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
|
/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 only,
* as published by the Free Software Foundation.
*
* 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 version 2 for more details (a copy is included
* in the LICENSE file that accompanied this code).
*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
* http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
*
* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
* CA 95054 USA or visit www.sun.com if you need additional information or
* have any questions.
*
* GPL HEADER END
*/
/*
* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
* Copyright (c) 2010, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Lustre is a trademark of Sun Microsystems, Inc.
*/
/**
* This file contains Asynchronous System Trap (AST) handlers and related
* LDLM request-processing routines.
*
* An AST is a callback issued on a lock when its state is changed. There are
* several different types of ASTs (callbacks) registered for each lock:
*
* - completion AST: when a lock is enqueued by some process, but cannot be
* granted immediately due to other conflicting locks on the same resource,
* the completion AST is sent to notify the caller when the lock is
* eventually granted
*
* - blocking AST: when a lock is granted to some process, if another process
* enqueues a conflicting (blocking) lock on a resource, a blocking AST is
* sent to notify the holder(s) of the lock(s) of the conflicting lock
* request. The lock holder(s) must release their lock(s) on that resource in
* a timely manner or be evicted by the server.
*
* - glimpse AST: this is used when a process wants information about a lock
* (i.e. the lock value block (LVB)) but does not necessarily require holding
* the lock. If the resource is locked, the lock holder(s) are sent glimpse
* ASTs and the LVB is returned to the caller, and lock holder(s) may CANCEL
* their lock(s) if they are idle. If the resource is not locked, the server
* may grant the lock.
*/
#define DEBUG_SUBSYSTEM S_LDLM
#include <lustre_dlm.h>
#include <obd_class.h>
#include <obd.h>
#include "ldlm_internal.h"
int ldlm_enqueue_min = OBD_TIMEOUT_DEFAULT;
CFS_MODULE_PARM(ldlm_enqueue_min, "i", int, 0644,
"lock enqueue timeout minimum");
/* in client side, whether the cached locks will be canceled before replay */
unsigned int ldlm_cancel_unused_locks_before_replay = 1;
static void interrupted_completion_wait(void *data)
{
}
struct lock_wait_data {
struct ldlm_lock *lwd_lock;
__u32 lwd_conn_cnt;
};
struct ldlm_async_args {
struct lustre_handle lock_handle;
};
int ldlm_expired_completion_wait(void *data)
{
struct lock_wait_data *lwd = data;
struct ldlm_lock *lock = lwd->lwd_lock;
struct obd_import *imp;
struct obd_device *obd;
ENTRY;
if (lock->l_conn_export == NULL) {
static cfs_time_t next_dump = 0, last_dump = 0;
if (ptlrpc_check_suspend())
RETURN(0);
LCONSOLE_WARN("lock timed out (enqueued at "CFS_TIME_T", "
CFS_DURATION_T"s ago)\n",
lock->l_last_activity,
cfs_time_sub(cfs_time_current_sec(),
lock->l_last_activity));
LDLM_DEBUG(lock, "lock timed out (enqueued at "CFS_TIME_T", "
CFS_DURATION_T"s ago); not entering recovery in "
"server code, just going back to sleep",
lock->l_last_activity,
cfs_time_sub(cfs_time_current_sec(),
lock->l_last_activity));
if (cfs_time_after(cfs_time_current(), next_dump)) {
last_dump = next_dump;
next_dump = cfs_time_shift(300);
ldlm_namespace_dump(D_DLMTRACE,
ldlm_lock_to_ns(lock));
if (last_dump == 0)
libcfs_debug_dumplog();
}
RETURN(0);
}
obd = lock->l_conn_export->exp_obd;
imp = obd->u.cli.cl_import;
ptlrpc_fail_import(imp, lwd->lwd_conn_cnt);
LDLM_ERROR(lock, "lock timed out (enqueued at "CFS_TIME_T", "
CFS_DURATION_T"s ago), entering recovery for %s@%s",
lock->l_last_activity,
cfs_time_sub(cfs_time_current_sec(), lock->l_last_activity),
obd2cli_tgt(obd), imp->imp_connection->c_remote_uuid.uuid);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_expired_completion_wait);
/* We use the same basis for both server side and client side functions
from a single node. */
int ldlm_get_enq_timeout(struct ldlm_lock *lock)
{
int timeout = at_get(ldlm_lock_to_ns_at(lock));
if (AT_OFF)
return obd_timeout / 2;
/* Since these are non-updating timeouts, we should be conservative.
It would be nice to have some kind of "early reply" mechanism for
lock callbacks too... */
timeout = min_t(int, at_max, timeout + (timeout >> 1)); /* 150% */
return max(timeout, ldlm_enqueue_min);
}
EXPORT_SYMBOL(ldlm_get_enq_timeout);
/**
* Helper function for ldlm_completion_ast(), updating timings when lock is
* actually granted.
*/
static int ldlm_completion_tail(struct ldlm_lock *lock)
{
long delay;
int result;
if (lock->l_destroyed || lock->l_flags & LDLM_FL_FAILED) {
LDLM_DEBUG(lock, "client-side enqueue: destroyed");
result = -EIO;
} else {
delay = cfs_time_sub(cfs_time_current_sec(),
lock->l_last_activity);
LDLM_DEBUG(lock, "client-side enqueue: granted after "
CFS_DURATION_T"s", delay);
/* Update our time estimate */
at_measured(ldlm_lock_to_ns_at(lock),
delay);
result = 0;
}
return result;
}
/**
* Implementation of ->l_completion_ast() for a client, that doesn't wait
* until lock is granted. Suitable for locks enqueued through ptlrpcd, of
* other threads that cannot block for long.
*/
int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data)
{
ENTRY;
if (flags == LDLM_FL_WAIT_NOREPROC) {
LDLM_DEBUG(lock, "client-side enqueue waiting on pending lock");
RETURN(0);
}
if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
LDLM_FL_BLOCK_CONV))) {
wake_up(&lock->l_waitq);
RETURN(ldlm_completion_tail(lock));
}
LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
"going forward");
ldlm_reprocess_all(lock->l_resource);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_completion_ast_async);
/**
* Generic LDLM "completion" AST. This is called in several cases:
*
* - when a reply to an ENQUEUE RPC is received from the server
* (ldlm_cli_enqueue_fini()). Lock might be granted or not granted at
* this point (determined by flags);
*
* - when LDLM_CP_CALLBACK RPC comes to client to notify it that lock has
* been granted;
*
* - when ldlm_lock_match(LDLM_FL_LVB_READY) is about to wait until lock
* gets correct lvb;
*
* - to force all locks when resource is destroyed (cleanup_resource());
*
* - during lock conversion (not used currently).
*
* If lock is not granted in the first case, this function waits until second
* or penultimate cases happen in some other thread.
*
*/
int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
{
/* XXX ALLOCATE - 160 bytes */
struct lock_wait_data lwd;
struct obd_device *obd;
struct obd_import *imp = NULL;
struct l_wait_info lwi;
__u32 timeout;
int rc = 0;
ENTRY;
if (flags == LDLM_FL_WAIT_NOREPROC) {
LDLM_DEBUG(lock, "client-side enqueue waiting on pending lock");
goto noreproc;
}
if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
LDLM_FL_BLOCK_CONV))) {
wake_up(&lock->l_waitq);
RETURN(0);
}
LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
"sleeping");
noreproc:
obd = class_exp2obd(lock->l_conn_export);
/* if this is a local lock, then there is no import */
if (obd != NULL) {
imp = obd->u.cli.cl_import;
}
/* Wait a long time for enqueue - server may have to callback a
lock from another client. Server will evict the other client if it
doesn't respond reasonably, and then give us the lock. */
timeout = ldlm_get_enq_timeout(lock) * 2;
lwd.lwd_lock = lock;
if (lock->l_flags & LDLM_FL_NO_TIMEOUT) {
LDLM_DEBUG(lock, "waiting indefinitely because of NO_TIMEOUT");
lwi = LWI_INTR(interrupted_completion_wait, &lwd);
} else {
lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(timeout),
ldlm_expired_completion_wait,
interrupted_completion_wait, &lwd);
}
if (imp != NULL) {
spin_lock(&imp->imp_lock);
lwd.lwd_conn_cnt = imp->imp_conn_cnt;
spin_unlock(&imp->imp_lock);
}
if (ns_is_client(ldlm_lock_to_ns(lock)) &&
OBD_FAIL_CHECK_RESET(OBD_FAIL_LDLM_INTR_CP_AST,
OBD_FAIL_LDLM_CP_BL_RACE | OBD_FAIL_ONCE)) {
lock->l_flags |= LDLM_FL_FAIL_LOC;
rc = -EINTR;
} else {
/* Go to sleep until the lock is granted or cancelled. */
rc = l_wait_event(lock->l_waitq,
is_granted_or_cancelled(lock), &lwi);
}
if (rc) {
LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
rc);
RETURN(rc);
}
RETURN(ldlm_completion_tail(lock));
}
EXPORT_SYMBOL(ldlm_completion_ast);
/**
* A helper to build a blocking AST function
*
* Perform a common operation for blocking ASTs:
* defferred lock cancellation.
*
* \param lock the lock blocking or canceling AST was called on
* \retval 0
* \see mdt_blocking_ast
* \see ldlm_blocking_ast
*/
int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock)
{
int do_ast;
ENTRY;
lock->l_flags |= LDLM_FL_CBPENDING;
do_ast = (!lock->l_readers && !lock->l_writers);
unlock_res_and_lock(lock);
if (do_ast) {
struct lustre_handle lockh;
int rc;
LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
ldlm_lock2handle(lock, &lockh);
rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
if (rc < 0)
CERROR("ldlm_cli_cancel: %d\n", rc);
} else {
LDLM_DEBUG(lock, "Lock still has references, will be "
"cancelled later");
}
RETURN(0);
}
EXPORT_SYMBOL(ldlm_blocking_ast_nocheck);
/**
* Server blocking AST
*
* ->l_blocking_ast() callback for LDLM locks acquired by server-side
* OBDs.
*
* \param lock the lock which blocks a request or cancelling lock
* \param desc unused
* \param data unused
* \param flag indicates whether this cancelling or blocking callback
* \retval 0
* \see ldlm_blocking_ast_nocheck
*/
int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
void *data, int flag)
{
ENTRY;
if (flag == LDLM_CB_CANCELING) {
/* Don't need to do anything here. */
RETURN(0);
}
lock_res_and_lock(lock);
/* Get this: if ldlm_blocking_ast is racing with intent_policy, such
* that ldlm_blocking_ast is called just before intent_policy method
* takes the lr_lock, then by the time we get the lock, we might not
* be the correct blocking function anymore. So check, and return
* early, if so. */
if (lock->l_blocking_ast != ldlm_blocking_ast) {
unlock_res_and_lock(lock);
RETURN(0);
}
RETURN(ldlm_blocking_ast_nocheck(lock));
}
EXPORT_SYMBOL(ldlm_blocking_ast);
/**
* ->l_glimpse_ast() for DLM extent locks acquired on the server-side. See
* comment in filter_intent_policy() on why you may need this.
*/
int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp)
{
/*
* Returning -ELDLM_NO_LOCK_DATA actually works, but the reason for
* that is rather subtle: with OST-side locking, it may so happen that
* _all_ extent locks are held by the OST. If client wants to obtain
* current file size it calls ll{,u}_glimpse_size(), and (as locks are
* on the server), dummy glimpse callback fires and does
* nothing. Client still receives correct file size due to the
* following fragment in filter_intent_policy():
*
* rc = l->l_glimpse_ast(l, NULL); // this will update the LVB
* if (rc != 0 && res->lr_namespace->ns_lvbo &&
* res->lr_namespace->ns_lvbo->lvbo_update) {
* res->lr_namespace->ns_lvbo->lvbo_update(res, NULL, 0, 1);
* }
*
* that is, after glimpse_ast() fails, filter_lvbo_update() runs, and
* returns correct file size to the client.
*/
return -ELDLM_NO_LOCK_DATA;
}
EXPORT_SYMBOL(ldlm_glimpse_ast);
/**
* Enqueue a local lock (typically on a server).
*/
int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
const struct ldlm_res_id *res_id,
ldlm_type_t type, ldlm_policy_data_t *policy,
ldlm_mode_t mode, __u64 *flags,
ldlm_blocking_callback blocking,
ldlm_completion_callback completion,
ldlm_glimpse_callback glimpse,
void *data, __u32 lvb_len, enum lvb_type lvb_type,
const __u64 *client_cookie,
struct lustre_handle *lockh)
{
struct ldlm_lock *lock;
int err;
const struct ldlm_callback_suite cbs = { .lcs_completion = completion,
.lcs_blocking = blocking,
.lcs_glimpse = glimpse,
};
ENTRY;
LASSERT(!(*flags & LDLM_FL_REPLAY));
if (unlikely(ns_is_client(ns))) {
CERROR("Trying to enqueue local lock in a shadow namespace\n");
LBUG();
}
lock = ldlm_lock_create(ns, res_id, type, mode, &cbs, data, lvb_len,
lvb_type);
if (unlikely(!lock))
GOTO(out_nolock, err = -ENOMEM);
ldlm_lock2handle(lock, lockh);
/* NB: we don't have any lock now (lock_res_and_lock)
* because it's a new lock */
ldlm_lock_addref_internal_nolock(lock, mode);
lock->l_flags |= LDLM_FL_LOCAL;
if (*flags & LDLM_FL_ATOMIC_CB)
lock->l_flags |= LDLM_FL_ATOMIC_CB;
if (policy != NULL)
lock->l_policy_data = *policy;
if (client_cookie != NULL)
lock->l_client_cookie = *client_cookie;
if (type == LDLM_EXTENT)
lock->l_req_extent = policy->l_extent;
err = ldlm_lock_enqueue(ns, &lock, policy, flags);
if (unlikely(err != ELDLM_OK))
GOTO(out, err);
if (policy != NULL)
*policy = lock->l_policy_data;
if (lock->l_completion_ast)
lock->l_completion_ast(lock, *flags, NULL);
LDLM_DEBUG(lock, "client-side local enqueue handler, new lock created");
EXIT;
out:
LDLM_LOCK_RELEASE(lock);
out_nolock:
return err;
}
EXPORT_SYMBOL(ldlm_cli_enqueue_local);
static void failed_lock_cleanup(struct ldlm_namespace *ns,
struct ldlm_lock *lock, int mode)
{
int need_cancel = 0;
/* Set a flag to prevent us from sending a CANCEL (bug 407) */
lock_res_and_lock(lock);
/* Check that lock is not granted or failed, we might race. */
if ((lock->l_req_mode != lock->l_granted_mode) &&
!(lock->l_flags & LDLM_FL_FAILED)) {
/* Make sure that this lock will not be found by raced
* bl_ast and -EINVAL reply is sent to server anyways.
* bug 17645 */
lock->l_flags |= LDLM_FL_LOCAL_ONLY | LDLM_FL_FAILED |
LDLM_FL_ATOMIC_CB | LDLM_FL_CBPENDING;
need_cancel = 1;
}
unlock_res_and_lock(lock);
if (need_cancel)
LDLM_DEBUG(lock,
"setting FL_LOCAL_ONLY | LDLM_FL_FAILED | "
"LDLM_FL_ATOMIC_CB | LDLM_FL_CBPENDING");
else
LDLM_DEBUG(lock, "lock was granted or failed in race");
ldlm_lock_decref_internal(lock, mode);
/* XXX - HACK because we shouldn't call ldlm_lock_destroy()
* from llite/file.c/ll_file_flock(). */
/* This code makes for the fact that we do not have blocking handler on
* a client for flock locks. As such this is the place where we must
* completely kill failed locks. (interrupted and those that
* were waiting to be granted when server evicted us. */
if (lock->l_resource->lr_type == LDLM_FLOCK) {
lock_res_and_lock(lock);
ldlm_resource_unlink_lock(lock);
ldlm_lock_destroy_nolock(lock);
unlock_res_and_lock(lock);
}
}
/**
* Finishing portion of client lock enqueue code.
*
* Called after receiving reply from server.
*/
int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
__u64 *flags, void *lvb, __u32 lvb_len,
struct lustre_handle *lockh,int rc)
{
struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
int is_replay = *flags & LDLM_FL_REPLAY;
struct ldlm_lock *lock;
struct ldlm_reply *reply;
int cleanup_phase = 1;
int size = 0;
ENTRY;
lock = ldlm_handle2lock(lockh);
/* ldlm_cli_enqueue is holding a reference on this lock. */
if (!lock) {
LASSERT(type == LDLM_FLOCK);
RETURN(-ENOLCK);
}
LASSERTF(ergo(lvb_len != 0, lvb_len == lock->l_lvb_len),
"lvb_len = %d, l_lvb_len = %d\n", lvb_len, lock->l_lvb_len);
if (rc != ELDLM_OK) {
LASSERT(!is_replay);
LDLM_DEBUG(lock, "client-side enqueue END (%s)",
rc == ELDLM_LOCK_ABORTED ? "ABORTED" : "FAILED");
if (rc != ELDLM_LOCK_ABORTED)
GOTO(cleanup, rc);
}
/* Before we return, swab the reply */
reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
if (reply == NULL)
GOTO(cleanup, rc = -EPROTO);
if (lvb_len != 0) {
LASSERT(lvb != NULL);
size = req_capsule_get_size(&req->rq_pill, &RMF_DLM_LVB,
RCL_SERVER);
if (size < 0) {
LDLM_ERROR(lock, "Fail to get lvb_len, rc = %d", size);
GOTO(cleanup, rc = size);
} else if (unlikely(size > lvb_len)) {
LDLM_ERROR(lock, "Replied LVB is larger than "
"expectation, expected = %d, replied = %d",
lvb_len, size);
GOTO(cleanup, rc = -EINVAL);
}
}
if (rc == ELDLM_LOCK_ABORTED) {
if (lvb_len != 0)
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_SERVER,
lvb, size);
GOTO(cleanup, rc = (rc != 0 ? rc : ELDLM_LOCK_ABORTED));
}
/* lock enqueued on the server */
cleanup_phase = 0;
lock_res_and_lock(lock);
/* Key change rehash lock in per-export hash with new key */
if (exp->exp_lock_hash) {
/* In the function below, .hs_keycmp resolves to
* ldlm_export_lock_keycmp() */
/* coverity[overrun-buffer-val] */
cfs_hash_rehash_key(exp->exp_lock_hash,
&lock->l_remote_handle,
&reply->lock_handle,
&lock->l_exp_hash);
} else {
lock->l_remote_handle = reply->lock_handle;
}
*flags = ldlm_flags_from_wire(reply->lock_flags);
lock->l_flags |= ldlm_flags_from_wire(reply->lock_flags &
LDLM_INHERIT_FLAGS);
/* move NO_TIMEOUT flag to the lock to force ldlm_lock_match()
* to wait with no timeout as well */
lock->l_flags |= ldlm_flags_from_wire(reply->lock_flags &
LDLM_FL_NO_TIMEOUT);
unlock_res_and_lock(lock);
CDEBUG(D_INFO, "local: %p, remote cookie: "LPX64", flags: 0x%llx\n",
lock, reply->lock_handle.cookie, *flags);
/* If enqueue returned a blocked lock but the completion handler has
* already run, then it fixed up the resource and we don't need to do it
* again. */
if ((*flags) & LDLM_FL_LOCK_CHANGED) {
int newmode = reply->lock_desc.l_req_mode;
LASSERT(!is_replay);
if (newmode && newmode != lock->l_req_mode) {
LDLM_DEBUG(lock, "server returned different mode %s",
ldlm_lockname[newmode]);
lock->l_req_mode = newmode;
}
if (memcmp(reply->lock_desc.l_resource.lr_name.name,
lock->l_resource->lr_name.name,
sizeof(struct ldlm_res_id))) {
CDEBUG(D_INFO, "remote intent success, locking "
"(%ld,%ld,%ld) instead of "
"(%ld,%ld,%ld)\n",
(long)reply->lock_desc.l_resource.lr_name.name[0],
(long)reply->lock_desc.l_resource.lr_name.name[1],
(long)reply->lock_desc.l_resource.lr_name.name[2],
(long)lock->l_resource->lr_name.name[0],
(long)lock->l_resource->lr_name.name[1],
(long)lock->l_resource->lr_name.name[2]);
rc = ldlm_lock_change_resource(ns, lock,
&reply->lock_desc.l_resource.lr_name);
if (rc || lock->l_resource == NULL)
GOTO(cleanup, rc = -ENOMEM);
LDLM_DEBUG(lock, "client-side enqueue, new resource");
}
if (with_policy)
if (!(type == LDLM_IBITS &&
!(exp_connect_flags(exp) & OBD_CONNECT_IBITS)))
/* We assume lock type cannot change on server*/
ldlm_convert_policy_to_local(exp,
lock->l_resource->lr_type,
&reply->lock_desc.l_policy_data,
&lock->l_policy_data);
if (type != LDLM_PLAIN)
LDLM_DEBUG(lock,"client-side enqueue, new policy data");
}
if ((*flags) & LDLM_FL_AST_SENT ||
/* Cancel extent locks as soon as possible on a liblustre client,
* because it cannot handle asynchronous ASTs robustly (see
* bug 7311). */
(LIBLUSTRE_CLIENT && type == LDLM_EXTENT)) {
lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_BL_AST;
unlock_res_and_lock(lock);
LDLM_DEBUG(lock, "enqueue reply includes blocking AST");
}
/* If the lock has already been granted by a completion AST, don't
* clobber the LVB with an older one. */
if (lvb_len != 0) {
/* We must lock or a racing completion might update lvb without
* letting us know and we'll clobber the correct value.
* Cannot unlock after the check either, a that still leaves
* a tiny window for completion to get in */
lock_res_and_lock(lock);
if (lock->l_req_mode != lock->l_granted_mode)
rc = ldlm_fill_lvb(lock, &req->rq_pill, RCL_SERVER,
lock->l_lvb_data, size);
unlock_res_and_lock(lock);
if (rc < 0) {
cleanup_phase = 1;
GOTO(cleanup, rc);
}
}
if (!is_replay) {
rc = ldlm_lock_enqueue(ns, &lock, NULL, flags);
if (lock->l_completion_ast != NULL) {
int err = lock->l_completion_ast(lock, *flags, NULL);
if (!rc)
rc = err;
if (rc)
cleanup_phase = 1;
}
}
if (lvb_len && lvb != NULL) {
/* Copy the LVB here, and not earlier, because the completion
* AST (if any) can override what we got in the reply */
memcpy(lvb, lock->l_lvb_data, lvb_len);
}
LDLM_DEBUG(lock, "client-side enqueue END");
EXIT;
cleanup:
if (cleanup_phase == 1 && rc)
failed_lock_cleanup(ns, lock, mode);
/* Put lock 2 times, the second reference is held by ldlm_cli_enqueue */
LDLM_LOCK_PUT(lock);
LDLM_LOCK_RELEASE(lock);
return rc;
}
EXPORT_SYMBOL(ldlm_cli_enqueue_fini);
/**
* Estimate number of lock handles that would fit into request of given
* size. PAGE_SIZE-512 is to allow TCP/IP and LNET headers to fit into
* a single page on the send/receive side. XXX: 512 should be changed to
* more adequate value.
*/
static inline int ldlm_req_handles_avail(int req_size, int off)
{
int avail;
avail = min_t(int, LDLM_MAXREQSIZE, PAGE_CACHE_SIZE - 512) - req_size;
if (likely(avail >= 0))
avail /= (int)sizeof(struct lustre_handle);
else
avail = 0;
avail += LDLM_LOCKREQ_HANDLES - off;
return avail;
}
static inline int ldlm_capsule_handles_avail(struct req_capsule *pill,
enum req_location loc,
int off)
{
int size = req_capsule_msg_size(pill, loc);
return ldlm_req_handles_avail(size, off);
}
static inline int ldlm_format_handles_avail(struct obd_import *imp,
const struct req_format *fmt,
enum req_location loc, int off)
{
int size = req_capsule_fmt_size(imp->imp_msg_magic, fmt, loc);
return ldlm_req_handles_avail(size, off);
}
/**
* Cancel LRU locks and pack them into the enqueue request. Pack there the given
* \a count locks in \a cancels.
*
* This is to be called by functions preparing their own requests that
* might contain lists of locks to cancel in addition to actual operation
* that needs to be performed.
*/
int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
int version, int opc, int canceloff,
struct list_head *cancels, int count)
{
struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
struct req_capsule *pill = &req->rq_pill;
struct ldlm_request *dlm = NULL;
int flags, avail, to_free, pack = 0;
LIST_HEAD(head);
int rc;
ENTRY;
if (cancels == NULL)
cancels = &head;
if (ns_connect_cancelset(ns)) {
/* Estimate the amount of available space in the request. */
req_capsule_filled_sizes(pill, RCL_CLIENT);
avail = ldlm_capsule_handles_avail(pill, RCL_CLIENT, canceloff);
flags = ns_connect_lru_resize(ns) ?
LDLM_CANCEL_LRUR : LDLM_CANCEL_AGED;
to_free = !ns_connect_lru_resize(ns) &&
opc == LDLM_ENQUEUE ? 1 : 0;
/* Cancel LRU locks here _only_ if the server supports
* EARLY_CANCEL. Otherwise we have to send extra CANCEL
* RPC, which will make us slower. */
if (avail > count)
count += ldlm_cancel_lru_local(ns, cancels, to_free,
avail - count, 0, flags);
if (avail > count)
pack = count;
else
pack = avail;
req_capsule_set_size(pill, &RMF_DLM_REQ, RCL_CLIENT,
ldlm_request_bufsize(pack, opc));
}
rc = ptlrpc_request_pack(req, version, opc);
if (rc) {
ldlm_lock_list_put(cancels, l_bl_ast, count);
RETURN(rc);
}
if (ns_connect_cancelset(ns)) {
if (canceloff) {
dlm = req_capsule_client_get(pill, &RMF_DLM_REQ);
LASSERT(dlm);
/* Skip first lock handler in ldlm_request_pack(),
* this method will incrment @lock_count according
* to the lock handle amount actually written to
* the buffer. */
dlm->lock_count = canceloff;
}
/* Pack into the request @pack lock handles. */
ldlm_cli_cancel_list(cancels, pack, req, 0);
/* Prepare and send separate cancel RPC for others. */
ldlm_cli_cancel_list(cancels, count - pack, NULL, 0);
} else {
ldlm_lock_list_put(cancels, l_bl_ast, count);
}
RETURN(0);
}
EXPORT_SYMBOL(ldlm_prep_elc_req);
int ldlm_prep_enqueue_req(struct obd_export *exp, struct ptlrpc_request *req,
struct list_head *cancels, int count)
{
return ldlm_prep_elc_req(exp, req, LUSTRE_DLM_VERSION, LDLM_ENQUEUE,
LDLM_ENQUEUE_CANCEL_OFF, cancels, count);
}
EXPORT_SYMBOL(ldlm_prep_enqueue_req);
struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len)
{
struct ptlrpc_request *req;
int rc;
ENTRY;
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
if (req == NULL)
RETURN(ERR_PTR(-ENOMEM));
rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
if (rc) {
ptlrpc_request_free(req);
RETURN(ERR_PTR(rc));
}
req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
ptlrpc_request_set_replen(req);
RETURN(req);
}
EXPORT_SYMBOL(ldlm_enqueue_pack);
/**
* Client-side lock enqueue.
*
* If a request has some specific initialisation it is passed in \a reqp,
* otherwise it is created in ldlm_cli_enqueue.
*
* Supports sync and async requests, pass \a async flag accordingly. If a
* request was created in ldlm_cli_enqueue and it is the async request,
* pass it to the caller in \a reqp.
*/
int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
struct ldlm_enqueue_info *einfo,
const struct ldlm_res_id *res_id,
ldlm_policy_data_t const *policy, __u64 *flags,
void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
struct lustre_handle *lockh, int async)
{
struct ldlm_namespace *ns;
struct ldlm_lock *lock;
struct ldlm_request *body;
int is_replay = *flags & LDLM_FL_REPLAY;
int req_passed_in = 1;
int rc, err;
struct ptlrpc_request *req;
ENTRY;
LASSERT(exp != NULL);
ns = exp->exp_obd->obd_namespace;
/* If we're replaying this lock, just check some invariants.
* If we're creating a new lock, get everything all setup nice. */
if (is_replay) {
lock = ldlm_handle2lock_long(lockh, 0);
LASSERT(lock != NULL);
LDLM_DEBUG(lock, "client-side enqueue START");
LASSERT(exp == lock->l_conn_export);
} else {
const struct ldlm_callback_suite cbs = {
.lcs_completion = einfo->ei_cb_cp,
.lcs_blocking = einfo->ei_cb_bl,
.lcs_glimpse = einfo->ei_cb_gl,
.lcs_weigh = einfo->ei_cb_wg
};
lock = ldlm_lock_create(ns, res_id, einfo->ei_type,
einfo->ei_mode, &cbs, einfo->ei_cbdata,
lvb_len, lvb_type);
if (lock == NULL)
RETURN(-ENOMEM);
/* for the local lock, add the reference */
ldlm_lock_addref_internal(lock, einfo->ei_mode);
ldlm_lock2handle(lock, lockh);
if (policy != NULL) {
/* INODEBITS_INTEROP: If the server does not support
* inodebits, we will request a plain lock in the
* descriptor (ldlm_lock2desc() below) but use an
* inodebits lock internally with both bits set.
*/
if (einfo->ei_type == LDLM_IBITS &&
!(exp_connect_flags(exp) &
OBD_CONNECT_IBITS))
lock->l_policy_data.l_inodebits.bits =
MDS_INODELOCK_LOOKUP |
MDS_INODELOCK_UPDATE;
else
lock->l_policy_data = *policy;
}
if (einfo->ei_type == LDLM_EXTENT)
lock->l_req_extent = policy->l_extent;
LDLM_DEBUG(lock, "client-side enqueue START, flags %llx\n",
*flags);
}
lock->l_conn_export = exp;
lock->l_export = NULL;
lock->l_blocking_ast = einfo->ei_cb_bl;
lock->l_flags |= (*flags & LDLM_FL_NO_LRU);
/* lock not sent to server yet */
if (reqp == NULL || *reqp == NULL) {
req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
&RQF_LDLM_ENQUEUE,
LUSTRE_DLM_VERSION,
LDLM_ENQUEUE);
if (req == NULL) {
failed_lock_cleanup(ns, lock, einfo->ei_mode);
LDLM_LOCK_RELEASE(lock);
RETURN(-ENOMEM);
}
req_passed_in = 0;
if (reqp)
*reqp = req;
} else {
int len;
req = *reqp;
len = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ,
RCL_CLIENT);
LASSERTF(len >= sizeof(*body), "buflen[%d] = %d, not %d\n",
DLM_LOCKREQ_OFF, len, (int)sizeof(*body));
}
/* Dump lock data into the request buffer */
body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
ldlm_lock2desc(lock, &body->lock_desc);
body->lock_flags = ldlm_flags_to_wire(*flags);
body->lock_handle[0] = *lockh;
/* Continue as normal. */
if (!req_passed_in) {
if (lvb_len > 0)
req_capsule_extend(&req->rq_pill,
&RQF_LDLM_ENQUEUE_LVB);
req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
lvb_len);
ptlrpc_request_set_replen(req);
}
/*
* Liblustre client doesn't get extent locks, except for O_APPEND case
* where [0, OBD_OBJECT_EOF] lock is taken, or truncate, where
* [i_size, OBD_OBJECT_EOF] lock is taken.
*/
LASSERT(ergo(LIBLUSTRE_CLIENT, einfo->ei_type != LDLM_EXTENT ||
policy->l_extent.end == OBD_OBJECT_EOF));
if (async) {
LASSERT(reqp != NULL);
RETURN(0);
}
LDLM_DEBUG(lock, "sending request");
rc = ptlrpc_queue_wait(req);
err = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, policy ? 1 : 0,
einfo->ei_mode, flags, lvb, lvb_len,
lockh, rc);
/* If ldlm_cli_enqueue_fini did not find the lock, we need to free
* one reference that we took */
if (err == -ENOLCK)
LDLM_LOCK_RELEASE(lock);
else
rc = err;
if (!req_passed_in && req != NULL) {
ptlrpc_req_finished(req);
if (reqp)
*reqp = NULL;
}
RETURN(rc);
}
EXPORT_SYMBOL(ldlm_cli_enqueue);
static int ldlm_cli_convert_local(struct ldlm_lock *lock, int new_mode,
__u32 *flags)
{
struct ldlm_resource *res;
int rc;
ENTRY;
if (ns_is_client(ldlm_lock_to_ns(lock))) {
CERROR("Trying to cancel local lock\n");
LBUG();
}
LDLM_DEBUG(lock, "client-side local convert");
res = ldlm_lock_convert(lock, new_mode, flags);
if (res) {
ldlm_reprocess_all(res);
rc = 0;
} else {
rc = LUSTRE_EDEADLK;
}
LDLM_DEBUG(lock, "client-side local convert handler END");
LDLM_LOCK_PUT(lock);
RETURN(rc);
}
/* FIXME: one of ldlm_cli_convert or the server side should reject attempted
* conversion of locks which are on the waiting or converting queue */
/* Caller of this code is supposed to take care of lock readers/writers
accounting */
int ldlm_cli_convert(struct lustre_handle *lockh, int new_mode, __u32 *flags)
{
struct ldlm_request *body;
struct ldlm_reply *reply;
struct ldlm_lock *lock;
struct ldlm_resource *res;
struct ptlrpc_request *req;
int rc;
ENTRY;
lock = ldlm_handle2lock(lockh);
if (!lock) {
LBUG();
RETURN(-EINVAL);
}
*flags = 0;
if (lock->l_conn_export == NULL)
RETURN(ldlm_cli_convert_local(lock, new_mode, flags));
LDLM_DEBUG(lock, "client-side convert");
req = ptlrpc_request_alloc_pack(class_exp2cliimp(lock->l_conn_export),
&RQF_LDLM_CONVERT, LUSTRE_DLM_VERSION,
LDLM_CONVERT);
if (req == NULL) {
LDLM_LOCK_PUT(lock);
RETURN(-ENOMEM);
}
body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
body->lock_handle[0] = lock->l_remote_handle;
body->lock_desc.l_req_mode = new_mode;
body->lock_flags = ldlm_flags_to_wire(*flags);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc != ELDLM_OK)
GOTO(out, rc);
reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
if (reply == NULL)
GOTO(out, rc = -EPROTO);
if (req->rq_status)
GOTO(out, rc = req->rq_status);
res = ldlm_lock_convert(lock, new_mode, &reply->lock_flags);
if (res != NULL) {
ldlm_reprocess_all(res);
/* Go to sleep until the lock is granted. */
/* FIXME: or cancelled. */
if (lock->l_completion_ast) {
rc = lock->l_completion_ast(lock, LDLM_FL_WAIT_NOREPROC,
NULL);
if (rc)
GOTO(out, rc);
}
} else {
rc = LUSTRE_EDEADLK;
}
EXIT;
out:
LDLM_LOCK_PUT(lock);
ptlrpc_req_finished(req);
return rc;
}
EXPORT_SYMBOL(ldlm_cli_convert);
/**
* Cancel locks locally.
* Returns:
* \retval LDLM_FL_LOCAL_ONLY if there is no need for a CANCEL RPC to the server
* \retval LDLM_FL_CANCELING otherwise;
* \retval LDLM_FL_BL_AST if there is a need for a separate CANCEL RPC.
*/
static __u64 ldlm_cli_cancel_local(struct ldlm_lock *lock)
{
__u64 rc = LDLM_FL_LOCAL_ONLY;
ENTRY;
if (lock->l_conn_export) {
bool local_only;
LDLM_DEBUG(lock, "client-side cancel");
/* Set this flag to prevent others from getting new references*/
lock_res_and_lock(lock);
lock->l_flags |= LDLM_FL_CBPENDING;
local_only = !!(lock->l_flags &
(LDLM_FL_LOCAL_ONLY|LDLM_FL_CANCEL_ON_BLOCK));
ldlm_cancel_callback(lock);
rc = (lock->l_flags & LDLM_FL_BL_AST) ?
LDLM_FL_BL_AST : LDLM_FL_CANCELING;
unlock_res_and_lock(lock);
if (local_only) {
CDEBUG(D_DLMTRACE, "not sending request (at caller's "
"instruction)\n");
rc = LDLM_FL_LOCAL_ONLY;
}
ldlm_lock_cancel(lock);
} else {
if (ns_is_client(ldlm_lock_to_ns(lock))) {
LDLM_ERROR(lock, "Trying to cancel local lock");
LBUG();
}
LDLM_DEBUG(lock, "server-side local cancel");
ldlm_lock_cancel(lock);
ldlm_reprocess_all(lock->l_resource);
}
RETURN(rc);
}
/**
* Pack \a count locks in \a head into ldlm_request buffer of request \a req.
*/
static void ldlm_cancel_pack(struct ptlrpc_request *req,
struct list_head *head, int count)
{
struct ldlm_request *dlm;
struct ldlm_lock *lock;
int max, packed = 0;
ENTRY;
dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
LASSERT(dlm != NULL);
/* Check the room in the request buffer. */
max = req_capsule_get_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT) -
sizeof(struct ldlm_request);
max /= sizeof(struct lustre_handle);
max += LDLM_LOCKREQ_HANDLES;
LASSERT(max >= dlm->lock_count + count);
/* XXX: it would be better to pack lock handles grouped by resource.
* so that the server cancel would call filter_lvbo_update() less
* frequently. */
list_for_each_entry(lock, head, l_bl_ast) {
if (!count--)
break;
LASSERT(lock->l_conn_export);
/* Pack the lock handle to the given request buffer. */
LDLM_DEBUG(lock, "packing");
dlm->lock_handle[dlm->lock_count++] = lock->l_remote_handle;
packed++;
}
CDEBUG(D_DLMTRACE, "%d locks packed\n", packed);
EXIT;
}
/**
* Prepare and send a batched cancel RPC. It will include \a count lock
* handles of locks given in \a cancels list. */
int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *cancels,
int count, ldlm_cancel_flags_t flags)
{
struct ptlrpc_request *req = NULL;
struct obd_import *imp;
int free, sent = 0;
int rc = 0;
ENTRY;
LASSERT(exp != NULL);
LASSERT(count > 0);
CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_PAUSE_CANCEL, cfs_fail_val);
if (CFS_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_RACE))
RETURN(count);
free = ldlm_format_handles_avail(class_exp2cliimp(exp),
&RQF_LDLM_CANCEL, RCL_CLIENT, 0);
if (count > free)
count = free;
while (1) {
imp = class_exp2cliimp(exp);
if (imp == NULL || imp->imp_invalid) {
CDEBUG(D_DLMTRACE,
"skipping cancel on invalid import %p\n", imp);
RETURN(count);
}
req = ptlrpc_request_alloc(imp, &RQF_LDLM_CANCEL);
if (req == NULL)
GOTO(out, rc = -ENOMEM);
req_capsule_filled_sizes(&req->rq_pill, RCL_CLIENT);
req_capsule_set_size(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT,
ldlm_request_bufsize(count, LDLM_CANCEL));
rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_CANCEL);
if (rc) {
ptlrpc_request_free(req);
GOTO(out, rc);
}
req->rq_request_portal = LDLM_CANCEL_REQUEST_PORTAL;
req->rq_reply_portal = LDLM_CANCEL_REPLY_PORTAL;
ptlrpc_at_set_req_timeout(req);
ldlm_cancel_pack(req, cancels, count);
ptlrpc_request_set_replen(req);
if (flags & LCF_ASYNC) {
ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
sent = count;
GOTO(out, 0);
} else {
rc = ptlrpc_queue_wait(req);
}
if (rc == LUSTRE_ESTALE) {
CDEBUG(D_DLMTRACE, "client/server (nid %s) "
"out of sync -- not fatal\n",
libcfs_nid2str(req->rq_import->
imp_connection->c_peer.nid));
rc = 0;
} else if (rc == -ETIMEDOUT && /* check there was no reconnect*/
req->rq_import_generation == imp->imp_generation) {
ptlrpc_req_finished(req);
continue;
} else if (rc != ELDLM_OK) {
/* -ESHUTDOWN is common on umount */
CDEBUG_LIMIT(rc == -ESHUTDOWN ? D_DLMTRACE : D_ERROR,
"Got rc %d from cancel RPC: "
"canceling anyway\n", rc);
break;
}
sent = count;
break;
}
ptlrpc_req_finished(req);
EXIT;
out:
return sent ? sent : rc;
}
EXPORT_SYMBOL(ldlm_cli_cancel_req);
static inline struct ldlm_pool *ldlm_imp2pl(struct obd_import *imp)
{
LASSERT(imp != NULL);
return &imp->imp_obd->obd_namespace->ns_pool;
}
/**
* Update client's OBD pool related fields with new SLV and Limit from \a req.
*/
int ldlm_cli_update_pool(struct ptlrpc_request *req)
{
struct obd_device *obd;
__u64 new_slv;
__u32 new_limit;
ENTRY;
if (unlikely(!req->rq_import || !req->rq_import->imp_obd ||
!imp_connect_lru_resize(req->rq_import)))
{
/*
* Do nothing for corner cases.
*/
RETURN(0);
}
/* In some cases RPC may contain SLV and limit zeroed out. This
* is the case when server does not support LRU resize feature.
* This is also possible in some recovery cases when server-side
* reqs have no reference to the OBD export and thus access to
* server-side namespace is not possible. */
if (lustre_msg_get_slv(req->rq_repmsg) == 0 ||
lustre_msg_get_limit(req->rq_repmsg) == 0) {
DEBUG_REQ(D_HA, req, "Zero SLV or Limit found "
"(SLV: "LPU64", Limit: %u)",
lustre_msg_get_slv(req->rq_repmsg),
lustre_msg_get_limit(req->rq_repmsg));
RETURN(0);
}
new_limit = lustre_msg_get_limit(req->rq_repmsg);
new_slv = lustre_msg_get_slv(req->rq_repmsg);
obd = req->rq_import->imp_obd;
/* Set new SLV and limit in OBD fields to make them accessible
* to the pool thread. We do not access obd_namespace and pool
* directly here as there is no reliable way to make sure that
* they are still alive at cleanup time. Evil races are possible
* which may cause Oops at that time. */
write_lock(&obd->obd_pool_lock);
obd->obd_pool_slv = new_slv;
obd->obd_pool_limit = new_limit;
write_unlock(&obd->obd_pool_lock);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_update_pool);
/**
* Client side lock cancel.
*
* Lock must not have any readers or writers by this time.
*/
int ldlm_cli_cancel(struct lustre_handle *lockh,
ldlm_cancel_flags_t cancel_flags)
{
struct obd_export *exp;
int avail, flags, count = 1;
__u64 rc = 0;
struct ldlm_namespace *ns;
struct ldlm_lock *lock;
LIST_HEAD(cancels);
ENTRY;
/* concurrent cancels on the same handle can happen */
lock = ldlm_handle2lock_long(lockh, LDLM_FL_CANCELING);
if (lock == NULL) {
LDLM_DEBUG_NOLOCK("lock is already being destroyed\n");
RETURN(0);
}
rc = ldlm_cli_cancel_local(lock);
if (rc == LDLM_FL_LOCAL_ONLY) {
LDLM_LOCK_RELEASE(lock);
RETURN(0);
}
/* Even if the lock is marked as LDLM_FL_BL_AST, this is a LDLM_CANCEL
* RPC which goes to canceld portal, so we can cancel other LRU locks
* here and send them all as one LDLM_CANCEL RPC. */
LASSERT(list_empty(&lock->l_bl_ast));
list_add(&lock->l_bl_ast, &cancels);
exp = lock->l_conn_export;
if (exp_connect_cancelset(exp)) {
avail = ldlm_format_handles_avail(class_exp2cliimp(exp),
&RQF_LDLM_CANCEL,
RCL_CLIENT, 0);
LASSERT(avail > 0);
ns = ldlm_lock_to_ns(lock);
flags = ns_connect_lru_resize(ns) ?
LDLM_CANCEL_LRUR : LDLM_CANCEL_AGED;
count += ldlm_cancel_lru_local(ns, &cancels, 0, avail - 1,
LCF_BL_AST, flags);
}
ldlm_cli_cancel_list(&cancels, count, NULL, cancel_flags);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_cancel);
/**
* Locally cancel up to \a count locks in list \a cancels.
* Return the number of cancelled locks.
*/
int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
ldlm_cancel_flags_t flags)
{
LIST_HEAD(head);
struct ldlm_lock *lock, *next;
int left = 0, bl_ast = 0;
__u64 rc;
left = count;
list_for_each_entry_safe(lock, next, cancels, l_bl_ast) {
if (left-- == 0)
break;
if (flags & LCF_LOCAL) {
rc = LDLM_FL_LOCAL_ONLY;
ldlm_lock_cancel(lock);
} else {
rc = ldlm_cli_cancel_local(lock);
}
/* Until we have compound requests and can send LDLM_CANCEL
* requests batched with generic RPCs, we need to send cancels
* with the LDLM_FL_BL_AST flag in a separate RPC from
* the one being generated now. */
if (!(flags & LCF_BL_AST) && (rc == LDLM_FL_BL_AST)) {
LDLM_DEBUG(lock, "Cancel lock separately");
list_del_init(&lock->l_bl_ast);
list_add(&lock->l_bl_ast, &head);
bl_ast++;
continue;
}
if (rc == LDLM_FL_LOCAL_ONLY) {
/* CANCEL RPC should not be sent to server. */
list_del_init(&lock->l_bl_ast);
LDLM_LOCK_RELEASE(lock);
count--;
}
}
if (bl_ast > 0) {
count -= bl_ast;
ldlm_cli_cancel_list(&head, bl_ast, NULL, 0);
}
RETURN(count);
}
EXPORT_SYMBOL(ldlm_cli_cancel_list_local);
/**
* Cancel as many locks as possible w/o sending any RPCs (e.g. to write back
* dirty data, to close a file, ...) or waiting for any RPCs in-flight (e.g.
* readahead requests, ...)
*/
static ldlm_policy_res_t ldlm_cancel_no_wait_policy(struct ldlm_namespace *ns,
struct ldlm_lock *lock,
int unused, int added,
int count)
{
ldlm_policy_res_t result = LDLM_POLICY_CANCEL_LOCK;
ldlm_cancel_for_recovery cb = ns->ns_cancel_for_recovery;
lock_res_and_lock(lock);
/* don't check added & count since we want to process all locks
* from unused list */
switch (lock->l_resource->lr_type) {
case LDLM_EXTENT:
case LDLM_IBITS:
if (cb && cb(lock))
break;
default:
result = LDLM_POLICY_SKIP_LOCK;
lock->l_flags |= LDLM_FL_SKIPPED;
break;
}
unlock_res_and_lock(lock);
RETURN(result);
}
/**
* Callback function for LRU-resize policy. Decides whether to keep
* \a lock in LRU for current \a LRU size \a unused, added in current
* scan \a added and number of locks to be preferably canceled \a count.
*
* \retval LDLM_POLICY_KEEP_LOCK keep lock in LRU in stop scanning
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
static ldlm_policy_res_t ldlm_cancel_lrur_policy(struct ldlm_namespace *ns,
struct ldlm_lock *lock,
int unused, int added,
int count)
{
cfs_time_t cur = cfs_time_current();
struct ldlm_pool *pl = &ns->ns_pool;
__u64 slv, lvf, lv;
cfs_time_t la;
/* Stop LRU processing when we reach past @count or have checked all
* locks in LRU. */
if (count && added >= count)
return LDLM_POLICY_KEEP_LOCK;
slv = ldlm_pool_get_slv(pl);
lvf = ldlm_pool_get_lvf(pl);
la = cfs_duration_sec(cfs_time_sub(cur,
lock->l_last_used));
lv = lvf * la * unused;
/* Inform pool about current CLV to see it via proc. */
ldlm_pool_set_clv(pl, lv);
/* Stop when SLV is not yet come from server or lv is smaller than
* it is. */
return (slv == 0 || lv < slv) ?
LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
}
/**
* Callback function for proc used policy. Makes decision whether to keep
* \a lock in LRU for current \a LRU size \a unused, added in current scan \a
* added and number of locks to be preferably canceled \a count.
*
* \retval LDLM_POLICY_KEEP_LOCK keep lock in LRU in stop scanning
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
static ldlm_policy_res_t ldlm_cancel_passed_policy(struct ldlm_namespace *ns,
struct ldlm_lock *lock,
int unused, int added,
int count)
{
/* Stop LRU processing when we reach past @count or have checked all
* locks in LRU. */
return (added >= count) ?
LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
}
/**
* Callback function for aged policy. Makes decision whether to keep \a lock in
* LRU for current LRU size \a unused, added in current scan \a added and
* number of locks to be preferably canceled \a count.
*
* \retval LDLM_POLICY_KEEP_LOCK keep lock in LRU in stop scanning
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
static ldlm_policy_res_t ldlm_cancel_aged_policy(struct ldlm_namespace *ns,
struct ldlm_lock *lock,
int unused, int added,
int count)
{
/* Stop LRU processing if young lock is found and we reach past count */
return ((added >= count) &&
cfs_time_before(cfs_time_current(),
cfs_time_add(lock->l_last_used,
ns->ns_max_age))) ?
LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
}
/**
* Callback function for default policy. Makes decision whether to keep \a lock
* in LRU for current LRU size \a unused, added in current scan \a added and
* number of locks to be preferably canceled \a count.
*
* \retval LDLM_POLICY_KEEP_LOCK keep lock in LRU in stop scanning
*
* \retval LDLM_POLICY_CANCEL_LOCK cancel lock from LRU
*/
static ldlm_policy_res_t ldlm_cancel_default_policy(struct ldlm_namespace *ns,
struct ldlm_lock *lock,
int unused, int added,
int count)
{
/* Stop LRU processing when we reach past count or have checked all
* locks in LRU. */
return (added >= count) ?
LDLM_POLICY_KEEP_LOCK : LDLM_POLICY_CANCEL_LOCK;
}
typedef ldlm_policy_res_t (*ldlm_cancel_lru_policy_t)(struct ldlm_namespace *,
struct ldlm_lock *, int,
int, int);
static ldlm_cancel_lru_policy_t
ldlm_cancel_lru_policy(struct ldlm_namespace *ns, int flags)
{
if (flags & LDLM_CANCEL_NO_WAIT)
return ldlm_cancel_no_wait_policy;
if (ns_connect_lru_resize(ns)) {
if (flags & LDLM_CANCEL_SHRINK)
/* We kill passed number of old locks. */
return ldlm_cancel_passed_policy;
else if (flags & LDLM_CANCEL_LRUR)
return ldlm_cancel_lrur_policy;
else if (flags & LDLM_CANCEL_PASSED)
return ldlm_cancel_passed_policy;
} else {
if (flags & LDLM_CANCEL_AGED)
return ldlm_cancel_aged_policy;
}
return ldlm_cancel_default_policy;
}
/**
* - Free space in LRU for \a count new locks,
* redundant unused locks are canceled locally;
* - also cancel locally unused aged locks;
* - do not cancel more than \a max locks;
* - GET the found locks and add them into the \a cancels list.
*
* A client lock can be added to the l_bl_ast list only when it is
* marked LDLM_FL_CANCELING. Otherwise, somebody is already doing
* CANCEL. There are the following use cases:
* ldlm_cancel_resource_local(), ldlm_cancel_lru_local() and
* ldlm_cli_cancel(), which check and set this flag properly. As any
* attempt to cancel a lock rely on this flag, l_bl_ast list is accessed
* later without any special locking.
*
* Calling policies for enabled LRU resize:
* ----------------------------------------
* flags & LDLM_CANCEL_LRUR - use LRU resize policy (SLV from server) to
* cancel not more than \a count locks;
*
* flags & LDLM_CANCEL_PASSED - cancel \a count number of old locks (located at
* the beginning of LRU list);
*
* flags & LDLM_CANCEL_SHRINK - cancel not more than \a count locks according to
* memory pressre policy function;
*
* flags & LDLM_CANCEL_AGED - cancel \a count locks according to "aged policy".
*
* flags & LDLM_CANCEL_NO_WAIT - cancel as many unused locks as possible
* (typically before replaying locks) w/o
* sending any RPCs or waiting for any
* outstanding RPC to complete.
*/
static int ldlm_prepare_lru_list(struct ldlm_namespace *ns, struct list_head *cancels,
int count, int max, int flags)
{
ldlm_cancel_lru_policy_t pf;
struct ldlm_lock *lock, *next;
int added = 0, unused, remained;
ENTRY;
spin_lock(&ns->ns_lock);
unused = ns->ns_nr_unused;
remained = unused;
if (!ns_connect_lru_resize(ns))
count += unused - ns->ns_max_unused;
pf = ldlm_cancel_lru_policy(ns, flags);
LASSERT(pf != NULL);
while (!list_empty(&ns->ns_unused_list)) {
ldlm_policy_res_t result;
/* all unused locks */
if (remained-- <= 0)
break;
/* For any flags, stop scanning if @max is reached. */
if (max && added >= max)
break;
list_for_each_entry_safe(lock, next, &ns->ns_unused_list,
l_lru) {
/* No locks which got blocking requests. */
LASSERT(!(lock->l_flags & LDLM_FL_BL_AST));
if (flags & LDLM_CANCEL_NO_WAIT &&
lock->l_flags & LDLM_FL_SKIPPED)
/* already processed */
continue;
/* Somebody is already doing CANCEL. No need for this
* lock in LRU, do not traverse it again. */
if (!(lock->l_flags & LDLM_FL_CANCELING))
break;
ldlm_lock_remove_from_lru_nolock(lock);
}
if (&lock->l_lru == &ns->ns_unused_list)
break;
LDLM_LOCK_GET(lock);
spin_unlock(&ns->ns_lock);
lu_ref_add(&lock->l_reference, __FUNCTION__, current);
/* Pass the lock through the policy filter and see if it
* should stay in LRU.
*
* Even for shrinker policy we stop scanning if
* we find a lock that should stay in the cache.
* We should take into account lock age anyway
* as a new lock is a valuable resource even if
* it has a low weight.
*
* That is, for shrinker policy we drop only
* old locks, but additionally choose them by
* their weight. Big extent locks will stay in
* the cache. */
result = pf(ns, lock, unused, added, count);
if (result == LDLM_POLICY_KEEP_LOCK) {
lu_ref_del(&lock->l_reference,
__FUNCTION__, current);
LDLM_LOCK_RELEASE(lock);
spin_lock(&ns->ns_lock);
break;
}
if (result == LDLM_POLICY_SKIP_LOCK) {
lu_ref_del(&lock->l_reference,
__func__, current);
LDLM_LOCK_RELEASE(lock);
spin_lock(&ns->ns_lock);
continue;
}
lock_res_and_lock(lock);
/* Check flags again under the lock. */
if ((lock->l_flags & LDLM_FL_CANCELING) ||
(ldlm_lock_remove_from_lru(lock) == 0)) {
/* Another thread is removing lock from LRU, or
* somebody is already doing CANCEL, or there
* is a blocking request which will send cancel
* by itself, or the lock is no longer unused. */
unlock_res_and_lock(lock);
lu_ref_del(&lock->l_reference,
__FUNCTION__, current);
LDLM_LOCK_RELEASE(lock);
spin_lock(&ns->ns_lock);
continue;
}
LASSERT(!lock->l_readers && !lock->l_writers);
/* If we have chosen to cancel this lock voluntarily, we
* better send cancel notification to server, so that it
* frees appropriate state. This might lead to a race
* where while we are doing cancel here, server is also
* silently cancelling this lock. */
lock->l_flags &= ~LDLM_FL_CANCEL_ON_BLOCK;
/* Setting the CBPENDING flag is a little misleading,
* but prevents an important race; namely, once
* CBPENDING is set, the lock can accumulate no more
* readers/writers. Since readers and writers are
* already zero here, ldlm_lock_decref() won't see
* this flag and call l_blocking_ast */
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_CANCELING;
/* We can't re-add to l_lru as it confuses the
* refcounting in ldlm_lock_remove_from_lru() if an AST
* arrives after we drop lr_lock below. We use l_bl_ast
* and can't use l_pending_chain as it is used both on
* server and client nevertheless bug 5666 says it is
* used only on server */
LASSERT(list_empty(&lock->l_bl_ast));
list_add(&lock->l_bl_ast, cancels);
unlock_res_and_lock(lock);
lu_ref_del(&lock->l_reference, __FUNCTION__, current);
spin_lock(&ns->ns_lock);
added++;
unused--;
}
spin_unlock(&ns->ns_lock);
RETURN(added);
}
int ldlm_cancel_lru_local(struct ldlm_namespace *ns, struct list_head *cancels,
int count, int max, ldlm_cancel_flags_t cancel_flags,
int flags)
{
int added;
added = ldlm_prepare_lru_list(ns, cancels, count, max, flags);
if (added <= 0)
return added;
return ldlm_cli_cancel_list_local(cancels, added, cancel_flags);
}
/**
* Cancel at least \a nr locks from given namespace LRU.
*
* When called with LCF_ASYNC the blocking callback will be handled
* in a thread and this function will return after the thread has been
* asked to call the callback. When called with LCF_ASYNC the blocking
* callback will be performed in this function.
*/
int ldlm_cancel_lru(struct ldlm_namespace *ns, int nr,
ldlm_cancel_flags_t cancel_flags,
int flags)
{
LIST_HEAD(cancels);
int count, rc;
ENTRY;
/* Just prepare the list of locks, do not actually cancel them yet.
* Locks are cancelled later in a separate thread. */
count = ldlm_prepare_lru_list(ns, &cancels, nr, 0, flags);
rc = ldlm_bl_to_thread_list(ns, NULL, &cancels, count, cancel_flags);
if (rc == 0)
RETURN(count);
RETURN(0);
}
/**
* Find and cancel locally unused locks found on resource, matched to the
* given policy, mode. GET the found locks and add them into the \a cancels
* list.
*/
int ldlm_cancel_resource_local(struct ldlm_resource *res,
struct list_head *cancels,
ldlm_policy_data_t *policy,
ldlm_mode_t mode, int lock_flags,
ldlm_cancel_flags_t cancel_flags, void *opaque)
{
struct ldlm_lock *lock;
int count = 0;
ENTRY;
lock_res(res);
list_for_each_entry(lock, &res->lr_granted, l_res_link) {
if (opaque != NULL && lock->l_ast_data != opaque) {
LDLM_ERROR(lock, "data %p doesn't match opaque %p",
lock->l_ast_data, opaque);
//LBUG();
continue;
}
if (lock->l_readers || lock->l_writers)
continue;
/* If somebody is already doing CANCEL, or blocking AST came,
* skip this lock. */
if (lock->l_flags & LDLM_FL_BL_AST ||
lock->l_flags & LDLM_FL_CANCELING)
continue;
if (lockmode_compat(lock->l_granted_mode, mode))
continue;
/* If policy is given and this is IBITS lock, add to list only
* those locks that match by policy. */
if (policy && (lock->l_resource->lr_type == LDLM_IBITS) &&
!(lock->l_policy_data.l_inodebits.bits &
policy->l_inodebits.bits))
continue;
/* See CBPENDING comment in ldlm_cancel_lru */
lock->l_flags |= LDLM_FL_CBPENDING | LDLM_FL_CANCELING |
lock_flags;
LASSERT(list_empty(&lock->l_bl_ast));
list_add(&lock->l_bl_ast, cancels);
LDLM_LOCK_GET(lock);
count++;
}
unlock_res(res);
RETURN(ldlm_cli_cancel_list_local(cancels, count, cancel_flags));
}
EXPORT_SYMBOL(ldlm_cancel_resource_local);
/**
* Cancel client-side locks from a list and send/prepare cancel RPCs to the
* server.
* If \a req is NULL, send CANCEL request to server with handles of locks
* in the \a cancels. If EARLY_CANCEL is not supported, send CANCEL requests
* separately per lock.
* If \a req is not NULL, put handles of locks in \a cancels into the request
* buffer at the offset \a off.
* Destroy \a cancels at the end.
*/
int ldlm_cli_cancel_list(struct list_head *cancels, int count,
struct ptlrpc_request *req, ldlm_cancel_flags_t flags)
{
struct ldlm_lock *lock;
int res = 0;
ENTRY;
if (list_empty(cancels) || count == 0)
RETURN(0);
/* XXX: requests (both batched and not) could be sent in parallel.
* Usually it is enough to have just 1 RPC, but it is possible that
* there are too many locks to be cancelled in LRU or on a resource.
* It would also speed up the case when the server does not support
* the feature. */
while (count > 0) {
LASSERT(!list_empty(cancels));
lock = list_entry(cancels->next, struct ldlm_lock,
l_bl_ast);
LASSERT(lock->l_conn_export);
if (exp_connect_cancelset(lock->l_conn_export)) {
res = count;
if (req)
ldlm_cancel_pack(req, cancels, count);
else
res = ldlm_cli_cancel_req(lock->l_conn_export,
cancels, count,
flags);
} else {
res = ldlm_cli_cancel_req(lock->l_conn_export,
cancels, 1, flags);
}
if (res < 0) {
CDEBUG_LIMIT(res == -ESHUTDOWN ? D_DLMTRACE : D_ERROR,
"ldlm_cli_cancel_list: %d\n", res);
res = count;
}
count -= res;
ldlm_lock_list_put(cancels, l_bl_ast, res);
}
LASSERT(count == 0);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_cancel_list);
/**
* Cancel all locks on a resource that have 0 readers/writers.
*
* If flags & LDLM_FL_LOCAL_ONLY, throw the locks away without trying
* to notify the server. */
int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
const struct ldlm_res_id *res_id,
ldlm_policy_data_t *policy,
ldlm_mode_t mode,
ldlm_cancel_flags_t flags,
void *opaque)
{
struct ldlm_resource *res;
LIST_HEAD(cancels);
int count;
int rc;
ENTRY;
res = ldlm_resource_get(ns, NULL, res_id, 0, 0);
if (res == NULL) {
/* This is not a problem. */
CDEBUG(D_INFO, "No resource "LPU64"\n", res_id->name[0]);
RETURN(0);
}
LDLM_RESOURCE_ADDREF(res);
count = ldlm_cancel_resource_local(res, &cancels, policy, mode,
0, flags | LCF_BL_AST, opaque);
rc = ldlm_cli_cancel_list(&cancels, count, NULL, flags);
if (rc != ELDLM_OK)
CERROR("ldlm_cli_cancel_unused_resource: %d\n", rc);
LDLM_RESOURCE_DELREF(res);
ldlm_resource_putref(res);
RETURN(0);
}
EXPORT_SYMBOL(ldlm_cli_cancel_unused_resource);
struct ldlm_cli_cancel_arg {
int lc_flags;
void *lc_opaque;
};
static int ldlm_cli_hash_cancel_unused(cfs_hash_t *hs, cfs_hash_bd_t *bd,
struct hlist_node *hnode, void *arg)
{
struct ldlm_resource *res = cfs_hash_object(hs, hnode);
struct ldlm_cli_cancel_arg *lc = arg;
int rc;
rc = ldlm_cli_cancel_unused_resource(ldlm_res_to_ns(res), &res->lr_name,
NULL, LCK_MINMODE,
lc->lc_flags, lc->lc_opaque);
if (rc != 0) {
CERROR("ldlm_cli_cancel_unused ("LPU64"): %d\n",
res->lr_name.name[0], rc);
}
/* must return 0 for hash iteration */
return 0;
}
/**
* Cancel all locks on a namespace (or a specific resource, if given)
* that have 0 readers/writers.
*
* If flags & LCF_LOCAL, throw the locks away without trying
* to notify the server. */
int ldlm_cli_cancel_unused(struct ldlm_namespace *ns,
const struct ldlm_res_id *res_id,
ldlm_cancel_flags_t flags, void *opaque)
{
struct ldlm_cli_cancel_arg arg = {
.lc_flags = flags,
.lc_opaque = opaque,
};
ENTRY;
if (ns == NULL)
RETURN(ELDLM_OK);
if (res_id != NULL) {
RETURN(ldlm_cli_cancel_unused_resource(ns, res_id, NULL,
LCK_MINMODE, flags,
opaque));
} else {
cfs_hash_for_each_nolock(ns->ns_rs_hash,
ldlm_cli_hash_cancel_unused, &arg);
RETURN(ELDLM_OK);
}
}
EXPORT_SYMBOL(ldlm_cli_cancel_unused);
/* Lock iterators. */
int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
void *closure)
{
struct list_head *tmp, *next;
struct ldlm_lock *lock;
int rc = LDLM_ITER_CONTINUE;
ENTRY;
if (!res)
RETURN(LDLM_ITER_CONTINUE);
lock_res(res);
list_for_each_safe(tmp, next, &res->lr_granted) {
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
if (iter(lock, closure) == LDLM_ITER_STOP)
GOTO(out, rc = LDLM_ITER_STOP);
}
list_for_each_safe(tmp, next, &res->lr_converting) {
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
if (iter(lock, closure) == LDLM_ITER_STOP)
GOTO(out, rc = LDLM_ITER_STOP);
}
list_for_each_safe(tmp, next, &res->lr_waiting) {
lock = list_entry(tmp, struct ldlm_lock, l_res_link);
if (iter(lock, closure) == LDLM_ITER_STOP)
GOTO(out, rc = LDLM_ITER_STOP);
}
out:
unlock_res(res);
RETURN(rc);
}
EXPORT_SYMBOL(ldlm_resource_foreach);
struct iter_helper_data {
ldlm_iterator_t iter;
void *closure;
};
static int ldlm_iter_helper(struct ldlm_lock *lock, void *closure)
{
struct iter_helper_data *helper = closure;
return helper->iter(lock, helper->closure);
}
static int ldlm_res_iter_helper(cfs_hash_t *hs, cfs_hash_bd_t *bd,
struct hlist_node *hnode, void *arg)
{
struct ldlm_resource *res = cfs_hash_object(hs, hnode);
return ldlm_resource_foreach(res, ldlm_iter_helper, arg) ==
LDLM_ITER_STOP;
}
void ldlm_namespace_foreach(struct ldlm_namespace *ns,
ldlm_iterator_t iter, void *closure)
{
struct iter_helper_data helper = { iter: iter, closure: closure };
cfs_hash_for_each_nolock(ns->ns_rs_hash,
ldlm_res_iter_helper, &helper);
}
EXPORT_SYMBOL(ldlm_namespace_foreach);
/* non-blocking function to manipulate a lock whose cb_data is being put away.
* return 0: find no resource
* > 0: must be LDLM_ITER_STOP/LDLM_ITER_CONTINUE.
* < 0: errors
*/
int ldlm_resource_iterate(struct ldlm_namespace *ns,
const struct ldlm_res_id *res_id,
ldlm_iterator_t iter, void *data)
{
struct ldlm_resource *res;
int rc;
ENTRY;
if (ns == NULL) {
CERROR("must pass in namespace\n");
LBUG();
}
res = ldlm_resource_get(ns, NULL, res_id, 0, 0);
if (res == NULL)
RETURN(0);
LDLM_RESOURCE_ADDREF(res);
rc = ldlm_resource_foreach(res, iter, data);
LDLM_RESOURCE_DELREF(res);
ldlm_resource_putref(res);
RETURN(rc);
}
EXPORT_SYMBOL(ldlm_resource_iterate);
/* Lock replay */
static int ldlm_chain_lock_for_replay(struct ldlm_lock *lock, void *closure)
{
struct list_head *list = closure;
/* we use l_pending_chain here, because it's unused on clients. */
LASSERTF(list_empty(&lock->l_pending_chain),
"lock %p next %p prev %p\n",
lock, &lock->l_pending_chain.next,&lock->l_pending_chain.prev);
/* bug 9573: don't replay locks left after eviction, or
* bug 17614: locks being actively cancelled. Get a reference
* on a lock so that it does not disapear under us (e.g. due to cancel)
*/
if (!(lock->l_flags & (LDLM_FL_FAILED|LDLM_FL_CANCELING))) {
list_add(&lock->l_pending_chain, list);
LDLM_LOCK_GET(lock);
}
return LDLM_ITER_CONTINUE;
}
static int replay_lock_interpret(const struct lu_env *env,
struct ptlrpc_request *req,
struct ldlm_async_args *aa, int rc)
{
struct ldlm_lock *lock;
struct ldlm_reply *reply;
struct obd_export *exp;
ENTRY;
atomic_dec(&req->rq_import->imp_replay_inflight);
if (rc != ELDLM_OK)
GOTO(out, rc);
reply = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
if (reply == NULL)
GOTO(out, rc = -EPROTO);
lock = ldlm_handle2lock(&aa->lock_handle);
if (!lock) {
CERROR("received replay ack for unknown local cookie "LPX64
" remote cookie "LPX64 " from server %s id %s\n",
aa->lock_handle.cookie, reply->lock_handle.cookie,
req->rq_export->exp_client_uuid.uuid,
libcfs_id2str(req->rq_peer));
GOTO(out, rc = -ESTALE);
}
/* Key change rehash lock in per-export hash with new key */
exp = req->rq_export;
if (exp && exp->exp_lock_hash) {
/* In the function below, .hs_keycmp resolves to
* ldlm_export_lock_keycmp() */
/* coverity[overrun-buffer-val] */
cfs_hash_rehash_key(exp->exp_lock_hash,
&lock->l_remote_handle,
&reply->lock_handle,
&lock->l_exp_hash);
} else {
lock->l_remote_handle = reply->lock_handle;
}
LDLM_DEBUG(lock, "replayed lock:");
ptlrpc_import_recovery_state_machine(req->rq_import);
LDLM_LOCK_PUT(lock);
out:
if (rc != ELDLM_OK)
ptlrpc_connect_import(req->rq_import);
RETURN(rc);
}
static int replay_one_lock(struct obd_import *imp, struct ldlm_lock *lock)
{
struct ptlrpc_request *req;
struct ldlm_async_args *aa;
struct ldlm_request *body;
int flags;
ENTRY;
/* Bug 11974: Do not replay a lock which is actively being canceled */
if (lock->l_flags & LDLM_FL_CANCELING) {
LDLM_DEBUG(lock, "Not replaying canceled lock:");
RETURN(0);
}
/* If this is reply-less callback lock, we cannot replay it, since
* server might have long dropped it, but notification of that event was
* lost by network. (and server granted conflicting lock already) */
if (lock->l_flags & LDLM_FL_CANCEL_ON_BLOCK) {
LDLM_DEBUG(lock, "Not replaying reply-less lock:");
ldlm_lock_cancel(lock);
RETURN(0);
}
/*
* If granted mode matches the requested mode, this lock is granted.
*
* If they differ, but we have a granted mode, then we were granted
* one mode and now want another: ergo, converting.
*
* If we haven't been granted anything and are on a resource list,
* then we're blocked/waiting.
*
* If we haven't been granted anything and we're NOT on a resource list,
* then we haven't got a reply yet and don't have a known disposition.
* This happens whenever a lock enqueue is the request that triggers
* recovery.
*/
if (lock->l_granted_mode == lock->l_req_mode)
flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_GRANTED;
else if (lock->l_granted_mode)
flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_CONV;
else if (!list_empty(&lock->l_res_link))
flags = LDLM_FL_REPLAY | LDLM_FL_BLOCK_WAIT;
else
flags = LDLM_FL_REPLAY;
req = ptlrpc_request_alloc_pack(imp, &RQF_LDLM_ENQUEUE,
LUSTRE_DLM_VERSION, LDLM_ENQUEUE);
if (req == NULL)
RETURN(-ENOMEM);
/* We're part of recovery, so don't wait for it. */
req->rq_send_state = LUSTRE_IMP_REPLAY_LOCKS;
body = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
ldlm_lock2desc(lock, &body->lock_desc);
body->lock_flags = ldlm_flags_to_wire(flags);
ldlm_lock2handle(lock, &body->lock_handle[0]);
if (lock->l_lvb_len > 0)
req_capsule_extend(&req->rq_pill, &RQF_LDLM_ENQUEUE_LVB);
req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
lock->l_lvb_len);
ptlrpc_request_set_replen(req);
/* notify the server we've replayed all requests.
* also, we mark the request to be put on a dedicated
* queue to be processed after all request replayes.
* bug 6063 */
lustre_msg_set_flags(req->rq_reqmsg, MSG_REQ_REPLAY_DONE);
LDLM_DEBUG(lock, "replaying lock:");
atomic_inc(&req->rq_import->imp_replay_inflight);
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
aa->lock_handle = body->lock_handle[0];
req->rq_interpret_reply = (ptlrpc_interpterer_t)replay_lock_interpret;
ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
RETURN(0);
}
/**
* Cancel as many unused locks as possible before replay. since we are
* in recovery, we can't wait for any outstanding RPCs to send any RPC
* to the server.
*
* Called only in recovery before replaying locks. there is no need to
* replay locks that are unused. since the clients may hold thousands of
* cached unused locks, dropping the unused locks can greatly reduce the
* load on the servers at recovery time.
*/
static void ldlm_cancel_unused_locks_for_replay(struct ldlm_namespace *ns)
{
int canceled;
LIST_HEAD(cancels);
CDEBUG(D_DLMTRACE, "Dropping as many unused locks as possible before"
"replay for namespace %s (%d)\n",
ldlm_ns_name(ns), ns->ns_nr_unused);
/* We don't need to care whether or not LRU resize is enabled
* because the LDLM_CANCEL_NO_WAIT policy doesn't use the
* count parameter */
canceled = ldlm_cancel_lru_local(ns, &cancels, ns->ns_nr_unused, 0,
LCF_LOCAL, LDLM_CANCEL_NO_WAIT);
CDEBUG(D_DLMTRACE, "Canceled %d unused locks from namespace %s\n",
canceled, ldlm_ns_name(ns));
}
int ldlm_replay_locks(struct obd_import *imp)
{
struct ldlm_namespace *ns = imp->imp_obd->obd_namespace;
LIST_HEAD(list);
struct ldlm_lock *lock, *next;
int rc = 0;
ENTRY;
LASSERT(atomic_read(&imp->imp_replay_inflight) == 0);
/* don't replay locks if import failed recovery */
if (imp->imp_vbr_failed)
RETURN(0);
/* ensure this doesn't fall to 0 before all have been queued */
atomic_inc(&imp->imp_replay_inflight);
if (ldlm_cancel_unused_locks_before_replay)
ldlm_cancel_unused_locks_for_replay(ns);
ldlm_namespace_foreach(ns, ldlm_chain_lock_for_replay, &list);
list_for_each_entry_safe(lock, next, &list, l_pending_chain) {
list_del_init(&lock->l_pending_chain);
if (rc) {
LDLM_LOCK_RELEASE(lock);
continue; /* or try to do the rest? */
}
rc = replay_one_lock(imp, lock);
LDLM_LOCK_RELEASE(lock);
}
atomic_dec(&imp->imp_replay_inflight);
RETURN(rc);
}
EXPORT_SYMBOL(ldlm_replay_locks);
|