Hi Sergey,

Thank you for testing the report and for providing the exact diagnostics.

I reviewed my frozen laboratory procedure and found that the reproduction
package I sent was incomplete. The README incorrectly instructed extraction
of graal-v6.tar into an empty directory.

The successful reproduction requires an incremental restore pre-state
created by two earlier extraction steps on the same destination tree:

   1.

   extract the level-0 archive using --listed-incremental=/dev/null;
   2.

   extract a setup archive that creates root/alias as a symlink resolving
   outside the logical root subtree;
   3.

   extract graal-v6.tar using --listed-incremental=/dev/null.

The diagnostic you observed:

Cannot rename 'root/alias' to 'root/b': No such file or directory

is consistent with the missing pre-state: the GNU.dumpdir transition
contained in the archive expects root/alias to exist before it is renamed
to root/b.

This was my mistake in preparing the disclosure package, and I apologize
for the incomplete reproduction instructions.

I am rebuilding the package so that it contains all three archives, a
deterministic reproduce.sh, the expected filesystem layout, integrity
hashes, and automatic verification of the exit status, stderr, resulting
inode relationship, and outside-subtree object.

I will send the corrected self-contained package after reproducing it again
from a clean directory using only the files included in that package.

Kind regards,

Federico

Il giorno gio 23 lug 2026 alle ore 11:41 Sergey Poznyakoff <[email protected]>
ha scritto:

> Hi Federico,
>
> Trying to reproduce your results on current git HEAD, I get the
> following:
>
> 1. Extraction fails.
> 2. Exit status is 2.
> 3. Diagnostics on stderr:
>    tar: Cannot rename 'root/alias' to 'root/b': No such file or directory
>    tar: Exiting with failure status due to previous errors
> 4. After extraction, root/new/parent/b/victim is a hardlink to
>    root/b/victim
> 5. No files outside the extraction directory are created.
>
> FWIW, I got the same results when running tar 1.35.
>
> Best regards,
> Sergey
>
>

Reply via email to