Re: [libvirt] [PATCHv4 9/9] virCaps: get rid of macPrefix field
On 03/29/13 18:39, Eric Blake wrote:
On 03/15/2013 09:26 AM, Peter Krempa wrote:
> Use the virDomainXMLConf structure to hold this data and tweak the code
> to avoid semantic change.
>
> Without configuration the KVM mac prefix is used by default. I chose it
> as it's in the privately administered segment so it should be usable for
> any purposes.
> ---
>
> Notes:
> Version 4:
> - new in series
>
> src/conf/capabilities.c | 14 --------------
> src/conf/capabilities.h | 9 ---------
> src/conf/domain_conf.c | 26 ++++++++++++++++++++++----
> src/conf/domain_conf.h | 3 +++
> src/esx/esx_driver.c | 1 -
> src/libvirt_private.syms | 3 +--
> src/libxl/libxl_conf.c | 2 --
> src/libxl/libxl_driver.c | 6 +++++-
> src/lxc/lxc_conf.c | 3 ---
> src/openvz/openvz_conf.c | 2 --
> src/openvz/openvz_driver.c | 2 +-
> src/parallels/parallels_driver.c | 12 ++++++++----
> src/phyp/phyp_driver.c | 4 ----
> src/qemu/qemu_capabilities.c | 3 ---
> src/qemu/qemu_command.c | 6 +++---
> src/vbox/vbox_tmpl.c | 10 +++++++---
> src/vmware/vmware_conf.c | 2 --
> src/vmx/vmx.c | 1 +
> src/xen/xen_driver.c | 7 ++++++-
> src/xen/xen_hypervisor.c | 2 --
> tests/vmx2xmltest.c | 1 -
> tests/xml2vmxtest.c | 1 -
> 22 files changed, 57 insertions(+), 63 deletions(-)
>
> +++ b/src/conf/domain_conf.c
> @@ -800,6 +800,16 @@ virDomainXMLConfNew(virDomainDefParserConfigPtr config,
> if (xmlns)
> xmlconf->ns = *xmlns;
>
> + /* Technically this forbids to use one of Xerox's MAC address prefixes in
> + * our hypervisor drivers. This shouldn't ever be a problem.
> + *
> + * Use the KVM prefix as default as it's in the privately administered
> + * range */
> + if (memcmp(xmlconf->config.macPrefix,
> + (unsigned char[]) {0x00, 0x00, 0x00}, 3))
> + memcpy(xmlconf->config.macPrefix,
> + (unsigned char[]) {0x54, 0x52, 0x00}, 3);
Regarding a comment later in the review. This line is flawed and the
default and most commonly used prefix should be 0x52, 0x54. ...
I don't see this C99 construct used very often; would it be any more
straightforward to write:
if (xmlconf->conf.macPrefix[0] == 0 &&
xmlconf->conf.macPrefix[1] == 0 &&
xmlconf->conf.macPrefix[2] == 0) {
xmlconf->conf.macPrefix[0] = 0x54;
xmlconf->conf.macPrefix[1] = 0x52;
}
But that's bikeshedding, so you don't necessarily have to agree.
More importantly, why the magic number '3', instead of...
> +++ b/src/conf/domain_conf.h
> @@ -1975,6 +1975,7 @@ struct _virDomainDefParserConfig {
>
> /* data */
> bool hasWideScsiBus;
> + unsigned char macPrefix[VIR_MAC_PREFIX_BUFLEN];
...the symbolic constant VIR_MAC_PREFIX_BUFLEN?
> +++ b/src/esx/esx_driver.c
> @@ -598,7 +598,6 @@ esxCapsInit(esxPrivate *priv)
> return NULL;
> }
>
> - virCapabilitiesSetMacPrefix(caps, (unsigned char[]){ 0x00, 0x0c, 0x29 });
You are removing this prefix from esx://, ...
esx uses the VMX backend to do the parsing and other stuff ... [2]
>
> +virDomainDefParserConfig libxlDomainDefParserConfig = {
> + .macPrefix = { 0x00, 0x16, 0x3e },
> +};
> +
>
> - /* XXX shouldn't 'borrow' KVM's prefix */
> - virCapabilitiesSetMacPrefix(caps, (unsigned char []){ 0x52, 0x54, 0x00 });
This is a change to the default lxc:// prefix; since that is a semantic
change, it should probably be a separate patch.
... thus this doesn't make a semantic change. The funny stuff is that
the test suite doesn't care about this kind of mistake ...
> +++ b/src/vmx/vmx.c
> @@ -521,6 +521,7 @@ VIR_ENUM_IMPL(virVMXControllerModelSCSI,
VIR_DOMAIN_CONTROLLER_MODEL_SCSI_LAST,
>
> virDomainDefParserConfig virVMXDomainDefParserConfig = {
> .hasWideScsiBus = true,
> + .macPrefix = {0x00, 0x0c, 0x29},
> };
...but adding it to vmx://. Are you sure you did this correctly?
[2] ... so adding here is in fact accurate and also the changes in the
tests confirm this.
I think you need to fix the esx vs. vmx issue in v5.
Peter