Free Electrons

Embedded Linux Experts

   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
/*
 * 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) 2007, 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.
 *
 * lustre/ldlm/ldlm_pool.c
 *
 * Author: Yury Umanets <umka@clusterfs.com>
 */

/*
 * Idea of this code is rather simple. Each second, for each server namespace
 * we have SLV - server lock volume which is calculated on current number of
 * granted locks, grant speed for past period, etc - that is, locking load.
 * This SLV number may be thought as a flow definition for simplicity. It is
 * sent to clients with each occasion to let them know what is current load
 * situation on the server. By default, at the beginning, SLV on server is
 * set max value which is calculated as the following: allow to one client
 * have all locks of limit ->pl_limit for 10h.
 *
 * Next, on clients, number of cached locks is not limited artificially in any
 * way as it was before. Instead, client calculates CLV, that is, client lock
 * volume for each lock and compares it with last SLV from the server. CLV is
 * calculated as the number of locks in LRU * lock live time in seconds. If
 * CLV > SLV - lock is canceled.
 *
 * Client has LVF, that is, lock volume factor which regulates how much sensitive
 * client should be about last SLV from server. The higher LVF is the more locks
 * will be canceled on client. Default value for it is 1. Setting LVF to 2 means
 * that client will cancel locks 2 times faster.
 *
 * Locks on a client will be canceled more intensively in these cases:
 * (1) if SLV is smaller, that is, load is higher on the server;
 * (2) client has a lot of locks (the more locks are held by client, the bigger
 *     chances that some of them should be canceled);
 * (3) client has old locks (taken some time ago);
 *
 * Thus, according to flow paradigm that we use for better understanding SLV,
 * CLV is the volume of particle in flow described by SLV. According to this,
 * if flow is getting thinner, more and more particles become outside of it and
 * as particles are locks, they should be canceled.
 *
 * General idea of this belongs to Vitaly Fertman (vitaly@clusterfs.com). Andreas
 * Dilger (adilger@clusterfs.com) proposed few nice ideas like using LVF and many
 * cleanups. Flow definition to allow more easy understanding of the logic belongs
 * to Nikita Danilov (nikita@clusterfs.com) as well as many cleanups and fixes.
 * And design and implementation are done by Yury Umanets (umka@clusterfs.com).
 *
 * Glossary for terms used:
 *
 * pl_limit - Number of allowed locks in pool. Applies to server and client
 * side (tunable);
 *
 * pl_granted - Number of granted locks (calculated);
 * pl_grant_rate - Number of granted locks for last T (calculated);
 * pl_cancel_rate - Number of canceled locks for last T (calculated);
 * pl_grant_speed - Grant speed (GR - CR) for last T (calculated);
 * pl_grant_plan - Planned number of granted locks for next T (calculated);
 * pl_server_lock_volume - Current server lock volume (calculated);
 *
 * As it may be seen from list above, we have few possible tunables which may
 * affect behavior much. They all may be modified via proc. However, they also
 * give a possibility for constructing few pre-defined behavior policies. If
 * none of predefines is suitable for a working pattern being used, new one may
 * be "constructed" via proc tunables.
 */

#define DEBUG_SUBSYSTEM S_LDLM

# include <lustre_dlm.h>

#include <cl_object.h>

#include <obd_class.h>
#include <obd_support.h>
#include "ldlm_internal.h"


/*
 * 50 ldlm locks for 1MB of RAM.
 */
#define LDLM_POOL_HOST_L ((NUM_CACHEPAGES >> (20 - PAGE_CACHE_SHIFT)) * 50)

/*
 * Maximal possible grant step plan in %.
 */
#define LDLM_POOL_MAX_GSP (30)

/*
 * Minimal possible grant step plan in %.
 */
#define LDLM_POOL_MIN_GSP (1)

/*
 * This controls the speed of reaching LDLM_POOL_MAX_GSP
 * with increasing thread period.
 */
#define LDLM_POOL_GSP_STEP_SHIFT (2)

/*
 * LDLM_POOL_GSP% of all locks is default GP.
 */
#define LDLM_POOL_GP(L)   (((L) * LDLM_POOL_MAX_GSP) / 100)

/*
 * Max age for locks on clients.
 */
#define LDLM_POOL_MAX_AGE (36000)

/*
 * The granularity of SLV calculation.
 */
#define LDLM_POOL_SLV_SHIFT (10)

extern struct proc_dir_entry *ldlm_ns_proc_dir;

static inline __u64 dru(__u64 val, __u32 shift, int round_up)
{
	return (val + (round_up ? (1 << shift) - 1 : 0)) >> shift;
}

static inline __u64 ldlm_pool_slv_max(__u32 L)
{
	/*
	 * Allow to have all locks for 1 client for 10 hrs.
	 * Formula is the following: limit * 10h / 1 client.
	 */
	__u64 lim = (__u64)L *  LDLM_POOL_MAX_AGE / 1;
	return lim;
}

static inline __u64 ldlm_pool_slv_min(__u32 L)
{
	return 1;
}

enum {
	LDLM_POOL_FIRST_STAT = 0,
	LDLM_POOL_GRANTED_STAT = LDLM_POOL_FIRST_STAT,
	LDLM_POOL_GRANT_STAT,
	LDLM_POOL_CANCEL_STAT,
	LDLM_POOL_GRANT_RATE_STAT,
	LDLM_POOL_CANCEL_RATE_STAT,
	LDLM_POOL_GRANT_PLAN_STAT,
	LDLM_POOL_SLV_STAT,
	LDLM_POOL_SHRINK_REQTD_STAT,
	LDLM_POOL_SHRINK_FREED_STAT,
	LDLM_POOL_RECALC_STAT,
	LDLM_POOL_TIMING_STAT,
	LDLM_POOL_LAST_STAT
};

static inline struct ldlm_namespace *ldlm_pl2ns(struct ldlm_pool *pl)
{
	return container_of(pl, struct ldlm_namespace, ns_pool);
}

/**
 * Calculates suggested grant_step in % of available locks for passed
 * \a period. This is later used in grant_plan calculations.
 */
static inline int ldlm_pool_t2gsp(unsigned int t)
{
	/*
	 * This yields 1% grant step for anything below LDLM_POOL_GSP_STEP
	 * and up to 30% for anything higher than LDLM_POOL_GSP_STEP.
	 *
	 * How this will affect execution is the following:
	 *
	 * - for thread period 1s we will have grant_step 1% which good from
	 * pov of taking some load off from server and push it out to clients.
	 * This is like that because 1% for grant_step means that server will
	 * not allow clients to get lots of locks in short period of time and
	 * keep all old locks in their caches. Clients will always have to
	 * get some locks back if they want to take some new;
	 *
	 * - for thread period 10s (which is default) we will have 23% which
	 * means that clients will have enough of room to take some new locks
	 * without getting some back. All locks from this 23% which were not
	 * taken by clients in current period will contribute in SLV growing.
	 * SLV growing means more locks cached on clients until limit or grant
	 * plan is reached.
	 */
	return LDLM_POOL_MAX_GSP -
		((LDLM_POOL_MAX_GSP - LDLM_POOL_MIN_GSP) >>
		 (t >> LDLM_POOL_GSP_STEP_SHIFT));
}

/**
 * Recalculates next grant limit on passed \a pl.
 *
 * \pre ->pl_lock is locked.
 */
static void ldlm_pool_recalc_grant_plan(struct ldlm_pool *pl)
{
	int granted, grant_step, limit;

	limit = ldlm_pool_get_limit(pl);
	granted = atomic_read(&pl->pl_granted);

	grant_step = ldlm_pool_t2gsp(pl->pl_recalc_period);
	grant_step = ((limit - granted) * grant_step) / 100;
	pl->pl_grant_plan = granted + grant_step;
	limit = (limit * 5) >> 2;
	if (pl->pl_grant_plan > limit)
		pl->pl_grant_plan = limit;
}

/**
 * Recalculates next SLV on passed \a pl.
 *
 * \pre ->pl_lock is locked.
 */
static void ldlm_pool_recalc_slv(struct ldlm_pool *pl)
{
	int granted;
	int grant_plan;
	int round_up;
	__u64 slv;
	__u64 slv_factor;
	__u64 grant_usage;
	__u32 limit;

	slv = pl->pl_server_lock_volume;
	grant_plan = pl->pl_grant_plan;
	limit = ldlm_pool_get_limit(pl);
	granted = atomic_read(&pl->pl_granted);
	round_up = granted < limit;

	grant_usage = max_t(int, limit - (granted - grant_plan), 1);

	/*
	 * Find out SLV change factor which is the ratio of grant usage
	 * from limit. SLV changes as fast as the ratio of grant plan
	 * consumption. The more locks from grant plan are not consumed
	 * by clients in last interval (idle time), the faster grows
	 * SLV. And the opposite, the more grant plan is over-consumed
	 * (load time) the faster drops SLV.
	 */
	slv_factor = (grant_usage << LDLM_POOL_SLV_SHIFT);
	do_div(slv_factor, limit);
	slv = slv * slv_factor;
	slv = dru(slv, LDLM_POOL_SLV_SHIFT, round_up);

	if (slv > ldlm_pool_slv_max(limit)) {
		slv = ldlm_pool_slv_max(limit);
	} else if (slv < ldlm_pool_slv_min(limit)) {
		slv = ldlm_pool_slv_min(limit);
	}

	pl->pl_server_lock_volume = slv;
}

/**
 * Recalculates next stats on passed \a pl.
 *
 * \pre ->pl_lock is locked.
 */
static void ldlm_pool_recalc_stats(struct ldlm_pool *pl)
{
	int grant_plan = pl->pl_grant_plan;
	__u64 slv = pl->pl_server_lock_volume;
	int granted = atomic_read(&pl->pl_granted);
	int grant_rate = atomic_read(&pl->pl_grant_rate);
	int cancel_rate = atomic_read(&pl->pl_cancel_rate);

	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_SLV_STAT,
			    slv);
	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_GRANTED_STAT,
			    granted);
	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_GRANT_RATE_STAT,
			    grant_rate);
	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_GRANT_PLAN_STAT,
			    grant_plan);
	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_CANCEL_RATE_STAT,
			    cancel_rate);
}

/**
 * Sets current SLV into obd accessible via ldlm_pl2ns(pl)->ns_obd.
 */
static void ldlm_srv_pool_push_slv(struct ldlm_pool *pl)
{
	struct obd_device *obd;

	/*
	 * Set new SLV in obd field for using it later without accessing the
	 * pool. This is required to avoid race between sending reply to client
	 * with new SLV and cleanup server stack in which we can't guarantee
	 * that namespace is still alive. We know only that obd is alive as
	 * long as valid export is alive.
	 */
	obd = ldlm_pl2ns(pl)->ns_obd;
	LASSERT(obd != NULL);
	write_lock(&obd->obd_pool_lock);
	obd->obd_pool_slv = pl->pl_server_lock_volume;
	write_unlock(&obd->obd_pool_lock);
}

/**
 * Recalculates all pool fields on passed \a pl.
 *
 * \pre ->pl_lock is not locked.
 */
static int ldlm_srv_pool_recalc(struct ldlm_pool *pl)
{
	time_t recalc_interval_sec;

	recalc_interval_sec = cfs_time_current_sec() - pl->pl_recalc_time;
	if (recalc_interval_sec < pl->pl_recalc_period)
		return 0;

	spin_lock(&pl->pl_lock);
	recalc_interval_sec = cfs_time_current_sec() - pl->pl_recalc_time;
	if (recalc_interval_sec < pl->pl_recalc_period) {
		spin_unlock(&pl->pl_lock);
		return 0;
	}
	/*
	 * Recalc SLV after last period. This should be done
	 * _before_ recalculating new grant plan.
	 */
	ldlm_pool_recalc_slv(pl);

	/*
	 * Make sure that pool informed obd of last SLV changes.
	 */
	ldlm_srv_pool_push_slv(pl);

	/*
	 * Update grant_plan for new period.
	 */
	ldlm_pool_recalc_grant_plan(pl);

	pl->pl_recalc_time = cfs_time_current_sec();
	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_TIMING_STAT,
			    recalc_interval_sec);
	spin_unlock(&pl->pl_lock);
	return 0;
}

/**
 * This function is used on server side as main entry point for memory
 * pressure handling. It decreases SLV on \a pl according to passed
 * \a nr and \a gfp_mask.
 *
 * Our goal here is to decrease SLV such a way that clients hold \a nr
 * locks smaller in next 10h.
 */
static int ldlm_srv_pool_shrink(struct ldlm_pool *pl,
				int nr, unsigned int gfp_mask)
{
	__u32 limit;

	/*
	 * VM is asking how many entries may be potentially freed.
	 */
	if (nr == 0)
		return atomic_read(&pl->pl_granted);

	/*
	 * Client already canceled locks but server is already in shrinker
	 * and can't cancel anything. Let's catch this race.
	 */
	if (atomic_read(&pl->pl_granted) == 0)
		return 0;

	spin_lock(&pl->pl_lock);

	/*
	 * We want shrinker to possibly cause cancellation of @nr locks from
	 * clients or grant approximately @nr locks smaller next intervals.
	 *
	 * This is why we decreased SLV by @nr. This effect will only be as
	 * long as one re-calc interval (1s these days) and this should be
	 * enough to pass this decreased SLV to all clients. On next recalc
	 * interval pool will either increase SLV if locks load is not high
	 * or will keep on same level or even decrease again, thus, shrinker
	 * decreased SLV will affect next recalc intervals and this way will
	 * make locking load lower.
	 */
	if (nr < pl->pl_server_lock_volume) {
		pl->pl_server_lock_volume = pl->pl_server_lock_volume - nr;
	} else {
		limit = ldlm_pool_get_limit(pl);
		pl->pl_server_lock_volume = ldlm_pool_slv_min(limit);
	}

	/*
	 * Make sure that pool informed obd of last SLV changes.
	 */
	ldlm_srv_pool_push_slv(pl);
	spin_unlock(&pl->pl_lock);

	/*
	 * We did not really free any memory here so far, it only will be
	 * freed later may be, so that we return 0 to not confuse VM.
	 */
	return 0;
}

/**
 * Setup server side pool \a pl with passed \a limit.
 */
static int ldlm_srv_pool_setup(struct ldlm_pool *pl, int limit)
{
	struct obd_device *obd;

	obd = ldlm_pl2ns(pl)->ns_obd;
	LASSERT(obd != NULL && obd != LP_POISON);
	LASSERT(obd->obd_type != LP_POISON);
	write_lock(&obd->obd_pool_lock);
	obd->obd_pool_limit = limit;
	write_unlock(&obd->obd_pool_lock);

	ldlm_pool_set_limit(pl, limit);
	return 0;
}

/**
 * Sets SLV and Limit from ldlm_pl2ns(pl)->ns_obd tp passed \a pl.
 */
static void ldlm_cli_pool_pop_slv(struct ldlm_pool *pl)
{
	struct obd_device *obd;

	/*
	 * Get new SLV and Limit from obd which is updated with coming
	 * RPCs.
	 */
	obd = ldlm_pl2ns(pl)->ns_obd;
	LASSERT(obd != NULL);
	read_lock(&obd->obd_pool_lock);
	pl->pl_server_lock_volume = obd->obd_pool_slv;
	ldlm_pool_set_limit(pl, obd->obd_pool_limit);
	read_unlock(&obd->obd_pool_lock);
}

/**
 * Recalculates client size pool \a pl according to current SLV and Limit.
 */
static int ldlm_cli_pool_recalc(struct ldlm_pool *pl)
{
	time_t recalc_interval_sec;

	recalc_interval_sec = cfs_time_current_sec() - pl->pl_recalc_time;
	if (recalc_interval_sec < pl->pl_recalc_period)
		return 0;

	spin_lock(&pl->pl_lock);
	/*
	 * Check if we need to recalc lists now.
	 */
	recalc_interval_sec = cfs_time_current_sec() - pl->pl_recalc_time;
	if (recalc_interval_sec < pl->pl_recalc_period) {
		spin_unlock(&pl->pl_lock);
		return 0;
	}

	/*
	 * Make sure that pool knows last SLV and Limit from obd.
	 */
	ldlm_cli_pool_pop_slv(pl);

	pl->pl_recalc_time = cfs_time_current_sec();
	lprocfs_counter_add(pl->pl_stats, LDLM_POOL_TIMING_STAT,
			    recalc_interval_sec);
	spin_unlock(&pl->pl_lock);

	/*
	 * Do not cancel locks in case lru resize is disabled for this ns.
	 */
	if (!ns_connect_lru_resize(ldlm_pl2ns(pl)))
		return 0;

	/*
	 * In the time of canceling locks on client we do not need to maintain
	 * sharp timing, we only want to cancel locks asap according to new SLV.
	 * It may be called when SLV has changed much, this is why we do not
	 * take into account pl->pl_recalc_time here.
	 */
	return ldlm_cancel_lru(ldlm_pl2ns(pl), 0, LCF_ASYNC, LDLM_CANCEL_LRUR);
}

/**
 * This function is main entry point for memory pressure handling on client
 * side.  Main goal of this function is to cancel some number of locks on
 * passed \a pl according to \a nr and \a gfp_mask.
 */
static int ldlm_cli_pool_shrink(struct ldlm_pool *pl,
				int nr, unsigned int gfp_mask)
{
	struct ldlm_namespace *ns;
	int unused;

	ns = ldlm_pl2ns(pl);

	/*
	 * Do not cancel locks in case lru resize is disabled for this ns.
	 */
	if (!ns_connect_lru_resize(ns))
		return 0;

	/*
	 * Make sure that pool knows last SLV and Limit from obd.
	 */
	ldlm_cli_pool_pop_slv(pl);

	spin_lock(&ns->ns_lock);
	unused = ns->ns_nr_unused;
	spin_unlock(&ns->ns_lock);

	if (nr == 0)
		return (unused / 100) * sysctl_vfs_cache_pressure;
	else
		return ldlm_cancel_lru(ns, nr, LCF_ASYNC, LDLM_CANCEL_SHRINK);
}

struct ldlm_pool_ops ldlm_srv_pool_ops = {
	.po_recalc = ldlm_srv_pool_recalc,
	.po_shrink = ldlm_srv_pool_shrink,
	.po_setup  = ldlm_srv_pool_setup
};

struct ldlm_pool_ops ldlm_cli_pool_ops = {
	.po_recalc = ldlm_cli_pool_recalc,
	.po_shrink = ldlm_cli_pool_shrink
};

/**
 * Pool recalc wrapper. Will call either client or server pool recalc callback
 * depending what pool \a pl is used.
 */
int ldlm_pool_recalc(struct ldlm_pool *pl)
{
	time_t recalc_interval_sec;
	int count;

	recalc_interval_sec = cfs_time_current_sec() - pl->pl_recalc_time;
	if (recalc_interval_sec <= 0)
		goto recalc;

	spin_lock(&pl->pl_lock);
	if (recalc_interval_sec > 0) {
		/*
		 * Update pool statistics every 1s.
		 */
		ldlm_pool_recalc_stats(pl);

		/*
		 * Zero out all rates and speed for the last period.
		 */
		atomic_set(&pl->pl_grant_rate, 0);
		atomic_set(&pl->pl_cancel_rate, 0);
	}
	spin_unlock(&pl->pl_lock);

 recalc:
	if (pl->pl_ops->po_recalc != NULL) {
		count = pl->pl_ops->po_recalc(pl);
		lprocfs_counter_add(pl->pl_stats, LDLM_POOL_RECALC_STAT,
				    count);
	}
	recalc_interval_sec = pl->pl_recalc_time - cfs_time_current_sec() +
			      pl->pl_recalc_period;

	return recalc_interval_sec;
}

/*
 * Pool shrink wrapper. Will call either client or server pool recalc callback
 * depending what pool pl is used. When nr == 0, just return the number of
 * freeable locks. Otherwise, return the number of canceled locks.
 */
int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
		     unsigned int gfp_mask)
{
	int cancel = 0;

	if (pl->pl_ops->po_shrink != NULL) {
		cancel = pl->pl_ops->po_shrink(pl, nr, gfp_mask);
		if (nr > 0) {
			lprocfs_counter_add(pl->pl_stats,
					    LDLM_POOL_SHRINK_REQTD_STAT,
					    nr);
			lprocfs_counter_add(pl->pl_stats,
					    LDLM_POOL_SHRINK_FREED_STAT,
					    cancel);
			CDEBUG(D_DLMTRACE, "%s: request to shrink %d locks, "
			       "shrunk %d\n", pl->pl_name, nr, cancel);
		}
	}
	return cancel;
}
EXPORT_SYMBOL(ldlm_pool_shrink);

/**
 * Pool setup wrapper. Will call either client or server pool recalc callback
 * depending what pool \a pl is used.
 *
 * Sets passed \a limit into pool \a pl.
 */
int ldlm_pool_setup(struct ldlm_pool *pl, int limit)
{
	if (pl->pl_ops->po_setup != NULL)
		return(pl->pl_ops->po_setup(pl, limit));
	return 0;
}
EXPORT_SYMBOL(ldlm_pool_setup);

static int lprocfs_pool_state_seq_show(struct seq_file *m, void *unused)
{
	int granted, grant_rate, cancel_rate, grant_step;
	int grant_speed, grant_plan, lvf;
	struct ldlm_pool *pl = m->private;
	__u64 slv, clv;
	__u32 limit;

	spin_lock(&pl->pl_lock);
	slv = pl->pl_server_lock_volume;
	clv = pl->pl_client_lock_volume;
	limit = ldlm_pool_get_limit(pl);
	grant_plan = pl->pl_grant_plan;
	granted = atomic_read(&pl->pl_granted);
	grant_rate = atomic_read(&pl->pl_grant_rate);
	cancel_rate = atomic_read(&pl->pl_cancel_rate);
	grant_speed = grant_rate - cancel_rate;
	lvf = atomic_read(&pl->pl_lock_volume_factor);
	grant_step = ldlm_pool_t2gsp(pl->pl_recalc_period);
	spin_unlock(&pl->pl_lock);

	seq_printf(m, "LDLM pool state (%s):\n"
		      "  SLV: "LPU64"\n"
		      "  CLV: "LPU64"\n"
		      "  LVF: %d\n",
		      pl->pl_name, slv, clv, lvf);

	if (ns_is_server(ldlm_pl2ns(pl))) {
		seq_printf(m, "  GSP: %d%%\n"
			      "  GP:  %d\n",
			      grant_step, grant_plan);
	}
	seq_printf(m, "  GR:  %d\n" "  CR:  %d\n" "  GS:  %d\n"
		      "  G:   %d\n" "  L:   %d\n",
		      grant_rate, cancel_rate, grant_speed,
		      granted, limit);

	return 0;
}
LPROC_SEQ_FOPS_RO(lprocfs_pool_state);

static int lprocfs_grant_speed_seq_show(struct seq_file *m, void *unused)
{
	struct ldlm_pool *pl = m->private;
	int	       grant_speed;

	spin_lock(&pl->pl_lock);
	/* serialize with ldlm_pool_recalc */
	grant_speed = atomic_read(&pl->pl_grant_rate) -
			atomic_read(&pl->pl_cancel_rate);
	spin_unlock(&pl->pl_lock);
	return lprocfs_rd_uint(m, &grant_speed);
}

LDLM_POOL_PROC_READER_SEQ_SHOW(grant_plan, int);
LPROC_SEQ_FOPS_RO(lprocfs_grant_plan);

LDLM_POOL_PROC_READER_SEQ_SHOW(recalc_period, int);
LDLM_POOL_PROC_WRITER(recalc_period, int);
static ssize_t lprocfs_recalc_period_seq_write(struct file *file, const char *buf,
					   size_t len, loff_t *off)
{
	struct seq_file *seq = file->private_data;

	return lprocfs_wr_recalc_period(file, buf, len, seq->private);
}
LPROC_SEQ_FOPS(lprocfs_recalc_period);

LPROC_SEQ_FOPS_RO_TYPE(ldlm_pool, u64);
LPROC_SEQ_FOPS_RO_TYPE(ldlm_pool, atomic);
LPROC_SEQ_FOPS_RW_TYPE(ldlm_pool_rw, atomic);

LPROC_SEQ_FOPS_RO(lprocfs_grant_speed);

#define LDLM_POOL_ADD_VAR(name, var, ops)			\
	do {							\
		snprintf(var_name, MAX_STRING_SIZE, #name);	\
		pool_vars[0].data = var;			\
		pool_vars[0].fops = ops;			\
		lprocfs_add_vars(pl->pl_proc_dir, pool_vars, 0);\
	} while (0)

static int ldlm_pool_proc_init(struct ldlm_pool *pl)
{
	struct ldlm_namespace *ns = ldlm_pl2ns(pl);
	struct proc_dir_entry *parent_ns_proc;
	struct lprocfs_vars pool_vars[2];
	char *var_name = NULL;
	int rc = 0;

	OBD_ALLOC(var_name, MAX_STRING_SIZE + 1);
	if (!var_name)
		return -ENOMEM;

	parent_ns_proc = ns->ns_proc_dir_entry;
	if (parent_ns_proc == NULL) {
		CERROR("%s: proc entry is not initialized\n",
		       ldlm_ns_name(ns));
		GOTO(out_free_name, rc = -EINVAL);
	}
	pl->pl_proc_dir = lprocfs_register("pool", parent_ns_proc,
					   NULL, NULL);
	if (IS_ERR(pl->pl_proc_dir)) {
		CERROR("LProcFS failed in ldlm-pool-init\n");
		rc = PTR_ERR(pl->pl_proc_dir);
		pl->pl_proc_dir = NULL;
		GOTO(out_free_name, rc);
	}

	var_name[MAX_STRING_SIZE] = '\0';
	memset(pool_vars, 0, sizeof(pool_vars));
	pool_vars[0].name = var_name;

	LDLM_POOL_ADD_VAR("server_lock_volume", &pl->pl_server_lock_volume,
			  &ldlm_pool_u64_fops);
	LDLM_POOL_ADD_VAR("limit", &pl->pl_limit, &ldlm_pool_rw_atomic_fops);
	LDLM_POOL_ADD_VAR("granted", &pl->pl_granted, &ldlm_pool_atomic_fops);
	LDLM_POOL_ADD_VAR("grant_speed", pl, &lprocfs_grant_speed_fops);
	LDLM_POOL_ADD_VAR("cancel_rate", &pl->pl_cancel_rate,
			  &ldlm_pool_atomic_fops);
	LDLM_POOL_ADD_VAR("grant_rate", &pl->pl_grant_rate,
			  &ldlm_pool_atomic_fops);
	LDLM_POOL_ADD_VAR("grant_plan", pl, &lprocfs_grant_plan_fops);
	LDLM_POOL_ADD_VAR("recalc_period", pl, &lprocfs_recalc_period_fops);
	LDLM_POOL_ADD_VAR("lock_volume_factor", &pl->pl_lock_volume_factor,
			  &ldlm_pool_rw_atomic_fops);
	LDLM_POOL_ADD_VAR("state", pl, &lprocfs_pool_state_fops);

	pl->pl_stats = lprocfs_alloc_stats(LDLM_POOL_LAST_STAT -
					   LDLM_POOL_FIRST_STAT, 0);
	if (!pl->pl_stats)
		GOTO(out_free_name, rc = -ENOMEM);

	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANTED_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "granted", "locks");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANT_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "grant", "locks");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_CANCEL_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "cancel", "locks");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANT_RATE_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "grant_rate", "locks/s");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_CANCEL_RATE_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "cancel_rate", "locks/s");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_GRANT_PLAN_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "grant_plan", "locks/s");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_SLV_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "slv", "slv");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_SHRINK_REQTD_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "shrink_request", "locks");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_SHRINK_FREED_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "shrink_freed", "locks");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_RECALC_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "recalc_freed", "locks");
	lprocfs_counter_init(pl->pl_stats, LDLM_POOL_TIMING_STAT,
			     LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV,
			     "recalc_timing", "sec");
	rc = lprocfs_register_stats(pl->pl_proc_dir, "stats", pl->pl_stats);

out_free_name:
	OBD_FREE(var_name, MAX_STRING_SIZE + 1);
	return rc;
}

static void ldlm_pool_proc_fini(struct ldlm_pool *pl)
{
	if (pl->pl_stats != NULL) {
		lprocfs_free_stats(&pl->pl_stats);
		pl->pl_stats = NULL;
	}
	if (pl->pl_proc_dir != NULL) {
		lprocfs_remove(&pl->pl_proc_dir);
		pl->pl_proc_dir = NULL;
	}
}

int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
		   int idx, ldlm_side_t client)
{
	int rc;

	spin_lock_init(&pl->pl_lock);
	atomic_set(&pl->pl_granted, 0);
	pl->pl_recalc_time = cfs_time_current_sec();
	atomic_set(&pl->pl_lock_volume_factor, 1);

	atomic_set(&pl->pl_grant_rate, 0);
	atomic_set(&pl->pl_cancel_rate, 0);
	pl->pl_grant_plan = LDLM_POOL_GP(LDLM_POOL_HOST_L);

	snprintf(pl->pl_name, sizeof(pl->pl_name), "ldlm-pool-%s-%d",
		 ldlm_ns_name(ns), idx);

	if (client == LDLM_NAMESPACE_SERVER) {
		pl->pl_ops = &ldlm_srv_pool_ops;
		ldlm_pool_set_limit(pl, LDLM_POOL_HOST_L);
		pl->pl_recalc_period = LDLM_POOL_SRV_DEF_RECALC_PERIOD;
		pl->pl_server_lock_volume = ldlm_pool_slv_max(LDLM_POOL_HOST_L);
	} else {
		ldlm_pool_set_limit(pl, 1);
		pl->pl_server_lock_volume = 0;
		pl->pl_ops = &ldlm_cli_pool_ops;
		pl->pl_recalc_period = LDLM_POOL_CLI_DEF_RECALC_PERIOD;
	}
	pl->pl_client_lock_volume = 0;
	rc = ldlm_pool_proc_init(pl);
	if (rc)
		return rc;

	CDEBUG(D_DLMTRACE, "Lock pool %s is initialized\n", pl->pl_name);

	return rc;
}
EXPORT_SYMBOL(ldlm_pool_init);

void ldlm_pool_fini(struct ldlm_pool *pl)
{
	ldlm_pool_proc_fini(pl);

	/*
	 * Pool should not be used after this point. We can't free it here as
	 * it lives in struct ldlm_namespace, but still interested in catching
	 * any abnormal using cases.
	 */
	POISON(pl, 0x5a, sizeof(*pl));
}
EXPORT_SYMBOL(ldlm_pool_fini);

/**
 * Add new taken ldlm lock \a lock into pool \a pl accounting.
 */
void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock)
{
	/*
	 * FLOCK locks are special in a sense that they are almost never
	 * cancelled, instead special kind of lock is used to drop them.
	 * also there is no LRU for flock locks, so no point in tracking
	 * them anyway.
	 */
	if (lock->l_resource->lr_type == LDLM_FLOCK)
		return;

	atomic_inc(&pl->pl_granted);
	atomic_inc(&pl->pl_grant_rate);
	lprocfs_counter_incr(pl->pl_stats, LDLM_POOL_GRANT_STAT);
	/*
	 * Do not do pool recalc for client side as all locks which
	 * potentially may be canceled has already been packed into
	 * enqueue/cancel rpc. Also we do not want to run out of stack
	 * with too long call paths.
	 */
	if (ns_is_server(ldlm_pl2ns(pl)))
		ldlm_pool_recalc(pl);
}
EXPORT_SYMBOL(ldlm_pool_add);

/**
 * Remove ldlm lock \a lock from pool \a pl accounting.
 */
void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock)
{
	/*
	 * Filter out FLOCK locks. Read above comment in ldlm_pool_add().
	 */
	if (lock->l_resource->lr_type == LDLM_FLOCK)
		return;

	LASSERT(atomic_read(&pl->pl_granted) > 0);
	atomic_dec(&pl->pl_granted);
	atomic_inc(&pl->pl_cancel_rate);

	lprocfs_counter_incr(pl->pl_stats, LDLM_POOL_CANCEL_STAT);

	if (ns_is_server(ldlm_pl2ns(pl)))
		ldlm_pool_recalc(pl);
}
EXPORT_SYMBOL(ldlm_pool_del);

/**
 * Returns current \a pl SLV.
 *
 * \pre ->pl_lock is not locked.
 */
__u64 ldlm_pool_get_slv(struct ldlm_pool *pl)
{
	__u64 slv;
	spin_lock(&pl->pl_lock);
	slv = pl->pl_server_lock_volume;
	spin_unlock(&pl->pl_lock);
	return slv;
}
EXPORT_SYMBOL(ldlm_pool_get_slv);

/**
 * Sets passed \a slv to \a pl.
 *
 * \pre ->pl_lock is not locked.
 */
void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv)
{
	spin_lock(&pl->pl_lock);
	pl->pl_server_lock_volume = slv;
	spin_unlock(&pl->pl_lock);
}
EXPORT_SYMBOL(ldlm_pool_set_slv);

/**
 * Returns current \a pl CLV.
 *
 * \pre ->pl_lock is not locked.
 */
__u64 ldlm_pool_get_clv(struct ldlm_pool *pl)
{
	__u64 slv;
	spin_lock(&pl->pl_lock);
	slv = pl->pl_client_lock_volume;
	spin_unlock(&pl->pl_lock);
	return slv;
}
EXPORT_SYMBOL(ldlm_pool_get_clv);

/**
 * Sets passed \a clv to \a pl.
 *
 * \pre ->pl_lock is not locked.
 */
void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv)
{
	spin_lock(&pl->pl_lock);
	pl->pl_client_lock_volume = clv;
	spin_unlock(&pl->pl_lock);
}
EXPORT_SYMBOL(ldlm_pool_set_clv);

/**
 * Returns current \a pl limit.
 */
__u32 ldlm_pool_get_limit(struct ldlm_pool *pl)
{
	return atomic_read(&pl->pl_limit);
}
EXPORT_SYMBOL(ldlm_pool_get_limit);

/**
 * Sets passed \a limit to \a pl.
 */
void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit)
{
	atomic_set(&pl->pl_limit, limit);
}
EXPORT_SYMBOL(ldlm_pool_set_limit);

/**
 * Returns current LVF from \a pl.
 */
__u32 ldlm_pool_get_lvf(struct ldlm_pool *pl)
{
	return atomic_read(&pl->pl_lock_volume_factor);
}
EXPORT_SYMBOL(ldlm_pool_get_lvf);

static int ldlm_pool_granted(struct ldlm_pool *pl)
{
	return atomic_read(&pl->pl_granted);
}

static struct ptlrpc_thread *ldlm_pools_thread;
static struct completion ldlm_pools_comp;

/*
 * count locks from all namespaces (if possible). Returns number of
 * cached locks.
 */
static unsigned long ldlm_pools_count(ldlm_side_t client, unsigned int gfp_mask)
{
	int total = 0, nr_ns;
	struct ldlm_namespace *ns;
	struct ldlm_namespace *ns_old = NULL; /* loop detection */
	void *cookie;

	if (client == LDLM_NAMESPACE_CLIENT && !(gfp_mask & __GFP_FS))
		return 0;

	CDEBUG(D_DLMTRACE, "Request to count %s locks from all pools\n",
	       client == LDLM_NAMESPACE_CLIENT ? "client" : "server");

	cookie = cl_env_reenter();

	/*
	 * Find out how many resources we may release.
	 */
	for (nr_ns = ldlm_namespace_nr_read(client);
	     nr_ns > 0; nr_ns--) {
		mutex_lock(ldlm_namespace_lock(client));
		if (list_empty(ldlm_namespace_list(client))) {
			mutex_unlock(ldlm_namespace_lock(client));
			cl_env_reexit(cookie);
			return 0;
		}
		ns = ldlm_namespace_first_locked(client);

		if (ns == ns_old) {
			mutex_unlock(ldlm_namespace_lock(client));
			break;
		}

		if (ldlm_ns_empty(ns)) {
			ldlm_namespace_move_to_inactive_locked(ns, client);
			mutex_unlock(ldlm_namespace_lock(client));
			continue;
		}

		if (ns_old == NULL)
			ns_old = ns;

		ldlm_namespace_get(ns);
		ldlm_namespace_move_to_active_locked(ns, client);
		mutex_unlock(ldlm_namespace_lock(client));
		total += ldlm_pool_shrink(&ns->ns_pool, 0, gfp_mask);
		ldlm_namespace_put(ns);
	}

	cl_env_reexit(cookie);
	return total;
}

static unsigned long ldlm_pools_scan(ldlm_side_t client, int nr, unsigned int gfp_mask)
{
	unsigned long freed = 0;
	int tmp, nr_ns;
	struct ldlm_namespace *ns;
	void *cookie;

	if (client == LDLM_NAMESPACE_CLIENT && !(gfp_mask & __GFP_FS))
		return -1;

	cookie = cl_env_reenter();

	/*
	 * Shrink at least ldlm_namespace_nr_read(client) namespaces.
	 */
	for (tmp = nr_ns = ldlm_namespace_nr_read(client);
	     tmp > 0; tmp--) {
		int cancel, nr_locks;

		/*
		 * Do not call shrink under ldlm_namespace_lock(client)
		 */
		mutex_lock(ldlm_namespace_lock(client));
		if (list_empty(ldlm_namespace_list(client))) {
			mutex_unlock(ldlm_namespace_lock(client));
			break;
		}
		ns = ldlm_namespace_first_locked(client);
		ldlm_namespace_get(ns);
		ldlm_namespace_move_to_active_locked(ns, client);
		mutex_unlock(ldlm_namespace_lock(client));

		nr_locks = ldlm_pool_granted(&ns->ns_pool);
		/*
		 * We use to shrink propotionally but with new shrinker API,
		 * we lost the total number of freeable locks.
		 */
		cancel = 1 + min_t(int, nr_locks, nr / nr_ns);
		freed += ldlm_pool_shrink(&ns->ns_pool, cancel, gfp_mask);
		ldlm_namespace_put(ns);
	}
	cl_env_reexit(cookie);
	/*
	 * we only decrease the SLV in server pools shrinker, return
	 * SHRINK_STOP to kernel to avoid needless loop. LU-1128
	 */
	return (client == LDLM_NAMESPACE_SERVER) ? SHRINK_STOP : freed;
}

static unsigned long ldlm_pools_srv_count(struct shrinker *s, struct shrink_control *sc)
{
	return ldlm_pools_count(LDLM_NAMESPACE_SERVER, sc->gfp_mask);
}

static unsigned long ldlm_pools_srv_scan(struct shrinker *s, struct shrink_control *sc)
{
	return ldlm_pools_scan(LDLM_NAMESPACE_SERVER, sc->nr_to_scan,
			       sc->gfp_mask);
}

static unsigned long ldlm_pools_cli_count(struct shrinker *s, struct shrink_control *sc)
{
	return ldlm_pools_count(LDLM_NAMESPACE_CLIENT, sc->gfp_mask);
}

static unsigned long ldlm_pools_cli_scan(struct shrinker *s, struct shrink_control *sc)
{
	return ldlm_pools_scan(LDLM_NAMESPACE_CLIENT, sc->nr_to_scan,
			       sc->gfp_mask);
}

int ldlm_pools_recalc(ldlm_side_t client)
{
	__u32 nr_l = 0, nr_p = 0, l;
	struct ldlm_namespace *ns;
	struct ldlm_namespace *ns_old = NULL;
	int nr, equal = 0;
	int time = 50; /* seconds of sleep if no active namespaces */

	/*
	 * No need to setup pool limit for client pools.
	 */
	if (client == LDLM_NAMESPACE_SERVER) {
		/*
		 * Check all modest namespaces first.
		 */
		mutex_lock(ldlm_namespace_lock(client));
		list_for_each_entry(ns, ldlm_namespace_list(client),
					ns_list_chain)
		{
			if (ns->ns_appetite != LDLM_NAMESPACE_MODEST)
				continue;

			l = ldlm_pool_granted(&ns->ns_pool);
			if (l == 0)
				l = 1;

			/*
			 * Set the modest pools limit equal to their avg granted
			 * locks + ~6%.
			 */
			l += dru(l, LDLM_POOLS_MODEST_MARGIN_SHIFT, 0);
			ldlm_pool_setup(&ns->ns_pool, l);
			nr_l += l;
			nr_p++;
		}

		/*
		 * Make sure that modest namespaces did not eat more that 2/3
		 * of limit.
		 */
		if (nr_l >= 2 * (LDLM_POOL_HOST_L / 3)) {
			CWARN("\"Modest\" pools eat out 2/3 of server locks "
			      "limit (%d of %lu). This means that you have too "
			      "many clients for this amount of server RAM. "
			      "Upgrade server!\n", nr_l, LDLM_POOL_HOST_L);
			equal = 1;
		}

		/*
		 * The rest is given to greedy namespaces.
		 */
		list_for_each_entry(ns, ldlm_namespace_list(client),
					ns_list_chain)
		{
			if (!equal && ns->ns_appetite != LDLM_NAMESPACE_GREEDY)
				continue;

			if (equal) {
				/*
				 * In the case 2/3 locks are eaten out by
				 * modest pools, we re-setup equal limit
				 * for _all_ pools.
				 */
				l = LDLM_POOL_HOST_L /
					ldlm_namespace_nr_read(client);
			} else {
				/*
				 * All the rest of greedy pools will have
				 * all locks in equal parts.
				 */
				l = (LDLM_POOL_HOST_L - nr_l) /
					(ldlm_namespace_nr_read(client) -
					 nr_p);
			}
			ldlm_pool_setup(&ns->ns_pool, l);
		}
		mutex_unlock(ldlm_namespace_lock(client));
	}

	/*
	 * Recalc at least ldlm_namespace_nr_read(client) namespaces.
	 */
	for (nr = ldlm_namespace_nr_read(client); nr > 0; nr--) {
		int     skip;
		/*
		 * Lock the list, get first @ns in the list, getref, move it
		 * to the tail, unlock and call pool recalc. This way we avoid
		 * calling recalc under @ns lock what is really good as we get
		 * rid of potential deadlock on client nodes when canceling
		 * locks synchronously.
		 */
		mutex_lock(ldlm_namespace_lock(client));
		if (list_empty(ldlm_namespace_list(client))) {
			mutex_unlock(ldlm_namespace_lock(client));
			break;
		}
		ns = ldlm_namespace_first_locked(client);

		if (ns_old == ns) { /* Full pass complete */
			mutex_unlock(ldlm_namespace_lock(client));
			break;
		}

		/* We got an empty namespace, need to move it back to inactive
		 * list.
		 * The race with parallel resource creation is fine:
		 * - If they do namespace_get before our check, we fail the
		 *   check and they move this item to the end of the list anyway
		 * - If we do the check and then they do namespace_get, then
		 *   we move the namespace to inactive and they will move
		 *   it back to active (synchronised by the lock, so no clash
		 *   there).
		 */
		if (ldlm_ns_empty(ns)) {
			ldlm_namespace_move_to_inactive_locked(ns, client);
			mutex_unlock(ldlm_namespace_lock(client));
			continue;
		}

		if (ns_old == NULL)
			ns_old = ns;

		spin_lock(&ns->ns_lock);
		/*
		 * skip ns which is being freed, and we don't want to increase
		 * its refcount again, not even temporarily. bz21519 & LU-499.
		 */
		if (ns->ns_stopping) {
			skip = 1;
		} else {
			skip = 0;
			ldlm_namespace_get(ns);
		}
		spin_unlock(&ns->ns_lock);

		ldlm_namespace_move_to_active_locked(ns, client);
		mutex_unlock(ldlm_namespace_lock(client));

		/*
		 * After setup is done - recalc the pool.
		 */
		if (!skip) {
			int ttime = ldlm_pool_recalc(&ns->ns_pool);

			if (ttime < time)
				time = ttime;

			ldlm_namespace_put(ns);
		}
	}
	return time;
}
EXPORT_SYMBOL(ldlm_pools_recalc);

static int ldlm_pools_thread_main(void *arg)
{
	struct ptlrpc_thread *thread = (struct ptlrpc_thread *)arg;
	int s_time, c_time;

	thread_set_flags(thread, SVC_RUNNING);
	wake_up(&thread->t_ctl_waitq);

	CDEBUG(D_DLMTRACE, "%s: pool thread starting, process %d\n",
		"ldlm_poold", current_pid());

	while (1) {
		struct l_wait_info lwi;

		/*
		 * Recal all pools on this tick.
		 */
		s_time = ldlm_pools_recalc(LDLM_NAMESPACE_SERVER);
		c_time = ldlm_pools_recalc(LDLM_NAMESPACE_CLIENT);

		/*
		 * Wait until the next check time, or until we're
		 * stopped.
		 */
		lwi = LWI_TIMEOUT(cfs_time_seconds(min(s_time, c_time)),
				  NULL, NULL);
		l_wait_event(thread->t_ctl_waitq,
			     thread_is_stopping(thread) ||
			     thread_is_event(thread),
			     &lwi);

		if (thread_test_and_clear_flags(thread, SVC_STOPPING))
			break;
		else
			thread_test_and_clear_flags(thread, SVC_EVENT);
	}

	thread_set_flags(thread, SVC_STOPPED);
	wake_up(&thread->t_ctl_waitq);

	CDEBUG(D_DLMTRACE, "%s: pool thread exiting, process %d\n",
		"ldlm_poold", current_pid());

	complete_and_exit(&ldlm_pools_comp, 0);
}

static int ldlm_pools_thread_start(void)
{
	struct l_wait_info lwi = { 0 };
	struct task_struct *task;

	if (ldlm_pools_thread != NULL)
		return -EALREADY;

	OBD_ALLOC_PTR(ldlm_pools_thread);
	if (ldlm_pools_thread == NULL)
		return -ENOMEM;

	init_completion(&ldlm_pools_comp);
	init_waitqueue_head(&ldlm_pools_thread->t_ctl_waitq);

	task = kthread_run(ldlm_pools_thread_main, ldlm_pools_thread,
			   "ldlm_poold");
	if (IS_ERR(task)) {
		CERROR("Can't start pool thread, error %ld\n", PTR_ERR(task));
		OBD_FREE(ldlm_pools_thread, sizeof(*ldlm_pools_thread));
		ldlm_pools_thread = NULL;
		return PTR_ERR(task);
	}
	l_wait_event(ldlm_pools_thread->t_ctl_waitq,
		     thread_is_running(ldlm_pools_thread), &lwi);
	return 0;
}

static void ldlm_pools_thread_stop(void)
{
	if (ldlm_pools_thread == NULL) {
		return;
	}

	thread_set_flags(ldlm_pools_thread, SVC_STOPPING);
	wake_up(&ldlm_pools_thread->t_ctl_waitq);

	/*
	 * Make sure that pools thread is finished before freeing @thread.
	 * This fixes possible race and oops due to accessing freed memory
	 * in pools thread.
	 */
	wait_for_completion(&ldlm_pools_comp);
	OBD_FREE_PTR(ldlm_pools_thread);
	ldlm_pools_thread = NULL;
}

static struct shrinker ldlm_pools_srv_shrinker = {
	.count_objects	= ldlm_pools_srv_count,
	.scan_objects	= ldlm_pools_srv_scan,
	.seeks		= DEFAULT_SEEKS,
};

static struct shrinker ldlm_pools_cli_shrinker = {
	.count_objects	= ldlm_pools_cli_count,
	.scan_objects	= ldlm_pools_cli_scan,
	.seeks		= DEFAULT_SEEKS,
};

int ldlm_pools_init(void)
{
	int rc;

	rc = ldlm_pools_thread_start();
	if (rc == 0) {
		register_shrinker(&ldlm_pools_srv_shrinker);
		register_shrinker(&ldlm_pools_cli_shrinker);
	}
	return rc;
}
EXPORT_SYMBOL(ldlm_pools_init);

void ldlm_pools_fini(void)
{
	unregister_shrinker(&ldlm_pools_srv_shrinker);
	unregister_shrinker(&ldlm_pools_cli_shrinker);
	ldlm_pools_thread_stop();
}
EXPORT_SYMBOL(ldlm_pools_fini);