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