On 06/21/2011 03:25 PM, Eric Blake wrote:
On 06/21/2011 12:21 PM, Stefan Berger wrote:
> On 06/21/2011 01:16 PM, Eric Blake wrote:
>> On 06/21/2011 06:01 AM, Gerhard Stenzel wrote:
>>> + rc = ifaceGetMacaddr(linkdev, oldmac);
>>> +
>>> + if (rc) {
>>> + virReportSystemError(rc,
>> Sorry for not catching this sooner. This should probably be
>> virReportSystemError(errno,...), assuming that ifaceGetMacaddr
>> guarantees a sane errno setting (at any rate, you used errno for
>> ifaceSetMacaddr later on).
>>
> It should be ok the way it is since ifaceGetMacAddr() returns either
> EINVAL or the content of errno.
Oh, I see. But then we have the converse problem:
+ rc = ifaceSetMacaddr(linkdev, macaddress);
+ if (rc) {
+ virReportSystemError(errno,
+ _("Setting MAC address on '%s' to "
+ "'%02x:%02x:%02x:%02x:%02x:%02x'
failed."),
But ifaceSetMacaddr returns EINVAL without setting errno on at least one
error path, which means this can print a bogus errno value.
So we should either fix the iface* functions to follow convention of -1
return and valid errno used in other libvirt interfaces (rather than
positive return), as well as fix its clients; or we should fix this
particular ifaceSetMacaddr call to fit in with the different convention.
I vote for consistency with the rest of the project, although I think
that should be a separate patch (done as a prerequisite to this one)