Re: [Qemu-devel] pty/tty functions for BSD too

["Blue Swirl" <blauwirbel@gmail.com>]

On 8/18/08, Blue Swirl <blauwirbel@gmail.com> wrote:
> On 8/18/08, Daniel P. Berrange <berrange@redhat.com> wrote:
>  > On Mon, Aug 18, 2008 at 07:16:36PM +0300, Blue Swirl wrote:
>  >  > On 8/18/08, Samuel Thibault <samuel.thibault@ens-lyon.org> wrote:
>  >  > > Anthony Liguori, le Mon 18 Aug 2008 09:06:41 -0500, a écrit :
>  >  > >
>  >  > > > Samuel Thibault wrote:
>  >  > >  > >In Xen, pty/tty functions are enabled for BSD too, shouldn't we enable
>  >  > >  > >them in upstream qemu too, as patched below?
>  >  > >  > >
>  >  > >  >
>  >  > >  > And you're sure that these functions compile/work on NetBSD/OpenBSD?
>  >  > >
>  >  > >
>  >  > > The defines are explicit in Xen, so I guess somebody tested it.  I
>  >  > >  haven't myself.  I wonder why there is no FreeBSD however.
>  >  >
>  >  > The patch does not work on OpenBSD, because while openpty() is
>  >  > available, ptsname() isn't.
>  >  >
>  >  > I tested the attached version on OpenBSD and Linux, pty name is
>  >  > printed correctly.
>  >
>  >
>  > Passing a non-NULL value to openpty()'s name parameter is not safe
>  >
>  >  [quote openpty(1)]
>  >  BUGS
>  >        Nobody knows how much space should be reserved for name.  So, call-
>  >        ing openpty() or forkpty() with non-NULL name may not be secure.
>  >  [/quote]
>
>
> Maybe in theory, but in practice the name will be
>  /dev/pty[0-9a-z][a-z] or /dev/pts/[0-9]* or something similar. Even if
>  they are not, PATH_MAX should be enough.
>
>
>  >  If BSD has no other way to determine the PTY name, then at least it
>  >  should be conditionalized so that systems with ptsname() use it, only
>  >  falling back to using the 'name' arg to openpty() for OS lacking ptsname
>
>
> I'm not convinced (yet?) this will be worth it.

Glibc uses an internal PATH_MAX buffer. If the name does not fit, the
buffer will be doubled in size:
http://sourceware.org/cgi-bin/cvsweb.cgi/libc/login/openpty.c?rev=1.8&content-type=text/x-cvsweb-markup&cvsroot=glibc

But in practice the pty will be allocated using getpt(), which uses
hard coded ptmx name:
http://sourceware.org/cgi-bin/cvsweb.cgi/libc/sysdeps/unix/sysv/linux/getpt.c?rev=1.10&content-type=text/x-cvsweb-markup&cvsroot=glibc

And ptsname() is used to get the pty name, which uses either hard
coded buffer length of _PATH_DEVPTS or _PATH_TTY:
http://sourceware.org/cgi-bin/cvsweb.cgi/libc/sysdeps/unix/sysv/linux/ptsname.c?rev=1.15&content-type=text/x-cvsweb-markup&cvsroot=glibc

OpenBSD forces the name to be in format "/dev/ptyXX":
http://www.openbsd.org/cgi-bin/cvsweb/src/lib/libutil/pty.c?rev=1.15&content-type=text/x-cvsweb-markup

NetBSD is similar, except also /dev/ttyXX is possible:
http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libutil/pty.c?rev=1.29&content-type=text/x-cvsweb-markup

I still maintain that in reality, PATH_MAX should be enough.