xref: /freebsd/sys/contrib/openzfs/scripts/zloop.sh (revision dc318a4ffabcbfa23bb56a33403aad36e6de30af) 
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
24BASE_DIR=$(dirname "$0")
25SCRIPT_COMMON=common.sh
26if [ -f "${BASE_DIR}/${SCRIPT_COMMON}" ]; then
27	. "${BASE_DIR}/${SCRIPT_COMMON}"
28else
29	echo "Missing helper script ${SCRIPT_COMMON}" && exit 1
30fi
31
32# shellcheck disable=SC2034
33PROG=zloop.sh
34GDB=${GDB:-gdb}
35
36DEFAULTWORKDIR=/var/tmp
37DEFAULTCOREDIR=/var/tmp/zloop
38
39function usage
40{
41	echo -e "\n$0 [-t <timeout>] [ -s <vdev size> ] [-c <dump directory>]" \
42	    "[ -- [extra ztest parameters]]\n" \
43	    "\n" \
44	    "  This script runs ztest repeatedly with randomized arguments.\n" \
45	    "  If a crash is encountered, the ztest logs, any associated\n" \
46	    "  vdev files, and core file (if one exists) are moved to the\n" \
47	    "  output directory ($DEFAULTCOREDIR by default). Any options\n" \
48	    "  after the -- end-of-options marker will be passed to ztest.\n" \
49	    "\n" \
50	    "  Options:\n" \
51	    "    -t  Total time to loop for, in seconds. If not provided,\n" \
52	    "        zloop runs forever.\n" \
53	    "    -s  Size of vdev devices.\n" \
54	    "    -f  Specify working directory for ztest vdev files.\n" \
55	    "    -c  Specify a core dump directory to use.\n" \
56	    "    -m  Max number of core dumps to allow before exiting.\n" \
57	    "    -l  Create 'ztest.core.N' symlink to core directory.\n" \
58	    "    -h  Print this help message.\n" \
59	    "" >&2
60}
61
62function or_die
63{
64	# shellcheck disable=SC2068
65	if ! $@; then
66		echo "Command failed: $*"
67		exit 1
68	fi
69}
70
71case $(uname) in
72FreeBSD)
73	coreglob="z*.core"
74	;;
75Linux)
76	# core file helpers
77	origcorepattern="$(cat /proc/sys/kernel/core_pattern)"
78	coreglob="$(grep -E -o '^([^|%[:space:]]*)' /proc/sys/kernel/core_pattern)*"
79
80	if [[ $coreglob = "*" ]]; then
81		echo "Setting core file pattern..."
82		echo "core" > /proc/sys/kernel/core_pattern
83		coreglob="$(grep -E -o '^([^|%[:space:]]*)' \
84		    /proc/sys/kernel/core_pattern)*"
85	fi
86	;;
87*)
88	exit 1
89	;;
90esac
91
92function core_file
93{
94	# shellcheck disable=SC2012,SC2086
95	ls -tr1 $coreglob 2>/dev/null | head -1
96}
97
98function core_prog
99{
100	prog=$ZTEST
101	core_id=$($GDB --batch -c "$1" | grep "Core was generated by" | \
102	    tr  \' ' ')
103	if [[ "$core_id" == *"zdb "* ]]; then
104		prog=$ZDB
105	fi
106	printf "%s" "$prog"
107}
108
109function store_core
110{
111	core="$(core_file)"
112	if [[ $ztrc -ne 0 ]] || [[ -f "$core" ]]; then
113		df -h "$workdir" >>ztest.out
114		coreid=$(date "+zloop-%y%m%d-%H%M%S")
115		foundcrashes=$((foundcrashes + 1))
116
117		# zdb debugging
118		zdbcmd="$ZDB -U "$workdir/zpool.cache" -dddMmDDG ztest"
119		zdbdebug=$($zdbcmd 2>&1)
120		echo -e "$zdbcmd\n" >>ztest.zdb
121		echo "$zdbdebug" >>ztest.zdb
122
123		dest=$coredir/$coreid
124		or_die mkdir -p "$dest"
125		or_die mkdir -p "$dest/vdev"
126
127		if [[ $symlink -ne 0 ]]; then
128			or_die ln -sf "$dest" ztest.core.$foundcrashes
129		fi
130
131		echo "*** ztest crash found - moving logs to $dest"
132
133		or_die mv ztest.history "$dest/"
134		or_die mv ztest.zdb "$dest/"
135		or_die mv ztest.out "$dest/"
136		or_die mv "$workdir/ztest*" "$dest/vdev/"
137
138		if [[ -e "$workdir/zpool.cache" ]]; then
139			or_die mv "$workdir/zpool.cache" "$dest/vdev/"
140		fi
141
142		# check for core
143		if [[ -f "$core" ]]; then
144			coreprog=$(core_prog "$core")
145			coredebug=$($GDB --batch --quiet \
146			    -ex "set print thread-events off" \
147			    -ex "printf \"*\n* Backtrace \n*\n\"" \
148			    -ex "bt" \
149			    -ex "printf \"*\n* Libraries \n*\n\"" \
150			    -ex "info sharedlib" \
151			    -ex "printf \"*\n* Threads (full) \n*\n\"" \
152			    -ex "info threads" \
153			    -ex "printf \"*\n* Backtraces \n*\n\"" \
154			    -ex "thread apply all bt" \
155			    -ex "printf \"*\n* Backtraces (full) \n*\n\"" \
156			    -ex "thread apply all bt full" \
157			    -ex "quit" "$coreprog" "$core" 2>&1 | \
158			    grep -v "New LWP")
159
160			# Dump core + logs to stored directory
161			echo "$coredebug" >>"$dest/ztest.gdb"
162			or_die mv "$core" "$dest/"
163
164			# Record info in cores logfile
165			echo "*** core @ $coredir/$coreid/$core:" | \
166			    tee -a ztest.cores
167		fi
168
169		if [[ $coremax -gt 0 ]] &&
170		   [[ $foundcrashes -ge $coremax ]]; then
171			echo "exiting... max $coremax allowed cores"
172			exit 1
173		else
174			echo "continuing..."
175		fi
176	fi
177}
178
179# parse arguments
180# expected format: zloop [-t timeout] [-c coredir] [-- extra ztest args]
181coredir=$DEFAULTCOREDIR
182basedir=$DEFAULTWORKDIR
183rundir="zloop-run"
184timeout=0
185size="512m"
186coremax=0
187symlink=0
188while getopts ":ht:m:s:c:f:l" opt; do
189	case $opt in
190		t ) [[ $OPTARG -gt 0 ]] && timeout=$OPTARG ;;
191		m ) [[ $OPTARG -gt 0 ]] && coremax=$OPTARG ;;
192		s ) [[ $OPTARG ]] && size=$OPTARG ;;
193		c ) [[ $OPTARG ]] && coredir=$OPTARG ;;
194		f ) [[ $OPTARG ]] && basedir=$(readlink -f "$OPTARG") ;;
195		l ) symlink=1 ;;
196		h ) usage
197		    exit 2
198		    ;;
199		* ) echo "Invalid argument: -$OPTARG";
200		    usage
201		    exit 1
202	esac
203done
204# pass remaining arguments on to ztest
205shift $((OPTIND - 1))
206
207# enable core dumps
208ulimit -c unlimited
209export ASAN_OPTIONS=abort_on_error=1:disable_coredump=0
210
211if [[ -f "$(core_file)" ]]; then
212	echo -n "There's a core dump here you might want to look at first... "
213	core_file
214	echo
215	exit 1
216fi
217
218if [[ ! -d $coredir ]]; then
219	echo "core dump directory ($coredir) does not exist, creating it."
220	or_die mkdir -p "$coredir"
221fi
222
223if [[ ! -w $coredir ]]; then
224	echo "core dump directory ($coredir) is not writable."
225	exit 1
226fi
227
228or_die rm -f ztest.history
229or_die rm -f ztest.zdb
230or_die rm -f ztest.cores
231
232ztrc=0		# ztest return value
233foundcrashes=0	# number of crashes found so far
234starttime=$(date +%s)
235curtime=$starttime
236
237# if no timeout was specified, loop forever.
238while [[ $timeout -eq 0 ]] || [[ $curtime -le $((starttime + timeout)) ]]; do
239	zopt="-G -VVVVV"
240
241	# start each run with an empty directory
242	workdir="$basedir/$rundir"
243	or_die rm -rf "$workdir"
244	or_die mkdir "$workdir"
245
246	# switch between three types of configs
247	# 1/3 basic, 1/3 raidz mix, and 1/3 draid mix
248	choice=$((RANDOM % 3))
249
250	# ashift range 9 - 15
251	align=$(((RANDOM % 2) * 3 + 9))
252
253	# randomly use special classes
254	class="special=random"
255
256	if [[ $choice -eq 0 ]]; then
257		# basic mirror only
258		parity=1
259		mirrors=2
260		draid_data=0
261		draid_spares=0
262		raid_children=0
263		vdevs=2
264		raid_type="raidz"
265	elif [[ $choice -eq 1 ]]; then
266		# fully randomized mirror/raidz (sans dRAID)
267		parity=$(((RANDOM % 3) + 1))
268		mirrors=$(((RANDOM % 3) * 1))
269		draid_data=0
270		draid_spares=0
271		raid_children=$((((RANDOM % 9) + parity + 1) * (RANDOM % 2)))
272		vdevs=$(((RANDOM % 3) + 3))
273		raid_type="raidz"
274	else
275		# fully randomized dRAID (sans mirror/raidz)
276		parity=$(((RANDOM % 3) + 1))
277		mirrors=0
278		draid_data=$(((RANDOM % 8) + 3))
279		draid_spares=$(((RANDOM % 2) + parity))
280		stripe=$((draid_data + parity))
281		extra=$((draid_spares + (RANDOM % 4)))
282		raid_children=$(((((RANDOM % 4) + 1) * stripe) + extra))
283		vdevs=$((RANDOM % 3))
284		raid_type="draid"
285	fi
286
287	# run from 30 to 120 seconds
288	runtime=$(((RANDOM % 90) + 30))
289	passtime=$((RANDOM % (runtime / 3 + 1) + 10))
290
291	zopt="$zopt -K $raid_type"
292	zopt="$zopt -m $mirrors"
293	zopt="$zopt -r $raid_children"
294	zopt="$zopt -D $draid_data"
295	zopt="$zopt -S $draid_spares"
296	zopt="$zopt -R $parity"
297	zopt="$zopt -v $vdevs"
298	zopt="$zopt -a $align"
299	zopt="$zopt -C $class"
300	zopt="$zopt -T $runtime"
301	zopt="$zopt -P $passtime"
302	zopt="$zopt -s $size"
303	zopt="$zopt -f $workdir"
304
305	cmd="$ZTEST $zopt $*"
306	desc="$(date '+%m/%d %T') $cmd"
307	echo "$desc" | tee -a ztest.history
308	echo "$desc" >>ztest.out
309	$cmd >>ztest.out 2>&1
310	ztrc=$?
311	grep -E '===|WARNING' ztest.out >>ztest.history
312
313	store_core
314
315	curtime=$(date +%s)
316done
317
318echo "zloop finished, $foundcrashes crashes found"
319
320# restore core pattern.
321case $(uname) in
322Linux)
323	echo "$origcorepattern" > /proc/sys/kernel/core_pattern
324	;;
325*)
326	;;
327esac
328
329uptime >>ztest.out
330
331if [[ $foundcrashes -gt 0 ]]; then
332	exit 1
333fi
334