Vulnerabilities found for Exception#to_s, NameError#to_s, and
name_err_mesg_to_s() which is Ruby interpreter-internal API. A
malicious user code can bypass $SAFE check by utilizing one of
those security holes.
Ruby's $SAFE mechanism enables untrusted user codes to run in
$SAFE >= 4 mode. This is a kind of sandboxing so some operations
are restricted in that mode to protect other data outside the
sandbox.
The problem found was around this mechanism. Exception#to_s,
NameError#to_s, and name_err_mesg_to_s() interpreter-internal API
was not correctly handling the $SAFE bits so a String object which
is not tainted can destructively be marked as tainted using them.
By using this an untrusted code in a sandbox can modify a
formerly-untainted string destructively.
Ruby 1.8 once had a similar security issue. It fixed
Exception#to_s and NameError#to_s, but name_err_mesg_to_str() issue
survived previous security fix