Re: [libvirt] [PATCH] qemu: Fix serial stub console allocation
On Mon, May 29, 2017 at 02:33:04PM +0200, Peter Krempa wrote:
On Mon, May 29, 2017 at 13:11:09 +0200, Erik Skultety wrote:
> When adding the aliased serial stub console, the structure wasn't
> properly allocated (VIR_ALLOC instead of virDomainChrDefNew) which then
> resulted in SIGSEGV in virDomainChrSourceIsEqual during a serial device
> coldplug.
>
> https://bugzilla.redhat.com/show_bug.cgi?id=1434278
>
> Signed-off-by: Erik Skultety <eskultet(a)redhat.com>
> ---
> src/qemu/qemu_hotplug.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/src/qemu/qemu_hotplug.c b/src/qemu/qemu_hotplug.c
> index 4a7d99725..34ddb95f8 100644
> --- a/src/qemu/qemu_hotplug.c
> +++ b/src/qemu/qemu_hotplug.c
> @@ -1810,7 +1810,7 @@ qemuDomainChrPreInsert(virDomainDefPtr vmdef,
> if (!vmdef->consoles && VIR_ALLOC(vmdef->consoles) < 0)
> return -1;
>
> - if (VIR_ALLOC(vmdef->consoles[0]) < 0) {
> + if (!(vmdef->consoles[0] = virDomainChrDefNew(NULL))) {
With this the code will not call qemuDomainChrSourcePrivateNew in the
qemu driver, which is called everywhere. Are you sure this is okay? If
so please add a comment that xmlopt is not necessary here.
I hopefully checked all places where we're handling char devices and when
dealing with consoles, we always special case the backcompat serial console 0
in a way where instead of the console's own def, we rather pass the serial[0]'s
data. I'll add a comment as suggested.
Erik
Otherwise it should be simple to pass xmlopt here.