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 BASE_DIR=$(dirname "$0")
  24 SCRIPT_COMMON=common.sh
  25 if [ -f "${BASE_DIR}/${SCRIPT_COMMON}" ]; then
  26         . "${BASE_DIR}/${SCRIPT_COMMON}"
  27 else
  28         echo "Missing helper script ${SCRIPT_COMMON}" && exit 1
  29 fi
  30
  31 # shellcheck disable=SC2034
  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         # shellcheck disable=SC2068
  60         $@
  61         # shellcheck disable=SC2181
  62         if [[ $? -ne 0 ]]; then
  63                 # shellcheck disable=SC2145
  64                 echo "Command failed: $@"
  65                 exit 1
  66         fi
  67 }
  68
  69 # core file helpers
  70 origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
  71 coreglob="$(egrep -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
  72
  73 if [[ $coreglob = "*" ]]; then
  74         echo "Setting core file pattern..."
  75         echo "core" > /proc/sys/kernel/core_pattern
  76         coreglob="$(egrep -o '^([^|%[:space:]]*)' \
  77             /proc/sys/kernel/core_pattern)*"
  78 fi
  79
  80 function core_file
  81 {
  82         # shellcheck disable=SC2012 disable=2086
  83         printf "%s" "$(ls -tr1 $coreglob 2> /dev/null | head -1)"
  84 }
  85
  86 function core_prog
  87 {
  88         prog=$ZTEST
  89         core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
  90             tr  \' ' ')
  91         # shellcheck disable=SC2076
  92         if [[ "$core_id" =~ "zdb "  ]]; then
  93                 prog=$ZDB
  94         fi
  95         printf "%s" "$prog"
  96 }
  97
  98 function store_core
  99 {
 100         core="$(core_file)"
 101         if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
 102                 df -h "$workdir" >>ztest.out
 103                 coreid=$(date "+zloop-%y%m%d-%H%M%S")
 104                 foundcrashes=$((foundcrashes + 1))
 105
 106                 dest=$coredir/$coreid
 107                 or_die mkdir -p "$dest"
 108                 or_die mkdir -p "$dest/vdev"
 109
 110                 echo "*** ztest crash found - moving logs to $dest"
 111
 112                 or_die mv ztest.history "$dest/"
 113                 or_die mv ztest.ddt "$dest/"
 114                 or_die mv ztest.out "$dest/"
 115                 or_die mv "$workdir/ztest*" "$dest/vdev/"
 116                 or_die mv "$workdir/zpool.cache" "$dest/vdev/"
 117
 118                 # check for core
 119                 if [[ -f "$core" ]]; then
 120                         coreprog=$(core_prog "$core")
 121                         corestatus=$($GDB --batch --quiet \
 122                             -ex "set print thread-events off" \
 123                             -ex "printf \"*\n* Backtrace \n*\n\"" \
 124                             -ex "bt" \
 125                             -ex "printf \"*\n* Libraries \n*\n\"" \
 126                             -ex "info sharedlib" \
 127                             -ex "printf \"*\n* Threads (full) \n*\n\"" \
 128                             -ex "info threads" \
 129                             -ex "printf \"*\n* Backtraces \n*\n\"" \
 130                             -ex "thread apply all bt" \
 131                             -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
 132                             -ex "thread apply all bt full" \
 133                             -ex "quit" "$coreprog" "$core" | grep -v "New LWP")
 134
 135                         # Dump core + logs to stored directory
 136                         echo "$corestatus" >>"$dest/status"
 137                         or_die mv "$core" "$dest/"
 138
 139                         # Record info in cores logfile
 140                         echo "*** core @ $coredir/$coreid/$core:" | \
 141                             tee -a ztest.cores
 142                         echo "$corestatus" | tee -a ztest.cores
 143                         echo "" | tee -a ztest.cores
 144                 fi
 145                 echo "continuing..."
 146         fi
 147 }
 148
 149 rngdpid=""
 150 function on_exit
 151 {
 152         if [ -n "$rngdpid" ]; then
 153                 kill -9 "$rngdpid"
 154         fi
 155 }
 156 trap on_exit EXIT
 157
 158 # parse arguments
 159 # expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
 160 coredir=$DEFAULTCOREDIR
 161 basedir=$DEFAULTWORKDIR
 162 rundir="zloop-run"
 163 timeout=0
 164 while getopts ":ht:c:f:" opt; do
 165         case $opt in
 166                 t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
 167                 c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
 168                 f ) [[ $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
 169                 h ) usage
 170                     exit 2
 171                     ;;
 172                 * ) echo "Invalid argument: -$OPTARG";
 173                     usage
 174                     exit 1
 175         esac
 176 done
 177 # pass remaining arguments on to ztest
 178 shift $((OPTIND - 1))
 179
 180 # enable core dumps
 181 ulimit -c unlimited
 182
 183 if [[ -f "$(core_file)" ]]; then
 184         echo -n "There's a core dump here you might want to look at first... "
 185         core_file
 186         exit 1
 187 fi
 188
 189 if [[ ! -d $coredir ]]; then
 190         echo "core dump directory ($coredir) does not exist, creating it."
 191         or_die mkdir -p "$coredir"
 192 fi
 193
 194 if [[ ! -w $coredir ]]; then
 195         echo "core dump directory ($coredir) is not writable."
 196         exit 1
 197 fi
 198
 199 or_die rm -f ztest.history
 200 or_die rm -f ztest.ddt
 201 or_die rm -f ztest.cores
 202
 203 # start rngd in the background so we don't run out of entropy
 204 or_die read -r rngdpid < <(rngd -f -r /dev/urandom & echo $!)
 205
 206 ztrc=0          # ztest return value
 207 foundcrashes=0  # number of crashes found so far
 208 starttime=$(date +%s)
 209 curtime=$starttime
 210
 211 # if no timeout was specified, loop forever.
 212 while [[ $timeout -eq 0 ]] || [[ $curtime -le $((starttime + timeout)) ]]; do
 213         zopt="-VVVVV"
 214
 215         # start each run with an empty directory
 216         workdir="$basedir/$rundir"
 217         or_die rm -rf "$workdir"
 218         or_die mkdir "$workdir"
 219
 220         # switch between common arrangements & fully randomized
 221         if [[ $((RANDOM % 2)) -eq 0 ]]; then
 222                 mirrors=2
 223                 raidz=0
 224                 parity=1
 225                 vdevs=2
 226         else
 227                 mirrors=$(((RANDOM % 3) * 1))
 228                 parity=$(((RANDOM % 3) + 1))
 229                 raidz=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
 230                 vdevs=$(((RANDOM % 3) + 3))
 231         fi
 232         align=$(((RANDOM % 2) * 3 + 9))
 233         runtime=$((RANDOM % 100))
 234         passtime=$((RANDOM % (runtime / 3 + 1) + 10))
 235         size=128m
 236
 237         zopt="$zopt -m $mirrors"
 238         zopt="$zopt -r $raidz"
 239         zopt="$zopt -R $parity"
 240         zopt="$zopt -v $vdevs"
 241         zopt="$zopt -a $align"
 242         zopt="$zopt -T $runtime"
 243         zopt="$zopt -P $passtime"
 244         zopt="$zopt -s $size"
 245         zopt="$zopt -f $workdir"
 246
 247         # shellcheck disable=SC2124
 248         cmd="$ZTEST $zopt $@"
 249         desc="$(date '+%m/%d %T') $cmd"
 250         echo "$desc" | tee -a ztest.history
 251         echo "$desc" >>ztest.out
 252         $cmd >>ztest.out 2>&1
 253         ztrc=$?
 254         egrep '===|WARNING' ztest.out >>ztest.history
 255         $ZDB -U "$workdir/zpool.cache" -DD ztest >>ztest.ddt
 256
 257         store_core
 258
 259         curtime=$(date +%s)
 260 done
 261
 262 echo "zloop finished, $foundcrashes crashes found"
 263
 264 #restore core pattern
 265 echo "$origcorepattern" > /proc/sys/kernel/core_pattern
 266
 267 uptime >>ztest.out
 268
 269 if [[ $foundcrashes -gt 0 ]]; then
 270         exit 1
 271 fi