On Tue, Apr 30, 2024 at 11:08:13AM +0200, Martin Pieuchot wrote: > > With the patch, the nfsnode-vmmaplk reversal looks like this: > > So the issue here is due to NFS entering the network stack after the > VFS. Alexander, Vitaly are we far from a NET_LOCK()-free sosend()? > Is something we should consider?
The way I am panning is to require shared netlock instead of exclusive netlock in sosend(). I want to keep exclusive netlock for network configuration. Sending and receiving NFS UDP traffic with shared netlock should be possible. Switching TCP to shared netlock is much more work. I don't know if witness knows that two shared rwlocks do not interfere. > On the other side, would that make sense to have a NET_LOCK()-free > sysctl path? I am working to remove netlock for integer type sysctl. That is the majority, others need case by case analysis. bluhm
