PDA

View Full Version : small script to re-hardlink *ALL* equal files



cehteh
01-02-2003, 05:35 PM
Have fun, this is not well tested, it will prolly work fine with readonly CD-Images but certainly have problems with hd-installs since there are equal files which *should not* be hardlinked! Take care.
One could probably add filters to the 'find' command to exclude such files. Since I did this as fast hack for CD remastering so i have no need for better selections. I'm working on a much more advanced file-unifier for the vserver project. When that is finished i will announce it here too (ETA: Mid Feburary, i'm in hollyday next time)

Results (compressed image):
691M Jan 2 13:49 KNOPPIX.new
664M Jan 2 17:25 KNOPPIX.new.unified



PS: Klaus ... if you use this for the main Knoppix, the free space you get is dedicated for a recent version of 'distcc', thanks.



#!/bin/sh

find $1 -type f -exec md5sum {} \; |\
sort +0 |\
{
read sum file;
while [ "$sum" != "" ]; do
while read sum2 file2 && [ $sum = $sum2 ]; do
ln -f "$file" "$file2"
done
sum=$sum2
file="$file2"
done
}

knopper
01-02-2003, 06:17 PM
Have fun, this is not well tested, it will prolly work fine with readonly CD-Images but certainly have problems with hd-installs since there are equal files which *should not* be hardlinked! Take care.
One could probably add filters to the 'find' command to exclude such files. Since I did this as fast hack for CD remastering so i have no need for better selections. I'm working on a much more advanced file-unifier for the vserver project. When that is finished i will announce it here too (ETA: Mid Feburary, i'm in hollyday next time)

Results (compressed image):
691M Jan 2 13:49 KNOPPIX.new
664M Jan 2 17:25 KNOPPIX.new.unified



PS: Klaus ... if you use this for the main Knoppix, the free space you get is dedicated for a recent version of 'distcc', thanks.



#!/bin/sh

find $1 -type f -exec md5sum {} \; |\
sort +0 |\
{
read sum file;
while [ "$sum" != "" ]; do
while read sum2 file2 && [ $sum = $sum2 ]; do
ln -f "$file" "$file2"
done
sum=$sum2
file="$file2"
done
}


This looks a little dangerous. Do you realize that this makes all empty (0 bytes) files hardlinks to each other, even locks and logfiles? Also, some files with the same checksum may NOT be actually the same file or have desirably the same contents.

I use a harddisk-installed Debian system that should have all hardlinks intact. If you copied your system from CD, this may not always be the case. You should use rsync -Ha for copying to preserve the hardlinks.

-KK

cehteh
01-02-2003, 06:33 PM
sure .. i know it is a fast hack ... the next tool (vserver unifier) will be much better. Resolving *All* problems u showed.

>This looks a little dangerous. Do you realize that this makes all empty (0 bytes) files hardlinks to each other,

use '! -size 0' within the find

> locks and logfiles
there are no locks when mastering,
logs are not important when kept only in ram ..
! -path '*/var/log*' will fix it anyways

>Also, some files with the same checksum may NOT be actually the same file or have desirably the same contents.
with md5sum the chance of doubletes are microscopic (even less), but you are right .. for the vserver unifier i dont use checksuming

remember this is meant as fast hack and to show a 'general' concept for reducing the diskusage of knoppix.

cheers :)

cehteh
01-03-2003, 07:18 AM
- compares files
- only hardlink files which are not hardlinked yet
- permission and uid/gid must be equal
- size 0 files are excluded
- can be called with multiple arguments which are used directly by find
- writes a script which can be used to undo all hardlinks

i will not do any more updates on this and focus on the vserver unifier now (it will get a small control language and tons of more sophisticated options)


example usage:
./knoppix.unify Path/To/KNOPPIX ! -path "*/var/log/*" >undo_hardlinks.sh

File knoppix.unify:


#!/bin/bash

find "$@" -type f ! -size 0 -exec md5sum {} \; |\
sort +0 |\
{
read sum file;
while [ "$sum" != "" ]; do
while read sum2 file2 &&\
[ $sum = $sum2 ] &&\
[ $(cut -d " " -f 1 <(ls -i "$file")) != $(cut -d " " -f 1 <(ls -i "$file2")) ] &&\
[ $(cut -c -32 <(ls -ln "$file")) = $(cut -c -32 <(ls -ln "$file2")) ] &&\
cmp "$file" "$file2";
do
ln -f "$file" "$file2" &&
echo "rm -f \"$file2\"; cp -pf \"$file\" \"$file2\""
done
sum=$sum2
file="$file2"
done
}