summaryrefslogtreecommitdiffstats
path: root/sys/contrib/softupdates/ffs_softdep.c
blob: 433cafe5faf50d85ee4dc1ca88d605797078d145 (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

/*
 * Copyright 1998 Marshall Kirk McKusick. All Rights Reserved.
 *
 * The soft updates code is derived from the appendix of a University
 * of Michigan technical report (Gregory R. Ganger and Yale N. Patt,
 * "Soft Updates: A Solution to the Metadata Update Problem in File
 * Systems", CSE-TR-254-95, August 1995).
 *
 * The following are the copyrights and redistribution conditions that
 * apply to this copy of the soft update software. For a license
 * to use, redistribute or sell the soft update software under
 * conditions other than those described here, please contact the
 * author at one of the following addresses:
 *
 *	Marshall Kirk McKusick		mckusick@mckusick.com
 *	1614 Oxford Street		+1-510-843-9542
 *	Berkeley, CA 94709-1608
 *	USA
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. None of the names of McKusick, Ganger, Patt, or the University of
 *    Michigan may be used to endorse or promote products derived from
 *    this software without specific prior written permission.
 * 4. Redistributions in any form must be accompanied by information on
 *    how to obtain complete source code for any accompanying software
 *    that uses this software. This source code must either be included
 *    in the distribution or be available for no more than the cost of
 *    distribution plus a nominal fee, and must be freely redistributable
 *    under reasonable conditions. For an executable file, complete
 *    source code means the source code for all modules it contains.
 *    It does not mean source code for modules or files that typically
 *    accompany the operating system on which the executable file runs,
 *    e.g., standard library modules or system header files.
 *
 * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY
 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * DISCLAIMED.  IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR
 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *	from: @(#)ffs_softdep.c	9.28 (McKusick) 8/8/98
 *	$Id: ffs_softdep.c,v 1.19 1999/01/06 18:18:04 bde Exp $
 */

/*
 * For now we want the safety net that the DIAGNOSTIC and DEBUG flags provide.
 */
#ifndef DIAGNOSTIC
#define DIAGNOSTIC
#endif
#ifndef DEBUG
#define DEBUG
#endif

#include <sys/param.h>
#include <sys/buf.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/proc.h>
#include <sys/syslog.h>
#include <sys/systm.h>
#include <sys/vnode.h>
#include <miscfs/specfs/specdev.h>
#include <ufs/ufs/dir.h>
#include <ufs/ufs/quota.h>
#include <ufs/ufs/inode.h>
#include <ufs/ufs/ufsmount.h>
#include <ufs/ffs/fs.h>
#include <ufs/ffs/softdep.h>
#include <ufs/ffs/ffs_extern.h>
#include <ufs/ufs/ufs_extern.h>

/*
 * These definitions need to be adapted to the system to which
 * this file is being ported.
 */
/*
 * malloc types defined for the softdep system.
 */
MALLOC_DEFINE(M_PAGEDEP, "pagedep","File page dependencies");
MALLOC_DEFINE(M_INODEDEP, "inodedep","Inode dependencies");
MALLOC_DEFINE(M_NEWBLK, "newblk","New block allocation");
MALLOC_DEFINE(M_BMSAFEMAP, "bmsafemap","Block or frag allocated from cyl group map");
MALLOC_DEFINE(M_ALLOCDIRECT, "allocdirect","Block or frag dependency for an inode");
MALLOC_DEFINE(M_INDIRDEP, "indirdep","Indirect block dependencies");
MALLOC_DEFINE(M_ALLOCINDIR, "allocindir","Block dependency for an indirect block");
MALLOC_DEFINE(M_FREEFRAG, "freefrag","Previously used frag for an inode");
MALLOC_DEFINE(M_FREEBLKS, "freeblks","Blocks freed from an inode");
MALLOC_DEFINE(M_FREEFILE, "freefile","Inode deallocated");
MALLOC_DEFINE(M_DIRADD, "diradd","New directory entry");
MALLOC_DEFINE(M_MKDIR, "mkdir","New directory");
MALLOC_DEFINE(M_DIRREM, "dirrem","Directory entry deleted");

#define	D_PAGEDEP	0
#define	D_INODEDEP	1
#define	D_NEWBLK	2
#define	D_BMSAFEMAP	3
#define	D_ALLOCDIRECT	4
#define	D_INDIRDEP	5
#define	D_ALLOCINDIR	6
#define	D_FREEFRAG	7
#define	D_FREEBLKS	8
#define	D_FREEFILE	9
#define	D_DIRADD	10
#define	D_MKDIR		11
#define	D_DIRREM	12
#define D_LAST		D_DIRREM

/* 
 * translate from workitem type to memory type
 * MUST match the defines above, such that memtype[D_XXX] == M_XXX
 */
static struct malloc_type *memtype[] = {
	M_PAGEDEP,
	M_INODEDEP,
	M_NEWBLK,
	M_BMSAFEMAP,
	M_ALLOCDIRECT,
	M_INDIRDEP,
	M_ALLOCINDIR,
	M_FREEFRAG,
	M_FREEBLKS,
	M_FREEFILE,
	M_DIRADD,
	M_MKDIR,
	M_DIRREM
};

#define DtoM(type) (memtype[type])

/*
 * Names of malloc types.
 */
#define TYPENAME(type)  \
	((unsigned)(type) < D_LAST ? memtype[type]->ks_shortdesc : "???")
#define CURPROC curproc
/*
 * End system adaptaion definitions.
 */

/*
 * Internal function prototypes.
 */
static	void softdep_error __P((char *, int));
static	int getdirtybuf __P((struct buf **, int));
static	int flush_pagedep_deps __P((struct vnode *, struct mount *,
	    struct diraddhd *));
static	int flush_inodedep_deps __P((struct fs *, ino_t));
static	int handle_written_filepage __P((struct pagedep *, struct buf *));
static  void diradd_inode_written __P((struct diradd *, struct inodedep *));
static	int handle_written_inodeblock __P((struct inodedep *, struct buf *));
static	void handle_allocdirect_partdone __P((struct allocdirect *));
static	void handle_allocindir_partdone __P((struct allocindir *));
static	void initiate_write_filepage __P((struct pagedep *, struct buf *));
static	void handle_written_mkdir __P((struct mkdir *, int));
static	void initiate_write_inodeblock __P((struct inodedep *, struct buf *));
static	void handle_workitem_freefile __P((struct freefile *));
static	void handle_workitem_remove __P((struct dirrem *));
static	struct dirrem *newdirrem __P((struct buf *, struct inode *,
	    struct inode *, int));
static	void free_diradd __P((struct diradd *));
static	void free_allocindir __P((struct allocindir *, struct inodedep *));
static	int indir_trunc __P((struct inode *, ufs_daddr_t, int, ufs_lbn_t,
	    long *));
static	void deallocate_dependencies __P((struct buf *, struct inodedep *));
static	void free_allocdirect __P((struct allocdirectlst *,
	    struct allocdirect *, int));
static	int free_inodedep __P((struct inodedep *));
static	void handle_workitem_freeblocks __P((struct freeblks *));
static	void merge_inode_lists __P((struct inodedep *));
static	void setup_allocindir_phase2 __P((struct buf *, struct inode *,
	    struct allocindir *));
static	struct allocindir *newallocindir __P((struct inode *, int, ufs_daddr_t,
	    ufs_daddr_t));
static	void handle_workitem_freefrag __P((struct freefrag *));
static	struct freefrag *newfreefrag __P((struct inode *, ufs_daddr_t, long));
static	void allocdirect_merge __P((struct allocdirectlst *,
	    struct allocdirect *, struct allocdirect *));
static	struct bmsafemap *bmsafemap_lookup __P((struct buf *));
static	int newblk_lookup __P((struct fs *, ufs_daddr_t, int,
	    struct newblk **));
static	int inodedep_lookup __P((struct fs *, ino_t, int, struct inodedep **));
static	int pagedep_lookup __P((struct inode *, ufs_lbn_t, int,
	    struct pagedep **));
static	void pause_timer __P((void *));
static	int checklimit __P((long *, int));
static	void add_to_worklist __P((struct worklist *));

/*
 * Exported softdep operations.
 */
struct bio_ops bioops = {
	softdep_disk_io_initiation,		/* io_start */
	softdep_disk_write_complete,		/* io_complete */
	softdep_deallocate_dependencies,	/* io_deallocate */
	softdep_fsync,				/* io_fsync */
	softdep_process_worklist,		/* io_sync */
};

/*
 * Locking primitives.
 *
 * For a uniprocessor, all we need to do is protect against disk
 * interrupts. For a multiprocessor, this lock would have to be
 * a mutex. A single mutex is used throughout this file, though
 * finer grain locking could be used if contention warranted it.
 *
 * For a multiprocessor, the sleep call would accept a lock and
 * release it after the sleep processing was complete. In a uniprocessor
 * implementation there is no such interlock, so we simple mark
 * the places where it needs to be done with the `interlocked' form
 * of the lock calls. Since the uniprocessor sleep already interlocks
 * the spl, there is nothing that really needs to be done.
 */
#ifndef /* NOT */ DEBUG
static struct lockit {
	int	lkt_spl;
} lk = { 0 };
#define ACQUIRE_LOCK(lk)		(lk)->lkt_spl = splbio()
#define FREE_LOCK(lk)			splx((lk)->lkt_spl)
#define ACQUIRE_LOCK_INTERLOCKED(lk)
#define FREE_LOCK_INTERLOCKED(lk)

#else /* DEBUG */
static struct lockit {
	int	lkt_spl;
	pid_t	lkt_held;
} lk = { 0, -1 };
static int lockcnt;

static	void acquire_lock __P((struct lockit *));
static	void free_lock __P((struct lockit *));
static	void acquire_lock_interlocked __P((struct lockit *));
static	void free_lock_interlocked __P((struct lockit *));

#define ACQUIRE_LOCK(lk)		acquire_lock(lk)
#define FREE_LOCK(lk)			free_lock(lk)
#define ACQUIRE_LOCK_INTERLOCKED(lk)	acquire_lock_interlocked(lk)
#define FREE_LOCK_INTERLOCKED(lk)	free_lock_interlocked(lk)

static void
acquire_lock(lk)
	struct lockit *lk;
{

	if (lk->lkt_held != -1)
		if (lk->lkt_held == CURPROC->p_pid)
			panic("softdep_lock: locking against myself");
		else
			panic("softdep_lock: lock held by %d", lk->lkt_held);
	lk->lkt_spl = splbio();
	lk->lkt_held = CURPROC->p_pid;
	lockcnt++;
}

static void
free_lock(lk)
	struct lockit *lk;
{

	if (lk->lkt_held == -1)
		panic("softdep_unlock: lock not held");
	lk->lkt_held = -1;
	splx(lk->lkt_spl);
}

static void
acquire_lock_interlocked(lk)
	struct lockit *lk;
{

	if (lk->lkt_held != -1)
		if (lk->lkt_held == CURPROC->p_pid)
			panic("softdep_lock_interlocked: locking against self");
		else
			panic("softdep_lock_interlocked: lock held by %d",
			    lk->lkt_held);
	lk->lkt_held = CURPROC->p_pid;
	lockcnt++;
}

static void
free_lock_interlocked(lk)
	struct lockit *lk;
{

	if (lk->lkt_held == -1)
		panic("softdep_unlock_interlocked: lock not held");
	lk->lkt_held = -1;
}
#endif /* DEBUG */

/*
 * Place holder for real semaphores.
 */
struct sema {
	int	value;
	pid_t	holder;
	char	*name;
	int	prio;
	int	timo;
};
static	void sema_init __P((struct sema *, char *, int, int));
static	int sema_get __P((struct sema *, struct lockit *));
static	void sema_release __P((struct sema *));

static void
sema_init(semap, name, prio, timo)
	struct sema *semap;
	char *name;
	int prio, timo;
{

	semap->holder = -1;
	semap->value = 0;
	semap->name = name;
	semap->prio = prio;
	semap->timo = timo;
}

static int
sema_get(semap, interlock)
	struct sema *semap;
	struct lockit *interlock;
{

	if (semap->value++ > 0) {
		if (interlock != NULL)
			FREE_LOCK_INTERLOCKED(interlock);
		tsleep((caddr_t)semap, semap->prio, semap->name, semap->timo);
		if (interlock != NULL) {
			ACQUIRE_LOCK_INTERLOCKED(interlock);
			FREE_LOCK(interlock);
		}
		return (0);
	}
	semap->holder = CURPROC->p_pid;
	if (interlock != NULL)
		FREE_LOCK(interlock);
	return (1);
}

static void
sema_release(semap)
	struct sema *semap;
{

	if (semap->value <= 0 || semap->holder != CURPROC->p_pid)
		panic("sema_release: not held");
	if (--semap->value > 0) {
		semap->value = 0;
		wakeup(semap);
	}
	semap->holder = -1;
}

/*
 * Worklist queue management.
 * These routines require that the lock be held.
 */
#ifndef /* NOT */ DEBUG
#define WORKLIST_INSERT(head, item) do {	\
	(item)->wk_state |= ONWORKLIST;		\
	LIST_INSERT_HEAD(head, item, wk_list);	\
} while (0)
#define WORKLIST_REMOVE(item) do {		\
	(item)->wk_state &= ~ONWORKLIST;	\
	LIST_REMOVE(item, wk_list);		\
} while (0)
#define WORKITEM_FREE(item, type) FREE(item, DtoM(type))

#else /* DEBUG */
static	void worklist_insert __P((struct workhead *, struct worklist *));
static	void worklist_remove __P((struct worklist *));
static	void workitem_free __P((struct worklist *, int));

#define WORKLIST_INSERT(head, item) worklist_insert(head, item)
#define WORKLIST_REMOVE(item) worklist_remove(item)
#define WORKITEM_FREE(item, type) workitem_free((struct worklist *)item, type)

static void
worklist_insert(head, item)
	struct workhead *head;
	struct worklist *item;
{

	if (lk.lkt_held == -1)
		panic("worklist_insert: lock not held");
	if (item->wk_state & ONWORKLIST)
		panic("worklist_insert: already on list");
	item->wk_state |= ONWORKLIST;
	LIST_INSERT_HEAD(head, item, wk_list);
}

static void
worklist_remove(item)
	struct worklist *item;
{

	if (lk.lkt_held == -1)
		panic("worklist_remove: lock not held");
	if ((item->wk_state & ONWORKLIST) == 0)
		panic("worklist_remove: not on list");
	item->wk_state &= ~ONWORKLIST;
	LIST_REMOVE(item, wk_list);
}

static void
workitem_free(item, type)
	struct worklist *item;
	int type;
{

	if (item->wk_state & ONWORKLIST)
		panic("workitem_free: still on list");
	if (item->wk_type != type)
		panic("workitem_free: type mismatch");
	FREE(item, DtoM(type));
}
#endif /* DEBUG */

/*
 * Workitem queue management
 */
static struct workhead softdep_workitem_pending;
static int softdep_worklist_busy;
static int max_softdeps;	/* maximum number of structs before slowdown */
static int tickdelay = 2;	/* number of ticks to pause during slowdown */
static int max_limit_hit;	/* number of times slowdown imposed */
static int rush_requests;	/* number of times I/O speeded up */
static int proc_waiting;	/* tracks whether we have a timeout posted */
static pid_t filesys_syncer_pid;/* records pid of filesystem syncer process */
#ifdef DEBUG
#include <vm/vm.h>
#include <sys/sysctl.h>
#if defined(__FreeBSD__)
SYSCTL_INT(_debug, OID_AUTO, max_softdeps, CTLFLAG_RW, &max_softdeps, 0, "");
SYSCTL_INT(_debug, OID_AUTO, tickdelay, CTLFLAG_RW, &tickdelay, 0, "");
SYSCTL_INT(_debug, OID_AUTO, max_limit_hit, CTLFLAG_RW, &max_limit_hit, 0, "");
SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &rush_requests, 0, "");
#else /* !__FreeBSD__ */
struct ctldebug debug8 = { "max_softdeps", &max_softdeps };
struct ctldebug debug9 = { "tickdelay", &tickdelay };
struct ctldebug debug10 = { "max_limit_hit", &max_limit_hit };
struct ctldebug debug11 = { "rush_requests", &rush_requests };
#endif	/* !__FreeBSD__ */

#endif /* DEBUG */

/*
 * Add an item to the end of the work queue.
 * This routine requires that the lock be held.
 * This is the only routine that adds items to the list.
 * The following routine is the only one that removes items
 * and does so in order from first to last.
 */
static void
add_to_worklist(wk)
	struct worklist *wk;
{
	static struct worklist *worklist_tail;

	if (wk->wk_state & ONWORKLIST)
		panic("add_to_worklist: already on list");
	wk->wk_state |= ONWORKLIST;
	if (LIST_FIRST(&softdep_workitem_pending) == NULL) {
		LIST_INSERT_HEAD(&softdep_workitem_pending, wk, wk_list);
	} else {
		LIST_INSERT_AFTER(worklist_tail, wk, wk_list);
	}
	worklist_tail = wk;
}

/*
 * Process that runs once per second to handle items in the background queue.
 *
 * Note that we ensure that everything is done in the order in which they
 * appear in the queue. The code below depends on this property to ensure
 * that blocks of a file are freed before the inode itself is freed. This
 * ordering ensures that no new <vfsid, inum, lbn> triples will be generated
 * until all the old ones have been purged from the dependency lists.
 */
int 
softdep_process_worklist(matchmnt)
	struct mount *matchmnt;
{
	struct proc *p = CURPROC;
	struct worklist *wk;
	struct fs *matchfs;
	int matchcnt;

	/*
	 * Record the process identifier of our caller so that we can
	 * give this process preferential treatment in checklimit below.
	 */
	filesys_syncer_pid = p->p_pid;
	matchcnt = 0;
	matchfs = NULL;
	if (matchmnt != NULL)
		matchfs = VFSTOUFS(matchmnt)->um_fs;
	/*
	 * There is no danger of having multiple processes run this
	 * code. It is single threaded solely so that softdep_flushfiles
	 * (below) can get an accurate count of the number of items
	 * related to its mount point that are in the list.
	 */
	if (softdep_worklist_busy && matchmnt == NULL)
		return (-1);
	ACQUIRE_LOCK(&lk);
	while ((wk = LIST_FIRST(&softdep_workitem_pending)) != 0) {
		WORKLIST_REMOVE(wk);
		FREE_LOCK(&lk);
		switch (wk->wk_type) {

		case D_DIRREM:
			/* removal of a directory entry */
			if (WK_DIRREM(wk)->dm_mnt == matchmnt)
				matchcnt += 1;
			handle_workitem_remove(WK_DIRREM(wk));
			break;

		case D_FREEBLKS:
			/* releasing blocks and/or fragments from a file */
			if (WK_FREEBLKS(wk)->fb_fs == matchfs)
				matchcnt += 1;
			handle_workitem_freeblocks(WK_FREEBLKS(wk));
			break;

		case D_FREEFRAG:
			/* releasing a fragment when replaced as a file grows */
			if (WK_FREEFRAG(wk)->ff_fs == matchfs)
				matchcnt += 1;
			handle_workitem_freefrag(WK_FREEFRAG(wk));
			break;

		case D_FREEFILE:
			/* releasing an inode when its link count drops to 0 */
			if (WK_FREEFILE(wk)->fx_fs == matchfs)
				matchcnt += 1;
			handle_workitem_freefile(WK_FREEFILE(wk));
			break;

		default:
			panic("%s_process_worklist: Unknown type %s",
			    "softdep", TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
		if (softdep_worklist_busy && matchmnt == NULL)
			return (-1);
		ACQUIRE_LOCK(&lk);
	}
	FREE_LOCK(&lk);
	return (matchcnt);
}

/*
 * Purge the work list of all items associated with a particular mount point.
 */
int
softdep_flushfiles(oldmnt, flags, p)
	struct mount *oldmnt;
	int flags;
	struct proc *p;
{
	struct vnode *devvp;
	int error, loopcnt;

	/*
	 * Await our turn to clear out the queue.
	 */
	while (softdep_worklist_busy)
		tsleep(&lbolt, PRIBIO, "softflush", 0);
	softdep_worklist_busy = 1;
	if ((error = ffs_flushfiles(oldmnt, flags, p)) != 0) {
		softdep_worklist_busy = 0;
		return (error);
	}
	/*
	 * Alternately flush the block device associated with the mount
	 * point and process any dependencies that the flushing
	 * creates. In theory, this loop can happen at most twice,
	 * but we give it a few extra just to be sure.
	 */
	devvp = VFSTOUFS(oldmnt)->um_devvp;
	for (loopcnt = 10; loopcnt > 0; loopcnt--) {
		if (softdep_process_worklist(oldmnt) == 0) {
			/*
			 * Do another flush in case any vnodes were brought in
			 * as part of the cleanup operations.
			 */
			if ((error = ffs_flushfiles(oldmnt, flags, p)) != 0)
				break;
			/*
			 * If we still found nothing to do, we are really done.
			 */
			if (softdep_process_worklist(oldmnt) == 0)
				break;
		}
		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY, p);
		error = VOP_FSYNC(devvp, p->p_ucred, MNT_WAIT, p);
		VOP_UNLOCK(devvp, 0, p);
		if (error)
			break;
	}
	softdep_worklist_busy = 0;
	/*
	 * If we are unmounting then it is an error to fail. If we
	 * are simply trying to downgrade to read-only, then filesystem
	 * activity can keep us busy forever, so we just fail with EBUSY.
	 */
	if (loopcnt == 0) {
		if (oldmnt->mnt_kern_flag & MNTK_UNMOUNT)
			panic("softdep_flushfiles: looping");
		error = EBUSY;
	}
	return (error);
}

/*
 * A large burst of file addition or deletion activity can drive the
 * memory load excessively high. Therefore we deliberately slow things
 * down and speed up the I/O processing if we find ourselves with too
 * many dependencies in progress.
 */
static int
checklimit(resource, islocked)
	long *resource;
	int islocked;
{
	struct proc *p = CURPROC;

	/*
	 * If we are under our limit, just proceed.
	 */
	if (*resource < max_softdeps)
		return (0);
	/*
	 * We never hold up the filesystem syncer process.
	 */
	if (p->p_pid == filesys_syncer_pid)
		return (0);
	/*
	 * Our first approach is to speed up the syncer process.
	 * We never push it to speed up more than half of its
	 * normal turn time, otherwise it could take over the cpu.
	 */
	if (rushjob < syncdelay / 2) {
		rushjob += 1;
		rush_requests += 1;
		return (0);
	}
	/*
	 * Every trick has failed, so we pause momentarily to let
	 * the filesystem syncer process catch up.
	 */
	if (islocked == 0)
		ACQUIRE_LOCK(&lk);
	if (proc_waiting == 0) {
		proc_waiting = 1;
		timeout(pause_timer, NULL, tickdelay > 2 ? tickdelay : 2);
	}
	FREE_LOCK_INTERLOCKED(&lk);
	(void) tsleep((caddr_t)&proc_waiting, PPAUSE | PCATCH, "softupdate", 0);
	ACQUIRE_LOCK_INTERLOCKED(&lk);
	if (islocked == 0)
		FREE_LOCK(&lk);
	max_limit_hit += 1;
	return (1);
}

/*
 * Awaken processes pausing in checklimit and clear proc_waiting
 * to indicate that there is no longer a timer running.
 */
void
pause_timer(arg)
	void *arg;
{

	proc_waiting = 0;
	wakeup(&proc_waiting);
}

/*
 * Structure hashing.
 * 
 * There are three types of structures that can be looked up:
 *	1) pagedep structures identified by mount point, inode number,
 *	   and logical block.
 *	2) inodedep structures identified by mount point and inode number.
 *	3) newblk structures identified by mount point and
 *	   physical block number.
 *
 * The "pagedep" and "inodedep" dependency structures are hashed
 * separately from the file blocks and inodes to which they correspond.
 * This separation helps when the in-memory copy of an inode or
 * file block must be replaced. It also obviates the need to access
 * an inode or file page when simply updating (or de-allocating)
 * dependency structures. Lookup of newblk structures is needed to
 * find newly allocated blocks when trying to associate them with
 * their allocdirect or allocindir structure.
 *
 * The lookup routines optionally create and hash a new instance when
 * an existing entry is not found.
 */
#define DEPALLOC	0x0001	/* allocate structure if lookup fails */

/*
 * Structures and routines associated with pagedep caching.
 */
LIST_HEAD(pagedep_hashhead, pagedep) *pagedep_hashtbl;
u_long	pagedep_hash;		/* size of hash table - 1 */
#define	PAGEDEP_HASH(mp, inum, lbn) \
	(&pagedep_hashtbl[((((register_t)(mp)) >> 13) + (inum) + (lbn)) & \
	    pagedep_hash])
static struct sema pagedep_in_progress;

/*
 * Look up a pagedep. Return 1 if found, 0 if not found.
 * If not found, allocate if DEPALLOC flag is passed.
 * Found or allocated entry is returned in pagedeppp.
 * This routine must be called with splbio interrupts blocked.
 */
static int
pagedep_lookup(ip, lbn, flags, pagedeppp)
	struct inode *ip;
	ufs_lbn_t lbn;
	int flags;
	struct pagedep **pagedeppp;
{
	struct pagedep *pagedep;
	struct pagedep_hashhead *pagedephd;
	struct mount *mp;
	int i;

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("pagedep_lookup: lock not held");
#endif
	mp = ITOV(ip)->v_mount;
	pagedephd = PAGEDEP_HASH(mp, ip->i_number, lbn);
top:
	for (pagedep = LIST_FIRST(pagedephd); pagedep;
	     pagedep = LIST_NEXT(pagedep, pd_hash))
		if (ip->i_number == pagedep->pd_ino &&
		    lbn == pagedep->pd_lbn &&
		    mp == pagedep->pd_mnt)
			break;
	if (pagedep) {
		*pagedeppp = pagedep;
		return (1);
	}
	if ((flags & DEPALLOC) == 0) {
		*pagedeppp = NULL;
		return (0);
	}
	if (sema_get(&pagedep_in_progress, &lk) == 0) {
		ACQUIRE_LOCK(&lk);
		goto top;
	}
	MALLOC(pagedep, struct pagedep *, sizeof(struct pagedep), M_PAGEDEP,
		M_WAITOK);
	bzero(pagedep, sizeof(struct pagedep));
	pagedep->pd_list.wk_type = D_PAGEDEP;
	pagedep->pd_mnt = mp;
	pagedep->pd_ino = ip->i_number;
	pagedep->pd_lbn = lbn;
	LIST_INIT(&pagedep->pd_dirremhd);
	LIST_INIT(&pagedep->pd_pendinghd);
	for (i = 0; i < DAHASHSZ; i++)
		LIST_INIT(&pagedep->pd_diraddhd[i]);
	ACQUIRE_LOCK(&lk);
	LIST_INSERT_HEAD(pagedephd, pagedep, pd_hash);
	sema_release(&pagedep_in_progress);
	*pagedeppp = pagedep;
	return (0);
}

/*
 * Structures and routines associated with inodedep caching.
 */
LIST_HEAD(inodedep_hashhead, inodedep) *inodedep_hashtbl;
static u_long	inodedep_hash;	/* size of hash table - 1 */
static long	num_inodedep;	/* number of inodedep allocated */
#define	INODEDEP_HASH(fs, inum) \
      (&inodedep_hashtbl[((((register_t)(fs)) >> 13) + (inum)) & inodedep_hash])
static struct sema inodedep_in_progress;

/*
 * Look up a inodedep. Return 1 if found, 0 if not found.
 * If not found, allocate if DEPALLOC flag is passed.
 * Found or allocated entry is returned in inodedeppp.
 * This routine must be called with splbio interrupts blocked.
 */
static int
inodedep_lookup(fs, inum, flags, inodedeppp)
	struct fs *fs;
	ino_t inum;
	int flags;
	struct inodedep **inodedeppp;
{
	struct inodedep *inodedep;
	struct inodedep_hashhead *inodedephd;
	int firsttry;

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("inodedep_lookup: lock not held");
#endif
	firsttry = 1;
	inodedephd = INODEDEP_HASH(fs, inum);
top:
	for (inodedep = LIST_FIRST(inodedephd); inodedep;
	     inodedep = LIST_NEXT(inodedep, id_hash))
		if (inum == inodedep->id_ino && fs == inodedep->id_fs)
			break;
	if (inodedep) {
		*inodedeppp = inodedep;
		return (1);
	}
	if ((flags & DEPALLOC) == 0) {
		*inodedeppp = NULL;
		return (0);
	}
	if (firsttry && checklimit(&num_inodedep, 1) == 1) {
		firsttry = 0;
		goto top;
	}
	if (sema_get(&inodedep_in_progress, &lk) == 0) {
		ACQUIRE_LOCK(&lk);
		goto top;
	}
	num_inodedep += 1;
	MALLOC(inodedep, struct inodedep *, sizeof(struct inodedep),
		M_INODEDEP, M_WAITOK);
	inodedep->id_list.wk_type = D_INODEDEP;
	inodedep->id_fs = fs;
	inodedep->id_ino = inum;
	inodedep->id_state = ALLCOMPLETE;
	inodedep->id_nlinkdelta = 0;
	inodedep->id_savedino = NULL;
	inodedep->id_savedsize = -1;
	inodedep->id_buf = NULL;
	LIST_INIT(&inodedep->id_pendinghd);
	LIST_INIT(&inodedep->id_inowait);
	LIST_INIT(&inodedep->id_bufwait);
	TAILQ_INIT(&inodedep->id_inoupdt);
	TAILQ_INIT(&inodedep->id_newinoupdt);
	ACQUIRE_LOCK(&lk);
	LIST_INSERT_HEAD(inodedephd, inodedep, id_hash);
	sema_release(&inodedep_in_progress);
	*inodedeppp = inodedep;
	return (0);
}

/*
 * Structures and routines associated with newblk caching.
 */
LIST_HEAD(newblk_hashhead, newblk) *newblk_hashtbl;
u_long	newblk_hash;		/* size of hash table - 1 */
#define	NEWBLK_HASH(fs, inum) \
	(&newblk_hashtbl[((((register_t)(fs)) >> 13) + (inum)) & newblk_hash])
static struct sema newblk_in_progress;

/*
 * Look up a newblk. Return 1 if found, 0 if not found.
 * If not found, allocate if DEPALLOC flag is passed.
 * Found or allocated entry is returned in newblkpp.
 */
static int
newblk_lookup(fs, newblkno, flags, newblkpp)
	struct fs *fs;
	ufs_daddr_t newblkno;
	int flags;
	struct newblk **newblkpp;
{
	struct newblk *newblk;
	struct newblk_hashhead *newblkhd;

	newblkhd = NEWBLK_HASH(fs, newblkno);
top:
	for (newblk = LIST_FIRST(newblkhd); newblk;
	     newblk = LIST_NEXT(newblk, nb_hash))
		if (newblkno == newblk->nb_newblkno && fs == newblk->nb_fs)
			break;
	if (newblk) {
		*newblkpp = newblk;
		return (1);
	}
	if ((flags & DEPALLOC) == 0) {
		*newblkpp = NULL;
		return (0);
	}
	if (sema_get(&newblk_in_progress, 0) == 0)
		goto top;
	MALLOC(newblk, struct newblk *, sizeof(struct newblk),
		M_NEWBLK, M_WAITOK);
	newblk->nb_state = 0;
	newblk->nb_fs = fs;
	newblk->nb_newblkno = newblkno;
	LIST_INSERT_HEAD(newblkhd, newblk, nb_hash);
	sema_release(&newblk_in_progress);
	*newblkpp = newblk;
	return (0);
}

/*
 * Executed during filesystem system initialization before
 * mounting any file systems.
 */
void 
softdep_initialize()
{

	LIST_INIT(&mkdirlisthd);
	LIST_INIT(&softdep_workitem_pending);
	max_softdeps = desiredvnodes * 8;
	pagedep_hashtbl = hashinit(desiredvnodes / 5, M_PAGEDEP,
	    &pagedep_hash);
	sema_init(&pagedep_in_progress, "pagedep", PRIBIO, 0);
	inodedep_hashtbl = hashinit(desiredvnodes, M_INODEDEP, &inodedep_hash);
	sema_init(&inodedep_in_progress, "inodedep", PRIBIO, 0);
	newblk_hashtbl = hashinit(64, M_NEWBLK, &newblk_hash);
	sema_init(&newblk_in_progress, "newblk", PRIBIO, 0);
}

/*
 * Called at mount time to notify the dependency code that a
 * filesystem wishes to use it.
 */
int
softdep_mount(devvp, mp, fs, cred)
	struct vnode *devvp;
	struct mount *mp;
	struct fs *fs;
	struct ucred *cred;
{
	struct csum cstotal;
	struct cg *cgp;
	struct buf *bp;
	int error, cyl;

	mp->mnt_flag &= ~MNT_ASYNC;
	mp->mnt_flag |= MNT_SOFTDEP;
	/*
	 * When doing soft updates, the counters in the
	 * superblock may have gotten out of sync, so we have
	 * to scan the cylinder groups and recalculate them.
	 */
	if (fs->fs_clean != 0)
		return (0);
	bzero(&cstotal, sizeof cstotal);
	for (cyl = 0; cyl < fs->fs_ncg; cyl++) {
		if ((error = bread(devvp, fsbtodb(fs, cgtod(fs, cyl)),
		    fs->fs_cgsize, cred, &bp)) != 0) {
			brelse(bp);
			return (error);
		}
		cgp = (struct cg *)bp->b_data;
		cstotal.cs_nffree += cgp->cg_cs.cs_nffree;
		cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree;
		cstotal.cs_nifree += cgp->cg_cs.cs_nifree;
		cstotal.cs_ndir += cgp->cg_cs.cs_ndir;
		fs->fs_cs(fs, cyl) = cgp->cg_cs;
		brelse(bp);
	}
#ifdef DEBUG
	if (!bcmp(&cstotal, &fs->fs_cstotal, sizeof cstotal))
		printf("ffs_mountfs: superblock updated for soft updates\n");
#endif
	bcopy(&cstotal, &fs->fs_cstotal, sizeof cstotal);
	return (0);
}

/*
 * Protecting the freemaps (or bitmaps).
 * 
 * To eliminate the need to execute fsck before mounting a file system
 * after a power failure, one must (conservatively) guarantee that the
 * on-disk copy of the bitmaps never indicate that a live inode or block is
 * free.  So, when a block or inode is allocated, the bitmap should be
 * updated (on disk) before any new pointers.  When a block or inode is
 * freed, the bitmap should not be updated until all pointers have been
 * reset.  The latter dependency is handled by the delayed de-allocation
 * approach described below for block and inode de-allocation.  The former
 * dependency is handled by calling the following procedure when a block or
 * inode is allocated. When an inode is allocated an "inodedep" is created
 * with its DEPCOMPLETE flag cleared until its bitmap is written to disk.
 * Each "inodedep" is also inserted into the hash indexing structure so
 * that any additional link additions can be made dependent on the inode
 * allocation.
 * 
 * The ufs file system maintains a number of free block counts (e.g., per
 * cylinder group, per cylinder and per <cylinder, rotational position> pair)
 * in addition to the bitmaps.  These counts are used to improve efficiency
 * during allocation and therefore must be consistent with the bitmaps.
 * There is no convenient way to guarantee post-crash consistency of these
 * counts with simple update ordering, for two main reasons: (1) The counts
 * and bitmaps for a single cylinder group block are not in the same disk
 * sector.  If a disk write is interrupted (e.g., by power failure), one may
 * be written and the other not.  (2) Some of the counts are located in the
 * superblock rather than the cylinder group block. So, we focus our soft
 * updates implementation on protecting the bitmaps. When mounting a
 * filesystem, we recompute the auxiliary counts from the bitmaps.
 */

/*
 * Called just after updating the cylinder group block to allocate an inode.
 */
void
softdep_setup_inomapdep(bp, ip, newinum)
	struct buf *bp;		/* buffer for cylgroup block with inode map */
	struct inode *ip;	/* inode related to allocation */
	ino_t newinum;		/* new inode number being allocated */
{
	struct inodedep *inodedep;
	struct bmsafemap *bmsafemap;

	/*
	 * Create a dependency for the newly allocated inode.
	 * Panic if it already exists as something is seriously wrong.
	 * Otherwise add it to the dependency list for the buffer holding
	 * the cylinder group map from which it was allocated.
	 */
	ACQUIRE_LOCK(&lk);
	if (inodedep_lookup(ip->i_fs, newinum, DEPALLOC, &inodedep) != 0)
		panic("softdep_setup_inomapdep: found inode");
	inodedep->id_buf = bp;
	inodedep->id_state &= ~DEPCOMPLETE;
	bmsafemap = bmsafemap_lookup(bp);
	LIST_INSERT_HEAD(&bmsafemap->sm_inodedephd, inodedep, id_deps);
	FREE_LOCK(&lk);
}

/*
 * Called just after updating the cylinder group block to
 * allocate block or fragment.
 */
void
softdep_setup_blkmapdep(bp, fs, newblkno)
	struct buf *bp;		/* buffer for cylgroup block with block map */
	struct fs *fs;		/* filesystem doing allocation */
	ufs_daddr_t newblkno;	/* number of newly allocated block */
{
	struct newblk *newblk;
	struct bmsafemap *bmsafemap;

	/*
	 * Create a dependency for the newly allocated block.
	 * Add it to the dependency list for the buffer holding
	 * the cylinder group map from which it was allocated.
	 */
	if (newblk_lookup(fs, newblkno, DEPALLOC, &newblk) != 0)
		panic("softdep_setup_blkmapdep: found block");
	ACQUIRE_LOCK(&lk);
	newblk->nb_bmsafemap = bmsafemap = bmsafemap_lookup(bp);
	LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps);
	FREE_LOCK(&lk);
}

/*
 * Find the bmsafemap associated with a cylinder group buffer.
 * If none exists, create one. The buffer must be locked when
 * this routine is called and this routine must be called with
 * splbio interrupts blocked.
 */
static struct bmsafemap *
bmsafemap_lookup(bp)
	struct buf *bp;
{
	struct bmsafemap *bmsafemap;
	struct worklist *wk;

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("bmsafemap_lookup: lock not held");
#endif
	for (wk = LIST_FIRST(&bp->b_dep); wk; wk = LIST_NEXT(wk, wk_list))
		if (wk->wk_type == D_BMSAFEMAP)
			return (WK_BMSAFEMAP(wk));
	FREE_LOCK(&lk);
	MALLOC(bmsafemap, struct bmsafemap *, sizeof(struct bmsafemap),
		M_BMSAFEMAP, M_WAITOK);
	bmsafemap->sm_list.wk_type = D_BMSAFEMAP;
	bmsafemap->sm_list.wk_state = 0;
	bmsafemap->sm_buf = bp;
	LIST_INIT(&bmsafemap->sm_allocdirecthd);
	LIST_INIT(&bmsafemap->sm_allocindirhd);
	LIST_INIT(&bmsafemap->sm_inodedephd);
	LIST_INIT(&bmsafemap->sm_newblkhd);
	ACQUIRE_LOCK(&lk);
	WORKLIST_INSERT(&bp->b_dep, &bmsafemap->sm_list);
	return (bmsafemap);
}

/*
 * Direct block allocation dependencies.
 * 
 * When a new block is allocated, the corresponding disk locations must be
 * initialized (with zeros or new data) before the on-disk inode points to
 * them.  Also, the freemap from which the block was allocated must be
 * updated (on disk) before the inode's pointer. These two dependencies are
 * independent of each other and are needed for all file blocks and indirect
 * blocks that are pointed to directly by the inode.  Just before the
 * "in-core" version of the inode is updated with a newly allocated block
 * number, a procedure (below) is called to setup allocation dependency
 * structures.  These structures are removed when the corresponding
 * dependencies are satisfied or when the block allocation becomes obsolete
 * (i.e., the file is deleted, the block is de-allocated, or the block is a
 * fragment that gets upgraded).  All of these cases are handled in
 * procedures described later.
 * 
 * When a file extension causes a fragment to be upgraded, either to a larger
 * fragment or to a full block, the on-disk location may change (if the
 * previous fragment could not simply be extended). In this case, the old
 * fragment must be de-allocated, but not until after the inode's pointer has
 * been updated. In most cases, this is handled by later procedures, which
 * will construct a "freefrag" structure to be added to the workitem queue
 * when the inode update is complete (or obsolete).  The main exception to
 * this is when an allocation occurs while a pending allocation dependency
 * (for the same block pointer) remains.  This case is handled in the main
 * allocation dependency setup procedure by immediately freeing the
 * unreferenced fragments.
 */ 
void 
softdep_setup_allocdirect(ip, lbn, newblkno, oldblkno, newsize, oldsize, bp)
	struct inode *ip;	/* inode to which block is being added */
	ufs_lbn_t lbn;		/* block pointer within inode */
	ufs_daddr_t newblkno;	/* disk block number being added */
	ufs_daddr_t oldblkno;	/* previous block number, 0 unless frag */
	long newsize;		/* size of new block */
	long oldsize;		/* size of new block */
	struct buf *bp;		/* bp for allocated block */
{
	struct allocdirect *adp, *oldadp;
	struct allocdirectlst *adphead;
	struct bmsafemap *bmsafemap;
	struct inodedep *inodedep;
	struct pagedep *pagedep;
	struct newblk *newblk;

	MALLOC(adp, struct allocdirect *, sizeof(struct allocdirect),
		M_ALLOCDIRECT, M_WAITOK);
	bzero(adp, sizeof(struct allocdirect));
	adp->ad_list.wk_type = D_ALLOCDIRECT;
	adp->ad_lbn = lbn;
	adp->ad_newblkno = newblkno;
	adp->ad_oldblkno = oldblkno;
	adp->ad_newsize = newsize;
	adp->ad_oldsize = oldsize;
	adp->ad_state = ATTACHED;
	if (newblkno == oldblkno)
		adp->ad_freefrag = NULL;
	else
		adp->ad_freefrag = newfreefrag(ip, oldblkno, oldsize);

	if (newblk_lookup(ip->i_fs, newblkno, 0, &newblk) == 0)
		panic("softdep_setup_allocdirect: lost block");

	ACQUIRE_LOCK(&lk);
	(void) inodedep_lookup(ip->i_fs, ip->i_number, DEPALLOC, &inodedep);
	adp->ad_inodedep = inodedep;

	if (newblk->nb_state == DEPCOMPLETE) {
		adp->ad_state |= DEPCOMPLETE;
		adp->ad_buf = NULL;
	} else {
		bmsafemap = newblk->nb_bmsafemap;
		adp->ad_buf = bmsafemap->sm_buf;
		LIST_REMOVE(newblk, nb_deps);
		LIST_INSERT_HEAD(&bmsafemap->sm_allocdirecthd, adp, ad_deps);
	}
	LIST_REMOVE(newblk, nb_hash);
	FREE(newblk, M_NEWBLK);

	WORKLIST_INSERT(&bp->b_dep, &adp->ad_list);
	if (lbn >= NDADDR) {
		/* allocating an indirect block */
		if (oldblkno != 0)
			panic("softdep_setup_allocdirect: non-zero indir");
	} else {
		/*
		 * Allocating a direct block.
		 *
		 * If we are allocating a directory block, then we must
		 * allocate an associated pagedep to track additions and
		 * deletions.
		 */
		if ((ip->i_mode & IFMT) == IFDIR &&
		    pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0)
			WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
	}
	/*
	 * The list of allocdirects must be kept in sorted and ascending
	 * order so that the rollback routines can quickly determine the
	 * first uncommitted block (the size of the file stored on disk
	 * ends at the end of the lowest committed fragment, or if there
	 * are no fragments, at the end of the highest committed block).
	 * Since files generally grow, the typical case is that the new
	 * block is to be added at the end of the list. We speed this
	 * special case by checking against the last allocdirect in the
	 * list before laboriously traversing the list looking for the
	 * insertion point.
	 */
	adphead = &inodedep->id_newinoupdt;
	oldadp = TAILQ_LAST(adphead, allocdirectlst);
	if (oldadp == NULL || oldadp->ad_lbn <= lbn) {
		/* insert at end of list */
		TAILQ_INSERT_TAIL(adphead, adp, ad_next);
		if (oldadp != NULL && oldadp->ad_lbn == lbn)
			allocdirect_merge(adphead, adp, oldadp);
		FREE_LOCK(&lk);
		return;
	}
	for (oldadp = TAILQ_FIRST(adphead); oldadp;
	     oldadp = TAILQ_NEXT(oldadp, ad_next)) {
		if (oldadp->ad_lbn >= lbn)
			break;
	}
	if (oldadp == NULL)
		panic("softdep_setup_allocdirect: lost entry");
	/* insert in middle of list */
	TAILQ_INSERT_BEFORE(oldadp, adp, ad_next);
	if (oldadp->ad_lbn == lbn)
		allocdirect_merge(adphead, adp, oldadp);
	FREE_LOCK(&lk);
}

/*
 * Replace an old allocdirect dependency with a newer one.
 * This routine must be called with splbio interrupts blocked.
 */
static void
allocdirect_merge(adphead, newadp, oldadp)
	struct allocdirectlst *adphead;	/* head of list holding allocdirects */
	struct allocdirect *newadp;	/* allocdirect being added */
	struct allocdirect *oldadp;	/* existing allocdirect being checked */
{
	struct freefrag *freefrag;

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("allocdirect_merge: lock not held");
#endif
	if (newadp->ad_oldblkno != oldadp->ad_newblkno ||
	    newadp->ad_oldsize != oldadp->ad_newsize ||
	    newadp->ad_lbn >= NDADDR)
		panic("allocdirect_check: old %d != new %d || lbn %ld >= %d",
		    newadp->ad_oldblkno, oldadp->ad_newblkno, newadp->ad_lbn,
		    NDADDR);
	newadp->ad_oldblkno = oldadp->ad_oldblkno;
	newadp->ad_oldsize = oldadp->ad_oldsize;
	/*
	 * If the old dependency had a fragment to free or had never
	 * previously had a block allocated, then the new dependency
	 * can immediately post its freefrag and adopt the old freefrag.
	 * This action is done by swapping the freefrag dependencies.
	 * The new dependency gains the old one's freefrag, and the
	 * old one gets the new one and then immediately puts it on
	 * the worklist when it is freed by free_allocdirect. It is
	 * not possible to do this swap when the old dependency had a
	 * non-zero size but no previous fragment to free. This condition
	 * arises when the new block is an extension of the old block.
	 * Here, the first part of the fragment allocated to the new
	 * dependency is part of the block currently claimed on disk by
	 * the old dependency, so cannot legitimately be freed until the
	 * conditions for the new dependency are fulfilled.
	 */
	if (oldadp->ad_freefrag != NULL || oldadp->ad_oldblkno == 0) {
		freefrag = newadp->ad_freefrag;
		newadp->ad_freefrag = oldadp->ad_freefrag;
		oldadp->ad_freefrag = freefrag;
	}
	free_allocdirect(adphead, oldadp, 0);
}
		
/*
 * Allocate a new freefrag structure if needed.
 */
static struct freefrag *
newfreefrag(ip, blkno, size)
	struct inode *ip;
	ufs_daddr_t blkno;
	long size;
{
	struct freefrag *freefrag;
	struct fs *fs;

	if (blkno == 0)
		return (NULL);
	fs = ip->i_fs;
	if (fragnum(fs, blkno) + numfrags(fs, size) > fs->fs_frag)
		panic("newfreefrag: frag size");
	MALLOC(freefrag, struct freefrag *, sizeof(struct freefrag),
		M_FREEFRAG, M_WAITOK);
	freefrag->ff_list.wk_type = D_FREEFRAG;
	freefrag->ff_state = ip->i_uid & ~ONWORKLIST;	/* XXX - used below */
	freefrag->ff_inum = ip->i_number;
	freefrag->ff_fs = fs;
	freefrag->ff_devvp = ip->i_devvp;
	freefrag->ff_blkno = blkno;
	freefrag->ff_fragsize = size;
	return (freefrag);
}

/*
 * This workitem de-allocates fragments that were replaced during
 * file block allocation.
 */
static void 
handle_workitem_freefrag(freefrag)
	struct freefrag *freefrag;
{
	struct inode tip;

	tip.i_fs = freefrag->ff_fs;
	tip.i_devvp = freefrag->ff_devvp;
	tip.i_dev = freefrag->ff_devvp->v_rdev;
	tip.i_number = freefrag->ff_inum;
	tip.i_uid = freefrag->ff_state & ~ONWORKLIST;	/* XXX - set above */
	ffs_blkfree(&tip, freefrag->ff_blkno, freefrag->ff_fragsize);
	FREE(freefrag, M_FREEFRAG);
}

/*
 * Indirect block allocation dependencies.
 * 
 * The same dependencies that exist for a direct block also exist when
 * a new block is allocated and pointed to by an entry in a block of
 * indirect pointers. The undo/redo states described above are also
 * used here. Because an indirect block contains many pointers that
 * may have dependencies, a second copy of the entire in-memory indirect
 * block is kept. The buffer cache copy is always completely up-to-date.
 * The second copy, which is used only as a source for disk writes,
 * contains only the safe pointers (i.e., those that have no remaining
 * update dependencies). The second copy is freed when all pointers
 * are safe. The cache is not allowed to replace indirect blocks with
 * pending update dependencies. If a buffer containing an indirect
 * block with dependencies is written, these routines will mark it
 * dirty again. It can only be successfully written once all the
 * dependencies are removed. The ffs_fsync routine in conjunction with
 * softdep_sync_metadata work together to get all the dependencies
 * removed so that a file can be successfully written to disk. Three
 * procedures are used when setting up indirect block pointer
 * dependencies. The division is necessary because of the organization
 * of the "balloc" routine and because of the distinction between file
 * pages and file metadata blocks.
 */

/*
 * Allocate a new allocindir structure.
 */
static struct allocindir *
newallocindir(ip, ptrno, newblkno, oldblkno)
	struct inode *ip;	/* inode for file being extended */
	int ptrno;		/* offset of pointer in indirect block */
	ufs_daddr_t newblkno;	/* disk block number being added */
	ufs_daddr_t oldblkno;	/* previous block number, 0 if none */
{
	struct allocindir *aip;

	MALLOC(aip, struct allocindir *, sizeof(struct allocindir),
		M_ALLOCINDIR, M_WAITOK);
	bzero(aip, sizeof(struct allocindir));
	aip->ai_list.wk_type = D_ALLOCINDIR;
	aip->ai_state = ATTACHED;
	aip->ai_offset = ptrno;
	aip->ai_newblkno = newblkno;
	aip->ai_oldblkno = oldblkno;
	aip->ai_freefrag = newfreefrag(ip, oldblkno, ip->i_fs->fs_bsize);
	return (aip);
}

/*
 * Called just before setting an indirect block pointer
 * to a newly allocated file page.
 */
void
softdep_setup_allocindir_page(ip, lbn, bp, ptrno, newblkno, oldblkno, nbp)
	struct inode *ip;	/* inode for file being extended */
	ufs_lbn_t lbn;		/* allocated block number within file */
	struct buf *bp;		/* buffer with indirect blk referencing page */
	int ptrno;		/* offset of pointer in indirect block */
	ufs_daddr_t newblkno;	/* disk block number being added */
	ufs_daddr_t oldblkno;	/* previous block number, 0 if none */
	struct buf *nbp;	/* buffer holding allocated page */
{
	struct allocindir *aip;
	struct pagedep *pagedep;

	aip = newallocindir(ip, ptrno, newblkno, oldblkno);
	ACQUIRE_LOCK(&lk);
	/*
	 * If we are allocating a directory page, then we must
	 * allocate an associated pagedep to track additions and
	 * deletions.
	 */
	if ((ip->i_mode & IFMT) == IFDIR &&
	    pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0)
		WORKLIST_INSERT(&nbp->b_dep, &pagedep->pd_list);
	WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list);
	FREE_LOCK(&lk);
	setup_allocindir_phase2(bp, ip, aip);
}

/*
 * Called just before setting an indirect block pointer to a
 * newly allocated indirect block.
 */
void
softdep_setup_allocindir_meta(nbp, ip, bp, ptrno, newblkno)
	struct buf *nbp;	/* newly allocated indirect block */
	struct inode *ip;	/* inode for file being extended */
	struct buf *bp;		/* indirect block referencing allocated block */
	int ptrno;		/* offset of pointer in indirect block */
	ufs_daddr_t newblkno;	/* disk block number being added */
{
	struct allocindir *aip;

	aip = newallocindir(ip, ptrno, newblkno, 0);
	ACQUIRE_LOCK(&lk);
	WORKLIST_INSERT(&nbp->b_dep, &aip->ai_list);
	FREE_LOCK(&lk);
	setup_allocindir_phase2(bp, ip, aip);
}

/*
 * Called to finish the allocation of the "aip" allocated
 * by one of the two routines above.
 */
static void 
setup_allocindir_phase2(bp, ip, aip)
	struct buf *bp;		/* in-memory copy of the indirect block */
	struct inode *ip;	/* inode for file being extended */
	struct allocindir *aip;	/* allocindir allocated by the above routines */
{
	struct worklist *wk;
	struct indirdep *indirdep, *newindirdep;
	struct bmsafemap *bmsafemap;
	struct allocindir *oldaip;
	struct freefrag *freefrag;
	struct newblk *newblk;

	if (bp->b_lblkno >= 0)
		panic("setup_allocindir_phase2: not indir blk");
	for (indirdep = NULL, newindirdep = NULL; ; ) {
		ACQUIRE_LOCK(&lk);
		for (wk = LIST_FIRST(&bp->b_dep); wk;
		     wk = LIST_NEXT(wk, wk_list)) {
			if (wk->wk_type != D_INDIRDEP)
				continue;
			indirdep = WK_INDIRDEP(wk);
			break;
		}
		if (indirdep == NULL && newindirdep) {
			indirdep = newindirdep;
			WORKLIST_INSERT(&bp->b_dep, &indirdep->ir_list);
			newindirdep = NULL;
		}
		FREE_LOCK(&lk);
		if (indirdep) {
			if (newblk_lookup(ip->i_fs, aip->ai_newblkno, 0,
			    &newblk) == 0)
				panic("setup_allocindir: lost block");
			ACQUIRE_LOCK(&lk);
			if (newblk->nb_state == DEPCOMPLETE) {
				aip->ai_state |= DEPCOMPLETE;
				aip->ai_buf = NULL;
			} else {
				bmsafemap = newblk->nb_bmsafemap;
				aip->ai_buf = bmsafemap->sm_buf;
				LIST_REMOVE(newblk, nb_deps);
				LIST_INSERT_HEAD(&bmsafemap->sm_allocindirhd,
				    aip, ai_deps);
			}
			LIST_REMOVE(newblk, nb_hash);
			FREE(newblk, M_NEWBLK);
			aip->ai_indirdep = indirdep;
			/*
			 * Check to see if there is an existing dependency
			 * for this block. If there is, merge the old
			 * dependency into the new one.
			 */
			if (aip->ai_oldblkno == 0)
				oldaip = NULL;
			else
				for (oldaip=LIST_FIRST(&indirdep->ir_deplisthd);
				    oldaip; oldaip = LIST_NEXT(oldaip, ai_next))
					if (oldaip->ai_offset == aip->ai_offset)
						break;
			if (oldaip != NULL) {
				if (oldaip->ai_newblkno != aip->ai_oldblkno)
					panic("setup_allocindir_phase2: blkno");
				aip->ai_oldblkno = oldaip->ai_oldblkno;
				freefrag = oldaip->ai_freefrag;
				oldaip->ai_freefrag = aip->ai_freefrag;
				aip->ai_freefrag = freefrag;
				free_allocindir(oldaip, NULL);
			}
			LIST_INSERT_HEAD(&indirdep->ir_deplisthd, aip, ai_next);
			((ufs_daddr_t *)indirdep->ir_savebp->b_data)
			    [aip->ai_offset] = aip->ai_oldblkno;
			FREE_LOCK(&lk);
		}
		if (newindirdep) {
			if (indirdep->ir_savebp != NULL)
				brelse(newindirdep->ir_savebp);
			WORKITEM_FREE((caddr_t)newindirdep, D_INDIRDEP);
		}
		if (indirdep)
			break;
		MALLOC(newindirdep, struct indirdep *, sizeof(struct indirdep),
			M_INDIRDEP, M_WAITOK);
		newindirdep->ir_list.wk_type = D_INDIRDEP;
		newindirdep->ir_state = ATTACHED;
		LIST_INIT(&newindirdep->ir_deplisthd);
		LIST_INIT(&newindirdep->ir_donehd);
#ifdef __FreeBSD__
		if (bp->b_blkno == bp->b_lblkno) {
#if 0 /* we know this happens.. research suggested.. */
			printf("setup_allocindir_phase2: need bmap, blk %d\n",
				bp->b_lblkno);
#endif
			VOP_BMAP(bp->b_vp, bp->b_lblkno, NULL, &bp->b_blkno,
				NULL, NULL);
		}
#endif /* __FreeBSD__ */
		newindirdep->ir_savebp =
		    getblk(ip->i_devvp, bp->b_blkno, bp->b_bcount, 0, 0);
		bp->b_flags |= B_XXX;
		bcopy(bp->b_data, newindirdep->ir_savebp->b_data, bp->b_bcount);
	}
}

/*
 * Block de-allocation dependencies.
 * 
 * When blocks are de-allocated, the on-disk pointers must be nullified before
 * the blocks are made available for use by other files.  (The true
 * requirement is that old pointers must be nullified before new on-disk
 * pointers are set.  We chose this slightly more stringent requirement to
 * reduce complexity.) Our implementation handles this dependency by updating
 * the inode (or indirect block) appropriately but delaying the actual block
 * de-allocation (i.e., freemap and free space count manipulation) until
 * after the updated versions reach stable storage.  After the disk is
 * updated, the blocks can be safely de-allocated whenever it is convenient.
 * This implementation handles only the common case of reducing a file's
 * length to zero. Other cases are handled by the conventional synchronous
 * write approach.
 *
 * The ffs implementation with which we worked double-checks
 * the state of the block pointers and file size as it reduces
 * a file's length.  Some of this code is replicated here in our
 * soft updates implementation.  The freeblks->fb_chkcnt field is
 * used to transfer a part of this information to the procedure
 * that eventually de-allocates the blocks.
 *
 * This routine should be called from the routine that shortens
 * a file's length, before the inode's size or block pointers
 * are modified. It will save the block pointer information for
 * later release and zero the inode so that the calling routine
 * can release it.
 */
static long num_freeblks;	/* number of freeblks allocated */
void
softdep_setup_freeblocks(ip, length)
	struct inode *ip;	/* The inode whose length is to be reduced */
	off_t length;		/* The new length for the file */
{
	struct freeblks *freeblks;
	struct inodedep *inodedep;
	struct allocdirect *adp;
	struct vnode *vp;
	struct buf *bp;
	struct fs *fs;
	int i, error;

	fs = ip->i_fs;
	if (length != 0)
		panic("softde_setup_freeblocks: non-zero length");
	(void) checklimit(&num_freeblks, 0);
	num_freeblks += 1;
	MALLOC(freeblks, struct freeblks *, sizeof(struct freeblks),
		M_FREEBLKS, M_WAITOK);
	bzero(freeblks, sizeof(struct freeblks));
	freeblks->fb_list.wk_type = D_FREEBLKS;
	freeblks->fb_uid = ip->i_uid;
	freeblks->fb_previousinum = ip->i_number;
	freeblks->fb_devvp = ip->i_devvp;
	freeblks->fb_fs = fs;
	freeblks->fb_oldsize = ip->i_size;
	freeblks->fb_newsize = length;
	freeblks->fb_chkcnt = ip->i_blocks;
	for (i = 0; i < NDADDR; i++) {
		freeblks->fb_dblks[i] = ip->i_db[i];
		ip->i_db[i] = 0;
	}
	for (i = 0; i < NIADDR; i++) {
		freeblks->fb_iblks[i] = ip->i_ib[i];
		ip->i_ib[i] = 0;
	}
	ip->i_blocks = 0;
	ip->i_size = 0;
	/*
	 * Push the zero'ed inode to to its disk buffer so that we are free
	 * to delete its dependencies below. Once the dependencies are gone
	 * the buffer can be safely released.
	 */
	if ((error = bread(ip->i_devvp,
	    fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
	    (int)fs->fs_bsize, NOCRED, &bp)) != 0)
		softdep_error("softdep_setup_freeblocks", error);
	*((struct dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) =
	    ip->i_din;
	/*
	 * Find and eliminate any inode dependencies.
	 */
	ACQUIRE_LOCK(&lk);
	(void) inodedep_lookup(fs, ip->i_number, DEPALLOC, &inodedep);
	if ((inodedep->id_state & IOSTARTED) != 0)
		panic("softdep_setup_freeblocks: inode busy");
	/*
	 * Add the freeblks structure to the list of operations that
	 * must await the zero'ed inode being written to disk.
	 */
	WORKLIST_INSERT(&inodedep->id_bufwait, &freeblks->fb_list);
	/*
	 * Because the file length has been truncated to zero, any
	 * pending block allocation dependency structures associated
	 * with this inode are obsolete and can simply be de-allocated.
	 * We must first merge the two dependency lists to get rid of
	 * any duplicate freefrag structures, then purge the merged list.
	 */
	merge_inode_lists(inodedep);
	while ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != 0)
		free_allocdirect(&inodedep->id_inoupdt, adp, 1);
	bdwrite(bp);
	/*
	 * We must wait for any I/O in progress to finish so that
	 * all potential buffers on the dirty list will be visible.
	 * Once they are all there, walk the list and get rid of
	 * any dependencies.
	 */
	vp = ITOV(ip);
	while (vp->v_numoutput) {
		vp->v_flag |= VBWAIT;
		FREE_LOCK_INTERLOCKED(&lk);
		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "sdsetf", 0);
		ACQUIRE_LOCK_INTERLOCKED(&lk);
	}
	while (getdirtybuf(&TAILQ_FIRST(&vp->v_dirtyblkhd), MNT_WAIT)) {
		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
		(void) inodedep_lookup(fs, ip->i_number, 0, &inodedep);
		deallocate_dependencies(bp, inodedep);
		bp->b_flags |= B_INVAL | B_NOCACHE;
		brelse(bp);
	}
	/*
	 * Try freeing the inodedep in case that was the last dependency.
	 */
	if ((inodedep_lookup(fs, ip->i_number, 0, &inodedep)) != 0)
		(void) free_inodedep(inodedep);
	FREE_LOCK(&lk);
}

/*
 * Reclaim any dependency structures from a buffer that is about to
 * be reallocated to a new vnode. The buffer must be locked, thus,
 * no I/O completion operations can occur while we are manipulating
 * its associated dependencies. The mutex is held so that other I/O's
 * associated with related dependencies do not occur.
 */
static void
deallocate_dependencies(bp, inodedep)
	struct buf *bp;
	struct inodedep *inodedep;
{
	struct worklist *wk;
	struct indirdep *indirdep;
	struct allocindir *aip;
	struct pagedep *pagedep;
	struct dirrem *dirrem;
	struct diradd *dap;
	int i;

	while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) {
		switch (wk->wk_type) {

		case D_INDIRDEP:
			indirdep = WK_INDIRDEP(wk);
			/*
			 * None of the indirect pointers will ever be visible,
			 * so they can simply be tossed. GOINGAWAY ensures
			 * that allocated pointers will be saved in the buffer
			 * cache until they are freed. Note that they will
			 * only be able to be found by their physical address
			 * since the inode mapping the logical address will
			 * be gone. The save buffer used for the safe copy
			 * was allocated in setup_allocindir_phase2 using
			 * the physical address so it could be used for this
			 * purpose. Hence we swap the safe copy with the real
			 * copy, allowing the safe copy to be freed and holding
			 * on to the real copy for later use in indir_trunc.
			 */
			if (indirdep->ir_state & GOINGAWAY)
				panic("deallocate_dependencies: already gone");
			indirdep->ir_state |= GOINGAWAY;
			while ((aip = LIST_FIRST(&indirdep->ir_deplisthd)) != 0)
				free_allocindir(aip, inodedep);
			if (bp->b_lblkno >= 0 ||
			    bp->b_blkno != indirdep->ir_savebp->b_lblkno)
				panic("deallocate_dependencies: not indir");
			bcopy(bp->b_data, indirdep->ir_savebp->b_data,
			    bp->b_bcount);
			WORKLIST_REMOVE(wk);
			WORKLIST_INSERT(&indirdep->ir_savebp->b_dep, wk);
			continue;

		case D_PAGEDEP:
			pagedep = WK_PAGEDEP(wk);
			/*
			 * None of the directory additions will ever be
			 * visible, so they can simply be tossed.
			 */
			for (i = 0; i < DAHASHSZ; i++)
				while (dap=LIST_FIRST(&pagedep->pd_diraddhd[i]))
					free_diradd(dap);
			while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != 0)
				free_diradd(dap);
			/*
			 * Copy any directory remove dependencies to the list
			 * to be processed after the zero'ed inode is written.
			 * If the inode has already been written, then they 
			 * can be dumped directly onto the work list.
			 */
			for (dirrem = LIST_FIRST(&pagedep->pd_dirremhd); dirrem;
			     dirrem = LIST_NEXT(dirrem, dm_next)) {
				LIST_REMOVE(dirrem, dm_next);
				dirrem->dm_dirinum = pagedep->pd_ino;
				if (inodedep == NULL)
					add_to_worklist(&dirrem->dm_list);
				else
					WORKLIST_INSERT(&inodedep->id_bufwait,
					    &dirrem->dm_list);
			}
			WORKLIST_REMOVE(&pagedep->pd_list);
			LIST_REMOVE(pagedep, pd_hash);
			WORKITEM_FREE(pagedep, D_PAGEDEP);
			continue;

		case D_ALLOCINDIR:
			free_allocindir(WK_ALLOCINDIR(wk), inodedep);
			continue;

		case D_ALLOCDIRECT:
		case D_INODEDEP:
			panic("deallocate_dependencies: Unexpected type %s",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */

		default:
			panic("deallocate_dependencies: Unknown type %s",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
	}
}

/*
 * Free an allocdirect. Generate a new freefrag work request if appropriate.
 * This routine must be called with splbio interrupts blocked.
 */
static void
free_allocdirect(adphead, adp, delay)
	struct allocdirectlst *adphead;
	struct allocdirect *adp;
	int delay;
{

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("free_allocdirect: lock not held");
#endif
	if ((adp->ad_state & DEPCOMPLETE) == 0)
		LIST_REMOVE(adp, ad_deps);
	TAILQ_REMOVE(adphead, adp, ad_next);
	if ((adp->ad_state & COMPLETE) == 0)
		WORKLIST_REMOVE(&adp->ad_list);
	if (adp->ad_freefrag != NULL) {
		if (delay)
			WORKLIST_INSERT(&adp->ad_inodedep->id_bufwait,
			    &adp->ad_freefrag->ff_list);
		else
			add_to_worklist(&adp->ad_freefrag->ff_list);
	}
	WORKITEM_FREE(adp, D_ALLOCDIRECT);
}

/*
 * Prepare an inode to be freed. The actual free operation is not
 * done until the zero'ed inode has been written to disk.
 */
static long num_freefile;	/* number of freefile allocated */
void
softdep_freefile(pvp, ino, mode)
		struct vnode *pvp;
		ino_t ino;
		int mode;
{
	struct inode *ip = VTOI(pvp);
	struct inodedep *inodedep;
	struct freefile *freefile;

	/*
	 * This sets up the inode de-allocation dependency.
	 */
	(void) checklimit(&num_freefile, 0);
	num_freefile += 1;
	MALLOC(freefile, struct freefile *, sizeof(struct freefile),
		M_FREEFILE, M_WAITOK);
	freefile->fx_list.wk_type = D_FREEFILE;
	freefile->fx_list.wk_state = 0;
	freefile->fx_mode = mode;
	freefile->fx_oldinum = ino;
	freefile->fx_devvp = ip->i_devvp;
	freefile->fx_fs = ip->i_fs;

	/*
	 * If the inodedep does not exist, then the zero'ed inode has
	 * been written to disk and we can free the file immediately.
	 */
	ACQUIRE_LOCK(&lk);
	if (inodedep_lookup(ip->i_fs, ino, 0, &inodedep) == 0) {
		add_to_worklist(&freefile->fx_list);
		FREE_LOCK(&lk);
		return;
	}

	/*
	 * If we still have a bitmap dependency, then the inode has never
	 * been written to disk. Drop the dependency as it is no longer
	 * necessary since the inode is being deallocated. We could process
	 * the freefile immediately, but then we would have to clear the
	 * id_inowait dependencies here and it is easier just to let the
	 * zero'ed inode be written and let them be cleaned up in the
	 * normal followup actions that follow the inode write.
	 */
	 if ((inodedep->id_state & DEPCOMPLETE) == 0) {
		inodedep->id_state |= DEPCOMPLETE;
		LIST_REMOVE(inodedep, id_deps);
		inodedep->id_buf = NULL;
	}
	/*
	 * If the inodedep has no dependencies associated with it,
	 * then we must free it here and free the file immediately.
	 * This case arises when an early allocation fails (for
	 * example, the user is over their file quota).
	 */
	if (free_inodedep(inodedep) == 0)
		WORKLIST_INSERT(&inodedep->id_inowait, &freefile->fx_list);
	else
		add_to_worklist(&freefile->fx_list);
	FREE_LOCK(&lk);
}

/*
 * Try to free an inodedep structure. Return 1 if it could be freed.
 */
static int
free_inodedep(inodedep)
	struct inodedep *inodedep;
{

	if ((inodedep->id_state & ONWORKLIST) != 0 ||
	    (inodedep->id_state & ALLCOMPLETE) != ALLCOMPLETE ||
	    LIST_FIRST(&inodedep->id_pendinghd) != NULL ||
	    LIST_FIRST(&inodedep->id_bufwait) != NULL ||
	    LIST_FIRST(&inodedep->id_inowait) != NULL ||
	    TAILQ_FIRST(&inodedep->id_inoupdt) != NULL ||
	    TAILQ_FIRST(&inodedep->id_newinoupdt) != NULL ||
	    inodedep->id_nlinkdelta != 0 || inodedep->id_savedino != NULL)
		return (0);
	LIST_REMOVE(inodedep, id_hash);
	WORKITEM_FREE(inodedep, D_INODEDEP);
	num_inodedep -= 1;
	return (1);
}

/*
 * This workitem routine performs the block de-allocation.
 * The workitem is added to the pending list after the updated
 * inode block has been written to disk.  As mentioned above,
 * checks regarding the number of blocks de-allocated (compared
 * to the number of blocks allocated for the file) are also
 * performed in this function.
 */
static void
handle_workitem_freeblocks(freeblks)
	struct freeblks *freeblks;
{
	struct inode tip;
	ufs_daddr_t bn;
	struct fs *fs;
	int i, level, bsize;
	long nblocks, blocksreleased = 0;
	int error, allerror = 0;
	ufs_lbn_t baselbns[NIADDR], tmpval;

	tip.i_number = freeblks->fb_previousinum;
	tip.i_devvp = freeblks->fb_devvp;
	tip.i_dev = freeblks->fb_devvp->v_rdev;
	tip.i_fs = freeblks->fb_fs;
	tip.i_size = freeblks->fb_oldsize;
	tip.i_uid = freeblks->fb_uid;
	fs = freeblks->fb_fs;
	tmpval = 1;
	baselbns[0] = NDADDR;
	for (i = 1; i < NIADDR; i++) {
		tmpval *= NINDIR(fs);
		baselbns[i] = baselbns[i - 1] + tmpval;
	}
	nblocks = btodb(fs->fs_bsize);
	blocksreleased = 0;
	/*
	 * Indirect blocks first.
	 */
	for (level = (NIADDR - 1); level >= 0; level--) {
		if ((bn = freeblks->fb_iblks[level]) == 0)
			continue;
		if ((error = indir_trunc(&tip, fsbtodb(fs, bn), level,
		    baselbns[level], &blocksreleased)) == 0)
			allerror = error;
		ffs_blkfree(&tip, bn, fs->fs_bsize);
		blocksreleased += nblocks;
	}
	/*
	 * All direct blocks or frags.
	 */
	for (i = (NDADDR - 1); i >= 0; i--) {
		if ((bn = freeblks->fb_dblks[i]) == 0)
			continue;
		bsize = blksize(fs, &tip, i);
		ffs_blkfree(&tip, bn, bsize);
		blocksreleased += btodb(bsize);
	}

#ifdef DIAGNOSTIC
	if (freeblks->fb_chkcnt != blocksreleased)
		panic("handle_workitem_freeblocks: block count");
	if (allerror)
		softdep_error("handle_workitem_freeblks", allerror);
#endif /* DIAGNOSTIC */
	WORKITEM_FREE(freeblks, D_FREEBLKS);
	num_freeblks -= 1;
}

/*
 * Release blocks associated with the inode ip and stored in the indirect
 * block dbn. If level is greater than SINGLE, the block is an indirect block
 * and recursive calls to indirtrunc must be used to cleanse other indirect
 * blocks.
 */
static int
indir_trunc(ip, dbn, level, lbn, countp)
	struct inode *ip;
	ufs_daddr_t dbn;
	int level;
	ufs_lbn_t lbn;
	long *countp;
{
	struct buf *bp;
	ufs_daddr_t *bap;
	ufs_daddr_t nb;
	struct fs *fs;
	struct worklist *wk;
	struct indirdep *indirdep;
	int i, lbnadd, nblocks;
	int error, allerror = 0;

	fs = ip->i_fs;
	lbnadd = 1;
	for (i = level; i > 0; i--)
		lbnadd *= NINDIR(fs);
	/*
	 * Get buffer of block pointers to be freed. This routine is not
	 * called until the zero'ed inode has been written, so it is safe
	 * to free blocks as they are encountered. Because the inode has
	 * been zero'ed, calls to bmap on these blocks will fail. So, we
	 * have to use the on-disk address and the block device for the
	 * filesystem to look them up. If the file was deleted before its
	 * indirect blocks were all written to disk, the routine that set
	 * us up (deallocate_dependencies) will have arranged to leave
	 * a complete copy of the indirect block in memory for our use.
	 * Otherwise we have to read the blocks in from the disk.
	 */
	ACQUIRE_LOCK(&lk);
	if ((bp = incore(ip->i_devvp, dbn)) != NULL &&
	    (wk = LIST_FIRST(&bp->b_dep)) != NULL) {
		if (wk->wk_type != D_INDIRDEP ||
		    (indirdep = WK_INDIRDEP(wk))->ir_savebp != bp ||
		    (indirdep->ir_state & GOINGAWAY) == 0)
			panic("indir_trunc: lost indirdep");
		WORKLIST_REMOVE(wk);
		WORKITEM_FREE(indirdep, D_INDIRDEP);
		if (LIST_FIRST(&bp->b_dep) != NULL)
			panic("indir_trunc: dangling dep");
		FREE_LOCK(&lk);
	} else {
		FREE_LOCK(&lk);
		error = bread(ip->i_devvp, dbn, (int)fs->fs_bsize, NOCRED, &bp);
		if (error)
			return (error);
	}
	/*
	 * Recursively free indirect blocks.
	 */
	bap = (ufs_daddr_t *)bp->b_data;
	nblocks = btodb(fs->fs_bsize);
	for (i = NINDIR(fs) - 1; i >= 0; i--) {
		if ((nb = bap[i]) == 0)
			continue;
		if (level != 0) {
			if ((error = indir_trunc(ip, fsbtodb(fs, nb),
			     level - 1, lbn + (i * lbnadd), countp)) != 0)
				allerror = error;
		}
		ffs_blkfree(ip, nb, fs->fs_bsize);
		*countp += nblocks;
	}
	bp->b_flags |= B_INVAL | B_NOCACHE;
	bp->b_flags &= ~B_XXX;
	brelse(bp);
	return (allerror);
}

/*
 * Free an allocindir.
 * This routine must be called with splbio interrupts blocked.
 */
static void
free_allocindir(aip, inodedep)
	struct allocindir *aip;
	struct inodedep *inodedep;
{
	struct freefrag *freefrag;

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("free_allocindir: lock not held");
#endif
	if ((aip->ai_state & DEPCOMPLETE) == 0)
		LIST_REMOVE(aip, ai_deps);
	if (aip->ai_state & ONWORKLIST)
		WORKLIST_REMOVE(&aip->ai_list);
	LIST_REMOVE(aip, ai_next);
	if ((freefrag = aip->ai_freefrag) != NULL) {
		if (inodedep == NULL)
			add_to_worklist(&freefrag->ff_list);
		else
			WORKLIST_INSERT(&inodedep->id_bufwait,
			    &freefrag->ff_list);
	}
	WORKITEM_FREE(aip, D_ALLOCINDIR);
}

/*
 * Directory entry addition dependencies.
 * 
 * When adding a new directory entry, the inode (with its incremented link
 * count) must be written to disk before the directory entry's pointer to it.
 * Also, if the inode is newly allocated, the corresponding freemap must be
 * updated (on disk) before the directory entry's pointer. These requirements
 * are met via undo/redo on the directory entry's pointer, which consists
 * simply of the inode number.
 * 
 * As directory entries are added and deleted, the free space within a
 * directory block can become fragmented.  The ufs file system will compact
 * a fragmented directory block to make space for a new entry. When this
 * occurs, the offsets of previously added entries change. Any "diradd"
 * dependency structures corresponding to these entries must be updated with
 * the new offsets.
 */

/*
 * This routine is called after the in-memory inode's link
 * count has been incremented, but before the directory entry's
 * pointer to the inode has been set.
 */
void 
softdep_setup_directory_add(bp, dp, diroffset, newinum, newdirbp)
	struct buf *bp;		/* buffer containing directory block */
	struct inode *dp;	/* inode for directory */
	off_t diroffset;	/* offset of new entry in directory */
	long newinum;		/* inode referenced by new directory entry */
	struct buf *newdirbp;	/* non-NULL => contents of new mkdir */
{
	int offset;		/* offset of new entry within directory block */
	ufs_lbn_t lbn;		/* block in directory containing new entry */
	struct fs *fs;
	struct diradd *dap;
	struct pagedep *pagedep;
	struct inodedep *inodedep;
	struct mkdir *mkdir1, *mkdir2;

	/*
	 * Whiteouts have no dependencies.
	 */
	if (newinum == WINO) {
		if (newdirbp != NULL)
			bdwrite(newdirbp);
		return;
	}

	fs = dp->i_fs;
	lbn = lblkno(fs, diroffset);
	offset = blkoff(fs, diroffset);
	MALLOC(dap, struct diradd *, sizeof(struct diradd), M_DIRADD, M_WAITOK);
	bzero(dap, sizeof(struct diradd));
	dap->da_list.wk_type = D_DIRADD;
	dap->da_offset = offset;
	dap->da_newinum = newinum;
	dap->da_state = ATTACHED;
	if (newdirbp == NULL) {
		dap->da_state |= DEPCOMPLETE;
		ACQUIRE_LOCK(&lk);
	} else {
		dap->da_state |= MKDIR_BODY | MKDIR_PARENT;
		MALLOC(mkdir1, struct mkdir *, sizeof(struct mkdir), M_MKDIR,
		    M_WAITOK);
		mkdir1->md_list.wk_type = D_MKDIR;
		mkdir1->md_state = MKDIR_BODY;
		mkdir1->md_diradd = dap;
		MALLOC(mkdir2, struct mkdir *, sizeof(struct mkdir), M_MKDIR,
		    M_WAITOK);
		mkdir2->md_list.wk_type = D_MKDIR;
		mkdir2->md_state = MKDIR_PARENT;
		mkdir2->md_diradd = dap;
		ACQUIRE_LOCK(&lk);
		/*
		 * Dependency on "." and ".." being written to disk.
		 */
		LIST_INSERT_HEAD(&mkdirlisthd, mkdir1, md_mkdirs);
		WORKLIST_INSERT(&newdirbp->b_dep, &mkdir1->md_list);
		bdwrite(newdirbp);
		/*
		 * Dependency on link count increase for parent directory
		 */
		if (inodedep_lookup(dp->i_fs, dp->i_number, 0, &inodedep) == 0
		    || (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) {
			dap->da_state &= ~MKDIR_PARENT;
			WORKITEM_FREE(mkdir2, D_MKDIR);
		} else {
			LIST_INSERT_HEAD(&mkdirlisthd, mkdir2, md_mkdirs);
			WORKLIST_INSERT(&inodedep->id_bufwait,&mkdir2->md_list);
		}
	}
	/*
	 * Link into parent directory pagedep to await its being written.
	 */
	if (pagedep_lookup(dp, lbn, DEPALLOC, &pagedep) == 0)
		WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
	dap->da_pagedep = pagedep;
	LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap,
	    da_pdlist);
	/*
	 * Link into its inodedep. Put it on the id_bufwait list if the inode
	 * is not yet written. If it is written, do the post-inode write
	 * processing to put it on the id_pendinghd list.
	 */
	(void) inodedep_lookup(fs, newinum, DEPALLOC, &inodedep);
	if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE)
		diradd_inode_written(dap, inodedep);
	else
		WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list);
	FREE_LOCK(&lk);
}

/*
 * This procedure is called to change the offset of a directory
 * entry when compacting a directory block which must be owned
 * exclusively by the caller. Note that the actual entry movement
 * must be done in this procedure to ensure that no I/O completions
 * occur while the move is in progress.
 */
void 
softdep_change_directoryentry_offset(dp, base, oldloc, newloc, entrysize)
	struct inode *dp;	/* inode for directory */
	caddr_t base;		/* address of dp->i_offset */
	caddr_t oldloc;		/* address of old directory location */
	caddr_t newloc;		/* address of new directory location */
	int entrysize;		/* size of directory entry */
{
	int offset, oldoffset, newoffset;
	struct pagedep *pagedep;
	struct diradd *dap;
	ufs_lbn_t lbn;

	ACQUIRE_LOCK(&lk);
	lbn = lblkno(dp->i_fs, dp->i_offset);
	offset = blkoff(dp->i_fs, dp->i_offset);
	if (pagedep_lookup(dp, lbn, 0, &pagedep) == 0)
		goto done;
	oldoffset = offset + (oldloc - base);
	newoffset = offset + (newloc - base);
	for (dap = LIST_FIRST(&pagedep->pd_diraddhd[DIRADDHASH(oldoffset)]);
	     dap; dap = LIST_NEXT(dap, da_pdlist)) {
		if (dap->da_offset != oldoffset)
			continue;
		dap->da_offset = newoffset;
		if (DIRADDHASH(newoffset) == DIRADDHASH(oldoffset))
			break;
		LIST_REMOVE(dap, da_pdlist);
		LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(newoffset)],
		    dap, da_pdlist);
		break;
	}
	if (dap == NULL) { 
		for (dap = LIST_FIRST(&pagedep->pd_pendinghd); 
		     dap; dap = LIST_NEXT(dap, da_pdlist)) { 
			if (dap->da_offset == oldoffset) { 
				dap->da_offset = newoffset; 
				break; 
			} 
		} 
	} 
done:
	bcopy(oldloc, newloc, entrysize);
	FREE_LOCK(&lk);
}

/*
 * Free a diradd dependency structure. This routine must be called
 * with splbio interrupts blocked.
 */
static void
free_diradd(dap)
	struct diradd *dap;
{
	struct dirrem *dirrem;
	struct pagedep *pagedep;
	struct inodedep *inodedep;
	struct mkdir *mkdir, *nextmd;

#ifdef DEBUG
	if (lk.lkt_held == -1)
		panic("free_diradd: lock not held");
#endif
	WORKLIST_REMOVE(&dap->da_list);
	LIST_REMOVE(dap, da_pdlist);
	if ((dap->da_state & DIRCHG) == 0) {
		pagedep = dap->da_pagedep;
	} else {
		dirrem = dap->da_previous;
		pagedep = dirrem->dm_pagedep;
		dirrem->dm_dirinum = pagedep->pd_ino;
		add_to_worklist(&dirrem->dm_list);
	}
	if (inodedep_lookup(VFSTOUFS(pagedep->pd_mnt)->um_fs, dap->da_newinum,
	    0, &inodedep) != 0)
		(void) free_inodedep(inodedep);
	if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) {
		for (mkdir = LIST_FIRST(&mkdirlisthd); mkdir; mkdir = nextmd) {
			nextmd = LIST_NEXT(mkdir, md_mkdirs);
			if (mkdir->md_diradd != dap)
				continue;
			dap->da_state &= ~mkdir->md_state;
			WORKLIST_REMOVE(&mkdir->md_list);
			LIST_REMOVE(mkdir, md_mkdirs);
			WORKITEM_FREE(mkdir, D_MKDIR);
		}
		if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0)
			panic("free_diradd: unfound ref");
	}
	WORKITEM_FREE(dap, D_DIRADD);
}

/*
 * Directory entry removal dependencies.
 * 
 * When removing a directory entry, the entry's inode pointer must be
 * zero'ed on disk before the corresponding inode's link count is decremented
 * (possibly freeing the inode for re-use). This dependency is handled by
 * updating the directory entry but delaying the inode count reduction until
 * after the directory block has been written to disk. After this point, the
 * inode count can be decremented whenever it is convenient.
 */

/*
 * This routine should be called immediately after removing
 * a directory entry.  The inode's link count should not be
 * decremented by the calling procedure -- the soft updates
 * code will do this task when it is safe.
 */
void 
softdep_setup_remove(bp, dp, ip, isrmdir)
	struct buf *bp;		/* buffer containing directory block */
	struct inode *dp;	/* inode for the directory being modified */
	struct inode *ip;	/* inode for directory entry being removed */
	int isrmdir;		/* indicates if doing RMDIR */
{
	struct dirrem *dirrem;

	/*
	 * Allocate a new dirrem if appropriate and ACQUIRE_LOCK.
	 */
	dirrem = newdirrem(bp, dp, ip, isrmdir);
	if ((dirrem->dm_state & COMPLETE) == 0) {
		LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd, dirrem,
		    dm_next);
	} else {
		dirrem->dm_dirinum = dirrem->dm_pagedep->pd_ino;
		add_to_worklist(&dirrem->dm_list);
	}
	FREE_LOCK(&lk);
}

/*
 * Allocate a new dirrem if appropriate and return it along with
 * its associated pagedep. Called without a lock, returns with lock.
 */
static struct dirrem *
newdirrem(bp, dp, ip, isrmdir)
	struct buf *bp;		/* buffer containing directory block */
	struct inode *dp;	/* inode for the directory being modified */
	struct inode *ip;	/* inode for directory entry being removed */
	int isrmdir;		/* indicates if doing RMDIR */
{
	int offset;
	ufs_lbn_t lbn;
	struct diradd *dap;
	struct dirrem *dirrem;
	struct pagedep *pagedep;

	/*
	 * Whiteouts have no deletion dependencies.
	 */
	if (ip == NULL)
		panic("newdirrem: whiteout");
	MALLOC(dirrem, struct dirrem *, sizeof(struct dirrem),
		M_DIRREM, M_WAITOK);
	bzero(dirrem, sizeof(struct dirrem));
	dirrem->dm_list.wk_type = D_DIRREM;
	dirrem->dm_state = isrmdir ? RMDIR : 0;
	dirrem->dm_mnt = ITOV(ip)->v_mount;
	dirrem->dm_oldinum = ip->i_number;

	ACQUIRE_LOCK(&lk);
	lbn = lblkno(dp->i_fs, dp->i_offset);
	offset = blkoff(dp->i_fs, dp->i_offset);
	if (pagedep_lookup(dp, lbn, DEPALLOC, &pagedep) == 0)
		WORKLIST_INSERT(&bp->b_dep, &pagedep->pd_list);
	dirrem->dm_pagedep = pagedep;
	/*
	 * Check for a diradd dependency for the same directory entry.
	 * If present, then both dependencies become obsolete and can
	 * be de-allocated. Check for an entry on both the pd_dirraddhd
	 * list and the pd_pendinghd list.
	 */
	for (dap = LIST_FIRST(&pagedep->pd_diraddhd[DIRADDHASH(offset)]);
	     dap; dap = LIST_NEXT(dap, da_pdlist))
		if (dap->da_offset == offset)
			break;
	if (dap == NULL) {
		for (dap = LIST_FIRST(&pagedep->pd_pendinghd);
		     dap; dap = LIST_NEXT(dap, da_pdlist))
			if (dap->da_offset == offset)
				break;
		if (dap == NULL)
			return (dirrem);
	}
	/*
	 * Must be ATTACHED at this point, so just delete it.
	 */
	if ((dap->da_state & ATTACHED) == 0)
		panic("newdirrem: not ATTACHED");
	if (dap->da_newinum != ip->i_number)
		panic("newdirrem: inum %d should be %d",
		    ip->i_number, dap->da_newinum);
	free_diradd(dap);
	dirrem->dm_state |= COMPLETE;
	return (dirrem);
}

/*
 * Directory entry change dependencies.
 * 
 * Changing an existing directory entry requires that an add operation
 * be completed first followed by a deletion. The semantics for the addition
 * are identical to the description of adding a new entry above except
 * that the rollback is to the old inode number rather than zero. Once
 * the addition dependency is completed, the removal is done as described
 * in the removal routine above.
 */

/*
 * This routine should be called immediately after changing
 * a directory entry.  The inode's link count should not be
 * decremented by the calling procedure -- the soft updates
 * code will perform this task when it is safe.
 */
void 
softdep_setup_directory_change(bp, dp, ip, newinum, isrmdir)
	struct buf *bp;		/* buffer containing directory block */
	struct inode *dp;	/* inode for the directory being modified */
	struct inode *ip;	/* inode for directory entry being removed */
	long newinum;		/* new inode number for changed entry */
	int isrmdir;		/* indicates if doing RMDIR */
{
	int offset;
	struct diradd *dap = NULL;
	struct dirrem *dirrem;
	struct pagedep *pagedep;
	struct inodedep *inodedep;

	offset = blkoff(dp->i_fs, dp->i_offset);

	/*
	 * Whiteouts do not need diradd dependencies.
	 */
	if (newinum != WINO) {
		MALLOC(dap, struct diradd *, sizeof(struct diradd),
		    M_DIRADD, M_WAITOK);
		bzero(dap, sizeof(struct diradd));
		dap->da_list.wk_type = D_DIRADD;
		dap->da_state = DIRCHG | ATTACHED | DEPCOMPLETE;
		dap->da_offset = offset;
		dap->da_newinum = newinum;
	}

	/*
	 * Allocate a new dirrem and ACQUIRE_LOCK.
	 */
	dirrem = newdirrem(bp, dp, ip, isrmdir);
	pagedep = dirrem->dm_pagedep;
	/*
	 * The possible values for isrmdir:
	 *	0 - non-directory file rename
	 *	1 - directory rename within same directory
	 *   inum - directory rename to new directory of given inode number
	 * When renaming to a new directory, we are both deleting and
	 * creating a new directory entry, so the link count on the new
	 * directory should not change. Thus we do not need the followup
	 * dirrem which is usually done in handle_workitem_remove. We set
	 * the DIRCHG flag to tell handle_workitem_remove to skip the 
	 * followup dirrem.
	 */
	if (isrmdir > 1)
		dirrem->dm_state |= DIRCHG;

	/*
	 * Whiteouts have no additional dependencies,
	 * so just put the dirrem on the correct list.
	 */
	if (newinum == WINO) {
		if ((dirrem->dm_state & COMPLETE) == 0) {
			LIST_INSERT_HEAD(&pagedep->pd_dirremhd, dirrem,
			    dm_next);
		} else {
			dirrem->dm_dirinum = pagedep->pd_ino;
			add_to_worklist(&dirrem->dm_list);
		}
		FREE_LOCK(&lk);
		return;
	}

	/*
	 * Link into its inodedep. Put it on the id_bufwait list if the inode
	 * is not yet written. If it is written, do the post-inode write
	 * processing to put it on the id_pendinghd list.
	 */
	dap->da_previous = dirrem;
	if (inodedep_lookup(dp->i_fs, newinum, DEPALLOC, &inodedep) == 0 ||
	    (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) {
		dap->da_state |= COMPLETE;
		LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
		WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list);
	} else {
		LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)],
		    dap, da_pdlist);
		WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list);
	}
	/*
	 * If the previous inode was never written or its previous directory
	 * entry was never written, then we do not want to roll back to this
	 * previous value. Instead we want to roll back to zero and immediately
	 * free the unwritten or unreferenced inode.
	 */
	if (dirrem->dm_state & COMPLETE) {
		dap->da_state &= ~DIRCHG;
		dap->da_pagedep = pagedep;
		dirrem->dm_dirinum = pagedep->pd_ino;
		add_to_worklist(&dirrem->dm_list);
	}
	FREE_LOCK(&lk);
}

/*
 * Called whenever the link count on an inode is increased.
 * It creates an inode dependency so that the new reference(s)
 * to the inode cannot be committed to disk until the updated
 * inode has been written.
 */
void
softdep_increase_linkcnt(ip)
	struct inode *ip;	/* the inode with the increased link count */
{
	struct inodedep *inodedep;

	ACQUIRE_LOCK(&lk);
	(void) inodedep_lookup(ip->i_fs, ip->i_number, DEPALLOC, &inodedep);
	FREE_LOCK(&lk);
}

/*
 * This workitem decrements the inode's link count.
 * If the link count reaches zero, the file is removed.
 */
static void 
handle_workitem_remove(dirrem)
	struct dirrem *dirrem;
{
	struct proc *p = CURPROC;	/* XXX */
	struct inodedep *inodedep;
	struct vnode *vp;
	struct inode *ip;
	int error;

	if ((error = VFS_VGET(dirrem->dm_mnt, dirrem->dm_oldinum, &vp)) != 0) {
		softdep_error("handle_workitem_remove: vget", error);
		return;
	}
	ip = VTOI(vp);
	/*
	 * Normal file deletion.
	 */
	if ((dirrem->dm_state & RMDIR) == 0) {
		ip->i_nlink--;
		if (ip->i_nlink < ip->i_effnlink)
			panic("handle_workitem_remove: bad file delta");
		ip->i_flag |= IN_CHANGE;
		vput(vp);
		WORKITEM_FREE(dirrem, D_DIRREM);
		return;
	}
	/*
	 * Directory deletion. Decrement reference count for both the
	 * just deleted parent directory entry and the reference for ".".
	 * Next truncate the directory to length zero. When the
	 * truncation completes, arrange to have the reference count on
	 * the parent decremented to account for the loss of "..".
	 */
	ip->i_nlink -= 2;
	if (ip->i_nlink < ip->i_effnlink)
		panic("handle_workitem_remove: bad dir delta");
	ip->i_flag |= IN_CHANGE;
	if ((error = UFS_TRUNCATE(vp, (off_t)0, 0, p->p_ucred, p)) != 0)
		softdep_error("handle_workitem_remove: truncate", error);
	/*
	 * Rename a directory to a new parent. Since, we are both deleting
	 * and creating a new directory entry, the link count on the new
	 * directory should not change. Thus we skip the followup dirrem.
	 */
	if (dirrem->dm_state & DIRCHG) {
		vput(vp);
		WORKITEM_FREE(dirrem, D_DIRREM);
		return;
	}
	ACQUIRE_LOCK(&lk);
	(void) inodedep_lookup(ip->i_fs, dirrem->dm_oldinum, DEPALLOC,
	    &inodedep);
	dirrem->dm_state = 0;
	dirrem->dm_oldinum = dirrem->dm_dirinum;
	WORKLIST_INSERT(&inodedep->id_inowait, &dirrem->dm_list);
	FREE_LOCK(&lk);
	vput(vp);
}

/*
 * Inode de-allocation dependencies.
 * 
 * When an inode's link count is reduced to zero, it can be de-allocated. We
 * found it convenient to postpone de-allocation until after the inode is
 * written to disk with its new link count (zero).  At this point, all of the
 * on-disk inode's block pointers are nullified and, with careful dependency
 * list ordering, all dependencies related to the inode will be satisfied and
 * the corresponding dependency structures de-allocated.  So, if/when the
 * inode is reused, there will be no mixing of old dependencies with new
 * ones.  This artificial dependency is set up by the block de-allocation
 * procedure above (softdep_setup_freeblocks) and completed by the
 * following procedure.
 */
static void 
handle_workitem_freefile(freefile)
	struct freefile *freefile;
{
	struct vnode vp;
	struct inode tip;
	struct inodedep *idp;
	int error;

#ifdef DEBUG
	ACQUIRE_LOCK(&lk);
	if (inodedep_lookup(freefile->fx_fs, freefile->fx_oldinum, 0, &idp))
		panic("handle_workitem_freefile: inodedep survived");
	FREE_LOCK(&lk);
#endif
	tip.i_devvp = freefile->fx_devvp;
	tip.i_dev = freefile->fx_devvp->v_rdev;
	tip.i_fs = freefile->fx_fs;
	vp.v_data = &tip;
	if ((error = ffs_freefile(&vp, freefile->fx_oldinum, freefile->fx_mode)) != 0)
		softdep_error("handle_workitem_freefile", error);
	WORKITEM_FREE(freefile, D_FREEFILE);
	num_freefile -= 1;
}

/*
 * Disk writes.
 * 
 * The dependency structures constructed above are most actively used when file
 * system blocks are written to disk.  No constraints are placed on when a
 * block can be written, but unsatisfied update dependencies are made safe by
 * modifying (or replacing) the source memory for the duration of the disk
 * write.  When the disk write completes, the memory block is again brought
 * up-to-date.
 *
 * In-core inode structure reclamation.
 * 
 * Because there are a finite number of "in-core" inode structures, they are
 * reused regularly.  By transferring all inode-related dependencies to the
 * in-memory inode block and indexing them separately (via "inodedep"s), we
 * can allow "in-core" inode structures to be reused at any time and avoid
 * any increase in contention.
 *
 * Called just before entering the device driver to initiate a new disk I/O.
 * The buffer must be locked, thus, no I/O completion operations can occur
 * while we are manipulating its associated dependencies.
 */
void 
softdep_disk_io_initiation(bp)
	struct buf *bp;		/* structure describing disk write to occur */
{
	struct worklist *wk, *nextwk;
	struct indirdep *indirdep;

	/*
	 * We only care about write operations. There should never
	 * be dependencies for reads.
	 */
	if (bp->b_flags & B_READ)
		panic("softdep_disk_io_initiation: read");
	/*
	 * Do any necessary pre-I/O processing.
	 */
	for (wk = LIST_FIRST(&bp->b_dep); wk; wk = nextwk) {
		nextwk = LIST_NEXT(wk, wk_list);
		switch (wk->wk_type) {

		case D_PAGEDEP:
			initiate_write_filepage(WK_PAGEDEP(wk), bp);
			continue;

		case D_INODEDEP:
			initiate_write_inodeblock(WK_INODEDEP(wk), bp);
			continue;

		case D_INDIRDEP:
			indirdep = WK_INDIRDEP(wk);
			if (indirdep->ir_state & GOINGAWAY)
				panic("disk_io_initiation: indirdep gone");
			/*
			 * If there are no remaining dependencies, this
			 * will be writing the real pointers, so the
			 * dependency can be freed.
			 */
			if (LIST_FIRST(&indirdep->ir_deplisthd) == NULL) {
				indirdep->ir_savebp->b_flags &= ~B_XXX;
				indirdep->ir_savebp->b_flags |= B_INVAL | B_NOCACHE;
				brelse(indirdep->ir_savebp);
				/* inline expand WORKLIST_REMOVE(wk); */
				wk->wk_state &= ~ONWORKLIST;
				LIST_REMOVE(wk, wk_list);
				WORKITEM_FREE(indirdep, D_INDIRDEP);
				continue;
			}
			/*
			 * Replace up-to-date version with safe version.
			 */
			ACQUIRE_LOCK(&lk);
			indirdep->ir_state &= ~ATTACHED;
			indirdep->ir_state |= UNDONE;
			MALLOC(indirdep->ir_saveddata, caddr_t, bp->b_bcount,
			    M_INDIRDEP, M_WAITOK);
			bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount);
			bcopy(indirdep->ir_savebp->b_data, bp->b_data,
			    bp->b_bcount);
			FREE_LOCK(&lk);
			continue;

		case D_MKDIR:
		case D_BMSAFEMAP:
		case D_ALLOCDIRECT:
		case D_ALLOCINDIR:
			continue;

		default:
			panic("handle_disk_io_initiation: Unexpected type %s",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
	}
}

/*
 * Called from within the procedure above to deal with unsatisfied
 * allocation dependencies in a directory. The buffer must be locked,
 * thus, no I/O completion operations can occur while we are
 * manipulating its associated dependencies.
 */
static void
initiate_write_filepage(pagedep, bp)
	struct pagedep *pagedep;
	struct buf *bp;
{
	struct diradd *dap;
	struct direct *ep;
	int i;

	if (pagedep->pd_state & IOSTARTED) {
		/*
		 * This can only happen if there is a driver that does not
		 * understand chaining. Here biodone will reissue the call
		 * to strategy for the incomplete buffers.
		 */
		printf("initiate_write_filepage: already started\n");
		return;
	}
	pagedep->pd_state |= IOSTARTED;
	ACQUIRE_LOCK(&lk);
	for (i = 0; i < DAHASHSZ; i++) {
		for (dap = LIST_FIRST(&pagedep->pd_diraddhd[i]); dap;
		     dap = LIST_NEXT(dap, da_pdlist)) {
			ep = (struct direct *)
			    ((char *)bp->b_data + dap->da_offset);
			if (ep->d_ino != dap->da_newinum)
				panic("%s: dir inum %d != new %d",
				    "initiate_write_filepage",
				    ep->d_ino, dap->da_newinum);
			if (dap->da_state & DIRCHG)
				ep->d_ino = dap->da_previous->dm_oldinum;
			else
				ep->d_ino = 0;
			dap->da_state &= ~ATTACHED;
			dap->da_state |= UNDONE;
		}
	}
	FREE_LOCK(&lk);
}

/*
 * Called from within the procedure above to deal with unsatisfied
 * allocation dependencies in an inodeblock. The buffer must be
 * locked, thus, no I/O completion operations can occur while we
 * are manipulating its associated dependencies.
 */
static void 
initiate_write_inodeblock(inodedep, bp)
	struct inodedep *inodedep;
	struct buf *bp;			/* The inode block */
{
	struct allocdirect *adp, *lastadp;
	struct dinode *dp;
	struct fs *fs;
	ufs_lbn_t prevlbn = 0;
	int i, deplist;

	if (inodedep->id_state & IOSTARTED)
		panic("initiate_write_inodeblock: already started");
	inodedep->id_state |= IOSTARTED;
	fs = inodedep->id_fs;
	dp = (struct dinode *)bp->b_data +
	    ino_to_fsbo(fs, inodedep->id_ino);
	/*
	 * If the bitmap is not yet written, then the allocated
	 * inode cannot be written to disk.
	 */
	if ((inodedep->id_state & DEPCOMPLETE) == 0) {
		if (inodedep->id_savedino != NULL)
			panic("initiate_write_inodeblock: already doing I/O");
		MALLOC(inodedep->id_savedino, struct dinode *,
		    sizeof(struct dinode), M_INODEDEP, M_WAITOK);
		*inodedep->id_savedino = *dp;
		bzero((caddr_t)dp, sizeof(struct dinode));
		return;
	}
	/*
	 * If no dependencies, then there is nothing to roll back.
	 */
	inodedep->id_savedsize = dp->di_size;
	if (TAILQ_FIRST(&inodedep->id_inoupdt) == NULL)
		return;
	/*
	 * Set the dependencies to busy.
	 */
	ACQUIRE_LOCK(&lk);
	for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
	     adp = TAILQ_NEXT(adp, ad_next)) {
#ifdef DIAGNOSTIC
		if (deplist != 0 && prevlbn >= adp->ad_lbn)
			panic("softdep_write_inodeblock: lbn order");
		prevlbn = adp->ad_lbn;
		if (adp->ad_lbn < NDADDR &&
		    dp->di_db[adp->ad_lbn] != adp->ad_newblkno)
			panic("%s: direct pointer #%ld mismatch %d != %d",
			    "softdep_write_inodeblock", adp->ad_lbn,
			    dp->di_db[adp->ad_lbn], adp->ad_newblkno);
		if (adp->ad_lbn >= NDADDR &&
		    dp->di_ib[adp->ad_lbn - NDADDR] != adp->ad_newblkno)
			panic("%s: indirect pointer #%ld mismatch %d != %d",
			    "softdep_write_inodeblock", adp->ad_lbn - NDADDR,
			    dp->di_ib[adp->ad_lbn - NDADDR], adp->ad_newblkno);
		deplist |= 1 << adp->ad_lbn;
		if ((adp->ad_state & ATTACHED) == 0)
			panic("softdep_write_inodeblock: Unknown state 0x%x",
			    adp->ad_state);
#endif /* DIAGNOSTIC */
		adp->ad_state &= ~ATTACHED;
		adp->ad_state |= UNDONE;
	}
	/*
	 * The on-disk inode cannot claim to be any larger than the last
	 * fragment that has been written. Otherwise, the on-disk inode
	 * might have fragments that were not the last block in the file
	 * which would corrupt the filesystem.
	 */
	for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
	     lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) {
		if (adp->ad_lbn >= NDADDR)
			break;
		dp->di_db[adp->ad_lbn] = adp->ad_oldblkno;
		/* keep going until hitting a rollback to a frag */
		if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize)
			continue;
		dp->di_size = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize;
		for (i = adp->ad_lbn + 1; i < NDADDR; i++) {
#ifdef DIAGNOSTIC
			if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0)
				panic("softdep_write_inodeblock: lost dep1");
#endif /* DIAGNOSTIC */
			dp->di_db[i] = 0;
		}
		for (i = 0; i < NIADDR; i++) {
#ifdef DIAGNOSTIC
			if (dp->di_ib[i] != 0 &&
			    (deplist & ((1 << NDADDR) << i)) == 0)
				panic("softdep_write_inodeblock: lost dep2");
#endif /* DIAGNOSTIC */
			dp->di_ib[i] = 0;
		}
		FREE_LOCK(&lk);
		return;
	}
	/*
	 * If we have zero'ed out the last allocated block of the file,
	 * roll back the size to the last currently allocated block.
	 * We know that this last allocated block is a full-sized as
	 * we already checked for fragments in the loop above.
	 */
	if (lastadp != NULL &&
	    dp->di_size <= (lastadp->ad_lbn + 1) * fs->fs_bsize) {
		for (i = lastadp->ad_lbn; i >= 0; i--)
			if (dp->di_db[i] != 0)
				break;
		dp->di_size = (i + 1) * fs->fs_bsize;
	}
	/*
	 * The only dependencies are for indirect blocks.
	 *
	 * The file size for indirect block additions is not guaranteed.
	 * Such a guarantee would be non-trivial to achieve. The conventional
	 * synchronous write implementation also does not make this guarantee.
	 * Fsck should catch and fix discrepancies. Arguably, the file size
	 * can be over-estimated without destroying integrity when the file
	 * moves into the indirect blocks (i.e., is large). If we want to
	 * postpone fsck, we are stuck with this argument.
	 */
	for (; adp; adp = TAILQ_NEXT(adp, ad_next))
		dp->di_ib[adp->ad_lbn - NDADDR] = 0;
	FREE_LOCK(&lk);
}

/*
 * This routine is called during the completion interrupt
 * service routine for a disk write (from the procedure called
 * by the device driver to inform the file system caches of
 * a request completion).  It should be called early in this
 * procedure, before the block is made available to other
 * processes or other routines are called.
 */
void 
softdep_disk_write_complete(bp)
	struct buf *bp;		/* describes the completed disk write */
{
	struct worklist *wk;
	struct workhead reattach;
	struct newblk *newblk;
	struct allocindir *aip;
	struct allocdirect *adp;
	struct indirdep *indirdep;
	struct inodedep *inodedep;
	struct bmsafemap *bmsafemap;

#ifdef DEBUG
	if (lk.lkt_held != -1)
		panic("softdep_disk_write_complete: lock is held");
	lk.lkt_held = -2;
#endif
	LIST_INIT(&reattach);
	while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) {
		WORKLIST_REMOVE(wk);
		switch (wk->wk_type) {

		case D_PAGEDEP:
			if (handle_written_filepage(WK_PAGEDEP(wk), bp))
				WORKLIST_INSERT(&reattach, wk);
			continue;

		case D_INODEDEP:
			if (handle_written_inodeblock(WK_INODEDEP(wk), bp))
				WORKLIST_INSERT(&reattach, wk);
			continue;

		case D_BMSAFEMAP:
			bmsafemap = WK_BMSAFEMAP(wk);
			while (newblk = LIST_FIRST(&bmsafemap->sm_newblkhd)) {
				newblk->nb_state |= DEPCOMPLETE;
				newblk->nb_bmsafemap = NULL;
				LIST_REMOVE(newblk, nb_deps);
			}
			while (adp = LIST_FIRST(&bmsafemap->sm_allocdirecthd)) {
				adp->ad_state |= DEPCOMPLETE;
				adp->ad_buf = NULL;
				LIST_REMOVE(adp, ad_deps);
				handle_allocdirect_partdone(adp);
			}
			while (aip = LIST_FIRST(&bmsafemap->sm_allocindirhd)) {
				aip->ai_state |= DEPCOMPLETE;
				aip->ai_buf = NULL;
				LIST_REMOVE(aip, ai_deps);
				handle_allocindir_partdone(aip);
			}
			while ((inodedep =
			       LIST_FIRST(&bmsafemap->sm_inodedephd)) != NULL) {
				inodedep->id_state |= DEPCOMPLETE;
				LIST_REMOVE(inodedep, id_deps);
				inodedep->id_buf = NULL;
			}
			WORKITEM_FREE(bmsafemap, D_BMSAFEMAP);
			continue;

		case D_MKDIR:
			handle_written_mkdir(WK_MKDIR(wk), MKDIR_BODY);
			continue;

		case D_ALLOCDIRECT:
			adp = WK_ALLOCDIRECT(wk);
			adp->ad_state |= COMPLETE;
			handle_allocdirect_partdone(adp);
			continue;

		case D_ALLOCINDIR:
			aip = WK_ALLOCINDIR(wk);
			aip->ai_state |= COMPLETE;
			handle_allocindir_partdone(aip);
			continue;

		case D_INDIRDEP:
			indirdep = WK_INDIRDEP(wk);
			if (indirdep->ir_state & GOINGAWAY)
				panic("disk_write_complete: indirdep gone");
			bcopy(indirdep->ir_saveddata, bp->b_data, bp->b_bcount);
			FREE(indirdep->ir_saveddata, M_INDIRDEP);
			indirdep->ir_saveddata = 0;
			indirdep->ir_state &= ~UNDONE;
			indirdep->ir_state |= ATTACHED;
			while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != 0) {
				LIST_REMOVE(aip, ai_next);
				handle_allocindir_partdone(aip);
			}
			WORKLIST_INSERT(&reattach, wk);
			bdirty(bp);
			continue;

		default:
			panic("handle_disk_write_complete: Unknown type %s",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
	}
	/*
	 * Reattach any requests that must be redone.
	 */
	while ((wk = LIST_FIRST(&reattach)) != NULL) {
		WORKLIST_REMOVE(wk);
		WORKLIST_INSERT(&bp->b_dep, wk);
	}
#ifdef DEBUG
	if (lk.lkt_held != -2)
		panic("softdep_disk_write_complete: lock lost");
	lk.lkt_held = -1;
#endif
}

/*
 * Called from within softdep_disk_write_complete above. Note that
 * this routine is always called from interrupt level with further
 * splbio interrupts blocked.
 */
static void 
handle_allocdirect_partdone(adp)
	struct allocdirect *adp;	/* the completed allocdirect */
{
	struct allocdirect *listadp;
	struct inodedep *inodedep;
	long bsize;

	if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE)
		return;
	if (adp->ad_buf != NULL)
		panic("handle_allocdirect_partdone: dangling dep");
	/*
	 * The on-disk inode cannot claim to be any larger than the last
	 * fragment that has been written. Otherwise, the on-disk inode
	 * might have fragments that were not the last block in the file
	 * which would corrupt the filesystem. Thus, we cannot free any
	 * allocdirects after one whose ad_oldblkno claims a fragment as
	 * these blocks must be rolled back to zero before writing the inode.
	 * We check the currently active set of allocdirects in id_inoupdt.
	 */
	inodedep = adp->ad_inodedep;
	bsize = inodedep->id_fs->fs_bsize;
	for (listadp = TAILQ_FIRST(&inodedep->id_inoupdt); listadp;
	     listadp = TAILQ_NEXT(listadp, ad_next)) {
		/* found our block */
		if (listadp == adp)
			break;
		/* continue if ad_oldlbn is not a fragment */
		if (listadp->ad_oldsize == 0 ||
		    listadp->ad_oldsize == bsize)
			continue;
		/* hit a fragment */
		return;
	}
	/*
	 * If we have reached the end of the current list without
	 * finding the just finished dependency, then it must be
	 * on the future dependency list. Future dependencies cannot
	 * be freed until they are moved to the current list.
	 */
	if (listadp == NULL) {
#ifdef DEBUG
		for (listadp = TAILQ_FIRST(&inodedep->id_newinoupdt); listadp;
		     listadp = TAILQ_NEXT(listadp, ad_next))
			/* found our block */
			if (listadp == adp)
				break;
		if (listadp == NULL)
			panic("handle_allocdirect_partdone: lost dep");
#endif /* DEBUG */
		return;
	}
	/*
	 * If we have found the just finished dependency, then free
	 * it along with anything that follows it that is complete.
	 */
	for (; adp; adp = listadp) {
		listadp = TAILQ_NEXT(adp, ad_next);
		if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE)
			return;
		free_allocdirect(&inodedep->id_inoupdt, adp, 1);
	}
}

/*
 * Called from within softdep_disk_write_complete above. Note that
 * this routine is always called from interrupt level with further
 * splbio interrupts blocked.
 */
static void
handle_allocindir_partdone(aip)
	struct allocindir *aip;		/* the completed allocindir */
{
	struct indirdep *indirdep;

	if ((aip->ai_state & ALLCOMPLETE) != ALLCOMPLETE)
		return;
	if (aip->ai_buf != NULL)
		panic("handle_allocindir_partdone: dangling dependency");
	indirdep = aip->ai_indirdep;
	if (indirdep->ir_state & UNDONE) {
		LIST_REMOVE(aip, ai_next);
		LIST_INSERT_HEAD(&indirdep->ir_donehd, aip, ai_next);
		return;
	}
	((ufs_daddr_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] =
	    aip->ai_newblkno;
	LIST_REMOVE(aip, ai_next);
	if (aip->ai_freefrag != NULL)
		add_to_worklist(&aip->ai_freefrag->ff_list);
	WORKITEM_FREE(aip, D_ALLOCINDIR);
}

/*
 * Called from within softdep_disk_write_complete above to restore
 * in-memory inode block contents to their most up-to-date state. Note
 * that this routine is always called from interrupt level with further
 * splbio interrupts blocked.
 */
static int 
handle_written_inodeblock(inodedep, bp)
	struct inodedep *inodedep;
	struct buf *bp;		/* buffer containing the inode block */
{
	struct worklist *wk, *filefree;
	struct allocdirect *adp, *nextadp;
	struct dinode *dp;
	int hadchanges;

	if ((inodedep->id_state & IOSTARTED) == 0)
		panic("handle_written_inodeblock: not started");
	inodedep->id_state &= ~IOSTARTED;
	inodedep->id_state |= COMPLETE;
	dp = (struct dinode *)bp->b_data +
	    ino_to_fsbo(inodedep->id_fs, inodedep->id_ino);
	/*
	 * If we had to rollback the inode allocation because of
	 * bitmaps being incomplete, then simply restore it.
	 * Keep the block dirty so that it will not be reclaimed until
	 * all associated dependencies have been cleared and the
	 * corresponding updates written to disk.
	 */
	if (inodedep->id_savedino != NULL) {
		*dp = *inodedep->id_savedino;
		FREE(inodedep->id_savedino, M_INODEDEP);
		inodedep->id_savedino = NULL;
		bdirty(bp);
		return (1);
	}
	/*
	 * Roll forward anything that had to be rolled back before 
	 * the inode could be updated.
	 */
	hadchanges = 0;
	for (adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = nextadp) {
		nextadp = TAILQ_NEXT(adp, ad_next);
		if (adp->ad_state & ATTACHED)
			panic("handle_written_inodeblock: new entry");
		if (adp->ad_lbn < NDADDR) {
			if (dp->di_db[adp->ad_lbn] != adp->ad_oldblkno)
				panic("%s: %s #%ld mismatch %d != %d",
				    "handle_written_inodeblock",
				    "direct pointer", adp->ad_lbn,
				    dp->di_db[adp->ad_lbn], adp->ad_oldblkno);
			dp->di_db[adp->ad_lbn] = adp->ad_newblkno;
		} else {
			if (dp->di_ib[adp->ad_lbn - NDADDR] != 0)
				panic("%s: %s #%ld allocated as %d",
				    "handle_written_inodeblock",
				    "indirect pointer", adp->ad_lbn - NDADDR,
				    dp->di_ib[adp->ad_lbn - NDADDR]);
			dp->di_ib[adp->ad_lbn - NDADDR] = adp->ad_newblkno;
		}
		adp->ad_state &= ~UNDONE;
		adp->ad_state |= ATTACHED;
		hadchanges = 1;
	}
	/*
	 * Reset the file size to its most up-to-date value.
	 */
	if (inodedep->id_savedsize == -1)
		panic("handle_written_inodeblock: bad size");
	if (dp->di_size != inodedep->id_savedsize) {
		dp->di_size = inodedep->id_savedsize;
		hadchanges = 1;
	}
	inodedep->id_savedsize = -1;
	/*
	 * If there were any rollbacks in the inode block, then it must be
	 * marked dirty so that its will eventually get written back in
	 * its correct form.
	 */
	if (hadchanges)
		bdirty(bp);
	/*
	 * Process any allocdirects that completed during the update.
	 */
	if ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != NULL)
		handle_allocdirect_partdone(adp);
	/*
	 * Process deallocations that were held pending until the
	 * inode had been written to disk. Freeing of the inode
	 * is delayed until after all blocks have been freed to
	 * avoid creation of new <vfsid, inum, lbn> triples
	 * before the old ones have been deleted.
	 */
	filefree = NULL;
	while ((wk = LIST_FIRST(&inodedep->id_bufwait)) != NULL) {
		WORKLIST_REMOVE(wk);
		switch (wk->wk_type) {

		case D_FREEFILE:
			/*
			 * We defer adding filefree to the worklist until
			 * all other additions have been made to ensure
			 * that it will be done after all the old blocks
			 * have been freed.
			 */
			if (filefree != NULL)
				panic("handle_written_inodeblock: filefree");
			filefree = wk;
			continue;

		case D_MKDIR:
			handle_written_mkdir(WK_MKDIR(wk), MKDIR_PARENT);
			continue;

		case D_DIRADD:
			diradd_inode_written(WK_DIRADD(wk), inodedep);
			continue;

		case D_FREEBLKS:
		case D_FREEFRAG:
		case D_DIRREM:
			add_to_worklist(wk);
			continue;

		default:
			panic("handle_written_inodeblock: Unknown type %s",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
	}
	if (filefree != NULL) {
		if (free_inodedep(inodedep) == 0)
			panic("handle_written_inodeblock: live inodedep");
		add_to_worklist(filefree);
		return (0);
	}

	/*
	 * If no outstanding dependencies, free it.
	 */
	if (free_inodedep(inodedep) || TAILQ_FIRST(&inodedep->id_inoupdt) == 0)
		return (0);
	return (hadchanges);
}

/*
 * Process a diradd entry after its dependent inode has been written.
 * This routine must be called with splbio interrupts blocked.
 */
static void
diradd_inode_written(dap, inodedep)
	struct diradd *dap;
	struct inodedep *inodedep;
{
	struct pagedep *pagedep;

	dap->da_state |= COMPLETE;
	if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) {
		if (dap->da_state & DIRCHG)
			pagedep = dap->da_previous->dm_pagedep;
		else
			pagedep = dap->da_pagedep;
		LIST_REMOVE(dap, da_pdlist);
		LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
	}
	WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list);
}

/*
 * Handle the completion of a mkdir dependency.
 */
static void
handle_written_mkdir(mkdir, type)
	struct mkdir *mkdir;
	int type;
{
	struct diradd *dap;
	struct pagedep *pagedep;

	if (mkdir->md_state != type)
		panic("handle_written_mkdir: bad type");
	dap = mkdir->md_diradd;
	dap->da_state &= ~type;
	if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0)
		dap->da_state |= DEPCOMPLETE;
	if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) {
		if (dap->da_state & DIRCHG)
			pagedep = dap->da_previous->dm_pagedep;
		else
			pagedep = dap->da_pagedep;
		LIST_REMOVE(dap, da_pdlist);
		LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist);
	}
	LIST_REMOVE(mkdir, md_mkdirs);
	WORKITEM_FREE(mkdir, D_MKDIR);
}

/*
 * Called from within softdep_disk_write_complete above.
 * A write operation was just completed. Removed inodes can
 * now be freed and associated block pointers may be committed.
 * Note that this routine is always called from interrupt level
 * with further splbio interrupts blocked.
 */
static int 
handle_written_filepage(pagedep, bp)
	struct pagedep *pagedep;
	struct buf *bp;		/* buffer containing the written page */
{
	struct dirrem *dirrem;
	struct diradd *dap, *nextdap;
	struct direct *ep;
	int i, chgs;

	if ((pagedep->pd_state & IOSTARTED) == 0)
		panic("handle_written_filepage: not started");
	pagedep->pd_state &= ~IOSTARTED;
	/*
	 * Process any directory removals that have been committed.
	 */
	while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd)) != NULL) {
		LIST_REMOVE(dirrem, dm_next);
		dirrem->dm_dirinum = pagedep->pd_ino;
		add_to_worklist(&dirrem->dm_list);
	}
	/*
	 * Free any directory additions that have been committed.
	 */
	while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL)
		free_diradd(dap);
	/*
	 * Uncommitted directory entries must be restored.
	 */
	for (chgs = 0, i = 0; i < DAHASHSZ; i++) {
		for (dap = LIST_FIRST(&pagedep->pd_diraddhd[i]); dap;
		     dap = nextdap) {
			nextdap = LIST_NEXT(dap, da_pdlist);
			if (dap->da_state & ATTACHED)
				panic("handle_written_filepage: attached");
			ep = (struct direct *)
			    ((char *)bp->b_data + dap->da_offset);
			ep->d_ino = dap->da_newinum;
			dap->da_state &= ~UNDONE;
			dap->da_state |= ATTACHED;
			chgs = 1;
			/*
			 * If the inode referenced by the directory has
			 * been written out, then the dependency can be
			 * moved to the pending list.
			 */
			if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) {
				LIST_REMOVE(dap, da_pdlist);
				LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap,
				    da_pdlist);
			}
		}
	}
	/*
	 * If there were any rollbacks in the directory, then it must be
	 * marked dirty so that its will eventually get written back in
	 * its correct form.
	 */
	if (chgs)
		bdirty(bp);
	/*
	 * If no dependencies remain, the pagedep will be freed.
	 * Otherwise it will remain to update the page before it
	 * is written back to disk.
	 */
	if (LIST_FIRST(&pagedep->pd_pendinghd) == 0) {
		for (i = 0; i < DAHASHSZ; i++)
			if (LIST_FIRST(&pagedep->pd_diraddhd[i]) != NULL)
				break;
		if (i == DAHASHSZ) {
			LIST_REMOVE(pagedep, pd_hash);
			WORKITEM_FREE(pagedep, D_PAGEDEP);
			return (0);
		}
	}
	return (1);
}

/*
 * Writing back in-core inode structures.
 * 
 * The file system only accesses an inode's contents when it occupies an
 * "in-core" inode structure.  These "in-core" structures are separate from
 * the page frames used to cache inode blocks.  Only the latter are
 * transferred to/from the disk.  So, when the updated contents of the
 * "in-core" inode structure are copied to the corresponding in-memory inode
 * block, the dependencies are also transferred.  The following procedure is
 * called when copying a dirty "in-core" inode to a cached inode block.
 */

/*
 * Called when an inode is loaded from disk. If the effective link count
 * differed from the actual link count when it was last flushed, then we
 * need to ensure that the correct effective link count is put back.
 */
void 
softdep_load_inodeblock(ip)
	struct inode *ip;	/* the "in_core" copy of the inode */
{
	struct inodedep *inodedep;

	/*
	 * Check for alternate nlink count.
	 */
	ip->i_effnlink = ip->i_nlink;
	ACQUIRE_LOCK(&lk);
	if (inodedep_lookup(ip->i_fs, ip->i_number, 0, &inodedep) == 0) {
		FREE_LOCK(&lk);
		return;
	}
	if (inodedep->id_nlinkdelta != 0) {
		ip->i_effnlink -= inodedep->id_nlinkdelta;
		ip->i_flag |= IN_MODIFIED;
		inodedep->id_nlinkdelta = 0;
		(void) free_inodedep(inodedep);
	}
	FREE_LOCK(&lk);
}

/*
 * This routine is called just before the "in-core" inode
 * information is to be copied to the in-memory inode block.
 * Recall that an inode block contains several inodes. If
 * the force flag is set, then the dependencies will be
 * cleared so that the update can always be made. Note that
 * the buffer is locked when this routine is called, so we
 * will never be in the middle of writing the inode block 
 * to disk.
 */
void 
softdep_update_inodeblock(ip, bp, waitfor)
	struct inode *ip;	/* the "in_core" copy of the inode */
	struct buf *bp;		/* the buffer containing the inode block */
	int waitfor;		/* nonzero => update must be allowed */
{
	struct inodedep *inodedep;
	struct worklist *wk;
	int error, gotit;

	/*
	 * If the effective link count is not equal to the actual link
	 * count, then we must track the difference in an inodedep while
	 * the inode is (potentially) tossed out of the cache. Otherwise,
	 * if there is no existing inodedep, then there are no dependencies
	 * to track.
	 */
	ACQUIRE_LOCK(&lk);
	if (ip->i_effnlink != ip->i_nlink) {
		(void) inodedep_lookup(ip->i_fs, ip->i_number, DEPALLOC,
		    &inodedep);
	} else if (inodedep_lookup(ip->i_fs, ip->i_number, 0, &inodedep) == 0) {
		FREE_LOCK(&lk);
		return;
	}
	if (ip->i_nlink < ip->i_effnlink)
		panic("softdep_update_inodeblock: bad delta");
	inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink;
	/*
	 * Changes have been initiated. Anything depending on these
	 * changes cannot occur until this inode has been written.
	 */
	inodedep->id_state &= ~COMPLETE;
	if ((inodedep->id_state & ONWORKLIST) == 0)
		WORKLIST_INSERT(&bp->b_dep, &inodedep->id_list);
	/*
	 * Any new dependencies associated with the incore inode must 
	 * now be moved to the list associated with the buffer holding
	 * the in-memory copy of the inode. Once merged process any
	 * allocdirects that are completed by the merger.
	 */
	merge_inode_lists(inodedep);
	if (TAILQ_FIRST(&inodedep->id_inoupdt) != NULL)
		handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_inoupdt));
	/*
	 * Now that the inode has been pushed into the buffer, the
	 * operations dependent on the inode being written to disk
	 * can be moved to the id_bufwait so that they will be
	 * processed when the buffer I/O completes.
	 */
	while ((wk = LIST_FIRST(&inodedep->id_inowait)) != NULL) {
		WORKLIST_REMOVE(wk);
		WORKLIST_INSERT(&inodedep->id_bufwait, wk);
	}
	/*
	 * Newly allocated inodes cannot be written until the bitmap
	 * that allocates them have been written (indicated by
	 * DEPCOMPLETE being set in id_state). If we are doing a
	 * forced sync (e.g., an fsync on a file), we force the bitmap
	 * to be written so that the update can be done.
	 */
	if ((inodedep->id_state & DEPCOMPLETE) != 0 || waitfor == 0) {
		FREE_LOCK(&lk);
		return;
	}
	gotit = getdirtybuf(&inodedep->id_buf, MNT_WAIT);
	FREE_LOCK(&lk);
	if (gotit && (error = VOP_BWRITE(inodedep->id_buf)) != 0)
		softdep_error("softdep_update_inodeblock: bwrite", error);
	if ((inodedep->id_state & DEPCOMPLETE) == 0)
		panic("softdep_update_inodeblock: update failed");
}

/*
 * Merge the new inode dependency list (id_newinoupdt) into the old
 * inode dependency list (id_inoupdt). This routine must be called
 * with splbio interrupts blocked.
 */
static void
merge_inode_lists(inodedep)
	struct inodedep *inodedep;
{
	struct allocdirect *listadp, *newadp;

	newadp = TAILQ_FIRST(&inodedep->id_newinoupdt);
	for (listadp = TAILQ_FIRST(&inodedep->id_inoupdt); listadp && newadp;) {
		if (listadp->ad_lbn < newadp->ad_lbn) {
			listadp = TAILQ_NEXT(listadp, ad_next);
			continue;
		}
		TAILQ_REMOVE(&inodedep->id_newinoupdt, newadp, ad_next);
		TAILQ_INSERT_BEFORE(listadp, newadp, ad_next);
		if (listadp->ad_lbn == newadp->ad_lbn) {
			allocdirect_merge(&inodedep->id_inoupdt, newadp,
			    listadp);
			listadp = newadp;
		}
		newadp = TAILQ_FIRST(&inodedep->id_newinoupdt);
	}
	while ((newadp = TAILQ_FIRST(&inodedep->id_newinoupdt)) != NULL) {
		TAILQ_REMOVE(&inodedep->id_newinoupdt, newadp, ad_next);
		TAILQ_INSERT_TAIL(&inodedep->id_inoupdt, newadp, ad_next);
	}
}

/*
 * If we are doing an fsync, then we must ensure that any directory
 * entries for the inode have been written after the inode gets to disk.
 */
int
softdep_fsync(vp)
	struct vnode *vp;	/* the "in_core" copy of the inode */
{
	struct diradd *dap, *olddap;
	struct inodedep *inodedep;
	struct pagedep *pagedep;
	struct worklist *wk;
	struct mount *mnt;
	struct vnode *pvp;
	struct inode *ip;
	struct buf *bp;
	struct fs *fs;
	struct proc *p = CURPROC;		/* XXX */
	int error, ret, flushparent;
#ifndef __FreeBSD__
	struct timeval tv;
#endif
	ino_t parentino;
	ufs_lbn_t lbn;

	ip = VTOI(vp);
	fs = ip->i_fs;
	for (error = 0, flushparent = 0, olddap = NULL; ; ) {
		ACQUIRE_LOCK(&lk);
		if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) == 0)
			break;
		if (LIST_FIRST(&inodedep->id_inowait) != NULL ||
		    LIST_FIRST(&inodedep->id_bufwait) != NULL ||
		    TAILQ_FIRST(&inodedep->id_inoupdt) != NULL ||
		    TAILQ_FIRST(&inodedep->id_newinoupdt) != NULL)
			panic("softdep_fsync: pending ops");
		if ((wk = LIST_FIRST(&inodedep->id_pendinghd)) == NULL)
			break;
		if (wk->wk_type != D_DIRADD)
			panic("softdep_fsync: Unexpected type %s",
			    TYPENAME(wk->wk_type));
		dap = WK_DIRADD(wk);
		/*
		 * If we have failed to get rid of all the dependencies
		 * then something is seriously wrong.
		 */
		if (dap == olddap)
			panic("softdep_fsync: flush failed");
		olddap = dap;
		/*
		 * Flush our parent if this directory entry
		 * has a MKDIR_PARENT dependency.
		 */
		if (dap->da_state & DIRCHG)
			pagedep = dap->da_previous->dm_pagedep;
		else
			pagedep = dap->da_pagedep;
		mnt = pagedep->pd_mnt;
		parentino = pagedep->pd_ino;
		lbn = pagedep->pd_lbn;
		if ((dap->da_state & (MKDIR_BODY | COMPLETE)) != COMPLETE)
			panic("softdep_fsync: dirty");
		flushparent = dap->da_state & MKDIR_PARENT;
		/*
		 * If we are being fsync'ed as part of vgone'ing this vnode,
		 * then we will not be able to release and recover the
		 * vnode below, so we just have to give up on writing its
		 * directory entry out. It will eventually be written, just
		 * not now, but then the user was not asking to have it
		 * written, so we are not breaking any promises.
		 */
		if (vp->v_flag & VXLOCK)
			break;
		/*
		 * We prevent deadlock by always fetching inodes from the
		 * root, moving down the directory tree. Thus, when fetching
		 * our parent directory, we must unlock ourselves before
		 * requesting the lock on our parent. See the comment in
		 * ufs_lookup for details on possible races.
		 */
		FREE_LOCK(&lk);
		VOP_UNLOCK(vp, 0, p);
		if ((error = VFS_VGET(mnt, parentino, &pvp)) != 0) {
			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
			return (error);
		}
		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
		if (flushparent) {
#ifdef __FreeBSD__
			error = UFS_UPDATE(pvp, 1);
#else
			tv = time;
			error = UFS_UPDATE(pvp, &tv, &tv, 1);
#endif
			if (error) {
				vput(pvp);
				return (error);
			}
		}
		/*
		 * Flush directory page containing the inode's name.
		 */
		error = bread(pvp, lbn, blksize(fs, VTOI(pvp), lbn), p->p_ucred,
		    &bp);
		ret = VOP_BWRITE(bp);
		vput(pvp);
		if (error != 0)
			return (error);
		if (ret != 0)
			return (ret);
	}
	FREE_LOCK(&lk);
	return (0);
}

/*
 * This routine is called when we are trying to synchronously flush a
 * file. This routine must eliminate any filesystem metadata dependencies
 * so that the syncing routine can succeed by pushing the dirty blocks
 * associated with the file. If any I/O errors occur, they are returned.
 */
int
softdep_sync_metadata(ap)
	struct vop_fsync_args /* {
		struct vnode *a_vp;
		struct ucred *a_cred;
		int a_waitfor;
		struct proc *a_p;
	} */ *ap;
{
	struct vnode *vp = ap->a_vp;
	struct pagedep *pagedep;
	struct allocdirect *adp;
	struct allocindir *aip;
	struct buf *bp, *nbp;
	struct worklist *wk;
	int i, error, waitfor;

	/*
	 * Check whether this vnode is involved in a filesystem
	 * that is doing soft dependency processing.
	 */
	if (vp->v_type != VBLK) {
		if (!DOINGSOFTDEP(vp))
			return (0);
	} else
		if (vp->v_specmountpoint == NULL ||
		    (vp->v_specmountpoint->mnt_flag & MNT_SOFTDEP) == 0)
			return (0);
	/*
	 * Ensure that any direct block dependencies have been cleared.
	 */
	ACQUIRE_LOCK(&lk);
	if (error = flush_inodedep_deps(VTOI(vp)->i_fs, VTOI(vp)->i_number)) {
		FREE_LOCK(&lk);
		return (error);
	}
	/*
	 * For most files, the only metadata dependencies are the
	 * cylinder group maps that allocate their inode or blocks.
	 * The block allocation dependencies can be found by traversing
	 * the dependency lists for any buffers that remain on their
	 * dirty buffer list. The inode allocation dependency will
	 * be resolved when the inode is updated with MNT_WAIT.
	 * This work is done in two passes. The first pass grabs most
	 * of the buffers and begins asynchronously writing them. The
	 * only way to wait for these asynchronous writes is to sleep
	 * on the filesystem vnode which may stay busy for a long time
	 * if the filesystem is active. So, instead, we make a second
	 * pass over the dependencies blocking on each write. In the
	 * usual case we will be blocking against a write that we
	 * initiated, so when it is done the dependency will have been
	 * resolved. Thus the second pass is expected to end quickly.
	 */
	waitfor = MNT_NOWAIT;
top:
	if (getdirtybuf(&TAILQ_FIRST(&vp->v_dirtyblkhd), MNT_WAIT) == 0) {
		while (vp->v_numoutput) {
			vp->v_flag |= VBWAIT;
			FREE_LOCK_INTERLOCKED(&lk);
			tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1,
				"sdsynm", 0);
			ACQUIRE_LOCK_INTERLOCKED(&lk);
		}
		FREE_LOCK(&lk);
		return (0);
	}
	bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
loop:
	/*
	 * As we hold the buffer locked, none of its dependencies
	 * will disappear.
	 */
	for (wk = LIST_FIRST(&bp->b_dep); wk;
	     wk = LIST_NEXT(wk, wk_list)) {
		switch (wk->wk_type) {

		case D_ALLOCDIRECT:
			adp = WK_ALLOCDIRECT(wk);
			if (adp->ad_state & DEPCOMPLETE)
				break;
			nbp = adp->ad_buf;
			if (getdirtybuf(&nbp, waitfor) == 0)
				break;
			FREE_LOCK(&lk);
			if (waitfor == MNT_NOWAIT) {
				bawrite(nbp);
			} else if ((error = VOP_BWRITE(nbp)) != 0) {
				bawrite(bp);
				return (error);
			}
			ACQUIRE_LOCK(&lk);
			break;

		case D_ALLOCINDIR:
			aip = WK_ALLOCINDIR(wk);
			if (aip->ai_state & DEPCOMPLETE)
				break;
			nbp = aip->ai_buf;
			if (getdirtybuf(&nbp, waitfor) == 0)
				break;
			FREE_LOCK(&lk);
			if (waitfor == MNT_NOWAIT) {
				bawrite(nbp);
			} else if ((error = VOP_BWRITE(nbp)) != 0) {
				bawrite(bp);
				return (error);
			}
			ACQUIRE_LOCK(&lk);
			break;

		case D_INDIRDEP:
		restart:
			for (aip = LIST_FIRST(&WK_INDIRDEP(wk)->ir_deplisthd);
			     aip; aip = LIST_NEXT(aip, ai_next)) {
				if (aip->ai_state & DEPCOMPLETE)
					continue;
				nbp = aip->ai_buf;
				if (getdirtybuf(&nbp, MNT_WAIT) == 0)
					goto restart;
				FREE_LOCK(&lk);
				if ((error = VOP_BWRITE(nbp)) != 0) {
					bawrite(bp);
					return (error);
				}
				ACQUIRE_LOCK(&lk);
				goto restart;
			}
			break;

		case D_INODEDEP:
			if ((error = flush_inodedep_deps(WK_INODEDEP(wk)->id_fs,
			    WK_INODEDEP(wk)->id_ino)) != 0) {
				FREE_LOCK(&lk);
				bawrite(bp);
				return (error);
			}
			break;

		case D_PAGEDEP:
			/*
			 * We are trying to sync a directory that may
			 * have dependencies on both its own metadata
			 * and/or dependencies on the inodes of any
			 * recently allocated files. We walk its diradd
			 * lists pushing out the associated inode.
			 */
			pagedep = WK_PAGEDEP(wk);
			for (i = 0; i < DAHASHSZ; i++) {
				if (LIST_FIRST(&pagedep->pd_diraddhd[i]) == 0)
					continue;
				if (error = flush_pagedep_deps(vp,
				   pagedep->pd_mnt, &pagedep->pd_diraddhd[i])) {
					FREE_LOCK(&lk);
					bawrite(bp);
					return (error);
				}
			}
			break;

		default:
			panic("softdep_sync_metadata: Unknown type %s",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
	}
	(void) getdirtybuf(&TAILQ_NEXT(bp, b_vnbufs), MNT_WAIT);
	nbp = TAILQ_NEXT(bp, b_vnbufs);
	FREE_LOCK(&lk);
	bawrite(bp);
	ACQUIRE_LOCK(&lk);
	if (nbp != NULL) {
		bp = nbp;
		goto loop;
	}
	/*
	 * We must wait for any I/O in progress to finish so that
	 * all potential buffers on the dirty list will be visible.
	 * Once they are all there, proceed with the second pass
	 * which will wait for the I/O as per above.
	 */
	while (vp->v_numoutput) {
		vp->v_flag |= VBWAIT;
		FREE_LOCK_INTERLOCKED(&lk);
		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "sdsynm", 0);
		ACQUIRE_LOCK_INTERLOCKED(&lk);
	}
	/*
	 * The brief unlock is to allow any pent up dependency
	 * processing to be done.
	 */
	if (waitfor == MNT_NOWAIT) {
		waitfor = MNT_WAIT;
		FREE_LOCK(&lk);
		ACQUIRE_LOCK(&lk);
		goto top;
	}

	/*
	 * If we have managed to get rid of all the dirty buffers,
	 * then we are done. For certain directories and block
	 * devices, we may need to do further work.
	 */
	if (TAILQ_FIRST(&vp->v_dirtyblkhd) == NULL) {
		FREE_LOCK(&lk);
		return (0);
	}

	FREE_LOCK(&lk);
	/*
	 * If we are trying to sync a block device, some of its buffers may
	 * contain metadata that cannot be written until the contents of some
	 * partially written files have been written to disk. The only easy
	 * way to accomplish this is to sync the entire filesystem (luckily
	 * this happens rarely).
	 */
	if (vp->v_type == VBLK && vp->v_specmountpoint && !VOP_ISLOCKED(vp) &&
	    (error = VFS_SYNC(vp->v_specmountpoint, MNT_WAIT, ap->a_cred,
	     ap->a_p)) != 0)
		return (error);
	return (0);
}

/*
 * Flush the dependencies associated with an inodedep.
 * Called with splbio blocked.
 */
static int
flush_inodedep_deps(fs, ino)
	struct fs *fs;
	ino_t ino;
{
	struct inodedep *inodedep;
	struct allocdirect *adp;
	int error, waitfor;
	struct buf *bp;

	/*
	 * This work is done in two passes. The first pass grabs most
	 * of the buffers and begins asynchronously writing them. The
	 * only way to wait for these asynchronous writes is to sleep
	 * on the filesystem vnode which may stay busy for a long time
	 * if the filesystem is active. So, instead, we make a second
	 * pass over the dependencies blocking on each write. In the
	 * usual case we will be blocking against a write that we
	 * initiated, so when it is done the dependency will have been
	 * resolved. Thus the second pass is expected to end quickly.
	 * We give a brief window at the top of the loop to allow
	 * any pending I/O to complete.
	 */
	for (waitfor = MNT_NOWAIT; ; ) {
		FREE_LOCK(&lk);
		ACQUIRE_LOCK(&lk);
		if (inodedep_lookup(fs, ino, 0, &inodedep) == 0)
			return (0);
		for (adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp;
		     adp = TAILQ_NEXT(adp, ad_next)) {
			if (adp->ad_state & DEPCOMPLETE)
				continue;
			bp = adp->ad_buf;
			if (getdirtybuf(&bp, waitfor) == 0) {
				if (waitfor == MNT_NOWAIT)
					continue;
				break;
			}
			FREE_LOCK(&lk);
			if (waitfor == MNT_NOWAIT) {
				bawrite(bp);
			} else if ((error = VOP_BWRITE(bp)) != 0) {
				ACQUIRE_LOCK(&lk);
				return (error);
			}
			ACQUIRE_LOCK(&lk);
			break;
		}
		if (adp != NULL)
			continue;
		for (adp = TAILQ_FIRST(&inodedep->id_newinoupdt); adp;
		     adp = TAILQ_NEXT(adp, ad_next)) {
			if (adp->ad_state & DEPCOMPLETE)
				continue;
			bp = adp->ad_buf;
			if (getdirtybuf(&bp, waitfor) == 0) {
				if (waitfor == MNT_NOWAIT)
					continue;
				break;
			}
			FREE_LOCK(&lk);
			if (waitfor == MNT_NOWAIT) {
				bawrite(bp);
			} else if ((error = VOP_BWRITE(bp)) != 0) {
				ACQUIRE_LOCK(&lk);
				return (error);
			}
			ACQUIRE_LOCK(&lk);
			break;
		}
		if (adp != NULL)
			continue;
		/*
		 * If pass2, we are done, otherwise do pass 2.
		 */
		if (waitfor == MNT_WAIT)
			break;
		waitfor = MNT_WAIT;
	}
	/*
	 * Try freeing inodedep in case all dependencies have been removed.
	 */
	if (inodedep_lookup(fs, ino, 0, &inodedep) != 0)
		(void) free_inodedep(inodedep);
	return (0);
}

/*
 * Eliminate a pagedep dependency by flushing out all its diradd dependencies.
 * Called with splbio blocked.
 */
static int
flush_pagedep_deps(pvp, mp, diraddhdp)
	struct vnode *pvp;
	struct mount *mp;
	struct diraddhd *diraddhdp;
{
	struct proc *p = CURPROC;	/* XXX */
	struct inodedep *inodedep;
	struct ufsmount *ump;
	struct diradd *dap;
#ifndef __FreeBSD__
	struct timeval tv;
#endif
	struct vnode *vp;
	int gotit, error = 0;
	struct buf *bp;
	ino_t inum;

	ump = VFSTOUFS(mp);
	while ((dap = LIST_FIRST(diraddhdp)) != NULL) {
		/*
		 * Flush ourselves if this directory entry
		 * has a MKDIR_PARENT dependency.
		 */
		if (dap->da_state & MKDIR_PARENT) {
			FREE_LOCK(&lk);
#ifdef __FreeBSD__
			error = UFS_UPDATE(pvp, 1);
#else
			tv = time;
			error = UFS_UPDATE(pvp, &tv, &tv, 1);
#endif
			if (error)
				break;
			ACQUIRE_LOCK(&lk);
			/*
			 * If that cleared dependencies, go on to next.
			 */
			if (dap != LIST_FIRST(diraddhdp))
				continue;
			if (dap->da_state & MKDIR_PARENT)
				panic("flush_pagedep_deps: MKDIR");
		}
		/*
		 * Flush the file on which the directory entry depends.
		 * If the inode has already been pushed out of the cache,
		 * then all the block dependencies will have been flushed
		 * leaving only inode dependencies (e.g., bitmaps). Thus,
		 * we do a ufs_ihashget to check for the vnode in the cache.
		 * If it is there, we do a full flush. If it is no longer
		 * there we need only dispose of any remaining bitmap
		 * dependencies and write the inode to disk.
		 */
		inum = dap->da_newinum;
		FREE_LOCK(&lk);
		if ((vp = ufs_ihashget(ump->um_dev, inum)) == NULL) {
			ACQUIRE_LOCK(&lk);
			if (inodedep_lookup(ump->um_fs, inum, 0, &inodedep) == 0
			    && dap == LIST_FIRST(diraddhdp))
				panic("flush_pagedep_deps: flush 1 failed");
			/*
			 * If the inode still has bitmap dependencies,
			 * push them to disk.
			 */
			if ((inodedep->id_state & DEPCOMPLETE) == 0) {
				gotit = getdirtybuf(&inodedep->id_buf,MNT_WAIT);
				FREE_LOCK(&lk);
				if (gotit &&
				    (error = VOP_BWRITE(inodedep->id_buf)) != 0)
					break;
				ACQUIRE_LOCK(&lk);
			}
			if (dap != LIST_FIRST(diraddhdp))
				continue;
			/*
			 * If the inode is still sitting in a buffer waiting
			 * to be written, push it to disk.
			 */
			FREE_LOCK(&lk);
			if ((error = bread(ump->um_devvp,
			    fsbtodb(ump->um_fs, ino_to_fsba(ump->um_fs, inum)),
			    (int)ump->um_fs->fs_bsize, NOCRED, &bp)) != 0)
				break;
			if ((error = VOP_BWRITE(bp)) != 0)
				break;
			ACQUIRE_LOCK(&lk);
			if (dap == LIST_FIRST(diraddhdp))
				panic("flush_pagedep_deps: flush 2 failed");
			continue;
		}
		if (vp->v_type == VDIR) {
			/*
			 * A newly allocated directory must have its "." and
			 * ".." entries written out before its name can be
			 * committed in its parent. We do not want or need
			 * the full semantics of a synchronous VOP_FSYNC as
			 * that may end up here again, once for each directory
			 * level in the filesystem. Instead, we push the blocks
			 * and wait for them to clear.
			 */
			if (error = VOP_FSYNC(vp, p->p_ucred, MNT_NOWAIT, p)) {
				vput(vp);
				break;
			}
			ACQUIRE_LOCK(&lk);
			while (vp->v_numoutput) {
				vp->v_flag |= VBWAIT;
				FREE_LOCK_INTERLOCKED(&lk);
				tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, 
					"sdflpd", 0);
				ACQUIRE_LOCK_INTERLOCKED(&lk);
			}
			FREE_LOCK(&lk);
		}
#ifdef __FreeBSD__
		error = UFS_UPDATE(vp, 1);
#else
		tv = time;
		error = UFS_UPDATE(vp, &tv, &tv, 1);
#endif
		vput(vp);
		if (error)
			break;
		/*
		 * If we have failed to get rid of all the dependencies
		 * then something is seriously wrong.
		 */
		if (dap == LIST_FIRST(diraddhdp))
			panic("flush_pagedep_deps: flush 3 failed");
		ACQUIRE_LOCK(&lk);
	}
	if (error)
		ACQUIRE_LOCK(&lk);
	return (error);
}

/*
 * Acquire exclusive access to a buffer.
 * Must be called with splbio blocked.
 * Return 1 if buffer was acquired.
 */
static int
getdirtybuf(bpp, waitfor)
	struct buf **bpp;
	int waitfor;
{
	struct buf *bp;

	for (;;) {
		if ((bp = *bpp) == NULL)
			return (0);
		if ((bp->b_flags & B_BUSY) == 0)
			break;
		if (waitfor != MNT_WAIT)
			return (0);
		bp->b_flags |= B_WANTED;
		FREE_LOCK_INTERLOCKED(&lk);
		tsleep((caddr_t)bp, PRIBIO + 1, "sdsdty", 0);
		ACQUIRE_LOCK_INTERLOCKED(&lk);
	}
	if ((bp->b_flags & B_DELWRI) == 0)
		return (0);
	bremfree(bp);
	bp->b_flags |= B_BUSY;
	return (1);
}

/*
 * Called whenever a buffer that is being invalidated or reallocated
 * contains dependencies. This should only happen if an I/O error has
 * occurred. The routine is called with the buffer locked.
 */ 
void
softdep_deallocate_dependencies(bp)
	struct buf *bp;
{
	struct worklist *wk;

	if ((bp->b_flags & B_ERROR) == 0)
		panic("softdep_deallocate_dependencies: dangling deps");
	softdep_error(bp->b_vp->v_mount->mnt_stat.f_mntonname, bp->b_error);
	ACQUIRE_LOCK(&lk);
	while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) {
		WORKLIST_REMOVE(wk);
		FREE_LOCK(&lk);
		switch (wk->wk_type) {
		/*
		 * XXX - should really clean up, but for now we will
		 * just leak memory and not worry about it. Also should
		 * mark the filesystem permanently dirty so that it will
		 * force fsck to be run (though this would best be done
		 * in the mainline code).
		 */
		case D_PAGEDEP:
		case D_INODEDEP:
		case D_BMSAFEMAP:
		case D_ALLOCDIRECT:
		case D_INDIRDEP:
		case D_ALLOCINDIR:
		case D_MKDIR:
#ifdef DEBUG
			printf("Lost type %s\n", TYPENAME(wk->wk_type));
#endif
			break;
		default:
			panic("%s: Unexpected type %s",
			    "softdep_deallocate_dependencies",
			    TYPENAME(wk->wk_type));
			/* NOTREACHED */
		}
		ACQUIRE_LOCK(&lk);
	}
	FREE_LOCK(&lk);
}

/*
 * Function to handle asynchronous write errors in the filesystem.
 */
void
softdep_error(func, error)
	char *func;
	int error;
{

	/* XXX should do something better! */
	log(LOG_ERR, "%s: got error %d while accessing filesystem\n",
	    func, error);
}

OpenPOWER on IntegriCloud