On Mon, 07 May 2012 18:30:08 +0200
Sasha Levin [off-list ref] wrote:

On Mon, 2012-05-07 at 17:11 +0100, Alan Cox wrote:

quoted

quoted

I don't believe that this change is correct.

Consider the following scenario:

tty_release -> tty_lock -> pty_close -> tty_vhangup -> tty_lock

We hang up tty->link not tty.

It's now a per tty lock. So I think we are ok.

Unless we can cause tty->link == tty, in which case:

We should not be able to cause tty->link == tty. So that's a different
problem altogether.

tty->link is set to point to the other half of the pty in pty_install and
in pty98_unix_install. It's never assigned to the same tty and ptys
simply wouldn't work if this wasn't the case.

So whatever your trace is showing, that's not the bug. Something more
complicated would appear to be afoot.

Alan