On 2/2/21 1:35 PM, Daniel P. Berrangé wrote:
On Tue, Feb 02, 2021 at 05:32:14PM +0100, Ján Tomko wrote:
On a Tuesday in 2021, Daniel P. Berrangé wrote:
On Tue, Feb 02, 2021 at 05:18:51PM +0100, Ján Tomko wrote:
On a Monday in 2021, Daniel Henrique Barboza wrote:
libxlNodeDeviceDetachFlags() and qemuNodeDeviceDetachFlags() are mostly
equal, aside from how the virHostdevmanager pointer is retrieved and
the PCI stub driver used.
Now that the PCI stub driver verification is done early in both functions,
we can use the virDomainDriverNodeDeviceDetachFlags() helper to reduce
code duplication between them. 'driverName' is checked inside the helper
to set the appropriate stub driver.
Signed-off-by: Daniel Henrique Barboza <danielhb413@xxxxxxxxx>
---
src/hypervisor/domain_driver.c | 60 ++++++++++++++++++++++++++++++++++
src/hypervisor/domain_driver.h | 4 +++
src/libvirt_private.syms | 1 +
src/libxl/libxl_driver.c | 54 ++----------------------------
src/qemu/qemu_driver.c | 49 ++-------------------------
5 files changed, 71 insertions(+), 97 deletions(-)
diff --git a/src/hypervisor/domain_driver.c b/src/hypervisor/domain_driver.c
index ea4c3c9466..6ee74d6dff 100644
--- a/src/hypervisor/domain_driver.c
+++ b/src/hypervisor/domain_driver.c
@@ -459,3 +459,63 @@ virDomainDriverNodeDeviceReAttach(virNodeDevicePtr dev,
return virHostdevPCINodeDeviceReAttach(hostdevMgr, pci);
}
+
+int
+virDomainDriverNodeDeviceDetachFlags(virNodeDevicePtr dev,
+ virHostdevManagerPtr hostdevMgr,
+ const char *driverName)
This helper does not even take flags, so the only reason to keep the
'Flags' in its name is to be consistent with the driver method it
implements...
Fair point. I can remove the 'Flags' name of the helper since the flags
are now being considered by the callers.
+{
+ virPCIDevicePtr pci = NULL;
+ virPCIDeviceAddress devAddr;
+ int ret = -1;
+ virNodeDeviceDefPtr def = NULL;
+ g_autofree char *xml = NULL;
+ virConnectPtr nodeconn = NULL;
+ virNodeDevicePtr nodedev = NULL;
+
+ if (!driverName)
+ return -1;
+
+ if (!(nodeconn = virGetConnectNodeDev()))
+ goto cleanup;
+
+ /* 'dev' is associated with virConnectPtr, so for split
+ * daemons, we need to get a copy that is associated with
+ * the virnodedevd daemon. */
+ if (!(nodedev = virNodeDeviceLookupByName(nodeconn,
+ virNodeDeviceGetName(dev))))
+ goto cleanup;
+
+ xml = virNodeDeviceGetXMLDesc(nodedev, 0);
+ if (!xml)
+ goto cleanup;
+
+ def = virNodeDeviceDefParseString(xml, EXISTING_DEVICE, NULL);
+ if (!def)
+ goto cleanup;
+
+ /* ACL check must happen against original 'dev',
+ * not the new 'nodedev' we acquired */
+ if (virNodeDeviceDetachFlagsEnsureACL(dev->conn, def) < 0)
+ goto cleanup;
+
... and the ACL check required.
But moving the ACL checks into src/hypervisor means they are no longer
covered by the check-aclrules check.
I'd rather keep the ACL checks repetitive in each driver than risk the
chance of missing them, but that invalidates most of these refactors.
Any ideas?
Make the check-aclrules also validate this new source file ?
I appreciate any pointers in doing that. I haven't found any 'not horrible' way
(e.g. creating a new condition to add domain_driver.c as a valid driver implementation
file) of doing it in check-aclrules.py.
Thanks,
DHB
Ah, I thought they also verify that each driver method has at least one
ACL call, but we have plenty of methods that are just wrappers for
MethodFlags(..., 0);
The check-aclrules.py script has some logic which looks to see if
the method contains a function call that resolves to another
public API entrypoint, as a special case.
So basically we'll need to process the hypervisor_driver.c file to
extract a list of public API entrpoints with ACL checks, and then
in special case the real drivers if they call one of these shared
functions.
Regards,
Daniel
