    pCur->nKey = sqlite3BtreePayloadSize(pCur);
    pKey = sqlite3Malloc( pCur->nKey + 9 + 8 );
    if( pKey ){
      rc = sqlite3BtreePayload(pCur, 0, (int)pCur->nKey, pKey);
      if( rc==SQLITE_OK ){
        memset(((u8*)pKey)+pCur->nKey, 0, 9+8);
        pCur->pKey = pKey;
      }else{
        sqlite3_free(pKey);
      }
    }else{
      rc = SQLITE_NOMEM_BKPT;
    }
  }
  assert( !pCur->curIntKey || !pCur->pKey );
  return rc;
}

/*
** Save the current cursor position in the variables BtCursor.nKey
** and BtCursor.pKey. The cursor's state is set to CURSOR_REQUIRESEEK.
**
** The caller must ensure that the cursor is valid (has eState==CURSOR_VALID)
** prior to calling this routine. 
*/
static int saveCursorPosition(BtCursor *pCur){
  int rc;

  assert( CURSOR_VALID==pCur->eState || CURSOR_SKIPNEXT==pCur->eState );
  assert( 0==pCur->pKey );
  assert( cursorHoldsMutex(pCur) );

  if( pCur->curFlags & BTCF_Pinned ){
    return SQLITE_CONSTRAINT_PINNED;
  }
  if( pCur->eState==CURSOR_SKIPNEXT ){
    pCur->eState = CURSOR_VALID;
  }else{
    pCur->skipNext = 0;
  }

  rc = saveCursorKey(pCur);
  if( rc==SQLITE_OK ){
    btreeReleaseAllCursorPages(pCur);
    pCur->eState = CURSOR_REQUIRESEEK;
  }

  pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl|BTCF_AtLast);
  return rc;
}

/* Forward reference */
static int SQLITE_NOINLINE saveCursorsOnList(BtCursor*,Pgno,BtCursor*);

/*
** Save the positions of all cursors (except pExcept) that are open on
** the table with root-page iRoot.  "Saving the cursor position" means that
** the location in the btree is remembered in such a way that it can be
** moved back to the same spot after the btree has been modified.  This
** routine is called just before cursor pExcept is used to modify the
** table, for example in BtreeDelete() or BtreeInsert().
**
** If there are two or more cursors on the same btree, then all such
** cursors should have their BTCF_Multiple flag set.  The btreeCursor()
** routine enforces that rule.  This routine only needs to be called in
** the uncommon case when pExpect has the BTCF_Multiple flag set.
**
** If pExpect!=NULL and if no other cursors are found on the same root-page,
** then the BTCF_Multiple flag on pExpect is cleared, to avoid another
** pointless call to this routine.
**
** Implementation note:  This routine merely checks to see if any cursors
** need to be saved.  It calls out to saveCursorsOnList() in the (unusual)
** event that cursors are in need to being saved.
*/
static int saveAllCursors(BtShared *pBt, Pgno iRoot, BtCursor *pExcept){
  BtCursor *p;
  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( pExcept==0 || pExcept->pBt==pBt );
  for(p=pBt->pCursor; p; p=p->pNext){
    if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) ) break;
  }
  if( p ) return saveCursorsOnList(p, iRoot, pExcept);
  if( pExcept ) pExcept->curFlags &= ~BTCF_Multiple;
  return SQLITE_OK;
}

/* This helper routine to saveAllCursors does the actual work of saving
** the cursors if and when a cursor is found that actually requires saving.
** The common case is that no cursors need to be saved, so this routine is
** broken out from its caller to avoid unnecessary stack pointer movement.
*/
static int SQLITE_NOINLINE saveCursorsOnList(
  BtCursor *p,         /* The first cursor that needs saving */
  Pgno iRoot,          /* Only save cursor with this iRoot. Save all if zero */
  BtCursor *pExcept    /* Do not save this cursor */
){
  do{
    if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) ){
      if( p->eState==CURSOR_VALID || p->eState==CURSOR_SKIPNEXT ){
        int rc = saveCursorPosition(p);
        if( SQLITE_OK!=rc ){
          return rc;
        }
      }else{
        testcase( p->iPage>=0 );
        btreeReleaseAllCursorPages(p);
      }
    }
    p = p->pNext;
  }while( p );
  return SQLITE_OK;
}

/*
** Clear the current cursor position.
*/
void sqlite3BtreeClearCursor(BtCursor *pCur){
  assert( cursorHoldsMutex(pCur) );
  sqlite3_free(pCur->pKey);
  pCur->pKey = 0;
  pCur->eState = CURSOR_INVALID;
}

/*
** In this version of BtreeMoveto, pKey is a packed index record
** such as is generated by the OP_MakeRecord opcode.  Unpack the
** record and then call sqlite3BtreeIndexMoveto() to do the work.
*/
static int btreeMoveto(
  BtCursor *pCur,     /* Cursor open on the btree to be searched */
  const void *pKey,   /* Packed key if the btree is an index */
  i64 nKey,           /* Integer key for tables.  Size of pKey for indices */
  int bias,           /* Bias search to the high end */
  int *pRes           /* Write search results here */
){
  int rc;                    /* Status code */
  UnpackedRecord *pIdxKey;   /* Unpacked index key */

  if( pKey ){
    KeyInfo *pKeyInfo = pCur->pKeyInfo;
    assert( nKey==(i64)(int)nKey );
    pIdxKey = sqlite3VdbeAllocUnpackedRecord(pKeyInfo);
    if( pIdxKey==0 ) return SQLITE_NOMEM_BKPT;
    sqlite3VdbeRecordUnpack(pKeyInfo, (int)nKey, pKey, pIdxKey);
    if( pIdxKey->nField==0 || pIdxKey->nField>pKeyInfo->nAllField ){
      rc = SQLITE_CORRUPT_BKPT;
    }else{
      rc = sqlite3BtreeIndexMoveto(pCur, pIdxKey, pRes);
    }
    sqlite3DbFree(pCur->pKeyInfo->db, pIdxKey);
  }else{
    pIdxKey = 0;
    rc = sqlite3BtreeTableMoveto(pCur, nKey, bias, pRes);
  }
  return rc;
}

/*
** Restore the cursor to the position it was in (or as close to as possible)
** when saveCursorPosition() was called. Note that this call deletes the
** saved position info stored by saveCursorPosition(), so there can be
** at most one effective restoreCursorPosition() call after each
** saveCursorPosition().
*/
static int btreeRestoreCursorPosition(BtCursor *pCur){
  int rc;
  int skipNext = 0;
  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->eState>=CURSOR_REQUIRESEEK );
  if( pCur->eState==CURSOR_FAULT ){
    return pCur->skipNext;
  }
  pCur->eState = CURSOR_INVALID;
  if( sqlite3FaultSim(410) ){
    rc = SQLITE_IOERR;
  }else{
    rc = btreeMoveto(pCur, pCur->pKey, pCur->nKey, 0, &skipNext);
  }
  if( rc==SQLITE_OK ){
    sqlite3_free(pCur->pKey);
    pCur->pKey = 0;
    assert( pCur->eState==CURSOR_VALID || pCur->eState==CURSOR_INVALID );
    if( skipNext ) pCur->skipNext = skipNext;
    if( pCur->skipNext && pCur->eState==CURSOR_VALID ){
      pCur->eState = CURSOR_SKIPNEXT;
    }
  }
  return rc;
}

#define restoreCursorPosition(p) \
  (p->eState>=CURSOR_REQUIRESEEK ? \
         btreeRestoreCursorPosition(p) : \
         SQLITE_OK)

/*
** Determine whether or not a cursor has moved from the position where
** it was last placed, or has been invalidated for any other reason.
** Cursors can move when the row they are pointing at is deleted out
** from under them, for example.  Cursor might also move if a btree
** is rebalanced.
**
** Calling this routine with a NULL cursor pointer returns false.
**
** Use the separate sqlite3BtreeCursorRestore() routine to restore a cursor
** back to where it ought to be if this routine returns true.
*/
int sqlite3BtreeCursorHasMoved(BtCursor *pCur){
  assert( EIGHT_BYTE_ALIGNMENT(pCur)
       || pCur==sqlite3BtreeFakeValidCursor() );
  assert( offsetof(BtCursor, eState)==0 );
  assert( sizeof(pCur->eState)==1 );
  return CURSOR_VALID != *(u8*)pCur;
}

/*
** Return a pointer to a fake BtCursor object that will always answer
** false to the sqlite3BtreeCursorHasMoved() routine above.  The fake
** cursor returned must not be used with any other Btree interface.
*/
BtCursor *sqlite3BtreeFakeValidCursor(void){
  static u8 fakeCursor = CURSOR_VALID;
  assert( offsetof(BtCursor, eState)==0 );
  return (BtCursor*)&fakeCursor;
}

/*
** This routine restores a cursor back to its original position after it
** has been moved by some outside activity (such as a btree rebalance or
** a row having been deleted out from under the cursor). 
**
** On success, the *pDifferentRow parameter is false if the cursor is left
** pointing at exactly the same row.  *pDifferntRow is the row the cursor
** was pointing to has been deleted, forcing the cursor to point to some
** nearby row.
**
** This routine should only be called for a cursor that just returned
** TRUE from sqlite3BtreeCursorHasMoved().
*/
int sqlite3BtreeCursorRestore(BtCursor *pCur, int *pDifferentRow){
  int rc;

  assert( pCur!=0 );
  assert( pCur->eState!=CURSOR_VALID );
  rc = restoreCursorPosition(pCur);
  if( rc ){
    *pDifferentRow = 1;
    return rc;
  }
  if( pCur->eState!=CURSOR_VALID ){
    *pDifferentRow = 1;
  }else{
    *pDifferentRow = 0;
  }
  return SQLITE_OK;
}

#ifdef SQLITE_ENABLE_CURSOR_HINTS
/*
** Provide hints to the cursor.  The particular hint given (and the type
** and number of the varargs parameters) is determined by the eHintType
** parameter.  See the definitions of the BTREE_HINT_* macros for details.
*/
void sqlite3BtreeCursorHint(BtCursor *pCur, int eHintType, ...){
  /* Used only by system that substitute their own storage engine */
#ifdef SQLITE_DEBUG
  if( ALWAYS(eHintType==BTREE_HINT_RANGE) ){
    va_list ap;
    Expr *pExpr;
    Walker w;
    memset(&w, 0, sizeof(w));
    w.xExprCallback = sqlite3CursorRangeHintExprCheck;
    va_start(ap, eHintType);
    pExpr = va_arg(ap, Expr*);
    w.u.aMem = va_arg(ap, Mem*);
    va_end(ap);
    assert( pExpr!=0 );
    assert( w.u.aMem!=0 );
    sqlite3WalkExpr(&w, pExpr);
  }
#endif /* SQLITE_DEBUG */
}
#endif /* SQLITE_ENABLE_CURSOR_HINTS */


/*
** Provide flag hints to the cursor.
*/
void sqlite3BtreeCursorHintFlags(BtCursor *pCur, unsigned x){
  assert( x==BTREE_SEEK_EQ || x==BTREE_BULKLOAD || x==0 );
  pCur->hints = x;
}


#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** Given a page number of a regular database page, return the page
** number for the pointer-map page that contains the entry for the
** input page number.
**
** Return 0 (not a valid page) for pgno==1 since there is
** no pointer map associated with page 1.  The integrity_check logic
** requires that ptrmapPageno(*,1)!=1.
*/
static Pgno ptrmapPageno(BtShared *pBt, Pgno pgno){
  int nPagesPerMapPage;
  Pgno iPtrMap, ret;
  assert( sqlite3_mutex_held(pBt->mutex) );
  if( pgno<2 ) return 0;
  nPagesPerMapPage = (pBt->usableSize/5)+1;
  iPtrMap = (pgno-2)/nPagesPerMapPage;
  ret = (iPtrMap*nPagesPerMapPage) + 2;
  if( ret==PENDING_BYTE_PAGE(pBt) ){
    ret++;
  }
  return ret;
}

/*
** Write an entry into the pointer map.
**
** This routine updates the pointer map entry for page number 'key'
** so that it maps to type 'eType' and parent page number 'pgno'.
**
** If *pRC is initially non-zero (non-SQLITE_OK) then this routine is
** a no-op.  If an error occurs, the appropriate error code is written
** into *pRC.
*/
static void ptrmapPut(BtShared *pBt, Pgno key, u8 eType, Pgno parent, int *pRC){
  DbPage *pDbPage;  /* The pointer map page */
  u8 *pPtrmap;      /* The pointer map data */
  Pgno iPtrmap;     /* The pointer map page number */
  int offset;       /* Offset in pointer map page */
  int rc;           /* Return code from subfunctions */

  if( *pRC ) return;

  assert( sqlite3_mutex_held(pBt->mutex) );
  /* The super-journal page number must never be used as a pointer map page */
  assert( 0==PTRMAP_ISPAGE(pBt, PENDING_BYTE_PAGE(pBt)) );

  assert( pBt->autoVacuum );
  if( key==0 ){
    *pRC = SQLITE_CORRUPT_BKPT;
    return;
  }
  iPtrmap = PTRMAP_PAGENO(pBt, key);
  rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage, 0);
  if( rc!=SQLITE_OK ){
    *pRC = rc;
    return;
  }
  if( ((char*)sqlite3PagerGetExtra(pDbPage))[0]!=0 ){
    /* The first byte of the extra data is the MemPage.isInit byte.
    ** If that byte is set, it means this page is also being used
    ** as a btree page. */
    *pRC = SQLITE_CORRUPT_BKPT;
    goto ptrmap_exit;
  }
  offset = PTRMAP_PTROFFSET(iPtrmap, key);
  if( offset<0 ){
    *pRC = SQLITE_CORRUPT_BKPT;
    goto ptrmap_exit;
  }
  assert( offset <= (int)pBt->usableSize-5 );
  pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);

  if( eType!=pPtrmap[offset] || get4byte(&pPtrmap[offset+1])!=parent ){
    TRACE(("PTRMAP_UPDATE: %u->(%u,%u)\n", key, eType, parent));
    *pRC= rc = sqlite3PagerWrite(pDbPage);
    if( rc==SQLITE_OK ){
      pPtrmap[offset] = eType;
      put4byte(&pPtrmap[offset+1], parent);
    }
  }

ptrmap_exit:
  sqlite3PagerUnref(pDbPage);
}

/*
** Read an entry from the pointer map.
**
** This routine retrieves the pointer map entry for page 'key', writing
** the type and parent page number to *pEType and *pPgno respectively.
** An error code is returned if something goes wrong, otherwise SQLITE_OK.
*/
static int ptrmapGet(BtShared *pBt, Pgno key, u8 *pEType, Pgno *pPgno){
  DbPage *pDbPage;   /* The pointer map page */
  int iPtrmap;       /* Pointer map page index */
  u8 *pPtrmap;       /* Pointer map page data */
  int offset;        /* Offset of entry in pointer map */
  int rc;

  assert( sqlite3_mutex_held(pBt->mutex) );

  iPtrmap = PTRMAP_PAGENO(pBt, key);
  rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage, 0);
  if( rc!=0 ){
    return rc;
  }
  pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);

  offset = PTRMAP_PTROFFSET(iPtrmap, key);
  if( offset<0 ){
    sqlite3PagerUnref(pDbPage);
    return SQLITE_CORRUPT_BKPT;
  }
  assert( offset <= (int)pBt->usableSize-5 );
  assert( pEType!=0 );
  *pEType = pPtrmap[offset];
  if( pPgno ) *pPgno = get4byte(&pPtrmap[offset+1]);

  sqlite3PagerUnref(pDbPage);
  if( *pEType<1 || *pEType>5 ) return SQLITE_CORRUPT_PGNO(iPtrmap);
  return SQLITE_OK;
}

#else /* if defined SQLITE_OMIT_AUTOVACUUM */
  #define ptrmapPut(w,x,y,z,rc)
  #define ptrmapGet(w,x,y,z) SQLITE_OK
  #define ptrmapPutOvflPtr(x, y, z, rc)
#endif

/*
** Given a btree page and a cell index (0 means the first cell on
** the page, 1 means the second cell, and so forth) return a pointer
** to the cell content.
**
** findCellPastPtr() does the same except it skips past the initial
** 4-byte child pointer found on interior pages, if there is one.
**
** This routine works only for pages that do not contain overflow cells.
*/
#define findCell(P,I) \
  ((P)->aData + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)])))
#define findCellPastPtr(P,I) \
  ((P)->aDataOfst + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)])))


/*
** This is common tail processing for btreeParseCellPtr() and
** btreeParseCellPtrIndex() for the case when the cell does not fit entirely
** on a single B-tree page.  Make necessary adjustments to the CellInfo
** structure.
*/
static SQLITE_NOINLINE void btreeParseCellAdjustSizeForOverflow(
  MemPage *pPage,         /* Page containing the cell */
  u8 *pCell,              /* Pointer to the cell text. */
  CellInfo *pInfo         /* Fill in this structure */
){
  /* If the payload will not fit completely on the local page, we have
  ** to decide how much to store locally and how much to spill onto
  ** overflow pages.  The strategy is to minimize the amount of unused
  ** space on overflow pages while keeping the amount of local storage
  ** in between minLocal and maxLocal.
  **
  ** Warning:  changing the way overflow payload is distributed in any
  ** way will result in an incompatible file format.
  */
  int minLocal;  /* Minimum amount of payload held locally */
  int maxLocal;  /* Maximum amount of payload held locally */
  int surplus;   /* Overflow payload available for local storage */

  minLocal = pPage->minLocal;
  maxLocal = pPage->maxLocal;
  surplus = minLocal + (pInfo->nPayload - minLocal)%(pPage->pBt->usableSize-4);
  testcase( surplus==maxLocal );
  testcase( surplus==maxLocal+1 );
  if( surplus <= maxLocal ){
    pInfo->nLocal = (u16)surplus;
  }else{
    pInfo->nLocal = (u16)minLocal;
  }
  pInfo->nSize = (u16)(&pInfo->pPayload[pInfo->nLocal] - pCell) + 4;
}

/*
** Given a record with nPayload bytes of payload stored within btree
** page pPage, return the number of bytes of payload stored locally.
*/
static int btreePayloadToLocal(MemPage *pPage, i64 nPayload){
  int maxLocal;  /* Maximum amount of payload held locally */
  maxLocal = pPage->maxLocal;
  if( nPayload<=maxLocal ){
    return nPayload;
  }else{
    int minLocal;  /* Minimum amount of payload held locally */
    int surplus;   /* Overflow payload available for local storage */
    minLocal = pPage->minLocal;
    surplus = minLocal + (nPayload - minLocal)%(pPage->pBt->usableSize-4);
    return ( surplus <= maxLocal ) ? surplus : minLocal;
  }
}

/*
** The following routines are implementations of the MemPage.xParseCell()
** method.
**
** Parse a cell content block and fill in the CellInfo structure.
**
** btreeParseCellPtr()        =>   table btree leaf nodes
** btreeParseCellNoPayload()  =>   table btree internal nodes
** btreeParseCellPtrIndex()   =>   index btree nodes
**
** There is also a wrapper function btreeParseCell() that works for
** all MemPage types and that references the cell by index rather than
** by pointer.
*/
static void btreeParseCellPtrNoPayload(
  MemPage *pPage,         /* Page containing the cell */
  u8 *pCell,              /* Pointer to the cell text. */
  CellInfo *pInfo         /* Fill in this structure */
){
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pPage->leaf==0 );
  assert( pPage->childPtrSize==4 );
#ifndef SQLITE_DEBUG
  UNUSED_PARAMETER(pPage);
#endif
  pInfo->nSize = 4 + getVarint(&pCell[4], (u64*)&pInfo->nKey);
  pInfo->nPayload = 0;
  pInfo->nLocal = 0;
  pInfo->pPayload = 0;
  return;
}
static void btreeParseCellPtr(
  MemPage *pPage,         /* Page containing the cell */
  u8 *pCell,              /* Pointer to the cell text. */
  CellInfo *pInfo         /* Fill in this structure */
){
  u8 *pIter;              /* For scanning through pCell */
  u32 nPayload;           /* Number of bytes of cell payload */
  u64 iKey;               /* Extracted Key value */

  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pPage->leaf==0 || pPage->leaf==1 );
  assert( pPage->intKeyLeaf );
  assert( pPage->childPtrSize==0 );
  pIter = pCell;

  /* The next block of code is equivalent to:
  **
  **     pIter += getVarint32(pIter, nPayload);
  **
  ** The code is inlined to avoid a function call.
  */
  nPayload = *pIter;
  if( nPayload>=0x80 ){
    u8 *pEnd = &pIter[8];
    nPayload &= 0x7f;
    do{
      nPayload = (nPayload<<7) | (*++pIter & 0x7f);
    }while( (*pIter)>=0x80 && pIter<pEnd );
  }
  pIter++;

  /* The next block of code is equivalent to:
  **
  **     pIter += getVarint(pIter, (u64*)&pInfo->nKey);
  **
  ** The code is inlined and the loop is unrolled for performance.
  ** This routine is a high-runner.
  */
  iKey = *pIter;
  if( iKey>=0x80 ){
    u8 x;
    iKey = (iKey<<7) ^ (x = *++pIter);
    if( x>=0x80 ){
      iKey = (iKey<<7) ^ (x = *++pIter);
      if( x>=0x80 ){
        iKey = (iKey<<7) ^ 0x10204000 ^ (x = *++pIter);
        if( x>=0x80 ){
          iKey = (iKey<<7) ^ 0x4000 ^ (x = *++pIter);
          if( x>=0x80 ){
            iKey = (iKey<<7) ^ 0x4000 ^ (x = *++pIter);
            if( x>=0x80 ){
              iKey = (iKey<<7) ^ 0x4000 ^ (x = *++pIter);
              if( x>=0x80 ){
                iKey = (iKey<<7) ^ 0x4000 ^ (x = *++pIter);
                if( x>=0x80 ){
                  iKey = (iKey<<8) ^ 0x8000 ^ (*++pIter);
                }
              }
            }
          }
        }
      }else{
        iKey ^= 0x204000;
      }
    }else{
      iKey ^= 0x4000;
    }
  }
  pIter++;

  pInfo->nKey = *(i64*)&iKey;
  pInfo->nPayload = nPayload;
  pInfo->pPayload = pIter;
  testcase( nPayload==pPage->maxLocal );
  testcase( nPayload==(u32)pPage->maxLocal+1 );
  if( nPayload<=pPage->maxLocal ){
    /* This is the (easy) common case where the entire payload fits
    ** on the local page.  No overflow is required.
    */
    pInfo->nSize = nPayload + (u16)(pIter - pCell);
    if( pInfo->nSize<4 ) pInfo->nSize = 4;
    pInfo->nLocal = (u16)nPayload;
  }else{
    btreeParseCellAdjustSizeForOverflow(pPage, pCell, pInfo);
  }
}
static void btreeParseCellPtrIndex(
  MemPage *pPage,         /* Page containing the cell */
  u8 *pCell,              /* Pointer to the cell text. */
  CellInfo *pInfo         /* Fill in this structure */
){
  u8 *pIter;              /* For scanning through pCell */
  u32 nPayload;           /* Number of bytes of cell payload */

  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pPage->leaf==0 || pPage->leaf==1 );
  assert( pPage->intKeyLeaf==0 );
  pIter = pCell + pPage->childPtrSize;
  nPayload = *pIter;
  if( nPayload>=0x80 ){
    u8 *pEnd = &pIter[8];
    nPayload &= 0x7f;
    do{
      nPayload = (nPayload<<7) | (*++pIter & 0x7f);
    }while( *(pIter)>=0x80 && pIter<pEnd );
  }
  pIter++;
  pInfo->nKey = nPayload;
  pInfo->nPayload = nPayload;
  pInfo->pPayload = pIter;
  testcase( nPayload==pPage->maxLocal );
  testcase( nPayload==(u32)pPage->maxLocal+1 );
  if( nPayload<=pPage->maxLocal ){
    /* This is the (easy) common case where the entire payload fits
    ** on the local page.  No overflow is required.
    */
    pInfo->nSize = nPayload + (u16)(pIter - pCell);
    if( pInfo->nSize<4 ) pInfo->nSize = 4;
    pInfo->nLocal = (u16)nPayload;
  }else{
    btreeParseCellAdjustSizeForOverflow(pPage, pCell, pInfo);
  }
}
static void btreeParseCell(
  MemPage *pPage,         /* Page containing the cell */
  int iCell,              /* The cell index.  First cell is 0 */
  CellInfo *pInfo         /* Fill in this structure */
){
  pPage->xParseCell(pPage, findCell(pPage, iCell), pInfo);
}

/*
** The following routines are implementations of the MemPage.xCellSize
** method.
**
** Compute the total number of bytes that a Cell needs in the cell
** data area of the btree-page.  The return number includes the cell
** data header and the local payload, but not any overflow page or
** the space used by the cell pointer.
**
** cellSizePtrNoPayload()    =>   table internal nodes
** cellSizePtrTableLeaf()    =>   table leaf nodes
** cellSizePtr()             =>   index internal nodes
** cellSizeIdxLeaf()         =>   index leaf nodes
*/
static u16 cellSizePtr(MemPage *pPage, u8 *pCell){
  u8 *pIter = pCell + 4;                   /* For looping over bytes of pCell */
  u8 *pEnd;                                /* End mark for a varint */
  u32 nSize;                               /* Size value to return */

#ifdef SQLITE_DEBUG
  /* The value returned by this function should always be the same as
  ** the (CellInfo.nSize) value found by doing a full parse of the
  ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of
  ** this function verifies that this invariant is not violated. */
  CellInfo debuginfo;
  pPage->xParseCell(pPage, pCell, &debuginfo);
#endif

  assert( pPage->childPtrSize==4 );
  nSize = *pIter;
  if( nSize>=0x80 ){
    pEnd = &pIter[8];
    nSize &= 0x7f;
    do{
      nSize = (nSize<<7) | (*++pIter & 0x7f);
    }while( *(pIter)>=0x80 && pIter<pEnd );
  }
  pIter++;
  testcase( nSize==pPage->maxLocal );
  testcase( nSize==(u32)pPage->maxLocal+1 );
  if( nSize<=pPage->maxLocal ){
    nSize += (u32)(pIter - pCell);
    assert( nSize>4 );
  }else{
    int minLocal = pPage->minLocal;
    nSize = minLocal + (nSize - minLocal) % (pPage->pBt->usableSize - 4);
    testcase( nSize==pPage->maxLocal );
    testcase( nSize==(u32)pPage->maxLocal+1 );
    if( nSize>pPage->maxLocal ){
      nSize = minLocal;
    }
    nSize += 4 + (u16)(pIter - pCell);
  }
  assert( nSize==debuginfo.nSize || CORRUPT_DB );
  return (u16)nSize;
}
static u16 cellSizePtrIdxLeaf(MemPage *pPage, u8 *pCell){
  u8 *pIter = pCell;                       /* For looping over bytes of pCell */
  u8 *pEnd;                                /* End mark for a varint */
  u32 nSize;                               /* Size value to return */

#ifdef SQLITE_DEBUG
  /* The value returned by this function should always be the same as
  ** the (CellInfo.nSize) value found by doing a full parse of the
  ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of
  ** this function verifies that this invariant is not violated. */
  CellInfo debuginfo;
  pPage->xParseCell(pPage, pCell, &debuginfo);
#endif

  assert( pPage->childPtrSize==0 );
  nSize = *pIter;
  if( nSize>=0x80 ){
    pEnd = &pIter[8];
    nSize &= 0x7f;
    do{
      nSize = (nSize<<7) | (*++pIter & 0x7f);
    }while( *(pIter)>=0x80 && pIter<pEnd );
  }
  pIter++;
  testcase( nSize==pPage->maxLocal );
  testcase( nSize==(u32)pPage->maxLocal+1 );
  if( nSize<=pPage->maxLocal ){
    nSize += (u32)(pIter - pCell);
    if( nSize<4 ) nSize = 4;
  }else{
    int minLocal = pPage->minLocal;
    nSize = minLocal + (nSize - minLocal) % (pPage->pBt->usableSize - 4);
    testcase( nSize==pPage->maxLocal );
    testcase( nSize==(u32)pPage->maxLocal+1 );
    if( nSize>pPage->maxLocal ){
      nSize = minLocal;
    }
    nSize += 4 + (u16)(pIter - pCell);
  }
  assert( nSize==debuginfo.nSize || CORRUPT_DB );
  return (u16)nSize;
}
static u16 cellSizePtrNoPayload(MemPage *pPage, u8 *pCell){
  u8 *pIter = pCell + 4; /* For looping over bytes of pCell */
  u8 *pEnd;              /* End mark for a varint */

#ifdef SQLITE_DEBUG
  /* The value returned by this function should always be the same as
  ** the (CellInfo.nSize) value found by doing a full parse of the
  ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of
  ** this function verifies that this invariant is not violated. */
  CellInfo debuginfo;
  pPage->xParseCell(pPage, pCell, &debuginfo);
#else
  UNUSED_PARAMETER(pPage);
#endif

  assert( pPage->childPtrSize==4 );
  pEnd = pIter + 9;
  while( (*pIter++)&0x80 && pIter<pEnd );
  assert( debuginfo.nSize==(u16)(pIter - pCell) || CORRUPT_DB );
  return (u16)(pIter - pCell);
}
static u16 cellSizePtrTableLeaf(MemPage *pPage, u8 *pCell){
  u8 *pIter = pCell;   /* For looping over bytes of pCell */
  u8 *pEnd;            /* End mark for a varint */
  u32 nSize;           /* Size value to return */

#ifdef SQLITE_DEBUG
  /* The value returned by this function should always be the same as
  ** the (CellInfo.nSize) value found by doing a full parse of the
  ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of
  ** this function verifies that this invariant is not violated. */
  CellInfo debuginfo;
  pPage->xParseCell(pPage, pCell, &debuginfo);
#endif

  nSize = *pIter;
  if( nSize>=0x80 ){
    pEnd = &pIter[8];
    nSize &= 0x7f;
    do{
      nSize = (nSize<<7) | (*++pIter & 0x7f);
    }while( *(pIter)>=0x80 && pIter<pEnd );
  }
  pIter++;
  /* pIter now points at the 64-bit integer key value, a variable length
  ** integer. The following block moves pIter to point at the first byte
  ** past the end of the key value. */
  if( (*pIter++)&0x80
   && (*pIter++)&0x80
   && (*pIter++)&0x80
   && (*pIter++)&0x80
   && (*pIter++)&0x80
   && (*pIter++)&0x80
   && (*pIter++)&0x80
   && (*pIter++)&0x80 ){ pIter++; }
  testcase( nSize==pPage->maxLocal );
  testcase( nSize==(u32)pPage->maxLocal+1 );
  if( nSize<=pPage->maxLocal ){
    nSize += (u32)(pIter - pCell);
    if( nSize<4 ) nSize = 4;
  }else{
    int minLocal = pPage->minLocal;
    nSize = minLocal + (nSize - minLocal) % (pPage->pBt->usableSize - 4);
    testcase( nSize==pPage->maxLocal );
    testcase( nSize==(u32)pPage->maxLocal+1 );
    if( nSize>pPage->maxLocal ){
      nSize = minLocal;
    }
    nSize += 4 + (u16)(pIter - pCell);
  }
  assert( nSize==debuginfo.nSize || CORRUPT_DB );
  return (u16)nSize;
}


#ifdef SQLITE_DEBUG
/* This variation on cellSizePtr() is used inside of assert() statements
** only. */
static u16 cellSize(MemPage *pPage, int iCell){
  return pPage->xCellSize(pPage, findCell(pPage, iCell));
}
#endif

#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** The cell pCell is currently part of page pSrc but will ultimately be part
** of pPage.  (pSrc and pPage are often the same.)  If pCell contains a
** pointer to an overflow page, insert an entry into the pointer-map for
** the overflow page that will be valid after pCell has been moved to pPage.
*/
static void ptrmapPutOvflPtr(MemPage *pPage, MemPage *pSrc, u8 *pCell,int *pRC){
  CellInfo info;
  if( *pRC ) return;
  assert( pCell!=0 );
  pPage->xParseCell(pPage, pCell, &info);
  if( info.nLocal<info.nPayload ){
    Pgno ovfl;
    if( SQLITE_OVERFLOW(pSrc->aDataEnd, pCell, pCell+info.nLocal) ){
      testcase( pSrc!=pPage );
      *pRC = SQLITE_CORRUPT_BKPT;
      return;
    }
    ovfl = get4byte(&pCell[info.nSize-4]);
    ptrmapPut(pPage->pBt, ovfl, PTRMAP_OVERFLOW1, pPage->pgno, pRC);
  }
}
#endif


/*
** Defragment the page given. This routine reorganizes cells within the
** page so that there are no free-blocks on the free-block list.
**
** Parameter nMaxFrag is the maximum amount of fragmented space that may be
** present in the page after this routine returns.
**
** EVIDENCE-OF: R-44582-60138 SQLite may from time to time reorganize a
** b-tree page so that there are no freeblocks or fragment bytes, all
** unused bytes are contained in the unallocated space region, and all
** cells are packed tightly at the end of the page.
*/
static int defragmentPage(MemPage *pPage, int nMaxFrag){
  int i;                     /* Loop counter */
  int pc;                    /* Address of the i-th cell */
  int hdr;                   /* Offset to the page header */
  int size;                  /* Size of a cell */
  int usableSize;            /* Number of usable bytes on a page */
  int cellOffset;            /* Offset to the cell pointer array */
  int cbrk;                  /* Offset to the cell content area */
  int nCell;                 /* Number of cells on the page */
  unsigned char *data;       /* The page data */
  unsigned char *temp;       /* Temp area for cell content */
  unsigned char *src;        /* Source of content */
  int iCellFirst;            /* First allowable cell index */
  int iCellLast;             /* Last possible cell index */
  int iCellStart;            /* First cell offset in input */

  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( pPage->pBt!=0 );
  assert( pPage->pBt->usableSize <= SQLITE_MAX_PAGE_SIZE );
  assert( pPage->nOverflow==0 );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  data = pPage->aData;
  hdr = pPage->hdrOffset;
  cellOffset = pPage->cellOffset;
  nCell = pPage->nCell;
  assert( nCell==get2byte(&data[hdr+3]) || CORRUPT_DB );
  iCellFirst = cellOffset + 2*nCell;
  usableSize = pPage->pBt->usableSize;

  /* This block handles pages with two or fewer free blocks and nMaxFrag
  ** or fewer fragmented bytes. In this case it is faster to move the
  ** two (or one) blocks of cells using memmove() and add the required
  ** offsets to each pointer in the cell-pointer array than it is to
  ** reconstruct the entire page.  */
  if( (int)data[hdr+7]<=nMaxFrag ){
    int iFree = get2byte(&data[hdr+1]);
    if( iFree>usableSize-4 ) return SQLITE_CORRUPT_PAGE(pPage);
    if( iFree ){
      int iFree2 = get2byte(&data[iFree]);
      if( iFree2>usableSize-4 ) return SQLITE_CORRUPT_PAGE(pPage);
      if( 0==iFree2 || (data[iFree2]==0 && data[iFree2+1]==0) ){
        u8 *pEnd = &data[cellOffset + nCell*2];
        u8 *pAddr;
        int sz2 = 0;
        int sz = get2byte(&data[iFree+2]);
        int top = get2byte(&data[hdr+5]);
        if( top>=iFree ){
          return SQLITE_CORRUPT_PAGE(pPage);
        }
        if( iFree2 ){
          if( iFree+sz>iFree2 ) return SQLITE_CORRUPT_PAGE(pPage);
          sz2 = get2byte(&data[iFree2+2]);
          if( iFree2+sz2 > usableSize ) return SQLITE_CORRUPT_PAGE(pPage);
          memmove(&data[iFree+sz+sz2], &data[iFree+sz], iFree2-(iFree+sz));
          sz += sz2;
        }else if( iFree+sz>usableSize ){
          return SQLITE_CORRUPT_PAGE(pPage);
        }

        cbrk = top+sz;
        assert( cbrk+(iFree-top) <= usableSize );
        memmove(&data[cbrk], &data[top], iFree-top);
        for(pAddr=&data[cellOffset]; pAddr<pEnd; pAddr+=2){
          pc = get2byte(pAddr);
          if( pc<iFree ){ put2byte(pAddr, pc+sz); }
          else if( pc<iFree2 ){ put2byte(pAddr, pc+sz2); }
        }
        goto defragment_out;
      }
    }
  }

  cbrk = usableSize;
  iCellLast = usableSize - 4;
  iCellStart = get2byte(&data[hdr+5]);
  if( nCell>0 ){
    temp = sqlite3PagerTempSpace(pPage->pBt->pPager);
    memcpy(temp, data, usableSize);
    src = temp;
    for(i=0; i<nCell; i++){
      u8 *pAddr;     /* The i-th cell pointer */
      pAddr = &data[cellOffset + i*2];
      pc = get2byte(pAddr);
      testcase( pc==iCellFirst );
      testcase( pc==iCellLast );
      /* These conditions have already been verified in btreeInitPage()
      ** if PRAGMA cell_size_check=ON.
      */
      if( pc>iCellLast ){
        return SQLITE_CORRUPT_PAGE(pPage);
      }
      assert( pc>=0 && pc<=iCellLast );
      size = pPage->xCellSize(pPage, &src[pc]);
      cbrk -= size;
      if( cbrk<iCellStart || pc+size>usableSize ){
        return SQLITE_CORRUPT_PAGE(pPage);
      }
      assert( cbrk+size<=usableSize && cbrk>=iCellStart );
      testcase( cbrk+size==usableSize );
      testcase( pc+size==usableSize );
      put2byte(pAddr, cbrk);
      memcpy(&data[cbrk], &src[pc], size);
    }
  }
  data[hdr+7] = 0;

defragment_out:
  assert( pPage->nFree>=0 );
  if( data[hdr+7]+cbrk-iCellFirst!=pPage->nFree ){
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  assert( cbrk>=iCellFirst );
  put2byte(&data[hdr+5], cbrk);
  data[hdr+1] = 0;
  data[hdr+2] = 0;
  memset(&data[iCellFirst], 0, cbrk-iCellFirst);
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  return SQLITE_OK;
}

/*
** Search the free-list on page pPg for space to store a cell nByte bytes in
** size. If one can be found, return a pointer to the space and remove it
** from the free-list.
**
** If no suitable space can be found on the free-list, return NULL.
**
** This function may detect corruption within pPg.  If corruption is
** detected then *pRc is set to SQLITE_CORRUPT and NULL is returned.
**
** Slots on the free list that are between 1 and 3 bytes larger than nByte
** will be ignored if adding the extra space to the fragmentation count
** causes the fragmentation count to exceed 60.
*/
static u8 *pageFindSlot(MemPage *pPg, int nByte, int *pRc){
  const int hdr = pPg->hdrOffset;            /* Offset to page header */
  u8 * const aData = pPg->aData;             /* Page data */
  int iAddr = hdr + 1;                       /* Address of ptr to pc */
  u8 *pTmp = &aData[iAddr];                  /* Temporary ptr into aData[] */
  int pc = get2byte(pTmp);                   /* Address of a free slot */
  int x;                                     /* Excess size of the slot */
  int maxPC = pPg->pBt->usableSize - nByte;  /* Max address for a usable slot */
  int size;                                  /* Size of the free slot */

  assert( pc>0 );
  while( pc<=maxPC ){
    /* EVIDENCE-OF: R-22710-53328 The third and fourth bytes of each
    ** freeblock form a big-endian integer which is the size of the freeblock
    ** in bytes, including the 4-byte header. */
    pTmp = &aData[pc+2];
    size = get2byte(pTmp);
    if( (x = size - nByte)>=0 ){
      testcase( x==4 );
      testcase( x==3 );
      if( x<4 ){
        /* EVIDENCE-OF: R-11498-58022 In a well-formed b-tree page, the total
        ** number of bytes in fragments may not exceed 60. */
        if( aData[hdr+7]>57 ) return 0;

        /* Remove the slot from the free-list. Update the number of
        ** fragmented bytes within the page. */
        memcpy(&aData[iAddr], &aData[pc], 2);
        aData[hdr+7] += (u8)x;
        return &aData[pc];
      }else if( x+pc > maxPC ){
        /* This slot extends off the end of the usable part of the page */
        *pRc = SQLITE_CORRUPT_PAGE(pPg);
        return 0;
      }else{
        /* The slot remains on the free-list. Reduce its size to account
        ** for the portion used by the new allocation. */
        put2byte(&aData[pc+2], x);
      }
      return &aData[pc + x];
    }
    iAddr = pc;
    pTmp = &aData[pc];
    pc = get2byte(pTmp);
    if( pc<=iAddr ){
      if( pc ){
        /* The next slot in the chain comes before the current slot */
        *pRc = SQLITE_CORRUPT_PAGE(pPg);
      }
      return 0;
    }
  }
  if( pc>maxPC+nByte-4 ){
    /* The free slot chain extends off the end of the page */
    *pRc = SQLITE_CORRUPT_PAGE(pPg);
  }
  return 0;
}

/*
** Allocate nByte bytes of space from within the B-Tree page passed
** as the first argument. Write into *pIdx the index into pPage->aData[]
** of the first byte of allocated space. Return either SQLITE_OK or
** an error code (usually SQLITE_CORRUPT).
**
** The caller guarantees that there is sufficient space to make the
** allocation.  This routine might need to defragment in order to bring
** all the space together, however.  This routine will avoid using
** the first two bytes past the cell pointer area since presumably this
** allocation is being made in order to insert a new cell, so we will
** also end up needing a new cell pointer.
*/
static SQLITE_INLINE int allocateSpace(MemPage *pPage, int nByte, int *pIdx){
  const int hdr = pPage->hdrOffset;    /* Local cache of pPage->hdrOffset */
  u8 * const data = pPage->aData;      /* Local cache of pPage->aData */
  int top;                             /* First byte of cell content area */
  int rc = SQLITE_OK;                  /* Integer return code */
  u8 *pTmp;                            /* Temp ptr into data[] */
  int gap;        /* First byte of gap between cell pointers and cell content */
 
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( pPage->pBt );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( nByte>=0 );  /* Minimum cell size is 4 */
  assert( pPage->nFree>=nByte );
  assert( pPage->nOverflow==0 );
  assert( nByte < (int)(pPage->pBt->usableSize-8) );

  assert( pPage->cellOffset == hdr + 12 - 4*pPage->leaf );
  gap = pPage->cellOffset + 2*pPage->nCell;
  assert( gap<=65536 );
  /* EVIDENCE-OF: R-29356-02391 If the database uses a 65536-byte page size
  ** and the reserved space is zero (the usual value for reserved space)
  ** then the cell content offset of an empty page wants to be 65536.
  ** However, that integer is too large to be stored in a 2-byte unsigned
  ** integer, so a value of 0 is used in its place. */
  pTmp = &data[hdr+5];
  top = get2byte(pTmp);
  if( gap>top ){
    if( top==0 && pPage->pBt->usableSize==65536 ){
      top = 65536;
    }else{
      return SQLITE_CORRUPT_PAGE(pPage);
    }
  }else if( top>(int)pPage->pBt->usableSize ){
    return SQLITE_CORRUPT_PAGE(pPage);
  }

  /* If there is enough space between gap and top for one more cell pointer,
  ** and if the freelist is not empty, then search the
  ** freelist looking for a slot big enough to satisfy the request.
  */
  testcase( gap+2==top );
  testcase( gap+1==top );
  testcase( gap==top );
  if( (data[hdr+2] || data[hdr+1]) && gap+2<=top ){
    u8 *pSpace = pageFindSlot(pPage, nByte, &rc);
    if( pSpace ){
      int g2;
      assert( pSpace+nByte<=data+pPage->pBt->usableSize );
      *pIdx = g2 = (int)(pSpace-data);
      if( g2<=gap ){
        return SQLITE_CORRUPT_PAGE(pPage);
      }else{
        return SQLITE_OK;
      }
    }else if( rc ){
      return rc;
    }
  }

  /* The request could not be fulfilled using a freelist slot.  Check
  ** to see if defragmentation is necessary.
  */
  testcase( gap+2+nByte==top );
  if( gap+2+nByte>top ){
    assert( pPage->nCell>0 || CORRUPT_DB );
    assert( pPage->nFree>=0 );
    rc = defragmentPage(pPage, MIN(4, pPage->nFree - (2+nByte)));
    if( rc ) return rc;
    top = get2byteNotZero(&data[hdr+5]);
    assert( gap+2+nByte<=top );
  }


  /* Allocate memory from the gap in between the cell pointer array
  ** and the cell content area.  The btreeComputeFreeSpace() call has already
  ** validated the freelist.  Given that the freelist is valid, there
  ** is no way that the allocation can extend off the end of the page.
  ** The assert() below verifies the previous sentence.
  */
  top -= nByte;
  put2byte(&data[hdr+5], top);
  assert( top+nByte <= (int)pPage->pBt->usableSize );
  *pIdx = top;
  return SQLITE_OK;
}

/*
** Return a section of the pPage->aData to the freelist.
** The first byte of the new free block is pPage->aData[iStart]
** and the size of the block is iSize bytes.
**
** Adjacent freeblocks are coalesced.
**
** Even though the freeblock list was checked by btreeComputeFreeSpace(),
** that routine will not detect overlap between cells or freeblocks.  Nor
** does it detect cells or freeblocks that encroach into the reserved bytes
** at the end of the page.  So do additional corruption checks inside this
** routine and return SQLITE_CORRUPT if any problems are found.
*/
static int freeSpace(MemPage *pPage, u16 iStart, u16 iSize){
  u16 iPtr;                             /* Address of ptr to next freeblock */
  u16 iFreeBlk;                         /* Address of the next freeblock */
  u8 hdr;                               /* Page header size.  0 or 100 */
  u8 nFrag = 0;                         /* Reduction in fragmentation */
  u16 iOrigSize = iSize;                /* Original value of iSize */
  u16 x;                                /* Offset to cell content area */
  u32 iEnd = iStart + iSize;            /* First byte past the iStart buffer */
  unsigned char *data = pPage->aData;   /* Page content */
  u8 *pTmp;                             /* Temporary ptr into data[] */

  assert( pPage->pBt!=0 );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( CORRUPT_DB || iStart>=pPage->hdrOffset+6+pPage->childPtrSize );
  assert( CORRUPT_DB || iEnd <= pPage->pBt->usableSize );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( iSize>=4 );   /* Minimum cell size is 4 */
  assert( CORRUPT_DB || iStart<=pPage->pBt->usableSize-4 );

  /* The list of freeblocks must be in ascending order.  Find the
  ** spot on the list where iStart should be inserted.
  */
  hdr = pPage->hdrOffset;
  iPtr = hdr + 1;
  if( data[iPtr+1]==0 && data[iPtr]==0 ){
    iFreeBlk = 0;  /* Shortcut for the case when the freelist is empty */
  }else{
    while( (iFreeBlk = get2byte(&data[iPtr]))<iStart ){
      if( iFreeBlk<=iPtr ){
        if( iFreeBlk==0 ) break; /* TH3: corrupt082.100 */
        return SQLITE_CORRUPT_PAGE(pPage);
      }
      iPtr = iFreeBlk;
    }
    if( iFreeBlk>pPage->pBt->usableSize-4 ){ /* TH3: corrupt081.100 */
      return SQLITE_CORRUPT_PAGE(pPage);
    }
    assert( iFreeBlk>iPtr || iFreeBlk==0 || CORRUPT_DB );
 
    /* At this point:
    **    iFreeBlk:   First freeblock after iStart, or zero if none
    **    iPtr:       The address of a pointer to iFreeBlk
    **
    ** Check to see if iFreeBlk should be coalesced onto the end of iStart.
    */
    if( iFreeBlk && iEnd+3>=iFreeBlk ){
      nFrag = iFreeBlk - iEnd;
      if( iEnd>iFreeBlk ) return SQLITE_CORRUPT_PAGE(pPage);
      iEnd = iFreeBlk + get2byte(&data[iFreeBlk+2]);
      if( iEnd > pPage->pBt->usableSize ){
        return SQLITE_CORRUPT_PAGE(pPage);
      }
      iSize = iEnd - iStart;
      iFreeBlk = get2byte(&data[iFreeBlk]);
    }
 
    /* If iPtr is another freeblock (that is, if iPtr is not the freelist
    ** pointer in the page header) then check to see if iStart should be
    ** coalesced onto the end of iPtr.
    */
    if( iPtr>hdr+1 ){
      int iPtrEnd = iPtr + get2byte(&data[iPtr+2]);
      if( iPtrEnd+3>=iStart ){
        if( iPtrEnd>iStart ) return SQLITE_CORRUPT_PAGE(pPage);
        nFrag += iStart - iPtrEnd;
        iSize = iEnd - iPtr;
        iStart = iPtr;
      }
    }
    if( nFrag>data[hdr+7] ) return SQLITE_CORRUPT_PAGE(pPage);
    data[hdr+7] -= nFrag;
  }
  pTmp = &data[hdr+5];
  x = get2byte(pTmp);
  if( pPage->pBt->btsFlags & BTS_FAST_SECURE ){
    /* Overwrite deleted information with zeros when the secure_delete
    ** option is enabled */
    memset(&data[iStart], 0, iSize);
  }
  if( iStart<=x ){
    /* The new freeblock is at the beginning of the cell content area,
    ** so just extend the cell content area rather than create another
    ** freelist entry */
    if( iStart<x ) return SQLITE_CORRUPT_PAGE(pPage);
    if( iPtr!=hdr+1 ) return SQLITE_CORRUPT_PAGE(pPage);
    put2byte(&data[hdr+1], iFreeBlk);
    put2byte(&data[hdr+5], iEnd);
  }else{
    /* Insert the new freeblock into the freelist */
    put2byte(&data[iPtr], iStart);
    put2byte(&data[iStart], iFreeBlk);
    put2byte(&data[iStart+2], iSize);
  }
  pPage->nFree += iOrigSize;
  return SQLITE_OK;
}

/*
** Decode the flags byte (the first byte of the header) for a page
** and initialize fields of the MemPage structure accordingly.
**
** Only the following combinations are supported.  Anything different
** indicates a corrupt database files:
**
**         PTF_ZERODATA                             (0x02,  2)
**         PTF_LEAFDATA | PTF_INTKEY                (0x05,  5)
**         PTF_ZERODATA | PTF_LEAF                  (0x0a, 10)
**         PTF_LEAFDATA | PTF_INTKEY | PTF_LEAF     (0x0d, 13)
*/
static int decodeFlags(MemPage *pPage, int flagByte){
  BtShared *pBt;     /* A copy of pPage->pBt */

  assert( pPage->hdrOffset==(pPage->pgno==1 ? 100 : 0) );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  pBt = pPage->pBt;
  pPage->max1bytePayload = pBt->max1bytePayload;
  if( flagByte>=(PTF_ZERODATA | PTF_LEAF) ){
    pPage->childPtrSize = 0;
    pPage->leaf = 1;
    if( flagByte==(PTF_LEAFDATA | PTF_INTKEY | PTF_LEAF) ){
      pPage->intKeyLeaf = 1;
      pPage->xCellSize = cellSizePtrTableLeaf;
      pPage->xParseCell = btreeParseCellPtr;
      pPage->intKey = 1;
      pPage->maxLocal = pBt->maxLeaf;
      pPage->minLocal = pBt->minLeaf;
    }else if( flagByte==(PTF_ZERODATA | PTF_LEAF) ){
      pPage->intKey = 0;
      pPage->intKeyLeaf = 0;
      pPage->xCellSize = cellSizePtrIdxLeaf;
      pPage->xParseCell = btreeParseCellPtrIndex;
      pPage->maxLocal = pBt->maxLocal;
      pPage->minLocal = pBt->minLocal;
    }else{
      pPage->intKey = 0;
      pPage->intKeyLeaf = 0;
      pPage->xCellSize = cellSizePtrIdxLeaf;
      pPage->xParseCell = btreeParseCellPtrIndex;
      return SQLITE_CORRUPT_PAGE(pPage);
    }
  }else{
    pPage->childPtrSize = 4;
    pPage->leaf = 0;
    if( flagByte==(PTF_ZERODATA) ){
      pPage->intKey = 0;
      pPage->intKeyLeaf = 0;
      pPage->xCellSize = cellSizePtr;
      pPage->xParseCell = btreeParseCellPtrIndex;
      pPage->maxLocal = pBt->maxLocal;
      pPage->minLocal = pBt->minLocal;
    }else if( flagByte==(PTF_LEAFDATA | PTF_INTKEY) ){
      pPage->intKeyLeaf = 0;
      pPage->xCellSize = cellSizePtrNoPayload;
      pPage->xParseCell = btreeParseCellPtrNoPayload;
      pPage->intKey = 1;
      pPage->maxLocal = pBt->maxLeaf;
      pPage->minLocal = pBt->minLeaf;
    }else{
      pPage->intKey = 0;
      pPage->intKeyLeaf = 0;
      pPage->xCellSize = cellSizePtr;
      pPage->xParseCell = btreeParseCellPtrIndex;
      return SQLITE_CORRUPT_PAGE(pPage);
    }
  }
  return SQLITE_OK;
}

/*
** Compute the amount of freespace on the page.  In other words, fill
** in the pPage->nFree field.
*/
static int btreeComputeFreeSpace(MemPage *pPage){
  int pc;            /* Address of a freeblock within pPage->aData[] */
  u8 hdr;            /* Offset to beginning of page header */
  u8 *data;          /* Equal to pPage->aData */
  int usableSize;    /* Amount of usable space on each page */
  int nFree;         /* Number of unused bytes on the page */
  int top;           /* First byte of the cell content area */
  int iCellFirst;    /* First allowable cell or freeblock offset */
  int iCellLast;     /* Last possible cell or freeblock offset */

  assert( pPage->pBt!=0 );
  assert( pPage->pBt->db!=0 );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) );
  assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) );
  assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) );
  assert( pPage->isInit==1 );
  assert( pPage->nFree<0 );

  usableSize = pPage->pBt->usableSize;
  hdr = pPage->hdrOffset;
  data = pPage->aData;
  /* EVIDENCE-OF: R-58015-48175 The two-byte integer at offset 5 designates
  ** the start of the cell content area. A zero value for this integer is
  ** interpreted as 65536. */
  top = get2byteNotZero(&data[hdr+5]);
  iCellFirst = hdr + 8 + pPage->childPtrSize + 2*pPage->nCell;
  iCellLast = usableSize - 4;

  /* Compute the total free space on the page
  ** EVIDENCE-OF: R-23588-34450 The two-byte integer at offset 1 gives the
  ** start of the first freeblock on the page, or is zero if there are no
  ** freeblocks. */
  pc = get2byte(&data[hdr+1]);
  nFree = data[hdr+7] + top;  /* Init nFree to non-freeblock free space */
  if( pc>0 ){
    u32 next, size;
    if( pc<top ){
      /* EVIDENCE-OF: R-55530-52930 In a well-formed b-tree page, there will
      ** always be at least one cell before the first freeblock.
      */
      return SQLITE_CORRUPT_PAGE(pPage);
    }
    while( 1 ){
      if( pc>iCellLast ){
        /* Freeblock off the end of the page */
        return SQLITE_CORRUPT_PAGE(pPage);
      }
      next = get2byte(&data[pc]);
      size = get2byte(&data[pc+2]);
      nFree = nFree + size;
      if( next<=pc+size+3 ) break;
      pc = next;
    }
    if( next>0 ){
      /* Freeblock not in ascending order */
      return SQLITE_CORRUPT_PAGE(pPage);
    }
    if( pc+size>(unsigned int)usableSize ){
      /* Last freeblock extends past page end */
      return SQLITE_CORRUPT_PAGE(pPage);
    }
  }

  /* At this point, nFree contains the sum of the offset to the start
  ** of the cell-content area plus the number of free bytes within
  ** the cell-content area. If this is greater than the usable-size
  ** of the page, then the page must be corrupted. This check also
  ** serves to verify that the offset to the start of the cell-content
  ** area, according to the page header, lies within the page.
  */
  if( nFree>usableSize || nFree<iCellFirst ){
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  pPage->nFree = (u16)(nFree - iCellFirst);
  return SQLITE_OK;
}

/*
** Do additional sanity check after btreeInitPage() if
** PRAGMA cell_size_check=ON
*/
static SQLITE_NOINLINE int btreeCellSizeCheck(MemPage *pPage){
  int iCellFirst;    /* First allowable cell or freeblock offset */
  int iCellLast;     /* Last possible cell or freeblock offset */
  int i;             /* Index into the cell pointer array */
  int sz;            /* Size of a cell */
  int pc;            /* Address of a freeblock within pPage->aData[] */
  u8 *data;          /* Equal to pPage->aData */
  int usableSize;    /* Maximum usable space on the page */
  int cellOffset;    /* Start of cell content area */

  iCellFirst = pPage->cellOffset + 2*pPage->nCell;
  usableSize = pPage->pBt->usableSize;
  iCellLast = usableSize - 4;
  data = pPage->aData;
  cellOffset = pPage->cellOffset;
  if( !pPage->leaf ) iCellLast--;
  for(i=0; i<pPage->nCell; i++){
    pc = get2byteAligned(&data[cellOffset+i*2]);
    testcase( pc==iCellFirst );
    testcase( pc==iCellLast );
    if( pc<iCellFirst || pc>iCellLast ){
      return SQLITE_CORRUPT_PAGE(pPage);
    }
    sz = pPage->xCellSize(pPage, &data[pc]);
    testcase( pc+sz==usableSize );
    if( pc+sz>usableSize ){
      return SQLITE_CORRUPT_PAGE(pPage);
    }
  }
  return SQLITE_OK;
}

/*
** Initialize the auxiliary information for a disk block.
**
** Return SQLITE_OK on success.  If we see that the page does
** not contain a well-formed database page, then return
** SQLITE_CORRUPT.  Note that a return of SQLITE_OK does not
** guarantee that the page is well-formed.  It only shows that
** we failed to detect any corruption.
*/
static int btreeInitPage(MemPage *pPage){
  u8 *data;          /* Equal to pPage->aData */
  BtShared *pBt;        /* The main btree structure */

  assert( pPage->pBt!=0 );
  assert( pPage->pBt->db!=0 );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) );
  assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) );
  assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) );
  assert( pPage->isInit==0 );

  pBt = pPage->pBt;
  data = pPage->aData + pPage->hdrOffset;
  /* EVIDENCE-OF: R-28594-02890 The one-byte flag at offset 0 indicating
  ** the b-tree page type. */
  if( decodeFlags(pPage, data[0]) ){
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  assert( pBt->pageSize>=512 && pBt->pageSize<=65536 );
  pPage->maskPage = (u16)(pBt->pageSize - 1);
  pPage->nOverflow = 0;
  pPage->cellOffset = pPage->hdrOffset + 8 + pPage->childPtrSize;
  pPage->aCellIdx = data + pPage->childPtrSize + 8;
  pPage->aDataEnd = pPage->aData + pBt->pageSize;
  pPage->aDataOfst = pPage->aData + pPage->childPtrSize;
  /* EVIDENCE-OF: R-37002-32774 The two-byte integer at offset 3 gives the
  ** number of cells on the page. */
  pPage->nCell = get2byte(&data[3]);
  if( pPage->nCell>MX_CELL(pBt) ){
    /* To many cells for a single page.  The page must be corrupt */
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  testcase( pPage->nCell==MX_CELL(pBt) );
  /* EVIDENCE-OF: R-24089-57979 If a page contains no cells (which is only
  ** possible for a root page of a table that contains no rows) then the
  ** offset to the cell content area will equal the page size minus the
  ** bytes of reserved space. */
  assert( pPage->nCell>0
       || get2byteNotZero(&data[5])==(int)pBt->usableSize
       || CORRUPT_DB );
  pPage->nFree = -1;  /* Indicate that this value is yet uncomputed */
  pPage->isInit = 1;
  if( pBt->db->flags & SQLITE_CellSizeCk ){
    return btreeCellSizeCheck(pPage);
  }
  return SQLITE_OK;
}

/*
** Set up a raw page so that it looks like a database page holding
** no entries.
*/
static void zeroPage(MemPage *pPage, int flags){
  unsigned char *data = pPage->aData;
  BtShared *pBt = pPage->pBt;
  u8 hdr = pPage->hdrOffset;
  u16 first;

  assert( sqlite3PagerPagenumber(pPage->pDbPage)==pPage->pgno || CORRUPT_DB );
  assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage );
  assert( sqlite3PagerGetData(pPage->pDbPage) == data );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( sqlite3_mutex_held(pBt->mutex) );
  if( pBt->btsFlags & BTS_FAST_SECURE ){
    memset(&data[hdr], 0, pBt->usableSize - hdr);
  }
  data[hdr] = (char)flags;
  first = hdr + ((flags&PTF_LEAF)==0 ? 12 : 8);
  memset(&data[hdr+1], 0, 4);
  data[hdr+7] = 0;
  put2byte(&data[hdr+5], pBt->usableSize);
  pPage->nFree = (u16)(pBt->usableSize - first);
  decodeFlags(pPage, flags);
  pPage->cellOffset = first;
  pPage->aDataEnd = &data[pBt->pageSize];
  pPage->aCellIdx = &data[first];
  pPage->aDataOfst = &data[pPage->childPtrSize];
  pPage->nOverflow = 0;
  assert( pBt->pageSize>=512 && pBt->pageSize<=65536 );
  pPage->maskPage = (u16)(pBt->pageSize - 1);
  pPage->nCell = 0;
  pPage->isInit = 1;
}


/*
** Convert a DbPage obtained from the pager into a MemPage used by
** the btree layer.
*/
static MemPage *btreePageFromDbPage(DbPage *pDbPage, Pgno pgno, BtShared *pBt){
  MemPage *pPage = (MemPage*)sqlite3PagerGetExtra(pDbPage);
  if( pgno!=pPage->pgno ){
    pPage->aData = sqlite3PagerGetData(pDbPage);
    pPage->pDbPage = pDbPage;
    pPage->pBt = pBt;
    pPage->pgno = pgno;
    pPage->hdrOffset = pgno==1 ? 100 : 0;
  }
  assert( pPage->aData==sqlite3PagerGetData(pDbPage) );
  return pPage;
}

/*
** Get a page from the pager.  Initialize the MemPage.pBt and
** MemPage.aData elements if needed.  See also: btreeGetUnusedPage().
**
** If the PAGER_GET_NOCONTENT flag is set, it means that we do not care
** about the content of the page at this time.  So do not go to the disk
** to fetch the content.  Just fill in the content with zeros for now.
** If in the future we call sqlite3PagerWrite() on this page, that
** means we have started to be concerned about content and the disk
** read should occur at that point.
*/
static int btreeGetPage(
  BtShared *pBt,       /* The btree */
  Pgno pgno,           /* Number of the page to fetch */
  MemPage **ppPage,    /* Return the page in this parameter */
  int flags            /* PAGER_GET_NOCONTENT or PAGER_GET_READONLY */
){
  int rc;
  DbPage *pDbPage;

  assert( flags==0 || flags==PAGER_GET_NOCONTENT || flags==PAGER_GET_READONLY );
  assert( sqlite3_mutex_held(pBt->mutex) );
  rc = sqlite3PagerGet(pBt->pPager, pgno, (DbPage**)&pDbPage, flags);
  if( rc ) return rc;
  *ppPage = btreePageFromDbPage(pDbPage, pgno, pBt);
  return SQLITE_OK;
}

/*
** Retrieve a page from the pager cache. If the requested page is not
** already in the pager cache return NULL. Initialize the MemPage.pBt and
** MemPage.aData elements if needed.
*/
static MemPage *btreePageLookup(BtShared *pBt, Pgno pgno){
  DbPage *pDbPage;
  assert( sqlite3_mutex_held(pBt->mutex) );
  pDbPage = sqlite3PagerLookup(pBt->pPager, pgno);
  if( pDbPage ){
    return btreePageFromDbPage(pDbPage, pgno, pBt);
  }
  return 0;
}

/*
** Return the size of the database file in pages. If there is any kind of
** error, return ((unsigned int)-1).
*/
static Pgno btreePagecount(BtShared *pBt){
  return pBt->nPage;
}
Pgno sqlite3BtreeLastPage(Btree *p){
  assert( sqlite3BtreeHoldsMutex(p) );
  return btreePagecount(p->pBt);
}

/*
** Get a page from the pager and initialize it.
*/
static int getAndInitPage(
  BtShared *pBt,                  /* The database file */
  Pgno pgno,                      /* Number of the page to get */
  MemPage **ppPage,               /* Write the page pointer here */
  int bReadOnly                   /* True for a read-only page */
){
  int rc;
  DbPage *pDbPage;
  MemPage *pPage;
  assert( sqlite3_mutex_held(pBt->mutex) );

  if( pgno>btreePagecount(pBt) ){
    *ppPage = 0;
    return SQLITE_CORRUPT_BKPT;
  }
  rc = sqlite3PagerGet(pBt->pPager, pgno, (DbPage**)&pDbPage, bReadOnly);
  if( rc ){
    *ppPage = 0;
    return rc;
  }
  pPage = (MemPage*)sqlite3PagerGetExtra(pDbPage);
  if( pPage->isInit==0 ){
    btreePageFromDbPage(pDbPage, pgno, pBt);
    rc = btreeInitPage(pPage);
    if( rc!=SQLITE_OK ){
      releasePage(pPage);
      *ppPage = 0;
      return rc;
    }
  }
  assert( pPage->pgno==pgno || CORRUPT_DB );
  assert( pPage->aData==sqlite3PagerGetData(pDbPage) );
  *ppPage = pPage;
  return SQLITE_OK;
}

/*
** Release a MemPage.  This should be called once for each prior
** call to btreeGetPage.
**
** Page1 is a special case and must be released using releasePageOne().
*/
static void releasePageNotNull(MemPage *pPage){
  assert( pPage->aData );
  assert( pPage->pBt );
  assert( pPage->pDbPage!=0 );
  assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage );
  assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  sqlite3PagerUnrefNotNull(pPage->pDbPage);
}
static void releasePage(MemPage *pPage){
  if( pPage ) releasePageNotNull(pPage);
}
static void releasePageOne(MemPage *pPage){
  assert( pPage!=0 );
  assert( pPage->aData );
  assert( pPage->pBt );
  assert( pPage->pDbPage!=0 );
  assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage );
  assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  sqlite3PagerUnrefPageOne(pPage->pDbPage);
}

/*
** Get an unused page.
**
** This works just like btreeGetPage() with the addition:
**
**   *  If the page is already in use for some other purpose, immediately
**      release it and return an SQLITE_CURRUPT error.
**   *  Make sure the isInit flag is clear
*/
static int btreeGetUnusedPage(
  BtShared *pBt,       /* The btree */
  Pgno pgno,           /* Number of the page to fetch */
  MemPage **ppPage,    /* Return the page in this parameter */
  int flags            /* PAGER_GET_NOCONTENT or PAGER_GET_READONLY */
){
  int rc = btreeGetPage(pBt, pgno, ppPage, flags);
  if( rc==SQLITE_OK ){
    if( sqlite3PagerPageRefcount((*ppPage)->pDbPage)>1 ){
      releasePage(*ppPage);
      *ppPage = 0;
      return SQLITE_CORRUPT_BKPT;
    }
    (*ppPage)->isInit = 0;
  }else{
    *ppPage = 0;
  }
  return rc;
}


/*
** During a rollback, when the pager reloads information into the cache
** so that the cache is restored to its original state at the start of
** the transaction, for each page restored this routine is called.
**
** This routine needs to reset the extra data section at the end of the
** page to agree with the restored data.
*/
static void pageReinit(DbPage *pData){
  MemPage *pPage;
  pPage = (MemPage *)sqlite3PagerGetExtra(pData);
  assert( sqlite3PagerPageRefcount(pData)>0 );
  if( pPage->isInit ){
    assert( sqlite3_mutex_held(pPage->pBt->mutex) );
    pPage->isInit = 0;
    if( sqlite3PagerPageRefcount(pData)>1 ){
      /* pPage might not be a btree page;  it might be an overflow page
      ** or ptrmap page or a free page.  In those cases, the following
      ** call to btreeInitPage() will likely return SQLITE_CORRUPT.
      ** But no harm is done by this.  And it is very important that
      ** btreeInitPage() be called on every btree page so we make
      ** the call for every page that comes in for re-initializing. */
      btreeInitPage(pPage);
    }
  }
}

/*
** Invoke the busy handler for a btree.
*/
static int btreeInvokeBusyHandler(void *pArg){
  BtShared *pBt = (BtShared*)pArg;
  assert( pBt->db );
  assert( sqlite3_mutex_held(pBt->db->mutex) );
  return sqlite3InvokeBusyHandler(&pBt->db->busyHandler);
}

/*
** Open a database file.
**
** zFilename is the name of the database file.  If zFilename is NULL
** then an ephemeral database is created.  The ephemeral database might
** be exclusively in memory, or it might use a disk-based memory cache.
** Either way, the ephemeral database will be automatically deleted
** when sqlite3BtreeClose() is called.
**
** If zFilename is ":memory:" then an in-memory database is created
** that is automatically destroyed when it is closed.
**
** The "flags" parameter is a bitmask that might contain bits like
** BTREE_OMIT_JOURNAL and/or BTREE_MEMORY.
**
** If the database is already opened in the same database connection
** and we are in shared cache mode, then the open will fail with an
** SQLITE_CONSTRAINT error.  We cannot allow two or more BtShared
** objects in the same database connection since doing so will lead
** to problems with locking.
*/
int sqlite3BtreeOpen(
  sqlite3_vfs *pVfs,      /* VFS to use for this b-tree */
  const char *zFilename,  /* Name of the file containing the BTree database */
  sqlite3 *db,            /* Associated database handle */
  Btree **ppBtree,        /* Pointer to new Btree object written here */
  int flags,              /* Options */
  int vfsFlags            /* Flags passed through to sqlite3_vfs.xOpen() */
){
  BtShared *pBt = 0;             /* Shared part of btree structure */
  Btree *p;                      /* Handle to return */
  sqlite3_mutex *mutexOpen = 0;  /* Prevents a race condition. Ticket #3537 */
  int rc = SQLITE_OK;            /* Result code from this function */
  u8 nReserve;                   /* Byte of unused space on each page */
  unsigned char zDbHeader[100];  /* Database header content */

  /* True if opening an ephemeral, temporary database */
  const int isTempDb = zFilename==0 || zFilename[0]==0;

  /* Set the variable isMemdb to true for an in-memory database, or
  ** false for a file-based database.
  */
#ifdef SQLITE_OMIT_MEMORYDB
  const int isMemdb = 0;
#else
  const int isMemdb = (zFilename && strcmp(zFilename, ":memory:")==0)
                       || (isTempDb && sqlite3TempInMemory(db))
                       || (vfsFlags & SQLITE_OPEN_MEMORY)!=0;
#endif

  assert( db!=0 );
  assert( pVfs!=0 );
  assert( sqlite3_mutex_held(db->mutex) );
  assert( (flags&0xff)==flags );   /* flags fit in 8 bits */

  /* Only a BTREE_SINGLE database can be BTREE_UNORDERED */
  assert( (flags & BTREE_UNORDERED)==0 || (flags & BTREE_SINGLE)!=0 );

  /* A BTREE_SINGLE database is always a temporary and/or ephemeral */
  assert( (flags & BTREE_SINGLE)==0 || isTempDb );

  if( isMemdb ){
    flags |= BTREE_MEMORY;
  }
  if( (vfsFlags & SQLITE_OPEN_MAIN_DB)!=0 && (isMemdb || isTempDb) ){
    vfsFlags = (vfsFlags & ~SQLITE_OPEN_MAIN_DB) | SQLITE_OPEN_TEMP_DB;
  }
  p = sqlite3MallocZero(sizeof(Btree));
  if( !p ){
    return SQLITE_NOMEM_BKPT;
  }
  p->inTrans = TRANS_NONE;
  p->db = db;
#ifndef SQLITE_OMIT_SHARED_CACHE
  p->lock.pBtree = p;
  p->lock.iTable = 1;
#endif

#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO)
  /*
  ** If this Btree is a candidate for shared cache, try to find an
  ** existing BtShared object that we can share with
  */
  if( isTempDb==0 && (isMemdb==0 || (vfsFlags&SQLITE_OPEN_URI)!=0) ){
    if( vfsFlags & SQLITE_OPEN_SHAREDCACHE ){
      int nFilename = sqlite3Strlen30(zFilename)+1;
      int nFullPathname = pVfs->mxPathname+1;
      char *zFullPathname = sqlite3Malloc(MAX(nFullPathname,nFilename));
      MUTEX_LOGIC( sqlite3_mutex *mutexShared; )

      p->sharable = 1;
      if( !zFullPathname ){
        sqlite3_free(p);
        return SQLITE_NOMEM_BKPT;
      }
      if( isMemdb ){
        memcpy(zFullPathname, zFilename, nFilename);
      }else{
        rc = sqlite3OsFullPathname(pVfs, zFilename,
                                   nFullPathname, zFullPathname);
        if( rc ){
          if( rc==SQLITE_OK_SYMLINK ){
            rc = SQLITE_OK;
          }else{
            sqlite3_free(zFullPathname);
            sqlite3_free(p);
            return rc;
          }
        }
      }
#if SQLITE_THREADSAFE
      mutexOpen = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_OPEN);
      sqlite3_mutex_enter(mutexOpen);
      mutexShared = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MAIN);
      sqlite3_mutex_enter(mutexShared);
#endif
      for(pBt=GLOBAL(BtShared*,sqlite3SharedCacheList); pBt; pBt=pBt->pNext){
        assert( pBt->nRef>0 );
        if( 0==strcmp(zFullPathname, sqlite3PagerFilename(pBt->pPager, 0))
                 && sqlite3PagerVfs(pBt->pPager)==pVfs ){
          int iDb;
          for(iDb=db->nDb-1; iDb>=0; iDb--){
            Btree *pExisting = db->aDb[iDb].pBt;
            if( pExisting && pExisting->pBt==pBt ){
              sqlite3_mutex_leave(mutexShared);
              sqlite3_mutex_leave(mutexOpen);
              sqlite3_free(zFullPathname);
              sqlite3_free(p);
              return SQLITE_CONSTRAINT;
            }
          }
          p->pBt = pBt;
          pBt->nRef++;
          break;
        }
      }
      sqlite3_mutex_leave(mutexShared);
      sqlite3_free(zFullPathname);
    }
#ifdef SQLITE_DEBUG
    else{
      /* In debug mode, we mark all persistent databases as sharable
      ** even when they are not.  This exercises the locking code and
      ** gives more opportunity for asserts(sqlite3_mutex_held())
      ** statements to find locking problems.
      */
      p->sharable = 1;
    }
#endif
  }
#endif
  if( pBt==0 ){
    /*
    ** The following asserts make sure that structures used by the btree are
    ** the right size.  This is to guard against size changes that result
    ** when compiling on a different architecture.
    */
    assert( sizeof(i64)==8 );
    assert( sizeof(u64)==8 );
    assert( sizeof(u32)==4 );
    assert( sizeof(u16)==2 );
    assert( sizeof(Pgno)==4 );

    /* Suppress false-positive compiler warning from PVS-Studio */
    memset(&zDbHeader[16], 0, 8);
 
    pBt = sqlite3MallocZero( sizeof(*pBt) );
    if( pBt==0 ){
      rc = SQLITE_NOMEM_BKPT;
      goto btree_open_out;
    }
    rc = sqlite3PagerOpen(pVfs, &pBt->pPager, zFilename,
                          sizeof(MemPage), flags, vfsFlags, pageReinit);
    if( rc==SQLITE_OK ){
      sqlite3PagerSetMmapLimit(pBt->pPager, db->szMmap);
      rc = sqlite3PagerReadFileheader(pBt->pPager,sizeof(zDbHeader),zDbHeader);
    }
    if( rc!=SQLITE_OK ){
      goto btree_open_out;
    }
    pBt->openFlags = (u8)flags;
    pBt->db = db;
    sqlite3PagerSetBusyHandler(pBt->pPager, btreeInvokeBusyHandler, pBt);
    p->pBt = pBt;
 
    pBt->pCursor = 0;
    pBt->pPage1 = 0;
    if( sqlite3PagerIsreadonly(pBt->pPager) ) pBt->btsFlags |= BTS_READ_ONLY;
#if defined(SQLITE_SECURE_DELETE)
    pBt->btsFlags |= BTS_SECURE_DELETE;
#elif defined(SQLITE_FAST_SECURE_DELETE)
    pBt->btsFlags |= BTS_OVERWRITE;
#endif
    /* EVIDENCE-OF: R-51873-39618 The page size for a database file is
    ** determined by the 2-byte integer located at an offset of 16 bytes from
    ** the beginning of the database file. */
    pBt->pageSize = (zDbHeader[16]<<8) | (zDbHeader[17]<<16);
    if( pBt->pageSize<512 || pBt->pageSize>SQLITE_MAX_PAGE_SIZE
         || ((pBt->pageSize-1)&pBt->pageSize)!=0 ){
      pBt->pageSize = 0;
#ifndef SQLITE_OMIT_AUTOVACUUM
      /* If the magic name ":memory:" will create an in-memory database, then
      ** leave the autoVacuum mode at 0 (do not auto-vacuum), even if
      ** SQLITE_DEFAULT_AUTOVACUUM is true. On the other hand, if
      ** SQLITE_OMIT_MEMORYDB has been defined, then ":memory:" is just a
      ** regular file-name. In this case the auto-vacuum applies as per normal.
      */
      if( zFilename && !isMemdb ){
        pBt->autoVacuum = (SQLITE_DEFAULT_AUTOVACUUM ? 1 : 0);
        pBt->incrVacuum = (SQLITE_DEFAULT_AUTOVACUUM==2 ? 1 : 0);
      }
#endif
      nReserve = 0;
    }else{
      /* EVIDENCE-OF: R-37497-42412 The size of the reserved region is
      ** determined by the one-byte unsigned integer found at an offset of 20
      ** into the database file header. */
      nReserve = zDbHeader[20];
      pBt->btsFlags |= BTS_PAGESIZE_FIXED;
#ifndef SQLITE_OMIT_AUTOVACUUM
      pBt->autoVacuum = (get4byte(&zDbHeader[36 + 4*4])?1:0);
      pBt->incrVacuum = (get4byte(&zDbHeader[36 + 7*4])?1:0);
#endif
    }
    rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, nReserve);
    if( rc ) goto btree_open_out;
    pBt->usableSize = pBt->pageSize - nReserve;
    assert( (pBt->pageSize & 7)==0 );  /* 8-byte alignment of pageSize */
  
#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO)
    /* Add the new BtShared object to the linked list sharable BtShareds.
    */
    pBt->nRef = 1;
    if( p->sharable ){
      MUTEX_LOGIC( sqlite3_mutex *mutexShared; )
      MUTEX_LOGIC( mutexShared = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MAIN);)
      if( SQLITE_THREADSAFE && sqlite3GlobalConfig.bCoreMutex ){
        pBt->mutex = sqlite3MutexAlloc(SQLITE_MUTEX_FAST);
        if( pBt->mutex==0 ){
          rc = SQLITE_NOMEM_BKPT;
          goto btree_open_out;
        }
      }
      sqlite3_mutex_enter(mutexShared);
      pBt->pNext = GLOBAL(BtShared*,sqlite3SharedCacheList);
      GLOBAL(BtShared*,sqlite3SharedCacheList) = pBt;
      sqlite3_mutex_leave(mutexShared);
    }
#endif
  }

#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO)
  /* If the new Btree uses a sharable pBtShared, then link the new
  ** Btree into the list of all sharable Btrees for the same connection.
  ** The list is kept in ascending order by pBt address.
  */
  if( p->sharable ){
    int i;
    Btree *pSib;
    for(i=0; i<db->nDb; i++){
      if( (pSib = db->aDb[i].pBt)!=0 && pSib->sharable ){
        while( pSib->pPrev ){ pSib = pSib->pPrev; }
        if( (uptr)p->pBt<(uptr)pSib->pBt ){
          p->pNext = pSib;
          p->pPrev = 0;
          pSib->pPrev = p;
        }else{
          while( pSib->pNext && (uptr)pSib->pNext->pBt<(uptr)p->pBt ){
            pSib = pSib->pNext;
          }
          p->pNext = pSib->pNext;
          p->pPrev = pSib;
          if( p->pNext ){
            p->pNext->pPrev = p;
          }
          pSib->pNext = p;
        }
        break;
      }
    }
  }
#endif
  *ppBtree = p;

btree_open_out:
  if( rc!=SQLITE_OK ){
    if( pBt && pBt->pPager ){
      sqlite3PagerClose(pBt->pPager, 0);
    }
    sqlite3_free(pBt);
    sqlite3_free(p);
    *ppBtree = 0;
  }else{
    sqlite3_file *pFile;

    /* If the B-Tree was successfully opened, set the pager-cache size to the
    ** default value. Except, when opening on an existing shared pager-cache,
    ** do not change the pager-cache size.
    */
    if( sqlite3BtreeSchema(p, 0, 0)==0 ){
      sqlite3BtreeSetCacheSize(p, SQLITE_DEFAULT_CACHE_SIZE);
    }

    pFile = sqlite3PagerFile(pBt->pPager);
    if( pFile->pMethods ){
      sqlite3OsFileControlHint(pFile, SQLITE_FCNTL_PDB, (void*)&pBt->db);
    }
  }
  if( mutexOpen ){
    assert( sqlite3_mutex_held(mutexOpen) );
    sqlite3_mutex_leave(mutexOpen);
  }
  assert( rc!=SQLITE_OK || sqlite3BtreeConnectionCount(*ppBtree)>0 );
  return rc;
}

/*
** Decrement the BtShared.nRef counter.  When it reaches zero,
** remove the BtShared structure from the sharing list.  Return
** true if the BtShared.nRef counter reaches zero and return
** false if it is still positive.
*/
static int removeFromSharingList(BtShared *pBt){
#ifndef SQLITE_OMIT_SHARED_CACHE
  MUTEX_LOGIC( sqlite3_mutex *pMainMtx; )
  BtShared *pList;
  int removed = 0;

  assert( sqlite3_mutex_notheld(pBt->mutex) );
  MUTEX_LOGIC( pMainMtx = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MAIN); )
  sqlite3_mutex_enter(pMainMtx);
  pBt->nRef--;
  if( pBt->nRef<=0 ){
    if( GLOBAL(BtShared*,sqlite3SharedCacheList)==pBt ){
      GLOBAL(BtShared*,sqlite3SharedCacheList) = pBt->pNext;
    }else{
      pList = GLOBAL(BtShared*,sqlite3SharedCacheList);
      while( ALWAYS(pList) && pList->pNext!=pBt ){
        pList=pList->pNext;
      }
      if( ALWAYS(pList) ){
        pList->pNext = pBt->pNext;
      }
    }
    if( SQLITE_THREADSAFE ){
      sqlite3_mutex_free(pBt->mutex);
    }
    removed = 1;
  }
  sqlite3_mutex_leave(pMainMtx);
  return removed;
#else
  return 1;
#endif
}

/*
** Make sure pBt->pTmpSpace points to an allocation of
** MX_CELL_SIZE(pBt) bytes with a 4-byte prefix for a left-child
** pointer.
*/
static SQLITE_NOINLINE int allocateTempSpace(BtShared *pBt){
  assert( pBt!=0 );
  assert( pBt->pTmpSpace==0 );
  /* This routine is called only by btreeCursor() when allocating the
  ** first write cursor for the BtShared object */
  assert( pBt->pCursor!=0 && (pBt->pCursor->curFlags & BTCF_WriteFlag)!=0 );
  pBt->pTmpSpace = sqlite3PageMalloc( pBt->pageSize );
  if( pBt->pTmpSpace==0 ){
    BtCursor *pCur = pBt->pCursor;
    pBt->pCursor = pCur->pNext;  /* Unlink the cursor */
    memset(pCur, 0, sizeof(*pCur));
    return SQLITE_NOMEM_BKPT;
  }

  /* One of the uses of pBt->pTmpSpace is to format cells before
  ** inserting them into a leaf page (function fillInCell()). If
  ** a cell is less than 4 bytes in size, it is rounded up to 4 bytes
  ** by the various routines that manipulate binary cells. Which
  ** can mean that fillInCell() only initializes the first 2 or 3
  ** bytes of pTmpSpace, but that the first 4 bytes are copied from
  ** it into a database page. This is not actually a problem, but it
  ** does cause a valgrind error when the 1 or 2 bytes of uninitialized
  ** data is passed to system call write(). So to avoid this error,
  ** zero the first 4 bytes of temp space here.
  **
  ** Also:  Provide four bytes of initialized space before the
  ** beginning of pTmpSpace as an area available to prepend the
  ** left-child pointer to the beginning of a cell.
  */
  memset(pBt->pTmpSpace, 0, 8);
  pBt->pTmpSpace += 4;
  return SQLITE_OK;
}

/*
** Free the pBt->pTmpSpace allocation
*/
static void freeTempSpace(BtShared *pBt){
  if( pBt->pTmpSpace ){
    pBt->pTmpSpace -= 4;
    sqlite3PageFree(pBt->pTmpSpace);
    pBt->pTmpSpace = 0;
  }
}

/*
** Close an open database and invalidate all cursors.
*/
int sqlite3BtreeClose(Btree *p){
  BtShared *pBt = p->pBt;

  /* Close all cursors opened via this handle.  */
  assert( sqlite3_mutex_held(p->db->mutex) );
  sqlite3BtreeEnter(p);

  /* Verify that no other cursors have this Btree open */
#ifdef SQLITE_DEBUG
  {
    BtCursor *pCur = pBt->pCursor;
    while( pCur ){
      BtCursor *pTmp = pCur;
      pCur = pCur->pNext;
      assert( pTmp->pBtree!=p );

    }
  }
#endif

  /* Rollback any active transaction and free the handle structure.
  ** The call to sqlite3BtreeRollback() drops any table-locks held by
  ** this handle.
  */
  sqlite3BtreeRollback(p, SQLITE_OK, 0);
  sqlite3BtreeLeave(p);

  /* If there are still other outstanding references to the shared-btree
  ** structure, return now. The remainder of this procedure cleans
  ** up the shared-btree.
  */
  assert( p->wantToLock==0 && p->locked==0 );
  if( !p->sharable || removeFromSharingList(pBt) ){
    /* The pBt is no longer on the sharing list, so we can access
    ** it without having to hold the mutex.
    **
    ** Clean out and delete the BtShared object.
    */
    assert( !pBt->pCursor );
    sqlite3PagerClose(pBt->pPager, p->db);
    if( pBt->xFreeSchema && pBt->pSchema ){
      pBt->xFreeSchema(pBt->pSchema);
    }
    sqlite3DbFree(0, pBt->pSchema);
    freeTempSpace(pBt);
    sqlite3_free(pBt);
  }

#ifndef SQLITE_OMIT_SHARED_CACHE
  assert( p->wantToLock==0 );
  assert( p->locked==0 );
  if( p->pPrev ) p->pPrev->pNext = p->pNext;
  if( p->pNext ) p->pNext->pPrev = p->pPrev;
#endif

  sqlite3_free(p);
  return SQLITE_OK;
}

/*
** Change the "soft" limit on the number of pages in the cache.
** Unused and unmodified pages will be recycled when the number of
** pages in the cache exceeds this soft limit.  But the size of the
** cache is allowed to grow larger than this limit if it contains
** dirty pages or pages still in active use.
*/
int sqlite3BtreeSetCacheSize(Btree *p, int mxPage){
  BtShared *pBt = p->pBt;
  assert( sqlite3_mutex_held(p->db->mutex) );
  sqlite3BtreeEnter(p);
  sqlite3PagerSetCachesize(pBt->pPager, mxPage);
  sqlite3BtreeLeave(p);
  return SQLITE_OK;
}

/*
** Change the "spill" limit on the number of pages in the cache.
** If the number of pages exceeds this limit during a write transaction,
** the pager might attempt to "spill" pages to the journal early in
** order to free up memory.
**
** The value returned is the current spill size.  If zero is passed
** as an argument, no changes are made to the spill size setting, so
** using mxPage of 0 is a way to query the current spill size.
*/
int sqlite3BtreeSetSpillSize(Btree *p, int mxPage){
  BtShared *pBt = p->pBt;
  int res;
  assert( sqlite3_mutex_held(p->db->mutex) );
  sqlite3BtreeEnter(p);
  res = sqlite3PagerSetSpillsize(pBt->pPager, mxPage);
  sqlite3BtreeLeave(p);
  return res;
}

#if SQLITE_MAX_MMAP_SIZE>0
/*
** Change the limit on the amount of the database file that may be
** memory mapped.
*/
int sqlite3BtreeSetMmapLimit(Btree *p, sqlite3_int64 szMmap){
  BtShared *pBt = p->pBt;
  assert( sqlite3_mutex_held(p->db->mutex) );
  sqlite3BtreeEnter(p);
  sqlite3PagerSetMmapLimit(pBt->pPager, szMmap);
  sqlite3BtreeLeave(p);
  return SQLITE_OK;
}
#endif /* SQLITE_MAX_MMAP_SIZE>0 */

/*
** Change the way data is synced to disk in order to increase or decrease
** how well the database resists damage due to OS crashes and power
** failures.  Level 1 is the same as asynchronous (no syncs() occur and
** there is a high probability of damage)  Level 2 is the default.  There
** is a very low but non-zero probability of damage.  Level 3 reduces the
** probability of damage to near zero but with a write performance reduction.
*/
#ifndef SQLITE_OMIT_PAGER_PRAGMAS
int sqlite3BtreeSetPagerFlags(
  Btree *p,              /* The btree to set the safety level on */
  unsigned pgFlags       /* Various PAGER_* flags */
){
  BtShared *pBt = p->pBt;
  assert( sqlite3_mutex_held(p->db->mutex) );
  sqlite3BtreeEnter(p);
  sqlite3PagerSetFlags(pBt->pPager, pgFlags);
  sqlite3BtreeLeave(p);
  return SQLITE_OK;
}
#endif

/*
** Change the default pages size and the number of reserved bytes per page.
** Or, if the page size has already been fixed, return SQLITE_READONLY
** without changing anything.
**
** The page size must be a power of 2 between 512 and 65536.  If the page
** size supplied does not meet this constraint then the page size is not
** changed.
**
** Page sizes are constrained to be a power of two so that the region
** of the database file used for locking (beginning at PENDING_BYTE,
** the first byte past the 1GB boundary, 0x40000000) needs to occur
** at the beginning of a page.
**
** If parameter nReserve is less than zero, then the number of reserved
** bytes per page is left unchanged.
**
** If the iFix!=0 then the BTS_PAGESIZE_FIXED flag is set so that the page size
** and autovacuum mode can no longer be changed.
*/
int sqlite3BtreeSetPageSize(Btree *p, int pageSize, int nReserve, int iFix){
  int rc = SQLITE_OK;
  int x;
  BtShared *pBt = p->pBt;
  assert( nReserve>=0 && nReserve<=255 );
  sqlite3BtreeEnter(p);
  pBt->nReserveWanted = nReserve;
  x = pBt->pageSize - pBt->usableSize;
  if( nReserve<x ) nReserve = x;
  if( pBt->btsFlags & BTS_PAGESIZE_FIXED ){
    sqlite3BtreeLeave(p);
    return SQLITE_READONLY;
  }
  assert( nReserve>=0 && nReserve<=255 );
  if( pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE &&
        ((pageSize-1)&pageSize)==0 ){
    assert( (pageSize & 7)==0 );
    assert( !pBt->pCursor );
    if( nReserve>32 && pageSize==512 ) pageSize = 1024;
    pBt->pageSize = (u32)pageSize;
    freeTempSpace(pBt);
  }
  rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, nReserve);
  pBt->usableSize = pBt->pageSize - (u16)nReserve;
  if( iFix ) pBt->btsFlags |= BTS_PAGESIZE_FIXED;
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** Return the currently defined page size
*/
int sqlite3BtreeGetPageSize(Btree *p){
  return p->pBt->pageSize;
}

/*
** This function is similar to sqlite3BtreeGetReserve(), except that it
** may only be called if it is guaranteed that the b-tree mutex is already
** held.
**
** This is useful in one special case in the backup API code where it is
** known that the shared b-tree mutex is held, but the mutex on the
** database handle that owns *p is not. In this case if sqlite3BtreeEnter()
** were to be called, it might collide with some other operation on the
** database handle that owns *p, causing undefined behavior.
*/
int sqlite3BtreeGetReserveNoMutex(Btree *p){
  int n;
  assert( sqlite3_mutex_held(p->pBt->mutex) );
  n = p->pBt->pageSize - p->pBt->usableSize;
  return n;
}

/*
** Return the number of bytes of space at the end of every page that
** are intentionally left unused.  This is the "reserved" space that is
** sometimes used by extensions.
**
** The value returned is the larger of the current reserve size and
** the latest reserve size requested by SQLITE_FILECTRL_RESERVE_BYTES.
** The amount of reserve can only grow - never shrink.
*/
int sqlite3BtreeGetRequestedReserve(Btree *p){
  int n1, n2;
  sqlite3BtreeEnter(p);
  n1 = (int)p->pBt->nReserveWanted;
  n2 = sqlite3BtreeGetReserveNoMutex(p);
  sqlite3BtreeLeave(p);
  return n1>n2 ? n1 : n2;
}


/*
** Set the maximum page count for a database if mxPage is positive.
** No changes are made if mxPage is 0 or negative.
** Regardless of the value of mxPage, return the maximum page count.
*/
Pgno sqlite3BtreeMaxPageCount(Btree *p, Pgno mxPage){
  Pgno n;
  sqlite3BtreeEnter(p);
  n = sqlite3PagerMaxPageCount(p->pBt->pPager, mxPage);
  sqlite3BtreeLeave(p);
  return n;
}

/*
** Change the values for the BTS_SECURE_DELETE and BTS_OVERWRITE flags:
**
**    newFlag==0       Both BTS_SECURE_DELETE and BTS_OVERWRITE are cleared
**    newFlag==1       BTS_SECURE_DELETE set and BTS_OVERWRITE is cleared
**    newFlag==2       BTS_SECURE_DELETE cleared and BTS_OVERWRITE is set
**    newFlag==(-1)    No changes
**
** This routine acts as a query if newFlag is less than zero
**
** With BTS_OVERWRITE set, deleted content is overwritten by zeros, but
** freelist leaf pages are not written back to the database.  Thus in-page
** deleted content is cleared, but freelist deleted content is not.
**
** With BTS_SECURE_DELETE, operation is like BTS_OVERWRITE with the addition
** that freelist leaf pages are written back into the database, increasing
** the amount of disk I/O.
*/
int sqlite3BtreeSecureDelete(Btree *p, int newFlag){
  int b;
  if( p==0 ) return 0;
  sqlite3BtreeEnter(p);
  assert( BTS_OVERWRITE==BTS_SECURE_DELETE*2 );
  assert( BTS_FAST_SECURE==(BTS_OVERWRITE|BTS_SECURE_DELETE) );
  if( newFlag>=0 ){
    p->pBt->btsFlags &= ~BTS_FAST_SECURE;
    p->pBt->btsFlags |= BTS_SECURE_DELETE*newFlag;
  }
  b = (p->pBt->btsFlags & BTS_FAST_SECURE)/BTS_SECURE_DELETE;
  sqlite3BtreeLeave(p);
  return b;
}

/*
** Change the 'auto-vacuum' property of the database. If the 'autoVacuum'
** parameter is non-zero, then auto-vacuum mode is enabled. If zero, it
** is disabled. The default value for the auto-vacuum property is
** determined by the SQLITE_DEFAULT_AUTOVACUUM macro.
*/
int sqlite3BtreeSetAutoVacuum(Btree *p, int autoVacuum){
#ifdef SQLITE_OMIT_AUTOVACUUM
  return SQLITE_READONLY;
#else
  BtShared *pBt = p->pBt;
  int rc = SQLITE_OK;
  u8 av = (u8)autoVacuum;

  sqlite3BtreeEnter(p);
  if( (pBt->btsFlags & BTS_PAGESIZE_FIXED)!=0 && (av ?1:0)!=pBt->autoVacuum ){
    rc = SQLITE_READONLY;
  }else{
    pBt->autoVacuum = av ?1:0;
    pBt->incrVacuum = av==2 ?1:0;
  }
  sqlite3BtreeLeave(p);
  return rc;
#endif
}

/*
** Return the value of the 'auto-vacuum' property. If auto-vacuum is
** enabled 1 is returned. Otherwise 0.
*/
int sqlite3BtreeGetAutoVacuum(Btree *p){
#ifdef SQLITE_OMIT_AUTOVACUUM
  return BTREE_AUTOVACUUM_NONE;
#else
  int rc;
  sqlite3BtreeEnter(p);
  rc = (
    (!p->pBt->autoVacuum)?BTREE_AUTOVACUUM_NONE:
    (!p->pBt->incrVacuum)?BTREE_AUTOVACUUM_FULL:
    BTREE_AUTOVACUUM_INCR
  );
  sqlite3BtreeLeave(p);
  return rc;
#endif
}

/*
** If the user has not set the safety-level for this database connection
** using "PRAGMA synchronous", and if the safety-level is not already
** set to the value passed to this function as the second parameter,
** set it so.
*/
#if SQLITE_DEFAULT_SYNCHRONOUS!=SQLITE_DEFAULT_WAL_SYNCHRONOUS \
    && !defined(SQLITE_OMIT_WAL)
static void setDefaultSyncFlag(BtShared *pBt, u8 safety_level){
  sqlite3 *db;
  Db *pDb;
  if( (db=pBt->db)!=0 && (pDb=db->aDb)!=0 ){
    while( pDb->pBt==0 || pDb->pBt->pBt!=pBt ){ pDb++; }
    if( pDb->bSyncSet==0
     && pDb->safety_level!=safety_level
     && pDb!=&db->aDb[1]
    ){
      pDb->safety_level = safety_level;
      sqlite3PagerSetFlags(pBt->pPager,
          pDb->safety_level | (db->flags & PAGER_FLAGS_MASK));
    }
  }
}
#else
# define setDefaultSyncFlag(pBt,safety_level)
#endif

/* Forward declaration */
static int newDatabase(BtShared*);


/*
** Get a reference to pPage1 of the database file.  This will
** also acquire a readlock on that file.
**
** SQLITE_OK is returned on success.  If the file is not a
** well-formed database file, then SQLITE_CORRUPT is returned.
** SQLITE_BUSY is returned if the database is locked.  SQLITE_NOMEM
** is returned if we run out of memory.
*/
static int lockBtree(BtShared *pBt){
  int rc;              /* Result code from subfunctions */
  MemPage *pPage1;     /* Page 1 of the database file */
  u32 nPage;           /* Number of pages in the database */
  u32 nPageFile = 0;   /* Number of pages in the database file */

  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( pBt->pPage1==0 );
  rc = sqlite3PagerSharedLock(pBt->pPager);
  if( rc!=SQLITE_OK ) return rc;
  rc = btreeGetPage(pBt, 1, &pPage1, 0);
  if( rc!=SQLITE_OK ) return rc;

  /* Do some checking to help insure the file we opened really is
  ** a valid database file.
  */
  nPage = get4byte(28+(u8*)pPage1->aData);
  sqlite3PagerPagecount(pBt->pPager, (int*)&nPageFile);
  if( nPage==0 || memcmp(24+(u8*)pPage1->aData, 92+(u8*)pPage1->aData,4)!=0 ){
    nPage = nPageFile;
  }
  if( (pBt->db->flags & SQLITE_ResetDatabase)!=0 ){
    nPage = 0;
  }
  if( nPage>0 ){
    u32 pageSize;
    u32 usableSize;
    u8 *page1 = pPage1->aData;
    rc = SQLITE_NOTADB;
    /* EVIDENCE-OF: R-43737-39999 Every valid SQLite database file begins
    ** with the following 16 bytes (in hex): 53 51 4c 69 74 65 20 66 6f 72 6d
    ** 61 74 20 33 00. */
    if( memcmp(page1, zMagicHeader, 16)!=0 ){
      goto page1_init_failed;
    }

#ifdef SQLITE_OMIT_WAL
    if( page1[18]>1 ){
      pBt->btsFlags |= BTS_READ_ONLY;
    }
    if( page1[19]>1 ){
      goto page1_init_failed;
    }
#else
    if( page1[18]>2 ){
      pBt->btsFlags |= BTS_READ_ONLY;
    }
    if( page1[19]>2 ){
      goto page1_init_failed;
    }

    /* If the read version is set to 2, this database should be accessed
    ** in WAL mode. If the log is not already open, open it now. Then
    ** return SQLITE_OK and return without populating BtShared.pPage1.
    ** The caller detects this and calls this function again. This is
    ** required as the version of page 1 currently in the page1 buffer
    ** may not be the latest version - there may be a newer one in the log
    ** file.
    */
    if( page1[19]==2 && (pBt->btsFlags & BTS_NO_WAL)==0 ){
      int isOpen = 0;
      rc = sqlite3PagerOpenWal(pBt->pPager, &isOpen);
      if( rc!=SQLITE_OK ){
        goto page1_init_failed;
      }else{
        setDefaultSyncFlag(pBt, SQLITE_DEFAULT_WAL_SYNCHRONOUS+1);
        if( isOpen==0 ){
          releasePageOne(pPage1);
          return SQLITE_OK;
        }
      }
      rc = SQLITE_NOTADB;
    }else{
      setDefaultSyncFlag(pBt, SQLITE_DEFAULT_SYNCHRONOUS+1);
    }
#endif

    /* EVIDENCE-OF: R-15465-20813 The maximum and minimum embedded payload
    ** fractions and the leaf payload fraction values must be 64, 32, and 32.
    **
    ** The original design allowed these amounts to vary, but as of
    ** version 3.6.0, we require them to be fixed.
    */
    if( memcmp(&page1[21], "\100\040\040",3)!=0 ){
      goto page1_init_failed;
    }
    /* EVIDENCE-OF: R-51873-39618 The page size for a database file is
    ** determined by the 2-byte integer located at an offset of 16 bytes from
    ** the beginning of the database file. */
    pageSize = (page1[16]<<8) | (page1[17]<<16);
    /* EVIDENCE-OF: R-25008-21688 The size of a page is a power of two
    ** between 512 and 65536 inclusive. */
    if( ((pageSize-1)&pageSize)!=0
     || pageSize>SQLITE_MAX_PAGE_SIZE
     || pageSize<=256
    ){
      goto page1_init_failed;
    }
    assert( (pageSize & 7)==0 );
    /* EVIDENCE-OF: R-59310-51205 The "reserved space" size in the 1-byte
    ** integer at offset 20 is the number of bytes of space at the end of
    ** each page to reserve for extensions.
    **
    ** EVIDENCE-OF: R-37497-42412 The size of the reserved region is
    ** determined by the one-byte unsigned integer found at an offset of 20
    ** into the database file header. */
    usableSize = pageSize - page1[20];
    if( (u32)pageSize!=pBt->pageSize ){
      /* After reading the first page of the database assuming a page size
      ** of BtShared.pageSize, we have discovered that the page-size is
      ** actually pageSize. Unlock the database, leave pBt->pPage1 at
      ** zero and return SQLITE_OK. The caller will call this function
      ** again with the correct page-size.
      */
      releasePageOne(pPage1);
      pBt->usableSize = usableSize;
      pBt->pageSize = pageSize;
      pBt->btsFlags |= BTS_PAGESIZE_FIXED;
      freeTempSpace(pBt);
      rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize,
                                   pageSize-usableSize);
      return rc;
    }
    if( nPage>nPageFile ){
      if( sqlite3WritableSchema(pBt->db)==0 ){
        rc = SQLITE_CORRUPT_BKPT;
        goto page1_init_failed;
      }else{
        nPage = nPageFile;
      }
    }
    /* EVIDENCE-OF: R-28312-64704 However, the usable size is not allowed to
    ** be less than 480. In other words, if the page size is 512, then the
    ** reserved space size cannot exceed 32. */
    if( usableSize<480 ){
      goto page1_init_failed;
    }
    pBt->btsFlags |= BTS_PAGESIZE_FIXED;
    pBt->pageSize = pageSize;
    pBt->usableSize = usableSize;
#ifndef SQLITE_OMIT_AUTOVACUUM
    pBt->autoVacuum = (get4byte(&page1[36 + 4*4])?1:0);
    pBt->incrVacuum = (get4byte(&page1[36 + 7*4])?1:0);
#endif
  }

  /* maxLocal is the maximum amount of payload to store locally for
  ** a cell.  Make sure it is small enough so that at least minFanout
  ** cells can will fit on one page.  We assume a 10-byte page header.
  ** Besides the payload, the cell must store:
  **     2-byte pointer to the cell
  **     4-byte child pointer
  **     9-byte nKey value
  **     4-byte nData value
  **     4-byte overflow page pointer
  ** So a cell consists of a 2-byte pointer, a header which is as much as
  ** 17 bytes long, 0 to N bytes of payload, and an optional 4 byte overflow
  ** page pointer.
  */
  pBt->maxLocal = (u16)((pBt->usableSize-12)*64/255 - 23);
  pBt->minLocal = (u16)((pBt->usableSize-12)*32/255 - 23);
  pBt->maxLeaf = (u16)(pBt->usableSize - 35);
  pBt->minLeaf = (u16)((pBt->usableSize-12)*32/255 - 23);
  if( pBt->maxLocal>127 ){
    pBt->max1bytePayload = 127;
  }else{
    pBt->max1bytePayload = (u8)pBt->maxLocal;
  }
  assert( pBt->maxLeaf + 23 <= MX_CELL_SIZE(pBt) );
  pBt->pPage1 = pPage1;
  pBt->nPage = nPage;
  return SQLITE_OK;

page1_init_failed:
  releasePageOne(pPage1);
  pBt->pPage1 = 0;
  return rc;
}

#ifndef NDEBUG
/*
** Return the number of cursors open on pBt. This is for use
** in assert() expressions, so it is only compiled if NDEBUG is not
** defined.
**
** Only write cursors are counted if wrOnly is true.  If wrOnly is
** false then all cursors are counted.
**
** For the purposes of this routine, a cursor is any cursor that
** is capable of reading or writing to the database.  Cursors that
** have been tripped into the CURSOR_FAULT state are not counted.
*/
static int countValidCursors(BtShared *pBt, int wrOnly){
  BtCursor *pCur;
  int r = 0;
  for(pCur=pBt->pCursor; pCur; pCur=pCur->pNext){
    if( (wrOnly==0 || (pCur->curFlags & BTCF_WriteFlag)!=0)
     && pCur->eState!=CURSOR_FAULT ) r++;
  }
  return r;
}
#endif

/*
** If there are no outstanding cursors and we are not in the middle
** of a transaction but there is a read lock on the database, then
** this routine unrefs the first page of the database file which
** has the effect of releasing the read lock.
**
** If there is a transaction in progress, this routine is a no-op.
*/
static void unlockBtreeIfUnused(BtShared *pBt){
  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( countValidCursors(pBt,0)==0 || pBt->inTransaction>TRANS_NONE );
  if( pBt->inTransaction==TRANS_NONE && pBt->pPage1!=0 ){
    MemPage *pPage1 = pBt->pPage1;
    assert( pPage1->aData );
    assert( sqlite3PagerRefcount(pBt->pPager)==1 );
    pBt->pPage1 = 0;
    releasePageOne(pPage1);
  }
}

/*
** If pBt points to an empty file then convert that empty file
** into a new empty database by initializing the first page of
** the database.
*/
static int newDatabase(BtShared *pBt){
  MemPage *pP1;
  unsigned char *data;
  int rc;

  assert( sqlite3_mutex_held(pBt->mutex) );
  if( pBt->nPage>0 ){
    return SQLITE_OK;
  }
  pP1 = pBt->pPage1;
  assert( pP1!=0 );
  data = pP1->aData;
  rc = sqlite3PagerWrite(pP1->pDbPage);
  if( rc ) return rc;
  memcpy(data, zMagicHeader, sizeof(zMagicHeader));
  assert( sizeof(zMagicHeader)==16 );
  data[16] = (u8)((pBt->pageSize>>8)&0xff);
  data[17] = (u8)((pBt->pageSize>>16)&0xff);
  data[18] = 1;
  data[19] = 1;
  assert( pBt->usableSize<=pBt->pageSize && pBt->usableSize+255>=pBt->pageSize);
  data[20] = (u8)(pBt->pageSize - pBt->usableSize);
  data[21] = 64;
  data[22] = 32;
  data[23] = 32;
  memset(&data[24], 0, 100-24);
  zeroPage(pP1, PTF_INTKEY|PTF_LEAF|PTF_LEAFDATA );
  pBt->btsFlags |= BTS_PAGESIZE_FIXED;
#ifndef SQLITE_OMIT_AUTOVACUUM
  assert( pBt->autoVacuum==1 || pBt->autoVacuum==0 );
  assert( pBt->incrVacuum==1 || pBt->incrVacuum==0 );
  put4byte(&data[36 + 4*4], pBt->autoVacuum);
  put4byte(&data[36 + 7*4], pBt->incrVacuum);
#endif
  pBt->nPage = 1;
  data[31] = 1;
  return SQLITE_OK;
}

/*
** Initialize the first page of the database file (creating a database
** consisting of a single page and no schema objects). Return SQLITE_OK
** if successful, or an SQLite error code otherwise.
*/
int sqlite3BtreeNewDb(Btree *p){
  int rc;
  sqlite3BtreeEnter(p);
  p->pBt->nPage = 0;
  rc = newDatabase(p->pBt);
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** Attempt to start a new transaction. A write-transaction
** is started if the second argument is nonzero, otherwise a read-
** transaction.  If the second argument is 2 or more and exclusive
** transaction is started, meaning that no other process is allowed
** to access the database.  A preexisting transaction may not be
** upgraded to exclusive by calling this routine a second time - the
** exclusivity flag only works for a new transaction.
**
** A write-transaction must be started before attempting any
** changes to the database.  None of the following routines
** will work unless a transaction is started first:
**
**      sqlite3BtreeCreateTable()
**      sqlite3BtreeCreateIndex()
**      sqlite3BtreeClearTable()
**      sqlite3BtreeDropTable()
**      sqlite3BtreeInsert()
**      sqlite3BtreeDelete()
**      sqlite3BtreeUpdateMeta()
**
** If an initial attempt to acquire the lock fails because of lock contention
** and the database was previously unlocked, then invoke the busy handler
** if there is one.  But if there was previously a read-lock, do not
** invoke the busy handler - just return SQLITE_BUSY.  SQLITE_BUSY is
** returned when there is already a read-lock in order to avoid a deadlock.
**
** Suppose there are two processes A and B.  A has a read lock and B has
** a reserved lock.  B tries to promote to exclusive but is blocked because
** of A's read lock.  A tries to promote to reserved but is blocked by B.
** One or the other of the two processes must give way or there can be
** no progress.  By returning SQLITE_BUSY and not invoking the busy callback
** when A already has a read lock, we encourage A to give up and let B
** proceed.
*/
static SQLITE_NOINLINE int btreeBeginTrans(
  Btree *p,                 /* The btree in which to start the transaction */
  int wrflag,               /* True to start a write transaction */
  int *pSchemaVersion       /* Put schema version number here, if not NULL */
){
  BtShared *pBt = p->pBt;
  Pager *pPager = pBt->pPager;
  int rc = SQLITE_OK;

  sqlite3BtreeEnter(p);
  btreeIntegrity(p);

  /* If the btree is already in a write-transaction, or it
  ** is already in a read-transaction and a read-transaction
  ** is requested, this is a no-op.
  */
  if( p->inTrans==TRANS_WRITE || (p->inTrans==TRANS_READ && !wrflag) ){
    goto trans_begun;
  }
  assert( pBt->inTransaction==TRANS_WRITE || IfNotOmitAV(pBt->bDoTruncate)==0 );

  if( (p->db->flags & SQLITE_ResetDatabase)
   && sqlite3PagerIsreadonly(pPager)==0
  ){
    pBt->btsFlags &= ~BTS_READ_ONLY;
  }

  /* Write transactions are not possible on a read-only database */
  if( (pBt->btsFlags & BTS_READ_ONLY)!=0 && wrflag ){
    rc = SQLITE_READONLY;
    goto trans_begun;
  }

#ifndef SQLITE_OMIT_SHARED_CACHE
  {
    sqlite3 *pBlock = 0;
    /* If another database handle has already opened a write transaction
    ** on this shared-btree structure and a second write transaction is
    ** requested, return SQLITE_LOCKED.
    */
    if( (wrflag && pBt->inTransaction==TRANS_WRITE)
     || (pBt->btsFlags & BTS_PENDING)!=0
    ){
      pBlock = pBt->pWriter->db;
    }else if( wrflag>1 ){
      BtLock *pIter;
      for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
        if( pIter->pBtree!=p ){
          pBlock = pIter->pBtree->db;
          break;
        }
      }
    }
    if( pBlock ){
      sqlite3ConnectionBlocked(p->db, pBlock);
      rc = SQLITE_LOCKED_SHAREDCACHE;
      goto trans_begun;
    }
  }
#endif

  /* Any read-only or read-write transaction implies a read-lock on
  ** page 1. So if some other shared-cache client already has a write-lock
  ** on page 1, the transaction cannot be opened. */
  rc = querySharedCacheTableLock(p, SCHEMA_ROOT, READ_LOCK);
  if( SQLITE_OK!=rc ) goto trans_begun;

  pBt->btsFlags &= ~BTS_INITIALLY_EMPTY;
  if( pBt->nPage==0 ) pBt->btsFlags |= BTS_INITIALLY_EMPTY;
  do {
    sqlite3PagerWalDb(pPager, p->db);

#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
    /* If transitioning from no transaction directly to a write transaction,
    ** block for the WRITER lock first if possible. */
    if( pBt->pPage1==0 && wrflag ){
      assert( pBt->inTransaction==TRANS_NONE );
      rc = sqlite3PagerWalWriteLock(pPager, 1);
      if( rc!=SQLITE_BUSY && rc!=SQLITE_OK ) break;
    }
#endif

    /* Call lockBtree() until either pBt->pPage1 is populated or
    ** lockBtree() returns something other than SQLITE_OK. lockBtree()
    ** may return SQLITE_OK but leave pBt->pPage1 set to 0 if after
    ** reading page 1 it discovers that the page-size of the database
    ** file is not pBt->pageSize. In this case lockBtree() will update
    ** pBt->pageSize to the page-size of the file on disk.
    */
    while( pBt->pPage1==0 && SQLITE_OK==(rc = lockBtree(pBt)) );

    if( rc==SQLITE_OK && wrflag ){
      if( (pBt->btsFlags & BTS_READ_ONLY)!=0 ){
        rc = SQLITE_READONLY;
      }else{
        rc = sqlite3PagerBegin(pPager, wrflag>1, sqlite3TempInMemory(p->db));
        if( rc==SQLITE_OK ){
          rc = newDatabase(pBt);
        }else if( rc==SQLITE_BUSY_SNAPSHOT && pBt->inTransaction==TRANS_NONE ){
          /* if there was no transaction opened when this function was
          ** called and SQLITE_BUSY_SNAPSHOT is returned, change the error
          ** code to SQLITE_BUSY. */
          rc = SQLITE_BUSY;
        }
      }
    }
 
    if( rc!=SQLITE_OK ){
      (void)sqlite3PagerWalWriteLock(pPager, 0);
      unlockBtreeIfUnused(pBt);
    }
  }while( (rc&0xFF)==SQLITE_BUSY && pBt->inTransaction==TRANS_NONE &&
          btreeInvokeBusyHandler(pBt) );
  sqlite3PagerWalDb(pPager, 0);
#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
  if( rc==SQLITE_BUSY_TIMEOUT ) rc = SQLITE_BUSY;
#endif

  if( rc==SQLITE_OK ){
    if( p->inTrans==TRANS_NONE ){
      pBt->nTransaction++;
#ifndef SQLITE_OMIT_SHARED_CACHE
      if( p->sharable ){
        assert( p->lock.pBtree==p && p->lock.iTable==1 );
        p->lock.eLock = READ_LOCK;
        p->lock.pNext = pBt->pLock;
        pBt->pLock = &p->lock;
      }
#endif
    }
    p->inTrans = (wrflag?TRANS_WRITE:TRANS_READ);
    if( p->inTrans>pBt->inTransaction ){
      pBt->inTransaction = p->inTrans;
    }
    if( wrflag ){
      MemPage *pPage1 = pBt->pPage1;
#ifndef SQLITE_OMIT_SHARED_CACHE
      assert( !pBt->pWriter );
      pBt->pWriter = p;
      pBt->btsFlags &= ~BTS_EXCLUSIVE;
      if( wrflag>1 ) pBt->btsFlags |= BTS_EXCLUSIVE;
#endif

      /* If the db-size header field is incorrect (as it may be if an old
      ** client has been writing the database file), update it now. Doing
      ** this sooner rather than later means the database size can safely
      ** re-read the database size from page 1 if a savepoint or transaction
      ** rollback occurs within the transaction.
      */
      if( pBt->nPage!=get4byte(&pPage1->aData[28]) ){
        rc = sqlite3PagerWrite(pPage1->pDbPage);
        if( rc==SQLITE_OK ){
          put4byte(&pPage1->aData[28], pBt->nPage);
        }
      }
    }
  }

trans_begun:
  if( rc==SQLITE_OK ){
    if( pSchemaVersion ){
      *pSchemaVersion = get4byte(&pBt->pPage1->aData[40]);
    }
    if( wrflag ){
      /* This call makes sure that the pager has the correct number of
      ** open savepoints. If the second parameter is greater than 0 and
      ** the sub-journal is not already open, then it will be opened here.
      */
      rc = sqlite3PagerOpenSavepoint(pPager, p->db->nSavepoint);
    }
  }

  btreeIntegrity(p);
  sqlite3BtreeLeave(p);
  return rc;
}
int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){
  BtShared *pBt;
  if( p->sharable
   || p->inTrans==TRANS_NONE
   || (p->inTrans==TRANS_READ && wrflag!=0)
  ){
    return btreeBeginTrans(p,wrflag,pSchemaVersion);
  }
  pBt = p->pBt;
  if( pSchemaVersion ){
    *pSchemaVersion = get4byte(&pBt->pPage1->aData[40]);
  }
  if( wrflag ){
    /* This call makes sure that the pager has the correct number of
    ** open savepoints. If the second parameter is greater than 0 and
    ** the sub-journal is not already open, then it will be opened here.
    */
    return sqlite3PagerOpenSavepoint(pBt->pPager, p->db->nSavepoint);
  }else{
    return SQLITE_OK;
  }
}

#ifndef SQLITE_OMIT_AUTOVACUUM

/*
** Set the pointer-map entries for all children of page pPage. Also, if
** pPage contains cells that point to overflow pages, set the pointer
** map entries for the overflow pages as well.
*/
static int setChildPtrmaps(MemPage *pPage){
  int i;                             /* Counter variable */
  int nCell;                         /* Number of cells in page pPage */
  int rc;                            /* Return code */
  BtShared *pBt = pPage->pBt;
  Pgno pgno = pPage->pgno;

  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  rc = pPage->isInit ? SQLITE_OK : btreeInitPage(pPage);
  if( rc!=SQLITE_OK ) return rc;
  nCell = pPage->nCell;

  for(i=0; i<nCell; i++){
    u8 *pCell = findCell(pPage, i);

    ptrmapPutOvflPtr(pPage, pPage, pCell, &rc);

    if( !pPage->leaf ){
      Pgno childPgno = get4byte(pCell);
      ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno, &rc);
    }
  }

  if( !pPage->leaf ){
    Pgno childPgno = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno, &rc);
  }

  return rc;
}

/*
** Somewhere on pPage is a pointer to page iFrom.  Modify this pointer so
** that it points to iTo. Parameter eType describes the type of pointer to
** be modified, as  follows:
**
** PTRMAP_BTREE:     pPage is a btree-page. The pointer points at a child
**                   page of pPage.
**
** PTRMAP_OVERFLOW1: pPage is a btree-page. The pointer points at an overflow
**                   page pointed to by one of the cells on pPage.
**
** PTRMAP_OVERFLOW2: pPage is an overflow-page. The pointer points at the next
**                   overflow page in the list.
*/
static int modifyPagePointer(MemPage *pPage, Pgno iFrom, Pgno iTo, u8 eType){
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  if( eType==PTRMAP_OVERFLOW2 ){
    /* The pointer is always the first 4 bytes of the page in this case.  */
    if( get4byte(pPage->aData)!=iFrom ){
      return SQLITE_CORRUPT_PAGE(pPage);
    }
    put4byte(pPage->aData, iTo);
  }else{
    int i;
    int nCell;
    int rc;

    rc = pPage->isInit ? SQLITE_OK : btreeInitPage(pPage);
    if( rc ) return rc;
    nCell = pPage->nCell;

    for(i=0; i<nCell; i++){
      u8 *pCell = findCell(pPage, i);
      if( eType==PTRMAP_OVERFLOW1 ){
        CellInfo info;
        pPage->xParseCell(pPage, pCell, &info);
        if( info.nLocal<info.nPayload ){
          if( pCell+info.nSize > pPage->aData+pPage->pBt->usableSize ){
            return SQLITE_CORRUPT_PAGE(pPage);
          }
          if( iFrom==get4byte(pCell+info.nSize-4) ){
            put4byte(pCell+info.nSize-4, iTo);
            break;
          }
        }
      }else{
        if( pCell+4 > pPage->aData+pPage->pBt->usableSize ){
          return SQLITE_CORRUPT_PAGE(pPage);
        }
        if( get4byte(pCell)==iFrom ){
          put4byte(pCell, iTo);
          break;
        }
      }
    }
 
    if( i==nCell ){
      if( eType!=PTRMAP_BTREE ||
          get4byte(&pPage->aData[pPage->hdrOffset+8])!=iFrom ){
        return SQLITE_CORRUPT_PAGE(pPage);
      }
      put4byte(&pPage->aData[pPage->hdrOffset+8], iTo);
    }
  }
  return SQLITE_OK;
}


/*
** Move the open database page pDbPage to location iFreePage in the
** database. The pDbPage reference remains valid.
**
** The isCommit flag indicates that there is no need to remember that
** the journal needs to be sync()ed before database page pDbPage->pgno
** can be written to. The caller has already promised not to write to that
** page.
*/
static int relocatePage(
  BtShared *pBt,           /* Btree */
  MemPage *pDbPage,        /* Open page to move */
  u8 eType,                /* Pointer map 'type' entry for pDbPage */
  Pgno iPtrPage,           /* Pointer map 'page-no' entry for pDbPage */
  Pgno iFreePage,          /* The location to move pDbPage to */
  int isCommit             /* isCommit flag passed to sqlite3PagerMovepage */
){
  MemPage *pPtrPage;   /* The page that contains a pointer to pDbPage */
  Pgno iDbPage = pDbPage->pgno;
  Pager *pPager = pBt->pPager;
  int rc;

  assert( eType==PTRMAP_OVERFLOW2 || eType==PTRMAP_OVERFLOW1 ||
      eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE );
  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( pDbPage->pBt==pBt );
  if( iDbPage<3 ) return SQLITE_CORRUPT_BKPT;

  /* Move page iDbPage from its current location to page number iFreePage */
  TRACE(("AUTOVACUUM: Moving %u to free page %u (ptr page %u type %u)\n",
      iDbPage, iFreePage, iPtrPage, eType));
  rc = sqlite3PagerMovepage(pPager, pDbPage->pDbPage, iFreePage, isCommit);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  pDbPage->pgno = iFreePage;

  /* If pDbPage was a btree-page, then it may have child pages and/or cells
  ** that point to overflow pages. The pointer map entries for all these
  ** pages need to be changed.
  **
  ** If pDbPage is an overflow page, then the first 4 bytes may store a
  ** pointer to a subsequent overflow page. If this is the case, then
  ** the pointer map needs to be updated for the subsequent overflow page.
  */
  if( eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ){
    rc = setChildPtrmaps(pDbPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
  }else{
    Pgno nextOvfl = get4byte(pDbPage->aData);
    if( nextOvfl!=0 ){
      ptrmapPut(pBt, nextOvfl, PTRMAP_OVERFLOW2, iFreePage, &rc);
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
  }

  /* Fix the database pointer on page iPtrPage that pointed at iDbPage so
  ** that it points at iFreePage. Also fix the pointer map entry for
  ** iPtrPage.
  */
  if( eType!=PTRMAP_ROOTPAGE ){
    rc = btreeGetPage(pBt, iPtrPage, &pPtrPage, 0);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    rc = sqlite3PagerWrite(pPtrPage->pDbPage);
    if( rc!=SQLITE_OK ){
      releasePage(pPtrPage);
      return rc;
    }
    rc = modifyPagePointer(pPtrPage, iDbPage, iFreePage, eType);
    releasePage(pPtrPage);
    if( rc==SQLITE_OK ){
      ptrmapPut(pBt, iFreePage, eType, iPtrPage, &rc);
    }
  }
  return rc;
}

/* Forward declaration required by incrVacuumStep(). */
static int allocateBtreePage(BtShared *, MemPage **, Pgno *, Pgno, u8);

/*
** Perform a single step of an incremental-vacuum. If successful, return
** SQLITE_OK. If there is no work to do (and therefore no point in
** calling this function again), return SQLITE_DONE. Or, if an error
** occurs, return some other error code.
**
** More specifically, this function attempts to re-organize the database so
** that the last page of the file currently in use is no longer in use.
**
** Parameter nFin is the number of pages that this database would contain
** were this function called until it returns SQLITE_DONE.
**
** If the bCommit parameter is non-zero, this function assumes that the
** caller will keep calling incrVacuumStep() until it returns SQLITE_DONE
** or an error. bCommit is passed true for an auto-vacuum-on-commit
** operation, or false for an incremental vacuum.
*/
static int incrVacuumStep(BtShared *pBt, Pgno nFin, Pgno iLastPg, int bCommit){
  Pgno nFreeList;           /* Number of pages still on the free-list */
  int rc;

  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( iLastPg>nFin );

  if( !PTRMAP_ISPAGE(pBt, iLastPg) && iLastPg!=PENDING_BYTE_PAGE(pBt) ){
    u8 eType;
    Pgno iPtrPage;

    nFreeList = get4byte(&pBt->pPage1->aData[36]);
    if( nFreeList==0 ){
      return SQLITE_DONE;
    }

    rc = ptrmapGet(pBt, iLastPg, &eType, &iPtrPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( eType==PTRMAP_ROOTPAGE ){
      return SQLITE_CORRUPT_BKPT;
    }

    if( eType==PTRMAP_FREEPAGE ){
      if( bCommit==0 ){
        /* Remove the page from the files free-list. This is not required
        ** if bCommit is non-zero. In that case, the free-list will be
        ** truncated to zero after this function returns, so it doesn't
        ** matter if it still contains some garbage entries.
        */
        Pgno iFreePg;
        MemPage *pFreePg;
        rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iLastPg, BTALLOC_EXACT);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        assert( iFreePg==iLastPg );
        releasePage(pFreePg);
      }
    } else {
      Pgno iFreePg;             /* Index of free page to move pLastPg to */
      MemPage *pLastPg;
      u8 eMode = BTALLOC_ANY;   /* Mode parameter for allocateBtreePage() */
      Pgno iNear = 0;           /* nearby parameter for allocateBtreePage() */

      rc = btreeGetPage(pBt, iLastPg, &pLastPg, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }

      /* If bCommit is zero, this loop runs exactly once and page pLastPg
      ** is swapped with the first free page pulled off the free list.
      **
      ** On the other hand, if bCommit is greater than zero, then keep
      ** looping until a free-page located within the first nFin pages
      ** of the file is found.
      */
      if( bCommit==0 ){
        eMode = BTALLOC_LE;
        iNear = nFin;
      }
      do {
        MemPage *pFreePg;
        Pgno dbSize = btreePagecount(pBt);
        rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iNear, eMode);
        if( rc!=SQLITE_OK ){
          releasePage(pLastPg);
          return rc;
        }
        releasePage(pFreePg);
        if( iFreePg>dbSize ){
          releasePage(pLastPg);
          return SQLITE_CORRUPT_BKPT;
        }
      }while( bCommit && iFreePg>nFin );
      assert( iFreePg<iLastPg );
     
      rc = relocatePage(pBt, pLastPg, eType, iPtrPage, iFreePg, bCommit);
      releasePage(pLastPg);
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
  }

  if( bCommit==0 ){
    do {
      iLastPg--;
    }while( iLastPg==PENDING_BYTE_PAGE(pBt) || PTRMAP_ISPAGE(pBt, iLastPg) );
    pBt->bDoTruncate = 1;
    pBt->nPage = iLastPg;
  }
  return SQLITE_OK;
}

/*
** The database opened by the first argument is an auto-vacuum database
** nOrig pages in size containing nFree free pages. Return the expected
** size of the database in pages following an auto-vacuum operation.
*/
static Pgno finalDbSize(BtShared *pBt, Pgno nOrig, Pgno nFree){
  int nEntry;                     /* Number of entries on one ptrmap page */
  Pgno nPtrmap;                   /* Number of PtrMap pages to be freed */
  Pgno nFin;                      /* Return value */

  nEntry = pBt->usableSize/5;
  nPtrmap = (nFree-nOrig+PTRMAP_PAGENO(pBt, nOrig)+nEntry)/nEntry;
  nFin = nOrig - nFree - nPtrmap;
  if( nOrig>PENDING_BYTE_PAGE(pBt) && nFin<PENDING_BYTE_PAGE(pBt) ){
    nFin--;
  }
  while( PTRMAP_ISPAGE(pBt, nFin) || nFin==PENDING_BYTE_PAGE(pBt) ){
    nFin--;
  }

  return nFin;
}

/*
** A write-transaction must be opened before calling this function.
** It performs a single unit of work towards an incremental vacuum.
**
** If the incremental vacuum is finished after this function has run,
** SQLITE_DONE is returned. If it is not finished, but no error occurred,
** SQLITE_OK is returned. Otherwise an SQLite error code.
*/
int sqlite3BtreeIncrVacuum(Btree *p){
  int rc;
  BtShared *pBt = p->pBt;

  sqlite3BtreeEnter(p);
  assert( pBt->inTransaction==TRANS_WRITE && p->inTrans==TRANS_WRITE );
  if( !pBt->autoVacuum ){
    rc = SQLITE_DONE;
  }else{
    Pgno nOrig = btreePagecount(pBt);
    Pgno nFree = get4byte(&pBt->pPage1->aData[36]);
    Pgno nFin = finalDbSize(pBt, nOrig, nFree);

    if( nOrig<nFin || nFree>=nOrig ){
      rc = SQLITE_CORRUPT_BKPT;
    }else if( nFree>0 ){
      rc = saveAllCursors(pBt, 0, 0);
      if( rc==SQLITE_OK ){
        invalidateAllOverflowCache(pBt);
        rc = incrVacuumStep(pBt, nFin, nOrig, 0);
      }
      if( rc==SQLITE_OK ){
        rc = sqlite3PagerWrite(pBt->pPage1->pDbPage);
        put4byte(&pBt->pPage1->aData[28], pBt->nPage);
      }
    }else{
      rc = SQLITE_DONE;
    }
  }
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** This routine is called prior to sqlite3PagerCommit when a transaction
** is committed for an auto-vacuum database.
*/
static int autoVacuumCommit(Btree *p){
  int rc = SQLITE_OK;
  Pager *pPager;
  BtShared *pBt;
  sqlite3 *db;
  VVA_ONLY( int nRef );

  assert( p!=0 );
  pBt = p->pBt; 
  pPager = pBt->pPager;
  VVA_ONLY( nRef = sqlite3PagerRefcount(pPager); )

  assert( sqlite3_mutex_held(pBt->mutex) );
  invalidateAllOverflowCache(pBt);
  assert(pBt->autoVacuum);
  if( !pBt->incrVacuum ){
    Pgno nFin;         /* Number of pages in database after autovacuuming */
    Pgno nFree;        /* Number of pages on the freelist initially */
    Pgno nVac;         /* Number of pages to vacuum */
    Pgno iFree;        /* The next page to be freed */
    Pgno nOrig;        /* Database size before freeing */

    nOrig = btreePagecount(pBt);
    if( PTRMAP_ISPAGE(pBt, nOrig) || nOrig==PENDING_BYTE_PAGE(pBt) ){
      /* It is not possible to create a database for which the final page
      ** is either a pointer-map page or the pending-byte page. If one
      ** is encountered, this indicates corruption.
      */
      return SQLITE_CORRUPT_BKPT;
    }

    nFree = get4byte(&pBt->pPage1->aData[36]);
    db = p->db;
    if( db->xAutovacPages ){
      int iDb;
      for(iDb=0; ALWAYS(iDb<db->nDb); iDb++){
        if( db->aDb[iDb].pBt==p ) break;
      }
      nVac = db->xAutovacPages(
        db->pAutovacPagesArg,
        db->aDb[iDb].zDbSName,
        nOrig,
        nFree,
        pBt->pageSize
      );
      if( nVac>nFree ){
        nVac = nFree;
      }
      if( nVac==0 ){
        return SQLITE_OK;
      }
    }else{
      nVac = nFree;
    }
    nFin = finalDbSize(pBt, nOrig, nVac);
    if( nFin>nOrig ) return SQLITE_CORRUPT_BKPT;
    if( nFin<nOrig ){
      rc = saveAllCursors(pBt, 0, 0);
    }
    for(iFree=nOrig; iFree>nFin && rc==SQLITE_OK; iFree--){
      rc = incrVacuumStep(pBt, nFin, iFree, nVac==nFree);
    }
    if( (rc==SQLITE_DONE || rc==SQLITE_OK) && nFree>0 ){
      rc = sqlite3PagerWrite(pBt->pPage1->pDbPage);
      if( nVac==nFree ){
        put4byte(&pBt->pPage1->aData[32], 0);
        put4byte(&pBt->pPage1->aData[36], 0);
      }
      put4byte(&pBt->pPage1->aData[28], nFin);
      pBt->bDoTruncate = 1;
      pBt->nPage = nFin;
    }
    if( rc!=SQLITE_OK ){
      sqlite3PagerRollback(pPager);
    }
  }

  assert( nRef>=sqlite3PagerRefcount(pPager) );
  return rc;
}

#else /* ifndef SQLITE_OMIT_AUTOVACUUM */
# define setChildPtrmaps(x) SQLITE_OK
#endif

/*
** This routine does the first phase of a two-phase commit.  This routine
** causes a rollback journal to be created (if it does not already exist)
** and populated with enough information so that if a power loss occurs
** the database can be restored to its original state by playing back
** the journal.  Then the contents of the journal are flushed out to
** the disk.  After the journal is safely on oxide, the changes to the
** database are written into the database file and flushed to oxide.
** At the end of this call, the rollback journal still exists on the
** disk and we are still holding all locks, so the transaction has not
** committed.  See sqlite3BtreeCommitPhaseTwo() for the second phase of the
** commit process.
**
** This call is a no-op if no write-transaction is currently active on pBt.
**
** Otherwise, sync the database file for the btree pBt. zSuperJrnl points to
** the name of a super-journal file that should be written into the
** individual journal file, or is NULL, indicating no super-journal file
** (single database transaction).
**
** When this is called, the super-journal should already have been
** created, populated with this journal pointer and synced to disk.
**
** Once this is routine has returned, the only thing required to commit
** the write-transaction for this database file is to delete the journal.
*/
int sqlite3BtreeCommitPhaseOne(Btree *p, const char *zSuperJrnl){
  int rc = SQLITE_OK;
  if( p->inTrans==TRANS_WRITE ){
    BtShared *pBt = p->pBt;
    sqlite3BtreeEnter(p);
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum ){
      rc = autoVacuumCommit(p);
      if( rc!=SQLITE_OK ){
        sqlite3BtreeLeave(p);
        return rc;
      }
    }
    if( pBt->bDoTruncate ){
      sqlite3PagerTruncateImage(pBt->pPager, pBt->nPage);
    }
#endif
    rc = sqlite3PagerCommitPhaseOne(pBt->pPager, zSuperJrnl, 0);
    sqlite3BtreeLeave(p);
  }
  return rc;
}

/*
** This function is called from both BtreeCommitPhaseTwo() and BtreeRollback()
** at the conclusion of a transaction.
*/
static void btreeEndTransaction(Btree *p){
  BtShared *pBt = p->pBt;
  sqlite3 *db = p->db;
  assert( sqlite3BtreeHoldsMutex(p) );

#ifndef SQLITE_OMIT_AUTOVACUUM
  pBt->bDoTruncate = 0;
#endif
  if( p->inTrans>TRANS_NONE && db->nVdbeRead>1 ){
    /* If there are other active statements that belong to this database
    ** handle, downgrade to a read-only transaction. The other statements
    ** may still be reading from the database.  */
    downgradeAllSharedCacheTableLocks(p);
    p->inTrans = TRANS_READ;
  }else{
    /* If the handle had any kind of transaction open, decrement the
    ** transaction count of the shared btree. If the transaction count
    ** reaches 0, set the shared state to TRANS_NONE. The unlockBtreeIfUnused()
    ** call below will unlock the pager.  */
    if( p->inTrans!=TRANS_NONE ){
      clearAllSharedCacheTableLocks(p);
      pBt->nTransaction--;
      if( 0==pBt->nTransaction ){
        pBt->inTransaction = TRANS_NONE;
      }
    }

    /* Set the current transaction state to TRANS_NONE and unlock the
    ** pager if this call closed the only read or write transaction.  */
    p->inTrans = TRANS_NONE;
    unlockBtreeIfUnused(pBt);
  }

  btreeIntegrity(p);
}

/*
** Commit the transaction currently in progress.
**
** This routine implements the second phase of a 2-phase commit.  The
** sqlite3BtreeCommitPhaseOne() routine does the first phase and should
** be invoked prior to calling this routine.  The sqlite3BtreeCommitPhaseOne()
** routine did all the work of writing information out to disk and flushing the
** contents so that they are written onto the disk platter.  All this
** routine has to do is delete or truncate or zero the header in the
** the rollback journal (which causes the transaction to commit) and
** drop locks.
**
** Normally, if an error occurs while the pager layer is attempting to
** finalize the underlying journal file, this function returns an error and
** the upper layer will attempt a rollback. However, if the second argument
** is non-zero then this b-tree transaction is part of a multi-file
** transaction. In this case, the transaction has already been committed
** (by deleting a super-journal file) and the caller will ignore this
** functions return code. So, even if an error occurs in the pager layer,
** reset the b-tree objects internal state to indicate that the write
** transaction has been closed. This is quite safe, as the pager will have
** transitioned to the error state.
**
** This will release the write lock on the database file.  If there
** are no active cursors, it also releases the read lock.
*/
int sqlite3BtreeCommitPhaseTwo(Btree *p, int bCleanup){

  if( p->inTrans==TRANS_NONE ) return SQLITE_OK;
  sqlite3BtreeEnter(p);
  btreeIntegrity(p);

  /* If the handle has a write-transaction open, commit the shared-btrees
  ** transaction and set the shared state to TRANS_READ.
  */
  if( p->inTrans==TRANS_WRITE ){
    int rc;
    BtShared *pBt = p->pBt;
    assert( pBt->inTransaction==TRANS_WRITE );
    assert( pBt->nTransaction>0 );
    rc = sqlite3PagerCommitPhaseTwo(pBt->pPager);
    if( rc!=SQLITE_OK && bCleanup==0 ){
      sqlite3BtreeLeave(p);
      return rc;
    }
    p->iBDataVersion--;  /* Compensate for pPager->iDataVersion++; */
    pBt->inTransaction = TRANS_READ;
    btreeClearHasContent(pBt);
  }

  btreeEndTransaction(p);
  sqlite3BtreeLeave(p);
  return SQLITE_OK;
}

/*
** Do both phases of a commit.
*/
int sqlite3BtreeCommit(Btree *p){
  int rc;
  sqlite3BtreeEnter(p);
  rc = sqlite3BtreeCommitPhaseOne(p, 0);
  if( rc==SQLITE_OK ){
    rc = sqlite3BtreeCommitPhaseTwo(p, 0);
  }
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** This routine sets the state to CURSOR_FAULT and the error
** code to errCode for every cursor on any BtShared that pBtree
** references.  Or if the writeOnly flag is set to 1, then only
** trip write cursors and leave read cursors unchanged.
**
** Every cursor is a candidate to be tripped, including cursors
** that belong to other database connections that happen to be
** sharing the cache with pBtree.
**
** This routine gets called when a rollback occurs. If the writeOnly
** flag is true, then only write-cursors need be tripped - read-only
** cursors save their current positions so that they may continue
** following the rollback. Or, if writeOnly is false, all cursors are
** tripped. In general, writeOnly is false if the transaction being
** rolled back modified the database schema. In this case b-tree root
** pages may be moved or deleted from the database altogether, making
** it unsafe for read cursors to continue.
**
** If the writeOnly flag is true and an error is encountered while
** saving the current position of a read-only cursor, all cursors,
** including all read-cursors are tripped.
**
** SQLITE_OK is returned if successful, or if an error occurs while
** saving a cursor position, an SQLite error code.
*/
int sqlite3BtreeTripAllCursors(Btree *pBtree, int errCode, int writeOnly){
  BtCursor *p;
  int rc = SQLITE_OK;

  assert( (writeOnly==0 || writeOnly==1) && BTCF_WriteFlag==1 );
  if( pBtree ){
    sqlite3BtreeEnter(pBtree);
    for(p=pBtree->pBt->pCursor; p; p=p->pNext){
      if( writeOnly && (p->curFlags & BTCF_WriteFlag)==0 ){
        if( p->eState==CURSOR_VALID || p->eState==CURSOR_SKIPNEXT ){
          rc = saveCursorPosition(p);
          if( rc!=SQLITE_OK ){
            (void)sqlite3BtreeTripAllCursors(pBtree, rc, 0);
            break;
          }
        }
      }else{
        sqlite3BtreeClearCursor(p);
        p->eState = CURSOR_FAULT;
        p->skipNext = errCode;
      }
      btreeReleaseAllCursorPages(p);
    }
    sqlite3BtreeLeave(pBtree);
  }
  return rc;
}

/*
** Set the pBt->nPage field correctly, according to the current
** state of the database.  Assume pBt->pPage1 is valid.
*/
static void btreeSetNPage(BtShared *pBt, MemPage *pPage1){
  int nPage = get4byte(&pPage1->aData[28]);
  testcase( nPage==0 );
  if( nPage==0 ) sqlite3PagerPagecount(pBt->pPager, &nPage);
  testcase( pBt->nPage!=(u32)nPage );
  pBt->nPage = nPage;
}

/*
** Rollback the transaction in progress.
**
** If tripCode is not SQLITE_OK then cursors will be invalidated (tripped).
** Only write cursors are tripped if writeOnly is true but all cursors are
** tripped if writeOnly is false.  Any attempt to use
** a tripped cursor will result in an error.
**
** This will release the write lock on the database file.  If there
** are no active cursors, it also releases the read lock.
*/
int sqlite3BtreeRollback(Btree *p, int tripCode, int writeOnly){
  int rc;
  BtShared *pBt = p->pBt;
  MemPage *pPage1;

  assert( writeOnly==1 || writeOnly==0 );
  assert( tripCode==SQLITE_ABORT_ROLLBACK || tripCode==SQLITE_OK );
  sqlite3BtreeEnter(p);
  if( tripCode==SQLITE_OK ){
    rc = tripCode = saveAllCursors(pBt, 0, 0);
    if( rc ) writeOnly = 0;
  }else{
    rc = SQLITE_OK;
  }
  if( tripCode ){
    int rc2 = sqlite3BtreeTripAllCursors(p, tripCode, writeOnly);
    assert( rc==SQLITE_OK || (writeOnly==0 && rc2==SQLITE_OK) );
    if( rc2!=SQLITE_OK ) rc = rc2;
  }
  btreeIntegrity(p);

  if( p->inTrans==TRANS_WRITE ){
    int rc2;

    assert( TRANS_WRITE==pBt->inTransaction );
    rc2 = sqlite3PagerRollback(pBt->pPager);
    if( rc2!=SQLITE_OK ){
      rc = rc2;
    }

    /* The rollback may have destroyed the pPage1->aData value.  So
    ** call btreeGetPage() on page 1 again to make
    ** sure pPage1->aData is set correctly. */
    if( btreeGetPage(pBt, 1, &pPage1, 0)==SQLITE_OK ){
      btreeSetNPage(pBt, pPage1);
      releasePageOne(pPage1);
    }
    assert( countValidCursors(pBt, 1)==0 );
    pBt->inTransaction = TRANS_READ;
    btreeClearHasContent(pBt);
  }

  btreeEndTransaction(p);
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** Start a statement subtransaction. The subtransaction can be rolled
** back independently of the main transaction. You must start a transaction
** before starting a subtransaction. The subtransaction is ended automatically
** if the main transaction commits or rolls back.
**
** Statement subtransactions are used around individual SQL statements
** that are contained within a BEGIN...COMMIT block.  If a constraint
** error occurs within the statement, the effect of that one statement
** can be rolled back without having to rollback the entire transaction.
**
** A statement sub-transaction is implemented as an anonymous savepoint. The
** value passed as the second parameter is the total number of savepoints,
** including the new anonymous savepoint, open on the B-Tree. i.e. if there
** are no active savepoints and no other statement-transactions open,
** iStatement is 1. This anonymous savepoint can be released or rolled back
** using the sqlite3BtreeSavepoint() function.
*/
int sqlite3BtreeBeginStmt(Btree *p, int iStatement){
  int rc;
  BtShared *pBt = p->pBt;
  sqlite3BtreeEnter(p);
  assert( p->inTrans==TRANS_WRITE );
  assert( (pBt->btsFlags & BTS_READ_ONLY)==0 );
  assert( iStatement>0 );
  assert( iStatement>p->db->nSavepoint );
  assert( pBt->inTransaction==TRANS_WRITE );
  /* At the pager level, a statement transaction is a savepoint with
  ** an index greater than all savepoints created explicitly using
  ** SQL statements. It is illegal to open, release or rollback any
  ** such savepoints while the statement transaction savepoint is active.
  */
  rc = sqlite3PagerOpenSavepoint(pBt->pPager, iStatement);
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** The second argument to this function, op, is always SAVEPOINT_ROLLBACK
** or SAVEPOINT_RELEASE. This function either releases or rolls back the
** savepoint identified by parameter iSavepoint, depending on the value
** of op.
**
** Normally, iSavepoint is greater than or equal to zero. However, if op is
** SAVEPOINT_ROLLBACK, then iSavepoint may also be -1. In this case the
** contents of the entire transaction are rolled back. This is different
** from a normal transaction rollback, as no locks are released and the
** transaction remains open.
*/
int sqlite3BtreeSavepoint(Btree *p, int op, int iSavepoint){
  int rc = SQLITE_OK;
  if( p && p->inTrans==TRANS_WRITE ){
    BtShared *pBt = p->pBt;
    assert( op==SAVEPOINT_RELEASE || op==SAVEPOINT_ROLLBACK );
    assert( iSavepoint>=0 || (iSavepoint==-1 && op==SAVEPOINT_ROLLBACK) );
    sqlite3BtreeEnter(p);
    if( op==SAVEPOINT_ROLLBACK ){
      rc = saveAllCursors(pBt, 0, 0);
    }
    if( rc==SQLITE_OK ){
      rc = sqlite3PagerSavepoint(pBt->pPager, op, iSavepoint);
    }
    if( rc==SQLITE_OK ){
      if( iSavepoint<0 && (pBt->btsFlags & BTS_INITIALLY_EMPTY)!=0 ){
        pBt->nPage = 0;
      }
      rc = newDatabase(pBt);
      btreeSetNPage(pBt, pBt->pPage1);

      /* pBt->nPage might be zero if the database was corrupt when
      ** the transaction was started. Otherwise, it must be at least 1.  */
      assert( CORRUPT_DB || pBt->nPage>0 );
    }
    sqlite3BtreeLeave(p);
  }
  return rc;
}

/*
** Create a new cursor for the BTree whose root is on the page
** iTable. If a read-only cursor is requested, it is assumed that
** the caller already has at least a read-only transaction open
** on the database already. If a write-cursor is requested, then
** the caller is assumed to have an open write transaction.
**
** If the BTREE_WRCSR bit of wrFlag is clear, then the cursor can only
** be used for reading.  If the BTREE_WRCSR bit is set, then the cursor
** can be used for reading or for writing if other conditions for writing
** are also met.  These are the conditions that must be met in order
** for writing to be allowed:
**
** 1:  The cursor must have been opened with wrFlag containing BTREE_WRCSR
**
** 2:  Other database connections that share the same pager cache
**     but which are not in the READ_UNCOMMITTED state may not have
**     cursors open with wrFlag==0 on the same table.  Otherwise
**     the changes made by this write cursor would be visible to
**     the read cursors in the other database connection.
**
** 3:  The database must be writable (not on read-only media)
**
** 4:  There must be an active transaction.
**
** The BTREE_FORDELETE bit of wrFlag may optionally be set if BTREE_WRCSR
** is set.  If FORDELETE is set, that is a hint to the implementation that
** this cursor will only be used to seek to and delete entries of an index
** as part of a larger DELETE statement.  The FORDELETE hint is not used by
** this implementation.  But in a hypothetical alternative storage engine
** in which index entries are automatically deleted when corresponding table
** rows are deleted, the FORDELETE flag is a hint that all SEEK and DELETE
** operations on this cursor can be no-ops and all READ operations can
** return a null row (2-bytes: 0x01 0x00).
**
** No checking is done to make sure that page iTable really is the
** root page of a b-tree.  If it is not, then the cursor acquired
** will not work correctly.
**
** It is assumed that the sqlite3BtreeCursorZero() has been called
** on pCur to initialize the memory space prior to invoking this routine.
*/
static int btreeCursor(
  Btree *p,                              /* The btree */
  Pgno iTable,                           /* Root page of table to open */
  int wrFlag,                            /* 1 to write. 0 read-only */
  struct KeyInfo *pKeyInfo,              /* First arg to comparison function */
  BtCursor *pCur                         /* Space for new cursor */
){
  BtShared *pBt = p->pBt;                /* Shared b-tree handle */
  BtCursor *pX;                          /* Looping over other all cursors */

  assert( sqlite3BtreeHoldsMutex(p) );
  assert( wrFlag==0
       || wrFlag==BTREE_WRCSR
       || wrFlag==(BTREE_WRCSR|BTREE_FORDELETE)
  );

  /* The following assert statements verify that if this is a sharable
  ** b-tree database, the connection is holding the required table locks,
  ** and that no other connection has any open cursor that conflicts with
  ** this lock.  The iTable<1 term disables the check for corrupt schemas. */
  assert( hasSharedCacheTableLock(p, iTable, pKeyInfo!=0, (wrFlag?2:1))
          || iTable<1 );
  assert( wrFlag==0 || !hasReadConflicts(p, iTable) );

  /* Assert that the caller has opened the required transaction. */
  assert( p->inTrans>TRANS_NONE );
  assert( wrFlag==0 || p->inTrans==TRANS_WRITE );
  assert( pBt->pPage1 && pBt->pPage1->aData );
  assert( wrFlag==0 || (pBt->btsFlags & BTS_READ_ONLY)==0 );

  if( iTable<=1 ){
    if( iTable<1 ){
      return SQLITE_CORRUPT_BKPT;
    }else if( btreePagecount(pBt)==0 ){
      assert( wrFlag==0 );
      iTable = 0;
    }
  }

  /* Now that no other errors can occur, finish filling in the BtCursor
  ** variables and link the cursor into the BtShared list.  */
  pCur->pgnoRoot = iTable;
  pCur->iPage = -1;
  pCur->pKeyInfo = pKeyInfo;
  pCur->pBtree = p;
  pCur->pBt = pBt;
  pCur->curFlags = 0;
  /* If there are two or more cursors on the same btree, then all such
  ** cursors *must* have the BTCF_Multiple flag set. */
  for(pX=pBt->pCursor; pX; pX=pX->pNext){
    if( pX->pgnoRoot==iTable ){
      pX->curFlags |= BTCF_Multiple;
      pCur->curFlags = BTCF_Multiple;
    }
  }
  pCur->eState = CURSOR_INVALID;
  pCur->pNext = pBt->pCursor;
  pBt->pCursor = pCur;
  if( wrFlag ){
    pCur->curFlags |= BTCF_WriteFlag;
    pCur->curPagerFlags = 0;
    if( pBt->pTmpSpace==0 ) return allocateTempSpace(pBt);
  }else{
    pCur->curPagerFlags = PAGER_GET_READONLY;
  }
  return SQLITE_OK;
}
static int btreeCursorWithLock(
  Btree *p,                              /* The btree */
  Pgno iTable,                           /* Root page of table to open */
  int wrFlag,                            /* 1 to write. 0 read-only */
  struct KeyInfo *pKeyInfo,              /* First arg to comparison function */
  BtCursor *pCur                         /* Space for new cursor */
){
  int rc;
  sqlite3BtreeEnter(p);
  rc = btreeCursor(p, iTable, wrFlag, pKeyInfo, pCur);
  sqlite3BtreeLeave(p);
  return rc;
}
int sqlite3BtreeCursor(
  Btree *p,                                   /* The btree */
  Pgno iTable,                                /* Root page of table to open */
  int wrFlag,                                 /* 1 to write. 0 read-only */
  struct KeyInfo *pKeyInfo,                   /* First arg to xCompare() */
  BtCursor *pCur                              /* Write new cursor here */
){
  if( p->sharable ){
    return btreeCursorWithLock(p, iTable, wrFlag, pKeyInfo, pCur);
  }else{
    return btreeCursor(p, iTable, wrFlag, pKeyInfo, pCur);
  }
}

/*
** Return the size of a BtCursor object in bytes.
**
** This interfaces is needed so that users of cursors can preallocate
** sufficient storage to hold a cursor.  The BtCursor object is opaque
** to users so they cannot do the sizeof() themselves - they must call
** this routine.
*/
int sqlite3BtreeCursorSize(void){
  return ROUND8(sizeof(BtCursor));
}

/*
** Initialize memory that will be converted into a BtCursor object.
**
** The simple approach here would be to memset() the entire object
** to zero.  But it turns out that the apPage[] and aiIdx[] arrays
** do not need to be zeroed and they are large, so we can save a lot
** of run-time by skipping the initialization of those elements.
*/
void sqlite3BtreeCursorZero(BtCursor *p){
  memset(p, 0, offsetof(BtCursor, BTCURSOR_FIRST_UNINIT));
}

/*
** Close a cursor.  The read lock on the database file is released
** when the last cursor is closed.
*/
int sqlite3BtreeCloseCursor(BtCursor *pCur){
  Btree *pBtree = pCur->pBtree;
  if( pBtree ){
    BtShared *pBt = pCur->pBt;
    sqlite3BtreeEnter(pBtree);
    assert( pBt->pCursor!=0 );
    if( pBt->pCursor==pCur ){
      pBt->pCursor = pCur->pNext;
    }else{
      BtCursor *pPrev = pBt->pCursor;
      do{
        if( pPrev->pNext==pCur ){
          pPrev->pNext = pCur->pNext;
          break;
        }
        pPrev = pPrev->pNext;
      }while( ALWAYS(pPrev) );
    }
    btreeReleaseAllCursorPages(pCur);
    unlockBtreeIfUnused(pBt);
    sqlite3_free(pCur->aOverflow);
    sqlite3_free(pCur->pKey);
    if( (pBt->openFlags & BTREE_SINGLE) && pBt->pCursor==0 ){
      /* Since the BtShared is not sharable, there is no need to
      ** worry about the missing sqlite3BtreeLeave() call here.  */
      assert( pBtree->sharable==0 );
      sqlite3BtreeClose(pBtree);
    }else{
      sqlite3BtreeLeave(pBtree);
    }
    pCur->pBtree = 0;
  }
  return SQLITE_OK;
}

/*
** Make sure the BtCursor* given in the argument has a valid
** BtCursor.info structure.  If it is not already valid, call
** btreeParseCell() to fill it in.
**
** BtCursor.info is a cache of the information in the current cell.
** Using this cache reduces the number of calls to btreeParseCell().
*/
#ifndef NDEBUG
  static int cellInfoEqual(CellInfo *a, CellInfo *b){
    if( a->nKey!=b->nKey ) return 0;
    if( a->pPayload!=b->pPayload ) return 0;
    if( a->nPayload!=b->nPayload ) return 0;
    if( a->nLocal!=b->nLocal ) return 0;
    if( a->nSize!=b->nSize ) return 0;
    return 1;
  }
  static void assertCellInfo(BtCursor *pCur){
    CellInfo info;
    memset(&info, 0, sizeof(info));
    btreeParseCell(pCur->pPage, pCur->ix, &info);
    assert( CORRUPT_DB || cellInfoEqual(&info, &pCur->info) );
  }
#else
  #define assertCellInfo(x)
#endif
static SQLITE_NOINLINE void getCellInfo(BtCursor *pCur){
  if( pCur->info.nSize==0 ){
    pCur->curFlags |= BTCF_ValidNKey;
    btreeParseCell(pCur->pPage,pCur->ix,&pCur->info);
  }else{
    assertCellInfo(pCur);
  }
}

#ifndef NDEBUG  /* The next routine used only within assert() statements */
/*
** Return true if the given BtCursor is valid.  A valid cursor is one
** that is currently pointing to a row in a (non-empty) table.
** This is a verification routine is used only within assert() statements.
*/
int sqlite3BtreeCursorIsValid(BtCursor *pCur){
  return pCur && pCur->eState==CURSOR_VALID;
}
#endif /* NDEBUG */
int sqlite3BtreeCursorIsValidNN(BtCursor *pCur){
  assert( pCur!=0 );
  return pCur->eState==CURSOR_VALID;
}

/*
** Return the value of the integer key or "rowid" for a table btree.
** This routine is only valid for a cursor that is pointing into a
** ordinary table btree.  If the cursor points to an index btree or
** is invalid, the result of this routine is undefined.
*/
i64 sqlite3BtreeIntegerKey(BtCursor *pCur){
  assert( cursorHoldsMutex(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->curIntKey );
  getCellInfo(pCur);
  return pCur->info.nKey;
}

/*
** Pin or unpin a cursor.
*/
void sqlite3BtreeCursorPin(BtCursor *pCur){
  assert( (pCur->curFlags & BTCF_Pinned)==0 );
  pCur->curFlags |= BTCF_Pinned;
}
void sqlite3BtreeCursorUnpin(BtCursor *pCur){
  assert( (pCur->curFlags & BTCF_Pinned)!=0 );
  pCur->curFlags &= ~BTCF_Pinned;
}

/*
** Return the offset into the database file for the start of the
** payload to which the cursor is pointing.
*/
i64 sqlite3BtreeOffset(BtCursor *pCur){
  assert( cursorHoldsMutex(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  getCellInfo(pCur);
  return (i64)pCur->pBt->pageSize*((i64)pCur->pPage->pgno - 1) +
         (i64)(pCur->info.pPayload - pCur->pPage->aData);
}

/*
** Return the number of bytes of payload for the entry that pCur is
** currently pointing to.  For table btrees, this will be the amount
** of data.  For index btrees, this will be the size of the key.
**
** The caller must guarantee that the cursor is pointing to a non-NULL
** valid entry.  In other words, the calling procedure must guarantee
** that the cursor has Cursor.eState==CURSOR_VALID.
*/
u32 sqlite3BtreePayloadSize(BtCursor *pCur){
  assert( cursorHoldsMutex(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  getCellInfo(pCur);
  return pCur->info.nPayload;
}

/*
** Return an upper bound on the size of any record for the table
** that the cursor is pointing into.
**
** This is an optimization.  Everything will still work if this
** routine always returns 2147483647 (which is the largest record
** that SQLite can handle) or more.  But returning a smaller value might
** prevent large memory allocations when trying to interpret a
** corrupt database.
**
** The current implementation merely returns the size of the underlying
** database file.
*/
sqlite3_int64 sqlite3BtreeMaxRecordSize(BtCursor *pCur){
  assert( cursorHoldsMutex(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  return pCur->pBt->pageSize * (sqlite3_int64)pCur->pBt->nPage;
}

/*
** Given the page number of an overflow page in the database (parameter
** ovfl), this function finds the page number of the next page in the
** linked list of overflow pages. If possible, it uses the auto-vacuum
** pointer-map data instead of reading the content of page ovfl to do so.
**
** If an error occurs an SQLite error code is returned. Otherwise:
**
** The page number of the next overflow page in the linked list is
** written to *pPgnoNext. If page ovfl is the last page in its linked
** list, *pPgnoNext is set to zero.
**
** If ppPage is not NULL, and a reference to the MemPage object corresponding
** to page number pOvfl was obtained, then *ppPage is set to point to that
** reference. It is the responsibility of the caller to call releasePage()
** on *ppPage to free the reference. In no reference was obtained (because
** the pointer-map was used to obtain the value for *pPgnoNext), then
** *ppPage is set to zero.
*/
static int getOverflowPage(
  BtShared *pBt,               /* The database file */
  Pgno ovfl,                   /* Current overflow page number */
  MemPage **ppPage,            /* OUT: MemPage handle (may be NULL) */
  Pgno *pPgnoNext              /* OUT: Next overflow page number */
){
  Pgno next = 0;
  MemPage *pPage = 0;
  int rc = SQLITE_OK;

  assert( sqlite3_mutex_held(pBt->mutex) );
  assert(pPgnoNext);

#ifndef SQLITE_OMIT_AUTOVACUUM
  /* Try to find the next page in the overflow list using the
  ** autovacuum pointer-map pages. Guess that the next page in
  ** the overflow list is page number (ovfl+1). If that guess turns
  ** out to be wrong, fall back to loading the data of page
  ** number ovfl to determine the next page number.
  */
  if( pBt->autoVacuum ){
    Pgno pgno;
    Pgno iGuess = ovfl+1;
    u8 eType;

    while( PTRMAP_ISPAGE(pBt, iGuess) || iGuess==PENDING_BYTE_PAGE(pBt) ){
      iGuess++;
    }

    if( iGuess<=btreePagecount(pBt) ){
      rc = ptrmapGet(pBt, iGuess, &eType, &pgno);
      if( rc==SQLITE_OK && eType==PTRMAP_OVERFLOW2 && pgno==ovfl ){
        next = iGuess;
        rc = SQLITE_DONE;
      }
    }
  }
#endif

  assert( next==0 || rc==SQLITE_DONE );
  if( rc==SQLITE_OK ){
    rc = btreeGetPage(pBt, ovfl, &pPage, (ppPage==0) ? PAGER_GET_READONLY : 0);
    assert( rc==SQLITE_OK || pPage==0 );
    if( rc==SQLITE_OK ){
      next = get4byte(pPage->aData);
    }
  }

  *pPgnoNext = next;
  if( ppPage ){
    *ppPage = pPage;
  }else{
    releasePage(pPage);
  }
  return (rc==SQLITE_DONE ? SQLITE_OK : rc);
}

/*
** Copy data from a buffer to a page, or from a page to a buffer.
**
** pPayload is a pointer to data stored on database page pDbPage.
** If argument eOp is false, then nByte bytes of data are copied
** from pPayload to the buffer pointed at by pBuf. If eOp is true,
** then sqlite3PagerWrite() is called on pDbPage and nByte bytes
** of data are copied from the buffer pBuf to pPayload.
**
** SQLITE_OK is returned on success, otherwise an error code.
*/
static int copyPayload(
  void *pPayload,           /* Pointer to page data */
  void *pBuf,               /* Pointer to buffer */
  int nByte,                /* Number of bytes to copy */
  int eOp,                  /* 0 -> copy from page, 1 -> copy to page */
  DbPage *pDbPage           /* Page containing pPayload */
){
  if( eOp ){
    /* Copy data from buffer to page (a write operation) */
    int rc = sqlite3PagerWrite(pDbPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    memcpy(pPayload, pBuf, nByte);
  }else{
    /* Copy data from page to buffer (a read operation) */
    memcpy(pBuf, pPayload, nByte);
  }
  return SQLITE_OK;
}

/*
** This function is used to read or overwrite payload information
** for the entry that the pCur cursor is pointing to. The eOp
** argument is interpreted as follows:
**
**   0: The operation is a read. Populate the overflow cache.
**   1: The operation is a write. Populate the overflow cache.
**
** A total of "amt" bytes are read or written beginning at "offset".
** Data is read to or from the buffer pBuf.
**
** The content being read or written might appear on the main page
** or be scattered out on multiple overflow pages.
**
** If the current cursor entry uses one or more overflow pages
** this function may allocate space for and lazily populate
** the overflow page-list cache array (BtCursor.aOverflow).
** Subsequent calls use this cache to make seeking to the supplied offset
** more efficient.
**
** Once an overflow page-list cache has been allocated, it must be
** invalidated if some other cursor writes to the same table, or if
** the cursor is moved to a different row. Additionally, in auto-vacuum
** mode, the following events may invalidate an overflow page-list cache.
**
**   * An incremental vacuum,
**   * A commit in auto_vacuum="full" mode,
**   * Creating a table (may require moving an overflow page).
*/
static int accessPayload(
  BtCursor *pCur,      /* Cursor pointing to entry to read from */
  u32 offset,          /* Begin reading this far into payload */
  u32 amt,             /* Read this many bytes */
  unsigned char *pBuf, /* Write the bytes into this buffer */
  int eOp              /* zero to read. non-zero to write. */
){
  unsigned char *aPayload;
  int rc = SQLITE_OK;
  int iIdx = 0;
  MemPage *pPage = pCur->pPage;               /* Btree page of current entry */
  BtShared *pBt = pCur->pBt;                  /* Btree this cursor belongs to */
#ifdef SQLITE_DIRECT_OVERFLOW_READ
  unsigned char * const pBufStart = pBuf;     /* Start of original out buffer */
#endif

  assert( pPage );
  assert( eOp==0 || eOp==1 );
  assert( pCur->eState==CURSOR_VALID );
  if( pCur->ix>=pPage->nCell ){
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  assert( cursorHoldsMutex(pCur) );

  getCellInfo(pCur);
  aPayload = pCur->info.pPayload;
  assert( offset+amt <= pCur->info.nPayload );

  assert( aPayload > pPage->aData );
  if( (uptr)(aPayload - pPage->aData) > (pBt->usableSize - pCur->info.nLocal) ){
    /* Trying to read or write past the end of the data is an error.  The
    ** conditional above is really:
    **    &aPayload[pCur->info.nLocal] > &pPage->aData[pBt->usableSize]
    ** but is recast into its current form to avoid integer overflow problems
    */
    return SQLITE_CORRUPT_PAGE(pPage);
  }

  /* Check if data must be read/written to/from the btree page itself. */
  if( offset<pCur->info.nLocal ){
    int a = amt;
    if( a+offset>pCur->info.nLocal ){
      a = pCur->info.nLocal - offset;
    }
    rc = copyPayload(&aPayload[offset], pBuf, a, eOp, pPage->pDbPage);
    offset = 0;
    pBuf += a;
    amt -= a;
  }else{
    offset -= pCur->info.nLocal;
  }


  if( rc==SQLITE_OK && amt>0 ){
    const u32 ovflSize = pBt->usableSize - 4;  /* Bytes content per ovfl page */
    Pgno nextPage;

    nextPage = get4byte(&aPayload[pCur->info.nLocal]);

    /* If the BtCursor.aOverflow[] has not been allocated, allocate it now.
    **
    ** The aOverflow[] array is sized at one entry for each overflow page
    ** in the overflow chain. The page number of the first overflow page is
    ** stored in aOverflow[0], etc. A value of 0 in the aOverflow[] array
    ** means "not yet known" (the cache is lazily populated).
    */
    if( (pCur->curFlags & BTCF_ValidOvfl)==0 ){
      int nOvfl = (pCur->info.nPayload-pCur->info.nLocal+ovflSize-1)/ovflSize;
      if( pCur->aOverflow==0
       || nOvfl*(int)sizeof(Pgno) > sqlite3MallocSize(pCur->aOverflow)
      ){
        Pgno *aNew = (Pgno*)sqlite3Realloc(
            pCur->aOverflow, nOvfl*2*sizeof(Pgno)
        );
        if( aNew==0 ){
          return SQLITE_NOMEM_BKPT;
        }else{
          pCur->aOverflow = aNew;
        }
      }
      memset(pCur->aOverflow, 0, nOvfl*sizeof(Pgno));
      pCur->curFlags |= BTCF_ValidOvfl;
    }else{
      /* If the overflow page-list cache has been allocated and the
      ** entry for the first required overflow page is valid, skip
      ** directly to it.
      */
      if( pCur->aOverflow[offset/ovflSize] ){
        iIdx = (offset/ovflSize);
        nextPage = pCur->aOverflow[iIdx];
        offset = (offset%ovflSize);
      }
    }

    assert( rc==SQLITE_OK && amt>0 );
    while( nextPage ){
      /* If required, populate the overflow page-list cache. */
      if( nextPage > pBt->nPage ) return SQLITE_CORRUPT_BKPT;
      assert( pCur->aOverflow[iIdx]==0
              || pCur->aOverflow[iIdx]==nextPage
              || CORRUPT_DB );
      pCur->aOverflow[iIdx] = nextPage;

      if( offset>=ovflSize ){
        /* The only reason to read this page is to obtain the page
        ** number for the next page in the overflow chain. The page
        ** data is not required. So first try to lookup the overflow
        ** page-list cache, if any, then fall back to the getOverflowPage()
        ** function.
        */
        assert( pCur->curFlags & BTCF_ValidOvfl );
        assert( pCur->pBtree->db==pBt->db );
        if( pCur->aOverflow[iIdx+1] ){
          nextPage = pCur->aOverflow[iIdx+1];
        }else{
          rc = getOverflowPage(pBt, nextPage, 0, &nextPage);
        }
        offset -= ovflSize;
      }else{
        /* Need to read this page properly. It contains some of the
        ** range of data that is being read (eOp==0) or written (eOp!=0).
        */
        int a = amt;
        if( a + offset > ovflSize ){
          a = ovflSize - offset;
        }

#ifdef SQLITE_DIRECT_OVERFLOW_READ
        /* If all the following are true:
        **
        **   1) this is a read operation, and
        **   2) data is required from the start of this overflow page, and
        **   3) there are no dirty pages in the page-cache
        **   4) the database is file-backed, and
        **   5) the page is not in the WAL file
        **   6) at least 4 bytes have already been read into the output buffer
        **
        ** then data can be read directly from the database file into the
        ** output buffer, bypassing the page-cache altogether. This speeds
        ** up loading large records that span many overflow pages.
        */
        if( eOp==0                                             /* (1) */
         && offset==0                                          /* (2) */
         && sqlite3PagerDirectReadOk(pBt->pPager, nextPage)    /* (3,4,5) */
         && &pBuf[-4]>=pBufStart                               /* (6) */
        ){
          sqlite3_file *fd = sqlite3PagerFile(pBt->pPager);
          u8 aSave[4];
          u8 *aWrite = &pBuf[-4];
          assert( aWrite>=pBufStart );                         /* due to (6) */
          memcpy(aSave, aWrite, 4);
          rc = sqlite3OsRead(fd, aWrite, a+4, (i64)pBt->pageSize*(nextPage-1));
          nextPage = get4byte(aWrite);
          memcpy(aWrite, aSave, 4);
        }else
#endif

        {
          DbPage *pDbPage;
          rc = sqlite3PagerGet(pBt->pPager, nextPage, &pDbPage,
              (eOp==0 ? PAGER_GET_READONLY : 0)
          );
          if( rc==SQLITE_OK ){
            aPayload = sqlite3PagerGetData(pDbPage);
            nextPage = get4byte(aPayload);
            rc = copyPayload(&aPayload[offset+4], pBuf, a, eOp, pDbPage);
            sqlite3PagerUnref(pDbPage);
            offset = 0;
          }
        }
        amt -= a;
        if( amt==0 ) return rc;
        pBuf += a;
      }
      if( rc ) break;
      iIdx++;
    }
  }

  if( rc==SQLITE_OK && amt>0 ){
    /* Overflow chain ends prematurely */
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  return rc;
}

/*
** Read part of the payload for the row at which that cursor pCur is currently
** pointing.  "amt" bytes will be transferred into pBuf[].  The transfer
** begins at "offset".
**
** pCur can be pointing to either a table or an index b-tree.
** If pointing to a table btree, then the content section is read.  If
** pCur is pointing to an index b-tree then the key section is read.
**
** For sqlite3BtreePayload(), the caller must ensure that pCur is pointing
** to a valid row in the table.  For sqlite3BtreePayloadChecked(), the
** cursor might be invalid or might need to be restored before being read.
**
** Return SQLITE_OK on success or an error code if anything goes
** wrong.  An error is returned if "offset+amt" is larger than
** the available payload.
*/
int sqlite3BtreePayload(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){
  assert( cursorHoldsMutex(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->iPage>=0 && pCur->pPage );
  return accessPayload(pCur, offset, amt, (unsigned char*)pBuf, 0);
}

/*
** This variant of sqlite3BtreePayload() works even if the cursor has not
** in the CURSOR_VALID state.  It is only used by the sqlite3_blob_read()
** interface.
*/
#ifndef SQLITE_OMIT_INCRBLOB
static SQLITE_NOINLINE int accessPayloadChecked(
  BtCursor *pCur,
  u32 offset,
  u32 amt,
  void *pBuf
){
  int rc;
  if ( pCur->eState==CURSOR_INVALID ){
    return SQLITE_ABORT;
  }
  assert( cursorOwnsBtShared(pCur) );
  rc = btreeRestoreCursorPosition(pCur);
  return rc ? rc : accessPayload(pCur, offset, amt, pBuf, 0);
}
int sqlite3BtreePayloadChecked(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){
  if( pCur->eState==CURSOR_VALID ){
    assert( cursorOwnsBtShared(pCur) );
    return accessPayload(pCur, offset, amt, pBuf, 0);
  }else{
    return accessPayloadChecked(pCur, offset, amt, pBuf);
  }
}
#endif /* SQLITE_OMIT_INCRBLOB */

/*
** Return a pointer to payload information from the entry that the
** pCur cursor is pointing to.  The pointer is to the beginning of
** the key if index btrees (pPage->intKey==0) and is the data for
** table btrees (pPage->intKey==1). The number of bytes of available
** key/data is written into *pAmt.  If *pAmt==0, then the value
** returned will not be a valid pointer.
**
** This routine is an optimization.  It is common for the entire key
** and data to fit on the local page and for there to be no overflow
** pages.  When that is so, this routine can be used to access the
** key and data without making a copy.  If the key and/or data spills
** onto overflow pages, then accessPayload() must be used to reassemble
** the key/data and copy it into a preallocated buffer.
**
** The pointer returned by this routine looks directly into the cached
** page of the database.  The data might change or move the next time
** any btree routine is called.
*/
static const void *fetchPayload(
  BtCursor *pCur,      /* Cursor pointing to entry to read from */
  u32 *pAmt            /* Write the number of available bytes here */
){
  int amt;
  assert( pCur!=0 && pCur->iPage>=0 && pCur->pPage);
  assert( pCur->eState==CURSOR_VALID );
  assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) );
  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->ix<pCur->pPage->nCell || CORRUPT_DB );
  assert( pCur->info.nSize>0 );
  assert( pCur->info.pPayload>pCur->pPage->aData || CORRUPT_DB );
  assert( pCur->info.pPayload<pCur->pPage->aDataEnd ||CORRUPT_DB);
  amt = pCur->info.nLocal;
  if( amt>(int)(pCur->pPage->aDataEnd - pCur->info.pPayload) ){
    /* There is too little space on the page for the expected amount
    ** of local content. Database must be corrupt. */
    assert( CORRUPT_DB );
    amt = MAX(0, (int)(pCur->pPage->aDataEnd - pCur->info.pPayload));
  }
  *pAmt = (u32)amt;
  return (void*)pCur->info.pPayload;
}


/*
** For the entry that cursor pCur is point to, return as
** many bytes of the key or data as are available on the local
** b-tree page.  Write the number of available bytes into *pAmt.
**
** The pointer returned is ephemeral.  The key/data may move
** or be destroyed on the next call to any Btree routine,
** including calls from other threads against the same cache.
** Hence, a mutex on the BtShared should be held prior to calling
** this routine.
**
** These routines is used to get quick access to key and data
** in the common case where no overflow pages are used.
*/
const void *sqlite3BtreePayloadFetch(BtCursor *pCur, u32 *pAmt){
  return fetchPayload(pCur, pAmt);
}


/*
** Move the cursor down to a new child page.  The newPgno argument is the
** page number of the child page to move to.
**
** This function returns SQLITE_CORRUPT if the page-header flags field of
** the new child page does not match the flags field of the parent (i.e.
** if an intkey page appears to be the parent of a non-intkey page, or
** vice-versa).
*/
static int moveToChild(BtCursor *pCur, u32 newPgno){
  int rc;
  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->iPage<BTCURSOR_MAX_DEPTH );
  assert( pCur->iPage>=0 );
  if( pCur->iPage>=(BTCURSOR_MAX_DEPTH-1) ){
    return SQLITE_CORRUPT_BKPT;
  }
  pCur->info.nSize = 0;
  pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl);
  pCur->aiIdx[pCur->iPage] = pCur->ix;
  pCur->apPage[pCur->iPage] = pCur->pPage;
  pCur->ix = 0;
  pCur->iPage++;
  rc = getAndInitPage(pCur->pBt, newPgno, &pCur->pPage, pCur->curPagerFlags);
  assert( pCur->pPage!=0 || rc!=SQLITE_OK );
  if( rc==SQLITE_OK
   && (pCur->pPage->nCell<1 || pCur->pPage->intKey!=pCur->curIntKey)
  ){
    releasePage(pCur->pPage);
    rc = SQLITE_CORRUPT_PGNO(newPgno);
  }
  if( rc ){
    pCur->pPage = pCur->apPage[--pCur->iPage];
  }
  return rc;
}

#ifdef SQLITE_DEBUG
/*
** Page pParent is an internal (non-leaf) tree page. This function
** asserts that page number iChild is the left-child if the iIdx'th
** cell in page pParent. Or, if iIdx is equal to the total number of
** cells in pParent, that page number iChild is the right-child of
** the page.
*/
static void assertParentIndex(MemPage *pParent, int iIdx, Pgno iChild){
  if( CORRUPT_DB ) return;  /* The conditions tested below might not be true
                            ** in a corrupt database */
  assert( iIdx<=pParent->nCell );
  if( iIdx==pParent->nCell ){
    assert( get4byte(&pParent->aData[pParent->hdrOffset+8])==iChild );
  }else{
    assert( get4byte(findCell(pParent, iIdx))==iChild );
  }
}
#else
#  define assertParentIndex(x,y,z)
#endif

/*
** Move the cursor up to the parent page.
**
** pCur->idx is set to the cell index that contains the pointer
** to the page we are coming from.  If we are coming from the
** right-most child page then pCur->idx is set to one more than
** the largest cell index.
*/
static void moveToParent(BtCursor *pCur){
  MemPage *pLeaf;
  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->iPage>0 );
  assert( pCur->pPage );
  assertParentIndex(
    pCur->apPage[pCur->iPage-1],
    pCur->aiIdx[pCur->iPage-1],
    pCur->pPage->pgno
  );
  testcase( pCur->aiIdx[pCur->iPage-1] > pCur->apPage[pCur->iPage-1]->nCell );
  pCur->info.nSize = 0;
  pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl);
  pCur->ix = pCur->aiIdx[pCur->iPage-1];
  pLeaf = pCur->pPage;
  pCur->pPage = pCur->apPage[--pCur->iPage];
  releasePageNotNull(pLeaf);
}

/*
** Move the cursor to point to the root page of its b-tree structure.
**
** If the table has a virtual root page, then the cursor is moved to point
** to the virtual root page instead of the actual root page. A table has a
** virtual root page when the actual root page contains no cells and a
** single child page. This can only happen with the table rooted at page 1.
**
** If the b-tree structure is empty, the cursor state is set to
** CURSOR_INVALID and this routine returns SQLITE_EMPTY. Otherwise,
** the cursor is set to point to the first cell located on the root
** (or virtual root) page and the cursor state is set to CURSOR_VALID.
**
** If this function returns successfully, it may be assumed that the
** page-header flags indicate that the [virtual] root-page is the expected
** kind of b-tree page (i.e. if when opening the cursor the caller did not
** specify a KeyInfo structure the flags byte is set to 0x05 or 0x0D,
** indicating a table b-tree, or if the caller did specify a KeyInfo
** structure the flags byte is set to 0x02 or 0x0A, indicating an index
** b-tree).
*/
static int moveToRoot(BtCursor *pCur){
  MemPage *pRoot;
  int rc = SQLITE_OK;

  assert( cursorOwnsBtShared(pCur) );
  assert( CURSOR_INVALID < CURSOR_REQUIRESEEK );
  assert( CURSOR_VALID   < CURSOR_REQUIRESEEK );
  assert( CURSOR_FAULT   > CURSOR_REQUIRESEEK );
  assert( pCur->eState < CURSOR_REQUIRESEEK || pCur->iPage<0 );
  assert( pCur->pgnoRoot>0 || pCur->iPage<0 );

  if( pCur->iPage>=0 ){
    if( pCur->iPage ){
      releasePageNotNull(pCur->pPage);
      while( --pCur->iPage ){
        releasePageNotNull(pCur->apPage[pCur->iPage]);
      }
      pRoot = pCur->pPage = pCur->apPage[0];
      goto skip_init;
    }
  }else if( pCur->pgnoRoot==0 ){
    pCur->eState = CURSOR_INVALID;
    return SQLITE_EMPTY;
  }else{
    assert( pCur->iPage==(-1) );
    if( pCur->eState>=CURSOR_REQUIRESEEK ){
      if( pCur->eState==CURSOR_FAULT ){
        assert( pCur->skipNext!=SQLITE_OK );
        return pCur->skipNext;
      }
      sqlite3BtreeClearCursor(pCur);
    }
    rc = getAndInitPage(pCur->pBt, pCur->pgnoRoot, &pCur->pPage,
                        pCur->curPagerFlags);
    if( rc!=SQLITE_OK ){
      pCur->eState = CURSOR_INVALID;
      return rc;
    }
    pCur->iPage = 0;
    pCur->curIntKey = pCur->pPage->intKey;
  }
  pRoot = pCur->pPage;
  assert( pRoot->pgno==pCur->pgnoRoot || CORRUPT_DB );

  /* If pCur->pKeyInfo is not NULL, then the caller that opened this cursor
  ** expected to open it on an index b-tree. Otherwise, if pKeyInfo is
  ** NULL, the caller expects a table b-tree. If this is not the case,
  ** return an SQLITE_CORRUPT error.
  **
  ** Earlier versions of SQLite assumed that this test could not fail
  ** if the root page was already loaded when this function was called (i.e.
  ** if pCur->iPage>=0). But this is not so if the database is corrupted
  ** in such a way that page pRoot is linked into a second b-tree table
  ** (or the freelist).  */
  assert( pRoot->intKey==1 || pRoot->intKey==0 );
  if( pRoot->isInit==0 || (pCur->pKeyInfo==0)!=pRoot->intKey ){
    return SQLITE_CORRUPT_PAGE(pCur->pPage);
  }

skip_init: 
  pCur->ix = 0;
  pCur->info.nSize = 0;
  pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidNKey|BTCF_ValidOvfl);

  if( pRoot->nCell>0 ){
    pCur->eState = CURSOR_VALID;
  }else if( !pRoot->leaf ){
    Pgno subpage;
    if( pRoot->pgno!=1 ) return SQLITE_CORRUPT_BKPT;
    subpage = get4byte(&pRoot->aData[pRoot->hdrOffset+8]);
    pCur->eState = CURSOR_VALID;
    rc = moveToChild(pCur, subpage);
  }else{
    pCur->eState = CURSOR_INVALID;
    rc = SQLITE_EMPTY;
  }
  return rc;
}

/*
** Move the cursor down to the left-most leaf entry beneath the
** entry to which it is currently pointing.
**
** The left-most leaf is the one with the smallest key - the first
** in ascending order.
*/
static int moveToLeftmost(BtCursor *pCur){
  Pgno pgno;
  int rc = SQLITE_OK;
  MemPage *pPage;

  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  while( rc==SQLITE_OK && !(pPage = pCur->pPage)->leaf ){
    assert( pCur->ix<pPage->nCell );
    pgno = get4byte(findCell(pPage, pCur->ix));
    rc = moveToChild(pCur, pgno);
  }
  return rc;
}

/*
** Move the cursor down to the right-most leaf entry beneath the
** page to which it is currently pointing.  Notice the difference
** between moveToLeftmost() and moveToRightmost().  moveToLeftmost()
** finds the left-most entry beneath the *entry* whereas moveToRightmost()
** finds the right-most entry beneath the *page*.
**
** The right-most entry is the one with the largest key - the last
** key in ascending order.
*/
static int moveToRightmost(BtCursor *pCur){
  Pgno pgno;
  int rc = SQLITE_OK;
  MemPage *pPage = 0;

  assert( cursorOwnsBtShared(pCur) );
  assert( pCur->eState==CURSOR_VALID );
  while( !(pPage = pCur->pPage)->leaf ){
    pgno = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    pCur->ix = pPage->nCell;
    rc = moveToChild(pCur, pgno);
    if( rc ) return rc;
  }
  pCur->ix = pPage->nCell-1;
  assert( pCur->info.nSize==0 );
  assert( (pCur->curFlags & BTCF_ValidNKey)==0 );
  return SQLITE_OK;
}

/* Move the cursor to the first entry in the table.  Return SQLITE_OK
** on success.  Set *pRes to 0 if the cursor actually points to something
** or set *pRes to 1 if the table is empty.
*/
int sqlite3BtreeFirst(BtCursor *pCur, int *pRes){
  int rc;

  assert( cursorOwnsBtShared(pCur) );
  assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) );
  rc = moveToRoot(pCur);
  if( rc==SQLITE_OK ){
    assert( pCur->pPage->nCell>0 );
    *pRes = 0;
    rc = moveToLeftmost(pCur);
  }else if( rc==SQLITE_EMPTY ){
    assert( pCur->pgnoRoot==0 || (pCur->pPage!=0 && pCur->pPage->nCell==0) );
    *pRes = 1;
    rc = SQLITE_OK;
  }
  return rc;
}

/* Move the cursor to the last entry in the table.  Return SQLITE_OK
** on success.  Set *pRes to 0 if the cursor actually points to something
** or set *pRes to 1 if the table is empty.
*/
static SQLITE_NOINLINE int btreeLast(BtCursor *pCur, int *pRes){
  int rc = moveToRoot(pCur);
  if( rc==SQLITE_OK ){
    assert( pCur->eState==CURSOR_VALID );
    *pRes = 0;
    rc = moveToRightmost(pCur);
    if( rc==SQLITE_OK ){
      pCur->curFlags |= BTCF_AtLast;
    }else{
      pCur->curFlags &= ~BTCF_AtLast;
    }
  }else if( rc==SQLITE_EMPTY ){
    assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 );
    *pRes = 1;
    rc = SQLITE_OK;
  }
  return rc;
}
int sqlite3BtreeLast(BtCursor *pCur, int *pRes){
  assert( cursorOwnsBtShared(pCur) );
  assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) );

  /* If the cursor already points to the last entry, this is a no-op. */
  if( CURSOR_VALID==pCur->eState && (pCur->curFlags & BTCF_AtLast)!=0 ){
#ifdef SQLITE_DEBUG
    /* This block serves to assert() that the cursor really does point
    ** to the last entry in the b-tree. */
    int ii;
    for(ii=0; ii<pCur->iPage; ii++){
      assert( pCur->aiIdx[ii]==pCur->apPage[ii]->nCell );
    }
    assert( pCur->ix==pCur->pPage->nCell-1 || CORRUPT_DB );
    testcase( pCur->ix!=pCur->pPage->nCell-1 );
    /* ^-- dbsqlfuzz b92b72e4de80b5140c30ab71372ca719b8feb618 */
    assert( pCur->pPage->leaf );
#endif
    *pRes = 0;
    return SQLITE_OK;
  }
  return btreeLast(pCur, pRes);
}

/* Move the cursor so that it points to an entry in a table (a.k.a INTKEY)
** table near the key intKey.   Return a success code.
**
** If an exact match is not found, then the cursor is always
** left pointing at a leaf page which would hold the entry if it
** were present.  The cursor might point to an entry that comes
** before or after the key.
**
** An integer is written into *pRes which is the result of
** comparing the key with the entry to which the cursor is
** pointing.  The meaning of the integer written into
** *pRes is as follows:
**
**     *pRes<0      The cursor is left pointing at an entry that
**                  is smaller than intKey or if the table is empty
**                  and the cursor is therefore left point to nothing.
**
**     *pRes==0     The cursor is left pointing at an entry that
**                  exactly matches intKey.
**
**     *pRes>0      The cursor is left pointing at an entry that
**                  is larger than intKey.
*/
int sqlite3BtreeTableMoveto(
  BtCursor *pCur,          /* The cursor to be moved */
  i64 intKey,              /* The table key */
  int biasRight,           /* If true, bias the search to the high end */
  int *pRes                /* Write search results here */
){
  int rc;

  assert( cursorOwnsBtShared(pCur) );
  assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) );
  assert( pRes );
  assert( pCur->pKeyInfo==0 );
  assert( pCur->eState!=CURSOR_VALID || pCur->curIntKey!=0 );

  /* If the cursor is already positioned at the point we are trying
  ** to move to, then just return without doing any work */
  if( pCur->eState==CURSOR_VALID && (pCur->curFlags & BTCF_ValidNKey)!=0 ){
    if( pCur->info.nKey==intKey ){
      *pRes = 0;
      return SQLITE_OK;
    }
    if( pCur->info.nKey<intKey ){
      if( (pCur->curFlags & BTCF_AtLast)!=0 ){
        *pRes = -1;
        return SQLITE_OK;
      }
      /* If the requested key is one more than the previous key, then
      ** try to get there using sqlite3BtreeNext() rather than a full
      ** binary search.  This is an optimization only.  The correct answer
      ** is still obtained without this case, only a little more slowly. */
      if( pCur->info.nKey+1==intKey ){
        *pRes = 0;
        rc = sqlite3BtreeNext(pCur, 0);
        if( rc==SQLITE_OK ){
          getCellInfo(pCur);
          if( pCur->info.nKey==intKey ){
            return SQLITE_OK;
          }
        }else if( rc!=SQLITE_DONE ){
          return rc;
        }
      }
    }
  }

#ifdef SQLITE_DEBUG
  pCur->pBtree->nSeek++;   /* Performance measurement during testing */
#endif

  rc = moveToRoot(pCur);
  if( rc ){
    if( rc==SQLITE_EMPTY ){
      assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 );
      *pRes = -1;
      return SQLITE_OK;
    }
    return rc;
  }
  assert( pCur->pPage );
  assert( pCur->pPage->isInit );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->pPage->nCell > 0 );
  assert( pCur->iPage==0 || pCur->apPage[0]->intKey==pCur->curIntKey );
  assert( pCur->curIntKey );

  for(;;){
    int lwr, upr, idx, c;
    Pgno chldPg;
    MemPage *pPage = pCur->pPage;
    u8 *pCell;                          /* Pointer to current cell in pPage */

    /* pPage->nCell must be greater than zero. If this is the root-page
    ** the cursor would have been INVALID above and this for(;;) loop
    ** not run. If this is not the root-page, then the moveToChild() routine
    ** would have already detected db corruption. Similarly, pPage must
    ** be the right kind (index or table) of b-tree page. Otherwise
    ** a moveToChild() or moveToRoot() call would have detected corruption.  */
    assert( pPage->nCell>0 );
    assert( pPage->intKey );
    lwr = 0;
    upr = pPage->nCell-1;
    assert( biasRight==0 || biasRight==1 );
    idx = upr>>(1-biasRight); /* idx = biasRight ? upr : (lwr+upr)/2; */
    for(;;){
      i64 nCellKey;
      pCell = findCellPastPtr(pPage, idx);
      if( pPage->intKeyLeaf ){
        while( 0x80 <= *(pCell++) ){
          if( pCell>=pPage->aDataEnd ){
            return SQLITE_CORRUPT_PAGE(pPage);
          }
        }
      }
      getVarint(pCell, (u64*)&nCellKey);
      if( nCellKey<intKey ){
        lwr = idx+1;
        if( lwr>upr ){ c = -1; break; }
      }else if( nCellKey>intKey ){
        upr = idx-1;
        if( lwr>upr ){ c = +1; break; }
      }else{
        assert( nCellKey==intKey );
        pCur->ix = (u16)idx;
        if( !pPage->leaf ){
          lwr = idx;
          goto moveto_table_next_layer;
        }else{
          pCur->curFlags |= BTCF_ValidNKey;
          pCur->info.nKey = nCellKey;
          pCur->info.nSize = 0;
          *pRes = 0;
          return SQLITE_OK;
        }
      }
      assert( lwr+upr>=0 );
      idx = (lwr+upr)>>1;  /* idx = (lwr+upr)/2; */
    }
    assert( lwr==upr+1 || !pPage->leaf );
    assert( pPage->isInit );
    if( pPage->leaf ){
      assert( pCur->ix<pCur->pPage->nCell );
      pCur->ix = (u16)idx;
      *pRes = c;
      rc = SQLITE_OK;
      goto moveto_table_finish;
    }
moveto_table_next_layer:
    if( lwr>=pPage->nCell ){
      chldPg = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    }else{
      chldPg = get4byte(findCell(pPage, lwr));
    }
    pCur->ix = (u16)lwr;
    rc = moveToChild(pCur, chldPg);
    if( rc ) break;
  }
moveto_table_finish:
  pCur->info.nSize = 0;
  assert( (pCur->curFlags & BTCF_ValidOvfl)==0 );
  return rc;
}

/*
** Compare the "idx"-th cell on the page the cursor pCur is currently
** pointing to to pIdxKey using xRecordCompare.  Return negative or
** zero if the cell is less than or equal pIdxKey.  Return positive
** if unknown.
**
**    Return value negative:     Cell at pCur[idx] less than pIdxKey
**
**    Return value is zero:      Cell at pCur[idx] equals pIdxKey
**
**    Return value positive:     Nothing is known about the relationship
**                               of the cell at pCur[idx] and pIdxKey.
**
** This routine is part of an optimization.  It is always safe to return
** a positive value as that will cause the optimization to be skipped.
*/
static int indexCellCompare(
  BtCursor *pCur,
  int idx,
  UnpackedRecord *pIdxKey,
  RecordCompare xRecordCompare
){
  MemPage *pPage = pCur->pPage;
  int c;
  int nCell;  /* Size of the pCell cell in bytes */
  u8 *pCell = findCellPastPtr(pPage, idx);

  nCell = pCell[0];
  if( nCell<=pPage->max1bytePayload ){
    /* This branch runs if the record-size field of the cell is a
    ** single byte varint and the record fits entirely on the main
    ** b-tree page.  */
    testcase( pCell+nCell+1==pPage->aDataEnd );
    c = xRecordCompare(nCell, (void*)&pCell[1], pIdxKey);
  }else if( !(pCell[1] & 0x80)
    && (nCell = ((nCell&0x7f)<<7) + pCell[1])<=pPage->maxLocal
  ){
    /* The record-size field is a 2 byte varint and the record
    ** fits entirely on the main b-tree page.  */
    testcase( pCell+nCell+2==pPage->aDataEnd );
    c = xRecordCompare(nCell, (void*)&pCell[2], pIdxKey);
  }else{
    /* If the record extends into overflow pages, do not attempt
    ** the optimization. */
    c = 99;
  }
  return c;
}

/*
** Return true (non-zero) if pCur is current pointing to the last
** page of a table.
*/
static int cursorOnLastPage(BtCursor *pCur){
  int i;
  assert( pCur->eState==CURSOR_VALID );
  for(i=0; i<pCur->iPage; i++){
    MemPage *pPage = pCur->apPage[i];
    if( pCur->aiIdx[i]<pPage->nCell ) return 0;
  }
  return 1;
}

/* Move the cursor so that it points to an entry in an index table
** near the key pIdxKey.   Return a success code.
**
** If an exact match is not found, then the cursor is always
** left pointing at a leaf page which would hold the entry if it
** were present.  The cursor might point to an entry that comes
** before or after the key.
**
** An integer is written into *pRes which is the result of
** comparing the key with the entry to which the cursor is
** pointing.  The meaning of the integer written into
** *pRes is as follows:
**
**     *pRes<0      The cursor is left pointing at an entry that
**                  is smaller than pIdxKey or if the table is empty
**                  and the cursor is therefore left point to nothing.
**
**     *pRes==0     The cursor is left pointing at an entry that
**                  exactly matches pIdxKey.
**
**     *pRes>0      The cursor is left pointing at an entry that
**                  is larger than pIdxKey.
**
** The pIdxKey->eqSeen field is set to 1 if there
** exists an entry in the table that exactly matches pIdxKey. 
*/
int sqlite3BtreeIndexMoveto(
  BtCursor *pCur,          /* The cursor to be moved */
  UnpackedRecord *pIdxKey, /* Unpacked index key */
  int *pRes                /* Write search results here */
){
  int rc;
  RecordCompare xRecordCompare;

  assert( cursorOwnsBtShared(pCur) );
  assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) );
  assert( pRes );
  assert( pCur->pKeyInfo!=0 );

#ifdef SQLITE_DEBUG
  pCur->pBtree->nSeek++;   /* Performance measurement during testing */
#endif

  xRecordCompare = sqlite3VdbeFindCompare(pIdxKey);
  pIdxKey->errCode = 0;
  assert( pIdxKey->default_rc==1
       || pIdxKey->default_rc==0
       || pIdxKey->default_rc==-1
  );


  /* Check to see if we can skip a lot of work.  Two cases:
  **
  **    (1) If the cursor is already pointing to the very last cell
  **        in the table and the pIdxKey search key is greater than or
  **        equal to that last cell, then no movement is required.
  **
  **    (2) If the cursor is on the last page of the table and the first
  **        cell on that last page is less than or equal to the pIdxKey
  **        search key, then we can start the search on the current page
  **        without needing to go back to root.
  */
  if( pCur->eState==CURSOR_VALID
   && pCur->pPage->leaf
   && cursorOnLastPage(pCur)
  ){
    int c;
    if( pCur->ix==pCur->pPage->nCell-1
     && (c = indexCellCompare(pCur, pCur->ix, pIdxKey, xRecordCompare))<=0
     && pIdxKey->errCode==SQLITE_OK
    ){
      *pRes = c;
      return SQLITE_OK;  /* Cursor already pointing at the correct spot */
    }
    if( pCur->iPage>0
     && indexCellCompare(pCur, 0, pIdxKey, xRecordCompare)<=0
     && pIdxKey->errCode==SQLITE_OK
    ){
      pCur->curFlags &= ~BTCF_ValidOvfl;
      if( !pCur->pPage->isInit ){
        return SQLITE_CORRUPT_BKPT;
      }
      goto bypass_moveto_root;  /* Start search on the current page */
    }
    pIdxKey->errCode = SQLITE_OK;
  }

  rc = moveToRoot(pCur);
  if( rc ){
    if( rc==SQLITE_EMPTY ){
      assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 );
      *pRes = -1;
      return SQLITE_OK;
    }
    return rc;
  }

bypass_moveto_root:
  assert( pCur->pPage );
  assert( pCur->pPage->isInit );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->pPage->nCell > 0 );
  assert( pCur->curIntKey==0 );
  assert( pIdxKey!=0 );
  for(;;){
    int lwr, upr, idx, c;
    Pgno chldPg;
    MemPage *pPage = pCur->pPage;
    u8 *pCell;                          /* Pointer to current cell in pPage */

    /* pPage->nCell must be greater than zero. If this is the root-page
    ** the cursor would have been INVALID above and this for(;;) loop
    ** not run. If this is not the root-page, then the moveToChild() routine
    ** would have already detected db corruption. Similarly, pPage must
    ** be the right kind (index or table) of b-tree page. Otherwise
    ** a moveToChild() or moveToRoot() call would have detected corruption.  */
    assert( pPage->nCell>0 );
    assert( pPage->intKey==0 );
    lwr = 0;
    upr = pPage->nCell-1;
    idx = upr>>1; /* idx = (lwr+upr)/2; */
    for(;;){
      int nCell;  /* Size of the pCell cell in bytes */
      pCell = findCellPastPtr(pPage, idx);

      /* The maximum supported page-size is 65536 bytes. This means that
      ** the maximum number of record bytes stored on an index B-Tree
      ** page is less than 16384 bytes and may be stored as a 2-byte
      ** varint. This information is used to attempt to avoid parsing
      ** the entire cell by checking for the cases where the record is
      ** stored entirely within the b-tree page by inspecting the first
      ** 2 bytes of the cell.
      */
      nCell = pCell[0];
      if( nCell<=pPage->max1bytePayload ){
        /* This branch runs if the record-size field of the cell is a
        ** single byte varint and the record fits entirely on the main
        ** b-tree page.  */
        testcase( pCell+nCell+1==pPage->aDataEnd );
        c = xRecordCompare(nCell, (void*)&pCell[1], pIdxKey);
      }else if( !(pCell[1] & 0x80)
        && (nCell = ((nCell&0x7f)<<7) + pCell[1])<=pPage->maxLocal
      ){
        /* The record-size field is a 2 byte varint and the record
        ** fits entirely on the main b-tree page.  */
        testcase( pCell+nCell+2==pPage->aDataEnd );
        c = xRecordCompare(nCell, (void*)&pCell[2], pIdxKey);
      }else{
        /* The record flows over onto one or more overflow pages. In
        ** this case the whole cell needs to be parsed, a buffer allocated
        ** and accessPayload() used to retrieve the record into the
        ** buffer before VdbeRecordCompare() can be called.
        **
        ** If the record is corrupt, the xRecordCompare routine may read
        ** up to two varints past the end of the buffer. An extra 18
        ** bytes of padding is allocated at the end of the buffer in
        ** case this happens.  */
        void *pCellKey;
        u8 * const pCellBody = pCell - pPage->childPtrSize;
        const int nOverrun = 18;  /* Size of the overrun padding */
        pPage->xParseCell(pPage, pCellBody, &pCur->info);
        nCell = (int)pCur->info.nKey;
        testcase( nCell<0 );   /* True if key size is 2^32 or more */
        testcase( nCell==0 );  /* Invalid key size:  0x80 0x80 0x00 */
        testcase( nCell==1 );  /* Invalid key size:  0x80 0x80 0x01 */
        testcase( nCell==2 );  /* Minimum legal index key size */
        if( nCell<2 || nCell/pCur->pBt->usableSize>pCur->pBt->nPage ){
          rc = SQLITE_CORRUPT_PAGE(pPage);
          goto moveto_index_finish;
        }
        pCellKey = sqlite3Malloc( nCell+nOverrun );
        if( pCellKey==0 ){
          rc = SQLITE_NOMEM_BKPT;
          goto moveto_index_finish;
        }
        pCur->ix = (u16)idx;
        rc = accessPayload(pCur, 0, nCell, (unsigned char*)pCellKey, 0);
        memset(((u8*)pCellKey)+nCell,0,nOverrun); /* Fix uninit warnings */
        pCur->curFlags &= ~BTCF_ValidOvfl;
        if( rc ){
          sqlite3_free(pCellKey);
          goto moveto_index_finish;
        }
        c = sqlite3VdbeRecordCompare(nCell, pCellKey, pIdxKey);
        sqlite3_free(pCellKey);
      }
      assert(
          (pIdxKey->errCode!=SQLITE_CORRUPT || c==0)
       && (pIdxKey->errCode!=SQLITE_NOMEM || pCur->pBtree->db->mallocFailed)
      );
      if( c<0 ){
        lwr = idx+1;
      }else if( c>0 ){
        upr = idx-1;
      }else{
        assert( c==0 );
        *pRes = 0;
        rc = SQLITE_OK;
        pCur->ix = (u16)idx;
        if( pIdxKey->errCode ) rc = SQLITE_CORRUPT_BKPT;
        goto moveto_index_finish;
      }
      if( lwr>upr ) break;
      assert( lwr+upr>=0 );
      idx = (lwr+upr)>>1;  /* idx = (lwr+upr)/2 */
    }
    assert( lwr==upr+1 || (pPage->intKey && !pPage->leaf) );
    assert( pPage->isInit );
    if( pPage->leaf ){
      assert( pCur->ix<pCur->pPage->nCell || CORRUPT_DB );
      pCur->ix = (u16)idx;
      *pRes = c;
      rc = SQLITE_OK;
      goto moveto_index_finish;
    }
    if( lwr>=pPage->nCell ){
      chldPg = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    }else{
      chldPg = get4byte(findCell(pPage, lwr));
    }

    /* This block is similar to an in-lined version of:
    **
    **    pCur->ix = (u16)lwr;
    **    rc = moveToChild(pCur, chldPg);
    **    if( rc ) break;
    */
    pCur->info.nSize = 0;
    pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl);
    if( pCur->iPage>=(BTCURSOR_MAX_DEPTH-1) ){
      return SQLITE_CORRUPT_BKPT;
    }
    pCur->aiIdx[pCur->iPage] = (u16)lwr;
    pCur->apPage[pCur->iPage] = pCur->pPage;
    pCur->ix = 0;
    pCur->iPage++;
    rc = getAndInitPage(pCur->pBt, chldPg, &pCur->pPage, pCur->curPagerFlags);
    if( rc==SQLITE_OK
     && (pCur->pPage->nCell<1 || pCur->pPage->intKey!=pCur->curIntKey)
    ){
      releasePage(pCur->pPage);
      rc = SQLITE_CORRUPT_PGNO(chldPg);
    }
    if( rc ){
      pCur->pPage = pCur->apPage[--pCur->iPage];
      break;
    }
    /*
    ***** End of in-lined moveToChild() call */
 }
moveto_index_finish:
  pCur->info.nSize = 0;
  assert( (pCur->curFlags & BTCF_ValidOvfl)==0 );
  return rc;
}


/*
** Return TRUE if the cursor is not pointing at an entry of the table.
**
** TRUE will be returned after a call to sqlite3BtreeNext() moves
** past the last entry in the table or sqlite3BtreePrev() moves past
** the first entry.  TRUE is also returned if the table is empty.
*/
int sqlite3BtreeEof(BtCursor *pCur){
  /* TODO: What if the cursor is in CURSOR_REQUIRESEEK but all table entries
  ** have been deleted? This API will need to change to return an error code
  ** as well as the boolean result value.
  */
  return (CURSOR_VALID!=pCur->eState);
}

/*
** Return an estimate for the number of rows in the table that pCur is
** pointing to.  Return a negative number if no estimate is currently
** available.
*/
i64 sqlite3BtreeRowCountEst(BtCursor *pCur){
  i64 n;
  u8 i;

  assert( cursorOwnsBtShared(pCur) );
  assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) );

  /* Currently this interface is only called by the OP_IfSmaller
  ** opcode, and it that case the cursor will always be valid and
  ** will always point to a leaf node. */
  if( NEVER(pCur->eState!=CURSOR_VALID) ) return -1;
  if( NEVER(pCur->pPage->leaf==0) ) return -1;

  n = pCur->pPage->nCell;
  for(i=0; i<pCur->iPage; i++){
    n *= pCur->apPage[i]->nCell;
  }
  return n;
}

/*
** Advance the cursor to the next entry in the database.
** Return value:
**
**    SQLITE_OK        success
**    SQLITE_DONE      cursor is already pointing at the last element
**    otherwise        some kind of error occurred
**
** The main entry point is sqlite3BtreeNext().  That routine is optimized
** for the common case of merely incrementing the cell counter BtCursor.aiIdx
** to the next cell on the current page.  The (slower) btreeNext() helper
** routine is called when it is necessary to move to a different page or
** to restore the cursor.
**
** If bit 0x01 of the F argument in sqlite3BtreeNext(C,F) is 1, then the
** cursor corresponds to an SQL index and this routine could have been
** skipped if the SQL index had been a unique index.  The F argument
** is a hint to the implement.  SQLite btree implementation does not use
** this hint, but COMDB2 does.
*/
static SQLITE_NOINLINE int btreeNext(BtCursor *pCur){
  int rc;
  int idx;
  MemPage *pPage;

  assert( cursorOwnsBtShared(pCur) );
  if( pCur->eState!=CURSOR_VALID ){
    assert( (pCur->curFlags & BTCF_ValidOvfl)==0 );
    rc = restoreCursorPosition(pCur);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( CURSOR_INVALID==pCur->eState ){
      return SQLITE_DONE;
    }
    if( pCur->eState==CURSOR_SKIPNEXT ){
      pCur->eState = CURSOR_VALID;
      if( pCur->skipNext>0 ) return SQLITE_OK;
    }
  }

  pPage = pCur->pPage;
  idx = ++pCur->ix;
  if( sqlite3FaultSim(412) ) pPage->isInit = 0;
  if( !pPage->isInit ){
    return SQLITE_CORRUPT_BKPT;
  }

  if( idx>=pPage->nCell ){
    if( !pPage->leaf ){
      rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8]));
      if( rc ) return rc;
      return moveToLeftmost(pCur);
    }
    do{
      if( pCur->iPage==0 ){
        pCur->eState = CURSOR_INVALID;
        return SQLITE_DONE;
      }
      moveToParent(pCur);
      pPage = pCur->pPage;
    }while( pCur->ix>=pPage->nCell );
    if( pPage->intKey ){
      return sqlite3BtreeNext(pCur, 0);
    }else{
      return SQLITE_OK;
    }
  }
  if( pPage->leaf ){
    return SQLITE_OK;
  }else{
    return moveToLeftmost(pCur);
  }
}
int sqlite3BtreeNext(BtCursor *pCur, int flags){
  MemPage *pPage;
  UNUSED_PARAMETER( flags );  /* Used in COMDB2 but not native SQLite */
  assert( cursorOwnsBtShared(pCur) );
  assert( flags==0 || flags==1 );
  pCur->info.nSize = 0;
  pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl);
  if( pCur->eState!=CURSOR_VALID ) return btreeNext(pCur);
  pPage = pCur->pPage;
  if( (++pCur->ix)>=pPage->nCell ){
    pCur->ix--;
    return btreeNext(pCur);
  }
  if( pPage->leaf ){
    return SQLITE_OK;
  }else{
    return moveToLeftmost(pCur);
  }
}

/*
** Step the cursor to the back to the previous entry in the database.
** Return values:
**
**     SQLITE_OK     success
**     SQLITE_DONE   the cursor is already on the first element of the table
**     otherwise     some kind of error occurred
**
** The main entry point is sqlite3BtreePrevious().  That routine is optimized
** for the common case of merely decrementing the cell counter BtCursor.aiIdx
** to the previous cell on the current page.  The (slower) btreePrevious()
** helper routine is called when it is necessary to move to a different page
** or to restore the cursor.
**
** If bit 0x01 of the F argument to sqlite3BtreePrevious(C,F) is 1, then
** the cursor corresponds to an SQL index and this routine could have been
** skipped if the SQL index had been a unique index.  The F argument is a
** hint to the implement.  The native SQLite btree implementation does not
** use this hint, but COMDB2 does.
*/
static SQLITE_NOINLINE int btreePrevious(BtCursor *pCur){
  int rc;
  MemPage *pPage;

  assert( cursorOwnsBtShared(pCur) );
  assert( (pCur->curFlags & (BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey))==0 );
  assert( pCur->info.nSize==0 );
  if( pCur->eState!=CURSOR_VALID ){
    rc = restoreCursorPosition(pCur);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( CURSOR_INVALID==pCur->eState ){
      return SQLITE_DONE;
    }
    if( CURSOR_SKIPNEXT==pCur->eState ){
      pCur->eState = CURSOR_VALID;
      if( pCur->skipNext<0 ) return SQLITE_OK;
    }
  }

  pPage = pCur->pPage;
  if( sqlite3FaultSim(412) ) pPage->isInit = 0;
  if( !pPage->isInit ){
    return SQLITE_CORRUPT_BKPT;
  }
  if( !pPage->leaf ){
    int idx = pCur->ix;
    rc = moveToChild(pCur, get4byte(findCell(pPage, idx)));
    if( rc ) return rc;
    rc = moveToRightmost(pCur);
  }else{
    while( pCur->ix==0 ){
      if( pCur->iPage==0 ){
        pCur->eState = CURSOR_INVALID;
        return SQLITE_DONE;
      }
      moveToParent(pCur);
    }
    assert( pCur->info.nSize==0 );
    assert( (pCur->curFlags & (BTCF_ValidOvfl))==0 );

    pCur->ix--;
    pPage = pCur->pPage;
    if( pPage->intKey && !pPage->leaf ){
      rc = sqlite3BtreePrevious(pCur, 0);
    }else{
      rc = SQLITE_OK;
    }
  }
  return rc;
}
int sqlite3BtreePrevious(BtCursor *pCur, int flags){
  assert( cursorOwnsBtShared(pCur) );
  assert( flags==0 || flags==1 );
  UNUSED_PARAMETER( flags );  /* Used in COMDB2 but not native SQLite */
  pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey);
  pCur->info.nSize = 0;
  if( pCur->eState!=CURSOR_VALID
   || pCur->ix==0
   || pCur->pPage->leaf==0
  ){
    return btreePrevious(pCur);
  }
  pCur->ix--;
  return SQLITE_OK;
}

/*
** Allocate a new page from the database file.
**
** The new page is marked as dirty.  (In other words, sqlite3PagerWrite()
** has already been called on the new page.)  The new page has also
** been referenced and the calling routine is responsible for calling
** sqlite3PagerUnref() on the new page when it is done.
**
** SQLITE_OK is returned on success.  Any other return value indicates
** an error.  *ppPage is set to NULL in the event of an error.
**
** If the "nearby" parameter is not 0, then an effort is made to
** locate a page close to the page number "nearby".  This can be used in an
** attempt to keep related pages close to each other in the database file,
** which in turn can make database access faster.
**
** If the eMode parameter is BTALLOC_EXACT and the nearby page exists
** anywhere on the free-list, then it is guaranteed to be returned.  If
** eMode is BTALLOC_LT then the page returned will be less than or equal
** to nearby if any such page exists.  If eMode is BTALLOC_ANY then there
** are no restrictions on which page is returned.
*/
static int allocateBtreePage(
  BtShared *pBt,         /* The btree */
  MemPage **ppPage,      /* Store pointer to the allocated page here */
  Pgno *pPgno,           /* Store the page number here */
  Pgno nearby,           /* Search for a page near this one */
  u8 eMode               /* BTALLOC_EXACT, BTALLOC_LT, or BTALLOC_ANY */
){
  MemPage *pPage1;
  int rc;
  u32 n;     /* Number of pages on the freelist */
  u32 k;     /* Number of leaves on the trunk of the freelist */
  MemPage *pTrunk = 0;
  MemPage *pPrevTrunk = 0;
  Pgno mxPage;     /* Total size of the database file */

  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( eMode==BTALLOC_ANY || (nearby>0 && IfNotOmitAV(pBt->autoVacuum)) );
  pPage1 = pBt->pPage1;
  mxPage = btreePagecount(pBt);
  /* EVIDENCE-OF: R-21003-45125 The 4-byte big-endian integer at offset 36
  ** stores the total number of pages on the freelist. */
  n = get4byte(&pPage1->aData[36]);
  testcase( n==mxPage-1 );
  if( n>=mxPage ){
    return SQLITE_CORRUPT_BKPT;
  }
  if( n>0 ){
    /* There are pages on the freelist.  Reuse one of those pages. */
    Pgno iTrunk;
    u8 searchList = 0; /* If the free-list must be searched for 'nearby' */
    u32 nSearch = 0;   /* Count of the number of search attempts */
   
    /* If eMode==BTALLOC_EXACT and a query of the pointer-map
    ** shows that the page 'nearby' is somewhere on the free-list, then
    ** the entire-list will be searched for that page.
    */
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( eMode==BTALLOC_EXACT ){
      if( nearby<=mxPage ){
        u8 eType;
        assert( nearby>0 );
        assert( pBt->autoVacuum );
        rc = ptrmapGet(pBt, nearby, &eType, 0);
        if( rc ) return rc;
        if( eType==PTRMAP_FREEPAGE ){
          searchList = 1;
        }
      }
    }else if( eMode==BTALLOC_LE ){
      searchList = 1;
    }
#endif

    /* Decrement the free-list count by 1. Set iTrunk to the index of the
    ** first free-list trunk page. iPrevTrunk is initially 1.
    */
    rc = sqlite3PagerWrite(pPage1->pDbPage);
    if( rc ) return rc;
    put4byte(&pPage1->aData[36], n-1);

    /* The code within this loop is run only once if the 'searchList' variable
    ** is not true. Otherwise, it runs once for each trunk-page on the
    ** free-list until the page 'nearby' is located (eMode==BTALLOC_EXACT)
    ** or until a page less than 'nearby' is located (eMode==BTALLOC_LT)
    */
    do {
      pPrevTrunk = pTrunk;
      if( pPrevTrunk ){
        /* EVIDENCE-OF: R-01506-11053 The first integer on a freelist trunk page
        ** is the page number of the next freelist trunk page in the list or
        ** zero if this is the last freelist trunk page. */
        iTrunk = get4byte(&pPrevTrunk->aData[0]);
      }else{
        /* EVIDENCE-OF: R-59841-13798 The 4-byte big-endian integer at offset 32
        ** stores the page number of the first page of the freelist, or zero if
        ** the freelist is empty. */
        iTrunk = get4byte(&pPage1->aData[32]);
      }
      testcase( iTrunk==mxPage );
      if( iTrunk>mxPage || nSearch++ > n ){
        rc = SQLITE_CORRUPT_PGNO(pPrevTrunk ? pPrevTrunk->pgno : 1);
      }else{
        rc = btreeGetUnusedPage(pBt, iTrunk, &pTrunk, 0);
      }
      if( rc ){
        pTrunk = 0;
        goto end_allocate_page;
      }
      assert( pTrunk!=0 );
      assert( pTrunk->aData!=0 );
      /* EVIDENCE-OF: R-13523-04394 The second integer on a freelist trunk page
      ** is the number of leaf page pointers to follow. */
      k = get4byte(&pTrunk->aData[4]);
      if( k==0 && !searchList ){
        /* The trunk has no leaves and the list is not being searched.
        ** So extract the trunk page itself and use it as the newly
        ** allocated page */
        assert( pPrevTrunk==0 );
        rc = sqlite3PagerWrite(pTrunk->pDbPage);
        if( rc ){
          goto end_allocate_page;
        }
        *pPgno = iTrunk;
        memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4);
        *ppPage = pTrunk;
        pTrunk = 0;
        TRACE(("ALLOCATE: %u trunk - %u free pages left\n", *pPgno, n-1));
      }else if( k>(u32)(pBt->usableSize/4 - 2) ){
        /* Value of k is out of range.  Database corruption */
        rc = SQLITE_CORRUPT_PGNO(iTrunk);
        goto end_allocate_page;
#ifndef SQLITE_OMIT_AUTOVACUUM
      }else if( searchList
            && (nearby==iTrunk || (iTrunk<nearby && eMode==BTALLOC_LE))
      ){
        /* The list is being searched and this trunk page is the page
        ** to allocate, regardless of whether it has leaves.
        */
        *pPgno = iTrunk;
        *ppPage = pTrunk;
        searchList = 0;
        rc = sqlite3PagerWrite(pTrunk->pDbPage);
        if( rc ){
          goto end_allocate_page;
        }
        if( k==0 ){
          if( !pPrevTrunk ){
            memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4);
          }else{
            rc = sqlite3PagerWrite(pPrevTrunk->pDbPage);
            if( rc!=SQLITE_OK ){
              goto end_allocate_page;
            }
            memcpy(&pPrevTrunk->aData[0], &pTrunk->aData[0], 4);
          }
        }else{
          /* The trunk page is required by the caller but it contains
          ** pointers to free-list leaves. The first leaf becomes a trunk
          ** page in this case.
          */
          MemPage *pNewTrunk;
          Pgno iNewTrunk = get4byte(&pTrunk->aData[8]);
          if( iNewTrunk>mxPage ){
            rc = SQLITE_CORRUPT_PGNO(iTrunk);
            goto end_allocate_page;
          }
          testcase( iNewTrunk==mxPage );
          rc = btreeGetUnusedPage(pBt, iNewTrunk, &pNewTrunk, 0);
          if( rc!=SQLITE_OK ){
            goto end_allocate_page;
          }
          rc = sqlite3PagerWrite(pNewTrunk->pDbPage);
          if( rc!=SQLITE_OK ){
            releasePage(pNewTrunk);
            goto end_allocate_page;
          }
          memcpy(&pNewTrunk->aData[0], &pTrunk->aData[0], 4);
          put4byte(&pNewTrunk->aData[4], k-1);
          memcpy(&pNewTrunk->aData[8], &pTrunk->aData[12], (k-1)*4);
          releasePage(pNewTrunk);
          if( !pPrevTrunk ){
            assert( sqlite3PagerIswriteable(pPage1->pDbPage) );
            put4byte(&pPage1->aData[32], iNewTrunk);
          }else{
            rc = sqlite3PagerWrite(pPrevTrunk->pDbPage);
            if( rc ){
              goto end_allocate_page;
            }
            put4byte(&pPrevTrunk->aData[0], iNewTrunk);
          }
        }
        pTrunk = 0;
        TRACE(("ALLOCATE: %u trunk - %u free pages left\n", *pPgno, n-1));
#endif
      }else if( k>0 ){
        /* Extract a leaf from the trunk */
        u32 closest;
        Pgno iPage;
        unsigned char *aData = pTrunk->aData;
        if( nearby>0 ){
          u32 i;
          closest = 0;
          if( eMode==BTALLOC_LE ){
            for(i=0; i<k; i++){
              iPage = get4byte(&aData[8+i*4]);
              if( iPage<=nearby ){
                closest = i;
                break;
              }
            }
          }else{
            int dist;
            dist = sqlite3AbsInt32(get4byte(&aData[8]) - nearby);
            for(i=1; i<k; i++){
              int d2 = sqlite3AbsInt32(get4byte(&aData[8+i*4]) - nearby);
              if( d2<dist ){
                closest = i;
                dist = d2;
              }
            }
          }
        }else{
          closest = 0;
        }

        iPage = get4byte(&aData[8+closest*4]);
        testcase( iPage==mxPage );
        if( iPage>mxPage || iPage<2 ){
          rc = SQLITE_CORRUPT_PGNO(iTrunk);
          goto end_allocate_page;
        }
        testcase( iPage==mxPage );
        if( !searchList
         || (iPage==nearby || (iPage<nearby && eMode==BTALLOC_LE))
        ){
          int noContent;
          *pPgno = iPage;
          TRACE(("ALLOCATE: %u was leaf %u of %u on trunk %u"
                 ": %u more free pages\n",
                 *pPgno, closest+1, k, pTrunk->pgno, n-1));
          rc = sqlite3PagerWrite(pTrunk->pDbPage);
          if( rc ) goto end_allocate_page;
          if( closest<k-1 ){
            memcpy(&aData[8+closest*4], &aData[4+k*4], 4);
          }
          put4byte(&aData[4], k-1);
          noContent = !btreeGetHasContent(pBt, *pPgno)? PAGER_GET_NOCONTENT : 0;
          rc = btreeGetUnusedPage(pBt, *pPgno, ppPage, noContent);
          if( rc==SQLITE_OK ){
            rc = sqlite3PagerWrite((*ppPage)->pDbPage);
            if( rc!=SQLITE_OK ){
              releasePage(*ppPage);
              *ppPage = 0;
            }
          }
          searchList = 0;
        }
      }
      releasePage(pPrevTrunk);
      pPrevTrunk = 0;
    }while( searchList );
  }else{
    /* There are no pages on the freelist, so append a new page to the
    ** database image.
    **
    ** Normally, new pages allocated by this block can be requested from the
    ** pager layer with the 'no-content' flag set. This prevents the pager
    ** from trying to read the pages content from disk. However, if the
    ** current transaction has already run one or more incremental-vacuum
    ** steps, then the page we are about to allocate may contain content
    ** that is required in the event of a rollback. In this case, do
    ** not set the no-content flag. This causes the pager to load and journal
    ** the current page content before overwriting it.
    **
    ** Note that the pager will not actually attempt to load or journal
    ** content for any page that really does lie past the end of the database
    ** file on disk. So the effects of disabling the no-content optimization
    ** here are confined to those pages that lie between the end of the
    ** database image and the end of the database file.
    */
    int bNoContent = (0==IfNotOmitAV(pBt->bDoTruncate))? PAGER_GET_NOCONTENT:0;

    rc = sqlite3PagerWrite(pBt->pPage1->pDbPage);
    if( rc ) return rc;
    pBt->nPage++;
    if( pBt->nPage==PENDING_BYTE_PAGE(pBt) ) pBt->nPage++;

#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum && PTRMAP_ISPAGE(pBt, pBt->nPage) ){
      /* If *pPgno refers to a pointer-map page, allocate two new pages
      ** at the end of the file instead of one. The first allocated page
      ** becomes a new pointer-map page, the second is used by the caller.
      */
      MemPage *pPg = 0;
      TRACE(("ALLOCATE: %u from end of file (pointer-map page)\n", pBt->nPage));
      assert( pBt->nPage!=PENDING_BYTE_PAGE(pBt) );
      rc = btreeGetUnusedPage(pBt, pBt->nPage, &pPg, bNoContent);
      if( rc==SQLITE_OK ){
        rc = sqlite3PagerWrite(pPg->pDbPage);
        releasePage(pPg);
      }
      if( rc ) return rc;
      pBt->nPage++;
      if( pBt->nPage==PENDING_BYTE_PAGE(pBt) ){ pBt->nPage++; }
    }
#endif
    put4byte(28 + (u8*)pBt->pPage1->aData, pBt->nPage);
    *pPgno = pBt->nPage;

    assert( *pPgno!=PENDING_BYTE_PAGE(pBt) );
    rc = btreeGetUnusedPage(pBt, *pPgno, ppPage, bNoContent);
    if( rc ) return rc;
    rc = sqlite3PagerWrite((*ppPage)->pDbPage);
    if( rc!=SQLITE_OK ){
      releasePage(*ppPage);
      *ppPage = 0;
    }
    TRACE(("ALLOCATE: %u from end of file\n", *pPgno));
  }

  assert( CORRUPT_DB || *pPgno!=PENDING_BYTE_PAGE(pBt) );

end_allocate_page:
  releasePage(pTrunk);
  releasePage(pPrevTrunk);
  assert( rc!=SQLITE_OK || sqlite3PagerPageRefcount((*ppPage)->pDbPage)<=1 );
  assert( rc!=SQLITE_OK || (*ppPage)->isInit==0 );
  return rc;
}

/*
** This function is used to add page iPage to the database file free-list.
** It is assumed that the page is not already a part of the free-list.
**
** The value passed as the second argument to this function is optional.
** If the caller happens to have a pointer to the MemPage object
** corresponding to page iPage handy, it may pass it as the second value.
** Otherwise, it may pass NULL.
**
** If a pointer to a MemPage object is passed as the second argument,
** its reference count is not altered by this function.
*/
static int freePage2(BtShared *pBt, MemPage *pMemPage, Pgno iPage){
  MemPage *pTrunk = 0;                /* Free-list trunk page */
  Pgno iTrunk = 0;                    /* Page number of free-list trunk page */
  MemPage *pPage1 = pBt->pPage1;      /* Local reference to page 1 */
  MemPage *pPage;                     /* Page being freed. May be NULL. */
  int rc;                             /* Return Code */
  u32 nFree;                          /* Initial number of pages on free-list */

  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( CORRUPT_DB || iPage>1 );
  assert( !pMemPage || pMemPage->pgno==iPage );

  if( iPage<2 || iPage>pBt->nPage ){
    return SQLITE_CORRUPT_BKPT;
  }
  if( pMemPage ){
    pPage = pMemPage;
    sqlite3PagerRef(pPage->pDbPage);
  }else{
    pPage = btreePageLookup(pBt, iPage);
  }

  /* Increment the free page count on pPage1 */
  rc = sqlite3PagerWrite(pPage1->pDbPage);
  if( rc ) goto freepage_out;
  nFree = get4byte(&pPage1->aData[36]);
  put4byte(&pPage1->aData[36], nFree+1);

  if( pBt->btsFlags & BTS_SECURE_DELETE ){
    /* If the secure_delete option is enabled, then
    ** always fully overwrite deleted information with zeros.
    */
    if( (!pPage && ((rc = btreeGetPage(pBt, iPage, &pPage, 0))!=0) )
     ||            ((rc = sqlite3PagerWrite(pPage->pDbPage))!=0)
    ){
      goto freepage_out;
    }
    memset(pPage->aData, 0, pPage->pBt->pageSize);
  }

  /* If the database supports auto-vacuum, write an entry in the pointer-map
  ** to indicate that the page is free.
  */
  if( ISAUTOVACUUM(pBt) ){
    ptrmapPut(pBt, iPage, PTRMAP_FREEPAGE, 0, &rc);
    if( rc ) goto freepage_out;
  }

  /* Now manipulate the actual database free-list structure. There are two
  ** possibilities. If the free-list is currently empty, or if the first
  ** trunk page in the free-list is full, then this page will become a
  ** new free-list trunk page. Otherwise, it will become a leaf of the
  ** first trunk page in the current free-list. This block tests if it
  ** is possible to add the page as a new free-list leaf.
  */
  if( nFree!=0 ){
    u32 nLeaf;                /* Initial number of leaf cells on trunk page */

    iTrunk = get4byte(&pPage1->aData[32]);
    if( iTrunk>btreePagecount(pBt) ){
      rc = SQLITE_CORRUPT_BKPT;
      goto freepage_out;
    }
    rc = btreeGetPage(pBt, iTrunk, &pTrunk, 0);
    if( rc!=SQLITE_OK ){
      goto freepage_out;
    }

    nLeaf = get4byte(&pTrunk->aData[4]);
    assert( pBt->usableSize>32 );
    if( nLeaf > (u32)pBt->usableSize/4 - 2 ){
      rc = SQLITE_CORRUPT_BKPT;
      goto freepage_out;
    }
    if( nLeaf < (u32)pBt->usableSize/4 - 8 ){
      /* In this case there is room on the trunk page to insert the page
      ** being freed as a new leaf.
      **
      ** Note that the trunk page is not really full until it contains
      ** usableSize/4 - 2 entries, not usableSize/4 - 8 entries as we have
      ** coded.  But due to a coding error in versions of SQLite prior to
      ** 3.6.0, databases with freelist trunk pages holding more than
      ** usableSize/4 - 8 entries will be reported as corrupt.  In order
      ** to maintain backwards compatibility with older versions of SQLite,
      ** we will continue to restrict the number of entries to usableSize/4 - 8
      ** for now.  At some point in the future (once everyone has upgraded
      ** to 3.6.0 or later) we should consider fixing the conditional above
      ** to read "usableSize/4-2" instead of "usableSize/4-8".
      **
      ** EVIDENCE-OF: R-19920-11576 However, newer versions of SQLite still
      ** avoid using the last six entries in the freelist trunk page array in
      ** order that database files created by newer versions of SQLite can be
      ** read by older versions of SQLite.
      */
      rc = sqlite3PagerWrite(pTrunk->pDbPage);
      if( rc==SQLITE_OK ){
        put4byte(&pTrunk->aData[4], nLeaf+1);
        put4byte(&pTrunk->aData[8+nLeaf*4], iPage);
        if( pPage && (pBt->btsFlags & BTS_SECURE_DELETE)==0 ){
          sqlite3PagerDontWrite(pPage->pDbPage);
        }
        rc = btreeSetHasContent(pBt, iPage);
      }
      TRACE(("FREE-PAGE: %u leaf on trunk page %u\n",pPage->pgno,pTrunk->pgno));
      goto freepage_out;
    }
  }

  /* If control flows to this point, then it was not possible to add the
  ** the page being freed as a leaf page of the first trunk in the free-list.
  ** Possibly because the free-list is empty, or possibly because the
  ** first trunk in the free-list is full. Either way, the page being freed
  ** will become the new first trunk page in the free-list.
  */
  if( pPage==0 && SQLITE_OK!=(rc = btreeGetPage(pBt, iPage, &pPage, 0)) ){
    goto freepage_out;
  }
  rc = sqlite3PagerWrite(pPage->pDbPage);
  if( rc!=SQLITE_OK ){
    goto freepage_out;
  }
  put4byte(pPage->aData, iTrunk);
  put4byte(&pPage->aData[4], 0);
  put4byte(&pPage1->aData[32], iPage);
  TRACE(("FREE-PAGE: %u new trunk page replacing %u\n", pPage->pgno, iTrunk));

freepage_out:
  if( pPage ){
    pPage->isInit = 0;
  }
  releasePage(pPage);
  releasePage(pTrunk);
  return rc;
}
static void freePage(MemPage *pPage, int *pRC){
  if( (*pRC)==SQLITE_OK ){
    *pRC = freePage2(pPage->pBt, pPage, pPage->pgno);
  }
}

/*
** Free the overflow pages associated with the given Cell.
*/
static SQLITE_NOINLINE int clearCellOverflow(
  MemPage *pPage,          /* The page that contains the Cell */
  unsigned char *pCell,    /* First byte of the Cell */
  CellInfo *pInfo          /* Size information about the cell */
){
  BtShared *pBt;
  Pgno ovflPgno;
  int rc;
  int nOvfl;
  u32 ovflPageSize;

  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pInfo->nLocal!=pInfo->nPayload );
  testcase( pCell + pInfo->nSize == pPage->aDataEnd );
  testcase( pCell + (pInfo->nSize-1) == pPage->aDataEnd );
  if( pCell + pInfo->nSize > pPage->aDataEnd ){
    /* Cell extends past end of page */
    return SQLITE_CORRUPT_PAGE(pPage);
  }
  ovflPgno = get4byte(pCell + pInfo->nSize - 4);
  pBt = pPage->pBt;
  assert( pBt->usableSize > 4 );
  ovflPageSize = pBt->usableSize - 4;
  nOvfl = (pInfo->nPayload - pInfo->nLocal + ovflPageSize - 1)/ovflPageSize;
  assert( nOvfl>0 ||
    (CORRUPT_DB && (pInfo->nPayload + ovflPageSize)<ovflPageSize)
  );
  while( nOvfl-- ){
    Pgno iNext = 0;
    MemPage *pOvfl = 0;
    if( ovflPgno<2 || ovflPgno>btreePagecount(pBt) ){
      /* 0 is not a legal page number and page 1 cannot be an
      ** overflow page. Therefore if ovflPgno<2 or past the end of the
      ** file the database must be corrupt. */
      return SQLITE_CORRUPT_BKPT;
    }
    if( nOvfl ){
      rc = getOverflowPage(pBt, ovflPgno, &pOvfl, &iNext);
      if( rc ) return rc;
    }

    if( ( pOvfl || ((pOvfl = btreePageLookup(pBt, ovflPgno))!=0) )
     && sqlite3PagerPageRefcount(pOvfl->pDbPage)!=1
    ){
      /* There is no reason any cursor should have an outstanding reference
      ** to an overflow page belonging to a cell that is being deleted/updated.
      ** So if there exists more than one reference to this page, then it
      ** must not really be an overflow page and the database must be corrupt.
      ** It is helpful to detect this before calling freePage2(), as
      ** freePage2() may zero the page contents if secure-delete mode is
      ** enabled. If this 'overflow' page happens to be a page that the
      ** caller is iterating through or using in some other way, this
      ** can be problematic.
      */
      rc = SQLITE_CORRUPT_BKPT;
    }else{
      rc = freePage2(pBt, pOvfl, ovflPgno);
    }

    if( pOvfl ){
      sqlite3PagerUnref(pOvfl->pDbPage);
    }
    if( rc ) return rc;
    ovflPgno = iNext;
  }
  return SQLITE_OK;
}

/* Call xParseCell to compute the size of a cell.  If the cell contains
** overflow, then invoke cellClearOverflow to clear out that overflow.
** Store the result code (SQLITE_OK or some error code) in rc.
**
** Implemented as macro to force inlining for performance.
*/
#define BTREE_CLEAR_CELL(rc, pPage, pCell, sInfo)   \
  pPage->xParseCell(pPage, pCell, &sInfo);          \
  if( sInfo.nLocal!=sInfo.nPayload ){               \
    rc = clearCellOverflow(pPage, pCell, &sInfo);   \
  }else{                                            \
    rc = SQLITE_OK;                                 \
  }


/*
** Create the byte sequence used to represent a cell on page pPage
** and write that byte sequence into pCell[].  Overflow pages are
** allocated and filled in as necessary.  The calling procedure
** is responsible for making sure sufficient space has been allocated
** for pCell[].
**
** Note that pCell does not necessary need to point to the pPage->aData
** area.  pCell might point to some temporary storage.  The cell will
** be constructed in this temporary area then copied into pPage->aData
** later.
*/
static int fillInCell(
  MemPage *pPage,                /* The page that contains the cell */
  unsigned char *pCell,          /* Complete text of the cell */
  const BtreePayload *pX,        /* Payload with which to construct the cell */
  int *pnSize                    /* Write cell size here */
){
  int nPayload;
  const u8 *pSrc;
  int nSrc, n, rc, mn;
  int spaceLeft;
  MemPage *pToRelease;
  unsigned char *pPrior;
  unsigned char *pPayload;
  BtShared *pBt;
  Pgno pgnoOvfl;
  int nHeader;

  assert( sqlite3_mutex_held(pPage->pBt->mutex) );

  /* pPage is not necessarily writeable since pCell might be auxiliary
  ** buffer space that is separate from the pPage buffer area */
  assert( pCell<pPage->aData || pCell>=&pPage->aData[pPage->pBt->pageSize]
            || sqlite3PagerIswriteable(pPage->pDbPage) );

  /* Fill in the header. */
  nHeader = pPage->childPtrSize;
  if( pPage->intKey ){
    nPayload = pX->nData + pX->nZero;
    pSrc = pX->pData;
    nSrc = pX->nData;
    assert( pPage->intKeyLeaf ); /* fillInCell() only called for leaves */
    nHeader += putVarint32(&pCell[nHeader], nPayload);
    nHeader += putVarint(&pCell[nHeader], *(u64*)&pX->nKey);
  }else{
    assert( pX->nKey<=0x7fffffff && pX->pKey!=0 );
    nSrc = nPayload = (int)pX->nKey;
    pSrc = pX->pKey;
    nHeader += putVarint32(&pCell[nHeader], nPayload);
  }
 
  /* Fill in the payload */
  pPayload = &pCell[nHeader];
  if( nPayload<=pPage->maxLocal ){
    /* This is the common case where everything fits on the btree page
    ** and no overflow pages are required. */
    n = nHeader + nPayload;
    testcase( n==3 );
    testcase( n==4 );
    if( n<4 ){
      n = 4;
      pPayload[nPayload] = 0;
    }
    *pnSize = n;
    assert( nSrc<=nPayload );
    testcase( nSrc<nPayload );
    memcpy(pPayload, pSrc, nSrc);
    memset(pPayload+nSrc, 0, nPayload-nSrc);
    return SQLITE_OK;
  }

  /* If we reach this point, it means that some of the content will need
  ** to spill onto overflow pages.
  */
  mn = pPage->minLocal;
  n = mn + (nPayload - mn) % (pPage->pBt->usableSize - 4);
  testcase( n==pPage->maxLocal );
  testcase( n==pPage->maxLocal+1 );
  if( n > pPage->maxLocal ) n = mn;
  spaceLeft = n;
  *pnSize = n + nHeader + 4;
  pPrior = &pCell[nHeader+n];
  pToRelease = 0;
  pgnoOvfl = 0;
  pBt = pPage->pBt;

  /* At this point variables should be set as follows:
  **
  **   nPayload           Total payload size in bytes
  **   pPayload           Begin writing payload here
  **   spaceLeft          Space available at pPayload.  If nPayload>spaceLeft,
  **                      that means content must spill into overflow pages.
  **   *pnSize            Size of the local cell (not counting overflow pages)
  **   pPrior             Where to write the pgno of the first overflow page
  **
  ** Use a call to btreeParseCellPtr() to verify that the values above
  ** were computed correctly.
  */
#ifdef SQLITE_DEBUG
  {
    CellInfo info;
    pPage->xParseCell(pPage, pCell, &info);
    assert( nHeader==(int)(info.pPayload - pCell) );
    assert( info.nKey==pX->nKey );
    assert( *pnSize == info.nSize );
    assert( spaceLeft == info.nLocal );
  }
#endif

  /* Write the payload into the local Cell and any extra into overflow pages */
  while( 1 ){
    n = nPayload;
    if( n>spaceLeft ) n = spaceLeft;

    /* If pToRelease is not zero than pPayload points into the data area
    ** of pToRelease.  Make sure pToRelease is still writeable. */
    assert( pToRelease==0 || sqlite3PagerIswriteable(pToRelease->pDbPage) );

    /* If pPayload is part of the data area of pPage, then make sure pPage
    ** is still writeable */
    assert( pPayload<pPage->aData || pPayload>=&pPage->aData[pBt->pageSize]
            || sqlite3PagerIswriteable(pPage->pDbPage) );

    if( nSrc>=n ){
      memcpy(pPayload, pSrc, n);
    }else if( nSrc>0 ){
      n = nSrc;
      memcpy(pPayload, pSrc, n);
    }else{
      memset(pPayload, 0, n);
    }
    nPayload -= n;
    if( nPayload<=0 ) break;
    pPayload += n;
    pSrc += n;
    nSrc -= n;
    spaceLeft -= n;
    if( spaceLeft==0 ){
      MemPage *pOvfl = 0;
#ifndef SQLITE_OMIT_AUTOVACUUM
      Pgno pgnoPtrmap = pgnoOvfl; /* Overflow page pointer-map entry page */
      if( pBt->autoVacuum ){
        do{
          pgnoOvfl++;
        } while(
          PTRMAP_ISPAGE(pBt, pgnoOvfl) || pgnoOvfl==PENDING_BYTE_PAGE(pBt)
        );
      }
#endif
      rc = allocateBtreePage(pBt, &pOvfl, &pgnoOvfl, pgnoOvfl, 0);
#ifndef SQLITE_OMIT_AUTOVACUUM
      /* If the database supports auto-vacuum, and the second or subsequent
      ** overflow page is being allocated, add an entry to the pointer-map
      ** for that page now.
      **
      ** If this is the first overflow page, then write a partial entry
      ** to the pointer-map. If we write nothing to this pointer-map slot,
      ** then the optimistic overflow chain processing in clearCell()
      ** may misinterpret the uninitialized values and delete the
      ** wrong pages from the database.
      */
      if( pBt->autoVacuum && rc==SQLITE_OK ){
        u8 eType = (pgnoPtrmap?PTRMAP_OVERFLOW2:PTRMAP_OVERFLOW1);
        ptrmapPut(pBt, pgnoOvfl, eType, pgnoPtrmap, &rc);
        if( rc ){
          releasePage(pOvfl);
        }
      }
#endif
      if( rc ){
        releasePage(pToRelease);
        return rc;
      }

      /* If pToRelease is not zero than pPrior points into the data area
      ** of pToRelease.  Make sure pToRelease is still writeable. */
      assert( pToRelease==0 || sqlite3PagerIswriteable(pToRelease->pDbPage) );

      /* If pPrior is part of the data area of pPage, then make sure pPage
      ** is still writeable */
      assert( pPrior<pPage->aData || pPrior>=&pPage->aData[pBt->pageSize]
            || sqlite3PagerIswriteable(pPage->pDbPage) );

      put4byte(pPrior, pgnoOvfl);
      releasePage(pToRelease);
      pToRelease = pOvfl;
      pPrior = pOvfl->aData;
      put4byte(pPrior, 0);
      pPayload = &pOvfl->aData[4];
      spaceLeft = pBt->usableSize - 4;
    }
  }
  releasePage(pToRelease);
  return SQLITE_OK;
}

/*
** Remove the i-th cell from pPage.  This routine effects pPage only.
** The cell content is not freed or deallocated.  It is assumed that
** the cell content has been copied someplace else.  This routine just
** removes the reference to the cell from pPage.
**
** "sz" must be the number of bytes in the cell.
*/
static void dropCell(MemPage *pPage, int idx, int sz, int *pRC){
  u32 pc;         /* Offset to cell content of cell being deleted */
  u8 *data;       /* pPage->aData */
  u8 *ptr;        /* Used to move bytes around within data[] */
  int rc;         /* The return code */
  int hdr;        /* Beginning of the header.  0 most pages.  100 page 1 */

  if( *pRC ) return;
  assert( idx>=0 );
  assert( idx<pPage->nCell );
  assert( CORRUPT_DB || sz==cellSize(pPage, idx) );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( pPage->nFree>=0 );
  data = pPage->aData;
  ptr = &pPage->aCellIdx[2*idx];
  assert( pPage->pBt->usableSize > (u32)(ptr-data) );
  pc = get2byte(ptr);
  hdr = pPage->hdrOffset;
  testcase( pc==(u32)get2byte(&data[hdr+5]) );
  testcase( pc+sz==pPage->pBt->usableSize );
  if( pc+sz > pPage->pBt->usableSize ){
    *pRC = SQLITE_CORRUPT_BKPT;
    return;
  }
  rc = freeSpace(pPage, pc, sz);
  if( rc ){
    *pRC = rc;
    return;
  }
  pPage->nCell--;
  if( pPage->nCell==0 ){
    memset(&data[hdr+1], 0, 4);
    data[hdr+7] = 0;
    put2byte(&data[hdr+5], pPage->pBt->usableSize);
    pPage->nFree = pPage->pBt->usableSize - pPage->hdrOffset
                       - pPage->childPtrSize - 8;
  }else{
    memmove(ptr, ptr+2, 2*(pPage->nCell - idx));
    put2byte(&data[hdr+3], pPage->nCell);
    pPage->nFree += 2;
  }
}

/*
** Insert a new cell on pPage at cell index "i".  pCell points to the
** content of the cell.
**
** If the cell content will fit on the page, then put it there.  If it
** will not fit, then make a copy of the cell content into pTemp if
** pTemp is not null.  Regardless of pTemp, allocate a new entry
** in pPage->apOvfl[] and make it point to the cell content (either
** in pTemp or the original pCell) and also record its index.
** Allocating a new entry in pPage->aCell[] implies that
** pPage->nOverflow is incremented.
**
** The insertCellFast() routine below works exactly the same as
** insertCell() except that it lacks the pTemp and iChild parameters
** which are assumed zero.  Other than that, the two routines are the
** same.
**
** Fixes or enhancements to this routine should be reflected in
** insertCellFast()!
*/
static int insertCell(
  MemPage *pPage,   /* Page into which we are copying */
  int i,            /* New cell becomes the i-th cell of the page */
  u8 *pCell,        /* Content of the new cell */
  int sz,           /* Bytes of content in pCell */
  u8 *pTemp,        /* Temp storage space for pCell, if needed */
  Pgno iChild       /* If non-zero, replace first 4 bytes with this value */
){
  int idx = 0;      /* Where to write new cell content in data[] */
  int j;            /* Loop counter */
  u8 *data;         /* The content of the whole page */
  u8 *pIns;         /* The point in pPage->aCellIdx[] where no cell inserted */

  assert( i>=0 && i<=pPage->nCell+pPage->nOverflow );
  assert( MX_CELL(pPage->pBt)<=10921 );
  assert( pPage->nCell<=MX_CELL(pPage->pBt) || CORRUPT_DB );
  assert( pPage->nOverflow<=ArraySize(pPage->apOvfl) );
  assert( ArraySize(pPage->apOvfl)==ArraySize(pPage->aiOvfl) );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( sz==pPage->xCellSize(pPage, pCell) || CORRUPT_DB );
  assert( pPage->nFree>=0 );
  assert( iChild>0 );
  if( pPage->nOverflow || sz+2>pPage->nFree ){
    if( pTemp ){
      memcpy(pTemp, pCell, sz);
      pCell = pTemp;
    }
    put4byte(pCell, iChild);
    j = pPage->nOverflow++;
    /* Comparison against ArraySize-1 since we hold back one extra slot
    ** as a contingency.  In other words, never need more than 3 overflow
    ** slots but 4 are allocated, just to be safe. */
    assert( j < ArraySize(pPage->apOvfl)-1 );
    pPage->apOvfl[j] = pCell;
    pPage->aiOvfl[j] = (u16)i;

    /* When multiple overflows occur, they are always sequential and in
    ** sorted order.  This invariants arise because multiple overflows can
    ** only occur when inserting divider cells into the parent page during
    ** balancing, and the dividers are adjacent and sorted.
    */
    assert( j==0 || pPage->aiOvfl[j-1]<(u16)i ); /* Overflows in sorted order */
    assert( j==0 || i==pPage->aiOvfl[j-1]+1 );   /* Overflows are sequential */
  }else{
    int rc = sqlite3PagerWrite(pPage->pDbPage);
    if( NEVER(rc!=SQLITE_OK) ){
      return rc;
    }
    assert( sqlite3PagerIswriteable(pPage->pDbPage) );
    data = pPage->aData;
    assert( &data[pPage->cellOffset]==pPage->aCellIdx );
    rc = allocateSpace(pPage, sz, &idx);
    if( rc ){ return rc; }
    /* The allocateSpace() routine guarantees the following properties
    ** if it returns successfully */
    assert( idx >= 0 );
    assert( idx >= pPage->cellOffset+2*pPage->nCell+2 || CORRUPT_DB );
    assert( idx+sz <= (int)pPage->pBt->usableSize );
    pPage->nFree -= (u16)(2 + sz);
    /* In a corrupt database where an entry in the cell index section of
    ** a btree page has a value of 3 or less, the pCell value might point
    ** as many as 4 bytes in front of the start of the aData buffer for
    ** the source page.  Make sure this does not cause problems by not
    ** reading the first 4 bytes */
    memcpy(&data[idx+4], pCell+4, sz-4);
    put4byte(&data[idx], iChild);
    pIns = pPage->aCellIdx + i*2;
    memmove(pIns+2, pIns, 2*(pPage->nCell - i));
    put2byte(pIns, idx);
    pPage->nCell++;
    /* increment the cell count */
    if( (++data[pPage->hdrOffset+4])==0 ) data[pPage->hdrOffset+3]++;
    assert( get2byte(&data[pPage->hdrOffset+3])==pPage->nCell || CORRUPT_DB );
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pPage->pBt->autoVacuum ){
      int rc2 = SQLITE_OK;
      /* The cell may contain a pointer to an overflow page. If so, write
      ** the entry for the overflow page into the pointer map.
      */
      ptrmapPutOvflPtr(pPage, pPage, pCell, &rc2);
      if( rc2 ) return rc2;
    }
#endif
  }
  return SQLITE_OK;
}

/*
** This variant of insertCell() assumes that the pTemp and iChild
** parameters are both zero.  Use this variant in sqlite3BtreeInsert()
** for performance improvement, and also so that this variant is only
** called from that one place, and is thus inlined, and thus runs must
** faster.
**
** Fixes or enhancements to this routine should be reflected into
** the insertCell() routine.
*/
static int insertCellFast(
  MemPage *pPage,   /* Page into which we are copying */
  int i,            /* New cell becomes the i-th cell of the page */
  u8 *pCell,        /* Content of the new cell */
  int sz            /* Bytes of content in pCell */
){
  int idx = 0;      /* Where to write new cell content in data[] */
  int j;            /* Loop counter */
  u8 *data;         /* The content of the whole page */
  u8 *pIns;         /* The point in pPage->aCellIdx[] where no cell inserted */

  assert( i>=0 && i<=pPage->nCell+pPage->nOverflow );
  assert( MX_CELL(pPage->pBt)<=10921 );
  assert( pPage->nCell<=MX_CELL(pPage->pBt) || CORRUPT_DB );
  assert( pPage->nOverflow<=ArraySize(pPage->apOvfl) );
  assert( ArraySize(pPage->apOvfl)==ArraySize(pPage->aiOvfl) );
  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( sz==pPage->xCellSize(pPage, pCell) || CORRUPT_DB );
  assert( pPage->nFree>=0 );
  assert( pPage->nOverflow==0 );
  if( sz+2>pPage->nFree ){
    j = pPage->nOverflow++;
    /* Comparison against ArraySize-1 since we hold back one extra slot
    ** as a contingency.  In other words, never need more than 3 overflow
    ** slots but 4 are allocated, just to be safe. */
    assert( j < ArraySize(pPage->apOvfl)-1 );
    pPage->apOvfl[j] = pCell;
    pPage->aiOvfl[j] = (u16)i;

    /* When multiple overflows occur, they are always sequential and in
    ** sorted order.  This invariants arise because multiple overflows can
    ** only occur when inserting divider cells into the parent page during
    ** balancing, and the dividers are adjacent and sorted.
    */
    assert( j==0 || pPage->aiOvfl[j-1]<(u16)i ); /* Overflows in sorted order */
    assert( j==0 || i==pPage->aiOvfl[j-1]+1 );   /* Overflows are sequential */
  }else{
    int rc = sqlite3PagerWrite(pPage->pDbPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    assert( sqlite3PagerIswriteable(pPage->pDbPage) );
    data = pPage->aData;
    assert( &data[pPage->cellOffset]==pPage->aCellIdx );
    rc = allocateSpace(pPage, sz, &idx);
    if( rc ){ return rc; }
    /* The allocateSpace() routine guarantees the following properties
    ** if it returns successfully */
    assert( idx >= 0 );
    assert( idx >= pPage->cellOffset+2*pPage->nCell+2 || CORRUPT_DB );
    assert( idx+sz <= (int)pPage->pBt->usableSize );
    pPage->nFree -= (u16)(2 + sz);
    memcpy(&data[idx], pCell, sz);
    pIns = pPage->aCellIdx + i*2;
    memmove(pIns+2, pIns, 2*(pPage->nCell - i));
    put2byte(pIns, idx);
    pPage->nCell++;
    /* increment the cell count */
    if( (++data[pPage->hdrOffset+4])==0 ) data[pPage->hdrOffset+3]++;
    assert( get2byte(&data[pPage->hdrOffset+3])==pPage->nCell || CORRUPT_DB );
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pPage->pBt->autoVacuum ){
      int rc2 = SQLITE_OK;
      /* The cell may contain a pointer to an overflow page. If so, write
      ** the entry for the overflow page into the pointer map.
      */
      ptrmapPutOvflPtr(pPage, pPage, pCell, &rc2);
      if( rc2 ) return rc2;
    }
#endif
  }
  return SQLITE_OK;
}

/*
** The following parameters determine how many adjacent pages get involved
** in a balancing operation.  NN is the number of neighbors on either side
** of the page that participate in the balancing operation.  NB is the
** total number of pages that participate, including the target page and
** NN neighbors on either side.
**
** The minimum value of NN is 1 (of course).  Increasing NN above 1
** (to 2 or 3) gives a modest improvement in SELECT and DELETE performance
** in exchange for a larger degradation in INSERT and UPDATE performance.
** The value of NN appears to give the best results overall.
**
** (Later:) The description above makes it seem as if these values are
** tunable - as if you could change them and recompile and it would all work.
** But that is unlikely.  NB has been 3 since the inception of SQLite and
** we have never tested any other value.
*/
#define NN 1             /* Number of neighbors on either side of pPage */
#define NB 3             /* (NN*2+1): Total pages involved in the balance */

/*
** A CellArray object contains a cache of pointers and sizes for a
** consecutive sequence of cells that might be held on multiple pages.
**
** The cells in this array are the divider cell or cells from the pParent
** page plus up to three child pages.  There are a total of nCell cells.
**
** pRef is a pointer to one of the pages that contributes cells.  This is
** used to access information such as MemPage.intKey and MemPage.pBt->pageSize
** which should be common to all pages that contribute cells to this array.
**
** apCell[] and szCell[] hold, respectively, pointers to the start of each
** cell and the size of each cell.  Some of the apCell[] pointers might refer
** to overflow cells.  In other words, some apCel[] pointers might not point
** to content area of the pages.
**
** A szCell[] of zero means the size of that cell has not yet been computed.
**
** The cells come from as many as four different pages:
**
**             -----------
**             | Parent  |
**             -----------
**            /     |     \
**           /      |      \
**  ---------   ---------   ---------
**  |Child-1|   |Child-2|   |Child-3|
**  ---------   ---------   ---------
**
** The order of cells is in the array is for an index btree is:
**
**       1.  All cells from Child-1 in order
**       2.  The first divider cell from Parent
**       3.  All cells from Child-2 in order
**       4.  The second divider cell from Parent
**       5.  All cells from Child-3 in order
**
** For a table-btree (with rowids) the items 2 and 4 are empty because
** content exists only in leaves and there are no divider cells.
**
** For an index btree, the apEnd[] array holds pointer to the end of page
** for Child-1, the Parent, Child-2, the Parent (again), and Child-3,
** respectively. The ixNx[] array holds the number of cells contained in
** each of these 5 stages, and all stages to the left.  Hence:
**
**    ixNx[0] = Number of cells in Child-1.
**    ixNx[1] = Number of cells in Child-1 plus 1 for first divider.
**    ixNx[2] = Number of cells in Child-1 and Child-2 + 1 for 1st divider.
**    ixNx[3] = Number of cells in Child-1 and Child-2 + both divider cells
**    ixNx[4] = Total number of cells.
**
** For a table-btree, the concept is similar, except only apEnd[0]..apEnd[2]
** are used and they point to the leaf pages only, and the ixNx value are:
**
**    ixNx[0] = Number of cells in Child-1.
**    ixNx[1] = Number of cells in Child-1 and Child-2.
**    ixNx[2] = Total number of cells.
**
** Sometimes when deleting, a child page can have zero cells.  In those
** cases, ixNx[] entries with higher indexes, and the corresponding apEnd[]
** entries, shift down.  The end result is that each ixNx[] entry should
** be larger than the previous
*/
typedef struct CellArray CellArray;
struct CellArray {
  int nCell;              /* Number of cells in apCell[] */
  MemPage *pRef;          /* Reference page */
  u8 **apCell;            /* All cells begin balanced */
  u16 *szCell;            /* Local size of all cells in apCell[] */
  u8 *apEnd[NB*2];        /* MemPage.aDataEnd values */
  int ixNx[NB*2];         /* Index of at which we move to the next apEnd[] */
};

/*
** Make sure the cell sizes at idx, idx+1, ..., idx+N-1 have been
** computed.
*/
static void populateCellCache(CellArray *p, int idx, int N){
  MemPage *pRef = p->pRef;
  u16 *szCell = p->szCell;
  assert( idx>=0 && idx+N<=p->nCell );
  while( N>0 ){
    assert( p->apCell[idx]!=0 );
    if( szCell[idx]==0 ){
      szCell[idx] = pRef->xCellSize(pRef, p->apCell[idx]);
    }else{
      assert( CORRUPT_DB ||
              szCell[idx]==pRef->xCellSize(pRef, p->apCell[idx]) );
    }
    idx++;
    N--;
  }
}

/*
** Return the size of the Nth element of the cell array
*/
static SQLITE_NOINLINE u16 computeCellSize(CellArray *p, int N){
  assert( N>=0 && N<p->nCell );
  assert( p->szCell[N]==0 );
  p->szCell[N] = p->pRef->xCellSize(p->pRef, p->apCell[N]);
  return p->szCell[N];
}
static u16 cachedCellSize(CellArray *p, int N){
  assert( N>=0 && N<p->nCell );
  if( p->szCell[N] ) return p->szCell[N];
  return computeCellSize(p, N);
}

/*
** Array apCell[] contains pointers to nCell b-tree page cells. The
** szCell[] array contains the size in bytes of each cell. This function
** replaces the current contents of page pPg with the contents of the cell
** array.
**
** Some of the cells in apCell[] may currently be stored in pPg. This
** function works around problems caused by this by making a copy of any
** such cells before overwriting the page data.
**
** The MemPage.nFree field is invalidated by this function. It is the
** responsibility of the caller to set it correctly.
*/
static int rebuildPage(
  CellArray *pCArray,             /* Content to be added to page pPg */
  int iFirst,                     /* First cell in pCArray to use */
  int nCell,                      /* Final number of cells on page */
  MemPage *pPg                    /* The page to be reconstructed */
){
  const int hdr = pPg->hdrOffset;          /* Offset of header on pPg */
  u8 * const aData = pPg->aData;           /* Pointer to data for pPg */
  const int usableSize = pPg->pBt->usableSize;
  u8 * const pEnd = &aData[usableSize];
  int i = iFirst;                 /* Which cell to copy from pCArray*/
  u32 j;                          /* Start of cell content area */
  int iEnd = i+nCell;             /* Loop terminator */
  u8 *pCellptr = pPg->aCellIdx;
  u8 *pTmp = sqlite3PagerTempSpace(pPg->pBt->pPager);
  u8 *pData;
  int k;                          /* Current slot in pCArray->apEnd[] */
  u8 *pSrcEnd;                    /* Current pCArray->apEnd[k] value */

  assert( nCell>0 );
  assert( i<iEnd );
  j = get2byte(&aData[hdr+5]);
  if( j>(u32)usableSize ){ j = 0; }
  memcpy(&pTmp[j], &aData[j], usableSize - j);

  for(k=0; ALWAYS(k<NB*2) && pCArray->ixNx[k]<=i; k++){}
  pSrcEnd = pCArray->apEnd[k];

  pData = pEnd;
  while( 1/*exit by break*/ ){
    u8 *pCell = pCArray->apCell[i];
    u16 sz = pCArray->szCell[i];
    assert( sz>0 );
    if( SQLITE_WITHIN(pCell,aData+j,pEnd) ){
      if( ((uptr)(pCell+sz))>(uptr)pEnd ) return SQLITE_CORRUPT_BKPT;
      pCell = &pTmp[pCell - aData];
    }else if( (uptr)(pCell+sz)>(uptr)pSrcEnd
           && (uptr)(pCell)<(uptr)pSrcEnd
    ){
      return SQLITE_CORRUPT_BKPT;
    }

    pData -= sz;
    put2byte(pCellptr, (pData - aData));
    pCellptr += 2;
    if( pData < pCellptr ) return SQLITE_CORRUPT_BKPT;
    memmove(pData, pCell, sz);
    assert( sz==pPg->xCellSize(pPg, pCell) || CORRUPT_DB );
    i++;
    if( i>=iEnd ) break;
    if( pCArray->ixNx[k]<=i ){
      k++;
      pSrcEnd = pCArray->apEnd[k];
    }
  }

  /* The pPg->nFree field is now set incorrectly. The caller will fix it. */
  pPg->nCell = nCell;
  pPg->nOverflow = 0;

  put2byte(&aData[hdr+1], 0);
  put2byte(&aData[hdr+3], pPg->nCell);
  put2byte(&aData[hdr+5], pData - aData);
  aData[hdr+7] = 0x00;
  return SQLITE_OK;
}

/*
** The pCArray objects contains pointers to b-tree cells and the cell sizes.
** This function attempts to add the cells stored in the array to page pPg.
** If it cannot (because the page needs to be defragmented before the cells
** will fit), non-zero is returned. Otherwise, if the cells are added
** successfully, zero is returned.
**
** Argument pCellptr points to the first entry in the cell-pointer array
** (part of page pPg) to populate. After cell apCell[0] is written to the
** page body, a 16-bit offset is written to pCellptr. And so on, for each
** cell in the array. It is the responsibility of the caller to ensure
** that it is safe to overwrite this part of the cell-pointer array.
**
** When this function is called, *ppData points to the start of the
** content area on page pPg. If the size of the content area is extended,
** *ppData is updated to point to the new start of the content area
** before returning.
**
** Finally, argument pBegin points to the byte immediately following the
** end of the space required by this page for the cell-pointer area (for
** all cells - not just those inserted by the current call). If the content
** area must be extended to before this point in order to accommodate all
** cells in apCell[], then the cells do not fit and non-zero is returned.
*/
static int pageInsertArray(
  MemPage *pPg,                   /* Page to add cells to */
  u8 *pBegin,                     /* End of cell-pointer array */
  u8 **ppData,                    /* IN/OUT: Page content-area pointer */
  u8 *pCellptr,                   /* Pointer to cell-pointer area */
  int iFirst,                     /* Index of first cell to add */
  int nCell,                      /* Number of cells to add to pPg */
  CellArray *pCArray              /* Array of cells */
){
  int i = iFirst;                 /* Loop counter - cell index to insert */
  u8 *aData = pPg->aData;         /* Complete page */
  u8 *pData = *ppData;            /* Content area.  A subset of aData[] */
  int iEnd = iFirst + nCell;      /* End of loop. One past last cell to ins */
  int k;                          /* Current slot in pCArray->apEnd[] */
  u8 *pEnd;                       /* Maximum extent of cell data */
  assert( CORRUPT_DB || pPg->hdrOffset==0 );    /* Never called on page 1 */
  if( iEnd<=iFirst ) return 0;
  for(k=0; ALWAYS(k<NB*2) && pCArray->ixNx[k]<=i ; k++){}
  pEnd = pCArray->apEnd[k];
  while( 1 /*Exit by break*/ ){
    int sz, rc;
    u8 *pSlot;
    assert( pCArray->szCell[i]!=0 );
    sz = pCArray->szCell[i];
    if( (aData[1]==0 && aData[2]==0) || (pSlot = pageFindSlot(pPg,sz,&rc))==0 ){
      if( (pData - pBegin)<sz ) return 1;
      pData -= sz;
      pSlot = pData;
    }
    /* pSlot and pCArray->apCell[i] will never overlap on a well-formed
    ** database.  But they might for a corrupt database.  Hence use memmove()
    ** since memcpy() sends SIGABORT with overlapping buffers on OpenBSD */
    assert( (pSlot+sz)<=pCArray->apCell[i]
         || pSlot>=(pCArray->apCell[i]+sz)
         || CORRUPT_DB );
    if( (uptr)(pCArray->apCell[i]+sz)>(uptr)pEnd
     && (uptr)(pCArray->apCell[i])<(uptr)pEnd
    ){
      assert( CORRUPT_DB );
      (void)SQLITE_CORRUPT_BKPT;
      return 1;
    }
    memmove(pSlot, pCArray->apCell[i], sz);
    put2byte(pCellptr, (pSlot - aData));
    pCellptr += 2;
    i++;
    if( i>=iEnd ) break;
    if( pCArray->ixNx[k]<=i ){
      k++;
      pEnd = pCArray->apEnd[k];
    }
  }
  *ppData = pData;
  return 0;
}

/*
** The pCArray object contains pointers to b-tree cells and their sizes.
**
** This function adds the space associated with each cell in the array
** that is currently stored within the body of pPg to the pPg free-list.
** The cell-pointers and other fields of the page are not updated.
**
** This function returns the total number of cells added to the free-list.
*/
static int pageFreeArray(
  MemPage *pPg,                   /* Page to edit */
  int iFirst,                     /* First cell to delete */
  int nCell,                      /* Cells to delete */
  CellArray *pCArray              /* Array of cells */
){
  u8 * const aData = pPg->aData;
  u8 * const pEnd = &aData[pPg->pBt->usableSize];
  u8 * const pStart = &aData[pPg->hdrOffset + 8 + pPg->childPtrSize];
  int nRet = 0;
  int i, j;
  int iEnd = iFirst + nCell;
  int nFree = 0;
  int aOfst[10];
  int aAfter[10];

  for(i=iFirst; i<iEnd; i++){
    u8 *pCell = pCArray->apCell[i];
    if( SQLITE_WITHIN(pCell, pStart, pEnd) ){
      int sz;
      int iAfter;
      int iOfst;
      /* No need to use cachedCellSize() here.  The sizes of all cells that
      ** are to be freed have already been computing while deciding which
      ** cells need freeing */
      sz = pCArray->szCell[i];  assert( sz>0 );
      iOfst = (u16)(pCell - aData);
      iAfter = iOfst+sz;
      for(j=0; j<nFree; j++){
        if( aOfst[j]==iAfter ){
          aOfst[j] = iOfst;
          break;
        }else if( aAfter[j]==iOfst ){
          aAfter[j] = iAfter;
          break;
        }
      }
      if( j>=nFree ){
        if( nFree>=(int)(sizeof(aOfst)/sizeof(aOfst[0])) ){
          for(j=0; j<nFree; j++){
            freeSpace(pPg, aOfst[j], aAfter[j]-aOfst[j]);
          }
          nFree = 0;
        }
        aOfst[nFree] = iOfst;
        aAfter[nFree] = iAfter;
        if( &aData[iAfter]>pEnd ) return 0;
        nFree++;
      }
      nRet++;
    }
  }
  for(j=0; j<nFree; j++){
    freeSpace(pPg, aOfst[j], aAfter[j]-aOfst[j]);
  }
  return nRet;
}

/*
** pCArray contains pointers to and sizes of all cells in the page being
** balanced.  The current page, pPg, has pPg->nCell cells starting with
** pCArray->apCell[iOld].  After balancing, this page should hold nNew cells
** starting at apCell[iNew].
**
** This routine makes the necessary adjustments to pPg so that it contains
** the correct cells after being balanced.
**
** The pPg->nFree field is invalid when this function returns. It is the
** responsibility of the caller to set it correctly.
*/
static int editPage(
  MemPage *pPg,                   /* Edit this page */
  int iOld,                       /* Index of first cell currently on page */
  int iNew,                       /* Index of new first cell on page */
  int nNew,                       /* Final number of cells on page */
  CellArray *pCArray              /* Array of cells and sizes */
){
  u8 * const aData = pPg->aData;
  const int hdr = pPg->hdrOffset;
  u8 *pBegin = &pPg->aCellIdx[nNew * 2];
  int nCell = pPg->nCell;       /* Cells stored on pPg */
  u8 *pData;
  u8 *pCellptr;
  int i;
  int iOldEnd = iOld + pPg->nCell + pPg->nOverflow;
  int iNewEnd = iNew + nNew;

#ifdef SQLITE_DEBUG
  u8 *pTmp = sqlite3PagerTempSpace(pPg->pBt->pPager);
  memcpy(pTmp, aData, pPg->pBt->usableSize);
#endif

  /* Remove cells from the start and end of the page */
  assert( nCell>=0 );
  if( iOld<iNew ){
    int nShift = pageFreeArray(pPg, iOld, iNew-iOld, pCArray);
    if( NEVER(nShift>nCell) ) return SQLITE_CORRUPT_BKPT;
    memmove(pPg->aCellIdx, &pPg->aCellIdx[nShift*2], nCell*2);
    nCell -= nShift;
  }
  if( iNewEnd < iOldEnd ){
    int nTail = pageFreeArray(pPg, iNewEnd, iOldEnd - iNewEnd, pCArray);
    assert( nCell>=nTail );
    nCell -= nTail;
  }

  pData = &aData[get2byte(&aData[hdr+5])];
  if( pData<pBegin ) goto editpage_fail;
  if( NEVER(pData>pPg->aDataEnd) ) goto editpage_fail;

  /* Add cells to the start of the page */
  if( iNew<iOld ){
    int nAdd = MIN(nNew,iOld-iNew);
    assert( (iOld-iNew)<nNew || nCell==0 || CORRUPT_DB );
    assert( nAdd>=0 );
    pCellptr = pPg->aCellIdx;
    memmove(&pCellptr[nAdd*2], pCellptr, nCell*2);
    if( pageInsertArray(
          pPg, pBegin, &pData, pCellptr,
          iNew, nAdd, pCArray
    ) ) goto editpage_fail;
    nCell += nAdd;
  }

  /* Add any overflow cells */
  for(i=0; i<pPg->nOverflow; i++){
    int iCell = (iOld + pPg->aiOvfl[i]) - iNew;
    if( iCell>=0 && iCell<nNew ){
      pCellptr = &pPg->aCellIdx[iCell * 2];
      if( nCell>iCell ){
        memmove(&pCellptr[2], pCellptr, (nCell - iCell) * 2);
      }
      nCell++;
      cachedCellSize(pCArray, iCell+iNew);
      if( pageInsertArray(
            pPg, pBegin, &pData, pCellptr,
            iCell+iNew, 1, pCArray
      ) ) goto editpage_fail;
    }
  }

  /* Append cells to the end of the page */
  assert( nCell>=0 );
  pCellptr = &pPg->aCellIdx[nCell*2];
  if( pageInsertArray(
        pPg, pBegin, &pData, pCellptr,
        iNew+nCell, nNew-nCell, pCArray
  ) ) goto editpage_fail;

  pPg->nCell = nNew;
  pPg->nOverflow = 0;

  put2byte(&aData[hdr+3], pPg->nCell);
  put2byte(&aData[hdr+5], pData - aData);

#ifdef SQLITE_DEBUG
  for(i=0; i<nNew && !CORRUPT_DB; i++){
    u8 *pCell = pCArray->apCell[i+iNew];
    int iOff = get2byteAligned(&pPg->aCellIdx[i*2]);
    if( SQLITE_WITHIN(pCell, aData, &aData[pPg->pBt->usableSize]) ){
      pCell = &pTmp[pCell - aData];
    }
    assert( 0==memcmp(pCell, &aData[iOff],
            pCArray->pRef->xCellSize(pCArray->pRef, pCArray->apCell[i+iNew])) );
  }
#endif

  return SQLITE_OK;
 editpage_fail:
  /* Unable to edit this page. Rebuild it from scratch instead. */
  if( nNew<1 ) return SQLITE_CORRUPT_BKPT;
  populateCellCache(pCArray, iNew, nNew);
  return rebuildPage(pCArray, iNew, nNew, pPg);
}


#ifndef SQLITE_OMIT_QUICKBALANCE
/*
** This version of balance() handles the common special case where
** a new entry is being inserted on the extreme right-end of the
** tree, in other words, when the new entry will become the largest
** entry in the tree.
**
** Instead of trying to balance the 3 right-most leaf pages, just add
** a new page to the right-hand side and put the one new entry in
** that page.  This leaves the right side of the tree somewhat
** unbalanced.  But odds are that we will be inserting new entries
** at the end soon afterwards so the nearly empty page will quickly
** fill up.  On average.
**
** pPage is the leaf page which is the right-most page in the tree.
** pParent is its parent.  pPage must have a single overflow entry
** which is also the right-most entry on the page.
**
** The pSpace buffer is used to store a temporary copy of the divider
** cell that will be inserted into pParent. Such a cell consists of a 4
** byte page number followed by a variable length integer. In other
** words, at most 13 bytes. Hence the pSpace buffer must be at
** least 13 bytes in size.
*/
static int balance_quick(MemPage *pParent, MemPage *pPage, u8 *pSpace){
  BtShared *const pBt = pPage->pBt;    /* B-Tree Database */
  MemPage *pNew;                       /* Newly allocated page */
  int rc;                              /* Return Code */
  Pgno pgnoNew;                        /* Page number of pNew */

  assert( sqlite3_mutex_held(pPage->pBt->mutex) );
  assert( sqlite3PagerIswriteable(pParent->pDbPage) );
  assert( pPage->nOverflow==1 );
 
  if( pPage->nCell==0 ) return SQLITE_CORRUPT_BKPT;  /* dbfuzz001.test */
  assert( pPage->nFree>=0 );
  assert( pParent->nFree>=0 );

  /* Allocate a new page. This page will become the right-sibling of
  ** pPage. Make the parent page writable, so that the new divider cell
  ** may be inserted. If both these operations are successful, proceed.
  */
  rc = allocateBtreePage(pBt, &pNew, &pgnoNew, 0, 0);

  if( rc==SQLITE_OK ){

    u8 *pOut = &pSpace[4];
    u8 *pCell = pPage->apOvfl[0];
    u16 szCell = pPage->xCellSize(pPage, pCell);
    u8 *pStop;
    CellArray b;

    assert( sqlite3PagerIswriteable(pNew->pDbPage) );
    assert( CORRUPT_DB || pPage->aData[0]==(PTF_INTKEY|PTF_LEAFDATA|PTF_LEAF) );
    zeroPage(pNew, PTF_INTKEY|PTF_LEAFDATA|PTF_LEAF);
    b.nCell = 1;
    b.pRef = pPage;
    b.apCell = &pCell;
    b.szCell = &szCell;
    b.apEnd[0] = pPage->aDataEnd;
    b.ixNx[0] = 2;
    rc = rebuildPage(&b, 0, 1, pNew);
    if( NEVER(rc) ){
      releasePage(pNew);
      return rc;
    }
    pNew->nFree = pBt->usableSize - pNew->cellOffset - 2 - szCell;

    /* If this is an auto-vacuum database, update the pointer map
    ** with entries for the new page, and any pointer from the
    ** cell on the page to an overflow page. If either of these
    ** operations fails, the return code is set, but the contents
    ** of the parent page are still manipulated by the code below.
    ** That is Ok, at this point the parent page is guaranteed to
    ** be marked as dirty. Returning an error code will cause a
    ** rollback, undoing any changes made to the parent page.
    */
    if( ISAUTOVACUUM(pBt) ){
      ptrmapPut(pBt, pgnoNew, PTRMAP_BTREE, pParent->pgno, &rc);
      if( szCell>pNew->minLocal ){
        ptrmapPutOvflPtr(pNew, pNew, pCell, &rc);
      }
    }
 
    /* Create a divider cell to insert into pParent. The divider cell
    ** consists of a 4-byte page number (the page number of pPage) and
    ** a variable length key value (which must be the same value as the
    ** largest key on pPage).
    **
    ** To find the largest key value on pPage, first find the right-most
    ** cell on pPage. The first two fields of this cell are the
    ** record-length (a variable length integer at most 32-bits in size)
    ** and the key value (a variable length integer, may have any value).
    ** The first of the while(...) loops below skips over the record-length
    ** field. The second while(...) loop copies the key value from the
    ** cell on pPage into the pSpace buffer.
    */
    pCell = findCell(pPage, pPage->nCell-1);
    pStop = &pCell[9];
    while( (*(pCell++)&0x80) && pCell<pStop );
    pStop = &pCell[9];
    while( ((*(pOut++) = *(pCell++))&0x80) && pCell<pStop );

    /* Insert the new divider cell into pParent. */
    if( rc==SQLITE_OK ){
      rc = insertCell(pParent, pParent->nCell, pSpace, (int)(pOut-pSpace),
                      0, pPage->pgno);
    }

    /* Set the right-child pointer of pParent to point to the new page. */
    put4byte(&pParent->aData[pParent->hdrOffset+8], pgnoNew);
 
    /* Release the reference to the new page. */
    releasePage(pNew);
  }

  return rc;
}
#endif /* SQLITE_OMIT_QUICKBALANCE */

#if 0
/*
** This function does not contribute anything to the operation of SQLite.
** it is sometimes activated temporarily while debugging code responsible
** for setting pointer-map entries.
*/
static int ptrmapCheckPages(MemPage **apPage, int nPage){
  int i, j;
  for(i=0; i<nPage; i++){
    Pgno n;
    u8 e;
    MemPage *pPage = apPage[i];
    BtShared *pBt = pPage->pBt;
    assert( pPage->isInit );

    for(j=0; j<pPage->nCell; j++){
      CellInfo info;
      u8 *z;
    
      z = findCell(pPage, j);
      pPage->xParseCell(pPage, z, &info);
      if( info.nLocal<info.nPayload ){
        Pgno ovfl = get4byte(&z[info.nSize-4]);
        ptrmapGet(pBt, ovfl, &e, &n);
        assert( n==pPage->pgno && e==PTRMAP_OVERFLOW1 );
      }
      if( !pPage->leaf ){
        Pgno child = get4byte(z);
        ptrmapGet(pBt, child, &e, &n);
        assert( n==pPage->pgno && e==PTRMAP_BTREE );
      }
    }
    if( !pPage->leaf ){
      Pgno child = get4byte(&pPage->aData[pPage->hdrOffset+8]);
      ptrmapGet(pBt, child, &e, &n);
      assert( n==pPage->pgno && e==PTRMAP_BTREE );
    }
  }
  return 1;
}
#endif

/*
** This function is used to copy the contents of the b-tree node stored
** on page pFrom to page pTo. If page pFrom was not a leaf page, then
** the pointer-map entries for each child page are updated so that the
** parent page stored in the pointer map is page pTo. If pFrom contained
** any cells with overflow page pointers, then the corresponding pointer
** map entries are also updated so that the parent page is page pTo.
**
** If pFrom is currently carrying any overflow cells (entries in the
** MemPage.apOvfl[] array), they are not copied to pTo.
**
** Before returning, page pTo is reinitialized using btreeInitPage().
**
** The performance of this function is not critical. It is only used by
** the balance_shallower() and balance_deeper() procedures, neither of
** which are called often under normal circumstances.
*/
static void copyNodeContent(MemPage *pFrom, MemPage *pTo, int *pRC){
  if( (*pRC)==SQLITE_OK ){
    BtShared * const pBt = pFrom->pBt;
    u8 * const aFrom = pFrom->aData;
    u8 * const aTo = pTo->aData;
    int const iFromHdr = pFrom->hdrOffset;
    int const iToHdr = ((pTo->pgno==1) ? 100 : 0);
    int rc;
    int iData;
 
 
    assert( pFrom->isInit );
    assert( pFrom->nFree>=iToHdr );
    assert( get2byte(&aFrom[iFromHdr+5]) <= (int)pBt->usableSize );
 
    /* Copy the b-tree node content from page pFrom to page pTo. */
    iData = get2byte(&aFrom[iFromHdr+5]);
    memcpy(&aTo[iData], &aFrom[iData], pBt->usableSize-iData);
    memcpy(&aTo[iToHdr], &aFrom[iFromHdr], pFrom->cellOffset + 2*pFrom->nCell);
 
    /* Reinitialize page pTo so that the contents of the MemPage structure
    ** match the new data. The initialization of pTo can actually fail under
    ** fairly obscure circumstances, even though it is a copy of initialized
    ** page pFrom.
    */
    pTo->isInit = 0;
    rc = btreeInitPage(pTo);
    if( rc==SQLITE_OK ) rc = btreeComputeFreeSpace(pTo);
    if( rc!=SQLITE_OK ){
      *pRC = rc;
      return;
    }
 
    /* If this is an auto-vacuum database, update the pointer-map entries
    ** for any b-tree or overflow pages that pTo now contains the pointers to.
    */
    if( ISAUTOVACUUM(pBt) ){
      *pRC = setChildPtrmaps(pTo);
    }
  }
}

/*
** This routine redistributes cells on the iParentIdx'th child of pParent
** (hereafter "the page") and up to 2 siblings so that all pages have about the
** same amount of free space. Usually a single sibling on either side of the
** page are used in the balancing, though both siblings might come from one
** side if the page is the first or last child of its parent. If the page
** has fewer than 2 siblings (something which can only happen if the page
** is a root page or a child of a root page) then all available siblings
** participate in the balancing.
**
** The number of siblings of the page might be increased or decreased by
** one or two in an effort to keep pages nearly full but not over full.
**
** Note that when this routine is called, some of the cells on the page
** might not actually be stored in MemPage.aData[]. This can happen
** if the page is overfull. This routine ensures that all cells allocated
** to the page and its siblings fit into MemPage.aData[] before returning.
**
** In the course of balancing the page and its siblings, cells may be
** inserted into or removed from the parent page (pParent). Doing so
** may cause the parent page to become overfull or underfull. If this
** happens, it is the responsibility of the caller to invoke the correct
** balancing routine to fix this problem (see the balance() routine).
**
** If this routine fails for any reason, it might leave the database
** in a corrupted state. So if this routine fails, the database should
** be rolled back.
**
** The third argument to this function, aOvflSpace, is a pointer to a
** buffer big enough to hold one page. If while inserting cells into the parent
** page (pParent) the parent page becomes overfull, this buffer is
** used to store the parent's overflow cells. Because this function inserts
** a maximum of four divider cells into the parent page, and the maximum
** size of a cell stored within an internal node is always less than 1/4
** of the page-size, the aOvflSpace[] buffer is guaranteed to be large
** enough for all overflow cells.
**
** If aOvflSpace is set to a null pointer, this function returns
** SQLITE_NOMEM.
*/
static int balance_nonroot(
  MemPage *pParent,               /* Parent page of siblings being balanced */
  int iParentIdx,                 /* Index of "the page" in pParent */
  u8 *aOvflSpace,                 /* page-size bytes of space for parent ovfl */
  int isRoot,                     /* True if pParent is a root-page */
  int bBulk                       /* True if this call is part of a bulk load */
){
  BtShared *pBt;               /* The whole database */
  int nMaxCells = 0;           /* Allocated size of apCell, szCell, aFrom. */
  int nNew = 0;                /* Number of pages in apNew[] */
  int nOld;                    /* Number of pages in apOld[] */
  int i, j, k;                 /* Loop counters */
  int nxDiv;                   /* Next divider slot in pParent->aCell[] */
  int rc = SQLITE_OK;          /* The return code */
  u16 leafCorrection;          /* 4 if pPage is a leaf.  0 if not */
  int leafData;                /* True if pPage is a leaf of a LEAFDATA tree */
  int usableSpace;             /* Bytes in pPage beyond the header */
  int pageFlags;               /* Value of pPage->aData[0] */
  int iSpace1 = 0;             /* First unused byte of aSpace1[] */
  int iOvflSpace = 0;          /* First unused byte of aOvflSpace[] */
  int szScratch;               /* Size of scratch memory requested */
  MemPage *apOld[NB];          /* pPage and up to two siblings */
  MemPage *apNew[NB+2];        /* pPage and up to NB siblings after balancing */
  u8 *pRight;                  /* Location in parent of right-sibling pointer */
  u8 *apDiv[NB-1];             /* Divider cells in pParent */
  int cntNew[NB+2];            /* Index in b.paCell[] of cell after i-th page */
  int cntOld[NB+2];            /* Old index in b.apCell[] */
  int szNew[NB+2];             /* Combined size of cells placed on i-th page */
  u8 *aSpace1;                 /* Space for copies of dividers cells */
  Pgno pgno;                   /* Temp var to store a page number in */
  u8 abDone[NB+2];             /* True after i'th new page is populated */
  Pgno aPgno[NB+2];            /* Page numbers of new pages before shuffling */
  CellArray b;                 /* Parsed information on cells being balanced */

  memset(abDone, 0, sizeof(abDone));
  memset(&b, 0, sizeof(b));
  pBt = pParent->pBt;
  assert( sqlite3_mutex_held(pBt->mutex) );
  assert( sqlite3PagerIswriteable(pParent->pDbPage) );

  /* At this point pParent may have at most one overflow cell. And if
  ** this overflow cell is present, it must be the cell with
  ** index iParentIdx. This scenario comes about when this function
  ** is called (indirectly) from sqlite3BtreeDelete().
  */
  assert( pParent->nOverflow==0 || pParent->nOverflow==1 );
  assert( pParent->nOverflow==0 || pParent->aiOvfl[0]==iParentIdx );

  if( !aOvflSpace ){
    return SQLITE_NOMEM_BKPT;
  }
  assert( pParent->nFree>=0 );

  /* Find the sibling pages to balance. Also locate the cells in pParent
  ** that divide the siblings. An attempt is made to find NN siblings on
  ** either side of pPage. More siblings are taken from one side, however,
  ** if there are fewer than NN siblings on the other side. If pParent
  ** has NB or fewer children then all children of pParent are taken. 
  **
  ** This loop also drops the divider cells from the parent page. This
  ** way, the remainder of the function does not have to deal with any
  ** overflow cells in the parent page, since if any existed they will
  ** have already been removed.
  */
  i = pParent->nOverflow + pParent->nCell;
  if( i<2 ){
    nxDiv = 0;
  }else{
    assert( bBulk==0 || bBulk==1 );
    if( iParentIdx==0 ){                
      nxDiv = 0;
    }else if( iParentIdx==i ){
      nxDiv = i-2+bBulk;
    }else{
      nxDiv = iParentIdx-1;
    }
    i = 2-bBulk;
  }
  nOld = i+1;
  if( (i+nxDiv-pParent->nOverflow)==pParent->nCell ){
    pRight = &pParent->aData[pParent->hdrOffset+8];
  }else{
    pRight = findCell(pParent, i+nxDiv-pParent->nOverflow);
  }
  pgno = get4byte(pRight);
  while( 1 ){
    if( rc==SQLITE_OK ){
      rc = getAndInitPage(pBt, pgno, &apOld[i], 0);
    }
    if( rc ){
      memset(apOld, 0, (i+1)*sizeof(MemPage*));
      goto balance_cleanup;
    }
    if( apOld[i]->nFree<0 ){
      rc = btreeComputeFreeSpace(apOld[i]);
      if( rc ){
        memset(apOld, 0, (i)*sizeof(MemPage*));
        goto balance_cleanup;
      }
    }
    nMaxCells += apOld[i]->nCell + ArraySize(pParent->apOvfl);
    if( (i--)==0 ) break;

    if( pParent->nOverflow && i+nxDiv==pParent->aiOvfl[0] ){
      apDiv[i] = pParent->apOvfl[0];
      pgno = get4byte(apDiv[i]);
      szNew[i] = pParent->xCellSize(pParent, apDiv[i]);
      pParent->nOverflow = 0;
    }else{
      apDiv[i] = findCell(pParent, i+nxDiv-pParent->nOverflow);
      pgno = get4byte(apDiv[i]);
      szNew[i] = pParent->xCellSize(pParent, apDiv[i]);

      /* Drop the cell from the parent page. apDiv[i] still points to
      ** the cell within the parent, even though it has been dropped.
      ** This is safe because dropping a cell only overwrites the first
      ** four bytes of it, and this function does not need the first
      ** four bytes of the divider cell. So the pointer is safe to use
      ** later on. 
      **
      ** But not if we are in secure-delete mode. In secure-delete mode,
      ** the dropCell() routine will overwrite the entire cell with zeroes.
      ** In this case, temporarily copy the cell into the aOvflSpace[]
      ** buffer. It will be copied out again as soon as the aSpace[] buffer
      ** is allocated.  */
      if( pBt->btsFlags & BTS_FAST_SECURE ){
        int iOff;

        /* If the following if() condition is not true, the db is corrupted.
        ** The call to dropCell() below will detect this.  */
        iOff = SQLITE_PTR_TO_INT(apDiv[i]) - SQLITE_PTR_TO_INT(pParent->aData);
        if( (iOff+szNew[i])<=(int)pBt->usableSize ){
          memcpy(&aOvflSpace[iOff], apDiv[i], szNew[i]);
          apDiv[i] = &aOvflSpace[apDiv[i]-pParent->aData];
        }
      }
      dropCell(pParent, i+nxDiv-pParent->nOverflow, szNew[i], &rc);
    }
  }

  /* Make nMaxCells a multiple of 4 in order to preserve 8-byte
  ** alignment */
  nMaxCells = (nMaxCells + 3)&~3;

  /*
  ** Allocate space for memory structures
  */
  szScratch =
       nMaxCells*sizeof(u8*)                       /* b.apCell */
     + nMaxCells*sizeof(u16)                       /* b.szCell */
     + pBt->pageSize;                              /* aSpace1 */

  assert( szScratch<=7*(int)pBt->pageSize );
  b.apCell = sqlite3StackAllocRaw(0, szScratch );
  if( b.apCell==0 ){
    rc = SQLITE_NOMEM_BKPT;
    goto balance_cleanup;
  }
  b.szCell = (u16*)&b.apCell[nMaxCells];
  aSpace1 = (u8*)&b.szCell[nMaxCells];
  assert( EIGHT_BYTE_ALIGNMENT(aSpace1) );

  /*
  ** Load pointers to all cells on sibling pages and the divider cells
  ** into the local b.apCell[] array.  Make copies of the divider cells
  ** into space obtained from aSpace1[]. The divider cells have already
  ** been removed from pParent.
  **
  ** If the siblings are on leaf pages, then the child pointers of the
  ** divider cells are stripped from the cells before they are copied
  ** into aSpace1[].  In this way, all cells in b.apCell[] are without
  ** child pointers.  If siblings are not leaves, then all cell in
  ** b.apCell[] include child pointers.  Either way, all cells in b.apCell[]
  ** are alike.
  **
  ** leafCorrection:  4 if pPage is a leaf.  0 if pPage is not a leaf.
  **       leafData:  1 if pPage holds key+data and pParent holds only keys.
  */
  b.pRef = apOld[0];
  leafCorrection = b.pRef->leaf*4;
  leafData = b.pRef->intKeyLeaf;
  for(i=0; i<nOld; i++){
    MemPage *pOld = apOld[i];
    int limit = pOld->nCell;
    u8 *aData = pOld->aData;
    u16 maskPage = pOld->maskPage;
    u8 *piCell = aData + pOld->cellOffset;
    u8 *piEnd;
    VVA_ONLY( int nCellAtStart = b.nCell; )

    /* Verify that all sibling pages are of the same "type" (table-leaf,
    ** table-interior, index-leaf, or index-interior).
    */
    if( pOld->aData[0]!=apOld[0]->aData[0] ){
      rc = SQLITE_CORRUPT_BKPT;
      goto balance_cleanup;
    }

    /* Load b.apCell[] with pointers to all cells in pOld.  If pOld
    ** contains overflow cells, include them in the b.apCell[] array
    ** in the correct spot.
    **
    ** Note that when there are multiple overflow cells, it is always the
    ** case that they are sequential and adjacent.  This invariant arises
    ** because multiple overflows can only occurs when inserting divider
    ** cells into a parent on a prior balance, and divider cells are always
    ** adjacent and are inserted in order.  There is an assert() tagged
    ** with "NOTE 1" in the overflow cell insertion loop to prove this
    ** invariant.
    **
    ** This must be done in advance.  Once the balance starts, the cell
    ** offset section of the btree page will be overwritten and we will no
    ** long be able to find the cells if a pointer to each cell is not saved
    ** first.
    */
    memset(&b.szCell[b.nCell], 0, sizeof(b.szCell[0])*(limit+pOld->nOverflow));
    if( pOld->nOverflow>0 ){
      if( NEVER(limit<pOld->aiOvfl[0]) ){
        rc = SQLITE_CORRUPT_BKPT;
        goto balance_cleanup;
      }
      limit = pOld->aiOvfl[0];
      for(j=0; j<limit; j++){
        b.apCell[b.nCell] = aData + (maskPage & get2byteAligned(piCell));
        piCell += 2;
        b.nCell++;
      }
      for(k=0; k<pOld->nOverflow; k++){
        assert( k==0 || pOld->aiOvfl[k-1]+1==pOld->aiOvfl[k] );/* NOTE 1 */
        b.apCell[b.nCell] = pOld->apOvfl[k];
        b.nCell++;
      }
    }
    piEnd = aData + pOld->cellOffset + 2*pOld->nCell;
    while( piCell<piEnd ){
      assert( b.nCell<nMaxCells );
      b.apCell[b.nCell] = aData + (maskPage & get2byteAligned(piCell));
      piCell += 2;
      b.nCell++;
    }
    assert( (b.nCell-nCellAtStart)==(pOld->nCell+pOld->nOverflow) );

    cntOld[i] = b.nCell;
    if( i<nOld-1 && !leafData){
      u16 sz = (u16)szNew[i];
      u8 *pTemp;
      assert( b.nCell<nMaxCells );
      b.szCell[b.nCell] = sz;
      pTemp = &aSpace1[iSpace1];
      iSpace1 += sz;
      assert( sz<=pBt->maxLocal+23 );
      assert( iSpace1 <= (int)pBt->pageSize );
      memcpy(pTemp, apDiv[i], sz);
      b.apCell[b.nCell] = pTemp+leafCorrection;
      assert( leafCorrection==0 || leafCorrection==4 );
      b.szCell[b.nCell] = b.szCell[b.nCell] - leafCorrection;
      if( !pOld->leaf ){
        assert( leafCorrection==0 );
        assert( pOld->hdrOffset==0 || CORRUPT_DB );
        /* The right pointer of the child page pOld becomes the left
        ** pointer of the divider cell */
        memcpy(b.apCell[b.nCell], &pOld->aData[8], 4);
      }else{
        assert( leafCorrection==4 );
        while( b.szCell[b.nCell]<4 ){
          /* Do not allow any cells smaller than 4 bytes. If a smaller cell
          ** does exist, pad it with 0x00 bytes. */
          assert( b.szCell[b.nCell]==3 || CORRUPT_DB );
          assert( b.apCell[b.nCell]==&aSpace1[iSpace1-3] || CORRUPT_DB );
          aSpace1[iSpace1++] = 0x00;
          b.szCell[b.nCell]++;
        }
      }
      b.nCell++;
    }
  }

  /*
  ** Figure out the number of pages needed to hold all b.nCell cells.
  ** Store this number in "k".  Also compute szNew[] which is the total
  ** size of all cells on the i-th page and cntNew[] which is the index
  ** in b.apCell[] of the cell that divides page i from page i+1. 
  ** cntNew[k] should equal b.nCell.
  **
  ** Values computed by this block:
  **
  **           k: The total number of sibling pages
  **    szNew[i]: Spaced used on the i-th sibling page.
  **   cntNew[i]: Index in b.apCell[] and b.szCell[] for the first cell to
  **              the right of the i-th sibling page.
  ** usableSpace: Number of bytes of space available on each sibling.
  **
  */
  usableSpace = pBt->usableSize - 12 + leafCorrection;
  for(i=k=0; i<nOld; i++, k++){
    MemPage *p = apOld[i];
    b.apEnd[k] = p->aDataEnd;
    b.ixNx[k] = cntOld[i];
    if( k && b.ixNx[k]==b.ixNx[k-1] ){
      k--;  /* Omit b.ixNx[] entry for child pages with no cells */
    }
    if( !leafData ){
      k++;
      b.apEnd[k] = pParent->aDataEnd;
      b.ixNx[k] = cntOld[i]+1;
    }
    assert( p->nFree>=0 );
    szNew[i] = usableSpace - p->nFree;
    for(j=0; j<p->nOverflow; j++){
      szNew[i] += 2 + p->xCellSize(p, p->apOvfl[j]);
    }
    cntNew[i] = cntOld[i];
  }
  k = nOld;
  for(i=0; i<k; i++){
    int sz;
    while( szNew[i]>usableSpace ){
      if( i+1>=k ){
        k = i+2;
        if( k>NB+2 ){ rc = SQLITE_CORRUPT_BKPT; goto balance_cleanup; }
        szNew[k-1] = 0;
        cntNew[k-1] = b.nCell;
      }
      sz = 2 + cachedCellSize(&b, cntNew[i]-1);
      szNew[i] -= sz;
      if( !leafData ){
        if( cntNew[i]<b.nCell ){
          sz = 2 + cachedCellSize(&b, cntNew[i]);
        }else{
          sz = 0;
        }
      }
      szNew[i+1] += sz;
      cntNew[i]--;
    }
    while( cntNew[i]<b.nCell ){
      sz = 2 + cachedCellSize(&b, cntNew[i]);
      if( szNew[i]+sz>usableSpace ) break;
      szNew[i] += sz;
      cntNew[i]++;
      if( !leafData ){
        if( cntNew[i]<b.nCell ){
          sz = 2 + cachedCellSize(&b, cntNew[i]);
        }else{
          sz = 0;
        }
      }
      szNew[i+1] -= sz;
    }
    if( cntNew[i]>=b.nCell ){
      k = i+1;
    }else if( cntNew[i] <= (i>0 ? cntNew[i-1] : 0) ){
      rc = SQLITE_CORRUPT_BKPT;
      goto balance_cleanup;
    }
  }

  /*
  ** The packing computed by the previous block is biased toward the siblings
  ** on the left side (siblings with smaller keys). The left siblings are
  ** always nearly full, while the right-most sibling might be nearly empty.
  ** The next block of code attempts to adjust the packing of siblings to
  ** get a better balance.
  **
  ** This adjustment is more than an optimization.  The packing above might
  ** be so out of balance as to be illegal.  For example, the right-most
  ** sibling might be completely empty.  This adjustment is not optional.
  */
  for(i=k-1; i>0; i--){
    int szRight = szNew[i];  /* Size of sibling on the right */
    int szLeft = szNew[i-1]; /* Size of sibling on the left */
    int r;              /* Index of right-most cell in left sibling */
    int d;              /* Index of first cell to the left of right sibling */

    r = cntNew[i-1] - 1;
    d = r + 1 - leafData;
    (void)cachedCellSize(&b, d);
    do{
      int szR, szD;
      assert( d<nMaxCells );
      assert( r<nMaxCells );
      szR = cachedCellSize(&b, r);
      szD = b.szCell[d];
      if( szRight!=0
       && (bBulk || szRight+szD+2 > szLeft-(szR+(i==k-1?0:2)))){
        break;
      }
      szRight += szD + 2;
      szLeft -= szR + 2;
      cntNew[i-1] = r;
      r--;
      d--;
    }while( r>=0 );
    szNew[i] = szRight;
    szNew[i-1] = szLeft;
    if( cntNew[i-1] <= (i>1 ? cntNew[i-2] : 0) ){
      rc = SQLITE_CORRUPT_BKPT;
      goto balance_cleanup;
    }
  }

  /* Sanity check:  For a non-corrupt database file one of the following
  ** must be true:
  **    (1) We found one or more cells (cntNew[0])>0), or
  **    (2) pPage is a virtual root page.  A virtual root page is when
  **        the real root page is page 1 and we are the only child of
  **        that page.
  */
  assert( cntNew[0]>0 || (pParent->pgno==1 && pParent->nCell==0) || CORRUPT_DB);
  TRACE(("BALANCE: old: %u(nc=%u) %u(nc=%u) %u(nc=%u)\n",
    apOld[0]->pgno, apOld[0]->nCell,
    nOld>=2 ? apOld[1]->pgno : 0, nOld>=2 ? apOld[1]->nCell : 0,
    nOld>=3 ? apOld[2]->pgno : 0, nOld>=3 ? apOld[2]->nCell : 0
  ));

  /*
  ** Allocate k new pages.  Reuse old pages where possible.
  */
  pageFlags = apOld[0]->aData[0];
  for(i=0; i<k; i++){
    MemPage *pNew;
    if( i<nOld ){
      pNew = apNew[i] = apOld[i];
      apOld[i] = 0;
      rc = sqlite3PagerWrite(pNew->pDbPage);
      nNew++;
      if( sqlite3PagerPageRefcount(pNew->pDbPage)!=1+(i==(iParentIdx-nxDiv))
       && rc==SQLITE_OK
      ){
        rc = SQLITE_CORRUPT_BKPT;
      }
      if( rc ) goto balance_cleanup;
    }else{
      assert( i>0 );
      rc = allocateBtreePage(pBt, &pNew, &pgno, (bBulk ? 1 : pgno), 0);
      if( rc ) goto balance_cleanup;
      zeroPage(pNew, pageFlags);
      apNew[i] = pNew;
      nNew++;
      cntOld[i] = b.nCell;

      /* Set the pointer-map entry for the new sibling page. */
      if( ISAUTOVACUUM(pBt) ){
        ptrmapPut(pBt, pNew->pgno, PTRMAP_BTREE, pParent->pgno, &rc);
        if( rc!=SQLITE_OK ){
          goto balance_cleanup;
        }
      }
    }
  }

  /*
  ** Reassign page numbers so that the new pages are in ascending order.
  ** This helps to keep entries in the disk file in order so that a scan
  ** of the table is closer to a linear scan through the file. That in turn
  ** helps the operating system to deliver pages from the disk more rapidly.
  **
  ** An O(N*N) sort algorithm is used, but since N is never more than NB+2
  ** (5), that is not a performance concern.
  **
  ** When NB==3, this one optimization makes the database about 25% faster
  ** for large insertions and deletions.
  */
  for(i=0; i<nNew; i++){
    aPgno[i] = apNew[i]->pgno;
    assert( apNew[i]->pDbPage->flags & PGHDR_WRITEABLE );
    assert( apNew[i]->pDbPage->flags & PGHDR_DIRTY );
  }
  for(i=0; i<nNew-1; i++){
    int iB = i;
    for(j=i+1; j<nNew; j++){
      if( apNew[j]->pgno < apNew[iB]->pgno ) iB = j;
    }

    /* If apNew[i] has a page number that is bigger than any of the
    ** subsequence apNew[i] entries, then swap apNew[i] with the subsequent
    ** entry that has the smallest page number (which we know to be
    ** entry apNew[iB]).
    */
    if( iB!=i ){
      Pgno pgnoA = apNew[i]->pgno;
      Pgno pgnoB = apNew[iB]->pgno;
      Pgno pgnoTemp = (PENDING_BYTE/pBt->pageSize)+1;
      u16 fgA = apNew[i]->pDbPage->flags;
      u16 fgB = apNew[iB]->pDbPage->flags;
      sqlite3PagerRekey(apNew[i]->pDbPage, pgnoTemp, fgB);
      sqlite3PagerRekey(apNew[iB]->pDbPage, pgnoA, fgA);
      sqlite3PagerRekey(apNew[i]->pDbPage, pgnoB, fgB);
      apNew[i]->pgno = pgnoB;
      apNew[iB]->pgno = pgnoA;
    }
  }

  TRACE(("BALANCE: new: %u(%u nc=%u) %u(%u nc=%u) %u(%u nc=%u) "
         "%u(%u nc=%u) %u(%u nc=%u)\n",
    apNew[0]->pgno, szNew[0], cntNew[0],
    nNew>=2 ? apNew[1]->pgno : 0, nNew>=2 ? szNew[1] : 0,
    nNew>=2 ? cntNew[1] - cntNew[0] - !leafData : 0,
    nNew>=3 ? apNew[2]->pgno : 0, nNew>=3 ? szNew[2] : 0,
    nNew>=3 ? cntNew[2] - cntNew[1] - !leafData : 0,
    nNew>=4 ? apNew[3]->pgno : 0, nNew>=4 ? szNew[3] : 0,
    nNew>=4 ? cntNew[3] - cntNew[2] - !leafData : 0,
    nNew>=5 ? apNew[4]->pgno : 0, nNew>=5 ? szNew[4] : 0,
    nNew>=5 ? cntNew[4] - cntNew[3] - !leafData : 0
  ));

  assert( sqlite3PagerIswriteable(pParent->pDbPage) );
  assert( nNew>=1 && nNew<=ArraySize(apNew) );
  assert( apNew[nNew-1]!=0 );
  put4byte(pRight, apNew[nNew-1]->pgno);

  /* If the sibling pages are not leaves, ensure that the right-child pointer
  ** of the right-most new sibling page is set to the value that was
  ** originally in the same field of the right-most old sibling page. */
  if( (pageFlags & PTF_LEAF)==0 && nOld!=nNew ){
    MemPage *pOld = (nNew>nOld ? apNew : apOld)[nOld-1];
    memcpy(&apNew[nNew-1]->aData[8], &pOld->aData[8], 4);
  }

  /* Make any required updates to pointer map entries associated with
  ** cells stored on sibling pages following the balance operation. Pointer
  ** map entries associated with divider cells are set by the insertCell()
  ** routine. The associated pointer map entries are:
  **
  **   a) if the cell contains a reference to an overflow chain, the
  **      entry associated with the first page in the overflow chain, and
  **
  **   b) if the sibling pages are not leaves, the child page associated
  **      with the cell.
  **
  ** If the sibling pages are not leaves, then the pointer map entry
  ** associated with the right-child of each sibling may also need to be
  ** updated. This happens below, after the sibling pages have been
  ** populated, not here.
  */
  if( ISAUTOVACUUM(pBt) ){
    MemPage *pOld;
    MemPage *pNew = pOld = apNew[0];
    int cntOldNext = pNew->nCell + pNew->nOverflow;
    int iNew = 0;
    int iOld = 0;

    for(i=0; i<b.nCell; i++){
      u8 *pCell = b.apCell[i];
      while( i==cntOldNext ){
        iOld++;
        assert( iOld<nNew || iOld<nOld );
        assert( iOld>=0 && iOld<NB );
        pOld = iOld<nNew ? apNew[iOld] : apOld[iOld];
        cntOldNext += pOld->nCell + pOld->nOverflow + !leafData;
      }
      if( i==cntNew[iNew] ){
        pNew = apNew[++iNew];
        if( !leafData ) continue;
      }

      /* Cell pCell is destined for new sibling page pNew. Originally, it
      ** was either part of sibling page iOld (possibly an overflow cell),
      ** or else the divider cell to the left of sibling page iOld. So,
      ** if sibling page iOld had the same page number as pNew, and if
      ** pCell really was a part of sibling page iOld (not a divider or
      ** overflow cell), we can skip updating the pointer map entries.  */
      if( iOld>=nNew
       || pNew->pgno!=aPgno[iOld]
       || !SQLITE_WITHIN(pCell,pOld->aData,pOld->aDataEnd)
      ){
        if( !leafCorrection ){
          ptrmapPut(pBt, get4byte(pCell), PTRMAP_BTREE, pNew->pgno, &rc);
        }
        if( cachedCellSize(&b,i)>pNew->minLocal ){
          ptrmapPutOvflPtr(pNew, pOld, pCell, &rc);
        }
        if( rc ) goto balance_cleanup;
      }
    }
  }

  /* Insert new divider cells into pParent. */
  for(i=0; i<nNew-1; i++){
    u8 *pCell;
    u8 *pTemp;
    int sz;
    u8 *pSrcEnd;
    MemPage *pNew = apNew[i];
    j = cntNew[i];

    assert( j<nMaxCells );
    assert( b.apCell[j]!=0 );
    pCell = b.apCell[j];
    sz = b.szCell[j] + leafCorrection;
    pTemp = &aOvflSpace[iOvflSpace];
    if( !pNew->leaf ){
      memcpy(&pNew->aData[8], pCell, 4);
    }else if( leafData ){
      /* If the tree is a leaf-data tree, and the siblings are leaves,
      ** then there is no divider cell in b.apCell[]. Instead, the divider
      ** cell consists of the integer key for the right-most cell of
      ** the sibling-page assembled above only.
      */
      CellInfo info;
      j--;
      pNew->xParseCell(pNew, b.apCell[j], &info);
      pCell = pTemp;
      sz = 4 + putVarint(&pCell[4], info.nKey);
      pTemp = 0;
    }else{
      pCell -= 4;
      /* Obscure case for non-leaf-data trees: If the cell at pCell was
      ** previously stored on a leaf node, and its reported size was 4
      ** bytes, then it may actually be smaller than this
      ** (see btreeParseCellPtr(), 4 bytes is the minimum size of
      ** any cell). But it is important to pass the correct size to
      ** insertCell(), so reparse the cell now.
      **
      ** This can only happen for b-trees used to evaluate "IN (SELECT ...)"
      ** and WITHOUT ROWID tables with exactly one column which is the
      ** primary key.
      */
      if( b.szCell[j]==4 ){
        assert(leafCorrection==4);
        sz = pParent->xCellSize(pParent, pCell);
      }
    }
    iOvflSpace += sz;
    assert( sz<=pBt->maxLocal+23 );
    assert( iOvflSpace <= (int)pBt->pageSize );
    for(k=0; ALWAYS(k<NB*2) && b.ixNx[k]<=j; k++){}
    pSrcEnd = b.apEnd[k];
    if( SQLITE_OVERFLOW(pSrcEnd, pCell, pCell+sz) ){
      rc = SQLITE_CORRUPT_BKPT;
      goto balance_cleanup;
    }
    rc = insertCell(pParent, nxDiv+i, pCell, sz, pTemp, pNew->pgno);
    if( rc!=SQLITE_OK ) goto balance_cleanup;
    assert( sqlite3PagerIswriteable(pParent->pDbPage) );
  }

  /* Now update the actual sibling pages. The order in which they are updated
  ** is important, as this code needs to avoid disrupting any page from which
  ** cells may still to be read. In practice, this means:
  **
  **  (1) If cells are moving left (from apNew[iPg] to apNew[iPg-1])
  **      then it is not safe to update page apNew[iPg] until after
  **      the left-hand sibling apNew[iPg-1] has been updated.
  **
  **  (2) If cells are moving right (from apNew[iPg] to apNew[iPg+1])
  **      then it is not safe to update page apNew[iPg] until after
  **      the right-hand sibling apNew[iPg+1] has been updated.
  **
  ** If neither of the above apply, the page is safe to update.
  **
  ** The iPg value in the following loop starts at nNew-1 goes down
  ** to 0, then back up to nNew-1 again, thus making two passes over
  ** the pages.  On the initial downward pass, only condition (1) above
  ** needs to be tested because (2) will always be true from the previous
  ** step.  On the upward pass, both conditions are always true, so the
  ** upwards pass simply processes pages that were missed on the downward
  ** pass.
  */
  for(i=1-nNew; i<nNew; i++){
    int iPg = i<0 ? -i : i;
    assert( iPg>=0 && iPg<nNew );
    assert( iPg>=1 || i>=0 );
    assert( iPg<ArraySize(cntOld) );
    if( abDone[iPg] ) continue;         /* Skip pages already processed */
    if( i>=0                            /* On the upwards pass, or... */
     || cntOld[iPg-1]>=cntNew[iPg-1]    /* Condition (1) is true */
    ){
      int iNew;
      int iOld;
      int nNewCell;

      /* Verify condition (1):  If cells are moving left, update iPg
      ** only after iPg-1 has already been updated. */
      assert( iPg==0 || cntOld[iPg-1]>=cntNew[iPg-1] || abDone[iPg-1] );

      /* Verify condition (2):  If cells are moving right, update iPg
      ** only after iPg+1 has already been updated. */
      assert( cntNew[iPg]>=cntOld[iPg] || abDone[iPg+1] );

      if( iPg==0 ){
        iNew = iOld = 0;
        nNewCell = cntNew[0];
      }else{
        iOld = iPg<nOld ? (cntOld[iPg-1] + !leafData) : b.nCell;
        iNew = cntNew[iPg-1] + !leafData;
        nNewCell = cntNew[iPg] - iNew;
      }

      rc = editPage(apNew[iPg], iOld, iNew, nNewCell, &b);
      if( rc ) goto balance_cleanup;
      abDone[iPg]++;
      apNew[iPg]->nFree = usableSpace-szNew[iPg];
      assert( apNew[iPg]->nOverflow==0 );
      assert( apNew[iPg]->nCell==nNewCell );
    }
  }

  /* All pages have been processed exactly once */
  assert( memcmp(abDone, "\01\01\01\01\01", nNew)==0 );

  assert( nOld>0 );
  assert( nNew>0 );

  if( isRoot && pParent->nCell==0 && pParent->hdrOffset<=apNew[0]->nFree ){
    /* The root page of the b-tree now contains no cells. The only sibling
    ** page is the right-child of the parent. Copy the contents of the
    ** child page into the parent, decreasing the overall height of the
    ** b-tree structure by one. This is described as the "balance-shallower"
    ** sub-algorithm in some documentation.
    **
    ** If this is an auto-vacuum database, the call to copyNodeContent()
    ** sets all pointer-map entries corresponding to database image pages
    ** for which the pointer is stored within the content being copied.
    **
    ** It is critical that the child page be defragmented before being
    ** copied into the parent, because if the parent is page 1 then it will
    ** by smaller than the child due to the database header, and so all the
    ** free space needs to be up front.
    */
    assert( nNew==1 || CORRUPT_DB );
    rc = defragmentPage(apNew[0], -1);
    testcase( rc!=SQLITE_OK );
    assert( apNew[0]->nFree ==
        (get2byteNotZero(&apNew[0]->aData[5]) - apNew[0]->cellOffset
          - apNew[0]->nCell*2)
      || rc!=SQLITE_OK
    );
    copyNodeContent(apNew[0], pParent, &rc);
    freePage(apNew[0], &rc);
  }else if( ISAUTOVACUUM(pBt) && !leafCorrection ){
    /* Fix the pointer map entries associated with the right-child of each
    ** sibling page. All other pointer map entries have already been taken
    ** care of.  */
    for(i=0; i<nNew; i++){
      u32 key = get4byte(&apNew[i]->aData[8]);
      ptrmapPut(pBt, key, PTRMAP_BTREE, apNew[i]->pgno, &rc);
    }
  }

  assert( pParent->isInit );
  TRACE(("BALANCE: finished: old=%u new=%u cells=%u\n",
          nOld, nNew, b.nCell));

  /* Free any old pages that were not reused as new pages.
  */
  for(i=nNew; i<nOld; i++){
    freePage(apOld[i], &rc);
  }

#if 0
  if( ISAUTOVACUUM(pBt) && rc==SQLITE_OK && apNew[0]->isInit ){
    /* The ptrmapCheckPages() contains assert() statements that verify that
    ** all pointer map pages are set correctly. This is helpful while
    ** debugging. This is usually disabled because a corrupt database may
    ** cause an assert() statement to fail.  */
    ptrmapCheckPages(apNew, nNew);
    ptrmapCheckPages(&pParent, 1);
  }
#endif

  /*
  ** Cleanup before returning.
  */
balance_cleanup:
  sqlite3StackFree(0, b.apCell);
  for(i=0; i<nOld; i++){
    releasePage(apOld[i]);
  }
  for(i=0; i<nNew; i++){
    releasePage(apNew[i]);
  }

  return rc;
}


/*
** This function is called when the root page of a b-tree structure is
** overfull (has one or more overflow pages).
**
** A new child page is allocated and the contents of the current root
** page, including overflow cells, are copied into the child. The root
** page is then overwritten to make it an empty page with the right-child
** pointer pointing to the new page.
**
** Before returning, all pointer-map entries corresponding to pages
** that the new child-page now contains pointers to are updated. The
** entry corresponding to the new right-child pointer of the root
** page is also updated.
**
** If successful, *ppChild is set to contain a reference to the child
** page and SQLITE_OK is returned. In this case the caller is required
** to call releasePage() on *ppChild exactly once. If an error occurs,
** an error code is returned and *ppChild is set to 0.
*/
static int balance_deeper(MemPage *pRoot, MemPage **ppChild){
  int rc;                        /* Return value from subprocedures */
  MemPage *pChild = 0;           /* Pointer to a new child page */
  Pgno pgnoChild = 0;            /* Page number of the new child page */
  BtShared *pBt = pRoot->pBt;    /* The BTree */

  assert( pRoot->nOverflow>0 );
  assert( sqlite3_mutex_held(pBt->mutex) );

  /* Make pRoot, the root page of the b-tree, writable. Allocate a new
  ** page that will become the new right-child of pPage. Copy the contents
  ** of the node stored on pRoot into the new child page.
  */
  rc = sqlite3PagerWrite(pRoot->pDbPage);
  if( rc==SQLITE_OK ){
    rc = allocateBtreePage(pBt,&pChild,&pgnoChild,pRoot->pgno,0);
    copyNodeContent(pRoot, pChild, &rc);
    if( ISAUTOVACUUM(pBt) ){
      ptrmapPut(pBt, pgnoChild, PTRMAP_BTREE, pRoot->pgno, &rc);
    }
  }
  if( rc ){
    *ppChild = 0;
    releasePage(pChild);
    return rc;
  }
  assert( sqlite3PagerIswriteable(pChild->pDbPage) );
  assert( sqlite3PagerIswriteable(pRoot->pDbPage) );
  assert( pChild->nCell==pRoot->nCell || CORRUPT_DB );

  TRACE(("BALANCE: copy root %u into %u\n", pRoot->pgno, pChild->pgno));

  /* Copy the overflow cells from pRoot to pChild */
  memcpy(pChild->aiOvfl, pRoot->aiOvfl,
         pRoot->nOverflow*sizeof(pRoot->aiOvfl[0]));
  memcpy(pChild->apOvfl, pRoot->apOvfl,
         pRoot->nOverflow*sizeof(pRoot->apOvfl[0]));
  pChild->nOverflow = pRoot->nOverflow;

  /* Zero the contents of pRoot. Then install pChild as the right-child. */
  zeroPage(pRoot, pChild->aData[0] & ~PTF_LEAF);
  put4byte(&pRoot->aData[pRoot->hdrOffset+8], pgnoChild);

  *ppChild = pChild;
  return SQLITE_OK;
}

/*
** Return SQLITE_CORRUPT if any cursor other than pCur is currently valid
** on the same B-tree as pCur.
**
** This can occur if a database is corrupt with two or more SQL tables
** pointing to the same b-tree.  If an insert occurs on one SQL table
** and causes a BEFORE TRIGGER to do a secondary insert on the other SQL
** table linked to the same b-tree.  If the secondary insert causes a
** rebalance, that can change content out from under the cursor on the
** first SQL table, violating invariants on the first insert.
*/
static int anotherValidCursor(BtCursor *pCur){
  BtCursor *pOther;
  for(pOther=pCur->pBt->pCursor; pOther; pOther=pOther->pNext){
    if( pOther!=pCur
     && pOther->eState==CURSOR_VALID
     && pOther->pPage==pCur->pPage
    ){
      return SQLITE_CORRUPT_BKPT;
    }
  }
  return SQLITE_OK;
}

/*
** The page that pCur currently points to has just been modified in
** some way. This function figures out if this modification means the
** tree needs to be balanced, and if so calls the appropriate balancing
** routine. Balancing routines are:
**
**   balance_quick()
**   balance_deeper()
**   balance_nonroot()
*/
static int balance(BtCursor *pCur){
  int rc = SQLITE_OK;
  u8 aBalanceQuickSpace[13];
  u8 *pFree = 0;

  VVA_ONLY( int balance_quick_called = 0 );
  VVA_ONLY( int balance_deeper_called = 0 );

  do {
    int iPage;
    MemPage *pPage = pCur->pPage;

    if( NEVER(pPage->nFree<0) && btreeComputeFreeSpace(pPage) ) break;
    if( pPage->nOverflow==0 && pPage->nFree*3<=(int)pCur->pBt->usableSize*2 ){
      /* No rebalance required as long as:
      **   (1) There are no overflow cells
      **   (2) The amount of free space on the page is less than 2/3rds of
      **       the total usable space on the page. */
      break;
    }else if( (iPage = pCur->iPage)==0 ){
      if( pPage->nOverflow && (rc = anotherValidCursor(pCur))==SQLITE_OK ){
        /* The root page of the b-tree is overfull. In this case call the
        ** balance_deeper() function to create a new child for the root-page
        ** and copy the current contents of the root-page to it. The
        ** next iteration of the do-loop will balance the child page.
        */
        assert( balance_deeper_called==0 );
        VVA_ONLY( balance_deeper_called++ );
        rc = balance_deeper(pPage, &pCur->apPage[1]);
        if( rc==SQLITE_OK ){
          pCur->iPage = 1;
          pCur->ix = 0;
          pCur->aiIdx[0] = 0;
          pCur->apPage[0] = pPage;
          pCur->pPage = pCur->apPage[1];
          assert( pCur->pPage->nOverflow );
        }
      }else{
        break;
      }
    }else if( sqlite3PagerPageRefcount(pPage->pDbPage)>1 ){
      /* The page being written is not a root page, and there is currently
      ** more than one reference to it. This only happens if the page is one
      ** of its own ancestor pages. Corruption. */
      rc = SQLITE_CORRUPT_BKPT;
    }else{
      MemPage * const pParent = pCur->apPage[iPage-1];
      int const iIdx = pCur->aiIdx[iPage-1];

      rc = sqlite3PagerWrite(pParent->pDbPage);
      if( rc==SQLITE_OK && pParent->nFree<0 ){
        rc = btreeComputeFreeSpace(pParent);
      }
      if( rc==SQLITE_OK ){
#ifndef SQLITE_OMIT_QUICKBALANCE
        if( pPage->intKeyLeaf
         && pPage->nOverflow==1
         && pPage->aiOvfl[0]==pPage->nCell
         && pParent->pgno!=1
         && pParent->nCell==iIdx
        ){
          /* Call balance_quick() to create a new sibling of pPage on which
          ** to store the overflow cell. balance_quick() inserts a new cell
          ** into pParent, which may cause pParent overflow. If this
          ** happens, the next iteration of the do-loop will balance pParent
          ** use either balance_nonroot() or balance_deeper(). Until this
          ** happens, the overflow cell is stored in the aBalanceQuickSpace[]
          ** buffer.
          **
          ** The purpose of the following assert() is to check that only a
          ** single call to balance_quick() is made for each call to this
          ** function. If this were not verified, a subtle bug involving reuse
          ** of the aBalanceQuickSpace[] might sneak in.
          */
          assert( balance_quick_called==0 );
          VVA_ONLY( balance_quick_called++ );
          rc = balance_quick(pParent, pPage, aBalanceQuickSpace);
        }else
#endif
        {
          /* In this case, call balance_nonroot() to redistribute cells
          ** between pPage and up to 2 of its sibling pages. This involves
          ** modifying the contents of pParent, which may cause pParent to
          ** become overfull or underfull. The next iteration of the do-loop
          ** will balance the parent page to correct this.
          **
          ** If the parent page becomes overfull, the overflow cell or cells
          ** are stored in the pSpace buffer allocated immediately below.
          ** A subsequent iteration of the do-loop will deal with this by
          ** calling balance_nonroot() (balance_deeper() may be called first,
          ** but it doesn't deal with overflow cells - just moves them to a
          ** different page). Once this subsequent call to balance_nonroot()
          ** has completed, it is safe to release the pSpace buffer used by
          ** the previous call, as the overflow cell data will have been
          ** copied either into the body of a database page or into the new
          ** pSpace buffer passed to the latter call to balance_nonroot().
          */
          u8 *pSpace = sqlite3PageMalloc(pCur->pBt->pageSize);
          rc = balance_nonroot(pParent, iIdx, pSpace, iPage==1,
                               pCur->hints&BTREE_BULKLOAD);
          if( pFree ){
            /* If pFree is not NULL, it points to the pSpace buffer used
            ** by a previous call to balance_nonroot(). Its contents are
            ** now stored either on real database pages or within the
            ** new pSpace buffer, so it may be safely freed here. */
            sqlite3PageFree(pFree);
          }

          /* The pSpace buffer will be freed after the next call to
          ** balance_nonroot(), or just before this function returns, whichever
          ** comes first. */
          pFree = pSpace;
        }
      }

      pPage->nOverflow = 0;

      /* The next iteration of the do-loop balances the parent page. */
      releasePage(pPage);
      pCur->iPage--;
      assert( pCur->iPage>=0 );
      pCur->pPage = pCur->apPage[pCur->iPage];
    }
  }while( rc==SQLITE_OK );

  if( pFree ){
    sqlite3PageFree(pFree);
  }
  return rc;
}

/* Overwrite content from pX into pDest.  Only do the write if the
** content is different from what is already there.
*/
static int btreeOverwriteContent(
  MemPage *pPage,           /* MemPage on which writing will occur */
  u8 *pDest,                /* Pointer to the place to start writing */
  const BtreePayload *pX,   /* Source of data to write */
  int iOffset,              /* Offset of first byte to write */
  int iAmt                  /* Number of bytes to be written */
){
  int nData = pX->nData - iOffset;
  if( nData<=0 ){
    /* Overwriting with zeros */
    int i;
    for(i=0; i<iAmt && pDest[i]==0; i++){}
    if( i<iAmt ){
      int rc = sqlite3PagerWrite(pPage->pDbPage);
      if( rc ) return rc;
      memset(pDest + i, 0, iAmt - i);
    }
  }else{
    if( nData<iAmt ){
      /* Mixed read data and zeros at the end.  Make a recursive call
      ** to write the zeros then fall through to write the real data */
      int rc = btreeOverwriteContent(pPage, pDest+nData, pX, iOffset+nData,
                                 iAmt-nData);
      if( rc ) return rc;
      iAmt = nData;
    }
    if( memcmp(pDest, ((u8*)pX->pData) + iOffset, iAmt)!=0 ){
      int rc = sqlite3PagerWrite(pPage->pDbPage);
      if( rc ) return rc;
      /* In a corrupt database, it is possible for the source and destination
      ** buffers to overlap.  This is harmless since the database is already
      ** corrupt but it does cause valgrind and ASAN warnings.  So use
      ** memmove(). */
      memmove(pDest, ((u8*)pX->pData) + iOffset, iAmt);
    }
  }
  return SQLITE_OK;
}

/*
** Overwrite the cell that cursor pCur is pointing to with fresh content
** contained in pX.  In this variant, pCur is pointing to an overflow
** cell.
*/
static SQLITE_NOINLINE int btreeOverwriteOverflowCell(
  BtCursor *pCur,                     /* Cursor pointing to cell to overwrite */
  const BtreePayload *pX              /* Content to write into the cell */
){
  int iOffset;                        /* Next byte of pX->pData to write */
  int nTotal = pX->nData + pX->nZero; /* Total bytes of to write */
  int rc;                             /* Return code */
  MemPage *pPage = pCur->pPage;       /* Page being written */
  BtShared *pBt;                      /* Btree */
  Pgno ovflPgno;                      /* Next overflow page to write */
  u32 ovflPageSize;                   /* Size to write on overflow page */

  assert( pCur->info.nLocal<nTotal );  /* pCur is an overflow cell */

  /* Overwrite the local portion first */
  rc = btreeOverwriteContent(pPage, pCur->info.pPayload, pX,
                             0, pCur->info.nLocal);
  if( rc ) return rc;

  /* Now overwrite the overflow pages */
  iOffset = pCur->info.nLocal;
  assert( nTotal>=0 );
  assert( iOffset>=0 );
  ovflPgno = get4byte(pCur->info.pPayload + iOffset);
  pBt = pPage->pBt;
  ovflPageSize = pBt->usableSize - 4;
  do{
    rc = btreeGetPage(pBt, ovflPgno, &pPage, 0);
    if( rc ) return rc;
    if( sqlite3PagerPageRefcount(pPage->pDbPage)!=1 || pPage->isInit ){
      rc = SQLITE_CORRUPT_BKPT;
    }else{
      if( iOffset+ovflPageSize<(u32)nTotal ){
        ovflPgno = get4byte(pPage->aData);
      }else{
        ovflPageSize = nTotal - iOffset;
      }
      rc = btreeOverwriteContent(pPage, pPage->aData+4, pX,
                                 iOffset, ovflPageSize);
    }
    sqlite3PagerUnref(pPage->pDbPage);
    if( rc ) return rc;
    iOffset += ovflPageSize;
  }while( iOffset<nTotal );
  return SQLITE_OK;   
}

/*
** Overwrite the cell that cursor pCur is pointing to with fresh content
** contained in pX.
*/
static int btreeOverwriteCell(BtCursor *pCur, const BtreePayload *pX){
  int nTotal = pX->nData + pX->nZero; /* Total bytes of to write */
  MemPage *pPage = pCur->pPage;       /* Page being written */

  if( pCur->info.pPayload + pCur->info.nLocal > pPage->aDataEnd
   || pCur->info.pPayload < pPage->aData + pPage->cellOffset
  ){
    return SQLITE_CORRUPT_BKPT;
  }
  if( pCur->info.nLocal==nTotal ){
    /* The entire cell is local */
    return btreeOverwriteContent(pPage, pCur->info.pPayload, pX,
                                 0, pCur->info.nLocal);
  }else{
    /* The cell contains overflow content */
    return btreeOverwriteOverflowCell(pCur, pX);
  }
}


/*
** Insert a new record into the BTree.  The content of the new record
** is described by the pX object.  The pCur cursor is used only to
** define what table the record should be inserted into, and is left
** pointing at a random location.
**
** For a table btree (used for rowid tables), only the pX.nKey value of
** the key is used. The pX.pKey value must be NULL.  The pX.nKey is the
** rowid or INTEGER PRIMARY KEY of the row.  The pX.nData,pData,nZero fields
** hold the content of the row.
**
** For an index btree (used for indexes and WITHOUT ROWID tables), the
** key is an arbitrary byte sequence stored in pX.pKey,nKey.  The
** pX.pData,nData,nZero fields must be zero.
**
** If the seekResult parameter is non-zero, then a successful call to
** sqlite3BtreeIndexMoveto() to seek cursor pCur to (pKey,nKey) has already
** been performed.  In other words, if seekResult!=0 then the cursor
** is currently pointing to a cell that will be adjacent to the cell
** to be inserted.  If seekResult<0 then pCur points to a cell that is
** smaller then (pKey,nKey).  If seekResult>0 then pCur points to a cell
** that is larger than (pKey,nKey).
**
** If seekResult==0, that means pCur is pointing at some unknown location.
** In that case, this routine must seek the cursor to the correct insertion
** point for (pKey,nKey) before doing the insertion.  For index btrees,
** if pX->nMem is non-zero, then pX->aMem contains pointers to the unpacked
** key values and pX->aMem can be used instead of pX->pKey to avoid having
** to decode the key.
*/
int sqlite3BtreeInsert(
  BtCursor *pCur,                /* Insert data into the table of this cursor */
  const BtreePayload *pX,        /* Content of the row to be inserted */
  int flags,                     /* True if this is likely an append */
  int seekResult                 /* Result of prior IndexMoveto() call */
){
  int rc;
  int loc = seekResult;          /* -1: before desired location  +1: after */
  int szNew = 0;
  int idx;
  MemPage *pPage;
  Btree *p = pCur->pBtree;
  unsigned char *oldCell;
  unsigned char *newCell = 0;

  assert( (flags & (BTREE_SAVEPOSITION|BTREE_APPEND|BTREE_PREFORMAT))==flags );
  assert( (flags & BTREE_PREFORMAT)==0 || seekResult || pCur->pKeyInfo==0 );

  /* Save the positions of any other cursors open on this table.
  **
  ** In some cases, the call to btreeMoveto() below is a no-op. For
  ** example, when inserting data into a table with auto-generated integer
  ** keys, the VDBE layer invokes sqlite3BtreeLast() to figure out the
  ** integer key to use. It then calls this function to actually insert the
  ** data into the intkey B-Tree. In this case btreeMoveto() recognizes
  ** that the cursor is already where it needs to be and returns without
  ** doing any work. To avoid thwarting these optimizations, it is important
  ** not to clear the cursor here.
  */
  if( pCur->curFlags & BTCF_Multiple ){
    rc = saveAllCursors(p->pBt, pCur->pgnoRoot, pCur);
    if( rc ) return rc;
    if( loc && pCur->iPage<0 ){
      /* This can only happen if the schema is corrupt such that there is more
      ** than one table or index with the same root page as used by the cursor.
      ** Which can only happen if the SQLITE_NoSchemaError flag was set when
      ** the schema was loaded. This cannot be asserted though, as a user might
      ** set the flag, load the schema, and then unset the flag.  */
      return SQLITE_CORRUPT_BKPT;
    }
  }

  /* Ensure that the cursor is not in the CURSOR_FAULT state and that it
  ** points to a valid cell.
  */
  if( pCur->eState>=CURSOR_REQUIRESEEK ){
    testcase( pCur->eState==CURSOR_REQUIRESEEK );
    testcase( pCur->eState==CURSOR_FAULT );
    rc = moveToRoot(pCur);
    if( rc && rc!=SQLITE_EMPTY ) return rc;
  }

  assert( cursorOwnsBtShared(pCur) );
  assert( (pCur->curFlags & BTCF_WriteFlag)!=0
              && p->pBt->inTransaction==TRANS_WRITE
              && (p->pBt->btsFlags & BTS_READ_ONLY)==0 );
  assert( hasSharedCacheTableLock(p, pCur->pgnoRoot, pCur->pKeyInfo!=0, 2) );

  /* Assert that the caller has been consistent. If this cursor was opened
  ** expecting an index b-tree, then the caller should be inserting blob
  ** keys with no associated data. If the cursor was opened expecting an
  ** intkey table, the caller should be inserting integer keys with a
  ** blob of associated data.  */
  assert( (flags & BTREE_PREFORMAT) || (pX->pKey==0)==(pCur->pKeyInfo==0) );

  if( pCur->pKeyInfo==0 ){
    assert( pX->pKey==0 );
    /* If this is an insert into a table b-tree, invalidate any incrblob
    ** cursors open on the row being replaced */
    if( p->hasIncrblobCur ){
      invalidateIncrblobCursors(p, pCur->pgnoRoot, pX->nKey, 0);
    }

    /* If BTREE_SAVEPOSITION is set, the cursor must already be pointing
    ** to a row with the same key as the new entry being inserted.
    */
#ifdef SQLITE_DEBUG
    if( flags & BTREE_SAVEPOSITION ){
      assert( pCur->curFlags & BTCF_ValidNKey );
      assert( pX->nKey==pCur->info.nKey );
      assert( loc==0 );
    }
#endif

    /* On the other hand, BTREE_SAVEPOSITION==0 does not imply
    ** that the cursor is not pointing to a row to be overwritten.
    ** So do a complete check.
    */
    if( (pCur->curFlags&BTCF_ValidNKey)!=0 && pX->nKey==pCur->info.nKey ){
      /* The cursor is pointing to the entry that is to be
      ** overwritten */
      assert( pX->nData>=0 && pX->nZero>=0 );
      if( pCur->info.nSize!=0
       && pCur->info.nPayload==(u32)pX->nData+pX->nZero
      ){
        /* New entry is the same size as the old.  Do an overwrite */
        return btreeOverwriteCell(pCur, pX);
      }
      assert( loc==0 );
    }else if( loc==0 ){
      /* The cursor is *not* pointing to the cell to be overwritten, nor
      ** to an adjacent cell.  Move the cursor so that it is pointing either
      ** to the cell to be overwritten or an adjacent cell.
      */
      rc = sqlite3BtreeTableMoveto(pCur, pX->nKey,
               (flags & BTREE_APPEND)!=0, &loc);
      if( rc ) return rc;
    }
  }else{
    /* This is an index or a WITHOUT ROWID table */

    /* If BTREE_SAVEPOSITION is set, the cursor must already be pointing
    ** to a row with the same key as the new entry being inserted.
    */
    assert( (flags & BTREE_SAVEPOSITION)==0 || loc==0 );

    /* If the cursor is not already pointing either to the cell to be
    ** overwritten, or if a new cell is being inserted, if the cursor is
    ** not pointing to an immediately adjacent cell, then move the cursor
    ** so that it does.
    */
    if( loc==0 && (flags & BTREE_SAVEPOSITION)==0 ){
      if( pX->nMem ){
        UnpackedRecord r;
        r.pKeyInfo = pCur->pKeyInfo;
        r.aMem = pX->aMem;
        r.nField = pX->nMem;
        r.default_rc = 0;
        r.eqSeen = 0;
        rc = sqlite3BtreeIndexMoveto(pCur, &r, &loc);
      }else{
        rc = btreeMoveto(pCur, pX->pKey, pX->nKey,
                    (flags & BTREE_APPEND)!=0, &loc);
      }
      if( rc ) return rc;
    }

    /* If the cursor is currently pointing to an entry to be overwritten
    ** and the new content is the same as as the old, then use the
    ** overwrite optimization.
    */
    if( loc==0 ){
      getCellInfo(pCur);
      if( pCur->info.nKey==pX->nKey ){
        BtreePayload x2;
        x2.pData = pX->pKey;
        x2.nData = pX->nKey;
        x2.nZero = 0;
        return btreeOverwriteCell(pCur, &x2);
      }
    }
  }
  assert( pCur->eState==CURSOR_VALID
       || (pCur->eState==CURSOR_INVALID && loc) || CORRUPT_DB );

  pPage = pCur->pPage;
  assert( pPage->intKey || pX->nKey>=0 || (flags & BTREE_PREFORMAT) );
  assert( pPage->leaf || !pPage->intKey );
  if( pPage->nFree<0 ){
    if( NEVER(pCur->eState>CURSOR_INVALID) ){
     /* ^^^^^--- due to the moveToRoot() call above */
      rc = SQLITE_CORRUPT_BKPT;
    }else{
      rc = btreeComputeFreeSpace(pPage);
    }
    if( rc ) return rc;
  }

  TRACE(("INSERT: table=%u nkey=%lld ndata=%u page=%u %s\n",
          pCur->pgnoRoot, pX->nKey, pX->nData, pPage->pgno,
          loc==0 ? "overwrite" : "new entry"));
  assert( pPage->isInit || CORRUPT_DB );
  newCell = p->pBt->pTmpSpace;
  assert( newCell!=0 );
  assert( BTREE_PREFORMAT==OPFLAG_PREFORMAT );
  if( flags & BTREE_PREFORMAT ){
    rc = SQLITE_OK;
    szNew = p->pBt->nPreformatSize;
    if( szNew<4 ){
      szNew = 4;
      newCell[3] = 0;
    }
    if( ISAUTOVACUUM(p->pBt) && szNew>pPage->maxLocal ){
      CellInfo info;
      pPage->xParseCell(pPage, newCell, &info);
      if( info.nPayload!=info.nLocal ){
        Pgno ovfl = get4byte(&newCell[szNew-4]);
        ptrmapPut(p->pBt, ovfl, PTRMAP_OVERFLOW1, pPage->pgno, &rc);
        if( NEVER(rc) ) goto end_insert;
      }
    }
  }else{
    rc = fillInCell(pPage, newCell, pX, &szNew);
    if( rc ) goto end_insert;
  }
  assert( szNew==pPage->xCellSize(pPage, newCell) );
  assert( szNew <= MX_CELL_SIZE(p->pBt) );
  idx = pCur->ix;
  pCur->info.nSize = 0;
  if( loc==0 ){
    CellInfo info;
    assert( idx>=0 );
    if( idx>=pPage->nCell ){
      return SQLITE_CORRUPT_BKPT;
    }
    rc = sqlite3PagerWrite(pPage->pDbPage);
    if( rc ){
      goto end_insert;
    }
    oldCell = findCell(pPage, idx);
    if( !pPage->leaf ){
      memcpy(newCell, oldCell, 4);
    }
    BTREE_CLEAR_CELL(rc, pPage, oldCell, info);
    testcase( pCur->curFlags & BTCF_ValidOvfl );
    invalidateOverflowCache(pCur);
    if( info.nSize==szNew && info.nLocal==info.nPayload
     && (!ISAUTOVACUUM(p->pBt) || szNew<pPage->minLocal)
    ){
      /* Overwrite the old cell with the new if they are the same size.
      ** We could also try to do this if the old cell is smaller, then add
      ** the leftover space to the free list.  But experiments show that
      ** doing that is no faster then skipping this optimization and just
      ** calling dropCell() and insertCell().
      **
      ** This optimization cannot be used on an autovacuum database if the
      ** new entry uses overflow pages, as the insertCell() call below is
      ** necessary to add the PTRMAP_OVERFLOW1 pointer-map entry.  */
      assert( rc==SQLITE_OK ); /* clearCell never fails when nLocal==nPayload */
      if( oldCell < pPage->aData+pPage->hdrOffset+10 ){
        return SQLITE_CORRUPT_BKPT;
      }
      if( oldCell+szNew > pPage->aDataEnd ){
        return SQLITE_CORRUPT_BKPT;
      }
      memcpy(oldCell, newCell, szNew);
      return SQLITE_OK;
    }
    dropCell(pPage, idx, info.nSize, &rc);
    if( rc ) goto end_insert;
  }else if( loc<0 && pPage->nCell>0 ){
    assert( pPage->leaf );
    idx = ++pCur->ix;
    pCur->curFlags &= ~BTCF_ValidNKey;
  }else{
    assert( pPage->leaf );
  }
  rc = insertCellFast(pPage, idx, newCell, szNew);
  assert( pPage->nOverflow==0 || rc==SQLITE_OK );
  assert( rc!=SQLITE_OK || pPage->nCell>0 || pPage->nOverflow>0 );

  /* If no error has occurred and pPage has an overflow cell, call balance()
  ** to redistribute the cells within the tree. Since balance() may move
  ** the cursor, zero the BtCursor.info.nSize and BTCF_ValidNKey
  ** variables.
  **
  ** Previous versions of SQLite called moveToRoot() to move the cursor
  ** back to the root page as balance() used to invalidate the contents
  ** of BtCursor.apPage[] and BtCursor.aiIdx[]. Instead of doing that,
  ** set the cursor state to "invalid". This makes common insert operations
  ** slightly faster.
  **
  ** There is a subtle but important optimization here too. When inserting
  ** multiple records into an intkey b-tree using a single cursor (as can
  ** happen while processing an "INSERT INTO ... SELECT" statement), it
  ** is advantageous to leave the cursor pointing to the last entry in
  ** the b-tree if possible. If the cursor is left pointing to the last
  ** entry in the table, and the next row inserted has an integer key
  ** larger than the largest existing key, it is possible to insert the
  ** row without seeking the cursor. This can be a big performance boost.
  */
  if( pPage->nOverflow ){
    assert( rc==SQLITE_OK );
    pCur->curFlags &= ~(BTCF_ValidNKey);
    rc = balance(pCur);

    /* Must make sure nOverflow is reset to zero even if the balance()
    ** fails. Internal data structure corruption will result otherwise.
    ** Also, set the cursor state to invalid. This stops saveCursorPosition()
    ** from trying to save the current position of the cursor.  */
    pCur->pPage->nOverflow = 0;
    pCur->eState = CURSOR_INVALID;
    if( (flags & BTREE_SAVEPOSITION) && rc==SQLITE_OK ){
      btreeReleaseAllCursorPages(pCur);
      if( pCur->pKeyInfo ){
        assert( pCur->pKey==0 );
        pCur->pKey = sqlite3Malloc( pX->nKey );
        if( pCur->pKey==0 ){
          rc = SQLITE_NOMEM;
        }else{
          memcpy(pCur->pKey, pX->pKey, pX->nKey);
        }
      }
      pCur->eState = CURSOR_REQUIRESEEK;
      pCur->nKey = pX->nKey;
    }
  }
  assert( pCur->iPage<0 || pCur->pPage->nOverflow==0 );

end_insert:
  return rc;
}

/*
** This function is used as part of copying the current row from cursor
** pSrc into cursor pDest. If the cursors are open on intkey tables, then
** parameter iKey is used as the rowid value when the record is copied
** into pDest. Otherwise, the record is copied verbatim.
**
** This function does not actually write the new value to cursor pDest.
** Instead, it creates and populates any required overflow pages and
** writes the data for the new cell into the BtShared.pTmpSpace buffer
** for the destination database. The size of the cell, in bytes, is left
** in BtShared.nPreformatSize. The caller completes the insertion by
** calling sqlite3BtreeInsert() with the BTREE_PREFORMAT flag specified.
**
** SQLITE_OK is returned if successful, or an SQLite error code otherwise.
*/
int sqlite3BtreeTransferRow(BtCursor *pDest, BtCursor *pSrc, i64 iKey){
  BtShared *pBt = pDest->pBt;
  u8 *aOut = pBt->pTmpSpace;    /* Pointer to next output buffer */
  const u8 *aIn;                /* Pointer to next input buffer */
  u32 nIn;                      /* Size of input buffer aIn[] */
  u32 nRem;                     /* Bytes of data still to copy */

  getCellInfo(pSrc);
  if( pSrc->info.nPayload<0x80 ){
    *(aOut++) = pSrc->info.nPayload;
  }else{
    aOut += sqlite3PutVarint(aOut, pSrc->info.nPayload);
  }
  if( pDest->pKeyInfo==0 ) aOut += putVarint(aOut, iKey);
  nIn = pSrc->info.nLocal;
  aIn = pSrc->info.pPayload;
  if( aIn+nIn>pSrc->pPage->aDataEnd ){
    return SQLITE_CORRUPT_BKPT;
  }
  nRem = pSrc->info.nPayload;
  if( nIn==nRem && nIn<pDest->pPage->maxLocal ){
    memcpy(aOut, aIn, nIn);
    pBt->nPreformatSize = nIn + (aOut - pBt->pTmpSpace);
    return SQLITE_OK;
  }else{
    int rc = SQLITE_OK;
    Pager *pSrcPager = pSrc->pBt->pPager;
    u8 *pPgnoOut = 0;
    Pgno ovflIn = 0;
    DbPage *pPageIn = 0;
    MemPage *pPageOut = 0;
    u32 nOut;                     /* Size of output buffer aOut[] */

    nOut = btreePayloadToLocal(pDest->pPage, pSrc->info.nPayload);
    pBt->nPreformatSize = nOut + (aOut - pBt->pTmpSpace);
    if( nOut<pSrc->info.nPayload ){
      pPgnoOut = &aOut[nOut];
      pBt->nPreformatSize += 4;
    }
 
    if( nRem>nIn ){
      if( aIn+nIn+4>pSrc->pPage->aDataEnd ){
        return SQLITE_CORRUPT_BKPT;
      }
      ovflIn = get4byte(&pSrc->info.pPayload[nIn]);
    }
 
    do {
      nRem -= nOut;
      do{
        assert( nOut>0 );
        if( nIn>0 ){
          int nCopy = MIN(nOut, nIn);
          memcpy(aOut, aIn, nCopy);
          nOut -= nCopy;
          nIn -= nCopy;
          aOut += nCopy;
          aIn += nCopy;
        }
        if( nOut>0 ){
          sqlite3PagerUnref(pPageIn);
          pPageIn = 0;
          rc = sqlite3PagerGet(pSrcPager, ovflIn, &pPageIn, PAGER_GET_READONLY);
          if( rc==SQLITE_OK ){
            aIn = (const u8*)sqlite3PagerGetData(pPageIn);
            ovflIn = get4byte(aIn);
            aIn += 4;
            nIn = pSrc->pBt->usableSize - 4;
          }
        }
      }while( rc==SQLITE_OK && nOut>0 );
 
      if( rc==SQLITE_OK && nRem>0 && ALWAYS(pPgnoOut) ){
        Pgno pgnoNew;
        MemPage *pNew = 0;
        rc = allocateBtreePage(pBt, &pNew, &pgnoNew, 0, 0);
        put4byte(pPgnoOut, pgnoNew);
        if( ISAUTOVACUUM(pBt) && pPageOut ){
          ptrmapPut(pBt, pgnoNew, PTRMAP_OVERFLOW2, pPageOut->pgno, &rc);
        }
        releasePage(pPageOut);
        pPageOut = pNew;
        if( pPageOut ){
          pPgnoOut = pPageOut->aData;
          put4byte(pPgnoOut, 0);
          aOut = &pPgnoOut[4];
          nOut = MIN(pBt->usableSize - 4, nRem);
        }
      }
    }while( nRem>0 && rc==SQLITE_OK );
 
    releasePage(pPageOut);
    sqlite3PagerUnref(pPageIn);
    return rc;
  }
}

/*
** Delete the entry that the cursor is pointing to.
**
** If the BTREE_SAVEPOSITION bit of the flags parameter is zero, then
** the cursor is left pointing at an arbitrary location after the delete.
** But if that bit is set, then the cursor is left in a state such that
** the next call to BtreeNext() or BtreePrev() moves it to the same row
** as it would have been on if the call to BtreeDelete() had been omitted.
**
** The BTREE_AUXDELETE bit of flags indicates that is one of several deletes
** associated with a single table entry and its indexes.  Only one of those
** deletes is considered the "primary" delete.  The primary delete occurs
** on a cursor that is not a BTREE_FORDELETE cursor.  All but one delete
** operation on non-FORDELETE cursors is tagged with the AUXDELETE flag.
** The BTREE_AUXDELETE bit is a hint that is not used by this implementation,
** but which might be used by alternative storage engines.
*/
int sqlite3BtreeDelete(BtCursor *pCur, u8 flags){
  Btree *p = pCur->pBtree;
  BtShared *pBt = p->pBt;             
  int rc;                    /* Return code */
  MemPage *pPage;            /* Page to delete cell from */
  unsigned char *pCell;      /* Pointer to cell to delete */
  int iCellIdx;              /* Index of cell to delete */
  int iCellDepth;            /* Depth of node containing pCell */
  CellInfo info;             /* Size of the cell being deleted */
  u8 bPreserve;              /* Keep cursor valid.  2 for CURSOR_SKIPNEXT */

  assert( cursorOwnsBtShared(pCur) );
  assert( pBt->inTransaction==TRANS_WRITE );
  assert( (pBt->btsFlags & BTS_READ_ONLY)==0 );
  assert( pCur->curFlags & BTCF_WriteFlag );
  assert( hasSharedCacheTableLock(p, pCur->pgnoRoot, pCur->pKeyInfo!=0, 2) );
  assert( !hasReadConflicts(p, pCur->pgnoRoot) );
  assert( (flags & ~(BTREE_SAVEPOSITION | BTREE_AUXDELETE))==0 );
  if( pCur->eState!=CURSOR_VALID ){
    if( pCur->eState>=CURSOR_REQUIRESEEK ){
      rc = btreeRestoreCursorPosition(pCur);
      assert( rc!=SQLITE_OK || CORRUPT_DB || pCur->eState==CURSOR_VALID );
      if( rc || pCur->eState!=CURSOR_VALID ) return rc;
    }else{
      return SQLITE_CORRUPT_BKPT;
    }
  }
  assert( pCur->eState==CURSOR_VALID );

  iCellDepth = pCur->iPage;
  iCellIdx = pCur->ix;
  pPage = pCur->pPage;
  if( pPage->nCell<=iCellIdx ){
    return SQLITE_CORRUPT_BKPT;
  }
  pCell = findCell(pPage, iCellIdx);
  if( pPage->nFree<0 && btreeComputeFreeSpace(pPage) ){
    return SQLITE_CORRUPT_BKPT;
  }
  if( pCell<&pPage->aCellIdx[pPage->nCell] ){
    return SQLITE_CORRUPT_BKPT;
  }

  /* If the BTREE_SAVEPOSITION bit is on, then the cursor position must
  ** be preserved following this delete operation. If the current delete
  ** will cause a b-tree rebalance, then this is done by saving the cursor
  ** key and leaving the cursor in CURSOR_REQUIRESEEK state before
  ** returning.
  **
  ** If the current delete will not cause a rebalance, then the cursor
  ** will be left in CURSOR_SKIPNEXT state pointing to the entry immediately
  ** before or after the deleted entry.
  **
  ** The bPreserve value records which path is required:
  **
  **    bPreserve==0         Not necessary to save the cursor position
  **    bPreserve==1         Use CURSOR_REQUIRESEEK to save the cursor position
  **    bPreserve==2         Cursor won't move.  Set CURSOR_SKIPNEXT.
  */
  bPreserve = (flags & BTREE_SAVEPOSITION)!=0;
  if( bPreserve ){
    if( !pPage->leaf
     || (pPage->nFree+pPage->xCellSize(pPage,pCell)+2) >
                                                   (int)(pBt->usableSize*2/3)
     || pPage->nCell==1  /* See dbfuzz001.test for a test case */
    ){
      /* A b-tree rebalance will be required after deleting this entry.
      ** Save the cursor key.  */
      rc = saveCursorKey(pCur);
      if( rc ) return rc;
    }else{
      bPreserve = 2;
    }
  }

  /* If the page containing the entry to delete is not a leaf page, move
  ** the cursor to the largest entry in the tree that is smaller than
  ** the entry being deleted. This cell will replace the cell being deleted
  ** from the internal node. The 'previous' entry is used for this instead
  ** of the 'next' entry, as the previous entry is always a part of the
  ** sub-tree headed by the child page of the cell being deleted. This makes
  ** balancing the tree following the delete operation easier.  */
  if( !pPage->leaf ){
    rc = sqlite3BtreePrevious(pCur, 0);
    assert( rc!=SQLITE_DONE );
    if( rc ) return rc;
  }

  /* Save the positions of any other cursors open on this table before
  ** making any modifications.  */
  if( pCur->curFlags & BTCF_Multiple ){
    rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur);
    if( rc ) return rc;
  }

  /* If this is a delete operation to remove a row from a table b-tree,
  ** invalidate any incrblob cursors open on the row being deleted.  */
  if( pCur->pKeyInfo==0 && p->hasIncrblobCur ){
    invalidateIncrblobCursors(p, pCur->pgnoRoot, pCur->info.nKey, 0);
  }

  /* Make the page containing the entry to be deleted writable. Then free any
  ** overflow pages associated with the entry and finally remove the cell
  ** itself from within the page.  */
  rc = sqlite3PagerWrite(pPage->pDbPage);
  if( rc ) return rc;
  BTREE_CLEAR_CELL(rc, pPage, pCell, info);
  dropCell(pPage, iCellIdx, info.nSize, &rc);
  if( rc ) return rc;

  /* If the cell deleted was not located on a leaf page, then the cursor
  ** is currently pointing to the largest entry in the sub-tree headed
  ** by the child-page of the cell that was just deleted from an internal
  ** node. The cell from the leaf node needs to be moved to the internal
  ** node to replace the deleted cell.  */
  if( !pPage->leaf ){
    MemPage *pLeaf = pCur->pPage;
    int nCell;
    Pgno n;
    unsigned char *pTmp;

    if( pLeaf->nFree<0 ){
      rc = btreeComputeFreeSpace(pLeaf);
      if( rc ) return rc;
    }
    if( iCellDepth<pCur->iPage-1 ){
      n = pCur->apPage[iCellDepth+1]->pgno;
    }else{
      n = pCur->pPage->pgno;
    }
    pCell = findCell(pLeaf, pLeaf->nCell-1);
    if( pCell<&pLeaf->aData[4] ) return SQLITE_CORRUPT_BKPT;
    nCell = pLeaf->xCellSize(pLeaf, pCell);
    assert( MX_CELL_SIZE(pBt) >= nCell );
    pTmp = pBt->pTmpSpace;
    assert( pTmp!=0 );
    rc = sqlite3PagerWrite(pLeaf->pDbPage);
    if( rc==SQLITE_OK ){
      rc = insertCell(pPage, iCellIdx, pCell-4, nCell+4, pTmp, n);
    }
    dropCell(pLeaf, pLeaf->nCell-1, nCell, &rc);
    if( rc ) return rc;
  }

  /* Balance the tree. If the entry deleted was located on a leaf page,
  ** then the cursor still points to that page. In this case the first
  ** call to balance() repairs the tree, and the if(...) condition is
  ** never true.
  **
  ** Otherwise, if the entry deleted was on an internal node page, then
  ** pCur is pointing to the leaf page from which a cell was removed to
  ** replace the cell deleted from the internal node. This is slightly
  ** tricky as the leaf node may be underfull, and the internal node may
  ** be either under or overfull. In this case run the balancing algorithm
  ** on the leaf node first. If the balance proceeds far enough up the
  ** tree that we can be sure that any problem in the internal node has
  ** been corrected, so be it. Otherwise, after balancing the leaf node,
  ** walk the cursor up the tree to the internal node and balance it as
  ** well.  */
  assert( pCur->pPage->nOverflow==0 );
  assert( pCur->pPage->nFree>=0 );
  if( pCur->pPage->nFree*3<=(int)pCur->pBt->usableSize*2 ){
    /* Optimization: If the free space is less than 2/3rds of the page,
    ** then balance() will always be a no-op.  No need to invoke it. */
    rc = SQLITE_OK;
  }else{
    rc = balance(pCur);
  }
  if( rc==SQLITE_OK && pCur->iPage>iCellDepth ){
    releasePageNotNull(pCur->pPage);
    pCur->iPage--;
    while( pCur->iPage>iCellDepth ){
      releasePage(pCur->apPage[pCur->iPage--]);
    }
    pCur->pPage = pCur->apPage[pCur->iPage];
    rc = balance(pCur);
  }

  if( rc==SQLITE_OK ){
    if( bPreserve>1 ){
      assert( (pCur->iPage==iCellDepth || CORRUPT_DB) );
      assert( pPage==pCur->pPage || CORRUPT_DB );
      assert( (pPage->nCell>0 || CORRUPT_DB) && iCellIdx<=pPage->nCell );
      pCur->eState = CURSOR_SKIPNEXT;
      if( iCellIdx>=pPage->nCell ){
        pCur->skipNext = -1;
        pCur->ix = pPage->nCell-1;
      }else{
        pCur->skipNext = 1;
      }
    }else{
      rc = moveToRoot(pCur);
      if( bPreserve ){
        btreeReleaseAllCursorPages(pCur);
        pCur->eState = CURSOR_REQUIRESEEK;
      }
      if( rc==SQLITE_EMPTY ) rc = SQLITE_OK;
    }
  }
  return rc;
}

/*
** Create a new BTree table.  Write into *piTable the page
** number for the root page of the new table.
**
** The type of type is determined by the flags parameter.  Only the
** following values of flags are currently in use.  Other values for
** flags might not work:
**
**     BTREE_INTKEY|BTREE_LEAFDATA     Used for SQL tables with rowid keys
**     BTREE_ZERODATA                  Used for SQL indices
*/
static int btreeCreateTable(Btree *p, Pgno *piTable, int createTabFlags){
  BtShared *pBt = p->pBt;
  MemPage *pRoot;
  Pgno pgnoRoot;
  int rc;
  int ptfFlags;          /* Page-type flags for the root page of new table */

  assert( sqlite3BtreeHoldsMutex(p) );
  assert( pBt->inTransaction==TRANS_WRITE );
  assert( (pBt->btsFlags & BTS_READ_ONLY)==0 );

#ifdef SQLITE_OMIT_AUTOVACUUM
  rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0);
  if( rc ){
    return rc;
  }
#else
  if( pBt->autoVacuum ){
    Pgno pgnoMove;      /* Move a page here to make room for the root-page */
    MemPage *pPageMove; /* The page to move to. */

    /* Creating a new table may probably require moving an existing database
    ** to make room for the new tables root page. In case this page turns
    ** out to be an overflow page, delete all overflow page-map caches
    ** held by open cursors.
    */
    invalidateAllOverflowCache(pBt);

    /* Read the value of meta[3] from the database to determine where the
    ** root page of the new table should go. meta[3] is the largest root-page
    ** created so far, so the new root-page is (meta[3]+1).
    */
    sqlite3BtreeGetMeta(p, BTREE_LARGEST_ROOT_PAGE, &pgnoRoot);
    if( pgnoRoot>btreePagecount(pBt) ){
      return SQLITE_CORRUPT_BKPT;
    }
    pgnoRoot++;

    /* The new root-page may not be allocated on a pointer-map page, or the
    ** PENDING_BYTE page.
    */
    while( pgnoRoot==PTRMAP_PAGENO(pBt, pgnoRoot) ||
        pgnoRoot==PENDING_BYTE_PAGE(pBt) ){
      pgnoRoot++;
    }
    assert( pgnoRoot>=3 );

    /* Allocate a page. The page that currently resides at pgnoRoot will
    ** be moved to the allocated page (unless the allocated page happens
    ** to reside at pgnoRoot).
    */
    rc = allocateBtreePage(pBt, &pPageMove, &pgnoMove, pgnoRoot, BTALLOC_EXACT);
    if( rc!=SQLITE_OK ){
      return rc;
    }

    if( pgnoMove!=pgnoRoot ){
      /* pgnoRoot is the page that will be used for the root-page of
      ** the new table (assuming an error did not occur). But we were
      ** allocated pgnoMove. If required (i.e. if it was not allocated
      ** by extending the file), the current page at position pgnoMove
      ** is already journaled.
      */
      u8 eType = 0;
      Pgno iPtrPage = 0;

      /* Save the positions of any open cursors. This is required in
      ** case they are holding a reference to an xFetch reference
      ** corresponding to page pgnoRoot.  */
      rc = saveAllCursors(pBt, 0, 0);
      releasePage(pPageMove);
      if( rc!=SQLITE_OK ){
        return rc;
      }

      /* Move the page currently at pgnoRoot to pgnoMove. */
      rc = btreeGetPage(pBt, pgnoRoot, &pRoot, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = ptrmapGet(pBt, pgnoRoot, &eType, &iPtrPage);
      if( eType==PTRMAP_ROOTPAGE || eType==PTRMAP_FREEPAGE ){
        rc = SQLITE_CORRUPT_BKPT;
      }
      if( rc!=SQLITE_OK ){
        releasePage(pRoot);
        return rc;
      }
      assert( eType!=PTRMAP_ROOTPAGE );
      assert( eType!=PTRMAP_FREEPAGE );
      rc = relocatePage(pBt, pRoot, eType, iPtrPage, pgnoMove, 0);
      releasePage(pRoot);

      /* Obtain the page at pgnoRoot */
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = btreeGetPage(pBt, pgnoRoot, &pRoot, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = sqlite3PagerWrite(pRoot->pDbPage);
      if( rc!=SQLITE_OK ){
        releasePage(pRoot);
        return rc;
      }
    }else{
      pRoot = pPageMove;
    }

    /* Update the pointer-map and meta-data with the new root-page number. */
    ptrmapPut(pBt, pgnoRoot, PTRMAP_ROOTPAGE, 0, &rc);
    if( rc ){
      releasePage(pRoot);
      return rc;
    }

    /* When the new root page was allocated, page 1 was made writable in
    ** order either to increase the database filesize, or to decrement the
    ** freelist count.  Hence, the sqlite3BtreeUpdateMeta() call cannot fail.
    */
    assert( sqlite3PagerIswriteable(pBt->pPage1->pDbPage) );
    rc = sqlite3BtreeUpdateMeta(p, 4, pgnoRoot);
    if( NEVER(rc) ){
      releasePage(pRoot);
      return rc;
    }

  }else{
    rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0);
    if( rc ) return rc;
  }
#endif
  assert( sqlite3PagerIswriteable(pRoot->pDbPage) );
  if( createTabFlags & BTREE_INTKEY ){
    ptfFlags = PTF_INTKEY | PTF_LEAFDATA | PTF_LEAF;
  }else{
    ptfFlags = PTF_ZERODATA | PTF_LEAF;
  }
  zeroPage(pRoot, ptfFlags);
  sqlite3PagerUnref(pRoot->pDbPage);
  assert( (pBt->openFlags & BTREE_SINGLE)==0 || pgnoRoot==2 );
  *piTable = pgnoRoot;
  return SQLITE_OK;
}
int sqlite3BtreeCreateTable(Btree *p, Pgno *piTable, int flags){
  int rc;
  sqlite3BtreeEnter(p);
  rc = btreeCreateTable(p, piTable, flags);
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** Erase the given database page and all its children.  Return
** the page to the freelist.
*/
static int clearDatabasePage(
  BtShared *pBt,           /* The BTree that contains the table */
  Pgno pgno,               /* Page number to clear */
  int freePageFlag,        /* Deallocate page if true */
  i64 *pnChange            /* Add number of Cells freed to this counter */
){
  MemPage *pPage;
  int rc;
  unsigned char *pCell;
  int i;
  int hdr;
  CellInfo info;

  assert( sqlite3_mutex_held(pBt->mutex) );
  if( pgno>btreePagecount(pBt) ){
    return SQLITE_CORRUPT_BKPT;
  }
  rc = getAndInitPage(pBt, pgno, &pPage, 0);
  if( rc ) return rc;
  if( (pBt->openFlags & BTREE_SINGLE)==0
   && sqlite3PagerPageRefcount(pPage->pDbPage) != (1 + (pgno==1))
  ){
    rc = SQLITE_CORRUPT_BKPT;
    goto cleardatabasepage_out;
  }
  hdr = pPage->hdrOffset;
  for(i=0; i<pPage->nCell; i++){
    pCell = findCell(pPage, i);
    if( !pPage->leaf ){
      rc = clearDatabasePage(pBt, get4byte(pCell), 1, pnChange);
      if( rc ) goto cleardatabasepage_out;
    }
    BTREE_CLEAR_CELL(rc, pPage, pCell, info);
    if( rc ) goto cleardatabasepage_out;
  }
  if( !pPage->leaf ){
    rc = clearDatabasePage(pBt, get4byte(&pPage->aData[hdr+8]), 1, pnChange);
    if( rc ) goto cleardatabasepage_out;
    if( pPage->intKey ) pnChange = 0;
  }
  if( pnChange ){
    testcase( !pPage->intKey );
    *pnChange += pPage->nCell;
  }
  if( freePageFlag ){
    freePage(pPage, &rc);
  }else if( (rc = sqlite3PagerWrite(pPage->pDbPage))==0 ){
    zeroPage(pPage, pPage->aData[hdr] | PTF_LEAF);
  }

cleardatabasepage_out:
  releasePage(pPage);
  return rc;
}

/*
** Delete all information from a single table in the database.  iTable is
** the page number of the root of the table.  After this routine returns,
** the root page is empty, but still exists.
**
** This routine will fail with SQLITE_LOCKED if there are any open
** read cursors on the table.  Open write cursors are moved to the
** root of the table.
**
** If pnChange is not NULL, then the integer value pointed to by pnChange
** is incremented by the number of entries in the table.
*/
int sqlite3BtreeClearTable(Btree *p, int iTable, i64 *pnChange){
  int rc;
  BtShared *pBt = p->pBt;
  sqlite3BtreeEnter(p);
  assert( p->inTrans==TRANS_WRITE );

  rc = saveAllCursors(pBt, (Pgno)iTable, 0);

  if( SQLITE_OK==rc ){
    /* Invalidate all incrblob cursors open on table iTable (assuming iTable
    ** is the root of a table b-tree - if it is not, the following call is
    ** a no-op).  */
    if( p->hasIncrblobCur ){
      invalidateIncrblobCursors(p, (Pgno)iTable, 0, 1);
    }
    rc = clearDatabasePage(pBt, (Pgno)iTable, 0, pnChange);
  }
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** Delete all information from the single table that pCur is open on.
**
** This routine only work for pCur on an ephemeral table.
*/
int sqlite3BtreeClearTableOfCursor(BtCursor *pCur){
  return sqlite3BtreeClearTable(pCur->pBtree, pCur->pgnoRoot, 0);
}

/*
** Erase all information in a table and add the root of the table to
** the freelist.  Except, the root of the principle table (the one on
** page 1) is never added to the freelist.
**
** This routine will fail with SQLITE_LOCKED if there are any open
** cursors on the table.
**
** If AUTOVACUUM is enabled and the page at iTable is not the last
** root page in the database file, then the last root page
** in the database file is moved into the slot formerly occupied by
** iTable and that last slot formerly occupied by the last root page
** is added to the freelist instead of iTable.  In this say, all
** root pages are kept at the beginning of the database file, which
** is necessary for AUTOVACUUM to work right.  *piMoved is set to the
** page number that used to be the last root page in the file before
** the move.  If no page gets moved, *piMoved is set to 0.
** The last root page is recorded in meta[3] and the value of
** meta[3] is updated by this procedure.
*/
static int btreeDropTable(Btree *p, Pgno iTable, int *piMoved){
  int rc;
  MemPage *pPage = 0;
  BtShared *pBt = p->pBt;

  assert( sqlite3BtreeHoldsMutex(p) );
  assert( p->inTrans==TRANS_WRITE );
  assert( iTable>=2 );
  if( iTable>btreePagecount(pBt) ){
    return SQLITE_CORRUPT_BKPT;
  }

  rc = sqlite3BtreeClearTable(p, iTable, 0);
  if( rc ) return rc;
  rc = btreeGetPage(pBt, (Pgno)iTable, &pPage, 0);
  if( NEVER(rc) ){
    releasePage(pPage);
    return rc;
  }

  *piMoved = 0;

#ifdef SQLITE_OMIT_AUTOVACUUM
  freePage(pPage, &rc);
  releasePage(pPage);
#else
  if( pBt->autoVacuum ){
    Pgno maxRootPgno;
    sqlite3BtreeGetMeta(p, BTREE_LARGEST_ROOT_PAGE, &maxRootPgno);

    if( iTable==maxRootPgno ){
      /* If the table being dropped is the table with the largest root-page
      ** number in the database, put the root page on the free list.
      */
      freePage(pPage, &rc);
      releasePage(pPage);
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }else{
      /* The table being dropped does not have the largest root-page
      ** number in the database. So move the page that does into the
      ** gap left by the deleted root-page.
      */
      MemPage *pMove;
      releasePage(pPage);
      rc = btreeGetPage(pBt, maxRootPgno, &pMove, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = relocatePage(pBt, pMove, PTRMAP_ROOTPAGE, 0, iTable, 0);
      releasePage(pMove);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      pMove = 0;
      rc = btreeGetPage(pBt, maxRootPgno, &pMove, 0);
      freePage(pMove, &rc);
      releasePage(pMove);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      *piMoved = maxRootPgno;
    }

    /* Set the new 'max-root-page' value in the database header. This
    ** is the old value less one, less one more if that happens to
    ** be a root-page number, less one again if that is the
    ** PENDING_BYTE_PAGE.
    */
    maxRootPgno--;
    while( maxRootPgno==PENDING_BYTE_PAGE(pBt)
           || PTRMAP_ISPAGE(pBt, maxRootPgno) ){
      maxRootPgno--;
    }
    assert( maxRootPgno!=PENDING_BYTE_PAGE(pBt) );

    rc = sqlite3BtreeUpdateMeta(p, 4, maxRootPgno);
  }else{
    freePage(pPage, &rc);
    releasePage(pPage);
  }
#endif
  return rc; 
}
int sqlite3BtreeDropTable(Btree *p, int iTable, int *piMoved){
  int rc;
  sqlite3BtreeEnter(p);
  rc = btreeDropTable(p, iTable, piMoved);
  sqlite3BtreeLeave(p);
  return rc;
}


/*
** This function may only be called if the b-tree connection already
** has a read or write transaction open on the database.
**
** Read the meta-information out of a database file.  Meta[0]
** is the number of free pages currently in the database.  Meta[1]
** through meta[15] are available for use by higher layers.  Meta[0]
** is read-only, the others are read/write.
**
** The schema layer numbers meta values differently.  At the schema
** layer (and the SetCookie and ReadCookie opcodes) the number of
** free pages is not visible.  So Cookie[0] is the same as Meta[1].
**
** This routine treats Meta[BTREE_DATA_VERSION] as a special case.  Instead
** of reading the value out of the header, it instead loads the "DataVersion"
** from the pager.  The BTREE_DATA_VERSION value is not actually stored in the
** database file.  It is a number computed by the pager.  But its access
** pattern is the same as header meta values, and so it is convenient to
** read it from this routine.
*/
void sqlite3BtreeGetMeta(Btree *p, int idx, u32 *pMeta){
  BtShared *pBt = p->pBt;

  sqlite3BtreeEnter(p);
  assert( p->inTrans>TRANS_NONE );
  assert( SQLITE_OK==querySharedCacheTableLock(p, SCHEMA_ROOT, READ_LOCK) );
  assert( pBt->pPage1 );
  assert( idx>=0 && idx<=15 );

  if( idx==BTREE_DATA_VERSION ){
    *pMeta = sqlite3PagerDataVersion(pBt->pPager) + p->iBDataVersion;
  }else{
    *pMeta = get4byte(&pBt->pPage1->aData[36 + idx*4]);
  }

  /* If auto-vacuum is disabled in this build and this is an auto-vacuum
  ** database, mark the database as read-only.  */
#ifdef SQLITE_OMIT_AUTOVACUUM
  if( idx==BTREE_LARGEST_ROOT_PAGE && *pMeta>0 ){
    pBt->btsFlags |= BTS_READ_ONLY;
  }
#endif

  sqlite3BtreeLeave(p);
}

/*
** Write meta-information back into the database.  Meta[0] is
** read-only and may not be written.
*/
int sqlite3BtreeUpdateMeta(Btree *p, int idx, u32 iMeta){
  BtShared *pBt = p->pBt;
  unsigned char *pP1;
  int rc;
  assert( idx>=1 && idx<=15 );
  sqlite3BtreeEnter(p);
  assert( p->inTrans==TRANS_WRITE );
  assert( pBt->pPage1!=0 );
  pP1 = pBt->pPage1->aData;
  rc = sqlite3PagerWrite(pBt->pPage1->pDbPage);
  if( rc==SQLITE_OK ){
    put4byte(&pP1[36 + idx*4], iMeta);
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( idx==BTREE_INCR_VACUUM ){
      assert( pBt->autoVacuum || iMeta==0 );
      assert( iMeta==0 || iMeta==1 );
      pBt->incrVacuum = (u8)iMeta;
    }
#endif
  }
  sqlite3BtreeLeave(p);
  return rc;
}

/*
** The first argument, pCur, is a cursor opened on some b-tree. Count the
** number of entries in the b-tree and write the result to *pnEntry.
**
** SQLITE_OK is returned if the operation is successfully executed.
** Otherwise, if an error is encountered (i.e. an IO error or database
** corruption) an SQLite error code is returned.
*/
int sqlite3BtreeCount(sqlite3 *db, BtCursor *pCur, i64 *pnEntry){
  i64 nEntry = 0;                      /* Value to return in *pnEntry */
  int rc;                              /* Return code */

  rc = moveToRoot(pCur);
  if( rc==SQLITE_EMPTY ){
    *pnEntry = 0;
    return SQLITE_OK;
  }

  /* Unless an error occurs, the following loop runs one iteration for each
  ** page in the B-Tree structure (not including overflow pages).
  */
  while( rc==SQLITE_OK && !AtomicLoad(&db->u1.isInterrupted) ){
    int iIdx;                          /* Index of child node in parent */
    MemPage *pPage;                    /* Current page of the b-tree */

    /* If this is a leaf page or the tree is not an int-key tree, then
    ** this page contains countable entries. Increment the entry counter
    ** accordingly.
    */
    pPage = pCur->pPage;
    if( pPage->leaf || !pPage->intKey ){
      nEntry += pPage->nCell;
    }

    /* pPage is a leaf node. This loop navigates the cursor so that it
    ** points to the first interior cell that it points to the parent of
    ** the next page in the tree that has not yet been visited. The
    ** pCur->aiIdx[pCur->iPage] value is set to the index of the parent cell
    ** of the page, or to the number of cells in the page if the next page
    ** to visit is the right-child of its parent.
    **
    ** If all pages in the tree have been visited, return SQLITE_OK to the
    ** caller.
    */
    if( pPage->leaf ){
      do {
        if( pCur->iPage==0 ){
          /* All pages of the b-tree have been visited. Return successfully. */
          *pnEntry = nEntry;
          return moveToRoot(pCur);
        }
        moveToParent(pCur);
      }while ( pCur->ix>=pCur->pPage->nCell );

      pCur->ix++;
      pPage = pCur->pPage;
    }

    /* Descend to the child node of the cell that the cursor currently
    ** points at. This is the right-child if (iIdx==pPage->nCell).
    */
    iIdx = pCur->ix;
    if( iIdx==pPage->nCell ){
      rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8]));
    }else{
      rc = moveToChild(pCur, get4byte(findCell(pPage, iIdx)));
    }
  }

  /* An error has occurred. Return an error code. */
  return rc;
}

/*
** Return the pager associated with a BTree.  This routine is used for
** testing and debugging only.
*/
Pager *sqlite3BtreePager(Btree *p){
  return p->pBt->pPager;
}

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** Record an OOM error during integrity_check
*/
static void checkOom(IntegrityCk *pCheck){
  pCheck->rc = SQLITE_NOMEM;
  pCheck->mxErr = 0;  /* Causes integrity_check processing to stop */
  if( pCheck->nErr==0 ) pCheck->nErr++;
}

/*
** Invoke the progress handler, if appropriate.  Also check for an
** interrupt.
*/
static void checkProgress(IntegrityCk *pCheck){
  sqlite3 *db = pCheck->db;
  if( AtomicLoad(&db->u1.isInterrupted) ){
    pCheck->rc = SQLITE_INTERRUPT;
    pCheck->nErr++;
    pCheck->mxErr = 0;
  }
#ifndef SQLITE_OMIT_PROGRESS_CALLBACK
  if( db->xProgress ){
    assert( db->nProgressOps>0 );
    pCheck->nStep++;
    if( (pCheck->nStep % db->nProgressOps)==0
     && db->xProgress(db->pProgressArg)
    ){
      pCheck->rc = SQLITE_INTERRUPT;
      pCheck->nErr++;
      pCheck->mxErr = 0;
    }
  }
#endif
}

/*
** Append a message to the error message string.
*/
static void checkAppendMsg(
  IntegrityCk *pCheck,
  const char *zFormat,
  ...
){
  va_list ap;
  checkProgress(pCheck);
  if( !pCheck->mxErr ) return;
  pCheck->mxErr--;
  pCheck->nErr++;
  va_start(ap, zFormat);
  if( pCheck->errMsg.nChar ){
    sqlite3_str_append(&pCheck->errMsg, "\n", 1);
  }
  if( pCheck->zPfx ){
    sqlite3_str_appendf(&pCheck->errMsg, pCheck->zPfx,
                        pCheck->v0, pCheck->v1, pCheck->v2);
  }
  sqlite3_str_vappendf(&pCheck->errMsg, zFormat, ap);
  va_end(ap);
  if( pCheck->errMsg.accError==SQLITE_NOMEM ){
    checkOom(pCheck);
  }
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

#ifndef SQLITE_OMIT_INTEGRITY_CHECK

/*
** Return non-zero if the bit in the IntegrityCk.aPgRef[] array that
** corresponds to page iPg is already set.
*/
static int getPageReferenced(IntegrityCk *pCheck, Pgno iPg){
  assert( pCheck->aPgRef!=0 );
  assert( iPg<=pCheck->nCkPage && sizeof(pCheck->aPgRef[0])==1 );
  return (pCheck->aPgRef[iPg/8] & (1 << (iPg & 0x07)));
}

/*
** Set the bit in the IntegrityCk.aPgRef[] array that corresponds to page iPg.
*/
static void setPageReferenced(IntegrityCk *pCheck, Pgno iPg){
  assert( pCheck->aPgRef!=0 );
  assert( iPg<=pCheck->nCkPage && sizeof(pCheck->aPgRef[0])==1 );
  pCheck->aPgRef[iPg/8] |= (1 << (iPg & 0x07));
}


/*
** Add 1 to the reference count for page iPage.  If this is the second
** reference to the page, add an error message to pCheck->zErrMsg.
** Return 1 if there are 2 or more references to the page and 0 if
** if this is the first reference to the page.
**
** Also check that the page number is in bounds.
*/
static int checkRef(IntegrityCk *pCheck, Pgno iPage){
  if( iPage>pCheck->nCkPage || iPage==0 ){
    checkAppendMsg(pCheck, "invalid page number %u", iPage);
    return 1;
  }
  if( getPageReferenced(pCheck, iPage) ){
    checkAppendMsg(pCheck, "2nd reference to page %u", iPage);
    return 1;
  }
  setPageReferenced(pCheck, iPage);
  return 0;
}

#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** Check that the entry in the pointer-map for page iChild maps to
** page iParent, pointer type ptrType. If not, append an error message
** to pCheck.
*/
static void checkPtrmap(
  IntegrityCk *pCheck,   /* Integrity check context */
  Pgno iChild,           /* Child page number */
  u8 eType,              /* Expected pointer map type */
  Pgno iParent           /* Expected pointer map parent page number */
){
  int rc;
  u8 ePtrmapType;
  Pgno iPtrmapParent;

  rc = ptrmapGet(pCheck->pBt, iChild, &ePtrmapType, &iPtrmapParent);
  if( rc!=SQLITE_OK ){
    if( rc==SQLITE_NOMEM || rc==SQLITE_IOERR_NOMEM ) checkOom(pCheck);
    checkAppendMsg(pCheck, "Failed to read ptrmap key=%u", iChild);
    return;
  }

  if( ePtrmapType!=eType || iPtrmapParent!=iParent ){
    checkAppendMsg(pCheck,
      "Bad ptr map entry key=%u expected=(%u,%u) got=(%u,%u)",
      iChild, eType, iParent, ePtrmapType, iPtrmapParent);
  }
}
#endif

/*
** Check the integrity of the freelist or of an overflow page list.
** Verify that the number of pages on the list is N.
*/
static void checkList(
  IntegrityCk *pCheck,  /* Integrity checking context */
  int isFreeList,       /* True for a freelist.  False for overflow page list */
  Pgno iPage,           /* Page number for first page in the list */
  u32 N                 /* Expected number of pages in the list */
){
  int i;
  u32 expected = N;
  int nErrAtStart = pCheck->nErr;
  while( iPage!=0 && pCheck->mxErr ){
    DbPage *pOvflPage;
    unsigned char *pOvflData;
    if( checkRef(pCheck, iPage) ) break;
    N--;
    if( sqlite3PagerGet(pCheck->pPager, (Pgno)iPage, &pOvflPage, 0) ){
      checkAppendMsg(pCheck, "failed to get page %u", iPage);
      break;
    }
    pOvflData = (unsigned char *)sqlite3PagerGetData(pOvflPage);
    if( isFreeList ){
      u32 n = (u32)get4byte(&pOvflData[4]);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pCheck->pBt->autoVacuum ){
        checkPtrmap(pCheck, iPage, PTRMAP_FREEPAGE, 0);
      }
#endif
      if( n>pCheck->pBt->usableSize/4-2 ){
        checkAppendMsg(pCheck,
           "freelist leaf count too big on page %u", iPage);
        N--;
      }else{
        for(i=0; i<(int)n; i++){
          Pgno iFreePage = get4byte(&pOvflData[8+i*4]);
#ifndef SQLITE_OMIT_AUTOVACUUM
          if( pCheck->pBt->autoVacuum ){
            checkPtrmap(pCheck, iFreePage, PTRMAP_FREEPAGE, 0);
          }
#endif
          checkRef(pCheck, iFreePage);
        }
        N -= n;
      }
    }
#ifndef SQLITE_OMIT_AUTOVACUUM
    else{
      /* If this database supports auto-vacuum and iPage is not the last
      ** page in this overflow list, check that the pointer-map entry for
      ** the following page matches iPage.
      */
      if( pCheck->pBt->autoVacuum && N>0 ){
        i = get4byte(pOvflData);
        checkPtrmap(pCheck, i, PTRMAP_OVERFLOW2, iPage);
      }
    }
#endif
    iPage = get4byte(pOvflData);
    sqlite3PagerUnref(pOvflPage);
  }
  if( N && nErrAtStart==pCheck->nErr ){
    checkAppendMsg(pCheck,
      "%s is %u but should be %u",
      isFreeList ? "size" : "overflow list length",
      expected-N, expected);
  }
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

/*
** An implementation of a min-heap.
**
** aHeap[0] is the number of elements on the heap.  aHeap[1] is the
** root element.  The daughter nodes of aHeap[N] are aHeap[N*2]
** and aHeap[N*2+1].
**
** The heap property is this:  Every node is less than or equal to both
** of its daughter nodes.  A consequence of the heap property is that the
** root node aHeap[1] is always the minimum value currently in the heap.
**
** The btreeHeapInsert() routine inserts an unsigned 32-bit number onto
** the heap, preserving the heap property.  The btreeHeapPull() routine
** removes the root element from the heap (the minimum value in the heap)
** and then moves other nodes around as necessary to preserve the heap
** property.
**
** This heap is used for cell overlap and coverage testing.  Each u32
** entry represents the span of a cell or freeblock on a btree page. 
** The upper 16 bits are the index of the first byte of a range and the
** lower 16 bits are the index of the last byte of that range.
*/
static void btreeHeapInsert(u32 *aHeap, u32 x){
  u32 j, i;
  assert( aHeap!=0 );
  i = ++aHeap[0];
  aHeap[i] = x;
  while( (j = i/2)>0 && aHeap[j]>aHeap[i] ){
    x = aHeap[j];
    aHeap[j] = aHeap[i];
    aHeap[i] = x;
    i = j;
  }
}
static int btreeHeapPull(u32 *aHeap, u32 *pOut){
  u32 j, i, x;
  if( (x = aHeap[0])==0 ) return 0;
  *pOut = aHeap[1];
  aHeap[1] = aHeap[x];
  aHeap[x] = 0xffffffff;
  aHeap[0]--;
  i = 1;
  while( (j = i*2)<=aHeap[0] ){
    if( aHeap[j]>aHeap[j+1] ) j++;
    if( aHeap[i]<aHeap[j] ) break;
    x = aHeap[i];
    aHeap[i] = aHeap[j];
    aHeap[j] = x;
    i = j;
  }
  return 1; 
}

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** Do various sanity checks on a single page of a tree.  Return
** the tree depth.  Root pages return 0.  Parents of root pages
** return 1, and so forth.
**
** These checks are done:
**
**      1.  Make sure that cells and freeblocks do not overlap
**          but combine to completely cover the page.
**      2.  Make sure integer cell keys are in order.
**      3.  Check the integrity of overflow pages.
**      4.  Recursively call checkTreePage on all children.
**      5.  Verify that the depth of all children is the same.
*/
static int checkTreePage(
  IntegrityCk *pCheck,  /* Context for the sanity check */
  Pgno iPage,           /* Page number of the page to check */
  i64 *piMinKey,        /* Write minimum integer primary key here */
  i64 maxKey            /* Error if integer primary key greater than this */
){
  MemPage *pPage = 0;      /* The page being analyzed */
  int i;                   /* Loop counter */
  int rc;                  /* Result code from subroutine call */
  int depth = -1, d2;      /* Depth of a subtree */
  int pgno;                /* Page number */
  int nFrag;               /* Number of fragmented bytes on the page */
  int hdr;                 /* Offset to the page header */
  int cellStart;           /* Offset to the start of the cell pointer array */
  int nCell;               /* Number of cells */
  int doCoverageCheck = 1; /* True if cell coverage checking should be done */
  int keyCanBeEqual = 1;   /* True if IPK can be equal to maxKey
                           ** False if IPK must be strictly less than maxKey */
  u8 *data;                /* Page content */
  u8 *pCell;               /* Cell content */
  u8 *pCellIdx;            /* Next element of the cell pointer array */
  BtShared *pBt;           /* The BtShared object that owns pPage */
  u32 pc;                  /* Address of a cell */
  u32 usableSize;          /* Usable size of the page */
  u32 contentOffset;       /* Offset to the start of the cell content area */
  u32 *heap = 0;           /* Min-heap used for checking cell coverage */
  u32 x, prev = 0;         /* Next and previous entry on the min-heap */
  const char *saved_zPfx = pCheck->zPfx;
  int saved_v1 = pCheck->v1;
  int saved_v2 = pCheck->v2;
  u8 savedIsInit = 0;

  /* Check that the page exists
  */
  checkProgress(pCheck);
  if( pCheck->mxErr==0 ) goto end_of_check;
  pBt = pCheck->pBt;
  usableSize = pBt->usableSize;
  if( iPage==0 ) return 0;
  if( checkRef(pCheck, iPage) ) return 0;
  pCheck->zPfx = "Tree %u page %u: ";
  pCheck->v1 = iPage;
  if( (rc = btreeGetPage(pBt, iPage, &pPage, 0))!=0 ){
    checkAppendMsg(pCheck,
       "unable to get the page. error code=%d", rc);
    if( rc==SQLITE_IOERR_NOMEM ) pCheck->rc = SQLITE_NOMEM;
    goto end_of_check;
  }

  /* Clear MemPage.isInit to make sure the corruption detection code in
  ** btreeInitPage() is executed.  */
  savedIsInit = pPage->isInit;
  pPage->isInit = 0;
  if( (rc = btreeInitPage(pPage))!=0 ){
    assert( rc==SQLITE_CORRUPT );  /* The only possible error from InitPage */
    checkAppendMsg(pCheck,
                   "btreeInitPage() returns error code %d", rc);
    goto end_of_check;
  }
  if( (rc = btreeComputeFreeSpace(pPage))!=0 ){
    assert( rc==SQLITE_CORRUPT );
    checkAppendMsg(pCheck, "free space corruption", rc);
    goto end_of_check;
  }
  data = pPage->aData;
  hdr = pPage->hdrOffset;

  /* Set up for cell analysis */
  pCheck->zPfx = "Tree %u page %u cell %u: ";
  contentOffset = get2byteNotZero(&data[hdr+5]);
  assert( contentOffset<=usableSize );  /* Enforced by btreeInitPage() */

  /* EVIDENCE-OF: R-37002-32774 The two-byte integer at offset 3 gives the
  ** number of cells on the page. */
  nCell = get2byte(&data[hdr+3]);
  assert( pPage->nCell==nCell );

  /* EVIDENCE-OF: R-23882-45353 The cell pointer array of a b-tree page
  ** immediately follows the b-tree page header. */
  cellStart = hdr + 12 - 4*pPage->leaf;
  assert( pPage->aCellIdx==&data[cellStart] );
  pCellIdx = &data[cellStart + 2*(nCell-1)];

  if( !pPage->leaf ){
    /* Analyze the right-child page of internal pages */
    pgno = get4byte(&data[hdr+8]);
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum ){
      pCheck->zPfx = "Tree %u page %u right child: ";
      checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage);
    }
#endif
    depth = checkTreePage(pCheck, pgno, &maxKey, maxKey);
    keyCanBeEqual = 0;
  }else{
    /* For leaf pages, the coverage check will occur in the same loop
    ** as the other cell checks, so initialize the heap.  */
    heap = pCheck->heap;
    heap[0] = 0;
  }

  /* EVIDENCE-OF: R-02776-14802 The cell pointer array consists of K 2-byte
  ** integer offsets to the cell contents. */
  for(i=nCell-1; i>=0 && pCheck->mxErr; i--){
    CellInfo info;

    /* Check cell size */
    pCheck->v2 = i;
    assert( pCellIdx==&data[cellStart + i*2] );
    pc = get2byteAligned(pCellIdx);
    pCellIdx -= 2;
    if( pc<contentOffset || pc>usableSize-4 ){
      checkAppendMsg(pCheck, "Offset %u out of range %u..%u",
                             pc, contentOffset, usableSize-4);
      doCoverageCheck = 0;
      continue;
    }
    pCell = &data[pc];
    pPage->xParseCell(pPage, pCell, &info);
    if( pc+info.nSize>usableSize ){
      checkAppendMsg(pCheck, "Extends off end of page");
      doCoverageCheck = 0;
      continue;
    }

    /* Check for integer primary key out of range */
    if( pPage->intKey ){
      if( keyCanBeEqual ? (info.nKey > maxKey) : (info.nKey >= maxKey) ){
        checkAppendMsg(pCheck, "Rowid %lld out of order", info.nKey);
      }
      maxKey = info.nKey;
      keyCanBeEqual = 0;     /* Only the first key on the page may ==maxKey */
    }

    /* Check the content overflow list */
    if( info.nPayload>info.nLocal ){
      u32 nPage;       /* Number of pages on the overflow chain */
      Pgno pgnoOvfl;   /* First page of the overflow chain */
      assert( pc + info.nSize - 4 <= usableSize );
      nPage = (info.nPayload - info.nLocal + usableSize - 5)/(usableSize - 4);
      pgnoOvfl = get4byte(&pCell[info.nSize - 4]);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pBt->autoVacuum ){
        checkPtrmap(pCheck, pgnoOvfl, PTRMAP_OVERFLOW1, iPage);
      }
#endif
      checkList(pCheck, 0, pgnoOvfl, nPage);
    }

    if( !pPage->leaf ){
      /* Check sanity of left child page for internal pages */
      pgno = get4byte(pCell);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pBt->autoVacuum ){
        checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage);
      }
#endif
      d2 = checkTreePage(pCheck, pgno, &maxKey, maxKey);
      keyCanBeEqual = 0;
      if( d2!=depth ){
        checkAppendMsg(pCheck, "Child page depth differs");
        depth = d2;
      }
    }else{
      /* Populate the coverage-checking heap for leaf pages */
      btreeHeapInsert(heap, (pc<<16)|(pc+info.nSize-1));
    }
  }
  *piMinKey = maxKey;

  /* Check for complete coverage of the page
  */
  pCheck->zPfx = 0;
  if( doCoverageCheck && pCheck->mxErr>0 ){
    /* For leaf pages, the min-heap has already been initialized and the
    ** cells have already been inserted.  But for internal pages, that has
    ** not yet been done, so do it now */
    if( !pPage->leaf ){
      heap = pCheck->heap;
      heap[0] = 0;
      for(i=nCell-1; i>=0; i--){
        u32 size;
        pc = get2byteAligned(&data[cellStart+i*2]);
        size = pPage->xCellSize(pPage, &data[pc]);
        btreeHeapInsert(heap, (pc<<16)|(pc+size-1));
      }
    }
    /* Add the freeblocks to the min-heap
    **
    ** EVIDENCE-OF: R-20690-50594 The second field of the b-tree page header
    ** is the offset of the first freeblock, or zero if there are no
    ** freeblocks on the page.
    */
    i = get2byte(&data[hdr+1]);
    while( i>0 ){
      int size, j;
      assert( (u32)i<=usableSize-4 ); /* Enforced by btreeComputeFreeSpace() */
      size = get2byte(&data[i+2]);
      assert( (u32)(i+size)<=usableSize ); /* due to btreeComputeFreeSpace() */
      btreeHeapInsert(heap, (((u32)i)<<16)|(i+size-1));
      /* EVIDENCE-OF: R-58208-19414 The first 2 bytes of a freeblock are a
      ** big-endian integer which is the offset in the b-tree page of the next
      ** freeblock in the chain, or zero if the freeblock is the last on the
      ** chain. */
      j = get2byte(&data[i]);
      /* EVIDENCE-OF: R-06866-39125 Freeblocks are always connected in order of
      ** increasing offset. */
      assert( j==0 || j>i+size );     /* Enforced by btreeComputeFreeSpace() */
      assert( (u32)j<=usableSize-4 ); /* Enforced by btreeComputeFreeSpace() */
      i = j;
    }
    /* Analyze the min-heap looking for overlap between cells and/or
    ** freeblocks, and counting the number of untracked bytes in nFrag.
    **
    ** Each min-heap entry is of the form:    (start_address<<16)|end_address.
    ** There is an implied first entry the covers the page header, the cell
    ** pointer index, and the gap between the cell pointer index and the start
    ** of cell content. 
    **
    ** The loop below pulls entries from the min-heap in order and compares
    ** the start_address against the previous end_address.  If there is an
    ** overlap, that means bytes are used multiple times.  If there is a gap,
    ** that gap is added to the fragmentation count.
    */
    nFrag = 0;
    prev = contentOffset - 1;   /* Implied first min-heap entry */
    while( btreeHeapPull(heap,&x) ){
      if( (prev&0xffff)>=(x>>16) ){
        checkAppendMsg(pCheck,
          "Multiple uses for byte %u of page %u", x>>16, iPage);
        break;
      }else{
        nFrag += (x>>16) - (prev&0xffff) - 1;
        prev = x;
      }
    }
    nFrag += usableSize - (prev&0xffff) - 1;
    /* EVIDENCE-OF: R-43263-13491 The total number of bytes in all fragments
    ** is stored in the fifth field of the b-tree page header.
    ** EVIDENCE-OF: R-07161-27322 The one-byte integer at offset 7 gives the
    ** number of fragmented free bytes within the cell content area.
    */
    if( heap[0]==0 && nFrag!=data[hdr+7] ){
      checkAppendMsg(pCheck,
          "Fragmentation of %u bytes reported as %u on page %u",
          nFrag, data[hdr+7], iPage);
    }
  }

end_of_check:
  if( !doCoverageCheck ) pPage->isInit = savedIsInit;
  releasePage(pPage);
  pCheck->zPfx = saved_zPfx;
  pCheck->v1 = saved_v1;
  pCheck->v2 = saved_v2;
  return depth+1;
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** This routine does a complete check of the given BTree file.  aRoot[] is
** an array of pages numbers were each page number is the root page of
** a table.  nRoot is the number of entries in aRoot.
**
** A read-only or read-write transaction must be opened before calling
** this function.
**
** Write the number of error seen in *pnErr.  Except for some memory
** allocation errors,  an error message held in memory obtained from
** malloc is returned if *pnErr is non-zero.  If *pnErr==0 then NULL is
** returned.  If a memory allocation error occurs, NULL is returned.
**
** If the first entry in aRoot[] is 0, that indicates that the list of
** root pages is incomplete.  This is a "partial integrity-check".  This
** happens when performing an integrity check on a single table.  The
** zero is skipped, of course.  But in addition, the freelist checks
** and the checks to make sure every page is referenced are also skipped,
** since obviously it is not possible to know which pages are covered by
** the unverified btrees.  Except, if aRoot[1] is 1, then the freelist
** checks are still performed.
*/
int sqlite3BtreeIntegrityCheck(
  sqlite3 *db,  /* Database connection that is running the check */
  Btree *p,     /* The btree to be checked */
  Pgno *aRoot,  /* An array of root pages numbers for individual trees */
  int nRoot,    /* Number of entries in aRoot[] */
  int mxErr,    /* Stop reporting errors after this many */
  int *pnErr,   /* OUT: Write number of errors seen to this variable */
  char **pzOut  /* OUT: Write the error message string here */
){
  Pgno i;
  IntegrityCk sCheck;
  BtShared *pBt = p->pBt;
  u64 savedDbFlags = pBt->db->flags;
  char zErr[100];
  int bPartial = 0;            /* True if not checking all btrees */
  int bCkFreelist = 1;         /* True to scan the freelist */
  VVA_ONLY( int nRef );
  assert( nRoot>0 );

  /* aRoot[0]==0 means this is a partial check */
  if( aRoot[0]==0 ){
    assert( nRoot>1 );
    bPartial = 1;
    if( aRoot[1]!=1 ) bCkFreelist = 0;
  }

  sqlite3BtreeEnter(p);
  assert( p->inTrans>TRANS_NONE && pBt->inTransaction>TRANS_NONE );
  VVA_ONLY( nRef = sqlite3PagerRefcount(pBt->pPager) );
  assert( nRef>=0 );
  memset(&sCheck, 0, sizeof(sCheck));
  sCheck.db = db;
  sCheck.pBt = pBt;
  sCheck.pPager = pBt->pPager;
  sCheck.nCkPage = btreePagecount(sCheck.pBt);
  sCheck.mxErr = mxErr;
  sqlite3StrAccumInit(&sCheck.errMsg, 0, zErr, sizeof(zErr), SQLITE_MAX_LENGTH);
  sCheck.errMsg.printfFlags = SQLITE_PRINTF_INTERNAL;
  if( sCheck.nCkPage==0 ){
    goto integrity_ck_cleanup;
  }

  sCheck.aPgRef = sqlite3MallocZero((sCheck.nCkPage / 8)+ 1);
  if( !sCheck.aPgRef ){
    checkOom(&sCheck);
    goto integrity_ck_cleanup;
  }
  sCheck.heap = (u32*)sqlite3PageMalloc( pBt->pageSize );
  if( sCheck.heap==0 ){
    checkOom(&sCheck);
    goto integrity_ck_cleanup;
  }

  i = PENDING_BYTE_PAGE(pBt);
  if( i<=sCheck.nCkPage ) setPageReferenced(&sCheck, i);

  /* Check the integrity of the freelist
  */
  if( bCkFreelist ){
    sCheck.zPfx = "Freelist: ";
    checkList(&sCheck, 1, get4byte(&pBt->pPage1->aData[32]),
              get4byte(&pBt->pPage1->aData[36]));
    sCheck.zPfx = 0;
  }

  /* Check all the tables.
  */
#ifndef SQLITE_OMIT_AUTOVACUUM
  if( !bPartial ){
    if( pBt->autoVacuum ){
      Pgno mx = 0;
      Pgno mxInHdr;
      for(i=0; (int)i<nRoot; i++) if( mx<aRoot[i] ) mx = aRoot[i];
      mxInHdr = get4byte(&pBt->pPage1->aData[52]);
      if( mx!=mxInHdr ){
        checkAppendMsg(&sCheck,
          "max rootpage (%u) disagrees with header (%u)",
          mx, mxInHdr
        );
      }
    }else if( get4byte(&pBt->pPage1->aData[64])!=0 ){
      checkAppendMsg(&sCheck,
        "incremental_vacuum enabled with a max rootpage of zero"
      );
    }
  }
#endif
  testcase( pBt->db->flags & SQLITE_CellSizeCk );
  pBt->db->flags &= ~(u64)SQLITE_CellSizeCk;
  for(i=0; (int)i<nRoot && sCheck.mxErr; i++){
    i64 notUsed;
    if( aRoot[i]==0 ) continue;
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum && aRoot[i]>1 && !bPartial ){
      checkPtrmap(&sCheck, aRoot[i], PTRMAP_ROOTPAGE, 0);
    }
#endif
    sCheck.v0 = aRoot[i];
    checkTreePage(&sCheck, aRoot[i], &notUsed, LARGEST_INT64);
  }
  pBt->db->flags = savedDbFlags;

  /* Make sure every page in the file is referenced
  */
  if( !bPartial ){
    for(i=1; i<=sCheck.nCkPage && sCheck.mxErr; i++){
#ifdef SQLITE_OMIT_AUTOVACUUM
      if( getPageReferenced(&sCheck, i)==0 ){
        checkAppendMsg(&sCheck, "Page %u: never used", i);
      }
#else
      /* If the database supports auto-vacuum, make sure no tables contain
      ** references to pointer-map pages.
      */
      if( getPageReferenced(&sCheck, i)==0 &&
         (PTRMAP_PAGENO(pBt, i)!=i || !pBt->autoVacuum) ){
        checkAppendMsg(&sCheck, "Page %u: never used", i);
      }
      if( getPageReferenced(&sCheck, i)!=0 &&
         (PTRMAP_PAGENO(pBt, i)==i && pBt->autoVacuum) ){
        checkAppendMsg(&sCheck, "Page %u: pointer map referenced", i);
      }
#endif
    }
  }

  /* Clean  up and report errors.
  */
integrity_ck_cleanup:
  sqlite3PageFree(sCheck.heap);
  sqlite3_free(sCheck.aPgRef);
  *pnErr = sCheck.nErr;
  if( sCheck.nErr==0 ){
    sqlite3_str_reset(&sCheck.errMsg);
    *pzOut = 0;
  }else{
    *pzOut = sqlite3StrAccumFinish(&sCheck.errMsg);
  }
  /* Make sure this analysis did not leave any unref() pages. */
  assert( nRef==sqlite3PagerRefcount(pBt->pPager) );
  sqlite3BtreeLeave(p);
  return sCheck.rc;
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

/*
** Return the full pathname of the underlying database file.  Return
** an empty string if the database is in-memory or a TEMP database.
**
** The pager filename is invariant as long as the pager is
** open so it is safe to access without the BtShared mutex.
*/
const char *sqlite3BtreeGetFilename(Btree *p){
  assert( p->pBt->pPager!=0 );
  return sqlite3PagerFilename(p->pBt->pPager, 1);
}

/*
** Return the pathname of the journal file for this database. The return
** value of this routine is the same regardless of whether the journal file
** has been created or not.
**
** The pager journal filename is invariant as long as the pager is
** open so it is safe to access without the BtShared mutex.
*/
const char *sqlite3BtreeGetJournalname(Btree *p){
  assert( p->pBt->pPager!=0 );
  return sqlite3PagerJournalname(p->pBt->pPager);
}

/*
** Return one of SQLITE_TXN_NONE, SQLITE_TXN_READ, or SQLITE_TXN_WRITE
** to describe the current transaction state of Btree p.
*/
int sqlite3BtreeTxnState(Btree *p){
  assert( p==0 || sqlite3_mutex_held(p->db->mutex) );
  return p ? p->inTrans : 0;
}

#ifndef SQLITE_OMIT_WAL
/*
** Run a checkpoint on the Btree passed as the first argument.
**
** Return SQLITE_LOCKED if this or any other connection has an open
** transaction on the shared-cache the argument Btree is connected to.
**
** Parameter eMode is one of SQLITE_CHECKPOINT_PASSIVE, FULL or RESTART.
*/
int sqlite3BtreeCheckpoint(Btree *p, int eMode, int *pnLog, int *pnCkpt){
  int rc = SQLITE_OK;
  if( p ){
    BtShared *pBt = p->pBt;
    sqlite3BtreeEnter(p);
    if( pBt->inTransaction!=TRANS_NONE ){
      rc = SQLITE_LOCKED;
    }else{
      rc = sqlite3PagerCheckpoint(pBt->pPager, p->db, eMode, pnLog, pnCkpt);
    }
    sqlite3BtreeLeave(p);
  }
  return rc;
}
#endif

/*
** Return true if there is currently a backup running on Btree p.
*/
int sqlite3BtreeIsInBackup(Btree *p){
  assert( p );
  assert( sqlite3_mutex_held(p->db->mutex) );
  return p->nBackup!=0;
}

/*
** This function returns a pointer to a blob of memory associated with
** a single shared-btree. The memory is used by client code for its own
** purposes (for example, to store a high-level schema associated with
** the shared-btree). The btree layer manages reference counting issues.
**
** The first time this is called on a shared-btree, nBytes bytes of memory
** are allocated, zeroed, and returned to the caller. For each subsequent
** call the nBytes parameter is ignored and a pointer to the same blob
** of memory returned.
**
** If the nBytes parameter is 0 and the blob of memory has not yet been
** allocated, a null pointer is returned. If the blob has already been
** allocated, it is returned as normal.
**
** Just before the shared-btree is closed, the function passed as the
** xFree argument when the memory allocation was made is invoked on the
** blob of allocated memory. The xFree function should not call sqlite3_free()
** on the memory, the btree layer does that.
*/
void *sqlite3BtreeSchema(Btree *p, int nBytes, void(*xFree)(void *)){
  BtShared *pBt = p->pBt;
  sqlite3BtreeEnter(p);
  if( !pBt->pSchema && nBytes ){
    pBt->pSchema = sqlite3DbMallocZero(0, nBytes);
    pBt->xFreeSchema = xFree;
  }
  sqlite3BtreeLeave(p);
  return pBt->pSchema;
}

/*
** Return SQLITE_LOCKED_SHAREDCACHE if another user of the same shared
** btree as the argument handle holds an exclusive lock on the
** sqlite_schema table. Otherwise SQLITE_OK.
*/
int sqlite3BtreeSchemaLocked(Btree *p){
  int rc;
  assert( sqlite3_mutex_held(p->db->mutex) );
  sqlite3BtreeEnter(p);
  rc = querySharedCacheTableLock(p, SCHEMA_ROOT, READ_LOCK);
  assert( rc==SQLITE_OK || rc==SQLITE_LOCKED_SHAREDCACHE );
  sqlite3BtreeLeave(p);
  return rc;
}


#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Obtain a lock on the table whose root page is iTab.  The
** lock is a write lock if isWritelock is true or a read lock
** if it is false.
*/
int sqlite3BtreeLockTable(Btree *p, int iTab, u8 isWriteLock){
  int rc = SQLITE_OK;
  assert( p->inTrans!=TRANS_NONE );
  if( p->sharable ){
    u8 lockType = READ_LOCK + isWriteLock;
    assert( READ_LOCK+1==WRITE_LOCK );
    assert( isWriteLock==0 || isWriteLock==1 );

    sqlite3BtreeEnter(p);
    rc = querySharedCacheTableLock(p, iTab, lockType);
    if( rc==SQLITE_OK ){
      rc = setSharedCacheTableLock(p, iTab, lockType);
    }
    sqlite3BtreeLeave(p);
  }
  return rc;
}
#endif

#ifndef SQLITE_OMIT_INCRBLOB
/*
** Argument pCsr must be a cursor opened for writing on an
** INTKEY table currently pointing at a valid table entry.
** This function modifies the data stored as part of that entry.
**
** Only the data content may only be modified, it is not possible to
** change the length of the data stored. If this function is called with
** parameters that attempt to write past the end of the existing data,
** no modifications are made and SQLITE_CORRUPT is returned.
*/
int sqlite3BtreePutData(BtCursor *pCsr, u32 offset, u32 amt, void *z){
  int rc;
  assert( cursorOwnsBtShared(pCsr) );
  assert( sqlite3_mutex_held(pCsr->pBtree->db->mutex) );
  assert( pCsr->curFlags & BTCF_Incrblob );

  rc = restoreCursorPosition(pCsr);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  assert( pCsr->eState!=CURSOR_REQUIRESEEK );
  if( pCsr->eState!=CURSOR_VALID ){
    return SQLITE_ABORT;
  }

  /* Save the positions of all other cursors open on this table. This is
  ** required in case any of them are holding references to an xFetch
  ** version of the b-tree page modified by the accessPayload call below.
  **
  ** Note that pCsr must be open on a INTKEY table and saveCursorPosition()
  ** and hence saveAllCursors() cannot fail on a BTREE_INTKEY table, hence
  ** saveAllCursors can only return SQLITE_OK.
  */
  VVA_ONLY(rc =) saveAllCursors(pCsr->pBt, pCsr->pgnoRoot, pCsr);
  assert( rc==SQLITE_OK );

  /* Check some assumptions:
  **   (a) the cursor is open for writing,
  **   (b) there is a read/write transaction open,
  **   (c) the connection holds a write-lock on the table (if required),
  **   (d) there are no conflicting read-locks, and
  **   (e) the cursor points at a valid row of an intKey table.
  */
  if( (pCsr->curFlags & BTCF_WriteFlag)==0 ){
    return SQLITE_READONLY;
  }
  assert( (pCsr->pBt->btsFlags & BTS_READ_ONLY)==0
              && pCsr->pBt->inTransaction==TRANS_WRITE );
  assert( hasSharedCacheTableLock(pCsr->pBtree, pCsr->pgnoRoot, 0, 2) );
  assert( !hasReadConflicts(pCsr->pBtree, pCsr->pgnoRoot) );
  assert( pCsr->pPage->intKey );

  return accessPayload(pCsr, offset, amt, (unsigned char *)z, 1);
}

/*
** Mark this cursor as an incremental blob cursor.
*/
void sqlite3BtreeIncrblobCursor(BtCursor *pCur){
  pCur->curFlags |= BTCF_Incrblob;
  pCur->pBtree->hasIncrblobCur = 1;
}
#endif

/*
** Set both the "read version" (single byte at byte offset 18) and
** "write version" (single byte at byte offset 19) fields in the database
** header to iVersion.
*/
int sqlite3BtreeSetVersion(Btree *pBtree, int iVersion){
  BtShared *pBt = pBtree->pBt;
  int rc;                         /* Return code */

  assert( iVersion==1 || iVersion==2 );

  /* If setting the version fields to 1, do not automatically open the
  ** WAL connection, even if the version fields are currently set to 2.
  */
  pBt->btsFlags &= ~BTS_NO_WAL;
  if( iVersion==1 ) pBt->btsFlags |= BTS_NO_WAL;

  rc = sqlite3BtreeBeginTrans(pBtree, 0, 0);
  if( rc==SQLITE_OK ){
    u8 *aData = pBt->pPage1->aData;
    if( aData[18]!=(u8)iVersion || aData[19]!=(u8)iVersion ){
      rc = sqlite3BtreeBeginTrans(pBtree, 2, 0);
      if( rc==SQLITE_OK ){
        rc = sqlite3PagerWrite(pBt->pPage1->pDbPage);
        if( rc==SQLITE_OK ){
          aData[18] = (u8)iVersion;
          aData[19] = (u8)iVersion;
        }
      }
    }
  }

  pBt->btsFlags &= ~BTS_NO_WAL;
  return rc;
}

/*
** Return true if the cursor has a hint specified.  This routine is
** only used from within assert() statements
*/
int sqlite3BtreeCursorHasHint(BtCursor *pCsr, unsigned int mask){
  return (pCsr->hints & mask)!=0;
}

/*
** Return true if the given Btree is read-only.
*/
int sqlite3BtreeIsReadonly(Btree *p){
  return (p->pBt->btsFlags & BTS_READ_ONLY)!=0;
}

/*
** Return the size of the header added to each page by this module.
*/
int sqlite3HeaderSizeBtree(void){ return ROUND8(sizeof(MemPage)); }

/*
** If no transaction is active and the database is not a temp-db, clear
** the in-memory pager cache.
*/
void sqlite3BtreeClearCache(Btree *p){
  BtShared *pBt = p->pBt;
  if( pBt->inTransaction==TRANS_NONE ){
    sqlite3PagerClearCache(pBt->pPager);
  }
}

#if !defined(SQLITE_OMIT_SHARED_CACHE)
/*
** Return true if the Btree passed as the only argument is sharable.
*/
int sqlite3BtreeSharable(Btree *p){
  return p->sharable;
}

/*
** Return the number of connections to the BtShared object accessed by
** the Btree handle passed as the only argument. For private caches
** this is always 1. For shared caches it may be 1 or greater.
*/
int sqlite3BtreeConnectionCount(Btree *p){
  testcase( p->sharable );
  return p->pBt->nRef;
}
#endif