Re: [PATCH] nfsstat01: Read client stats from netns rhost

Re: [PATCH] nfsstat01: Read client stats from netns rhost

[Petr Vorel]

Hi Martin,

Thanks for fixing the test.

> On newer kernels, network namespaces have separate NFS stats. Detect
> support for per-NS files and read stats from the correct NS.

I'll mention before merging that it was kernel 6.9 and it got backported to up
to 5.15 so far.


Reviewed-by: Petr Vorel <pvorel@suse.cz>
Tested-by: Petr Vorel <pvorel@suse.cz>

> Signed-off-by: Martin Doucha <mdoucha@suse.cz>
> ---

> The /proc/net/rpc/nfs file did not exist in nested network namespaces
> on older kernels. The per-NS stats patchset adds it so we need to check
> for its presence to read the correct stats on kernels where it was
> backported.

> Kernel devs have also asked for a test that'll ensure the patchset doesn't
> get accidentaly reverted. Since this test uses namespaces only when
> the server and client run on the same machine, it'll be better to create
> a separate test for that. I'll send it later.
Thanks!

Kind regards,
Petr

>  testcases/network/nfs/nfsstat01/nfsstat01.sh | 24 +++++++++++++++++---
>  1 file changed, 21 insertions(+), 3 deletions(-)

> diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh
> index c2856eff1..8d7202cf3 100755
> --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh
> +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh
> @@ -3,8 +3,19 @@
>  # Copyright (c) 2016-2018 Oracle and/or its affiliates. All Rights Reserved.
>  # Copyright (c) International Business Machines  Corp., 2001

> +TST_SETUP="nfsstat_setup"
>  TST_TESTFUNC="do_test"
>  TST_NEEDS_CMDS="nfsstat"
> +NS_STAT_RHOST=0
> +
> +nfsstat_setup()
> +{
> +	nfs_setup
> +
> +	if tst_net_use_netns && [ -z "$LTP_NFS_NETNS_USE_LO" ]; then
> +		tst_rhost_run -c "test -r /proc/net/rpc/nfs" && NS_STAT_RHOST=1
> +	fi
> +}

>  get_calls()
>  {
> @@ -15,15 +26,22 @@ get_calls()
>  	local calls opt

>  	[ "$name" = "rpc" ] && opt="r" || opt="n"
> -	! tst_net_use_netns && [ "$nfs_f" != "nfs" ] && type="rhost"
> +	[ "$nfs_f" = "nfsd" ] && opt="-s$opt" || opt="-c$opt"
> +
> +	if tst_net_use_netns; then
> +		# In netns setup, rhost is the client
> +		[ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost"
> +	else
> +		[ "$nfs_f" != "nfs" ] && type="rhost"
> +	fi

>  	if [ "$type" = "lhost" ]; then
>  		calls="$(grep $name /proc/net/rpc/$nfs_f | cut -d' ' -f$field)"
> -		ROD nfsstat -c$opt | grep -q "$calls"
> +		ROD nfsstat $opt | grep -q "$calls"
>  	else
>  		calls=$(tst_rhost_run -c "grep $name /proc/net/rpc/$nfs_f" | \
>  			cut -d' ' -f$field)
> -		tst_rhost_run -s -c "nfsstat -s$opt" | grep -q "$calls"
> +		tst_rhost_run -s -c "nfsstat $opt" | grep -q "$calls"
>  	fi

>  	if ! tst_is_int "$calls"; then

Re: [PATCH] nfsstat01: Read client stats from netns rhost

[Petr Vorel]

Hi all,

> On Wed, 28 Aug 2024, Martin Doucha wrote:
> > On newer kernels, network namespaces have separate NFS stats. Detect
> > support for per-NS files and read stats from the correct NS.

> > Signed-off-by: Martin Doucha <mdoucha@suse.cz>

> Thanks for doing this Martin.  I very much prefer this approach.

> NeilBrown

FYI patch merged. Thanks all for your time.

Kind regards,
Petr


> > ---

> > The /proc/net/rpc/nfs file did not exist in nested network namespaces
> > on older kernels. The per-NS stats patchset adds it so we need to check
> > for its presence to read the correct stats on kernels where it was
> > backported.

> > Kernel devs have also asked for a test that'll ensure the patchset doesn't
> > get accidentaly reverted. Since this test uses namespaces only when
> > the server and client run on the same machine, it'll be better to create
> > a separate test for that. I'll send it later.

> >  testcases/network/nfs/nfsstat01/nfsstat01.sh | 24 +++++++++++++++++---
> >  1 file changed, 21 insertions(+), 3 deletions(-)

> > diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh
> > index c2856eff1..8d7202cf3 100755
> > --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh
> > +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh
> > @@ -3,8 +3,19 @@
> >  # Copyright (c) 2016-2018 Oracle and/or its affiliates. All Rights Reserved.
> >  # Copyright (c) International Business Machines  Corp., 2001

> > +TST_SETUP="nfsstat_setup"
> >  TST_TESTFUNC="do_test"
> >  TST_NEEDS_CMDS="nfsstat"
> > +NS_STAT_RHOST=0
> > +
> > +nfsstat_setup()
> > +{
> > +	nfs_setup
> > +
> > +	if tst_net_use_netns && [ -z "$LTP_NFS_NETNS_USE_LO" ]; then
> > +		tst_rhost_run -c "test -r /proc/net/rpc/nfs" && NS_STAT_RHOST=1
> > +	fi
> > +}

> >  get_calls()
> >  {
> > @@ -15,15 +26,22 @@ get_calls()
> >  	local calls opt

> >  	[ "$name" = "rpc" ] && opt="r" || opt="n"
> > -	! tst_net_use_netns && [ "$nfs_f" != "nfs" ] && type="rhost"
> > +	[ "$nfs_f" = "nfsd" ] && opt="-s$opt" || opt="-c$opt"
> > +
> > +	if tst_net_use_netns; then
> > +		# In netns setup, rhost is the client
> > +		[ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost"
> > +	else
> > +		[ "$nfs_f" != "nfs" ] && type="rhost"
> > +	fi

> >  	if [ "$type" = "lhost" ]; then
> >  		calls="$(grep $name /proc/net/rpc/$nfs_f | cut -d' ' -f$field)"
> > -		ROD nfsstat -c$opt | grep -q "$calls"
> > +		ROD nfsstat $opt | grep -q "$calls"
> >  	else
> >  		calls=$(tst_rhost_run -c "grep $name /proc/net/rpc/$nfs_f" | \
> >  			cut -d' ' -f$field)
> > -		tst_rhost_run -s -c "nfsstat -s$opt" | grep -q "$calls"
> > +		tst_rhost_run -s -c "nfsstat $opt" | grep -q "$calls"
> >  	fi

> >  	if ! tst_is_int "$calls"; then
> > -- 
> > 2.46.0