On Fri, 12 Jul 2024, Jeff Layton wrote:
On Fri, 2024-07-12 at 16:12 +1000, NeilBrown wrote:
My point is that if we are going to change the kernel to accommodate LTP at all, we should accommodate LTP as it is today. If we are going to change LTP to accommodate the kernel, then it should accommodate the kernel as it is today.
The problem is that there is no way for userland tell the difference between the older and newer behavior. That was what I was suggesting we add.
To make sure I wasn't talking through my hat, I had a look at the ltp code.
The test in question simply tests that the count of RPC calls increases.
It can get the count of RPC calls in one of 2 ways : 1/ "lhost" - look directly in /proc/net/rpc/{nfs,nfsd} 2/ "rhost" - ssh to the server and look in that file.
The current test to "fix" this for kernels -ge "6.9" is to force the use of "rhost".
I'm guessing that always using "rhost" for the nfsd stats would always work. But if not, the code could get both the local and remote nfsd stats, and check that at least one of them increases (and neither decrease).
So ltp doesn't need to know which kernel is being used - it can be written to work safely on either.
NeilBrown
To be clear, I hold this opinion loosely. If the consensus is that we need to revert things then so be it. I just don't see the value of doing that in this particular situation. -- Jeff Layton jlayton@kernel.org