CA-413587: Checking feature for old FreeBSD driver

[BengangY]

The FreeBSD driver used by NetScaler supports all power actions. However, older versions of the FreeBSD driver do not explicitly advertise these support. As a result, xapi does not attempt to signal these power actions.
This Free BSD driver issue has been fixed by this commit: 9dd5105
However, there may be many FreeBSD VMs running there which still have this issue. To address this as a workaround, all power actions should be permitted for FreeBSD guests.

Additionally, virtual machines with an explicit data/cant_suspend_reason set aren't allowed to suspend, which would crash Windows and other UEFI VMs.

[last-genius]

Revert the commit "6118d47249aa9bd27a7ed5961b8c0bbdf2891437". This commit will cause issues in the NetScaler scenario due to the following reasons:

1. The VM is FreeBSD, not Windows.

2. The VM includes a PV driver.

3. The VM does not support these features. Therefore, I will revert this commit. We should then revisit the changes and determine how to implement it.

I do not understand which exact assumptions are broken by the commit.

Is it that the driver does not advertise these features in xenstore, but FreeBSD+driver actually supports them, therefore they should be allowed? Does xe vm-migrate force=true still work? That disregards the feature advertisement.

If that's the case, the driver is buggy. control/feature-* paths are documented upstream in xenstore-paths

[last-genius]

For context, the current revert will mean that VMs with explicit data/cant_suspend_reason set will be allowed to suspend, which will keep crashing Windows (and other UEFI) VMs.

This is the reason for 6118d47 ("make cant_suspend_reason critical")

[psafont]

We should then revisit the changes and determine how to implement it.

I've spoken to Roger Pau, the maintainer of the FreeBSD driver. This is an issue with the guest driver used by netscaler, he fixed the issue last year for the upstream drivers: https://cgit.freebsd.org/src/commit/sys/dev/xen/control/control.c?id=9dd5105f22a2276cf099f0e10e318294fcc4f6a7

This means that xapi will need some sort of workaround until netscaler decides to update the driver :/ Don't you have some sort of stick to make them include the patch?

[robhoes]

Is it that the driver does not advertise these features in xenstore, but FreeBSD+driver actually supports them, therefore they should be allowed?

Apparently so. We have a test failure because of this. Indeed this was a Netscaler VM. So we need a different plan to handle this, or the "can't suspend" case.

[last-genius]

Here's an alternative fix that will keep old FreeBSD happy and will not allow suspend for VM's with data/cant_suspend_reason: master...last-genius:xen-api:asv/freebsd-feature

[robhoes]

This means that xapi will need some sort of workaround until netscaler decides to update the driver

If a fix went upstream for FreeBSD just last year, then this will mean that there are probably many FreeBSD VMs out there in the world (Netscaler and others), which we still need to support for the time being.

[lindig]

The logic implementing this would benefit from a longer explanation. It probably makes sense when you wrapped your head around it but reading the code coming just in, it's more magic than logic. Maybe explain how Windows and FreeBSD differ to motivate the logic.

[BengangY]

Update the PR as the following solutions:

1. Add a workaround in xapi to support the old FreeBSD driver and data/cant_suspend_reason.
2. Update the comment in the code to explain this situation.

I'm running the testcase for netscaler and BST+BVT for ring3.

[last-genius]

Looks good to me aside from a comment nitpick