[LTP] [PATCH 1/1] nfsstat01: Update client RPC calls for kernel 6.9

[Chuck Lever III]

> On Jul 2, 2024, at 6:55 PM, Calum Mackay <calum.mackay@oracle.com> wrote:
> 
> To clarify…
> 
> On 02/07/2024 5:54 pm, Calum Mackay wrote:
>> hi Petr,
>> I noticed your LTP patch [1][2] which adjusts the nfsstat01 test on v6.9 kernels, to account for Josef's changes [3], which restrict the NFS/RPC stats per-namespace.
>> I see that Josef's changes were backported, as far back as longterm v5.4,
> 
> Sorry, that's not quite accurate.
> 
> Josef's NFS client changes were all backported from v6.9, as far as longterm v5.4.y:
> 
> 2057a48d0dd0 sunrpc: add a struct rpc_stats arg to rpc_create_args
> d47151b79e32 nfs: expose /proc/net/sunrpc/nfs in net namespaces
> 1548036ef120 nfs: make the rpc_stat per net namespace
> 
> 
> Of Josef's NFS server changes, four were backported from v6.9 to v6.8:
> 
> 418b9687dece sunrpc: use the struct net as the svc proc private
> d98416cc2154 nfsd: rename NFSD_NET_* to NFSD_STATS_*
> 93483ac5fec6 nfsd: expose /proc/net/sunrpc/nfsd in net namespaces
> 4b14885411f7 nfsd: make all of the nfsd stats per-network namespace
> 
> and the others remained only in v6.9:
> 
> ab42f4d9a26f sunrpc: don't change ->sv_stats if it doesn't exist
> a2214ed588fb nfsd: stop setting ->pg_stats for unused stats
> f09432386766 sunrpc: pass in the sv_stats struct through svc_create_pooled
> 3f6ef182f144 sunrpc: remove ->pg_stats from svc_program
> e41ee44cc6a4 nfsd: remove nfsd_stats, make th_cnt a global counter
> 16fb9808ab2c nfsd: make svc_stat per-network namespace instead of global
> 
> 
> 
> I'm wondering if this difference between NFS client, and NFS server, stat behaviour, across kernel versions, may perhaps cause some user confusion?

As a refresher for the stable folken, Josef's changes make
nfsstats silo'd, so they no longer show counts from the whole
system, but only for NFS operations relating to the local net
namespace. That is a surprising change for some users, tools,
and testing.

I'm not clear on whether there are any rules/guidelines around
LTS backports causing behavior changes that user tools, like
nfsstat, might be impacted by.

The client-side nfsstat changes are fully backported to all
TS kernels. But should they have been?

The server-side nfsstat changes appear in only v6.9. Should
they be backported to the other LTS kernels, or not?


>> so your check for kernel version "6.9" in the test may need to be adjusted, if LTP is intended to be run on stable kernels?
>> best wishes,
>> calum.
>> [1] https://lore.kernel.org/ltp/20240620111129.594449-1-pvorel@suse.cz/
>> [2] https://patchwork.ozlabs.org/project/ltp/ patch/20240620111129.594449-1-pvorel@suse.cz/
>> [3] https://lore.kernel.org/linux-nfs/ cover.1708026931.git.josef@toxicpanda.com/

--
Chuck Lever