[CRIU] Re: [PATCH 3/6] tty: Add checkpoint/restore for unix terminals v5

[Cyrill Gorcunov]

On Tue, Sep 11, 2012 at 06:14:01PM +0400, Pavel Emelyanov wrote:
> 
> This all looks very good. One question:
> 
> > +static int pty_get_flags(int lfd, int major, int index, TtyInfoEntry *e)
> > +{
> > +	int slave;
> > +
> > +	e->locked	= false;
> > +	e->exclusive	= false;
> > +	e->packet_mode	= false;
> 
...
> > +	/*
> > +	 * FIXME
> > +	 *
> > +	 * PTYs are safe to use packet mode. While there
> > +	 * is no way to fetch packet mode settings from
> > +	 * the kernel, without it we see echos missing
> > +	 * in `screen' application restore.
> > +	 */
> > +	if (ioctl(master, TIOCPKT, &pktmode) < 0) {
> > +		pr_perror("Can't set packed mode");
> > +		goto err;
> > +	}
> 
> Why do we set packet mode in the image to false and treat it as true on restore?

That's drawback from kernel. Look, the situation is that -- we can't fetch this
flag from kernel without kernel patching but it's been found that without this
flag set the "screen" doesn't wotk well. Thus I had to do a decision -- either
write this flag as "true" on dump time and put a fixme telling why we do that,
or write it to false since we can't fetch it but treat as true on restore and
add fixme in another code area.

I thought that write "true" in PB image will be more incorrect.

	Cyrill