xref: /freebsd/sys/contrib/openzfs/scripts/zloop.sh (revision eda14cbc264d6969b02f2b1994cef11148e914f1)
1*eda14cbcSMatt Macy#!/usr/bin/env bash
2*eda14cbcSMatt Macy
3*eda14cbcSMatt Macy#
4*eda14cbcSMatt Macy# CDDL HEADER START
5*eda14cbcSMatt Macy#
6*eda14cbcSMatt Macy# This file and its contents are supplied under the terms of the
7*eda14cbcSMatt Macy# Common Development and Distribution License ("CDDL"), version 1.0.
8*eda14cbcSMatt Macy# You may only use this file in accordance with the terms of version
9*eda14cbcSMatt Macy# 1.0 of the CDDL.
10*eda14cbcSMatt Macy#
11*eda14cbcSMatt Macy# A full copy of the text of the CDDL should have accompanied this
12*eda14cbcSMatt Macy# source.  A copy of the CDDL is also available via the Internet at
13*eda14cbcSMatt Macy# http://www.illumos.org/license/CDDL.
14*eda14cbcSMatt Macy#
15*eda14cbcSMatt Macy# CDDL HEADER END
16*eda14cbcSMatt Macy#
17*eda14cbcSMatt Macy
18*eda14cbcSMatt Macy#
19*eda14cbcSMatt Macy# Copyright (c) 2015 by Delphix. All rights reserved.
20*eda14cbcSMatt Macy# Copyright (C) 2016 Lawrence Livermore National Security, LLC.
21*eda14cbcSMatt Macy#
22*eda14cbcSMatt Macy
23*eda14cbcSMatt MacyBASE_DIR=$(dirname "$0")
24*eda14cbcSMatt MacySCRIPT_COMMON=common.sh
25*eda14cbcSMatt Macyif [ -f "${BASE_DIR}/${SCRIPT_COMMON}" ]; then
26*eda14cbcSMatt Macy	. "${BASE_DIR}/${SCRIPT_COMMON}"
27*eda14cbcSMatt Macyelse
28*eda14cbcSMatt Macy	echo "Missing helper script ${SCRIPT_COMMON}" && exit 1
29*eda14cbcSMatt Macyfi
30*eda14cbcSMatt Macy
31*eda14cbcSMatt Macy# shellcheck disable=SC2034
32*eda14cbcSMatt MacyPROG=zloop.sh
33*eda14cbcSMatt MacyGDB=${GDB:-gdb}
34*eda14cbcSMatt Macy
35*eda14cbcSMatt MacyDEFAULTWORKDIR=/var/tmp
36*eda14cbcSMatt MacyDEFAULTCOREDIR=/var/tmp/zloop
37*eda14cbcSMatt Macy
38*eda14cbcSMatt Macyfunction usage
39*eda14cbcSMatt Macy{
40*eda14cbcSMatt Macy	echo -e "\n$0 [-t <timeout>] [ -s <vdev size> ] [-c <dump directory>]" \
41*eda14cbcSMatt Macy	    "[ -- [extra ztest parameters]]\n" \
42*eda14cbcSMatt Macy	    "\n" \
43*eda14cbcSMatt Macy	    "  This script runs ztest repeatedly with randomized arguments.\n" \
44*eda14cbcSMatt Macy	    "  If a crash is encountered, the ztest logs, any associated\n" \
45*eda14cbcSMatt Macy	    "  vdev files, and core file (if one exists) are moved to the\n" \
46*eda14cbcSMatt Macy	    "  output directory ($DEFAULTCOREDIR by default). Any options\n" \
47*eda14cbcSMatt Macy	    "  after the -- end-of-options marker will be passed to ztest.\n" \
48*eda14cbcSMatt Macy	    "\n" \
49*eda14cbcSMatt Macy	    "  Options:\n" \
50*eda14cbcSMatt Macy	    "    -t  Total time to loop for, in seconds. If not provided,\n" \
51*eda14cbcSMatt Macy	    "        zloop runs forever.\n" \
52*eda14cbcSMatt Macy	    "    -s  Size of vdev devices.\n" \
53*eda14cbcSMatt Macy	    "    -f  Specify working directory for ztest vdev files.\n" \
54*eda14cbcSMatt Macy	    "    -c  Specify a core dump directory to use.\n" \
55*eda14cbcSMatt Macy	    "    -m  Max number of core dumps to allow before exiting.\n" \
56*eda14cbcSMatt Macy	    "    -l  Create 'ztest.core.N' symlink to core directory.\n" \
57*eda14cbcSMatt Macy	    "    -h  Print this help message.\n" \
58*eda14cbcSMatt Macy	    "" >&2
59*eda14cbcSMatt Macy}
60*eda14cbcSMatt Macy
61*eda14cbcSMatt Macyfunction or_die
62*eda14cbcSMatt Macy{
63*eda14cbcSMatt Macy	# shellcheck disable=SC2068
64*eda14cbcSMatt Macy	$@
65*eda14cbcSMatt Macy	# shellcheck disable=SC2181
66*eda14cbcSMatt Macy	if [[ $? -ne 0 ]]; then
67*eda14cbcSMatt Macy		# shellcheck disable=SC2145
68*eda14cbcSMatt Macy		echo "Command failed: $@"
69*eda14cbcSMatt Macy		exit 1
70*eda14cbcSMatt Macy	fi
71*eda14cbcSMatt Macy}
72*eda14cbcSMatt Macy
73*eda14cbcSMatt Macycase $(uname) in
74*eda14cbcSMatt MacyFreeBSD)
75*eda14cbcSMatt Macy	coreglob="z*.core"
76*eda14cbcSMatt Macy	;;
77*eda14cbcSMatt MacyLinux)
78*eda14cbcSMatt Macy	# core file helpers
79*eda14cbcSMatt Macy	origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
80*eda14cbcSMatt Macy	coreglob="$(grep -E -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
81*eda14cbcSMatt Macy
82*eda14cbcSMatt Macy	if [[ $coreglob = "*" ]]; then
83*eda14cbcSMatt Macy		echo "Setting core file pattern..."
84*eda14cbcSMatt Macy		echo "core" > /proc/sys/kernel/core_pattern
85*eda14cbcSMatt Macy		coreglob="$(grep -E -o '^([^|%[:space:]]*)' \
86*eda14cbcSMatt Macy		    /proc/sys/kernel/core_pattern)*"
87*eda14cbcSMatt Macy	fi
88*eda14cbcSMatt Macy	;;
89*eda14cbcSMatt Macy*)
90*eda14cbcSMatt Macy	exit 1
91*eda14cbcSMatt Macy	;;
92*eda14cbcSMatt Macyesac
93*eda14cbcSMatt Macy
94*eda14cbcSMatt Macyfunction core_file
95*eda14cbcSMatt Macy{
96*eda14cbcSMatt Macy	# shellcheck disable=SC2012 disable=2086
97*eda14cbcSMatt Macy        printf "%s" "$(ls -tr1 $coreglob 2> /dev/null | head -1)"
98*eda14cbcSMatt Macy}
99*eda14cbcSMatt Macy
100*eda14cbcSMatt Macyfunction core_prog
101*eda14cbcSMatt Macy{
102*eda14cbcSMatt Macy	prog=$ZTEST
103*eda14cbcSMatt Macy	core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
104*eda14cbcSMatt Macy	    tr  \' ' ')
105*eda14cbcSMatt Macy	# shellcheck disable=SC2076
106*eda14cbcSMatt Macy	if [[ "$core_id" =~ "zdb "  ]]; then
107*eda14cbcSMatt Macy		prog=$ZDB
108*eda14cbcSMatt Macy	fi
109*eda14cbcSMatt Macy	printf "%s" "$prog"
110*eda14cbcSMatt Macy}
111*eda14cbcSMatt Macy
112*eda14cbcSMatt Macyfunction store_core
113*eda14cbcSMatt Macy{
114*eda14cbcSMatt Macy	core="$(core_file)"
115*eda14cbcSMatt Macy	if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
116*eda14cbcSMatt Macy		df -h "$workdir" >>ztest.out
117*eda14cbcSMatt Macy		coreid=$(date "+zloop-%y%m%d-%H%M%S")
118*eda14cbcSMatt Macy		foundcrashes=$((foundcrashes + 1))
119*eda14cbcSMatt Macy
120*eda14cbcSMatt Macy		# zdb debugging
121*eda14cbcSMatt Macy		zdbcmd="$ZDB -U "$workdir/zpool.cache" -dddMmDDG ztest"
122*eda14cbcSMatt Macy		zdbdebug=$($zdbcmd 2>&1)
123*eda14cbcSMatt Macy		echo -e "$zdbcmd\n" >>ztest.zdb
124*eda14cbcSMatt Macy		echo "$zdbdebug" >>ztest.zdb
125*eda14cbcSMatt Macy
126*eda14cbcSMatt Macy		dest=$coredir/$coreid
127*eda14cbcSMatt Macy		or_die mkdir -p "$dest"
128*eda14cbcSMatt Macy		or_die mkdir -p "$dest/vdev"
129*eda14cbcSMatt Macy
130*eda14cbcSMatt Macy		if [[ $symlink -ne 0 ]]; then
131*eda14cbcSMatt Macy			or_die ln -sf "$dest" ztest.core.$foundcrashes
132*eda14cbcSMatt Macy		fi
133*eda14cbcSMatt Macy
134*eda14cbcSMatt Macy		echo "*** ztest crash found - moving logs to $dest"
135*eda14cbcSMatt Macy
136*eda14cbcSMatt Macy		or_die mv ztest.history "$dest/"
137*eda14cbcSMatt Macy		or_die mv ztest.zdb "$dest/"
138*eda14cbcSMatt Macy		or_die mv ztest.out "$dest/"
139*eda14cbcSMatt Macy		or_die mv "$workdir/ztest*" "$dest/vdev/"
140*eda14cbcSMatt Macy
141*eda14cbcSMatt Macy		if [[ -e "$workdir/zpool.cache" ]]; then
142*eda14cbcSMatt Macy			or_die mv "$workdir/zpool.cache" "$dest/vdev/"
143*eda14cbcSMatt Macy		fi
144*eda14cbcSMatt Macy
145*eda14cbcSMatt Macy		# check for core
146*eda14cbcSMatt Macy		if [[ -f "$core" ]]; then
147*eda14cbcSMatt Macy			coreprog=$(core_prog "$core")
148*eda14cbcSMatt Macy			coredebug=$($GDB --batch --quiet \
149*eda14cbcSMatt Macy			    -ex "set print thread-events off" \
150*eda14cbcSMatt Macy			    -ex "printf \"*\n* Backtrace \n*\n\"" \
151*eda14cbcSMatt Macy			    -ex "bt" \
152*eda14cbcSMatt Macy			    -ex "printf \"*\n* Libraries \n*\n\"" \
153*eda14cbcSMatt Macy			    -ex "info sharedlib" \
154*eda14cbcSMatt Macy			    -ex "printf \"*\n* Threads (full) \n*\n\"" \
155*eda14cbcSMatt Macy			    -ex "info threads" \
156*eda14cbcSMatt Macy			    -ex "printf \"*\n* Backtraces \n*\n\"" \
157*eda14cbcSMatt Macy			    -ex "thread apply all bt" \
158*eda14cbcSMatt Macy			    -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
159*eda14cbcSMatt Macy			    -ex "thread apply all bt full" \
160*eda14cbcSMatt Macy			    -ex "quit" "$coreprog" "$core" 2>&1 | \
161*eda14cbcSMatt Macy			    grep -v "New LWP")
162*eda14cbcSMatt Macy
163*eda14cbcSMatt Macy			# Dump core + logs to stored directory
164*eda14cbcSMatt Macy			echo "$coredebug" >>"$dest/ztest.gdb"
165*eda14cbcSMatt Macy			or_die mv "$core" "$dest/"
166*eda14cbcSMatt Macy
167*eda14cbcSMatt Macy			# Record info in cores logfile
168*eda14cbcSMatt Macy			echo "*** core @ $coredir/$coreid/$core:" | \
169*eda14cbcSMatt Macy			    tee -a ztest.cores
170*eda14cbcSMatt Macy		fi
171*eda14cbcSMatt Macy
172*eda14cbcSMatt Macy		if [[ $coremax -gt 0 ]] &&
173*eda14cbcSMatt Macy		   [[ $foundcrashes -ge $coremax ]]; then
174*eda14cbcSMatt Macy			echo "exiting... max $coremax allowed cores"
175*eda14cbcSMatt Macy			exit 1
176*eda14cbcSMatt Macy		else
177*eda14cbcSMatt Macy			echo "continuing..."
178*eda14cbcSMatt Macy		fi
179*eda14cbcSMatt Macy	fi
180*eda14cbcSMatt Macy}
181*eda14cbcSMatt Macy
182*eda14cbcSMatt Macy# parse arguments
183*eda14cbcSMatt Macy# expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
184*eda14cbcSMatt Macycoredir=$DEFAULTCOREDIR
185*eda14cbcSMatt Macybasedir=$DEFAULTWORKDIR
186*eda14cbcSMatt Macyrundir="zloop-run"
187*eda14cbcSMatt Macytimeout=0
188*eda14cbcSMatt Macysize="512m"
189*eda14cbcSMatt Macycoremax=0
190*eda14cbcSMatt Macysymlink=0
191*eda14cbcSMatt Macywhile getopts ":ht:m:s:c:f:l" opt; do
192*eda14cbcSMatt Macy	case $opt in
193*eda14cbcSMatt Macy		t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
194*eda14cbcSMatt Macy		m ) [[ $OPTARG -gt 0 ]] && coremax=$OPTARG ;;
195*eda14cbcSMatt Macy		s ) [[ $OPTARG ]] && size=$OPTARG ;;
196*eda14cbcSMatt Macy		c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
197*eda14cbcSMatt Macy		f ) [[ $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
198*eda14cbcSMatt Macy		l ) symlink=1 ;;
199*eda14cbcSMatt Macy		h ) usage
200*eda14cbcSMatt Macy		    exit 2
201*eda14cbcSMatt Macy		    ;;
202*eda14cbcSMatt Macy		* ) echo "Invalid argument: -$OPTARG";
203*eda14cbcSMatt Macy		    usage
204*eda14cbcSMatt Macy		    exit 1
205*eda14cbcSMatt Macy	esac
206*eda14cbcSMatt Macydone
207*eda14cbcSMatt Macy# pass remaining arguments on to ztest
208*eda14cbcSMatt Macyshift $((OPTIND - 1))
209*eda14cbcSMatt Macy
210*eda14cbcSMatt Macy# enable core dumps
211*eda14cbcSMatt Macyulimit -c unlimited
212*eda14cbcSMatt Macyexport ASAN_OPTIONS=abort_on_error=1:disable_coredump=0
213*eda14cbcSMatt Macy
214*eda14cbcSMatt Macyif [[ -f "$(core_file)" ]]; then
215*eda14cbcSMatt Macy	echo -n "There's a core dump here you might want to look at first... "
216*eda14cbcSMatt Macy	core_file
217*eda14cbcSMatt Macy	echo
218*eda14cbcSMatt Macy	exit 1
219*eda14cbcSMatt Macyfi
220*eda14cbcSMatt Macy
221*eda14cbcSMatt Macyif [[ ! -d $coredir ]]; then
222*eda14cbcSMatt Macy	echo "core dump directory ($coredir) does not exist, creating it."
223*eda14cbcSMatt Macy	or_die mkdir -p "$coredir"
224*eda14cbcSMatt Macyfi
225*eda14cbcSMatt Macy
226*eda14cbcSMatt Macyif [[ ! -w $coredir ]]; then
227*eda14cbcSMatt Macy	echo "core dump directory ($coredir) is not writable."
228*eda14cbcSMatt Macy	exit 1
229*eda14cbcSMatt Macyfi
230*eda14cbcSMatt Macy
231*eda14cbcSMatt Macyor_die rm -f ztest.history
232*eda14cbcSMatt Macyor_die rm -f ztest.zdb
233*eda14cbcSMatt Macyor_die rm -f ztest.cores
234*eda14cbcSMatt Macy
235*eda14cbcSMatt Macyztrc=0		# ztest return value
236*eda14cbcSMatt Macyfoundcrashes=0	# number of crashes found so far
237*eda14cbcSMatt Macystarttime=$(date +%s)
238*eda14cbcSMatt Macycurtime=$starttime
239*eda14cbcSMatt Macy
240*eda14cbcSMatt Macy# if no timeout was specified, loop forever.
241*eda14cbcSMatt Macywhile [[ $timeout -eq 0 ]] || [[ $curtime -le $((starttime + timeout)) ]]; do
242*eda14cbcSMatt Macy	zopt="-G -VVVVV"
243*eda14cbcSMatt Macy
244*eda14cbcSMatt Macy	# start each run with an empty directory
245*eda14cbcSMatt Macy	workdir="$basedir/$rundir"
246*eda14cbcSMatt Macy	or_die rm -rf "$workdir"
247*eda14cbcSMatt Macy	or_die mkdir "$workdir"
248*eda14cbcSMatt Macy
249*eda14cbcSMatt Macy	# switch between common arrangements & fully randomized
250*eda14cbcSMatt Macy	if [[ $((RANDOM % 2)) -eq 0 ]]; then
251*eda14cbcSMatt Macy		mirrors=2
252*eda14cbcSMatt Macy		raidz=0
253*eda14cbcSMatt Macy		parity=1
254*eda14cbcSMatt Macy		vdevs=2
255*eda14cbcSMatt Macy	else
256*eda14cbcSMatt Macy		mirrors=$(((RANDOM % 3) * 1))
257*eda14cbcSMatt Macy		parity=$(((RANDOM % 3) + 1))
258*eda14cbcSMatt Macy		raidz=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
259*eda14cbcSMatt Macy		vdevs=$(((RANDOM % 3) + 3))
260*eda14cbcSMatt Macy	fi
261*eda14cbcSMatt Macy	align=$(((RANDOM % 2) * 3 + 9))
262*eda14cbcSMatt Macy	runtime=$((RANDOM % 100))
263*eda14cbcSMatt Macy	passtime=$((RANDOM % (runtime / 3 + 1) + 10))
264*eda14cbcSMatt Macy
265*eda14cbcSMatt Macy	zopt="$zopt -m $mirrors"
266*eda14cbcSMatt Macy	zopt="$zopt -r $raidz"
267*eda14cbcSMatt Macy	zopt="$zopt -R $parity"
268*eda14cbcSMatt Macy	zopt="$zopt -v $vdevs"
269*eda14cbcSMatt Macy	zopt="$zopt -a $align"
270*eda14cbcSMatt Macy	zopt="$zopt -T $runtime"
271*eda14cbcSMatt Macy	zopt="$zopt -P $passtime"
272*eda14cbcSMatt Macy	zopt="$zopt -s $size"
273*eda14cbcSMatt Macy	zopt="$zopt -f $workdir"
274*eda14cbcSMatt Macy
275*eda14cbcSMatt Macy	# shellcheck disable=SC2124
276*eda14cbcSMatt Macy	cmd="$ZTEST $zopt $@"
277*eda14cbcSMatt Macy	desc="$(date '+%m/%d %T') $cmd"
278*eda14cbcSMatt Macy	echo "$desc" | tee -a ztest.history
279*eda14cbcSMatt Macy	echo "$desc" >>ztest.out
280*eda14cbcSMatt Macy	$cmd >>ztest.out 2>&1
281*eda14cbcSMatt Macy	ztrc=$?
282*eda14cbcSMatt Macy	grep -E '===|WARNING' ztest.out >>ztest.history
283*eda14cbcSMatt Macy
284*eda14cbcSMatt Macy	store_core
285*eda14cbcSMatt Macy
286*eda14cbcSMatt Macy	curtime=$(date +%s)
287*eda14cbcSMatt Macydone
288*eda14cbcSMatt Macy
289*eda14cbcSMatt Macyecho "zloop finished, $foundcrashes crashes found"
290*eda14cbcSMatt Macy
291*eda14cbcSMatt Macy# restore core pattern.
292*eda14cbcSMatt Macycase $(uname) in
293*eda14cbcSMatt MacyLinux)
294*eda14cbcSMatt Macy	echo "$origcorepattern" > /proc/sys/kernel/core_pattern
295*eda14cbcSMatt Macy	;;
296*eda14cbcSMatt Macy*)
297*eda14cbcSMatt Macy	;;
298*eda14cbcSMatt Macyesac
299*eda14cbcSMatt Macy
300*eda14cbcSMatt Macyuptime >>ztest.out
301*eda14cbcSMatt Macy
302*eda14cbcSMatt Macyif [[ $foundcrashes -gt 0 ]]; then
303*eda14cbcSMatt Macy	exit 1
304*eda14cbcSMatt Macyfi
305