tar excludes files too late to stop hardlink detection

[Harald van Dijk]

On 27/06/2021 15:15, Harald van Dijk wrote:
> On 26/06/2021 00:36, Harald van Dijk wrote:
>> Hi,
>>
>> tar --exclude results in bad archives when hardlinks are used. 
>> Consider the following:
>>
>>    $ mkdir tartest
>>    $ echo hello > tartest/a
>>    $ ln tartest/a tartest/b
>>    $ busybox tar cf - tartest | tar tvf -
>>    drwxr-xr-x harald/harald     0 2021-06-26 00:25 tartest/
>>    -rw-r--r-- harald/harald     6 2021-06-26 00:25 tartest/b
>>    hrw-r--r-- harald/harald     0 2021-06-26 00:25 tartest/a link to 
>> tartest/b
>>
>> This is okay. tar may either pick up a first and then detect b as a 
>> hardlink to a, or pick up b first and then detect a as a hardlink to 
>> b. On my system, it picks up b first. You can adjust the below 
>> accordingly if on your system a is picked up first. Now, exclude b:
>>
>>    $ busybox tar cf - --exclude=b tartest | tar tvf -
>>    drwxr-xr-x harald/harald     0 2021-06-26 00:25 tartest/
>>    hrw-r--r-- harald/harald     0 2021-06-26 00:25 tartest/a link to 
>> tartest/b
>>
>> This resulted in an archive where the contents of tartest/a are 
>> missing. Extracting the archive results in an attempt to hardlink 
>> tartest/b, which may or may not exist in the target directory. GNU tar 
>> does not do this, it stores the contents of the file instead, which 
>> seems like a better idea to me. Can busybox be modified to do that as 
>> well?
>>
>> Tested with busybox 1.33.1.
> 
> It seems like the fix is trivial, please see attached patch.

ping

> Cheers,
> Harald van Dijk