On Tue, Apr 30, 2024 at 11:08:13AM +0200, Martin Pieuchot wrote: > > 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? >
We are close enough, but I want to unlock raw and UNIX sockets before. However, without splicing ability, the diff for UDP sockets is pretty small and I tries it with regress/sys/ffs/nfs test. I was surprised, but I didn't see witness reports about so_snd.sb_lock. I want to push UNIX sockets before because it removes re-locking from uipc_rcvd() which is always called in reception path of connection oriented sockets. Also it allows me to clean some temporary stuff.
