Re: [PATCH] nsenter: add support for pty

[ebiederm@xmission.com (Eric W. Biederman)]

Karel Zak <kzak@redhat.com> writes:

> On Wed, Mar 18, 2015 at 10:53:19AM +0100, Jörg Thalheim wrote:
>> If mount namespaces are used, the issued command, will not have access to the
>> tty device attached to its stdin/stdout/stderr. This patch adds an option to
>> allocate a new pseudo tty in the entered mount namespace and bridge between the
>> origin standard file descriptors and the standard file descriptors of the
>> executed command.
>
> The original nsenter(1) purpose is to have command line interface to
> setns(2) syscall. Your patch is trying to push us to something more
> complex. Not sure if we really want it. Eric, any comment?

I certainly would not want it to be the default.
After seeing the ptsname() and gpg I see what is driving it.

However playing the pty games gets us smack dab in the middle of sending
and receiving trusted input.  I don't know that I want nsenter to be on
the trusted path for entering in passwords for unlocking gpg keys.

If gpg is the driving use case I don't think it is wise to add pty
support.  All of a sudden nsenter has to become robust from side channel
attacks when you are entering in passwords and I at least have no
interest in maintaining nsenter in that way.

Eric