Re: [libvirt] [PATCH] qemu: Copy console definition from serial

On 2012-03-09 19:02, Daniel P. Berrange wrote: > On Fri, Mar 09, 2012 at 06:58:06PM +0100, Jan Kiszka wrote: >> On 2012-03-09 18:53, Daniel P. Berrange wrote: >>> On Fri, Mar 09, 2012 at 06:48:42PM +0100, Jan Kiszka wrote: >>>> On 2011-11-16 14:14, Michal Privoznik wrote: >>>>> Now, when we support multiple consoles per domain, >>>>> the vm->def->console[0] can still remain an alias >>>>> for vm->def->serial[0]; However, we need to copy >>>>> it's source definition as well otherwise we'll regress >>>>> on virDomainOpenConsole. >>>> >>>> ... >>>> >>>>> diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c >>>>> index 2882ef8..e0b1824 100644 >>>>> --- a/src/qemu/qemu_process.c >>>>> +++ b/src/qemu/qemu_process.c >>>>> @@ -1163,11 +1163,22 @@ qemuProcessFindCharDevicePTYs(virDomainObjPtr vm, >>>>> >>>>> for (i = 0 ; i < vm->def->nconsoles ; i++) { >>>>> virDomainChrDefPtr chr = vm->def->consoles[i]; >>>>> - if (chr->source.type == VIR_DOMAIN_CHR_TYPE_PTY && >>>>> - chr->targetType == VIR_DOMAIN_CHR_CONSOLE_TARGET_TYPE_VIRTIO) { >>>>> - if ((ret = qemuProcessExtractTTYPath(output, &offset, >>>>> - &chr->source.data.file.path)) != 0) >>>>> + /* For historical reasons, console[0] can be just an alias >>>>> + * for serial[0]; That's why we need to update it as well */ >>>>> + if (i == 0 && vm->def->nserials && >>>>> + chr->deviceType == VIR_DOMAIN_CHR_DEVICE_TYPE_CONSOLE && >>>>> + chr->targetType == VIR_DOMAIN_CHR_CONSOLE_TARGET_TYPE_SERIAL) { >>>>> + if ((ret = virDomainChrSourceDefCopy(&chr->source, >>>>> + &((vm->def->serials[0])->source))) != 0) >>>>> return ret; >>>>> + chr->source.type = VIR_DOMAIN_CHR_TYPE_PTY; >>>> >>>> This unconditional setting of TYPE_PTY breaks serial on stdio (we use >>>> this to easily fold guest into host logs). Can you explain why the >>>> copied source.type of serial[0] is not always correct? Or are we already >>>> in the wrong branch for a >>>> >>>> <serial type='stdio'></serial> >>>> >>>> configuration? >>> >>> Yeah I think this is a bug. The first serial element should match the >>> first console exactly, with targetType==serial. We shouldn't be >>> forcing it to type=pty >> >> So, if vm->def->serials[0])->source.type != VIR_DOMAIN_CHR_TYPE_PTY, we >> should skip this fixup branch? > > No, we should do the fix for all types. serial[0] and console[0] should > be identical, unless the console[0] targetType != TARGET_TYPE_SERIAL Well, libvirt is alien to me, and at this point I'm lost in fixups. :) Do you mean simply dropping the source.type = TYPE_PTY line?