On Tue, Mar 06, 2018 at 05:58:47PM +0000, Daniel P. Berrangé wrote:
This is a workaround for the bizarre behaviour whereby the kernel throws away all fcntl() locks on execve()... if any threads are running. We simply make virtlockd & virlogd single threaded instead.
Daniel P. Berrangé (5): rpc: push ref acquisition into RPC dispatch function rpc: simplify calling convention of virNetServerClientDispatchFunc rpc: invoke the message dispatch callback with client unlocked rpc: avoid crashing in pre-exec if no workers are present rpc: switch virtlockd and virtlogd to use single-threaded dispatch
src/locking/lock_daemon.c | 4 +-- src/logging/log_daemon.c | 4 +-- src/rpc/virnetserver.c | 29 ++++++++++------ src/rpc/virnetserverclient.c | 81 ++++++++++++++++++++++++++++++-------------- src/rpc/virnetserverclient.h | 9 +++-- 5 files changed, 83 insertions(+), 44 deletions(-)
Forgot to say that we might want to still take a variety on Jim's pathc to re-aquire locks after execve() so that existing deployments recover their state. This patch series only fixes things when virtlockd is cold started, and RPM upgrades will warm-restart with re-exec Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|