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             "    -m  Max number of core dumps to allow before exiting.\n" \
  56             "    -l  Create 'ztest.core.N' symlink to core directory.\n" \
  57             "    -h  Print this help message.\n" \
  58             "" >&2
  59 }
  60
  61 function or_die
  62 {
  63         # shellcheck disable=SC2068
  64         $@
  65         # shellcheck disable=SC2181
  66         if [[ $? -ne 0 ]]; then
  67                 # shellcheck disable=SC2145
  68                 echo "Command failed: $@"
  69                 exit 1
  70         fi
  71 }
  72
  73 # core file helpers
  74 origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
  75 coreglob="$(grep -E -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
  76
  77 if [[ $coreglob = "*" ]]; then
  78         echo "Setting core file pattern..."
  79         echo "core" > /proc/sys/kernel/core_pattern
  80         coreglob="$(grep -E -o '^([^|%[:space:]]*)' \
  81             /proc/sys/kernel/core_pattern)*"
  82 fi
  83
  84 function core_file
  85 {
  86         # shellcheck disable=SC2012 disable=2086
  87         printf "%s" "$(ls -tr1 $coreglob 2> /dev/null | head -1)"
  88 }
  89
  90 function core_prog
  91 {
  92         prog=$ZTEST
  93         core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
  94             tr  \' ' ')
  95         # shellcheck disable=SC2076
  96         if [[ "$core_id" =~ "zdb "  ]]; then
  97                 prog=$ZDB
  98         fi
  99         printf "%s" "$prog"
 100 }
 101
 102 function store_core
 103 {
 104         core="$(core_file)"
 105         if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
 106                 df -h "$workdir" >>ztest.out
 107                 coreid=$(date "+zloop-%y%m%d-%H%M%S")
 108                 foundcrashes=$((foundcrashes + 1))
 109
 110                 # zdb debugging
 111                 zdbcmd="$ZDB -U "$workdir/zpool.cache" -dddMmDDG ztest"
 112                 zdbdebug=$($zdbcmd 2>&1)
 113                 echo -e "$zdbcmd\n" >>ztest.zdb
 114                 echo "$zdbdebug" >>ztest.zdb
 115
 116                 dest=$coredir/$coreid
 117                 or_die mkdir -p "$dest"
 118                 or_die mkdir -p "$dest/vdev"
 119
 120                 if [[ $symlink -ne 0 ]]; then
 121                         or_die ln -sf "$dest" ztest.core.$foundcrashes
 122                 fi
 123
 124                 echo "*** ztest crash found - moving logs to $dest"
 125
 126                 or_die mv ztest.history "$dest/"
 127                 or_die mv ztest.zdb "$dest/"
 128                 or_die mv ztest.out "$dest/"
 129                 or_die mv "$workdir/ztest*" "$dest/vdev/"
 130
 131                 if [[ -e "$workdir/zpool.cache" ]]; then
 132                         or_die mv "$workdir/zpool.cache" "$dest/vdev/"
 133                 fi
 134
 135                 # check for core
 136                 if [[ -f "$core" ]]; then
 137                         coreprog=$(core_prog "$core")
 138                         coredebug=$($GDB --batch --quiet \
 139                             -ex "set print thread-events off" \
 140                             -ex "printf \"*\n* Backtrace \n*\n\"" \
 141                             -ex "bt" \
 142                             -ex "printf \"*\n* Libraries \n*\n\"" \
 143                             -ex "info sharedlib" \
 144                             -ex "printf \"*\n* Threads (full) \n*\n\"" \
 145                             -ex "info threads" \
 146                             -ex "printf \"*\n* Backtraces \n*\n\"" \
 147                             -ex "thread apply all bt" \
 148                             -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
 149                             -ex "thread apply all bt full" \
 150                             -ex "quit" "$coreprog" "$core" 2>&1 | \
 151                             grep -v "New LWP")
 152
 153                         # Dump core + logs to stored directory
 154                         echo "$coredebug" >>"$dest/ztest.gdb"
 155                         or_die mv "$core" "$dest/"
 156
 157                         # Record info in cores logfile
 158                         echo "*** core @ $coredir/$coreid/$core:" | \
 159                             tee -a ztest.cores
 160                 fi
 161
 162                 if [[ $coremax -gt 0 ]] &&
 163                    [[ $foundcrashes -ge $coremax ]]; then
 164                         echo "exiting... max $coremax allowed cores"
 165                         exit 1
 166                 else
 167                         echo "continuing..."
 168                 fi
 169         fi
 170 }
 171
 172 # parse arguments
 173 # expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
 174 coredir=$DEFAULTCOREDIR
 175 basedir=$DEFAULTWORKDIR
 176 rundir="zloop-run"
 177 timeout=0
 178 size="512m"
 179 coremax=0
 180 symlink=0
 181 while getopts ":ht:m:s:c:f:l" opt; do
 182         case $opt in
 183                 t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
 184                 m ) [[ $OPTARG -gt 0 ]] && coremax=$OPTARG ;;
 185                 s ) [[ $OPTARG ]] && size=$OPTARG ;;
 186                 c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
 187                 f ) [[ $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
 188                 l ) symlink=1 ;;
 189                 h ) usage
 190                     exit 2
 191                     ;;
 192                 * ) echo "Invalid argument: -$OPTARG";
 193                     usage
 194                     exit 1
 195         esac
 196 done
 197 # pass remaining arguments on to ztest
 198 shift $((OPTIND - 1))
 199
 200 # enable core dumps
 201 ulimit -c unlimited
 202 export ASAN_OPTIONS=abort_on_error=1:disable_coredump=0
 203
 204 if [[ -f "$(core_file)" ]]; then
 205         echo -n "There's a core dump here you might want to look at first... "
 206         core_file
 207         echo
 208         exit 1
 209 fi
 210
 211 if [[ ! -d $coredir ]]; then
 212         echo "core dump directory ($coredir) does not exist, creating it."
 213         or_die mkdir -p "$coredir"
 214 fi
 215
 216 if [[ ! -w $coredir ]]; then
 217         echo "core dump directory ($coredir) is not writable."
 218         exit 1
 219 fi
 220
 221 or_die rm -f ztest.history
 222 or_die rm -f ztest.zdb
 223 or_die rm -f ztest.cores
 224
 225 ztrc=0          # ztest return value
 226 foundcrashes=0  # number of crashes found so far
 227 starttime=$(date +%s)
 228 curtime=$starttime
 229
 230 # if no timeout was specified, loop forever.
 231 while [[ $timeout -eq 0 ]] || [[ $curtime -le $((starttime + timeout)) ]]; do
 232         zopt="-G -VVVVV"
 233
 234         # start each run with an empty directory
 235         workdir="$basedir/$rundir"
 236         or_die rm -rf "$workdir"
 237         or_die mkdir "$workdir"
 238
 239         # switch between common arrangements & fully randomized
 240         if [[ $((RANDOM % 2)) -eq 0 ]]; then
 241                 mirrors=2
 242                 raidz=0
 243                 parity=1
 244                 vdevs=2
 245         else
 246                 mirrors=$(((RANDOM % 3) * 1))
 247                 parity=$(((RANDOM % 3) + 1))
 248                 raidz=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
 249                 vdevs=$(((RANDOM % 3) + 3))
 250         fi
 251         align=$(((RANDOM % 2) * 3 + 9))
 252         runtime=$((RANDOM % 100))
 253         passtime=$((RANDOM % (runtime / 3 + 1) + 10))
 254
 255         zopt="$zopt -m $mirrors"
 256         zopt="$zopt -r $raidz"
 257         zopt="$zopt -R $parity"
 258         zopt="$zopt -v $vdevs"
 259         zopt="$zopt -a $align"
 260         zopt="$zopt -T $runtime"
 261         zopt="$zopt -P $passtime"
 262         zopt="$zopt -s $size"
 263         zopt="$zopt -f $workdir"
 264
 265         # shellcheck disable=SC2124
 266         cmd="$ZTEST $zopt $@"
 267         desc="$(date '+%m/%d %T') $cmd"
 268         echo "$desc" | tee -a ztest.history
 269         echo "$desc" >>ztest.out
 270         $cmd >>ztest.out 2>&1
 271         ztrc=$?
 272         grep -E '===|WARNING' ztest.out >>ztest.history
 273
 274         store_core
 275
 276         curtime=$(date +%s)
 277 done
 278
 279 echo "zloop finished, $foundcrashes crashes found"
 280
 281 #restore core pattern
 282 echo "$origcorepattern" > /proc/sys/kernel/core_pattern
 283
 284 uptime >>ztest.out
 285
 286 if [[ $foundcrashes -gt 0 ]]; then
 287         exit 1
 288 fi