On Fri, Jul 03, 2015 at 02:01:46PM +0200, Erik Skultety wrote:
On 01/07/15 12:05, Pavel Hrdina wrote:
The copy of persistent definition is already done in virDomainLiveConfigHelperMethod few lines above.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com> --- src/qemu/qemu_driver.c | 6 ------ 1 file changed, 6 deletions(-)
diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index 4cfae03..ca93a1a 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -10336,12 +10336,6 @@ qemuDomainSetSchedulerParametersFlags(virDomainPtr dom, &vmdef) < 0) goto endjob;
- if (flags & VIR_DOMAIN_AFFECT_CONFIG) { - /* Make a copy for updated domain. */ - if (!(vmdef = virDomainObjCopyPersistentDef(vm, caps, driver->xmlopt))) - goto endjob; - } - if (flags & VIR_DOMAIN_AFFECT_LIVE) { if (!virCgroupHasController(priv->cgroup, VIR_CGROUP_CONTROLLER_CPU)) { virReportError(VIR_ERR_OPERATION_INVALID,
NACK. It's not true, that a copy is made in virDomainLiveConfigHelperMethod, only a reference to domain definition is returned. The problem is that we free vmdef at the end of the API which might not result in a desired behavior, let's consider an inactive persistent domain and you try to set one of scheduler params, instead of modifying an freeing a copy, you manipulate and free the original instance, what happens then? Yep, daemon crashes in the next API (OK, not every time, it requires a bit of luck but after a couple of minutes I managed to do that as well).
True, I've missed that in the helper we don't create a copy, just take a reference. This makes perfect sense and in fact it's an exception, where we use a copy to update the domain and in case of failure just drop the copy and leave the domain definition intact. I thought, that there is no API that have this behavior. I'll push only the first patch, thanks for review. Pavel
Erik
-- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list