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 <linux/random.h>
  33 #include <libcfs/libcfs.h>
  34
  35 unsigned long cfs_fail_loc = 0;
  36 unsigned int cfs_fail_val = 0;
  37 int cfs_fail_err;
  38 DECLARE_WAIT_QUEUE_HEAD(cfs_race_waitq);
  39 int cfs_race_state;
  40
  41 EXPORT_SYMBOL(cfs_fail_loc);
  42 EXPORT_SYMBOL(cfs_fail_val);
  43 EXPORT_SYMBOL(cfs_fail_err);
  44 EXPORT_SYMBOL(cfs_race_waitq);
  45 EXPORT_SYMBOL(cfs_race_state);
  46
  47 int __cfs_fail_check_set(__u32 id, __u32 value, int set)
  48 {
  49         static atomic_t cfs_fail_count = ATOMIC_INIT(0);
  50
  51         LASSERT(!(id & CFS_FAIL_ONCE));
  52
  53         if ((cfs_fail_loc & (CFS_FAILED | CFS_FAIL_ONCE)) ==
  54             (CFS_FAILED | CFS_FAIL_ONCE)) {
  55                 atomic_set(&cfs_fail_count, 0); /* paranoia */
  56                 return 0;
  57         }
  58
  59         /* Fail 1/cfs_fail_val times */
  60         if (cfs_fail_loc & CFS_FAIL_RAND) {
  61                 if (cfs_fail_val < 2 || prandom_u32_max(cfs_fail_val) > 0)
  62                         return 0;
  63         }
  64
  65         /* Skip the first cfs_fail_val, then fail */
  66         if (cfs_fail_loc & CFS_FAIL_SKIP) {
  67                 if (atomic_inc_return(&cfs_fail_count) <= cfs_fail_val)
  68                         return 0;
  69         }
  70
  71         /* check cfs_fail_val... */
  72         if (set == CFS_FAIL_LOC_VALUE) {
  73                 if (cfs_fail_val != -1 && cfs_fail_val != value)
  74                         return 0;
  75         }
  76
  77         /* Fail cfs_fail_val times, overridden by FAIL_ONCE */
  78         if (cfs_fail_loc & CFS_FAIL_SOME &&
  79             (!(cfs_fail_loc & CFS_FAIL_ONCE) || cfs_fail_val <= 1)) {
  80                 int count = atomic_inc_return(&cfs_fail_count);
  81
  82                 if (count >= cfs_fail_val) {
  83                         set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  84                         atomic_set(&cfs_fail_count, 0);
  85                         /* we are lost race to increase  */
  86                         if (count > cfs_fail_val)
  87                                 return 0;
  88                 }
  89         }
  90
  91         /* Take into account the current call for FAIL_ONCE for ORSET only,
  92          * as RESET is a new fail_loc, it does not change the current call */
  93         if ((set == CFS_FAIL_LOC_ORSET) && (value & CFS_FAIL_ONCE))
  94                 set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  95         /* Lost race to set CFS_FAILED_BIT. */
  96         if (test_and_set_bit(CFS_FAILED_BIT, &cfs_fail_loc)) {
  97                 /* If CFS_FAIL_ONCE is valid, only one process can fail,
  98                  * otherwise multi-process can fail at the same time. */
  99                 if (cfs_fail_loc & CFS_FAIL_ONCE)
 100                         return 0;
 101         }
 102
 103         switch (set) {
 104                 case CFS_FAIL_LOC_NOSET:
 105                 case CFS_FAIL_LOC_VALUE:
 106                         break;
 107                 case CFS_FAIL_LOC_ORSET:
 108                         cfs_fail_loc |= value & ~(CFS_FAILED | CFS_FAIL_ONCE);
 109                         break;
 110                 case CFS_FAIL_LOC_RESET:
 111                         cfs_fail_loc = value;
 112                         atomic_set(&cfs_fail_count, 0);
 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         ktime_t till = ktime_add_ms(ktime_get(), ms);
 126         int ret = 0;
 127
 128         ret = __cfs_fail_check_set(id, value, set);
 129         if (ret && likely(ms > 0)) {
 130                 CERROR("cfs_fail_timeout id %x sleeping for %dms\n", id, ms);
 131                 while (ktime_before(ktime_get(), till)) {
 132                         set_current_state(TASK_UNINTERRUPTIBLE);
 133                         schedule_timeout(msecs_to_jiffies(1000) / 10);
 134                         set_current_state(TASK_RUNNING);
 135                         if (!cfs_fail_loc) {
 136                                 CERROR("cfs_fail_timeout interrupted\n");
 137                                 break;
 138                         }
 139                 }
 140                 if (cfs_fail_loc)
 141                         CERROR("cfs_fail_timeout id %x awake\n", id);
 142         }
 143         return ret;
 144 }
 145 EXPORT_SYMBOL(__cfs_fail_timeout_set);