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 # creator
  26 owner=dmonakhov@openvz.org
  27
  28 seq=`basename $0`
  29 echo "QA output created by $seq"
  30
  31 here=`pwd`
  32 tmp=/tmp/$$
  33 status=1        # failure is the default!
  34
  35 # get standard environment, filters and checks
  36 . ./common.rc
  37 . ./common.filter
  38 _supported_fs ext3 ext4 xfs btrfs reiserfs
  39 _supported_os Linux
  40 _need_to_be_root
  41 _require_scratch
  42 _require_fail_make_request
  43
  44 # TODO: Function are common enough to be moved to common.blkdev
  45 SCRATCH_REAL_DEV=`readlink -f $SCRATCH_DEV`
  46 SCRATCH_BDEV=`basename $SCRATCH_REAL_DEV`
  47
  48 allow_fail_make_request()
  49 {
  50     echo "Allow global fail_make_request feature"
  51     echo 100 > $DEBUGFS_MNT/fail_make_request/probability
  52     echo 9999999 > $DEBUGFS_MNT/fail_make_request/times
  53     echo 0 >  /sys/kernel/debug/fail_make_request/verbose
  54 }
  55
  56 disallow_fail_make_request()
  57 {
  58     echo "Disallow global fail_make_request feature"
  59     echo 0 > $DEBUGFS_MNT/fail_make_request/probability
  60     echo 0 > $DEBUGFS_MNT/fail_make_request/times
  61 }
  62
  63 start_fail_scratch_dev()
  64 {
  65     echo "Force SCRATCH_DEV device failure"
  66     echo " echo 1 > /sys/block/$SCRATCH_BDEV/make-it-fail" >> $here/$seq.full
  67     echo 1 > /sys/block/$SCRATCH_BDEV/make-it-fail
  68
  69 }
  70
  71 stop_fail_scratch_dev()
  72 {
  73     echo "Make SCRATCH_DEV device operable again"
  74     echo " echo 0 > /sys/block/$SCRATCH_BDEV/make-it-fail" >> $here/$seq.full
  75     echo 0 > /sys/block/$SCRATCH_BDEV/make-it-fail
  76
  77 }
  78
  79 _cleanup()
  80 {
  81     poweron_scratch_dev
  82     disallow_fail_make_request
  83 }
  84 trap "_cleanup; exit \$status" 1 2 3 15
  85
  86 RUN_TIME=$((20+10*$TIME_FACTOR))
  87 NUM_JOBS=$((4*LOAD_FACTOR))
  88 BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
  89 FILE_SIZE=$((BLK_DEV_SIZE * 512))
  90
  91 cat >$tmp-$seq.fio <<EOF
  92 ###########
  93 # $seq test's fio activity
  94 # Filenames derived from jobsname and jobid like follows:
  95 # ${JOB_NAME}.${JOB_ID}.${ITERATION_ID}
  96 [global]
  97 ioengine=libaio
  98 bs=4k
  99 directory=${SCRATCH_MNT}
 100 filesize=${FILE_SIZE}
 101 size=9999T
 102 continue_on_error=write
 103 ignore_error=EIO,ENOSPC:EIO
 104 error_dump=0
 105
 106 [stress_dio_aio_activity]
 107 create_on_open=1
 108 fallocate=none
 109 iodepth=128*${LOAD_FACTOR}
 110 direct=1
 111 buffered=0
 112 numjobs=${NUM_JOBS}
 113 rw=randwrite
 114 runtime=40+${RUN_TIME}
 115 time_based
 116
 117 [stress_mmap_activity]
 118 ioengine=mmap
 119 create_on_open=0
 120 fallocate=1
 121 fdatasync=40960
 122 filesize=8M
 123 size=9999T
 124 numjobs=${NUM_JOBS}
 125 rw=randwrite
 126 runtime=40+${RUN_TIME}
 127 time_based
 128
 129 EOF
 130
 131 _require_fio $tmp-$seq.fio
 132
 133 # Disable all sync operations to get higher load
 134 FSSTRESS_AVOID="$FSSTRESS_AVOID -ffsync=0 -fsync=0 -ffdatasync=0 -f setattr=1"
 135
 136 _workout()
 137 {
 138         out=$SCRATCH_MNT/fsstress.$$
 139         args=`_scale_fsstress_args -p 1 -n999999999 -f setattr=0 $FSSTRESS_AVOID -d $out`
 140         echo ""
 141         echo "Start fsstress.."
 142         echo ""
 143         echo "fsstress $args" >> $here/$seq.full
 144         $FSSTRESS_PROG $args > /dev/null 2>&1 &
 145         fs_pid=$!
 146         echo "Start fio.."
 147         cat $tmp-$seq.fio >>  $seq.full
 148         $FIO_PROG $tmp-$seq.fio >> $here/$seq.full 2>&1 &
 149         fio_pid=$!
 150
 151         # Let's it work for awhile, and force device failure
 152         sleep $RUN_TIME
 153         start_fail_scratch_dev
 154         # After device turns in to failed state filesystem may yet not know about
 155         # that so buffered write(2) may succeed, but any integrity operations
 156         # such as (sync, fsync, fdatasync, direct-io) should fail.
 157         dd if=/dev/zero of=$SCRATCH_MNT/touch_failed_filesystem count=1 bs=4k conv=fsync \
 158             >> $here/$seq.full 2>&1 && \
 159             _fail "failed: still able to perform integrity fsync on $SCRATCH_MNT"
 160
 161         kill $fs_pid
 162         wait $fs_pid
 163         wait $fio_pid
 164
 165         # We expect that broken FS still can be umounted
 166         run_check umount $SCRATCH_DEV
 167         # Once filesystem was umounted no one is able to write to block device
 168         # It is now safe to bring device back to normal state
 169         stop_fail_scratch_dev
 170
 171         # In order to check that filesystem is able to recover journal on mount(2)
 172         # perform mount/umount, after that all errors should be fixed
 173         run_check _scratch_mount
 174         run_check _scratch_unmount
 175         _check_scratch_fs
 176 }
 177
 178 # real QA test starts here
 179
 180 _scratch_mkfs >> $here/$seq.full 2>&1 || _fail "mkfs failed"
 181 _scratch_mount || _fail "mount failed"
 182 allow_fail_make_request
 183 _workout
 184 status=$?
 185 disallow_fail_make_request
 186 exit