BUG 801241 - xfsdump can cause filesystem corruption

[pv@xxxxxxxxxxxxx (ivanr@xxxxxxxxxxxx)]

View Incident: 
http://co-op.engr.sgi.com/BugWorks/code/bwxquery.cgi?search=Search&wlong=1&view_type=Bug&wi=801241

Submitter : ivanr                     Submitter Domain : engr               
Assigned Engineer : nb                Assigned Domain : sgi.com             
Assigned Group : xfs-linux            Category : software                   
Customer Reported : F                 Priority : 1                          
Project : xfs-linux                   Status : open                         
Description :
QA for xfsdump/xfsrestore revealed that it is possible for
xfsdump to cause filesystem corruption.  The test involves
creating a filesystem, creating a whole buncha files, xfsdumping
them to a file somewhere, then removing all the files.  xfs_check
reports lots of inodes with zero link counts, etc.  It's probably
due to xfsdump using bulkstat on the entire fs.

The following is a script which will reliably (at least on
bruce.melbourne) produce the error.  Run it in cmd/xfs/stress -
it needs the src/fill program, and make sure you fix the
variables at the top.

=======================================
#!/bin/sh

#TEST_DIR=/mnt/xfs1
#TEST_DEV=/dev/sda10
TEST_DIR=<put something here>
TEST_DEV=<put something here>

tmp=/tmp/$$
dump_dir=$TEST_DIR/dump.$$
dump_file=$tmp.dumpfile
here=`pwd`

trap "rm -rf $tmp\*; exit" 0 1 2 3 15

_wipe_dev()
{
        umount $1
        mkfs.xfs -f $1
        mount $1
}

_create_dumpdir_fill()
{
    echo "Creating directory system to dump using src/fill."

    # wipe test dir clean first
    rm -rf $TEST_DIR/*

    cat <<End-of-File >$tmp.config
# pathname      size in bytes
#
small           10
big             102400
sub/small       10
sub/big         102400
#
sub/a           1
sub/b           2
sub/c           4
sub/d           8
sub/e           16
sub/f           32
sub/g           64
sub/h           128
sub/i           256
sub/j           512
sub/k           1024
sub/l           2048
sub/m           4096
sub/n           8192
#
sub/a00         100
sub/b00         200
sub/c00         400
sub/d00         800
sub/e00         1600
sub/f00         3200
sub/g00         6400
sub/h00         12800
sub/i00         25600
sub/j00         51200
sub/k00         102400
sub/l00         204800
sub/m00         409600
sub/n00         819200
#
sub/a000        1000
sub/e000        16000
sub/h000        128000
sub/k000        1024000
End-of-File

    if mkdir -p $dump_dir
    then
        :
    else
        echo "Error: cannot mkdir \"$dump_dir\""
        exit 1
    fi
    cd $dump_dir

    sed -e '/^#/d' $tmp.config \
    | while read file nbytes
    do
        dir=`dirname $file`
        if [ "$dir" != "." ]
        then
            if [ ! -d $dir ]
            then
                if mkdir $dir
                then
                    :
                else
                    echo "Error: cannot mkdir \"$dir\""
                    exit 1
                fi
            fi
        fi
        rm -f $file
        if $here/src/fill $file $file $nbytes
        then
            :
        else
            echo "Error: cannot create \"$file\""
            exit 1
        fi
    done

    cd $here
}

_fix_malloc()
{
    # filter out the Electric Fence notice
    perl -e '
        undef $/;
        $_ = <>;
        s/\n  Electric Fence .*\n//g;
        print'
}

NO_REMOUNT=1

_check_fs()
{
    device=$1

    xfs_check $device 2>&1 | _fix_malloc > $tmp.xfscheck
    if [ -s $tmp.xfscheck ]
    then
        echo "_check_fs: filesystem on $device is inconsistent (xfs_check)"
        rm -f $tmp.xfscheck
    fi

    if ! xfs_repair -n $device > /dev/null 2>&1
    then
        echo "_check_fs: filesystem on $device is inconsistent (xfs_repair)"
    fi
}

_wipe_dev $TEST_DEV

_create_dumpdir_fill

sync
sleep 15

xfsdump -f $dump_file -M med -L sess $TEST_DIR > /dev/null

rm -rf $dump_dir

_check_fs $TEST_DEV
=======================================