[archive] symlinks outside archive directory created by _make_leading_paths

[pmoravec]

When archive._make_leading_paths is called to a path containing a symlink to an absolute path, the method creates a symlink outside the archive directory (to the absolute path of the destination of the symlink). Copying that archive to a different machine then causes broken symlink.

Reproducer:

mv /var/lib/mistral{,.prev}
mkdir -p /var/lib/mistral/overcloud
ln -s /var/lib/mistral/overcloud /var/lib/mistral/config-download-latest
echo "some ansible log content" > /var/lib/mistral/overcloud/ansible.log

Now, call:

sosreport -o openstack_instack --batch --build

and see that /var/tmp/sosreport-myhostname-2019-06-25-wpyemcm/var/lib/mistral/config-download-latest is a symlink to /var/lib/mistral/overcloud, while the archive directory itself does not contain a copy of the logfile.

Root cause of the problem: https://github.com/sosreport/sos/blob/master/sos/archive.py#L221-L237 calls:

target = os.readlink(src_path)
..
os.symlink(target, abs_path)

This works well for symlinks with relative destination. For absolute path destination, something like:

if os.path.isabs(target):
    target = target.strip(os.path.commonpath([target, src_path])).strip(os.sep)

must be called in between. This will make the target relative from absolute.

BUT this works well only when target and src_path are in common subdirectory. If e.g. /var/lib/mistral/config-download-latest would be a symlink to /var/log, we must set target to ../../log and create that relative path via self._make_leading_paths first.

The above proposal is sufficient for the real world scenario (Openstack Instack having /var/lib/mistral/config-download-latest: symbolic link to /var/lib/mistral/overcloud per my understanding), but it is not ideal.

[pmoravec]

.. or we can create symlink to (not-tested):

target = os.readlink(src_path)
..
target = os.path.relpath(target, start=src_path)
os.symlink(target, abs_path)

(and somehow take care the target path is created)