On a Wednesday in 2022, Daniel P. Berrangé wrote:
On Wed, Jan 05, 2022 at 04:06:49PM +0100, Ján Tomko wrote:Add a new attribute to control the killpriv feature: <filesystem> ... <binary> <killpriv v2='off'/> </binary> </filesystem>Ewww, ewww, ewww. This is a horrible element & attribute name. Even with the docs I have little clue as to why we would ever need this or what its real effects are. Having now read the libvirt and related QEMU bugzilla, I now see that killpriv v2 is enabled by default in virtiofs because it is faster and more reliable. A limitation in NFS, however, means that some specific syscalls don't work correctly with it, so QEMU folks are saying killpriv v2 should not be used with NFS.
Thank you for the short summary of the bug, I did not realize it was private :( Jano
At this point I ask why doesn't virtiofsd just do the right thing. It can statfs() the root of the export it is configured with and see that it is NFS. I'm not convinced by a need to expose this knob in libvirt, especially with such an unintelligible name & difficult to understand behavioural impact.https://bugzilla.redhat.com/show_bug.cgi?id=1972571 Signed-off-by: Ján Tomko <jtomko@xxxxxxxxxx> --- docs/formatdomain.rst | 4 ++++ docs/schemas/domaincommon.rng | 7 +++++++ src/conf/domain_conf.c | 15 +++++++++++++++ src/conf/domain_conf.h | 1 + .../qemuxml2argvdata/vhost-user-fs-fd-memory.xml | 1 + 5 files changed, 28 insertions(+)
Attachment:
signature.asc
Description: PGP signature
