Re: [Qemu-devel] [PATCH] net: detect errors from probing vnet hdr flag for TAP devices

[Daniel P. Berrange]

On Fri, Oct 27, 2017 at 01:59:22PM +0100, Dr. David Alan Gilbert wrote:
> * Daniel P. Berrange (address@hidden) wrote:
> > When QEMU sets up a tap based network device backend, it mostly ignores 
> > errors
> > reported from various ioctl() calls it makes, assuming the TAP file 
> > descriptor
> > is valid. This assumption can easily be violated when the user is passing 
> > in a
> > pre-opened file descriptor. At best, the ioctls may fail with a -EBADF, but 
> > if
> > the user passes in a bogus FD number that happens to clash with a FD number 
> > that
> > QEMU has opened internally for another reason, a wide variety of errnos may
> > result, as the TUNGETIFF ioctl number may map to a completely different 
> > command
> > on a different type of file.
> > 
> > By ignoring all these errors, QEMU sets up a zombie network backend that 
> > will
> > never pass any data. Even worse, when QEMU shuts down, or that network 
> > backend
> > is hot-removed, it will close this bogus file descriptor, which could 
> > belong to
> > another QEMU device backend.
> > 
> > There's no obvious guaranteed reliable way to detect that a FD genuinely is 
> > a
> > TAP device, as opposed to a UNIX socket, or pipe, or something else. 
> > Checking
> > the errno from probing vnet hdr flag though, does catch the big common 
> > cases.
> > ie calling TUNGETIFF will return EBADF for an invalid FD, and ENOTTY when 
> > FD is
> > a UNIX socket, or pipe which catches accidental collisions with FDs used for
> > stdio, or monitor socket.
> > 
> > Previously the example below where bogus fd 9 collides with the FD used for 
> > the
> > chardev saw:
> > 
> > $ ./x86_64-softmmu/qemu-system-x86_64 -netdev tap,id=hostnet0,fd=9 \
> >   -chardev socket,id=charchannel0,path=/tmp/qga,server,nowait \
> >   -monitor stdio -vnc :0
> > qemu-system-x86_64: -netdev tap,id=hostnet0,fd=9: TUNGETIFF ioctl() failed: 
> > Inappropriate ioctl for device
> > TUNSETOFFLOAD ioctl() failed: Bad address
> > QEMU 2.9.1 monitor - type 'help' for more information
> > (qemu) Warning: netdev hostnet0 has no peer
> > 
> > which gives a running QEMU with a zombie network backend.
> > 
> > With this change applied we get an error message and QEMU immediately exits
> > before carrying on and making a bigger disaster:
> 
> Right, that does make a better error so;
> 
> Tested-by: Dr. David Alan Gilbert <address@hidden>
> 
> Is there anyway we could get that error before the -chardev goes and
> allocates the fd 9?

That is unfortunately determined by the order in which the QEMU command line
args are parsed, and chardevs are procssed before netdevs.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|