On 06/02/2018 02:23 PM, Ján Tomko wrote:
On Sat, Jun 02, 2018 at 12:57:39PM +0200, Michal Privoznik wrote:
> When parsing domain XML the virCapsDomainData lookup is performed
> in order to fill in missing def->os.arch and def->os.machine
> strings. Well, when doing copy of already existing virDomainDef
> we don't want any automagic fill in of defaults (and those two
> strings are going to be provided at this point anyway by first
> parse of the domain XML).
The sentence in parentheses documents an inefficiency -
If we have machine and arch, the DomainDataLookup call is pointless,
unless we want to validate it.
>
> What is even worse is that we do not look up capabilities for
> parsed emulator path rather than some generic capabilities for
s/than//
> parsed arch. Therefore, if emulator points to qemu under
> non-default path (say $HOME/qemu-system-arm) but there's no such
> qemu under the default path (say /usr/bin/qemu-system-arm) the
> capabilities lookup fails and creating the copy is denied.
>
This is downright a bug - we should be using the specified emulator
path to fill the machine type even for new domains. Are you able
to define a new domain without having a default emulator for it?
> Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com>
> ---
> src/conf/domain_conf.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c
> index 86814d5f64..f36a1bfe79 100644
> --- a/src/conf/domain_conf.c
> +++ b/src/conf/domain_conf.c
> @@ -28397,7 +28397,8 @@ virDomainDefCopy(virDomainDefPtr src,
> virDomainDefPtr ret;
> unsigned int format_flags = VIR_DOMAIN_DEF_FORMAT_SECURE;
> unsigned int parse_flags = VIR_DOMAIN_DEF_PARSE_INACTIVE |
> - VIR_DOMAIN_DEF_PARSE_SKIP_VALIDATE;
> + VIR_DOMAIN_DEF_PARSE_SKIP_VALIDATE |
> + VIR_DOMAIN_DEF_PARSE_SKIP_OSTYPE_CHECKS;
For missing QEMU emulator, we postpone all the stuff depending
on qemuCaps until domain startup:
commit 7726d1581f9e433a106f45ed87ec95ece575f817
qemu: Implement postParse callback skipping on config reload
Ideally, we'd remove this flag, move the automagic to PostParse
and make it non-fatal in the same way, but I don't know how much would
that break.
Thing is, virCaps is going to be empty at this point. I mean, there is
no default emulator. We can try to look up virQEMUCaps and probably
derive something from that but that would only work for qemu driver.
Anyway, I'll try to write a follow up patch that does something among
those lines.
Practically, this hunk is an improvement. Hoping you'll address the
newly-defined domain bug as well:
You mean like this?
diff --git i/src/conf/domain_conf.c w/src/conf/domain_conf.c
index f36a1bfe79..9f4ed22701 100644
--- i/src/conf/domain_conf.c
+++ w/src/conf/domain_conf.c
@@ -19045,7 +19045,8 @@ virDomainDefParseXML(xmlDocPtr xml,
def->os.machine = virXPathString("string(./os/type[1]/@machine)",
ctxt);
def->emulator = virXPathString("string(./devices/emulator[1])", ctxt);
- if (!(flags & VIR_DOMAIN_DEF_PARSE_SKIP_OSTYPE_CHECKS)) {
+ if ((!def->os.arch || !def->os.machine) &&
+ !(flags & VIR_DOMAIN_DEF_PARSE_SKIP_OSTYPE_CHECKS)) {
/* If the logic here seems fairly arbitrary, that's because it is :)
* This is duplicating how the code worked before
* CapabilitiesDomainDataLookup was added. We can simplify this,
Michal