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 ${SCRIPT_COMMON}" && exit 1
  29 fi
  30
  31 # shellcheck disable=SC2034
  32 PROG=zloop.sh
  33 GDB=${GDB:-gdb}
  34
  35 DEFAULTWORKDIR=/var/tmp
  36 DEFAULTCOREDIR=/var/tmp/zloop
  37
  38 function usage
  39 {
  40         echo -e "\n$0 [-t <timeout>] [ -s <vdev size> ] [-c <dump directory>]" \
  41             "[ -- [extra ztest parameters]]\n" \
  42             "\n" \
  43             "  This script runs ztest repeatedly with randomized arguments.\n" \
  44             "  If a crash is encountered, the ztest logs, any associated\n" \
  45             "  vdev files, and core file (if one exists) are moved to the\n" \
  46             "  output directory ($DEFAULTCOREDIR by default). Any options\n" \
  47             "  after the -- end-of-options marker will be passed to ztest.\n" \
  48             "\n" \
  49             "  Options:\n" \
  50             "    -t  Total time to loop for, in seconds. If not provided,\n" \
  51             "        zloop runs forever.\n" \
  52             "    -s  Size of vdev devices.\n" \
  53             "    -f  Specify working directory for ztest vdev files.\n" \
  54             "    -c  Specify a core dump directory to use.\n" \
  55             "    -h  Print this help message.\n" \
  56             "" >&2
  57 }
  58
  59 function or_die
  60 {
  61         # shellcheck disable=SC2068
  62         $@
  63         # shellcheck disable=SC2181
  64         if [[ $? -ne 0 ]]; then
  65                 # shellcheck disable=SC2145
  66                 echo "Command failed: $@"
  67                 exit 1
  68         fi
  69 }
  70
  71 # core file helpers
  72 origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
  73 coreglob="$(egrep -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
  74
  75 if [[ $coreglob = "*" ]]; then
  76         echo "Setting core file pattern..."
  77         echo "core" > /proc/sys/kernel/core_pattern
  78         coreglob="$(egrep -o '^([^|%[:space:]]*)' \
  79             /proc/sys/kernel/core_pattern)*"
  80 fi
  81
  82 function core_file
  83 {
  84         # shellcheck disable=SC2012 disable=2086
  85         printf "%s" "$(ls -tr1 $coreglob 2> /dev/null | head -1)"
  86 }
  87
  88 function core_prog
  89 {
  90         prog=$ZTEST
  91         core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
  92             tr  \' ' ')
  93         # shellcheck disable=SC2076
  94         if [[ "$core_id" =~ "zdb "  ]]; then
  95                 prog=$ZDB
  96         fi
  97         printf "%s" "$prog"
  98 }
  99
 100 function store_core
 101 {
 102         core="$(core_file)"
 103         if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
 104                 df -h "$workdir" >>ztest.out
 105                 coreid=$(date "+zloop-%y%m%d-%H%M%S")
 106                 foundcrashes=$((foundcrashes + 1))
 107
 108                 dest=$coredir/$coreid
 109                 or_die mkdir -p "$dest"
 110                 or_die mkdir -p "$dest/vdev"
 111
 112                 echo "*** ztest crash found - moving logs to $dest"
 113
 114                 or_die mv ztest.history "$dest/"
 115                 or_die mv ztest.ddt "$dest/"
 116                 or_die mv ztest.out "$dest/"
 117                 or_die mv "$workdir/ztest*" "$dest/vdev/"
 118                 or_die mv "$workdir/zpool.cache" "$dest/vdev/"
 119
 120                 # check for core
 121                 if [[ -f "$core" ]]; then
 122                         coreprog=$(core_prog "$core")
 123                         corestatus=$($GDB --batch --quiet \
 124                             -ex "set print thread-events off" \
 125                             -ex "printf \"*\n* Backtrace \n*\n\"" \
 126                             -ex "bt" \
 127                             -ex "printf \"*\n* Libraries \n*\n\"" \
 128                             -ex "info sharedlib" \
 129                             -ex "printf \"*\n* Threads (full) \n*\n\"" \
 130                             -ex "info threads" \
 131                             -ex "printf \"*\n* Backtraces \n*\n\"" \
 132                             -ex "thread apply all bt" \
 133                             -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
 134                             -ex "thread apply all bt full" \
 135                             -ex "quit" "$coreprog" "$core" | grep -v "New LWP")
 136
 137                         # Dump core + logs to stored directory
 138                         echo "$corestatus" >>"$dest/status"
 139                         or_die mv "$core" "$dest/"
 140
 141                         # Record info in cores logfile
 142                         echo "*** core @ $coredir/$coreid/$core:" | \
 143                             tee -a ztest.cores
 144                         echo "$corestatus" | tee -a ztest.cores
 145                         echo "" | tee -a ztest.cores
 146                 fi
 147                 echo "continuing..."
 148         fi
 149 }
 150
 151 rngdpid=""
 152 function on_exit
 153 {
 154         if [ -n "$rngdpid" ]; then
 155                 kill -9 "$rngdpid"
 156         fi
 157 }
 158 trap on_exit EXIT
 159
 160 # parse arguments
 161 # expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
 162 coredir=$DEFAULTCOREDIR
 163 basedir=$DEFAULTWORKDIR
 164 rundir="zloop-run"
 165 timeout=0
 166 size="512m"
 167 while getopts ":ht:s:c:f:" opt; do
 168         case $opt in
 169                 t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
 170                 s ) [[ $OPTARG ]] && size=$OPTARG ;;
 171                 c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
 172                 f ) [[ $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
 173                 h ) usage
 174                     exit 2
 175                     ;;
 176                 * ) echo "Invalid argument: -$OPTARG";
 177                     usage
 178                     exit 1
 179         esac
 180 done
 181 # pass remaining arguments on to ztest
 182 shift $((OPTIND - 1))
 183
 184 # enable core dumps
 185 ulimit -c unlimited
 186 export ASAN_OPTIONS=abort_on_error=1:disable_coredump=0
 187
 188 if [[ -f "$(core_file)" ]]; then
 189         echo -n "There's a core dump here you might want to look at first... "
 190         core_file
 191         exit 1
 192 fi
 193
 194 if [[ ! -d $coredir ]]; then
 195         echo "core dump directory ($coredir) does not exist, creating it."
 196         or_die mkdir -p "$coredir"
 197 fi
 198
 199 if [[ ! -w $coredir ]]; then
 200         echo "core dump directory ($coredir) is not writable."
 201         exit 1
 202 fi
 203
 204 or_die rm -f ztest.history
 205 or_die rm -f ztest.ddt
 206 or_die rm -f ztest.cores
 207
 208 # start rngd in the background so we don't run out of entropy
 209 or_die read -r rngdpid < <(rngd -f -r /dev/urandom & echo $!)
 210
 211 ztrc=0          # ztest return value
 212 foundcrashes=0  # number of crashes found so far
 213 starttime=$(date +%s)
 214 curtime=$starttime
 215
 216 # if no timeout was specified, loop forever.
 217 while [[ $timeout -eq 0 ]] || [[ $curtime -le $((starttime + timeout)) ]]; do
 218         zopt="-VVVVV"
 219
 220         # start each run with an empty directory
 221         workdir="$basedir/$rundir"
 222         or_die rm -rf "$workdir"
 223         or_die mkdir "$workdir"
 224
 225         # switch between common arrangements & fully randomized
 226         if [[ $((RANDOM % 2)) -eq 0 ]]; then
 227                 mirrors=2
 228                 raidz=0
 229                 parity=1
 230                 vdevs=2
 231         else
 232                 mirrors=$(((RANDOM % 3) * 1))
 233                 parity=$(((RANDOM % 3) + 1))
 234                 raidz=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
 235                 vdevs=$(((RANDOM % 3) + 3))
 236         fi
 237         align=$(((RANDOM % 2) * 3 + 9))
 238         runtime=$((RANDOM % 100))
 239         passtime=$((RANDOM % (runtime / 3 + 1) + 10))
 240
 241         zopt="$zopt -m $mirrors"
 242         zopt="$zopt -r $raidz"
 243         zopt="$zopt -R $parity"
 244         zopt="$zopt -v $vdevs"
 245         zopt="$zopt -a $align"
 246         zopt="$zopt -T $runtime"
 247         zopt="$zopt -P $passtime"
 248         zopt="$zopt -s $size"
 249         zopt="$zopt -f $workdir"
 250
 251         # shellcheck disable=SC2124
 252         cmd="$ZTEST $zopt $@"
 253         desc="$(date '+%m/%d %T') $cmd"
 254         echo "$desc" | tee -a ztest.history
 255         echo "$desc" >>ztest.out
 256         $cmd >>ztest.out 2>&1
 257         ztrc=$?
 258         egrep '===|WARNING' ztest.out >>ztest.history
 259         $ZDB -U "$workdir/zpool.cache" -DD ztest >>ztest.ddt 2>&1
 260
 261         store_core
 262
 263         curtime=$(date +%s)
 264 done
 265
 266 echo "zloop finished, $foundcrashes crashes found"
 267
 268 #restore core pattern
 269 echo "$origcorepattern" > /proc/sys/kernel/core_pattern
 270
 271 uptime >>ztest.out
 272
 273 if [[ $foundcrashes -gt 0 ]]; then
 274         exit 1
 275 fi