libcfs/libcfs/fail.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see http://www.gnu.org/licenses
  18  *
  19  * Please contact Oracle Corporation, Inc., 500 Oracle Parkway, Redwood Shores,
  20  * CA 94065 USA or visit www.oracle.com if you need additional information or
  21  * have any questions.
  22  *
  23  * GPL HEADER END
  24  */
  25 /*
  26  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  27  * Use is subject to license terms.
  28  *
  29  * Copyright (c) 2012, Intel Corporation.
  30  */
  31 /*
  32  * This file is part of Lustre, http://www.lustre.org/
  33  * Lustre is a trademark of Oracle Corporation, Inc.
  34  */
  35
  36 #ifndef __KERNEL__
  37 #include <liblustre.h>
  38 #else
  39 #include <libcfs/libcfs.h>
  40 #endif
  41
  42 unsigned long cfs_fail_loc = 0;
  43 unsigned int cfs_fail_val = 0;
  44 wait_queue_head_t cfs_race_waitq;
  45 int cfs_race_state;
  46
  47 EXPORT_SYMBOL(cfs_fail_loc);
  48 EXPORT_SYMBOL(cfs_fail_val);
  49 EXPORT_SYMBOL(cfs_race_waitq);
  50 EXPORT_SYMBOL(cfs_race_state);
  51
  52 int __cfs_fail_check_set(__u32 id, __u32 value, int set)
  53 {
  54         static atomic_t cfs_fail_count = ATOMIC_INIT(0);
  55
  56         LASSERT(!(id & CFS_FAIL_ONCE));
  57
  58         if ((cfs_fail_loc & (CFS_FAILED | CFS_FAIL_ONCE)) ==
  59             (CFS_FAILED | CFS_FAIL_ONCE)) {
  60                 atomic_set(&cfs_fail_count, 0); /* paranoia */
  61                 return 0;
  62         }
  63
  64         /* Fail 1/cfs_fail_val times */
  65         if (cfs_fail_loc & CFS_FAIL_RAND) {
  66                 if (cfs_fail_val < 2 || cfs_rand() % cfs_fail_val > 0)
  67                         return 0;
  68         }
  69
  70         /* Skip the first cfs_fail_val, then fail */
  71         if (cfs_fail_loc & CFS_FAIL_SKIP) {
  72                 if (atomic_inc_return(&cfs_fail_count) <= cfs_fail_val)
  73                         return 0;
  74         }
  75
  76         /* check cfs_fail_val... */
  77         if (set == CFS_FAIL_LOC_VALUE) {
  78                 if (cfs_fail_val != -1 && cfs_fail_val != value)
  79                         return 0;
  80         }
  81
  82         /* Fail cfs_fail_val times, overridden by FAIL_ONCE */
  83         if (cfs_fail_loc & CFS_FAIL_SOME &&
  84             (!(cfs_fail_loc & CFS_FAIL_ONCE) || cfs_fail_val <= 1)) {
  85                 int count = atomic_inc_return(&cfs_fail_count);
  86
  87                 if (count >= cfs_fail_val) {
  88                         set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  89                         atomic_set(&cfs_fail_count, 0);
  90                         /* we are lost race to increase  */
  91                         if (count > cfs_fail_val)
  92                                 return 0;
  93                 }
  94         }
  95
  96         /* Take into account the current call for FAIL_ONCE for ORSET only,
  97          * as RESET is a new fail_loc, it does not change the current call */
  98         if ((set == CFS_FAIL_LOC_ORSET) && (value & CFS_FAIL_ONCE))
  99                 set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
 100         /* Lost race to set CFS_FAILED_BIT. */
 101         if (test_and_set_bit(CFS_FAILED_BIT, &cfs_fail_loc)) {
 102                 /* If CFS_FAIL_ONCE is valid, only one process can fail,
 103                  * otherwise multi-process can fail at the same time. */
 104                 if (cfs_fail_loc & CFS_FAIL_ONCE)
 105                         return 0;
 106         }
 107
 108         switch (set) {
 109                 case CFS_FAIL_LOC_NOSET:
 110                 case CFS_FAIL_LOC_VALUE:
 111                         break;
 112                 case CFS_FAIL_LOC_ORSET:
 113                         cfs_fail_loc |= value & ~(CFS_FAILED | CFS_FAIL_ONCE);
 114                         break;
 115                 case CFS_FAIL_LOC_RESET:
 116                         cfs_fail_loc = value;
 117                         break;
 118                 default:
 119                         LASSERTF(0, "called with bad set %u\n", set);
 120                         break;
 121         }
 122
 123         return 1;
 124 }
 125 EXPORT_SYMBOL(__cfs_fail_check_set);
 126
 127 int __cfs_fail_timeout_set(__u32 id, __u32 value, int ms, int set)
 128 {
 129         int ret = 0;
 130
 131         ret = __cfs_fail_check_set(id, value, set);
 132         if (ret) {
 133                 CERROR("cfs_fail_timeout id %x sleeping for %dms\n",
 134                        id, ms);
 135                 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
 136                                                    cfs_time_seconds(ms) / 1000);
 137                 set_current_state(TASK_RUNNING);
 138                 CERROR("cfs_fail_timeout id %x awake\n", id);
 139         }
 140         return ret;
 141 }
 142 EXPORT_SYMBOL(__cfs_fail_timeout_set);