xref: /freebsd/sys/contrib/openzfs/scripts/zloop.sh (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1eda14cbcSMatt Macy#!/usr/bin/env bash
2*61145dc2SMartin Matuska# SPDX-License-Identifier: CDDL-1.0
3eda14cbcSMatt Macy
4eda14cbcSMatt Macy#
5eda14cbcSMatt Macy# CDDL HEADER START
6eda14cbcSMatt Macy#
7eda14cbcSMatt Macy# This file and its contents are supplied under the terms of the
8eda14cbcSMatt Macy# Common Development and Distribution License ("CDDL"), version 1.0.
9eda14cbcSMatt Macy# You may only use this file in accordance with the terms of version
10eda14cbcSMatt Macy# 1.0 of the CDDL.
11eda14cbcSMatt Macy#
12eda14cbcSMatt Macy# A full copy of the text of the CDDL should have accompanied this
13eda14cbcSMatt Macy# source.  A copy of the CDDL is also available via the Internet at
14eda14cbcSMatt Macy# http://www.illumos.org/license/CDDL.
15eda14cbcSMatt Macy#
16eda14cbcSMatt Macy# CDDL HEADER END
17eda14cbcSMatt Macy#
18eda14cbcSMatt Macy
19eda14cbcSMatt Macy#
20eda14cbcSMatt Macy# Copyright (c) 2015 by Delphix. All rights reserved.
21eda14cbcSMatt Macy# Copyright (C) 2016 Lawrence Livermore National Security, LLC.
227877fdebSMatt Macy# Copyright (c) 2017, Intel Corporation.
23eda14cbcSMatt Macy#
24eda14cbcSMatt Macy
25716fd348SMartin MatuskaBASE_DIR=${0%/*}
26eda14cbcSMatt MacySCRIPT_COMMON=common.sh
27bb2d13b6SMartin Matuskaif [[ -f "${BASE_DIR}/${SCRIPT_COMMON}" ]]; then
28eda14cbcSMatt Macy	. "${BASE_DIR}/${SCRIPT_COMMON}"
29eda14cbcSMatt Macyelse
30eda14cbcSMatt Macy	echo "Missing helper script ${SCRIPT_COMMON}" && exit 1
31eda14cbcSMatt Macyfi
32eda14cbcSMatt Macy
33eda14cbcSMatt Macy# shellcheck disable=SC2034
34eda14cbcSMatt MacyPROG=zloop.sh
35eda14cbcSMatt MacyGDB=${GDB:-gdb}
36eda14cbcSMatt Macy
37eda14cbcSMatt MacyDEFAULTWORKDIR=/var/tmp
38eda14cbcSMatt MacyDEFAULTCOREDIR=/var/tmp/zloop
39eda14cbcSMatt Macy
40eda14cbcSMatt Macyfunction usage
41eda14cbcSMatt Macy{
423f9d360cSMartin Matuska	cat >&2 <<EOF
433f9d360cSMartin Matuska
443f9d360cSMartin Matuska$0 [-hl] [-c <dump directory>] [-f <vdev directory>]
453f9d360cSMartin Matuska  [-m <max core dumps>] [-s <vdev size>] [-t <timeout>]
463f9d360cSMartin Matuska  [-I <max iterations>] [-- [extra ztest parameters]]
473f9d360cSMartin Matuska
483f9d360cSMartin Matuska  This script runs ztest repeatedly with randomized arguments.
493f9d360cSMartin Matuska  If a crash is encountered, the ztest logs, any associated
503f9d360cSMartin Matuska  vdev files, and core file (if one exists) are moved to the
513f9d360cSMartin Matuska  output directory ($DEFAULTCOREDIR by default). Any options
523f9d360cSMartin Matuska  after the -- end-of-options marker will be passed to ztest.
533f9d360cSMartin Matuska
543f9d360cSMartin Matuska  Options:
553f9d360cSMartin Matuska    -c  Specify a core dump directory to use.
563f9d360cSMartin Matuska    -f  Specify working directory for ztest vdev files.
573f9d360cSMartin Matuska    -h  Print this help message.
583f9d360cSMartin Matuska    -l  Create 'ztest.core.N' symlink to core directory.
593f9d360cSMartin Matuska    -m  Max number of core dumps to allow before exiting.
603f9d360cSMartin Matuska    -s  Size of vdev devices.
613f9d360cSMartin Matuska    -t  Total time to loop for, in seconds. If not provided,
623f9d360cSMartin Matuska        zloop runs forever.
633f9d360cSMartin Matuska    -I  Max number of iterations to loop before exiting.
643f9d360cSMartin Matuska
653f9d360cSMartin MatuskaEOF
66eda14cbcSMatt Macy}
67eda14cbcSMatt Macy
68eda14cbcSMatt Macyfunction or_die
69eda14cbcSMatt Macy{
70716fd348SMartin Matuska	if ! "$@"; then
7116038816SMartin Matuska		echo "Command failed: $*"
72eda14cbcSMatt Macy		exit 1
73eda14cbcSMatt Macy	fi
74eda14cbcSMatt Macy}
75eda14cbcSMatt Macy
76eda14cbcSMatt Macycase $(uname) in
77eda14cbcSMatt MacyFreeBSD)
78eda14cbcSMatt Macy	coreglob="z*.core"
79eda14cbcSMatt Macy	;;
80eda14cbcSMatt MacyLinux)
81eda14cbcSMatt Macy	# core file helpers
82716fd348SMartin Matuska	read -r origcorepattern </proc/sys/kernel/core_pattern
83eda14cbcSMatt Macy	coreglob="$(grep -E -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
84eda14cbcSMatt Macy
85eda14cbcSMatt Macy	if [[ $coreglob = "*" ]]; then
86eda14cbcSMatt Macy		echo "Setting core file pattern..."
87eda14cbcSMatt Macy		echo "core" > /proc/sys/kernel/core_pattern
88eda14cbcSMatt Macy		coreglob="$(grep -E -o '^([^|%[:space:]]*)' \
89eda14cbcSMatt Macy		    /proc/sys/kernel/core_pattern)*"
90eda14cbcSMatt Macy	fi
91eda14cbcSMatt Macy	;;
92eda14cbcSMatt Macy*)
93eda14cbcSMatt Macy	exit 1
94eda14cbcSMatt Macy	;;
95eda14cbcSMatt Macyesac
96eda14cbcSMatt Macy
97eda14cbcSMatt Macyfunction core_file
98eda14cbcSMatt Macy{
9916038816SMartin Matuska	# shellcheck disable=SC2012,SC2086
10016038816SMartin Matuska	ls -tr1 $coreglob 2>/dev/null | head -1
101eda14cbcSMatt Macy}
102eda14cbcSMatt Macy
103eda14cbcSMatt Macyfunction core_prog
104eda14cbcSMatt Macy{
105e92ffd9bSMartin Matuska	# shellcheck disable=SC2154
106eda14cbcSMatt Macy	prog=$ZTEST
107eda14cbcSMatt Macy	core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
108eda14cbcSMatt Macy	    tr  \' ' ')
10916038816SMartin Matuska	if [[ "$core_id" == *"zdb "* ]]; then
110e92ffd9bSMartin Matuska		# shellcheck disable=SC2154
111eda14cbcSMatt Macy		prog=$ZDB
112eda14cbcSMatt Macy	fi
113eda14cbcSMatt Macy	printf "%s" "$prog"
114eda14cbcSMatt Macy}
115eda14cbcSMatt Macy
116eda14cbcSMatt Macyfunction store_core
117eda14cbcSMatt Macy{
118eda14cbcSMatt Macy	core="$(core_file)"
119eda14cbcSMatt Macy	if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
120eda14cbcSMatt Macy		df -h "$workdir" >>ztest.out
121eda14cbcSMatt Macy		coreid=$(date "+zloop-%y%m%d-%H%M%S")
122eda14cbcSMatt Macy		foundcrashes=$((foundcrashes + 1))
123eda14cbcSMatt Macy
124eda14cbcSMatt Macy		# zdb debugging
125eda14cbcSMatt Macy		zdbcmd="$ZDB -U "$workdir/zpool.cache" -dddMmDDG ztest"
126eda14cbcSMatt Macy		zdbdebug=$($zdbcmd 2>&1)
127eda14cbcSMatt Macy		echo -e "$zdbcmd\n" >>ztest.zdb
128eda14cbcSMatt Macy		echo "$zdbdebug" >>ztest.zdb
129eda14cbcSMatt Macy
130eda14cbcSMatt Macy		dest=$coredir/$coreid
131eda14cbcSMatt Macy		or_die mkdir -p "$dest/vdev"
132eda14cbcSMatt Macy
133eda14cbcSMatt Macy		if [[ $symlink -ne 0 ]]; then
134e92ffd9bSMartin Matuska			or_die ln -sf "$dest" "ztest.core.${foundcrashes}"
135eda14cbcSMatt Macy		fi
136eda14cbcSMatt Macy
137eda14cbcSMatt Macy		echo "*** ztest crash found - moving logs to $dest"
138eda14cbcSMatt Macy
139716fd348SMartin Matuska		or_die mv ztest.history ztest.zdb ztest.out "$dest/"
140716fd348SMartin Matuska		or_die mv "$workdir/"ztest* "$dest/vdev/"
141eda14cbcSMatt Macy
142eda14cbcSMatt Macy		if [[ -e "$workdir/zpool.cache" ]]; then
143eda14cbcSMatt Macy			or_die mv "$workdir/zpool.cache" "$dest/vdev/"
144eda14cbcSMatt Macy		fi
145eda14cbcSMatt Macy
146eda14cbcSMatt Macy		# check for core
147eda14cbcSMatt Macy		if [[ -f "$core" ]]; then
148eda14cbcSMatt Macy			coreprog=$(core_prog "$core")
149eda14cbcSMatt Macy			coredebug=$($GDB --batch --quiet \
150eda14cbcSMatt Macy			    -ex "set print thread-events off" \
151eda14cbcSMatt Macy			    -ex "printf \"*\n* Backtrace \n*\n\"" \
152eda14cbcSMatt Macy			    -ex "bt" \
153eda14cbcSMatt Macy			    -ex "printf \"*\n* Libraries \n*\n\"" \
154eda14cbcSMatt Macy			    -ex "info sharedlib" \
155eda14cbcSMatt Macy			    -ex "printf \"*\n* Threads (full) \n*\n\"" \
156eda14cbcSMatt Macy			    -ex "info threads" \
157eda14cbcSMatt Macy			    -ex "printf \"*\n* Backtraces \n*\n\"" \
158eda14cbcSMatt Macy			    -ex "thread apply all bt" \
159eda14cbcSMatt Macy			    -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
160eda14cbcSMatt Macy			    -ex "thread apply all bt full" \
161eda14cbcSMatt Macy			    -ex "quit" "$coreprog" "$core" 2>&1 | \
162eda14cbcSMatt Macy			    grep -v "New LWP")
163eda14cbcSMatt Macy
164eda14cbcSMatt Macy			# Dump core + logs to stored directory
165eda14cbcSMatt Macy			echo "$coredebug" >>"$dest/ztest.gdb"
166eda14cbcSMatt Macy			or_die mv "$core" "$dest/"
167eda14cbcSMatt Macy
168eda14cbcSMatt Macy			# Record info in cores logfile
169eda14cbcSMatt Macy			echo "*** core @ $coredir/$coreid/$core:" | \
170eda14cbcSMatt Macy			    tee -a ztest.cores
171eda14cbcSMatt Macy		fi
172eda14cbcSMatt Macy
173eda14cbcSMatt Macy		if [[ $coremax -gt 0 ]] &&
174eda14cbcSMatt Macy		   [[ $foundcrashes -ge $coremax ]]; then
175eda14cbcSMatt Macy			echo "exiting... max $coremax allowed cores"
176eda14cbcSMatt Macy			exit 1
177eda14cbcSMatt Macy		else
178eda14cbcSMatt Macy			echo "continuing..."
179eda14cbcSMatt Macy		fi
180eda14cbcSMatt Macy	fi
181eda14cbcSMatt Macy}
182eda14cbcSMatt Macy
183eda14cbcSMatt Macy# parse arguments
184eda14cbcSMatt Macy# expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
185eda14cbcSMatt Macycoredir=$DEFAULTCOREDIR
186eda14cbcSMatt Macybasedir=$DEFAULTWORKDIR
187eda14cbcSMatt Macyrundir="zloop-run"
188eda14cbcSMatt Macytimeout=0
189eda14cbcSMatt Macysize="512m"
190eda14cbcSMatt Macycoremax=0
191eda14cbcSMatt Macysymlink=0
1923f9d360cSMartin Matuskaiterations=0
1933f9d360cSMartin Matuskawhile getopts ":ht:m:I:s:c:f:l" opt; do
194eda14cbcSMatt Macy	case $opt in
195eda14cbcSMatt Macy		t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
196eda14cbcSMatt Macy		m ) [[ $OPTARG -gt 0 ]] && coremax=$OPTARG ;;
197e92ffd9bSMartin Matuska		I ) [[ -n $OPTARG ]] && iterations=$OPTARG ;;
198e92ffd9bSMartin Matuska		s ) [[ -n $OPTARG ]] && size=$OPTARG ;;
199e92ffd9bSMartin Matuska		c ) [[ -n $OPTARG ]] && coredir=$OPTARG ;;
200e92ffd9bSMartin Matuska		f ) [[ -n $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
201eda14cbcSMatt Macy		l ) symlink=1 ;;
202eda14cbcSMatt Macy		h ) usage
203eda14cbcSMatt Macy		    exit 2
204eda14cbcSMatt Macy		    ;;
205eda14cbcSMatt Macy		* ) echo "Invalid argument: -$OPTARG";
206eda14cbcSMatt Macy		    usage
207eda14cbcSMatt Macy		    exit 1
208eda14cbcSMatt Macy	esac
209eda14cbcSMatt Macydone
210eda14cbcSMatt Macy# pass remaining arguments on to ztest
211eda14cbcSMatt Macyshift $((OPTIND - 1))
212eda14cbcSMatt Macy
213eda14cbcSMatt Macy# enable core dumps
214eda14cbcSMatt Macyulimit -c unlimited
215c03c5b1cSMartin Matuskaexport ASAN_OPTIONS=abort_on_error=true:halt_on_error=true:allocator_may_return_null=true:disable_coredump=false:detect_stack_use_after_return=true
216c03c5b1cSMartin Matuskaexport UBSAN_OPTIONS=abort_on_error=true:halt_on_error=true:print_stacktrace=true
217eda14cbcSMatt Macy
218eda14cbcSMatt Macyif [[ -f "$(core_file)" ]]; then
219eda14cbcSMatt Macy	echo -n "There's a core dump here you might want to look at first... "
220eda14cbcSMatt Macy	core_file
221eda14cbcSMatt Macy	echo
222eda14cbcSMatt Macy	exit 1
223eda14cbcSMatt Macyfi
224eda14cbcSMatt Macy
225eda14cbcSMatt Macyif [[ ! -d $coredir ]]; then
226eda14cbcSMatt Macy	echo "core dump directory ($coredir) does not exist, creating it."
227eda14cbcSMatt Macy	or_die mkdir -p "$coredir"
228eda14cbcSMatt Macyfi
229eda14cbcSMatt Macy
230eda14cbcSMatt Macyif [[ ! -w $coredir ]]; then
231eda14cbcSMatt Macy	echo "core dump directory ($coredir) is not writable."
232eda14cbcSMatt Macy	exit 1
233eda14cbcSMatt Macyfi
234eda14cbcSMatt Macy
235716fd348SMartin Matuskaor_die rm -f ztest.history ztest.zdb ztest.cores
236eda14cbcSMatt Macy
237eda14cbcSMatt Macyztrc=0		# ztest return value
238eda14cbcSMatt Macyfoundcrashes=0	# number of crashes found so far
239eda14cbcSMatt Macystarttime=$(date +%s)
240eda14cbcSMatt Macycurtime=$starttime
2413f9d360cSMartin Matuskaiteration=0
242eda14cbcSMatt Macy
243eda14cbcSMatt Macy# if no timeout was specified, loop forever.
2443f9d360cSMartin Matuskawhile (( timeout == 0 )) || (( curtime <= (starttime + timeout) )); do
2453f9d360cSMartin Matuska	if (( iterations > 0 )) && (( iteration++ == iterations )); then
2463f9d360cSMartin Matuska		break
2473f9d360cSMartin Matuska	fi
2483f9d360cSMartin Matuska
249eda14cbcSMatt Macy	zopt="-G -VVVVV"
250eda14cbcSMatt Macy
251eda14cbcSMatt Macy	# start each run with an empty directory
252eda14cbcSMatt Macy	workdir="$basedir/$rundir"
253eda14cbcSMatt Macy	or_die rm -rf "$workdir"
254eda14cbcSMatt Macy	or_die mkdir "$workdir"
255eda14cbcSMatt Macy
2567877fdebSMatt Macy	# ashift range 9 - 15
257eda14cbcSMatt Macy	align=$(((RANDOM % 2) * 3 + 9))
2587877fdebSMatt Macy
259e716630dSMartin Matuska	# choose parity value
260e716630dSMartin Matuska	parity=$(((RANDOM % 3) + 1))
261e716630dSMartin Matuska
262e716630dSMartin Matuska	draid_data=0
263e716630dSMartin Matuska	draid_spares=0
264e716630dSMartin Matuska
2657877fdebSMatt Macy	# randomly use special classes
2667877fdebSMatt Macy	class="special=random"
2677877fdebSMatt Macy
268e716630dSMartin Matuska	# choose between four types of configs
269e716630dSMartin Matuska	# (basic, raidz mix, raidz expansion, and draid mix)
270e716630dSMartin Matuska	case $((RANDOM % 4)) in
271e716630dSMartin Matuska
272e716630dSMartin Matuska	# basic mirror configuration
273e716630dSMartin Matuska	0)	parity=1
2747877fdebSMatt Macy		mirrors=2
2757877fdebSMatt Macy		raid_children=0
2767877fdebSMatt Macy		vdevs=2
2777877fdebSMatt Macy		raid_type="raidz"
278e716630dSMartin Matuska		;;
279e716630dSMartin Matuska
2807877fdebSMatt Macy	# fully randomized mirror/raidz (sans dRAID)
281e716630dSMartin Matuska	1)	mirrors=$(((RANDOM % 3) * 1))
2827877fdebSMatt Macy		raid_children=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
2837877fdebSMatt Macy		vdevs=$(((RANDOM % 3) + 3))
2847877fdebSMatt Macy		raid_type="raidz"
285e716630dSMartin Matuska		;;
286e716630dSMartin Matuska
287e716630dSMartin Matuska	# randomized raidz expansion (one top-level raidz vdev)
288e716630dSMartin Matuska	2)	mirrors=0
289e716630dSMartin Matuska		vdevs=1
290e716630dSMartin Matuska		# derive initial raidz disk count based on parity choice
291e716630dSMartin Matuska		#   P1: 3 - 7 disks
292e716630dSMartin Matuska		#   P2: 5 - 9 disks
293e716630dSMartin Matuska		#   P3: 7 - 11 disks
294e716630dSMartin Matuska		raid_children=$(((RANDOM % 5) + (parity * 2) + 1))
295e716630dSMartin Matuska
296e716630dSMartin Matuska		# 1/3 of the time use a dedicated '-X' raidz expansion test
297e716630dSMartin Matuska		if [[ $((RANDOM % 3)) -eq 0 ]]; then
298e716630dSMartin Matuska			zopt="$zopt -X -t 16"
299e716630dSMartin Matuska			raid_type="raidz"
3007877fdebSMatt Macy		else
301e716630dSMartin Matuska			raid_type="eraidz"
302e716630dSMartin Matuska		fi
303e716630dSMartin Matuska		;;
304e716630dSMartin Matuska
3057877fdebSMatt Macy	# fully randomized dRAID (sans mirror/raidz)
306e716630dSMartin Matuska	3)	mirrors=0
3077877fdebSMatt Macy		draid_data=$(((RANDOM % 8) + 3))
3087877fdebSMatt Macy		draid_spares=$(((RANDOM % 2) + parity))
3097877fdebSMatt Macy		stripe=$((draid_data + parity))
3107877fdebSMatt Macy		extra=$((draid_spares + (RANDOM % 4)))
3117877fdebSMatt Macy		raid_children=$(((((RANDOM % 4) + 1) * stripe) + extra))
3127877fdebSMatt Macy		vdevs=$((RANDOM % 3))
3137877fdebSMatt Macy		raid_type="draid"
314e716630dSMartin Matuska		;;
315e716630dSMartin Matuska	*)
316e716630dSMartin Matuska		# avoid shellcheck SC2249
317e716630dSMartin Matuska		;;
318e716630dSMartin Matuska	esac
3197877fdebSMatt Macy
3207877fdebSMatt Macy	zopt="$zopt -K $raid_type"
321eda14cbcSMatt Macy	zopt="$zopt -m $mirrors"
3227877fdebSMatt Macy	zopt="$zopt -r $raid_children"
3237877fdebSMatt Macy	zopt="$zopt -D $draid_data"
3247877fdebSMatt Macy	zopt="$zopt -S $draid_spares"
325eda14cbcSMatt Macy	zopt="$zopt -R $parity"
326eda14cbcSMatt Macy	zopt="$zopt -v $vdevs"
327eda14cbcSMatt Macy	zopt="$zopt -a $align"
3287877fdebSMatt Macy	zopt="$zopt -C $class"
329eda14cbcSMatt Macy	zopt="$zopt -s $size"
330eda14cbcSMatt Macy	zopt="$zopt -f $workdir"
331eda14cbcSMatt Macy
33216038816SMartin Matuska	cmd="$ZTEST $zopt $*"
333716fd348SMartin Matuska	echo "$(date '+%m/%d %T') $cmd" | tee -a ztest.history ztest.out
334eda14cbcSMatt Macy	$cmd >>ztest.out 2>&1
335eda14cbcSMatt Macy	ztrc=$?
336eda14cbcSMatt Macy	grep -E '===|WARNING' ztest.out >>ztest.history
337eda14cbcSMatt Macy
338eda14cbcSMatt Macy	store_core
339eda14cbcSMatt Macy
340eda14cbcSMatt Macy	curtime=$(date +%s)
341eda14cbcSMatt Macydone
342eda14cbcSMatt Macy
343eda14cbcSMatt Macyecho "zloop finished, $foundcrashes crashes found"
344eda14cbcSMatt Macy
345eda14cbcSMatt Macy# restore core pattern.
346eda14cbcSMatt Macycase $(uname) in
347eda14cbcSMatt MacyLinux)
348eda14cbcSMatt Macy	echo "$origcorepattern" > /proc/sys/kernel/core_pattern
349eda14cbcSMatt Macy	;;
350eda14cbcSMatt Macy*)
351eda14cbcSMatt Macy	;;
352eda14cbcSMatt Macyesac
353eda14cbcSMatt Macy
354eda14cbcSMatt Macyuptime >>ztest.out
355eda14cbcSMatt Macy
356eda14cbcSMatt Macyif [[ $foundcrashes -gt 0 ]]; then
357eda14cbcSMatt Macy	exit 1
358eda14cbcSMatt Macyfi
359