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, 2014, 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 #include <libcfs/libcfs.h>
  37
  38 unsigned long cfs_fail_loc = 0;
  39 unsigned int cfs_fail_val = 0;
  40 int cfs_fail_err;
  41 DECLARE_WAIT_QUEUE_HEAD(cfs_race_waitq);
  42 int cfs_race_state;
  43
  44 EXPORT_SYMBOL(cfs_fail_loc);
  45 EXPORT_SYMBOL(cfs_fail_val);
  46 EXPORT_SYMBOL(cfs_fail_err);
  47 EXPORT_SYMBOL(cfs_race_waitq);
  48 EXPORT_SYMBOL(cfs_race_state);
  49
  50 int __cfs_fail_check_set(__u32 id, __u32 value, int set)
  51 {
  52         static atomic_t cfs_fail_count = ATOMIC_INIT(0);
  53
  54         LASSERT(!(id & CFS_FAIL_ONCE));
  55
  56         if ((cfs_fail_loc & (CFS_FAILED | CFS_FAIL_ONCE)) ==
  57             (CFS_FAILED | CFS_FAIL_ONCE)) {
  58                 atomic_set(&cfs_fail_count, 0); /* paranoia */
  59                 return 0;
  60         }
  61
  62         /* Fail 1/cfs_fail_val times */
  63         if (cfs_fail_loc & CFS_FAIL_RAND) {
  64                 if (cfs_fail_val < 2 || cfs_rand() % cfs_fail_val > 0)
  65                         return 0;
  66         }
  67
  68         /* Skip the first cfs_fail_val, then fail */
  69         if (cfs_fail_loc & CFS_FAIL_SKIP) {
  70                 if (atomic_inc_return(&cfs_fail_count) <= cfs_fail_val)
  71                         return 0;
  72         }
  73
  74         /* check cfs_fail_val... */
  75         if (set == CFS_FAIL_LOC_VALUE) {
  76                 if (cfs_fail_val != -1 && cfs_fail_val != value)
  77                         return 0;
  78         }
  79
  80         /* Fail cfs_fail_val times, overridden by FAIL_ONCE */
  81         if (cfs_fail_loc & CFS_FAIL_SOME &&
  82             (!(cfs_fail_loc & CFS_FAIL_ONCE) || cfs_fail_val <= 1)) {
  83                 int count = atomic_inc_return(&cfs_fail_count);
  84
  85                 if (count >= cfs_fail_val) {
  86                         set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  87                         atomic_set(&cfs_fail_count, 0);
  88                         /* we are lost race to increase  */
  89                         if (count > cfs_fail_val)
  90                                 return 0;
  91                 }
  92         }
  93
  94         /* Take into account the current call for FAIL_ONCE for ORSET only,
  95          * as RESET is a new fail_loc, it does not change the current call */
  96         if ((set == CFS_FAIL_LOC_ORSET) && (value & CFS_FAIL_ONCE))
  97                 set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  98         /* Lost race to set CFS_FAILED_BIT. */
  99         if (test_and_set_bit(CFS_FAILED_BIT, &cfs_fail_loc)) {
 100                 /* If CFS_FAIL_ONCE is valid, only one process can fail,
 101                  * otherwise multi-process can fail at the same time. */
 102                 if (cfs_fail_loc & CFS_FAIL_ONCE)
 103                         return 0;
 104         }
 105
 106         switch (set) {
 107                 case CFS_FAIL_LOC_NOSET:
 108                 case CFS_FAIL_LOC_VALUE:
 109                         break;
 110                 case CFS_FAIL_LOC_ORSET:
 111                         cfs_fail_loc |= value & ~(CFS_FAILED | CFS_FAIL_ONCE);
 112                         break;
 113                 case CFS_FAIL_LOC_RESET:
 114                         cfs_fail_loc = value;
 115                         break;
 116                 default:
 117                         LASSERTF(0, "called with bad set %u\n", set);
 118                         break;
 119         }
 120
 121         return 1;
 122 }
 123 EXPORT_SYMBOL(__cfs_fail_check_set);
 124
 125 int __cfs_fail_timeout_set(__u32 id, __u32 value, int ms, int set)
 126 {
 127         int ret = 0;
 128
 129         ret = __cfs_fail_check_set(id, value, set);
 130         if (ret && likely(ms > 0)) {
 131                 CERROR("cfs_fail_timeout id %x sleeping for %dms\n",
 132                        id, ms);
 133                 set_current_state(TASK_UNINTERRUPTIBLE);
 134                 schedule_timeout(cfs_time_seconds(ms) / 1000);
 135                 set_current_state(TASK_RUNNING);
 136                 CERROR("cfs_fail_timeout id %x awake\n", id);
 137         }
 138         return ret;
 139 }
 140 EXPORT_SYMBOL(__cfs_fail_timeout_set);