Extracting LLVM bitcode from ELF files generated with -fembed-bitcode.

extractgc.sh

#!/bin/sh

if [ "*$1" == "*" ]; then
    echo "usage: extract.sh src dst"
fi

if [ "*$2" == "*" ]; then
    echo "usage: extract.sh src dst"
fi

section=.llvmbc
if [ "*$3" != "*" ]; then
    section=$3
fi

echo "extracting llvm bitcode from $1 to $2"

bc=`objdump -h $1 | grep $section`
if [ "*$bc" == "*" ]; then
    echo " - llvmbc section not found"
    exit
fi

sb=$((`echo $bc | awk '{print "0x" $6}'` + 0))
ss=$((`echo $bc | awk '{print "0x" $3}'` + 0))

echo "found llvm section at $sb [$ss bytes]"

dd if=$1 of=$2 bs=1 count=$ss skip=$sb 2> /dev/null
if [ ! -f $2 ]; then
    echo " - failed"
    exit
fi    

if [ "*$section" != "*.llvmbc" ]; then
    echo "done"
    exit
fi

list=`hexdump -v -C $2 |
        grep '42 43 c0 de 35' |
        awk '{print "0x"$1 }'`

n=0
p=""
for i in $list; do
    if [ "*$p" != "*" ]; then
        n=$(($n + 1))
        l=$(($i - $p))
        echo "$n - $p, $l bytes"
        dd if="$2" of="$n-$2" bs=1 count=$(($l)) skip=$(($p)) 2> /dev/null
    fi
    p=$i
done

size=$(($(wc -c < "$2") +0))
echo "$2 [$size bytes]:"

if [ "*p" != "" ]; then
    n=$(($n + 1))
    l=$(($size - $p))
    echo "$n - $p, $l bytes"
    dd if="$2" of="$n-$2" bs=1 count=$(($l)) skip=$(($p)) 2> /dev/null
fi

for i in *.bc; do
    hasPadding=`hexdump -v -C $i | tail -2 | head -1 | grep '00 00 00 00 00 00 00' | wc -l`
    pad=$(($hasPadding + 0))
    echo $i pad="$pad";
    if [ $pad == 1 ]; then
        echo " - removing trailing zeros"
        mv $i tmp-$i
        s1=$(($(wc -c < "tmp-$i") - 4))
        dd if="tmp-$i" of="$i" bs=1 count=$s1 skip=0 2> /dev/null
    fi
done