[PATCH 0/9] Another round of qemu:///embed fixes
Daniel Henrique Barboza
danielhb413 at gmail.com
Wed Mar 25 17:00:15 UTC 2020
On 3/25/20 7:18 AM, Michal Privoznik wrote:
> After my first round of patches got merged [1], some new problems
> arose. They are not as serious (except maybe for 7/9) but they are still
> problematic for sure. Mostly in scenarios where a mgmt application uses
> the embedded driver and has multiple forks/instances sharing essentially
> the same config (e.g. it writes the same qemu.conf into different
> roots). Thing is, we may generate conflicting paths across these
> instances.
>
> What I've done here is, I looked at what paths can be set from qemu.conf
> and on their usage. Then I've grepped for virDomainDefGetShortName() and
> looked at its usages, because that has proven to be common denominator
> in patches 7-9.
>
> Some usages of the function are safe though. When the usage involves a
> path that is derived from the root and can't be overridden in qemu.conf
> it is safe. For instance, qemuDBusGetAddress() calls
> virDomainDefGetShortName() to construct a path to a UNIX socket. But the
> path has cfg->dbusStateDir prefix which is derived from cfg->stateDir
> which in turn is derived from the root. At the same time, neither of
> these paths can be overridden in qemu.conf. Therefore, no conflicts can
> occur.
>
> Michal Prívozník (9):
> tests: Fix virQEMUDriverConfigNew() calling with respect to @root
> conf: Move virDomainGenerateMachineName to hypervisor/
> virDomainDriverGenerateMachineName: Factor out embed path hashing
> qemu: Introduce virQEMUDriverGetEmbedRoot
> qemuDomainGetMachineName: Access embeddedRoot from driver rather than
> cfg
> Revert "qemu_conf: Track embed root dir"
> qemu: Make hugepages path generation embed driver aware
> qemu: Make memory path generation embed driver aware
> qemu: Make auto dump path generation embed driver aware
>
LGTM
Reviewed-by: Daniel Henrique Barboza <danielhb413 at gmail.com>
More information about the libvir-list
mailing list