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/types.h>
  33 #include <linux/slab.h>
  34 #include <linux/module.h>
  35 #include <linux/random.h>
  36 #include <libcfs/libcfs.h>
  37
  38 unsigned long cfs_fail_loc;
  39 EXPORT_SYMBOL(cfs_fail_loc);
  40
  41 unsigned int cfs_fail_val;
  42 EXPORT_SYMBOL(cfs_fail_val);
  43
  44 int cfs_fail_err;
  45 EXPORT_SYMBOL(cfs_fail_err);
  46
  47 DECLARE_WAIT_QUEUE_HEAD(cfs_race_waitq);
  48 EXPORT_SYMBOL(cfs_race_waitq);
  49
  50 int cfs_race_state;
  51 EXPORT_SYMBOL(cfs_race_state);
  52
  53 int __cfs_fail_check_set(u32 id, u32 value, int set)
  54 {
  55         static atomic_t cfs_fail_count = ATOMIC_INIT(0);
  56
  57         LASSERT(!(id & CFS_FAIL_ONCE));
  58
  59         if ((cfs_fail_loc & (CFS_FAILED | CFS_FAIL_ONCE)) ==
  60             (CFS_FAILED | CFS_FAIL_ONCE)) {
  61                 atomic_set(&cfs_fail_count, 0); /* paranoia */
  62                 return 0;
  63         }
  64
  65         /* Fail 1/cfs_fail_val times */
  66         if (cfs_fail_loc & CFS_FAIL_RAND) {
  67                 if (cfs_fail_val < 2 || prandom_u32_max(cfs_fail_val) > 0)
  68                         return 0;
  69         }
  70
  71         /* Skip the first cfs_fail_val, then fail */
  72         if (cfs_fail_loc & CFS_FAIL_SKIP) {
  73                 if (atomic_inc_return(&cfs_fail_count) <= cfs_fail_val)
  74                         return 0;
  75         }
  76
  77         /* check cfs_fail_val... */
  78         if (set == CFS_FAIL_LOC_VALUE) {
  79                 if (cfs_fail_val != -1 && cfs_fail_val != value)
  80                         return 0;
  81         }
  82
  83         /* Fail cfs_fail_val times, overridden by FAIL_ONCE */
  84         if (cfs_fail_loc & CFS_FAIL_SOME &&
  85             (!(cfs_fail_loc & CFS_FAIL_ONCE) || cfs_fail_val <= 1)) {
  86                 int count = atomic_inc_return(&cfs_fail_count);
  87
  88                 if (count >= cfs_fail_val) {
  89                         set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
  90                         atomic_set(&cfs_fail_count, 0);
  91                         /* we are lost race to increase  */
  92                         if (count > cfs_fail_val)
  93                                 return 0;
  94                 }
  95         }
  96
  97         /* Take into account the current call for FAIL_ONCE for ORSET only,
  98          * as RESET is a new fail_loc, it does not change the current call
  99          */
 100         if ((set == CFS_FAIL_LOC_ORSET) && (value & CFS_FAIL_ONCE))
 101                 set_bit(CFS_FAIL_ONCE_BIT, &cfs_fail_loc);
 102         /* Lost race to set CFS_FAILED_BIT. */
 103         if (test_and_set_bit(CFS_FAILED_BIT, &cfs_fail_loc)) {
 104                 /* If CFS_FAIL_ONCE is valid, only one process can fail,
 105                  * otherwise multi-process can fail at the same time.
 106                  */
 107                 if (cfs_fail_loc & CFS_FAIL_ONCE)
 108                         return 0;
 109         }
 110
 111         switch (set) {
 112         case CFS_FAIL_LOC_NOSET:
 113         case CFS_FAIL_LOC_VALUE:
 114                 break;
 115         case CFS_FAIL_LOC_ORSET:
 116                 cfs_fail_loc |= value & ~(CFS_FAILED | CFS_FAIL_ONCE);
 117                 break;
 118         case CFS_FAIL_LOC_RESET:
 119                 cfs_fail_loc = value;
 120                 atomic_set(&cfs_fail_count, 0);
 121                 break;
 122         default:
 123                 LASSERTF(0, "called with bad set %u\n", set);
 124                 break;
 125         }
 126
 127         return 1;
 128 }
 129 EXPORT_SYMBOL(__cfs_fail_check_set);
 130
 131 int __cfs_fail_timeout_set(u32 id, u32 value, int ms, int set)
 132 {
 133         ktime_t till = ktime_add_ms(ktime_get(), ms);
 134         int ret;
 135
 136         ret = __cfs_fail_check_set(id, value, set);
 137         if (ret && likely(ms > 0)) {
 138                 CERROR("cfs_fail_timeout id %x sleeping for %dms\n", id, ms);
 139                 while (ktime_before(ktime_get(), till)) {
 140                         schedule_timeout_uninterruptible(cfs_time_seconds(1)
 141                                                          / 10);
 142                         set_current_state(TASK_RUNNING);
 143                         if (!cfs_fail_loc) {
 144                                 CERROR("cfs_fail_timeout interrupted\n");
 145                                 break;
 146                         }
 147                 }
 148                 if (cfs_fail_loc)
 149                         CERROR("cfs_fail_timeout id %x awake\n", id);
 150         }
 151         return ret;
 152 }
 153 EXPORT_SYMBOL(__cfs_fail_timeout_set);