tests/generic/019

   1 #! /bin/bash
   2 # SPDX-License-Identifier: GPL-2.0
   3
   4 #
   5 # FSQA Test No. generic/019
   6 #
   7 # Run fsstress and fio(dio/aio and mmap) and simulate disk failure
   8 # check filesystem consistency at the end.
   9 #
  10 . ./common/preamble
  11 _begin_fstest aio dangerous enospc rw stress
  12
  13 fio_config=$tmp.fio
  14
  15 # Import common functions.
  16 . ./common/filter
  17 _supported_fs generic
  18 _require_scratch
  19 _require_block_device $SCRATCH_DEV
  20 _require_fail_make_request
  21
  22 SYSFS_BDEV=`_sysfs_dev $SCRATCH_DEV`
  23
  24 allow_fail_make_request()
  25 {
  26     echo "Allow global fail_make_request feature"
  27     echo 100 > $DEBUGFS_MNT/fail_make_request/probability
  28     echo 9999999 > $DEBUGFS_MNT/fail_make_request/times
  29     echo 0 >  /sys/kernel/debug/fail_make_request/verbose
  30 }
  31
  32 disallow_fail_make_request()
  33 {
  34     echo "Disallow global fail_make_request feature"
  35     echo 0 > $DEBUGFS_MNT/fail_make_request/probability
  36     echo 0 > $DEBUGFS_MNT/fail_make_request/times
  37 }
  38
  39 start_fail_scratch_dev()
  40 {
  41     echo "Force SCRATCH_DEV device failure"
  42     echo " echo 1 > $SYSFS_BDEV/make-it-fail" >> $seqres.full
  43     echo 1 > $SYSFS_BDEV/make-it-fail
  44 }
  45
  46 stop_fail_scratch_dev()
  47 {
  48     echo "Make SCRATCH_DEV device operable again"
  49     echo " echo 0 > $SYSFS_BDEV/make-it-fail" >> $seqres.full
  50     echo 0 > $SYSFS_BDEV/make-it-fail
  51 }
  52
  53 # Override the default cleanup function.
  54 _cleanup()
  55 {
  56     poweron_scratch_dev
  57     disallow_fail_make_request
  58     rm -f $tmp.*
  59 }
  60
  61 RUN_TIME=$((20+10*$TIME_FACTOR))
  62 NUM_JOBS=$((4*LOAD_FACTOR))
  63 BLK_DEV_SIZE=`blockdev --getsz $SCRATCH_DEV`
  64 FILE_SIZE=$((BLK_DEV_SIZE * 512))
  65
  66 cat >$fio_config <<EOF
  67 ###########
  68 # $seq test's fio activity
  69 # Filenames derived from jobsname and jobid like follows:
  70 # ${JOB_NAME}.${JOB_ID}.${ITERATION_ID}
  71 [global]
  72 ioengine=libaio
  73 bs=4k
  74 directory=${SCRATCH_MNT}
  75 filesize=${FILE_SIZE}
  76 size=9999T
  77 continue_on_error=write
  78 ignore_error=EIO,ENOSPC:EIO
  79 error_dump=0
  80
  81 [stress_dio_aio_activity]
  82 create_on_open=1
  83 fallocate=none
  84 iodepth=128*${LOAD_FACTOR}
  85 direct=1
  86 buffered=0
  87 numjobs=${NUM_JOBS}
  88 rw=randwrite
  89 runtime=40+${RUN_TIME}
  90 time_based
  91
  92 [stress_mmap_activity]
  93 ioengine=mmap
  94 create_on_open=0
  95 fallocate=1
  96 fdatasync=40960
  97 filesize=8M
  98 size=9999T
  99 numjobs=${NUM_JOBS}
 100 rw=randwrite
 101 runtime=40+${RUN_TIME}
 102 time_based
 103
 104 EOF
 105
 106 _require_fio $fio_config
 107
 108 # Disable all sync operations to get higher load
 109 FSSTRESS_AVOID="$FSSTRESS_AVOID -ffsync=0 -fsync=0 -ffdatasync=0 -f setattr=1"
 110
 111 _workout()
 112 {
 113         out=$SCRATCH_MNT/fsstress.$$
 114         args=`_scale_fsstress_args -p 1 -n999999999 -f setattr=0 $FSSTRESS_AVOID -d $out`
 115         echo ""
 116         echo "Start fsstress.."
 117         echo ""
 118         echo "fsstress $args" >> $seqres.full
 119         $FSSTRESS_PROG $args > /dev/null 2>&1 &
 120         fs_pid=$!
 121         echo "Start fio.."
 122         cat $fio_config >>  $seqres.full
 123         $FIO_PROG $fio_config >> $seqres.full 2>&1 &
 124         fio_pid=$!
 125
 126         # Let's it work for awhile, and force device failure
 127         sleep $RUN_TIME
 128         start_fail_scratch_dev
 129         # After device turns in to failed state filesystem may yet not know about
 130         # that so buffered write(2) may succeed, but any integrity operations
 131         # such as (sync, fsync, fdatasync, direct-io) should fail.
 132         dd if=/dev/zero of=$SCRATCH_MNT/touch_failed_filesystem count=1 bs=4k conv=fsync \
 133             >> $seqres.full 2>&1 && \
 134             _fail "failed: still able to perform integrity fsync on $SCRATCH_MNT"
 135
 136         kill $fs_pid &> /dev/null
 137         wait $fs_pid
 138         wait $fio_pid
 139
 140         # We expect that broken FS still can be umounted
 141         run_check _scratch_unmount
 142         # Once filesystem was umounted no one is able to write to block device
 143         # It is now safe to bring device back to normal state
 144         stop_fail_scratch_dev
 145
 146         # In order to check that filesystem is able to recover journal on mount(2)
 147         # perform mount/umount, after that all errors should be fixed
 148         _scratch_mount
 149         run_check _scratch_unmount
 150 }
 151
 152 # real QA test starts here
 153
 154 _scratch_mkfs >> $seqres.full 2>&1 || _fail "mkfs failed"
 155 _scratch_mount
 156 allow_fail_make_request
 157 _workout
 158 status=$?
 159 disallow_fail_make_request
 160 exit