[Bug-tar] new/old extract problems with 6GB sparse file

[Joerg Delker]

Hi folks,

when trying to recover a very important backup (ntfsclone sparse file), 
I had to notice that tar (1.15.1) had trouble to extract that file.

This archive was created with tar (1.1[45].x ?) about a month ago with 
"tar -cjSf jake.tar.bz2 jake.ntfs", compressing the 6GB ntfs sparse 
image into a 1.6GB tar.bz2 file.

All attempts to recover this 6GB sparse file have failed so far (either 
sparsed or zero-filled).
So, I did some research on the list and found several issues with equal 
or similar cases. However, they all (?) should have been fixed in cvs 
already.

All further test were tried with the version from cvs then (as of 18 
June 2006):

$ ls -al jake_winxp.ntfs.tar.bz2
-rwxr-xr-x 1 root root 1610433554 Jun 16 13:33 jake.tar.bz2

$ tar --version
tar (GNU tar) 1.15.91
...

$ tar -tvjf jake.tar.bz2
-rwx------ user/group 6292304384 2006-05-07 14:26:28 jake.ntfs

$ tar -xvjf jake.tar.bz2 --totals
jake.ntfs
Total bytes read: 3155937280 (3.0GiB, 3.2MiB/s)

This results in an extracted file of length 1997337088 bytes (~1.9GB), 
which looks very suspicious like a 4GB overflow!?
When I watched the file size during extraction, I noticed that it grew 
in size up to 4333572096 bytes, stayed there for a minute while 
extraction continued, until it finally dropped down to above size (1.9GB).

Trying to pipe the stream results in this:

$ tar -xvjf jake.tar.bz2 -O --totals | cat >jake.ntfs
jake.ntfs
tar: jake.ntfs: Cannot seek to 61198336: Invalid argument
Total bytes read: 3155937280 (3.0GiB, 2.6MiB/s)

Resulting in an extracted file length of 4333572096 bytes for jake.ntfs, 
which is still far away from the correct file size.

All tests were done on ext3fs.

Any hints how to recover this archive is very welcome!

JD