summaryrefslogtreecommitdiff
path: root/nscd/connections.c
blob: cc1ed72077640a8b45b8d70f3abc13814a0e8faf (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
/* Inner loops of cache daemon.
   Copyright (C) 1998-2017 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998.

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published
   by the Free Software Foundation; version 2 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, see <http://www.gnu.org/licenses/>.  */

#include <alloca.h>
#include <assert.h>
#include <atomic.h>
#include <error.h>
#include <errno.h>
#include <fcntl.h>
#include <grp.h>
#include <ifaddrs.h>
#include <libintl.h>
#include <pthread.h>
#include <pwd.h>
#include <resolv.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <stdint.h>
#include <arpa/inet.h>
#ifdef HAVE_NETLINK
# include <linux/netlink.h>
# include <linux/rtnetlink.h>
#endif
#ifdef HAVE_EPOLL
# include <sys/epoll.h>
#endif
#ifdef HAVE_INOTIFY
# include <sys/inotify.h>
#endif
#include <sys/mman.h>
#include <sys/param.h>
#include <sys/poll.h>
#ifdef HAVE_SENDFILE
# include <sys/sendfile.h>
#endif
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/un.h>

#include "nscd.h"
#include "dbg_log.h"
#include "selinux.h"
#include <resolv/resolv.h>

#include <kernel-features.h>
#include <libc-diag.h>


/* Support to run nscd as an unprivileged user */
const char *server_user;
static uid_t server_uid;
static gid_t server_gid;
const char *stat_user;
uid_t stat_uid;
static gid_t *server_groups;
#ifndef NGROUPS
# define NGROUPS 32
#endif
static int server_ngroups;

static pthread_attr_t attr;

static void begin_drop_privileges (void);
static void finish_drop_privileges (void);

/* Map request type to a string.  */
const char *const serv2str[LASTREQ] =
{
  [GETPWBYNAME] = "GETPWBYNAME",
  [GETPWBYUID] = "GETPWBYUID",
  [GETGRBYNAME] = "GETGRBYNAME",
  [GETGRBYGID] = "GETGRBYGID",
  [GETHOSTBYNAME] = "GETHOSTBYNAME",
  [GETHOSTBYNAMEv6] = "GETHOSTBYNAMEv6",
  [GETHOSTBYADDR] = "GETHOSTBYADDR",
  [GETHOSTBYADDRv6] = "GETHOSTBYADDRv6",
  [SHUTDOWN] = "SHUTDOWN",
  [GETSTAT] = "GETSTAT",
  [INVALIDATE] = "INVALIDATE",
  [GETFDPW] = "GETFDPW",
  [GETFDGR] = "GETFDGR",
  [GETFDHST] = "GETFDHST",
  [GETAI] = "GETAI",
  [INITGROUPS] = "INITGROUPS",
  [GETSERVBYNAME] = "GETSERVBYNAME",
  [GETSERVBYPORT] = "GETSERVBYPORT",
  [GETFDSERV] = "GETFDSERV",
  [GETNETGRENT] = "GETNETGRENT",
  [INNETGR] = "INNETGR",
  [GETFDNETGR] = "GETFDNETGR"
};

/* The control data structures for the services.  */
struct database_dyn dbs[lastdb] =
{
  [pwddb] = {
    .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
    .prune_lock = PTHREAD_MUTEX_INITIALIZER,
    .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
    .enabled = 0,
    .check_file = 1,
    .persistent = 0,
    .propagate = 1,
    .shared = 0,
    .max_db_size = DEFAULT_MAX_DB_SIZE,
    .suggested_module = DEFAULT_SUGGESTED_MODULE,
    .db_filename = _PATH_NSCD_PASSWD_DB,
    .disabled_iov = &pwd_iov_disabled,
    .postimeout = 3600,
    .negtimeout = 20,
    .wr_fd = -1,
    .ro_fd = -1,
    .mmap_used = false
  },
  [grpdb] = {
    .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
    .prune_lock = PTHREAD_MUTEX_INITIALIZER,
    .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
    .enabled = 0,
    .check_file = 1,
    .persistent = 0,
    .propagate = 1,
    .shared = 0,
    .max_db_size = DEFAULT_MAX_DB_SIZE,
    .suggested_module = DEFAULT_SUGGESTED_MODULE,
    .db_filename = _PATH_NSCD_GROUP_DB,
    .disabled_iov = &grp_iov_disabled,
    .postimeout = 3600,
    .negtimeout = 60,
    .wr_fd = -1,
    .ro_fd = -1,
    .mmap_used = false
  },
  [hstdb] = {
    .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
    .prune_lock = PTHREAD_MUTEX_INITIALIZER,
    .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
    .enabled = 0,
    .check_file = 1,
    .persistent = 0,
    .propagate = 0,		/* Not used.  */
    .shared = 0,
    .max_db_size = DEFAULT_MAX_DB_SIZE,
    .suggested_module = DEFAULT_SUGGESTED_MODULE,
    .db_filename = _PATH_NSCD_HOSTS_DB,
    .disabled_iov = &hst_iov_disabled,
    .postimeout = 3600,
    .negtimeout = 20,
    .wr_fd = -1,
    .ro_fd = -1,
    .mmap_used = false
  },
  [servdb] = {
    .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
    .prune_lock = PTHREAD_MUTEX_INITIALIZER,
    .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
    .enabled = 0,
    .check_file = 1,
    .persistent = 0,
    .propagate = 0,		/* Not used.  */
    .shared = 0,
    .max_db_size = DEFAULT_MAX_DB_SIZE,
    .suggested_module = DEFAULT_SUGGESTED_MODULE,
    .db_filename = _PATH_NSCD_SERVICES_DB,
    .disabled_iov = &serv_iov_disabled,
    .postimeout = 28800,
    .negtimeout = 20,
    .wr_fd = -1,
    .ro_fd = -1,
    .mmap_used = false
  },
  [netgrdb] = {
    .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
    .prune_lock = PTHREAD_MUTEX_INITIALIZER,
    .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
    .enabled = 0,
    .check_file = 1,
    .persistent = 0,
    .propagate = 0,		/* Not used.  */
    .shared = 0,
    .max_db_size = DEFAULT_MAX_DB_SIZE,
    .suggested_module = DEFAULT_SUGGESTED_MODULE,
    .db_filename = _PATH_NSCD_NETGROUP_DB,
    .disabled_iov = &netgroup_iov_disabled,
    .postimeout = 28800,
    .negtimeout = 20,
    .wr_fd = -1,
    .ro_fd = -1,
    .mmap_used = false
  }
};


/* Mapping of request type to database.  */
static struct
{
  bool data_request;
  struct database_dyn *db;
} const reqinfo[LASTREQ] =
{
  [GETPWBYNAME] = { true, &dbs[pwddb] },
  [GETPWBYUID] = { true, &dbs[pwddb] },
  [GETGRBYNAME] = { true, &dbs[grpdb] },
  [GETGRBYGID] = { true, &dbs[grpdb] },
  [GETHOSTBYNAME] = { true, &dbs[hstdb] },
  [GETHOSTBYNAMEv6] = { true, &dbs[hstdb] },
  [GETHOSTBYADDR] = { true, &dbs[hstdb] },
  [GETHOSTBYADDRv6] = { true, &dbs[hstdb] },
  [SHUTDOWN] = { false, NULL },
  [GETSTAT] = { false, NULL },
  [SHUTDOWN] = { false, NULL },
  [GETFDPW] = { false, &dbs[pwddb] },
  [GETFDGR] = { false, &dbs[grpdb] },
  [GETFDHST] = { false, &dbs[hstdb] },
  [GETAI] = { true, &dbs[hstdb] },
  [INITGROUPS] = { true, &dbs[grpdb] },
  [GETSERVBYNAME] = { true, &dbs[servdb] },
  [GETSERVBYPORT] = { true, &dbs[servdb] },
  [GETFDSERV] = { false, &dbs[servdb] },
  [GETNETGRENT] = { true, &dbs[netgrdb] },
  [INNETGR] = { true, &dbs[netgrdb] },
  [GETFDNETGR] = { false, &dbs[netgrdb] }
};


/* Initial number of threads to use.  */
int nthreads = -1;
/* Maximum number of threads to use.  */
int max_nthreads = 32;

/* Socket for incoming connections.  */
static int sock;

#ifdef HAVE_INOTIFY
/* Inotify descriptor.  */
int inotify_fd = -1;
#endif

#ifdef HAVE_NETLINK
/* Descriptor for netlink status updates.  */
static int nl_status_fd = -1;
#endif

/* Number of times clients had to wait.  */
unsigned long int client_queued;


ssize_t
writeall (int fd, const void *buf, size_t len)
{
  size_t n = len;
  ssize_t ret;
  do
    {
      ret = TEMP_FAILURE_RETRY (send (fd, buf, n, MSG_NOSIGNAL));
      if (ret <= 0)
	break;
      buf = (const char *) buf + ret;
      n -= ret;
    }
  while (n > 0);
  return ret < 0 ? ret : len - n;
}


#ifdef HAVE_SENDFILE
ssize_t
sendfileall (int tofd, int fromfd, off_t off, size_t len)
{
  ssize_t n = len;
  ssize_t ret;

  do
    {
      ret = TEMP_FAILURE_RETRY (sendfile (tofd, fromfd, &off, n));
      if (ret <= 0)
	break;
      n -= ret;
    }
  while (n > 0);
  return ret < 0 ? ret : len - n;
}
#endif


enum usekey
  {
    use_not = 0,
    /* The following three are not really used, they are symbolic constants.  */
    use_first = 16,
    use_begin = 32,
    use_end = 64,

    use_he = 1,
    use_he_begin = use_he | use_begin,
    use_he_end = use_he | use_end,
    use_data = 3,
    use_data_begin = use_data | use_begin,
    use_data_end = use_data | use_end,
    use_data_first = use_data_begin | use_first
  };


static int
check_use (const char *data, nscd_ssize_t first_free, uint8_t *usemap,
	   enum usekey use, ref_t start, size_t len)
{
  assert (len >= 2);

  if (start > first_free || start + len > first_free
      || (start & BLOCK_ALIGN_M1))
    return 0;

  if (usemap[start] == use_not)
    {
      /* Add the start marker.  */
      usemap[start] = use | use_begin;
      use &= ~use_first;

      while (--len > 0)
	if (usemap[++start] != use_not)
	  return 0;
	else
	  usemap[start] = use;

      /* Add the end marker.  */
      usemap[start] = use | use_end;
    }
  else if ((usemap[start] & ~use_first) == ((use | use_begin) & ~use_first))
    {
      /* Hash entries can't be shared.  */
      if (use == use_he)
	return 0;

      usemap[start] |= (use & use_first);
      use &= ~use_first;

      while (--len > 1)
	if (usemap[++start] != use)
	  return 0;

      if (usemap[++start] != (use | use_end))
	return 0;
    }
  else
    /* Points to a wrong object or somewhere in the middle.  */
    return 0;

  return 1;
}


/* Verify data in persistent database.  */
static int
verify_persistent_db (void *mem, struct database_pers_head *readhead, int dbnr)
{
  assert (dbnr == pwddb || dbnr == grpdb || dbnr == hstdb || dbnr == servdb
	  || dbnr == netgrdb);

  time_t now = time (NULL);

  struct database_pers_head *head = mem;
  struct database_pers_head head_copy = *head;

  /* Check that the header that was read matches the head in the database.  */
  if (memcmp (head, readhead, sizeof (*head)) != 0)
    return 0;

  /* First some easy tests: make sure the database header is sane.  */
  if (head->version != DB_VERSION
      || head->header_size != sizeof (*head)
      /* We allow a timestamp to be one hour ahead of the current time.
	 This should cover daylight saving time changes.  */
      || head->timestamp > now + 60 * 60 + 60
      || (head->gc_cycle & 1)
      || head->module == 0
      || (size_t) head->module > INT32_MAX / sizeof (ref_t)
      || (size_t) head->data_size > INT32_MAX - head->module * sizeof (ref_t)
      || head->first_free < 0
      || head->first_free > head->data_size
      || (head->first_free & BLOCK_ALIGN_M1) != 0
      || head->maxnentries < 0
      || head->maxnsearched < 0)
    return 0;

  uint8_t *usemap = calloc (head->first_free, 1);
  if (usemap == NULL)
    return 0;

  const char *data = (char *) &head->array[roundup (head->module,
						    ALIGN / sizeof (ref_t))];

  nscd_ssize_t he_cnt = 0;
  for (nscd_ssize_t cnt = 0; cnt < head->module; ++cnt)
    {
      ref_t trail = head->array[cnt];
      ref_t work = trail;
      int tick = 0;

      while (work != ENDREF)
	{
	  if (! check_use (data, head->first_free, usemap, use_he, work,
			   sizeof (struct hashentry)))
	    goto fail;

	  /* Now we know we can dereference the record.  */
	  struct hashentry *here = (struct hashentry *) (data + work);

	  ++he_cnt;

	  /* Make sure the record is for this type of service.  */
	  if (here->type >= LASTREQ
	      || reqinfo[here->type].db != &dbs[dbnr])
	    goto fail;

	  /* Validate boolean field value.  */
	  if (here->first != false && here->first != true)
	    goto fail;

	  if (here->len < 0)
	    goto fail;

	  /* Now the data.  */
	  if (here->packet < 0
	      || here->packet > head->first_free
	      || here->packet + sizeof (struct datahead) > head->first_free)
	    goto fail;

	  struct datahead *dh = (struct datahead *) (data + here->packet);

	  if (! check_use (data, head->first_free, usemap,
			   use_data | (here->first ? use_first : 0),
			   here->packet, dh->allocsize))
	    goto fail;

	  if (dh->allocsize < sizeof (struct datahead)
	      || dh->recsize > dh->allocsize
	      || (dh->notfound != false && dh->notfound != true)
	      || (dh->usable != false && dh->usable != true))
	    goto fail;

	  if (here->key < here->packet + sizeof (struct datahead)
	      || here->key > here->packet + dh->allocsize
	      || here->key + here->len > here->packet + dh->allocsize)
	    goto fail;

	  work = here->next;

	  if (work == trail)
	    /* A circular list, this must not happen.  */
	    goto fail;
	  if (tick)
	    trail = ((struct hashentry *) (data + trail))->next;
	  tick = 1 - tick;
	}
    }

  if (he_cnt != head->nentries)
    goto fail;

  /* See if all data and keys had at least one reference from
     he->first == true hashentry.  */
  for (ref_t idx = 0; idx < head->first_free; ++idx)
    {
      if (usemap[idx] == use_data_begin)
	goto fail;
    }

  /* Finally, make sure the database hasn't changed since the first test.  */
  if (memcmp (mem, &head_copy, sizeof (*head)) != 0)
    goto fail;

  free (usemap);
  return 1;

fail:
  free (usemap);
  return 0;
}


/* Initialize database information structures.  */
void
nscd_init (void)
{
  /* Look up unprivileged uid/gid/groups before we start listening on the
     socket  */
  if (server_user != NULL)
    begin_drop_privileges ();

  if (nthreads == -1)
    /* No configuration for this value, assume a default.  */
    nthreads = 4;

  for (size_t cnt = 0; cnt < lastdb; ++cnt)
    if (dbs[cnt].enabled)
      {
	pthread_rwlock_init (&dbs[cnt].lock, NULL);
	pthread_mutex_init (&dbs[cnt].memlock, NULL);

	if (dbs[cnt].persistent)
	  {
	    /* Try to open the appropriate file on disk.  */
	    int fd = open (dbs[cnt].db_filename, O_RDWR | O_CLOEXEC);
	    if (fd != -1)
	      {
		char *msg = NULL;
		struct stat64 st;
		void *mem;
		size_t total;
		struct database_pers_head head;
		ssize_t n = TEMP_FAILURE_RETRY (read (fd, &head,
						      sizeof (head)));
		if (n != sizeof (head) || fstat64 (fd, &st) != 0)
		  {
		  fail_db_errno:
		    /* The code is single-threaded at this point so
		       using strerror is just fine.  */
		    msg = strerror (errno);
		  fail_db:
		    dbg_log (_("invalid persistent database file \"%s\": %s"),
			     dbs[cnt].db_filename, msg);
		    unlink (dbs[cnt].db_filename);
		  }
		else if (head.module == 0 && head.data_size == 0)
		  {
		    /* The file has been created, but the head has not
		       been initialized yet.  */
		    msg = _("uninitialized header");
		    goto fail_db;
		  }
		else if (head.header_size != (int) sizeof (head))
		  {
		    msg = _("header size does not match");
		    goto fail_db;
		  }
		else if ((total = (sizeof (head)
				   + roundup (head.module * sizeof (ref_t),
					      ALIGN)
				   + head.data_size))
			 > st.st_size
			 || total < sizeof (head))
		  {
		    msg = _("file size does not match");
		    goto fail_db;
		  }
		/* Note we map with the maximum size allowed for the
		   database.  This is likely much larger than the
		   actual file size.  This is OK on most OSes since
		   extensions of the underlying file will
		   automatically translate more pages available for
		   memory access.  */
		else if ((mem = mmap (NULL, dbs[cnt].max_db_size,
				      PROT_READ | PROT_WRITE,
				      MAP_SHARED, fd, 0))
			 == MAP_FAILED)
		  goto fail_db_errno;
		else if (!verify_persistent_db (mem, &head, cnt))
		  {
		    munmap (mem, total);
		    msg = _("verification failed");
		    goto fail_db;
		  }
		else
		  {
		    /* Success.  We have the database.  */
		    dbs[cnt].head = mem;
		    dbs[cnt].memsize = total;
		    dbs[cnt].data = (char *)
		      &dbs[cnt].head->array[roundup (dbs[cnt].head->module,
						     ALIGN / sizeof (ref_t))];
		    dbs[cnt].mmap_used = true;

		    if (dbs[cnt].suggested_module > head.module)
		      dbg_log (_("suggested size of table for database %s larger than the persistent database's table"),
			       dbnames[cnt]);

		    dbs[cnt].wr_fd = fd;
		    fd = -1;
		    /* We also need a read-only descriptor.  */
		    if (dbs[cnt].shared)
		      {
			dbs[cnt].ro_fd = open (dbs[cnt].db_filename,
					       O_RDONLY | O_CLOEXEC);
			if (dbs[cnt].ro_fd == -1)
			  dbg_log (_("\
cannot create read-only descriptor for \"%s\"; no mmap"),
				   dbs[cnt].db_filename);
		      }

		    // XXX Shall we test whether the descriptors actually
		    // XXX point to the same file?
		  }

		/* Close the file descriptors in case something went
		   wrong in which case the variable have not been
		   assigned -1.  */
		if (fd != -1)
		  close (fd);
	      }
	    else if (errno == EACCES)
	      do_exit (EXIT_FAILURE, 0, _("cannot access '%s'"),
		       dbs[cnt].db_filename);
	  }

	if (dbs[cnt].head == NULL)
	  {
	    /* No database loaded.  Allocate the data structure,
	       possibly on disk.  */
	    struct database_pers_head head;
	    size_t total = (sizeof (head)
			    + roundup (dbs[cnt].suggested_module
				       * sizeof (ref_t), ALIGN)
			    + (dbs[cnt].suggested_module
			       * DEFAULT_DATASIZE_PER_BUCKET));

	    /* Try to create the database.  If we do not need a
	       persistent database create a temporary file.  */
	    int fd;
	    int ro_fd = -1;
	    if (dbs[cnt].persistent)
	      {
		fd = open (dbs[cnt].db_filename,
			   O_RDWR | O_CREAT | O_EXCL | O_TRUNC | O_CLOEXEC,
			   S_IRUSR | S_IWUSR);
		if (fd != -1 && dbs[cnt].shared)
		  ro_fd = open (dbs[cnt].db_filename,
				O_RDONLY | O_CLOEXEC);
	      }
	    else
	      {
		char fname[] = _PATH_NSCD_XYZ_DB_TMP;
		fd = mkostemp (fname, O_CLOEXEC);

		/* We do not need the file name anymore after we
		   opened another file descriptor in read-only mode.  */
		if (fd != -1)
		  {
		    if (dbs[cnt].shared)
		      ro_fd = open (fname, O_RDONLY | O_CLOEXEC);

		    unlink (fname);
		  }
	      }

	    if (fd == -1)
	      {
		if (errno == EEXIST)
		  {
		    dbg_log (_("database for %s corrupted or simultaneously used; remove %s manually if necessary and restart"),
			     dbnames[cnt], dbs[cnt].db_filename);
		    do_exit (1, 0, NULL);
		  }

		if  (dbs[cnt].persistent)
		  dbg_log (_("cannot create %s; no persistent database used"),
			   dbs[cnt].db_filename);
		else
		  dbg_log (_("cannot create %s; no sharing possible"),
			   dbs[cnt].db_filename);

		dbs[cnt].persistent = 0;
		// XXX remember: no mmap
	      }
	    else
	      {
		/* Tell the user if we could not create the read-only
		   descriptor.  */
		if (ro_fd == -1 && dbs[cnt].shared)
		  dbg_log (_("\
cannot create read-only descriptor for \"%s\"; no mmap"),
			   dbs[cnt].db_filename);

		/* Before we create the header, initialize the hash
		   table.  That way if we get interrupted while writing
		   the header we can recognize a partially initialized
		   database.  */
		size_t ps = sysconf (_SC_PAGESIZE);
		char tmpbuf[ps];
		assert (~ENDREF == 0);
		memset (tmpbuf, '\xff', ps);

		size_t remaining = dbs[cnt].suggested_module * sizeof (ref_t);
		off_t offset = sizeof (head);

		size_t towrite;
		if (offset % ps != 0)
		  {
		    towrite = MIN (remaining, ps - (offset % ps));
		    if (pwrite (fd, tmpbuf, towrite, offset) != towrite)
		      goto write_fail;
		    offset += towrite;
		    remaining -= towrite;
		  }

		while (remaining > ps)
		  {
		    if (pwrite (fd, tmpbuf, ps, offset) == -1)
		      goto write_fail;
		    offset += ps;
		    remaining -= ps;
		  }

		if (remaining > 0
		    && pwrite (fd, tmpbuf, remaining, offset) != remaining)
		  goto write_fail;

		/* Create the header of the file.  */
		struct database_pers_head head =
		  {
		    .version = DB_VERSION,
		    .header_size = sizeof (head),
		    .module = dbs[cnt].suggested_module,
		    .data_size = (dbs[cnt].suggested_module
				  * DEFAULT_DATASIZE_PER_BUCKET),
		    .first_free = 0
		  };
		void *mem;

		if ((TEMP_FAILURE_RETRY (write (fd, &head, sizeof (head)))
		     != sizeof (head))
		    || (TEMP_FAILURE_RETRY_VAL (posix_fallocate (fd, 0, total))
			!= 0)
		    || (mem = mmap (NULL, dbs[cnt].max_db_size,
				    PROT_READ | PROT_WRITE,
				    MAP_SHARED, fd, 0)) == MAP_FAILED)
		  {
		  write_fail:
		    unlink (dbs[cnt].db_filename);
		    dbg_log (_("cannot write to database file %s: %s"),
			     dbs[cnt].db_filename, strerror (errno));
		    dbs[cnt].persistent = 0;
		  }
		else
		  {
		    /* Success.  */
		    dbs[cnt].head = mem;
		    dbs[cnt].data = (char *)
		      &dbs[cnt].head->array[roundup (dbs[cnt].head->module,
						     ALIGN / sizeof (ref_t))];
		    dbs[cnt].memsize = total;
		    dbs[cnt].mmap_used = true;

		    /* Remember the descriptors.  */
		    dbs[cnt].wr_fd = fd;
		    dbs[cnt].ro_fd = ro_fd;
		    fd = -1;
		    ro_fd = -1;
		  }

		if (fd != -1)
		  close (fd);
		if (ro_fd != -1)
		  close (ro_fd);
	      }
	  }

	if (dbs[cnt].head == NULL)
	  {
	    /* We do not use the persistent database.  Just
	       create an in-memory data structure.  */
	    assert (! dbs[cnt].persistent);

	    dbs[cnt].head = xmalloc (sizeof (struct database_pers_head)
				     + (dbs[cnt].suggested_module
					* sizeof (ref_t)));
	    memset (dbs[cnt].head, '\0', sizeof (struct database_pers_head));
	    assert (~ENDREF == 0);
	    memset (dbs[cnt].head->array, '\xff',
		    dbs[cnt].suggested_module * sizeof (ref_t));
	    dbs[cnt].head->module = dbs[cnt].suggested_module;
	    dbs[cnt].head->data_size = (DEFAULT_DATASIZE_PER_BUCKET
					* dbs[cnt].head->module);
	    dbs[cnt].data = xmalloc (dbs[cnt].head->data_size);
	    dbs[cnt].head->first_free = 0;

	    dbs[cnt].shared = 0;
	    assert (dbs[cnt].ro_fd == -1);
	  }
      }

  /* Create the socket.  */
  sock = socket (AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, 0);
  if (sock < 0)
    {
      dbg_log (_("cannot open socket: %s"), strerror (errno));
      do_exit (errno == EACCES ? 4 : 1, 0, NULL);
    }
  /* Bind a name to the socket.  */
  struct sockaddr_un sock_addr;
  sock_addr.sun_family = AF_UNIX;
  strcpy (sock_addr.sun_path, _PATH_NSCDSOCKET);
  if (bind (sock, (struct sockaddr *) &sock_addr, sizeof (sock_addr)) < 0)
    {
      dbg_log ("%s: %s", _PATH_NSCDSOCKET, strerror (errno));
      do_exit (errno == EACCES ? 4 : 1, 0, NULL);
    }

  /* Set permissions for the socket.  */
  chmod (_PATH_NSCDSOCKET, DEFFILEMODE);

  /* Set the socket up to accept connections.  */
  if (listen (sock, SOMAXCONN) < 0)
    {
      dbg_log (_("cannot enable socket to accept connections: %s"),
	       strerror (errno));
      do_exit (1, 0, NULL);
    }

#ifdef HAVE_NETLINK
  if (dbs[hstdb].enabled)
    {
      /* Try to open netlink socket to monitor network setting changes.  */
      nl_status_fd = socket (AF_NETLINK,
			     SOCK_RAW | SOCK_CLOEXEC | SOCK_NONBLOCK,
			     NETLINK_ROUTE);
      if (nl_status_fd != -1)
	{
	  struct sockaddr_nl snl;
	  memset (&snl, '\0', sizeof (snl));
	  snl.nl_family = AF_NETLINK;
	  /* XXX Is this the best set to use?  */
	  snl.nl_groups = (RTMGRP_IPV4_IFADDR | RTMGRP_TC | RTMGRP_IPV4_MROUTE
			   | RTMGRP_IPV4_ROUTE | RTMGRP_IPV4_RULE
			   | RTMGRP_IPV6_IFADDR | RTMGRP_IPV6_MROUTE
			   | RTMGRP_IPV6_ROUTE | RTMGRP_IPV6_IFINFO
			   | RTMGRP_IPV6_PREFIX);

	  if (bind (nl_status_fd, (struct sockaddr *) &snl, sizeof (snl)) != 0)
	    {
	      close (nl_status_fd);
	      nl_status_fd = -1;
	    }
	  else
	    {
	      /* Start the timestamp process.  */
	      dbs[hstdb].head->extra_data[NSCD_HST_IDX_CONF_TIMESTAMP]
		= __bump_nl_timestamp ();
	    }
	}
    }
#endif

  /* Change to unprivileged uid/gid/groups if specified in config file */
  if (server_user != NULL)
    finish_drop_privileges ();
}

#ifdef HAVE_INOTIFY
#define TRACED_FILE_MASK (IN_DELETE_SELF | IN_CLOSE_WRITE | IN_MOVE_SELF)
#define TRACED_DIR_MASK (IN_DELETE_SELF | IN_CREATE | IN_MOVED_TO | IN_MOVE_SELF)
void
install_watches (struct traced_file *finfo)
{
  /* Use inotify support if we have it.  */
  if (finfo->inotify_descr[TRACED_FILE] < 0)
    finfo->inotify_descr[TRACED_FILE] = inotify_add_watch (inotify_fd,
							   finfo->fname,
							   TRACED_FILE_MASK);
  if (finfo->inotify_descr[TRACED_FILE] < 0)
    {
      dbg_log (_("disabled inotify-based monitoring for file `%s': %s"),
		 finfo->fname, strerror (errno));
      return;
    }
  dbg_log (_("monitoring file `%s` (%d)"),
	   finfo->fname, finfo->inotify_descr[TRACED_FILE]);
  /* Additionally listen for events in the file's parent directory.
     We do this because the file to be watched might be
     deleted and then added back again.  When it is added back again
     we must re-add the watch.  We must also cover IN_MOVED_TO to
     detect a file being moved into the directory.  */
  if (finfo->inotify_descr[TRACED_DIR] < 0)
    finfo->inotify_descr[TRACED_DIR] = inotify_add_watch (inotify_fd,
							  finfo->dname,
							  TRACED_DIR_MASK);
  if (finfo->inotify_descr[TRACED_DIR] < 0)
    {
      dbg_log (_("disabled inotify-based monitoring for directory `%s': %s"),
		 finfo->fname, strerror (errno));
      return;
    }
  dbg_log (_("monitoring directory `%s` (%d)"),
	   finfo->dname, finfo->inotify_descr[TRACED_DIR]);
}
#endif

/* Register the file in FINFO as a traced file for the database DBS[DBIX].

   We support registering multiple files per database. Each call to
   register_traced_file adds to the list of registered files.

   When we prune the database, either through timeout or a request to
   invalidate, we will check to see if any of the registered files has changed.
   When we accept new connections to handle a cache request we will also
   check to see if any of the registered files has changed.

   If we have inotify support then we install an inotify fd to notify us of
   file deletion or modification, both of which will require we invalidate
   the cache for the database.  Without inotify support we stat the file and
   store st_mtime to determine if the file has been modified.  */
void
register_traced_file (size_t dbidx, struct traced_file *finfo)
{
  /* If the database is disabled or file checking is disabled
     then ignore the registration.  */
  if (! dbs[dbidx].enabled || ! dbs[dbidx].check_file)
    return;

  if (__glibc_unlikely (debug_level > 0))
    dbg_log (_("monitoring file %s for database %s"),
	     finfo->fname, dbnames[dbidx]);

#ifdef HAVE_INOTIFY
  install_watches (finfo);
#endif
  struct stat64 st;
  if (stat64 (finfo->fname, &st) < 0)
    {
      /* We cannot stat() the file. Set mtime to zero and try again later.  */
      dbg_log (_("stat failed for file `%s'; will try again later: %s"),
	       finfo->fname, strerror (errno));
      finfo->mtime = 0;
    }
  else
    finfo->mtime = st.st_mtime;

  /* Queue up the file name.  */
  finfo->next = dbs[dbidx].traced_files;
  dbs[dbidx].traced_files = finfo;
}


/* Close the connections.  */
void
close_sockets (void)
{
  close (sock);
}


static void
invalidate_cache (char *key, int fd)
{
  dbtype number;
  int32_t resp;

  for (number = pwddb; number < lastdb; ++number)
    if (strcmp (key, dbnames[number]) == 0)
      {
	struct traced_file *runp = dbs[number].traced_files;
	while (runp != NULL)
	  {
	    /* Make sure we reload from file when checking mtime.  */
	    runp->mtime = 0;
#ifdef HAVE_INOTIFY
	    /* During an invalidation we try to reload the traced
	       file watches.  This allows the user to re-sync if
	       inotify events were lost.  Similar to what we do during
	       pruning.  */
	    install_watches (runp);
#endif
	    if (runp->call_res_init)
	      {
		res_init ();
		break;
	      }
	    runp = runp->next;
	  }
	break;
      }

  if (number == lastdb)
    {
      resp = EINVAL;
      writeall (fd, &resp, sizeof (resp));
      return;
    }

  if (dbs[number].enabled)
    {
      pthread_mutex_lock (&dbs[number].prune_run_lock);
      prune_cache (&dbs[number], LONG_MAX, fd);
      pthread_mutex_unlock (&dbs[number].prune_run_lock);
    }
  else
    {
      resp = 0;
      writeall (fd, &resp, sizeof (resp));
    }
}


#ifdef SCM_RIGHTS
static void
send_ro_fd (struct database_dyn *db, char *key, int fd)
{
  /* If we do not have an read-only file descriptor do nothing.  */
  if (db->ro_fd == -1)
    return;

  /* We need to send some data along with the descriptor.  */
  uint64_t mapsize = (db->head->data_size
		      + roundup (db->head->module * sizeof (ref_t), ALIGN)
		      + sizeof (struct database_pers_head));
  struct iovec iov[2];
  iov[0].iov_base = key;
  iov[0].iov_len = strlen (key) + 1;
  iov[1].iov_base = &mapsize;
  iov[1].iov_len = sizeof (mapsize);

  /* Prepare the control message to transfer the descriptor.  */
  union
  {
    struct cmsghdr hdr;
    char bytes[CMSG_SPACE (sizeof (int))];
  } buf;
  struct msghdr msg = { .msg_iov = iov, .msg_iovlen = 2,
			.msg_control = buf.bytes,
			.msg_controllen = sizeof (buf) };
  struct cmsghdr *cmsg = CMSG_FIRSTHDR (&msg);

  cmsg->cmsg_level = SOL_SOCKET;
  cmsg->cmsg_type = SCM_RIGHTS;
  cmsg->cmsg_len = CMSG_LEN (sizeof (int));

  int *ip = (int *) CMSG_DATA (cmsg);
  *ip = db->ro_fd;

  msg.msg_controllen = cmsg->cmsg_len;

  /* Send the control message.  We repeat when we are interrupted but
     everything else is ignored.  */
#ifndef MSG_NOSIGNAL
# define MSG_NOSIGNAL 0
#endif
  (void) TEMP_FAILURE_RETRY (sendmsg (fd, &msg, MSG_NOSIGNAL));

  if (__glibc_unlikely (debug_level > 0))
    dbg_log (_("provide access to FD %d, for %s"), db->ro_fd, key);
}
#endif	/* SCM_RIGHTS */


/* Handle new request.  */
static void
handle_request (int fd, request_header *req, void *key, uid_t uid, pid_t pid)
{
  if (__builtin_expect (req->version, NSCD_VERSION) != NSCD_VERSION)
    {
      if (debug_level > 0)
	dbg_log (_("\
cannot handle old request version %d; current version is %d"),
		 req->version, NSCD_VERSION);
      return;
    }

  /* Perform the SELinux check before we go on to the standard checks.  */
  if (selinux_enabled && nscd_request_avc_has_perm (fd, req->type) != 0)
    {
      if (debug_level > 0)
	{
#ifdef SO_PEERCRED
# ifdef PATH_MAX
	  char buf[PATH_MAX];
# else
	  char buf[4096];
# endif

	  snprintf (buf, sizeof (buf), "/proc/%ld/exe", (long int) pid);
	  ssize_t n = readlink (buf, buf, sizeof (buf) - 1);

	  if (n <= 0)
	    dbg_log (_("\
request from %ld not handled due to missing permission"), (long int) pid);
	  else
	    {
	      buf[n] = '\0';
	      dbg_log (_("\
request from '%s' [%ld] not handled due to missing permission"),
		       buf, (long int) pid);
	    }
#else
	  dbg_log (_("request not handled due to missing permission"));
#endif
	}
      return;
    }

  struct database_dyn *db = reqinfo[req->type].db;

  /* See whether we can service the request from the cache.  */
  if (__builtin_expect (reqinfo[req->type].data_request, true))
    {
      if (__builtin_expect (debug_level, 0) > 0)
	{
	  if (req->type == GETHOSTBYADDR || req->type == GETHOSTBYADDRv6)
	    {
	      char buf[INET6_ADDRSTRLEN];

	      dbg_log ("\t%s (%s)", serv2str[req->type],
		       inet_ntop (req->type == GETHOSTBYADDR
				  ? AF_INET : AF_INET6,
				  key, buf, sizeof (buf)));
	    }
	  else
	    dbg_log ("\t%s (%s)", serv2str[req->type], (char *) key);
	}

      /* Is this service enabled?  */
      if (__glibc_unlikely (!db->enabled))
	{
	  /* No, sent the prepared record.  */
	  if (TEMP_FAILURE_RETRY (send (fd, db->disabled_iov->iov_base,
					db->disabled_iov->iov_len,
					MSG_NOSIGNAL))
	      != (ssize_t) db->disabled_iov->iov_len
	      && __builtin_expect (debug_level, 0) > 0)
	    {
	      /* We have problems sending the result.  */
	      char buf[256];
	      dbg_log (_("cannot write result: %s"),
		       strerror_r (errno, buf, sizeof (buf)));
	    }

	  return;
	}

      /* Be sure we can read the data.  */
      if (__glibc_unlikely (pthread_rwlock_tryrdlock (&db->lock) != 0))
	{
	  ++db->head->rdlockdelayed;
	  pthread_rwlock_rdlock (&db->lock);
	}

      /* See whether we can handle it from the cache.  */
      struct datahead *cached;
      cached = (struct datahead *) cache_search (req->type, key, req->key_len,
						 db, uid);
      if (cached != NULL)
	{
	  /* Hurray it's in the cache.  */
	  ssize_t nwritten;

#ifdef HAVE_SENDFILE
	  if (__glibc_likely (db->mmap_used))
	    {
	      assert (db->wr_fd != -1);
	      assert ((char *) cached->data > (char *) db->data);
	      assert ((char *) cached->data - (char *) db->head
		      + cached->recsize
		      <= (sizeof (struct database_pers_head)
			  + db->head->module * sizeof (ref_t)
			  + db->head->data_size));
	      nwritten = sendfileall (fd, db->wr_fd,
				      (char *) cached->data
				      - (char *) db->head, cached->recsize);
# ifndef __ASSUME_SENDFILE
	      if (nwritten == -1 && errno == ENOSYS)
		goto use_write;
# endif
	    }
	  else
# ifndef __ASSUME_SENDFILE
	  use_write:
# endif
#endif
	    nwritten = writeall (fd, cached->data, cached->recsize);

	  if (nwritten != cached->recsize
	      && __builtin_expect (debug_level, 0) > 0)
	    {
	      /* We have problems sending the result.  */
	      char buf[256];
	      dbg_log (_("cannot write result: %s"),
		       strerror_r (errno, buf, sizeof (buf)));
	    }

	  pthread_rwlock_unlock (&db->lock);

	  return;
	}

      pthread_rwlock_unlock (&db->lock);
    }
  else if (__builtin_expect (debug_level, 0) > 0)
    {
      if (req->type == INVALIDATE)
	dbg_log ("\t%s (%s)", serv2str[req->type], (char *) key);
      else
	dbg_log ("\t%s", serv2str[req->type]);
    }

  /* Handle the request.  */
  switch (req->type)
    {
    case GETPWBYNAME:
      addpwbyname (db, fd, req, key, uid);
      break;

    case GETPWBYUID:
      addpwbyuid (db, fd, req, key, uid);
      break;

    case GETGRBYNAME:
      addgrbyname (db, fd, req, key, uid);
      break;

    case GETGRBYGID:
      addgrbygid (db, fd, req, key, uid);
      break;

    case GETHOSTBYNAME:
      addhstbyname (db, fd, req, key, uid);
      break;

    case GETHOSTBYNAMEv6:
      addhstbynamev6 (db, fd, req, key, uid);
      break;

    case GETHOSTBYADDR:
      addhstbyaddr (db, fd, req, key, uid);
      break;

    case GETHOSTBYADDRv6:
      addhstbyaddrv6 (db, fd, req, key, uid);
      break;

    case GETAI:
      addhstai (db, fd, req, key, uid);
      break;

    case INITGROUPS:
      addinitgroups (db, fd, req, key, uid);
      break;

    case GETSERVBYNAME:
      addservbyname (db, fd, req, key, uid);
      break;

    case GETSERVBYPORT:
      addservbyport (db, fd, req, key, uid);
      break;

    case GETNETGRENT:
      addgetnetgrent (db, fd, req, key, uid);
      break;

    case INNETGR:
      addinnetgr (db, fd, req, key, uid);
      break;

    case GETSTAT:
    case SHUTDOWN:
    case INVALIDATE:
      {
	/* Get the callers credentials.  */
#ifdef SO_PEERCRED
	struct ucred caller;
	socklen_t optlen = sizeof (caller);

	if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &caller, &optlen) < 0)
	  {
	    char buf[256];

	    dbg_log (_("error getting caller's id: %s"),
		     strerror_r (errno, buf, sizeof (buf)));
	    break;
	  }

	uid = caller.uid;
#else
	/* Some systems have no SO_PEERCRED implementation.  They don't
	   care about security so we don't as well.  */
	uid = 0;
#endif
      }

      /* Accept shutdown, getstat and invalidate only from root.  For
	 the stat call also allow the user specified in the config file.  */
      if (req->type == GETSTAT)
	{
	  if (uid == 0 || uid == stat_uid)
	    send_stats (fd, dbs);
	}
      else if (uid == 0)
	{
	  if (req->type == INVALIDATE)
	    invalidate_cache (key, fd);
	  else
	    termination_handler (0);
	}
      break;

    case GETFDPW:
    case GETFDGR:
    case GETFDHST:
    case GETFDSERV:
    case GETFDNETGR:
#ifdef SCM_RIGHTS
      send_ro_fd (reqinfo[req->type].db, key, fd);
#endif
      break;

    default:
      /* Ignore the command, it's nothing we know.  */
      break;
    }
}


/* Restart the process.  */
static void
restart (void)
{
  /* First determine the parameters.  We do not use the parameters
     passed to main() since in case nscd is started by running the
     dynamic linker this will not work.  Yes, this is not the usual
     case but nscd is part of glibc and we occasionally do this.  */
  size_t buflen = 1024;
  char *buf = alloca (buflen);
  size_t readlen = 0;
  int fd = open ("/proc/self/cmdline", O_RDONLY);
  if (fd == -1)
    {
      dbg_log (_("\
cannot open /proc/self/cmdline: %s; disabling paranoia mode"),
	       strerror (errno));

      paranoia = 0;
      return;
    }

  while (1)
    {
      ssize_t n = TEMP_FAILURE_RETRY (read (fd, buf + readlen,
					    buflen - readlen));
      if (n == -1)
	{
	  dbg_log (_("\
cannot read /proc/self/cmdline: %s; disabling paranoia mode"),
		   strerror (errno));

	  close (fd);
	  paranoia = 0;
	  return;
	}

      readlen += n;

      if (readlen < buflen)
	break;

      /* We might have to extend the buffer.  */
      size_t old_buflen = buflen;
      char *newp = extend_alloca (buf, buflen, 2 * buflen);
      buf = memmove (newp, buf, old_buflen);
    }

  close (fd);

  /* Parse the command line.  Worst case scenario: every two
     characters form one parameter (one character plus NUL).  */
  char **argv = alloca ((readlen / 2 + 1) * sizeof (argv[0]));
  int argc = 0;

  char *cp = buf;
  while (cp < buf + readlen)
    {
      argv[argc++] = cp;
      cp = (char *) rawmemchr (cp, '\0') + 1;
    }
  argv[argc] = NULL;

  /* Second, change back to the old user if we changed it.  */
  if (server_user != NULL)
    {
      if (setresuid (old_uid, old_uid, old_uid) != 0)
	{
	  dbg_log (_("\
cannot change to old UID: %s; disabling paranoia mode"),
		   strerror (errno));

	  paranoia = 0;
	  return;
	}

      if (setresgid (old_gid, old_gid, old_gid) != 0)
	{
	  dbg_log (_("\
cannot change to old GID: %s; disabling paranoia mode"),
		   strerror (errno));

	  ignore_value (setuid (server_uid));
	  paranoia = 0;
	  return;
	}
    }

  /* Next change back to the old working directory.  */
  if (chdir (oldcwd) == -1)
    {
      dbg_log (_("\
cannot change to old working directory: %s; disabling paranoia mode"),
	       strerror (errno));

      if (server_user != NULL)
	{
	  ignore_value (setuid (server_uid));
	  ignore_value (setgid (server_gid));
	}
      paranoia = 0;
      return;
    }

  /* Synchronize memory.  */
  int32_t certainly[lastdb];
  for (int cnt = 0; cnt < lastdb; ++cnt)
    if (dbs[cnt].enabled)
      {
	/* Make sure nobody keeps using the database.  */
	dbs[cnt].head->timestamp = 0;
	certainly[cnt] = dbs[cnt].head->nscd_certainly_running;
	dbs[cnt].head->nscd_certainly_running = 0;

	if (dbs[cnt].persistent)
	  // XXX async OK?
	  msync (dbs[cnt].head, dbs[cnt].memsize, MS_ASYNC);
      }

  /* The preparations are done.  */
#ifdef PATH_MAX
  char pathbuf[PATH_MAX];
#else
  char pathbuf[256];
#endif
  /* Try to exec the real nscd program so the process name (as reported
     in /proc/PID/status) will be 'nscd', but fall back to /proc/self/exe
     if readlink or the exec with the result of the readlink call fails.  */
  ssize_t n = readlink ("/proc/self/exe", pathbuf, sizeof (pathbuf) - 1);
  if (n != -1)
    {
      pathbuf[n] = '\0';
      execv (pathbuf, argv);
    }
  execv ("/proc/self/exe", argv);

  /* If we come here, we will never be able to re-exec.  */
  dbg_log (_("re-exec failed: %s; disabling paranoia mode"),
	   strerror (errno));

  if (server_user != NULL)
    {
      ignore_value (setuid (server_uid));
      ignore_value (setgid (server_gid));
    }
  if (chdir ("/") != 0)
    dbg_log (_("cannot change current working directory to \"/\": %s"),
	     strerror (errno));
  paranoia = 0;

  /* Reenable the databases.  */
  time_t now = time (NULL);
  for (int cnt = 0; cnt < lastdb; ++cnt)
    if (dbs[cnt].enabled)
      {
	dbs[cnt].head->timestamp = now;
	dbs[cnt].head->nscd_certainly_running = certainly[cnt];
      }
}


/* List of file descriptors.  */
struct fdlist
{
  int fd;
  struct fdlist *next;
};
/* Memory allocated for the list.  */
static struct fdlist *fdlist;
/* List of currently ready-to-read file descriptors.  */
static struct fdlist *readylist;

/* Conditional variable and mutex to signal availability of entries in
   READYLIST.  The condvar is initialized dynamically since we might
   use a different clock depending on availability.  */
static pthread_cond_t readylist_cond = PTHREAD_COND_INITIALIZER;
static pthread_mutex_t readylist_lock = PTHREAD_MUTEX_INITIALIZER;

/* The clock to use with the condvar.  */
static clockid_t timeout_clock = CLOCK_REALTIME;

/* Number of threads ready to handle the READYLIST.  */
static unsigned long int nready;


/* Function for the clean-up threads.  */
static void *
__attribute__ ((__noreturn__))
nscd_run_prune (void *p)
{
  const long int my_number = (long int) p;
  assert (dbs[my_number].enabled);

  int dont_need_update = setup_thread (&dbs[my_number]);

  time_t now = time (NULL);

  /* We are running.  */
  dbs[my_number].head->timestamp = now;

  struct timespec prune_ts;
  if (__glibc_unlikely (clock_gettime (timeout_clock, &prune_ts) == -1))
    /* Should never happen.  */
    abort ();

  /* Compute the initial timeout time.  Prevent all the timers to go
     off at the same time by adding a db-based value.  */
  prune_ts.tv_sec += CACHE_PRUNE_INTERVAL + my_number;
  dbs[my_number].wakeup_time = now + CACHE_PRUNE_INTERVAL + my_number;

  pthread_mutex_t *prune_lock = &dbs[my_number].prune_lock;
  pthread_mutex_t *prune_run_lock = &dbs[my_number].prune_run_lock;
  pthread_cond_t *prune_cond = &dbs[my_number].prune_cond;

  pthread_mutex_lock (prune_lock);
  while (1)
    {
      /* Wait, but not forever.  */
      int e = 0;
      if (! dbs[my_number].clear_cache)
	e = pthread_cond_timedwait (prune_cond, prune_lock, &prune_ts);
      assert (__builtin_expect (e == 0 || e == ETIMEDOUT, 1));

      time_t next_wait;
      now = time (NULL);
      if (e == ETIMEDOUT || now >= dbs[my_number].wakeup_time
	  || dbs[my_number].clear_cache)
	{
	  /* We will determine the new timout values based on the
	     cache content.  Should there be concurrent additions to
	     the cache which are not accounted for in the cache
	     pruning we want to know about it.  Therefore set the
	     timeout to the maximum.  It will be descreased when adding
	     new entries to the cache, if necessary.  */
	  dbs[my_number].wakeup_time = MAX_TIMEOUT_VALUE;

	  /* Unconditionally reset the flag.  */
	  time_t prune_now = dbs[my_number].clear_cache ? LONG_MAX : now;
	  dbs[my_number].clear_cache = 0;

	  pthread_mutex_unlock (prune_lock);

	  /* We use a separate lock for running the prune function (instead
	     of keeping prune_lock locked) because this enables concurrent
	     invocations of cache_add which might modify the timeout value.  */
	  pthread_mutex_lock (prune_run_lock);
	  next_wait = prune_cache (&dbs[my_number], prune_now, -1);
	  pthread_mutex_unlock (prune_run_lock);

	  next_wait = MAX (next_wait, CACHE_PRUNE_INTERVAL);
	  /* If clients cannot determine for sure whether nscd is running
	     we need to wake up occasionally to update the timestamp.
	     Wait 90% of the update period.  */
#define UPDATE_MAPPING_TIMEOUT (MAPPING_TIMEOUT * 9 / 10)
	  if (__glibc_unlikely (! dont_need_update))
	    {
	      next_wait = MIN (UPDATE_MAPPING_TIMEOUT, next_wait);
	      dbs[my_number].head->timestamp = now;
	    }

	  pthread_mutex_lock (prune_lock);

	  /* Make it known when we will wake up again.  */
	  if (now + next_wait < dbs[my_number].wakeup_time)
	    dbs[my_number].wakeup_time = now + next_wait;
	  else
	    next_wait = dbs[my_number].wakeup_time - now;
	}
      else
	/* The cache was just pruned.  Do not do it again now.  Just
	   use the new timeout value.  */
	next_wait = dbs[my_number].wakeup_time - now;

      if (clock_gettime (timeout_clock, &prune_ts) == -1)
	/* Should never happen.  */
	abort ();

      /* Compute next timeout time.  */
      prune_ts.tv_sec += next_wait;
    }
}


/* This is the main loop.  It is replicated in different threads but
   the use of the ready list makes sure only one thread handles an
   incoming connection.  */
static void *
__attribute__ ((__noreturn__))
nscd_run_worker (void *p)
{
  char buf[256];

  /* Initial locking.  */
  pthread_mutex_lock (&readylist_lock);

  /* One more thread available.  */
  ++nready;

  while (1)
    {
      while (readylist == NULL)
	pthread_cond_wait (&readylist_cond, &readylist_lock);

      struct fdlist *it = readylist->next;
      if (readylist->next == readylist)
	/* Just one entry on the list.  */
	readylist = NULL;
      else
	readylist->next = it->next;

      /* Extract the information and mark the record ready to be used
	 again.  */
      int fd = it->fd;
      it->next = NULL;

      /* One more thread available.  */
      --nready;

      /* We are done with the list.  */
      pthread_mutex_unlock (&readylist_lock);

      /* Now read the request.  */
      request_header req;
      if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, &req, sizeof (req)))
			    != sizeof (req), 0))
	{
	  /* We failed to read data.  Note that this also might mean we
	     failed because we would have blocked.  */
	  if (debug_level > 0)
	    dbg_log (_("short read while reading request: %s"),
		     strerror_r (errno, buf, sizeof (buf)));
	  goto close_and_out;
	}

      /* Check whether this is a valid request type.  */
      if (req.type < GETPWBYNAME || req.type >= LASTREQ)
	goto close_and_out;

      /* Some systems have no SO_PEERCRED implementation.  They don't
	 care about security so we don't as well.  */
      uid_t uid = -1;
#ifdef SO_PEERCRED
      pid_t pid = 0;

      if (__glibc_unlikely (debug_level > 0))
	{
	  struct ucred caller;
	  socklen_t optlen = sizeof (caller);

	  if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &caller, &optlen) == 0)
	    pid = caller.pid;
	}
#else
      const pid_t pid = 0;
#endif

      /* It should not be possible to crash the nscd with a silly
	 request (i.e., a terribly large key).  We limit the size to 1kb.  */
      if (__builtin_expect (req.key_len, 1) < 0
	  || __builtin_expect (req.key_len, 1) > MAXKEYLEN)
	{
	  if (debug_level > 0)
	    dbg_log (_("key length in request too long: %d"), req.key_len);
	}
      else
	{
	  /* Get the key.  */
	  char keybuf[MAXKEYLEN + 1];

	  if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, keybuf,
							  req.key_len))
				!= req.key_len, 0))
	    {
	      /* Again, this can also mean we would have blocked.  */
	      if (debug_level > 0)
		dbg_log (_("short read while reading request key: %s"),
			 strerror_r (errno, buf, sizeof (buf)));
	      goto close_and_out;
	    }
	  keybuf[req.key_len] = '\0';

	  if (__builtin_expect (debug_level, 0) > 0)
	    {
#ifdef SO_PEERCRED
	      if (pid != 0)
		dbg_log (_("\
handle_request: request received (Version = %d) from PID %ld"),
			 req.version, (long int) pid);
	      else
#endif
		dbg_log (_("\
handle_request: request received (Version = %d)"), req.version);
	    }

	  /* Phew, we got all the data, now process it.  */
	  handle_request (fd, &req, keybuf, uid, pid);
	}

    close_and_out:
      /* We are done.  */
      close (fd);

      /* Re-locking.  */
      pthread_mutex_lock (&readylist_lock);

      /* One more thread available.  */
      ++nready;
    }
  /* NOTREACHED */
}


static unsigned int nconns;

static void
fd_ready (int fd)
{
  pthread_mutex_lock (&readylist_lock);

  /* Find an empty entry in FDLIST.  */
  size_t inner;
  for (inner = 0; inner < nconns; ++inner)
    if (fdlist[inner].next == NULL)
      break;
  assert (inner < nconns);

  fdlist[inner].fd = fd;

  if (readylist == NULL)
    readylist = fdlist[inner].next = &fdlist[inner];
  else
    {
      fdlist[inner].next = readylist->next;
      readylist = readylist->next = &fdlist[inner];
    }

  bool do_signal = true;
  if (__glibc_unlikely (nready == 0))
    {
      ++client_queued;
      do_signal = false;

      /* Try to start another thread to help out.  */
      pthread_t th;
      if (nthreads < max_nthreads
	  && pthread_create (&th, &attr, nscd_run_worker,
			     (void *) (long int) nthreads) == 0)
	{
	  /* We got another thread.  */
	  ++nthreads;
	  /* The new thread might need a kick.  */
	  do_signal = true;
	}

    }

  pthread_mutex_unlock (&readylist_lock);

  /* Tell one of the worker threads there is work to do.  */
  if (do_signal)
    pthread_cond_signal (&readylist_cond);
}


/* Check whether restarting should happen.  */
static bool
restart_p (time_t now)
{
  return (paranoia && readylist == NULL && nready == nthreads
	  && now >= restart_time);
}


/* Array for times a connection was accepted.  */
static time_t *starttime;

#ifdef HAVE_INOTIFY
/* Inotify event for changed file.  */
union __inev
{
  struct inotify_event i;
# ifndef PATH_MAX
#  define PATH_MAX 1024
# endif
  char buf[sizeof (struct inotify_event) + PATH_MAX];
};

/* Returns 0 if the file is there otherwise -1.  */
int
check_file (struct traced_file *finfo)
{
  struct stat64 st;
  /* We could check mtime and if different re-add
     the watches, and invalidate the database, but we
     don't because we are called from inotify_check_files
     which should be doing that work.  If sufficient inotify
     events were lost then the next pruning or invalidation
     will do the stat and mtime check.  We don't do it here to
     keep the logic simple.  */
  if (stat64 (finfo->fname, &st) < 0)
    return -1;
  return 0;
}

/* Process the inotify event in INEV. If the event matches any of the files
   registered with a database then mark that database as requiring its cache
   to be cleared. We indicate the cache needs clearing by setting
   TO_CLEAR[DBCNT] to true for the matching database.  */
static void
inotify_check_files (bool *to_clear, union __inev *inev)
{
  /* Check which of the files changed.  */
  for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
    {
      struct traced_file *finfo = dbs[dbcnt].traced_files;

      while (finfo != NULL)
	{
	  /* The configuration file was moved or deleted.
	     We stop watching it at that point, and reinitialize.  */
	  if (finfo->inotify_descr[TRACED_FILE] == inev->i.wd
	      && ((inev->i.mask & IN_MOVE_SELF)
		  || (inev->i.mask & IN_DELETE_SELF)
		  || (inev->i.mask & IN_IGNORED)))
	    {
	      int ret;
	      bool moved = (inev->i.mask & IN_MOVE_SELF) != 0;

	      if (check_file (finfo) == 0)
	        {
		  dbg_log (_("ignored inotify event for `%s` (file exists)"),
			   finfo->fname);
		  return;
		}

	      dbg_log (_("monitored file `%s` was %s, removing watch"),
		       finfo->fname, moved ? "moved" : "deleted");
	      /* File was moved out, remove the watch.  Watches are
		 automatically removed when the file is deleted.  */
	      if (moved)
		{
		  ret = inotify_rm_watch (inotify_fd, inev->i.wd);
		  if (ret < 0)
		    dbg_log (_("failed to remove file watch `%s`: %s"),
			     finfo->fname, strerror (errno));
		}
	      finfo->inotify_descr[TRACED_FILE] = -1;
	      to_clear[dbcnt] = true;
	      if (finfo->call_res_init)
	        res_init ();
	      return;
	    }
	  /* The configuration file was open for writing and has just closed.
	     We reset the cache and reinitialize.  */
	  if (finfo->inotify_descr[TRACED_FILE] == inev->i.wd
	      && inev->i.mask & IN_CLOSE_WRITE)
	    {
	      /* Mark cache as needing to be cleared and reinitialize.  */
	      dbg_log (_("monitored file `%s` was written to"), finfo->fname);
	      to_clear[dbcnt] = true;
	      if (finfo->call_res_init)
	        res_init ();
	      return;
	    }
	  /* The parent directory was moved or deleted.  We trigger one last
	     invalidation.  At the next pruning or invalidation we may add
	     this watch back if the file is present again.  */
	  if (finfo->inotify_descr[TRACED_DIR] == inev->i.wd
	      && ((inev->i.mask & IN_DELETE_SELF)
		  || (inev->i.mask & IN_MOVE_SELF)
		  || (inev->i.mask & IN_IGNORED)))
	    {
	      bool moved = (inev->i.mask & IN_MOVE_SELF) != 0;
	      /* The directory watch may have already been removed
		 but we don't know so we just remove it again and
		 ignore the error.  Then we remove the file watch.
		 Note: watches are automatically removed for deleted
		 files.  */
	      if (moved)
		inotify_rm_watch (inotify_fd, inev->i.wd);
	      if (finfo->inotify_descr[TRACED_FILE] != -1)
		{
		  dbg_log (_("monitored parent directory `%s` was %s, removing watch on `%s`"),
			   finfo->dname, moved ? "moved" : "deleted", finfo->fname);
		  if (inotify_rm_watch (inotify_fd, finfo->inotify_descr[TRACED_FILE]) < 0)
		    dbg_log (_("failed to remove file watch `%s`: %s"),
			     finfo->dname, strerror (errno));
		}
	      finfo->inotify_descr[TRACED_FILE] = -1;
	      finfo->inotify_descr[TRACED_DIR] = -1;
	      to_clear[dbcnt] = true;
	      if (finfo->call_res_init)
	        res_init ();
	      /* Continue to the next entry since this might be the
		 parent directory for multiple registered files and
		 we want to remove watches for all registered files.  */
	      continue;
	    }
	  /* The parent directory had a create or moved to event.  */
	  if (finfo->inotify_descr[TRACED_DIR] == inev->i.wd
	      && ((inev->i.mask & IN_MOVED_TO)
		  || (inev->i.mask & IN_CREATE))
	      && strcmp (inev->i.name, finfo->sfname) == 0)
	    {
	      /* We detected a directory change.  We look for the creation
		 of the file we are tracking or the move of the same file
		 into the directory.  */
	      int ret;
	      dbg_log (_("monitored file `%s` was %s, adding watch"),
		       finfo->fname,
		       inev->i.mask & IN_CREATE ? "created" : "moved into place");
	      /* File was moved in or created.  Regenerate the watch.  */
	      if (finfo->inotify_descr[TRACED_FILE] != -1)
		inotify_rm_watch (inotify_fd,
				  finfo->inotify_descr[TRACED_FILE]);

	      ret = inotify_add_watch (inotify_fd,
				       finfo->fname,
				       TRACED_FILE_MASK);
	      if (ret < 0)
		dbg_log (_("failed to add file watch `%s`: %s"),
			 finfo->fname, strerror (errno));

	      finfo->inotify_descr[TRACED_FILE] = ret;

	      /* The file is new or moved so mark cache as needing to
		 be cleared and reinitialize.  */
	      to_clear[dbcnt] = true;
	      if (finfo->call_res_init)
		res_init ();

	      /* Done re-adding the watch.  Don't return, we may still
		 have other files in this same directory, same watch
		 descriptor, and need to process them.  */
	    }
	  /* Other events are ignored, and we move on to the next file.  */
	  finfo = finfo->next;
        }
    }
}

/* If an entry in the array of booleans TO_CLEAR is TRUE then clear the cache
   for the associated database, otherwise do nothing. The TO_CLEAR array must
   have LASTDB entries.  */
static inline void
clear_db_cache (bool *to_clear)
{
  for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
    if (to_clear[dbcnt])
      {
	pthread_mutex_lock (&dbs[dbcnt].prune_lock);
	dbs[dbcnt].clear_cache = 1;
	pthread_mutex_unlock (&dbs[dbcnt].prune_lock);
	pthread_cond_signal (&dbs[dbcnt].prune_cond);
      }
}

int
handle_inotify_events (void)
{
  bool to_clear[lastdb] = { false, };
  union __inev inev;

  /* Read all inotify events for files registered via
     register_traced_file().  */
  while (1)
    {
      /* Potentially read multiple events into buf.  */
      ssize_t nb = TEMP_FAILURE_RETRY (read (inotify_fd,
					     &inev.buf,
					     sizeof (inev)));
      if (nb < (ssize_t) sizeof (struct inotify_event))
	{
	  /* Not even 1 event.  */
	  if (__glibc_unlikely (nb == -1 && errno != EAGAIN))
	    return -1;
	  /* Done reading events that are ready.  */
	  break;
	}
      /* Process all events.  The normal inotify interface delivers
	 complete events on a read and never a partial event.  */
      char *eptr = &inev.buf[0];
      ssize_t count;
      while (1)
	{
	  /* Check which of the files changed.  */
	  inotify_check_files (to_clear, &inev);
	  count = sizeof (struct inotify_event) + inev.i.len;
	  eptr += count;
	  nb -= count;
	  if (nb >= (ssize_t) sizeof (struct inotify_event))
	    memcpy (&inev, eptr, nb);
	  else
	    break;
	}
      continue;
    }
  /* Actually perform the cache clearing.  */
  clear_db_cache (to_clear);
  return 0;
}

#endif

static void
__attribute__ ((__noreturn__))
main_loop_poll (void)
{
  struct pollfd *conns = (struct pollfd *) xmalloc (nconns
						    * sizeof (conns[0]));

  conns[0].fd = sock;
  conns[0].events = POLLRDNORM;
  size_t nused = 1;
  size_t firstfree = 1;

#ifdef HAVE_INOTIFY
  if (inotify_fd != -1)
    {
      conns[1].fd = inotify_fd;
      conns[1].events = POLLRDNORM;
      nused = 2;
      firstfree = 2;
    }
#endif

#ifdef HAVE_NETLINK
  size_t idx_nl_status_fd = 0;
  if (nl_status_fd != -1)
    {
      idx_nl_status_fd = nused;
      conns[nused].fd = nl_status_fd;
      conns[nused].events = POLLRDNORM;
      ++nused;
      firstfree = nused;
    }
#endif

  while (1)
    {
      /* Wait for any event.  We wait at most a couple of seconds so
	 that we can check whether we should close any of the accepted
	 connections since we have not received a request.  */
#define MAX_ACCEPT_TIMEOUT 30
#define MIN_ACCEPT_TIMEOUT 5
#define MAIN_THREAD_TIMEOUT \
  (MAX_ACCEPT_TIMEOUT * 1000						      \
   - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * 1000 * nused) / (2 * nconns))

      int n = poll (conns, nused, MAIN_THREAD_TIMEOUT);

      time_t now = time (NULL);

      /* If there is a descriptor ready for reading or there is a new
	 connection, process this now.  */
      if (n > 0)
	{
	  if (conns[0].revents != 0)
	    {
	      /* We have a new incoming connection.  Accept the connection.  */
	      int fd = TEMP_FAILURE_RETRY (accept4 (sock, NULL, NULL,
						    SOCK_NONBLOCK));

	      /* Use the descriptor if we have not reached the limit.  */
	      if (fd >= 0)
		{
		  if (firstfree < nconns)
		    {
		      conns[firstfree].fd = fd;
		      conns[firstfree].events = POLLRDNORM;
		      starttime[firstfree] = now;
		      if (firstfree >= nused)
			nused = firstfree + 1;

		      do
			++firstfree;
		      while (firstfree < nused && conns[firstfree].fd != -1);
		    }
		  else
		    /* We cannot use the connection so close it.  */
		    close (fd);
		}

	      --n;
	    }

	  size_t first = 1;
#ifdef HAVE_INOTIFY
	  if (inotify_fd != -1 && conns[1].fd == inotify_fd)
	    {
	      if (conns[1].revents != 0)
		{
		  int ret;
		  ret = handle_inotify_events ();
		  if (ret == -1)
		    {
		      /* Something went wrong when reading the inotify
			 data.  Better disable inotify.  */
		      dbg_log (_("disabled inotify-based monitoring after read error %d"), errno);
		      conns[1].fd = -1;
		      firstfree = 1;
		      if (nused == 2)
			nused = 1;
		      close (inotify_fd);
		      inotify_fd = -1;
		    }
		  --n;
		}

	      first = 2;
	    }
#endif

#ifdef HAVE_NETLINK
	  if (idx_nl_status_fd != 0 && conns[idx_nl_status_fd].revents != 0)
	    {
	      char buf[4096];
	      /* Read all the data.  We do not interpret it here.  */
	      while (TEMP_FAILURE_RETRY (read (nl_status_fd, buf,
					       sizeof (buf))) != -1)
		;

	      dbs[hstdb].head->extra_data[NSCD_HST_IDX_CONF_TIMESTAMP]
		= __bump_nl_timestamp ();
	    }
#endif

	  for (size_t cnt = first; cnt < nused && n > 0; ++cnt)
	    if (conns[cnt].revents != 0)
	      {
		fd_ready (conns[cnt].fd);

		/* Clean up the CONNS array.  */
		conns[cnt].fd = -1;
		if (cnt < firstfree)
		  firstfree = cnt;
		if (cnt == nused - 1)
		  do
		    --nused;
		  while (conns[nused - 1].fd == -1);

		--n;
	      }
	}

      /* Now find entries which have timed out.  */
      assert (nused > 0);

      /* We make the timeout length depend on the number of file
	 descriptors currently used.  */
#define ACCEPT_TIMEOUT \
  (MAX_ACCEPT_TIMEOUT							      \
   - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * nused) / nconns)
      time_t laststart = now - ACCEPT_TIMEOUT;

      for (size_t cnt = nused - 1; cnt > 0; --cnt)
	{
	  if (conns[cnt].fd != -1 && starttime[cnt] < laststart)
	    {
	      /* Remove the entry, it timed out.  */
	      (void) close (conns[cnt].fd);
	      conns[cnt].fd = -1;

	      if (cnt < firstfree)
		firstfree = cnt;
	      if (cnt == nused - 1)
		do
		  --nused;
		while (conns[nused - 1].fd == -1);
	    }
	}

      if (restart_p (now))
	restart ();
    }
}


#ifdef HAVE_EPOLL
static void
main_loop_epoll (int efd)
{
  struct epoll_event ev = { 0, };
  int nused = 1;
  size_t highest = 0;

  /* Add the socket.  */
  ev.events = EPOLLRDNORM;
  ev.data.fd = sock;
  if (epoll_ctl (efd, EPOLL_CTL_ADD, sock, &ev) == -1)
    /* We cannot use epoll.  */
    return;

# ifdef HAVE_INOTIFY
  if (inotify_fd != -1)
    {
      ev.events = EPOLLRDNORM;
      ev.data.fd = inotify_fd;
      if (epoll_ctl (efd, EPOLL_CTL_ADD, inotify_fd, &ev) == -1)
	/* We cannot use epoll.  */
	return;
      nused = 2;
    }
# endif

# ifdef HAVE_NETLINK
  if (nl_status_fd != -1)
    {
      ev.events = EPOLLRDNORM;
      ev.data.fd = nl_status_fd;
      if (epoll_ctl (efd, EPOLL_CTL_ADD, nl_status_fd, &ev) == -1)
	/* We cannot use epoll.  */
	return;
    }
# endif

  while (1)
    {
      struct epoll_event revs[100];
# define nrevs (sizeof (revs) / sizeof (revs[0]))

      int n = epoll_wait (efd, revs, nrevs, MAIN_THREAD_TIMEOUT);

      time_t now = time (NULL);

      for (int cnt = 0; cnt < n; ++cnt)
	if (revs[cnt].data.fd == sock)
	  {
	    /* A new connection.  */
	    int fd = TEMP_FAILURE_RETRY (accept4 (sock, NULL, NULL,
						  SOCK_NONBLOCK));

	    /* Use the descriptor if we have not reached the limit.  */
	    if (fd >= 0)
	      {
		/* Try to add the  new descriptor.  */
		ev.data.fd = fd;
		if (fd >= nconns
		    || epoll_ctl (efd, EPOLL_CTL_ADD, fd, &ev) == -1)
		  /* The descriptor is too large or something went
		     wrong.  Close the descriptor.  */
		  close (fd);
		else
		  {
		    /* Remember when we accepted the connection.  */
		    starttime[fd] = now;

		    if (fd > highest)
		      highest = fd;

		    ++nused;
		  }
	      }
	  }
# ifdef HAVE_INOTIFY
	else if (revs[cnt].data.fd == inotify_fd)
	  {
	    int ret;
	    ret = handle_inotify_events ();
	    if (ret == -1)
	      {
		/* Something went wrong when reading the inotify
		   data.  Better disable inotify.  */
		dbg_log (_("disabled inotify-based monitoring after read error %d"), errno);
		(void) epoll_ctl (efd, EPOLL_CTL_DEL, inotify_fd, NULL);
		close (inotify_fd);
		inotify_fd = -1;
		break;
	      }
	  }
# endif
# ifdef HAVE_NETLINK
	else if (revs[cnt].data.fd == nl_status_fd)
	  {
	    char buf[4096];
	    /* Read all the data.  We do not interpret it here.  */
	    while (TEMP_FAILURE_RETRY (read (nl_status_fd, buf,
					     sizeof (buf))) != -1)
	      ;

	    __bump_nl_timestamp ();
	  }
# endif
	else
	  {
	    /* Remove the descriptor from the epoll descriptor.  */
	    (void) epoll_ctl (efd, EPOLL_CTL_DEL, revs[cnt].data.fd, NULL);

	    /* Get a worker to handle the request.  */
	    fd_ready (revs[cnt].data.fd);

	    /* Reset the time.  */
	    starttime[revs[cnt].data.fd] = 0;
	    if (revs[cnt].data.fd == highest)
	      do
		--highest;
	      while (highest > 0 && starttime[highest] == 0);

	    --nused;
	  }

      /*  Now look for descriptors for accepted connections which have
	  no reply in too long of a time.  */
      time_t laststart = now - ACCEPT_TIMEOUT;
      assert (starttime[sock] == 0);
# ifdef HAVE_INOTIFY
      assert (inotify_fd == -1 || starttime[inotify_fd] == 0);
# endif
      assert (nl_status_fd == -1 || starttime[nl_status_fd] == 0);
      for (int cnt = highest; cnt > STDERR_FILENO; --cnt)
	if (starttime[cnt] != 0 && starttime[cnt] < laststart)
	  {
	    /* We are waiting for this one for too long.  Close it.  */
	    (void) epoll_ctl (efd, EPOLL_CTL_DEL, cnt, NULL);

	    (void) close (cnt);

	    starttime[cnt] = 0;
	    if (cnt == highest)
	      --highest;
	  }
	else if (cnt != sock && starttime[cnt] == 0 && cnt == highest)
	  --highest;

      if (restart_p (now))
	restart ();
    }
}
#endif


/* Start all the threads we want.  The initial process is thread no. 1.  */
void
start_threads (void)
{
  /* Initialize the conditional variable we will use.  The only
     non-standard attribute we might use is the clock selection.  */
  pthread_condattr_t condattr;
  pthread_condattr_init (&condattr);

#if defined _POSIX_CLOCK_SELECTION && _POSIX_CLOCK_SELECTION >= 0 \
    && defined _POSIX_MONOTONIC_CLOCK && _POSIX_MONOTONIC_CLOCK >= 0
  /* Determine whether the monotonous clock is available.  */
  struct timespec dummy;
# if _POSIX_MONOTONIC_CLOCK == 0
  if (sysconf (_SC_MONOTONIC_CLOCK) > 0)
# endif
# if _POSIX_CLOCK_SELECTION == 0
    if (sysconf (_SC_CLOCK_SELECTION) > 0)
# endif
      if (clock_getres (CLOCK_MONOTONIC, &dummy) == 0
	  && pthread_condattr_setclock (&condattr, CLOCK_MONOTONIC) == 0)
	timeout_clock = CLOCK_MONOTONIC;
#endif

  /* Create the attribute for the threads.  They are all created
     detached.  */
  pthread_attr_init (&attr);
  pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED);
  /* Use 1MB stacks, twice as much for 64-bit architectures.  */
  pthread_attr_setstacksize (&attr, NSCD_THREAD_STACKSIZE);

  /* We allow less than LASTDB threads only for debugging.  */
  if (debug_level == 0)
    nthreads = MAX (nthreads, lastdb);

  /* Create the threads which prune the databases.  */
  // XXX Ideally this work would be done by some of the worker threads.
  // XXX But this is problematic since we would need to be able to wake
  // XXX them up explicitly as well as part of the group handling the
  // XXX ready-list.  This requires an operation where we can wait on
  // XXX two conditional variables at the same time.  This operation
  // XXX does not exist (yet).
  for (long int i = 0; i < lastdb; ++i)
    {
      /* Initialize the conditional variable.  */
      if (pthread_cond_init (&dbs[i].prune_cond, &condattr) != 0)
	{
	  dbg_log (_("could not initialize conditional variable"));
	  do_exit (1, 0, NULL);
	}

      pthread_t th;
      if (dbs[i].enabled
	  && pthread_create (&th, &attr, nscd_run_prune, (void *) i) != 0)
	{
	  dbg_log (_("could not start clean-up thread; terminating"));
	  do_exit (1, 0, NULL);
	}
    }

  pthread_condattr_destroy (&condattr);

  for (long int i = 0; i < nthreads; ++i)
    {
      pthread_t th;
      if (pthread_create (&th, &attr, nscd_run_worker, NULL) != 0)
	{
	  if (i == 0)
	    {
	      dbg_log (_("could not start any worker thread; terminating"));
	      do_exit (1, 0, NULL);
	    }

	  break;
	}
    }

  /* Now it is safe to let the parent know that we're doing fine and it can
     exit.  */
  notify_parent (0);

  /* Determine how much room for descriptors we should initially
     allocate.  This might need to change later if we cap the number
     with MAXCONN.  */
  const long int nfds = sysconf (_SC_OPEN_MAX);
#define MINCONN 32
#define MAXCONN 16384
  if (nfds == -1 || nfds > MAXCONN)
    nconns = MAXCONN;
  else if (nfds < MINCONN)
    nconns = MINCONN;
  else
    nconns = nfds;

  /* We need memory to pass descriptors on to the worker threads.  */
  fdlist = (struct fdlist *) xcalloc (nconns, sizeof (fdlist[0]));
  /* Array to keep track when connection was accepted.  */
  starttime = (time_t *) xcalloc (nconns, sizeof (starttime[0]));

  /* In the main thread we execute the loop which handles incoming
     connections.  */
#ifdef HAVE_EPOLL
  int efd = epoll_create (100);
  if (efd != -1)
    {
      main_loop_epoll (efd);
      close (efd);
    }
#endif

  main_loop_poll ();
}


/* Look up the uid, gid, and supplementary groups to run nscd as. When
   this function is called, we are not listening on the nscd socket yet so
   we can just use the ordinary lookup functions without causing a lockup  */
static void
begin_drop_privileges (void)
{
  struct passwd *pwd = getpwnam (server_user);

  if (pwd == NULL)
    {
      dbg_log (_("Failed to run nscd as user '%s'"), server_user);
      do_exit (EXIT_FAILURE, 0,
	       _("Failed to run nscd as user '%s'"), server_user);
    }

  server_uid = pwd->pw_uid;
  server_gid = pwd->pw_gid;

  /* Save the old UID/GID if we have to change back.  */
  if (paranoia)
    {
      old_uid = getuid ();
      old_gid = getgid ();
    }

  if (getgrouplist (server_user, server_gid, NULL, &server_ngroups) == 0)
    {
      /* This really must never happen.  */
      dbg_log (_("Failed to run nscd as user '%s'"), server_user);
      do_exit (EXIT_FAILURE, errno,
	       _("initial getgrouplist failed"));
    }

  server_groups = (gid_t *) xmalloc (server_ngroups * sizeof (gid_t));

  if (getgrouplist (server_user, server_gid, server_groups, &server_ngroups)
      == -1)
    {
      dbg_log (_("Failed to run nscd as user '%s'"), server_user);
      do_exit (EXIT_FAILURE, errno, _("getgrouplist failed"));
    }
}


/* Call setgroups(), setgid(), and setuid() to drop root privileges and
   run nscd as the user specified in the configuration file.  */
static void
finish_drop_privileges (void)
{
#if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP
  /* We need to preserve the capabilities to connect to the audit daemon.  */
  cap_t new_caps = preserve_capabilities ();
#endif

  if (setgroups (server_ngroups, server_groups) == -1)
    {
      dbg_log (_("Failed to run nscd as user '%s'"), server_user);
      do_exit (EXIT_FAILURE, errno, _("setgroups failed"));
    }

  int res;
  if (paranoia)
    res = setresgid (server_gid, server_gid, old_gid);
  else
    res = setgid (server_gid);
  if (res == -1)
    {
      dbg_log (_("Failed to run nscd as user '%s'"), server_user);
      do_exit (4, errno, "setgid");
    }

  if (paranoia)
    res = setresuid (server_uid, server_uid, old_uid);
  else
    res = setuid (server_uid);
  if (res == -1)
    {
      dbg_log (_("Failed to run nscd as user '%s'"), server_user);
      do_exit (4, errno, "setuid");
    }

#if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP
  /* Remove the temporary capabilities.  */
  install_real_capabilities (new_caps);
#endif
}