scripts/zloop.sh

   1 #!/bin/bash
   2
   3 #
   4 # CDDL HEADER START
   5 #
   6 # This file and its contents are supplied under the terms of the
   7 # Common Development and Distribution License ("CDDL"), version 1.0.
   8 # You may only use this file in accordance with the terms of version
   9 # 1.0 of the CDDL.
  10 #
  11 # A full copy of the text of the CDDL should have accompanied this
  12 # source.  A copy of the CDDL is also available via the Internet at
  13 # http://www.illumos.org/license/CDDL.
  14 #
  15 # CDDL HEADER END
  16 #
  17
  18 #
  19 # Copyright (c) 2015 by Delphix. All rights reserved.
  20 # Copyright (C) 2016 Lawrence Livermore National Security, LLC.
  21 #
  22
  23 basedir="$(dirname $0)"
  24
  25 SCRIPT_COMMON=common.sh
  26 if [ -f "${basedir}/${SCRIPT_COMMON}" ]; then
  27         . "${basedir}/${SCRIPT_COMMON}"
  28 else
  29         echo "Missing helper script ${SCRIPT_COMMON}" && exit 1
  30 fi
  31
  32 PROG=zloop.sh
  33
  34 DEFAULTWORKDIR=/var/tmp
  35 DEFAULTCOREDIR=/var/tmp/zloop
  36
  37 function usage
  38 {
  39         echo -e "\n$0 [-t <timeout>] [-c <dump directory>]" \
  40             "[ -- [extra ztest parameters]]\n" \
  41             "\n" \
  42             "  This script runs ztest repeatedly with randomized arguments.\n" \
  43             "  If a crash is encountered, the ztest logs, any associated\n" \
  44             "  vdev files, and core file (if one exists) are moved to the\n" \
  45             "  output directory ($DEFAULTCOREDIR by default). Any options\n" \
  46             "  after the -- end-of-options marker will be passed to ztest.\n" \
  47             "\n" \
  48             "  Options:\n" \
  49             "    -t  Total time to loop for, in seconds. If not provided,\n" \
  50             "        zloop runs forever.\n" \
  51             "    -f  Specify working directory for ztest vdev files.\n" \
  52             "    -c  Specify a core dump directory to use.\n" \
  53             "    -h  Print this help message.\n" \
  54             "" >&2
  55 }
  56
  57 function or_die
  58 {
  59         $@
  60         if [[ $? -ne 0 ]]; then
  61                 echo "Command failed: $@"
  62                 exit 1
  63         fi
  64 }
  65
  66 # core file helpers
  67 origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
  68 coreglob="$(egrep -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
  69
  70 if [[ $coreglob = "*" ]]; then
  71         echo "Setting core file pattern..."
  72         echo "core" > /proc/sys/kernel/core_pattern
  73         coreglob="$(egrep -o '^([^|%[:space:]]*)' \
  74             /proc/sys/kernel/core_pattern)*"
  75 fi
  76
  77 function core_file
  78 {
  79         printf "%s" "$(ls -tr1 $coreglob 2> /dev/null | head -1)"
  80 }
  81
  82 function core_prog
  83 {
  84         prog=$ZTEST
  85         core_id=$($GDB --batch -c $1 | grep "Core was generated by" | \
  86             tr  \' ' ')
  87         if [[ "$core_id" =~ "zdb "  ]]; then
  88                 prog=$ZDB
  89         fi
  90         printf "%s" "$prog"
  91 }
  92
  93 function store_core
  94 {
  95         core="$(core_file)"
  96         if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
  97                 coreid=$(date "+zloop-%y%m%d-%H%M%S")
  98                 foundcrashes=$(($foundcrashes + 1))
  99
 100                 dest=$coredir/$coreid
 101                 or_die mkdir -p $dest
 102                 or_die mkdir -p $dest/vdev
 103
 104                 echo "*** ztest crash found - moving logs to $dest"
 105
 106                 or_die mv ztest.history $dest/
 107                 or_die mv ztest.ddt $dest/
 108                 or_die mv ztest.out $dest/
 109                 or_die mv $workdir/ztest* $dest/vdev/
 110                 or_die mv $workdir/zpool.cache $dest/vdev/
 111
 112                 # check for core
 113                 if [[ -f "$core" ]]; then
 114                         coreprog=$(core_prog $core)
 115                         corestatus=$($GDB --batch --quiet \
 116                             -ex "set print thread-events off" \
 117                             -ex "printf \"*\n* Backtrace \n*\n\"" \
 118                             -ex "bt" \
 119                             -ex "printf \"*\n* Libraries \n*\n\"" \
 120                             -ex "info sharedlib" \
 121                             -ex "printf \"*\n* Threads (full) \n*\n\"" \
 122                             -ex "info threads" \
 123                             -ex "printf \"*\n* Backtraces \n*\n\"" \
 124                             -ex "thread apply all bt" \
 125                             -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
 126                             -ex "thread apply all bt full" \
 127                             -ex "quit" $coreprog "$core" | grep -v "New LWP")
 128
 129                         # Dump core + logs to stored directory
 130                         echo "$corestatus" >>$dest/status
 131                         or_die mv "$core" $dest/
 132
 133                         # Record info in cores logfile
 134                         echo "*** core @ $coredir/$coreid/$core:" | \
 135                             tee -a ztest.cores
 136                         echo "$corestatus" | tee -a ztest.cores
 137                         echo "" | tee -a ztest.cores
 138                 fi
 139                 echo "continuing..."
 140         fi
 141 }
 142
 143 # parse arguments
 144 # expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
 145 coredir=$DEFAULTCOREDIR
 146 workdir=$DEFAULTWORKDIR
 147 timeout=0
 148 while getopts ":ht:c:f:" opt; do
 149         case $opt in
 150                 t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
 151                 c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
 152                 f ) [[ $OPTARG ]] && workdir=$(readlink -f $OPTARG) ;;
 153                 h ) usage
 154                     exit 2
 155                     ;;
 156                 * ) echo "Invalid argument: -$OPTARG";
 157                     usage
 158                     exit 1
 159         esac
 160 done
 161 # pass remaining arguments on to ztest
 162 shift $((OPTIND - 1))
 163
 164 # enable core dumps
 165 ulimit -c unlimited
 166
 167 if [[ -f "$(core_file)" ]]; then
 168         echo -n "There's a core dump here you might want to look at first... "
 169         echo "$(core_file)"
 170         exit 1
 171 fi
 172
 173 if [[ ! -d $coredir ]]; then
 174         echo "core dump directory ($coredir) does not exist, creating it."
 175         or_die mkdir -p $coredir
 176 fi
 177
 178 if [[ ! -w $coredir ]]; then
 179         echo "core dump directory ($coredir) is not writable."
 180         exit 1
 181 fi
 182
 183 or_die rm -f ztest.history
 184 or_die rm -f ztest.ddt
 185 or_die rm -f ztest.cores
 186
 187 ztrc=0          # ztest return value
 188 foundcrashes=0  # number of crashes found so far
 189 starttime=$(date +%s)
 190 curtime=$starttime
 191
 192 # if no timeout was specified, loop forever.
 193 while [[ $timeout -eq 0 ]] || [[ $curtime -le $(($starttime + $timeout)) ]]; do
 194         zopt="-VVVVV"
 195
 196         # switch between common arrangements & fully randomized
 197         if [[ $((RANDOM % 2)) -eq 0 ]]; then
 198                 mirrors=2
 199                 raidz=0
 200                 parity=1
 201                 vdevs=2
 202         else
 203                 mirrors=$(((RANDOM % 3) * 1))
 204                 parity=$(((RANDOM % 3) + 1))
 205                 raidz=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
 206                 vdevs=$(((RANDOM % 3) + 3))
 207         fi
 208         align=$(((RANDOM % 2) * 3 + 9))
 209         runtime=$((RANDOM % 100))
 210         passtime=$((RANDOM % (runtime / 3 + 1) + 10))
 211         size=128m
 212
 213         zopt="$zopt -m $mirrors"
 214         zopt="$zopt -r $raidz"
 215         zopt="$zopt -R $parity"
 216         zopt="$zopt -v $vdevs"
 217         zopt="$zopt -a $align"
 218         zopt="$zopt -T $runtime"
 219         zopt="$zopt -P $passtime"
 220         zopt="$zopt -s $size"
 221         zopt="$zopt -f $workdir"
 222
 223         cmd="$ZTEST $zopt $@"
 224         desc="$(date '+%m/%d %T') $cmd"
 225         echo "$desc" | tee -a ztest.history
 226         echo "$desc" >>ztest.out
 227         $cmd >>ztest.out 2>&1
 228         ztrc=$?
 229         egrep '===|WARNING' ztest.out >>ztest.history
 230         $ZDB -U $workdir/zpool.cache -DD ztest >>ztest.ddt
 231
 232         store_core
 233
 234         curtime=$(date +%s)
 235 done
 236
 237 echo "zloop finished, $foundcrashes crashes found"
 238
 239 #restore core pattern
 240 echo "$origcorepattern" > /proc/sys/kernel/core_pattern
 241
 242 uptime >>ztest.out
 243
 244 if [[ $foundcrashes -gt 0 ]]; then
 245         exit 1
 246 fi