Re: [Qemu-devel] [PATCH] virtio-serial: Fix segfault on guest boot

[Amit Shah]

On (Fri) 17 Jun 2011 [10:16:44], Luiz Capitulino wrote:
> On Fri, 17 Jun 2011 12:17:36 +0530
> Amit Shah <address@hidden> wrote:
> 
> > On (Thu) 16 Jun 2011 [13:38:49], Luiz Capitulino wrote:
> > > If I start qemu with:
> > > 
> > >  # qemu -hda disks/test.img -enable-kvm -m 1G -snapshot \
> > >         -device virtio-serial \
> > >         -chardev socket,host=localhost,port=1234,server,nowait,id=foo \
> > >         -device virtserialport,chardev=foo,name=org.qemu.guest_agent
> > > 
> > > I get a segfault when booting a Fedora 14 guest. The backtrace says:
> > >
> > >   Program terminated with signal 11, Segmentation fault.
> > >   #0  0x0000000000420850 in handle_control_message (vser=0x3732bd0, 
> > > buf=0x2c173e0, len=8) at 
> > > /home/lcapitulino/src/qmp-unstable/hw/virtio-serial-bus.c:335
> > >   335     info = DO_UPCAST(VirtIOSerialPortInfo, qdev, port->dev.info);
> > 
> > Strange, I've not seen it so far in my testing (neither in the daily
> > test runs of the virtio-serial testsuite).
> > 
> > > I've also bisected this and git points out to commit:
> > > 
> > >   commit a15bb0d6a981de749452a5180fc8084d625671da
> > >   Author: Markus Armbruster <address@hidden>
> > >   Date:   Wed May 25 14:21:13 2011 +0200
> > > 
> > >       virtio-serial: Drop redundant VirtIOSerialPort member info
> > > 
> > > I think what's happening is that the device is not initialized on a
> > > VIRTIO_CONSOLE_DEVICE_READY event. Moving the DO_UPCAST() call to
> > > the other events fixes the problem to me.
> > > 
> > > Signed-off-by: Luiz Capitulino <address@hidden>
> > > ---
> > >  hw/virtio-serial-bus.c |    4 ++--
> > >  1 files changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/hw/virtio-serial-bus.c b/hw/virtio-serial-bus.c
> > > index 9a12104..579f676 100644
> > > --- a/hw/virtio-serial-bus.c
> > > +++ b/hw/virtio-serial-bus.c
> > > @@ -332,8 +332,6 @@ static void handle_control_message(VirtIOSerial 
> > > *vser, void *buf, size_t len)
> > >      if (!port && cpkt.event != VIRTIO_CONSOLE_DEVICE_READY)
> > >          return;
> > >  
> > > -    info = DO_UPCAST(VirtIOSerialPortInfo, qdev, port->dev.info);
> > > -
> > 
> > Ah - this missed the !port check.  It should be possible to do this in
> > a 'if (port)' block instead of replicating in the individual case
> > statements.
> > 
> > Thanks for the debugging and patch; please update with the above and
> > I'll apply it to the virtio-serial tree.
> 
> What about moving the VIRTIO_CONSOLE_DEVICE_READY handling out of the
> switch, like the patch below? This way the function is divided in a way
> that related events are handled together.
> 
> I'll implement your first suggestion if you don't like this...
> 
> diff --git a/hw/virtio-serial-bus.c b/hw/virtio-serial-bus.c
> index 579f676..5f96245 100644
> --- a/hw/virtio-serial-bus.c
> +++ b/hw/virtio-serial-bus.c
> @@ -325,19 +325,12 @@ static void handle_control_message(VirtIOSerial *vser, 
> void *buf, size_t len)
>          return;
>      }
>  
> -    cpkt.event = lduw_p(&gcpkt->event);
>      cpkt.value = lduw_p(&gcpkt->value);
> -
> -    port = find_port_by_id(vser, ldl_p(&gcpkt->id));
> -    if (!port && cpkt.event != VIRTIO_CONSOLE_DEVICE_READY)
> -        return;
> -
> -    switch(cpkt.event) {
> -    case VIRTIO_CONSOLE_DEVICE_READY:
> +    cpkt.event = lduw_p(&gcpkt->event);
> +    if (cpkt.event == VIRTIO_CONSOLE_DEVICE_READY) {
>          if (!cpkt.value) {
> -            error_report("virtio-serial-bus: Guest failure in adding device 
> %s\n",
> -                         vser->bus.qbus.name);
> -            break;
> +            error_report("virtio-serial-bus: Guest failure in adding device 
> %s\n", vser->bus.qbus.name);
> +            return;

The line split should remain -- else it goes beyond 80 chars.

>          }
>          /*
>           * The device is up, we can now tell the device about all the
> @@ -346,8 +339,13 @@ static void handle_control_message(VirtIOSerial *vser, 
> void *buf, size_t len)
>          QTAILQ_FOREACH(port, &vser->ports, next) {
>              send_control_event(port, VIRTIO_CONSOLE_PORT_ADD, 1);
>          }
> -        break;
> +        return;
> +    }

Makes me think of one case (totally unrelated to what you found)where
the guest can fool us: by sending multiple VIRTIO_CONSOLE_DEVICE_READY
messages.

> +    port = find_port_by_id(vser, ldl_p(&gcpkt->id));
> +    assert(port != NULL);

I doubt if assert is the right thing: if the guest sends bad data, we
shouldn't just kill it.  It's easier to ignore such data, and perhaps
just log it.

> +
> +    switch(cpkt.event) {
>      case VIRTIO_CONSOLE_PORT_READY:
>          if (!cpkt.value) {
>              error_report("virtio-serial-bus: Guest failure in adding port %u 
> for device %s\n",

I'm fine with this approach.

                Amit