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  * GPL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Use is subject to license terms.
  24  *
  25  * Copyright (c) 2012, 2015, Intel Corporation.
  26  */
  27 /*
  28  * This file is part of Lustre, http://www.lustre.org/
  29  * Lustre is a trademark of Oracle Corporation, Inc.
  30  */
  31
  32 #include <libcfs/libcfs.h>
  33
  34 unsigned long cfs_fail_loc = 0;
  35 unsigned int cfs_fail_val = 0;
  36 int cfs_fail_err;
  37 DECLARE_WAIT_QUEUE_HEAD(cfs_race_waitq);
  38 int cfs_race_state;
  39
  40 EXPORT_SYMBOL(cfs_fail_loc);
  41 EXPORT_SYMBOL(cfs_fail_val);
  42 EXPORT_SYMBOL(cfs_fail_err);
  43 EXPORT_SYMBOL(cfs_race_waitq);
  44 EXPORT_SYMBOL(cfs_race_state);
  45
  46 int __cfs_fail_check_set(__u32 id, __u32 value, int set)
  47 {
  48         static atomic_t cfs_fail_count = ATOMIC_INIT(0);
  49
  50         LASSERT(!(id & CFS_FAIL_ONCE));
  51
  52         if ((cfs_fail_loc & (CFS_FAILED | CFS_FAIL_ONCE)) ==
  53             (CFS_FAILED | CFS_FAIL_ONCE)) {
  54                 atomic_set(&cfs_fail_count, 0); /* paranoia */
  55                 return 0;
  56         }
  57
  58         /* Fail 1/cfs_fail_val times */
  59         if (cfs_fail_loc & CFS_FAIL_RAND) {
  60                 if (cfs_fail_val < 2 || cfs_rand() % cfs_fail_val > 0)
  61                         return 0;
  62         }
  63
  64         /* Skip the first cfs_fail_val, then fail */
  65         if (cfs_fail_loc & CFS_FAIL_SKIP) {
  66                 if (atomic_inc_return(&cfs_fail_count) <= cfs_fail_val)
  67                         return 0;
  68         }
  69
  70         /* check cfs_fail_val... */
  71         if (set == CFS_FAIL_LOC_VALUE) {
  72                 if (cfs_fail_val != -1 && cfs_fail_val != value)
  73                         return 0;
  74         }
  75
  76         /* Fail cfs_fail_val times, overridden by FAIL_ONCE */
  77         if (cfs_fail_loc & CFS_FAIL_SOME &&
  78             (!(cfs_fail_loc & CFS_FAIL_ONCE) || cfs_fail_val <= 1)) {
  79                 int count = atomic_inc_return(&cfs_fail_count);
  80
  81                 if (count >= cfs_fail_val) {
  82                         set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  83                         atomic_set(&cfs_fail_count, 0);
  84                         /* we are lost race to increase  */
  85                         if (count > cfs_fail_val)
  86                                 return 0;
  87                 }
  88         }
  89
  90         /* Take into account the current call for FAIL_ONCE for ORSET only,
  91          * as RESET is a new fail_loc, it does not change the current call */
  92         if ((set == CFS_FAIL_LOC_ORSET) && (value & CFS_FAIL_ONCE))
  93                 set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  94         /* Lost race to set CFS_FAILED_BIT. */
  95         if (test_and_set_bit(CFS_FAILED_BIT, &cfs_fail_loc)) {
  96                 /* If CFS_FAIL_ONCE is valid, only one process can fail,
  97                  * otherwise multi-process can fail at the same time. */
  98                 if (cfs_fail_loc & CFS_FAIL_ONCE)
  99                         return 0;
 100         }
 101
 102         switch (set) {
 103                 case CFS_FAIL_LOC_NOSET:
 104                 case CFS_FAIL_LOC_VALUE:
 105                         break;
 106                 case CFS_FAIL_LOC_ORSET:
 107                         cfs_fail_loc |= value & ~(CFS_FAILED | CFS_FAIL_ONCE);
 108                         break;
 109                 case CFS_FAIL_LOC_RESET:
 110                         cfs_fail_loc = value;
 111                         atomic_set(&cfs_fail_count, 0);
 112                         break;
 113                 default:
 114                         LASSERTF(0, "called with bad set %u\n", set);
 115                         break;
 116         }
 117
 118         return 1;
 119 }
 120 EXPORT_SYMBOL(__cfs_fail_check_set);
 121
 122 int __cfs_fail_timeout_set(__u32 id, __u32 value, int ms, int set)
 123 {
 124         ktime_t till = ktime_add_ms(ktime_get(), ms);
 125         int ret = 0;
 126
 127         ret = __cfs_fail_check_set(id, value, set);
 128         if (ret && likely(ms > 0)) {
 129                 CERROR("cfs_fail_timeout id %x sleeping for %dms\n", id, ms);
 130                 while (ktime_before(ktime_get(), till)) {
 131                         set_current_state(TASK_UNINTERRUPTIBLE);
 132                         schedule_timeout(msecs_to_jiffies(1000) / 10);
 133                         set_current_state(TASK_RUNNING);
 134                         if (!cfs_fail_loc) {
 135                                 CERROR("cfs_fail_timeout interrupted\n");
 136                                 break;
 137                         }
 138                 }
 139                 if (cfs_fail_loc)
 140                         CERROR("cfs_fail_timeout id %x awake\n", id);
 141         }
 142         return ret;
 143 }
 144 EXPORT_SYMBOL(__cfs_fail_timeout_set);