305

   1 #! /bin/bash
   2 # FSQA Test No. 305
   3 #
   4 # Run fsstress and fio(dio/aio and mmap) and simulate disk failure
   5 # check filesystem consistency at the end.
   6 #
   7 #-----------------------------------------------------------------------
   8 # (c) 2013 Dmitry Monakhov
   9 #
  10 # This program is free software; you can redistribute it and/or
  11 # modify it under the terms of the GNU General Public License as
  12 # published by the Free Software Foundation.
  13 #
  14 # This program is distributed in the hope that it would be useful,
  15 # but WITHOUT ANY WARRANTY; without even the implied warranty of
  16 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17 # GNU General Public License for more details.
  18 #
  19 # You should have received a copy of the GNU General Public License
  20 # along with this program; if not, write the Free Software Foundation,
  21 # Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  22 #
  23 #-----------------------------------------------------------------------
  24 #
  25
  26 seq=`basename $0`
  27 echo "QA output created by $seq"
  28
  29 here=`pwd`
  30 tmp=/tmp/$$
  31 status=1        # failure is the default!
  32
  33 # get standard environment, filters and checks
  34 . ./common.rc
  35 . ./common.filter
  36 _supported_fs ext3 ext4 xfs btrfs reiserfs
  37 _supported_os Linux
  38 _need_to_be_root
  39 _require_scratch
  40 _require_fail_make_request
  41
  42 # TODO: Function are common enough to be moved to common.blkdev
  43 SCRATCH_REAL_DEV=`readlink -f $SCRATCH_DEV`
  44 SCRATCH_BDEV=`basename $SCRATCH_REAL_DEV`
  45
  46 allow_fail_make_request()
  47 {
  48     echo "Allow global fail_make_request feature"
  49     echo 100 > $DEBUGFS_MNT/fail_make_request/probability
  50     echo 9999999 > $DEBUGFS_MNT/fail_make_request/times
  51     echo 0 >  /sys/kernel/debug/fail_make_request/verbose
  52 }
  53
  54 disallow_fail_make_request()
  55 {
  56     echo "Disallow global fail_make_request feature"
  57     echo 0 > $DEBUGFS_MNT/fail_make_request/probability
  58     echo 0 > $DEBUGFS_MNT/fail_make_request/times
  59 }
  60
  61 start_fail_scratch_dev()
  62 {
  63     echo "Force SCRATCH_DEV device failure"
  64     echo " echo 1 > /sys/block/$SCRATCH_BDEV/make-it-fail" >> $here/$seq.full
  65     echo 1 > /sys/block/$SCRATCH_BDEV/make-it-fail
  66
  67 }
  68
  69 stop_fail_scratch_dev()
  70 {
  71     echo "Make SCRATCH_DEV device operable again"
  72     echo " echo 0 > /sys/block/$SCRATCH_BDEV/make-it-fail" >> $here/$seq.full
  73     echo 0 > /sys/block/$SCRATCH_BDEV/make-it-fail
  74
  75 }
  76
  77 _cleanup()
  78 {
  79     poweron_scratch_dev
  80     disallow_fail_make_request
  81 }
  82 trap "_cleanup; exit \$status" 1 2 3 15
  83
  84 RUN_TIME=$((20+10*$TIME_FACTOR))
  85 NUM_JOBS=$((4*LOAD_FACTOR))
  86 BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
  87 FILE_SIZE=$((BLK_DEV_SIZE * 512))
  88
  89 cat >$tmp-$seq.fio <<EOF
  90 ###########
  91 # $seq test's fio activity
  92 # Filenames derived from jobsname and jobid like follows:
  93 # ${JOB_NAME}.${JOB_ID}.${ITERATION_ID}
  94 [global]
  95 ioengine=libaio
  96 bs=4k
  97 directory=${SCRATCH_MNT}
  98 filesize=${FILE_SIZE}
  99 size=9999T
 100 continue_on_error=write
 101 ignore_error=EIO,ENOSPC:EIO
 102 error_dump=0
 103
 104 [stress_dio_aio_activity]
 105 create_on_open=1
 106 fallocate=none
 107 iodepth=128*${LOAD_FACTOR}
 108 direct=1
 109 buffered=0
 110 numjobs=${NUM_JOBS}
 111 rw=randwrite
 112 runtime=40+${RUN_TIME}
 113 time_based
 114
 115 [stress_mmap_activity]
 116 ioengine=mmap
 117 create_on_open=0
 118 fallocate=1
 119 fdatasync=40960
 120 filesize=8M
 121 size=9999T
 122 numjobs=${NUM_JOBS}
 123 rw=randwrite
 124 runtime=40+${RUN_TIME}
 125 time_based
 126
 127 EOF
 128
 129 _require_fio $tmp-$seq.fio
 130
 131 # Disable all sync operations to get higher load
 132 FSSTRESS_AVOID="$FSSTRESS_AVOID -ffsync=0 -fsync=0 -ffdatasync=0 -f setattr=1"
 133
 134 _workout()
 135 {
 136         out=$SCRATCH_MNT/fsstress.$$
 137         args=`_scale_fsstress_args -p 1 -n999999999 -f setattr=0 $FSSTRESS_AVOID -d $out`
 138         echo ""
 139         echo "Start fsstress.."
 140         echo ""
 141         echo "fsstress $args" >> $here/$seq.full
 142         $FSSTRESS_PROG $args > /dev/null 2>&1 &
 143         fs_pid=$!
 144         echo "Start fio.."
 145         cat $tmp-$seq.fio >>  $seq.full
 146         $FIO_PROG $tmp-$seq.fio >> $here/$seq.full 2>&1 &
 147         fio_pid=$!
 148
 149         # Let's it work for awhile, and force device failure
 150         sleep $RUN_TIME
 151         start_fail_scratch_dev
 152         # After device turns in to failed state filesystem may yet not know about
 153         # that so buffered write(2) may succeed, but any integrity operations
 154         # such as (sync, fsync, fdatasync, direct-io) should fail.
 155         dd if=/dev/zero of=$SCRATCH_MNT/touch_failed_filesystem count=1 bs=4k conv=fsync \
 156             >> $here/$seq.full 2>&1 && \
 157             _fail "failed: still able to perform integrity fsync on $SCRATCH_MNT"
 158
 159         kill $fs_pid
 160         wait $fs_pid
 161         wait $fio_pid
 162
 163         # We expect that broken FS still can be umounted
 164         run_check umount $SCRATCH_DEV
 165         # Once filesystem was umounted no one is able to write to block device
 166         # It is now safe to bring device back to normal state
 167         stop_fail_scratch_dev
 168
 169         # In order to check that filesystem is able to recover journal on mount(2)
 170         # perform mount/umount, after that all errors should be fixed
 171         run_check _scratch_mount
 172         run_check _scratch_unmount
 173         _check_scratch_fs
 174 }
 175
 176 # real QA test starts here
 177
 178 _scratch_mkfs >> $here/$seq.full 2>&1 || _fail "mkfs failed"
 179 _scratch_mount || _fail "mount failed"
 180 allow_fail_make_request
 181 _workout
 182 status=$?
 183 disallow_fail_make_request
 184 exit