scripts/zloop.sh

   1 #!/usr/bin/env 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 # Copyright (c) 2017, Intel Corporation.
  22 #
  23
  24 BASE_DIR=$(dirname "$0")
  25 SCRIPT_COMMON=common.sh
  26 if [ -f "${BASE_DIR}/${SCRIPT_COMMON}" ]; then
  27         . "${BASE_DIR}/${SCRIPT_COMMON}"
  28 else
  29         echo "Missing helper script ${SCRIPT_COMMON}" && exit 1
  30 fi
  31
  32 # shellcheck disable=SC2034
  33 PROG=zloop.sh
  34 GDB=${GDB:-gdb}
  35
  36 DEFAULTWORKDIR=/var/tmp
  37 DEFAULTCOREDIR=/var/tmp/zloop
  38
  39 function usage
  40 {
  41         cat >&2 <<EOF
  42
  43 $0 [-hl] [-c <dump directory>] [-f <vdev directory>]
  44   [-m <max core dumps>] [-s <vdev size>] [-t <timeout>]
  45   [-I <max iterations>] [-- [extra ztest parameters]]
  46
  47   This script runs ztest repeatedly with randomized arguments.
  48   If a crash is encountered, the ztest logs, any associated
  49   vdev files, and core file (if one exists) are moved to the
  50   output directory ($DEFAULTCOREDIR by default). Any options
  51   after the -- end-of-options marker will be passed to ztest.
  52
  53   Options:
  54     -c  Specify a core dump directory to use.
  55     -f  Specify working directory for ztest vdev files.
  56     -h  Print this help message.
  57     -l  Create 'ztest.core.N' symlink to core directory.
  58     -m  Max number of core dumps to allow before exiting.
  59     -s  Size of vdev devices.
  60     -t  Total time to loop for, in seconds. If not provided,
  61         zloop runs forever.
  62     -I  Max number of iterations to loop before exiting.
  63
  64 EOF
  65 }
  66
  67 function or_die
  68 {
  69         # shellcheck disable=SC2068
  70         if ! $@; then
  71                 echo "Command failed: $*"
  72                 exit 1
  73         fi
  74 }
  75
  76 case $(uname) in
  77 FreeBSD)
  78         coreglob="z*.core"
  79         ;;
  80 Linux)
  81         # core file helpers
  82         origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
  83         coreglob="$(grep -E -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
  84
  85         if [[ $coreglob = "*" ]]; then
  86                 echo "Setting core file pattern..."
  87                 echo "core" > /proc/sys/kernel/core_pattern
  88                 coreglob="$(grep -E -o '^([^|%[:space:]]*)' \
  89                     /proc/sys/kernel/core_pattern)*"
  90         fi
  91         ;;
  92 *)
  93         exit 1
  94         ;;
  95 esac
  96
  97 function core_file
  98 {
  99         # shellcheck disable=SC2012,SC2086
 100         ls -tr1 $coreglob 2>/dev/null | head -1
 101 }
 102
 103 function core_prog
 104 {
 105         prog=$ZTEST
 106         core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
 107             tr  \' ' ')
 108         if [[ "$core_id" == *"zdb "* ]]; then
 109                 prog=$ZDB
 110         fi
 111         printf "%s" "$prog"
 112 }
 113
 114 function store_core
 115 {
 116         core="$(core_file)"
 117         if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
 118                 df -h "$workdir" >>ztest.out
 119                 coreid=$(date "+zloop-%y%m%d-%H%M%S")
 120                 foundcrashes=$((foundcrashes + 1))
 121
 122                 # zdb debugging
 123                 zdbcmd="$ZDB -U "$workdir/zpool.cache" -dddMmDDG ztest"
 124                 zdbdebug=$($zdbcmd 2>&1)
 125                 echo -e "$zdbcmd\n" >>ztest.zdb
 126                 echo "$zdbdebug" >>ztest.zdb
 127
 128                 dest=$coredir/$coreid
 129                 or_die mkdir -p "$dest"
 130                 or_die mkdir -p "$dest/vdev"
 131
 132                 if [[ $symlink -ne 0 ]]; then
 133                         or_die ln -sf "$dest" ztest.core.$foundcrashes
 134                 fi
 135
 136                 echo "*** ztest crash found - moving logs to $dest"
 137
 138                 or_die mv ztest.history "$dest/"
 139                 or_die mv ztest.zdb "$dest/"
 140                 or_die mv ztest.out "$dest/"
 141                 or_die mv "$workdir/ztest*" "$dest/vdev/"
 142
 143                 if [[ -e "$workdir/zpool.cache" ]]; then
 144                         or_die mv "$workdir/zpool.cache" "$dest/vdev/"
 145                 fi
 146
 147                 # check for core
 148                 if [[ -f "$core" ]]; then
 149                         coreprog=$(core_prog "$core")
 150                         coredebug=$($GDB --batch --quiet \
 151                             -ex "set print thread-events off" \
 152                             -ex "printf \"*\n* Backtrace \n*\n\"" \
 153                             -ex "bt" \
 154                             -ex "printf \"*\n* Libraries \n*\n\"" \
 155                             -ex "info sharedlib" \
 156                             -ex "printf \"*\n* Threads (full) \n*\n\"" \
 157                             -ex "info threads" \
 158                             -ex "printf \"*\n* Backtraces \n*\n\"" \
 159                             -ex "thread apply all bt" \
 160                             -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
 161                             -ex "thread apply all bt full" \
 162                             -ex "quit" "$coreprog" "$core" 2>&1 | \
 163                             grep -v "New LWP")
 164
 165                         # Dump core + logs to stored directory
 166                         echo "$coredebug" >>"$dest/ztest.gdb"
 167                         or_die mv "$core" "$dest/"
 168
 169                         # Record info in cores logfile
 170                         echo "*** core @ $coredir/$coreid/$core:" | \
 171                             tee -a ztest.cores
 172                 fi
 173
 174                 if [[ $coremax -gt 0 ]] &&
 175                    [[ $foundcrashes -ge $coremax ]]; then
 176                         echo "exiting... max $coremax allowed cores"
 177                         exit 1
 178                 else
 179                         echo "continuing..."
 180                 fi
 181         fi
 182 }
 183
 184 # parse arguments
 185 # expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
 186 coredir=$DEFAULTCOREDIR
 187 basedir=$DEFAULTWORKDIR
 188 rundir="zloop-run"
 189 timeout=0
 190 size="512m"
 191 coremax=0
 192 symlink=0
 193 iterations=0
 194 while getopts ":ht:m:I:s:c:f:l" opt; do
 195         case $opt in
 196                 t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
 197                 m ) [[ $OPTARG -gt 0 ]] && coremax=$OPTARG ;;
 198                 I ) [[ $OPTARG ]] && iterations=$OPTARG ;;
 199                 s ) [[ $OPTARG ]] && size=$OPTARG ;;
 200                 c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
 201                 f ) [[ $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
 202                 l ) symlink=1 ;;
 203                 h ) usage
 204                     exit 2
 205                     ;;
 206                 * ) echo "Invalid argument: -$OPTARG";
 207                     usage
 208                     exit 1
 209         esac
 210 done
 211 # pass remaining arguments on to ztest
 212 shift $((OPTIND - 1))
 213
 214 # enable core dumps
 215 ulimit -c unlimited
 216 export ASAN_OPTIONS=abort_on_error=1:disable_coredump=0
 217
 218 if [[ -f "$(core_file)" ]]; then
 219         echo -n "There's a core dump here you might want to look at first... "
 220         core_file
 221         echo
 222         exit 1
 223 fi
 224
 225 if [[ ! -d $coredir ]]; then
 226         echo "core dump directory ($coredir) does not exist, creating it."
 227         or_die mkdir -p "$coredir"
 228 fi
 229
 230 if [[ ! -w $coredir ]]; then
 231         echo "core dump directory ($coredir) is not writable."
 232         exit 1
 233 fi
 234
 235 or_die rm -f ztest.history
 236 or_die rm -f ztest.zdb
 237 or_die rm -f ztest.cores
 238
 239 ztrc=0          # ztest return value
 240 foundcrashes=0  # number of crashes found so far
 241 starttime=$(date +%s)
 242 curtime=$starttime
 243 iteration=0
 244
 245 # if no timeout was specified, loop forever.
 246 while (( timeout == 0 )) || (( curtime <= (starttime + timeout) )); do
 247         if (( iterations > 0 )) && (( iteration++ == iterations )); then
 248                 break
 249         fi
 250
 251         zopt="-G -VVVVV"
 252
 253         # start each run with an empty directory
 254         workdir="$basedir/$rundir"
 255         or_die rm -rf "$workdir"
 256         or_die mkdir "$workdir"
 257
 258         # switch between three types of configs
 259         # 1/3 basic, 1/3 raidz mix, and 1/3 draid mix
 260         choice=$((RANDOM % 3))
 261
 262         # ashift range 9 - 15
 263         align=$(((RANDOM % 2) * 3 + 9))
 264
 265         # randomly use special classes
 266         class="special=random"
 267
 268         if [[ $choice -eq 0 ]]; then
 269                 # basic mirror only
 270                 parity=1
 271                 mirrors=2
 272                 draid_data=0
 273                 draid_spares=0
 274                 raid_children=0
 275                 vdevs=2
 276                 raid_type="raidz"
 277         elif [[ $choice -eq 1 ]]; then
 278                 # fully randomized mirror/raidz (sans dRAID)
 279                 parity=$(((RANDOM % 3) + 1))
 280                 mirrors=$(((RANDOM % 3) * 1))
 281                 draid_data=0
 282                 draid_spares=0
 283                 raid_children=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
 284                 vdevs=$(((RANDOM % 3) + 3))
 285                 raid_type="raidz"
 286         else
 287                 # fully randomized dRAID (sans mirror/raidz)
 288                 parity=$(((RANDOM % 3) + 1))
 289                 mirrors=0
 290                 draid_data=$(((RANDOM % 8) + 3))
 291                 draid_spares=$(((RANDOM % 2) + parity))
 292                 stripe=$((draid_data + parity))
 293                 extra=$((draid_spares + (RANDOM % 4)))
 294                 raid_children=$(((((RANDOM % 4) + 1) * stripe) + extra))
 295                 vdevs=$((RANDOM % 3))
 296                 raid_type="draid"
 297         fi
 298
 299         zopt="$zopt -K $raid_type"
 300         zopt="$zopt -m $mirrors"
 301         zopt="$zopt -r $raid_children"
 302         zopt="$zopt -D $draid_data"
 303         zopt="$zopt -S $draid_spares"
 304         zopt="$zopt -R $parity"
 305         zopt="$zopt -v $vdevs"
 306         zopt="$zopt -a $align"
 307         zopt="$zopt -C $class"
 308         zopt="$zopt -s $size"
 309         zopt="$zopt -f $workdir"
 310
 311         cmd="$ZTEST $zopt $*"
 312         desc="$(date '+%m/%d %T') $cmd"
 313         echo "$desc" | tee -a ztest.history
 314         echo "$desc" >>ztest.out
 315         $cmd >>ztest.out 2>&1
 316         ztrc=$?
 317         grep -E '===|WARNING' ztest.out >>ztest.history
 318
 319         store_core
 320
 321         curtime=$(date +%s)
 322 done
 323
 324 echo "zloop finished, $foundcrashes crashes found"
 325
 326 # restore core pattern.
 327 case $(uname) in
 328 Linux)
 329         echo "$origcorepattern" > /proc/sys/kernel/core_pattern
 330         ;;
 331 *)
 332         ;;
 333 esac
 334
 335 uptime >>ztest.out
 336
 337 if [[ $foundcrashes -gt 0 ]]; then
 338         exit 1
 339 fi