| Message ID | 20240830141453.28379-2-mdoucha@suse.cz |
|---|---|
| State | Accepted |
| Headers | show |
| Series | [1/2] Add test for per-NS NFS client statistics | expand |
On Fri, Aug 30, 2024 at 04:13:40PM +0200, Martin Doucha wrote: > When the NFS server and client run on the same host in different net > namespaces, check that RPC calls from the client namespace don't > change RPC statistics in the root namespace. > > Signed-off-by: Martin Doucha <mdoucha@suse.cz> > --- > > I've initially tried to test both NFS and RPC client stats but it appears > that NFS client stats are still shared across all namespaces. Only RPC > client stats are separate for each net namespace. The kernel patchset[1] > which introduced per-NS stats confirms that only RPC stats have been changed. I believe that is correct, Josef changed only RPC counters. Which counters did you expect also would be containerized, exactly? Perhaps this issue should be raised on linux-nfs@vger, it could be considered to be another information leak. > If NFS client stats should be separate for each namespace as well, let > me know and I'll return the second set of NS checks in patch v2. > > Tested on kernel v5.14 with Neil's backports. > > [1] https://lore.kernel.org/linux-nfs/cover.1708026931.git.josef@toxicpanda.com/ > > testcases/network/nfs/nfsstat01/nfsstat01.sh | 18 ++++++++++++++++-- > 1 file changed, 16 insertions(+), 2 deletions(-) > > diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh > index 8d7202cf3..3379c4d46 100755 > --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh > +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh > @@ -22,6 +22,7 @@ get_calls() > local name=$1 > local field=$2 > local nfs_f=$3 > + local netns=${4:-rhost} > local type="lhost" > local calls opt > > @@ -30,7 +31,8 @@ get_calls() > > if tst_net_use_netns; then > # In netns setup, rhost is the client > - [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost" > + [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && \ > + type="$netns" > else > [ "$nfs_f" != "nfs" ] && type="rhost" > fi > @@ -64,13 +66,14 @@ get_calls() > do_test() > { > local client_calls server_calls new_server_calls new_client_calls > - local client_field server_field > + local client_field server_field root_calls new_root_calls > local client_v=$VERSION server_v=$VERSION > > tst_res TINFO "checking RPC calls for server/client" > > server_calls="$(get_calls rpc 2 nfsd)" > client_calls="$(get_calls rpc 2 nfs)" > + root_calls="$(get_calls rpc 2 nfs lhost)" > > tst_res TINFO "calls $server_calls/$client_calls" > > @@ -79,6 +82,7 @@ do_test() > > new_server_calls="$(get_calls rpc 2 nfsd)" > new_client_calls="$(get_calls rpc 2 nfs)" > + new_root_calls="$(get_calls rpc 2 nfs lhost)" > tst_res TINFO "new calls $new_server_calls/$new_client_calls" > > if [ "$new_server_calls" -le "$server_calls" ]; then > @@ -93,6 +97,16 @@ do_test() > tst_res TPASS "client RPC calls increased" > fi > > + if [ $NS_STAT_RHOST -ne 0 ]; then > + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" > + > + if [ $root_calls -ne $new_root_calls ]; then > + tst_res TFAIL "RPC stats leaked between net namespaces" > + else > + tst_res TPASS "RPC stats stay within net namespaces" > + fi > + fi > + > tst_res TINFO "checking NFS calls for server/client" > case $VERSION in > 2) client_field=13 server_field=13 > -- > 2.46.0 >
Hi all, > On Fri, Aug 30, 2024 at 04:13:40PM +0200, Martin Doucha wrote: > > When the NFS server and client run on the same host in different net > > namespaces, check that RPC calls from the client namespace don't > > change RPC statistics in the root namespace. > > Signed-off-by: Martin Doucha <mdoucha@suse.cz> > > --- > > I've initially tried to test both NFS and RPC client stats but it appears > > that NFS client stats are still shared across all namespaces. Only RPC > > client stats are separate for each net namespace. The kernel patchset[1] > > which introduced per-NS stats confirms that only RPC stats have been changed. Yes, only RPC client stats needed to be fixed in LTP test. Kind regards, Petr > I believe that is correct, Josef changed only RPC counters. Which > counters did you expect also would be containerized, exactly? > Perhaps this issue should be raised on linux-nfs@vger, it could be > considered to be another information leak. > > If NFS client stats should be separate for each namespace as well, let > > me know and I'll return the second set of NS checks in patch v2. > > Tested on kernel v5.14 with Neil's backports. > > [1] https://lore.kernel.org/linux-nfs/cover.1708026931.git.josef@toxicpanda.com/ > > testcases/network/nfs/nfsstat01/nfsstat01.sh | 18 ++++++++++++++++-- > > 1 file changed, 16 insertions(+), 2 deletions(-) > > diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh > > index 8d7202cf3..3379c4d46 100755 > > --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh > > +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh > > @@ -22,6 +22,7 @@ get_calls() > > local name=$1 > > local field=$2 > > local nfs_f=$3 > > + local netns=${4:-rhost} > > local type="lhost" > > local calls opt > > @@ -30,7 +31,8 @@ get_calls() > > if tst_net_use_netns; then > > # In netns setup, rhost is the client > > - [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost" > > + [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && \ > > + type="$netns" > > else > > [ "$nfs_f" != "nfs" ] && type="rhost" > > fi > > @@ -64,13 +66,14 @@ get_calls() > > do_test() > > { > > local client_calls server_calls new_server_calls new_client_calls > > - local client_field server_field > > + local client_field server_field root_calls new_root_calls > > local client_v=$VERSION server_v=$VERSION > > tst_res TINFO "checking RPC calls for server/client" > > server_calls="$(get_calls rpc 2 nfsd)" > > client_calls="$(get_calls rpc 2 nfs)" > > + root_calls="$(get_calls rpc 2 nfs lhost)" > > tst_res TINFO "calls $server_calls/$client_calls" > > @@ -79,6 +82,7 @@ do_test() > > new_server_calls="$(get_calls rpc 2 nfsd)" > > new_client_calls="$(get_calls rpc 2 nfs)" > > + new_root_calls="$(get_calls rpc 2 nfs lhost)" > > tst_res TINFO "new calls $new_server_calls/$new_client_calls" > > if [ "$new_server_calls" -le "$server_calls" ]; then > > @@ -93,6 +97,16 @@ do_test() > > tst_res TPASS "client RPC calls increased" > > fi > > + if [ $NS_STAT_RHOST -ne 0 ]; then > > + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" > > + > > + if [ $root_calls -ne $new_root_calls ]; then > > + tst_res TFAIL "RPC stats leaked between net namespaces" > > + else > > + tst_res TPASS "RPC stats stay within net namespaces" > > + fi > > + fi > > + > > tst_res TINFO "checking NFS calls for server/client" > > case $VERSION in > > 2) client_field=13 server_field=13 > > -- > > 2.46.0
> When the NFS server and client run on the same host in different net > namespaces, check that RPC calls from the client namespace don't > change RPC statistics in the root namespace. > Signed-off-by: Martin Doucha <mdoucha@suse.cz> > --- > I've initially tried to test both NFS and RPC client stats but it appears > that NFS client stats are still shared across all namespaces. Only RPC > client stats are separate for each net namespace. The kernel patchset[1] > which introduced per-NS stats confirms that only RPC stats have been changed. > If NFS client stats should be separate for each namespace as well, let > me know and I'll return the second set of NS checks in patch v2. > Tested on kernel v5.14 with Neil's backports. > [1] https://lore.kernel.org/linux-nfs/cover.1708026931.git.josef@toxicpanda.com/ > testcases/network/nfs/nfsstat01/nfsstat01.sh | 18 ++++++++++++++++-- > 1 file changed, 16 insertions(+), 2 deletions(-) > diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh > index 8d7202cf3..3379c4d46 100755 > --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh > +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh > @@ -22,6 +22,7 @@ get_calls() > local name=$1 > local field=$2 > local nfs_f=$3 > + local netns=${4:-rhost} > local type="lhost" > local calls opt > @@ -30,7 +31,8 @@ get_calls() > if tst_net_use_netns; then > # In netns setup, rhost is the client > - [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost" > + [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && \ > + type="$netns" > else > [ "$nfs_f" != "nfs" ] && type="rhost" > fi > @@ -64,13 +66,14 @@ get_calls() > do_test() > { > local client_calls server_calls new_server_calls new_client_calls > - local client_field server_field > + local client_field server_field root_calls new_root_calls > local client_v=$VERSION server_v=$VERSION > tst_res TINFO "checking RPC calls for server/client" > server_calls="$(get_calls rpc 2 nfsd)" > client_calls="$(get_calls rpc 2 nfs)" > + root_calls="$(get_calls rpc 2 nfs lhost)" > tst_res TINFO "calls $server_calls/$client_calls" > @@ -79,6 +82,7 @@ do_test() > new_server_calls="$(get_calls rpc 2 nfsd)" > new_client_calls="$(get_calls rpc 2 nfs)" > + new_root_calls="$(get_calls rpc 2 nfs lhost)" > tst_res TINFO "new calls $new_server_calls/$new_client_calls" > if [ "$new_server_calls" -le "$server_calls" ]; then > @@ -93,6 +97,16 @@ do_test() > tst_res TPASS "client RPC calls increased" > fi > + if [ $NS_STAT_RHOST -ne 0 ]; then > + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" > + > + if [ $root_calls -ne $new_root_calls ]; then > + tst_res TFAIL "RPC stats leaked between net namespaces" > + else > + tst_res TPASS "RPC stats stay within net namespaces" > + fi Maybe also add TCONF message? (can be added before merge) else tst_res TCONF "Not testing leak between root NS and net NS due old kernel" Reviewed-by: Petr Vorel <pvorel@suse.cz> Tested-by: Petr Vorel <pvorel@suse.cz> Thanks for this test! Kind regards, Petr > + fi > + > tst_res TINFO "checking NFS calls for server/client" > case $VERSION in > 2) client_field=13 server_field=13
> Hi all, > > On Fri, Aug 30, 2024 at 04:13:40PM +0200, Martin Doucha wrote: > > > When the NFS server and client run on the same host in different net > > > namespaces, check that RPC calls from the client namespace don't > > > change RPC statistics in the root namespace. > > > Signed-off-by: Martin Doucha <mdoucha@suse.cz> > > > --- > > > I've initially tried to test both NFS and RPC client stats but it appears > > > that NFS client stats are still shared across all namespaces. Only RPC > > > client stats are separate for each net namespace. The kernel patchset[1] > > > which introduced per-NS stats confirms that only RPC stats have been changed. > Yes, only RPC client stats needed to be fixed in LTP test. OTOH there is also nfsd stats namespaced [2] as a second part of whole "Make nfs and nfsd stats visible in network ns" patchset. I suppose we would need to reverse client and server to detect this (IMHO worth of doing it). Kind regards, Petr [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/log/?qt=author&q=4b14885411f74b2b0ce0eb2b39d0fffe54e5ca0d [3] https://lore.kernel.org/linux-nfs/cover.1706212207.git.josef@toxicpanda.com/ > Kind regards, > Petr > > I believe that is correct, Josef changed only RPC counters. Which > > counters did you expect also would be containerized, exactly? > > Perhaps this issue should be raised on linux-nfs@vger, it could be > > considered to be another information leak. > > > If NFS client stats should be separate for each namespace as well, let > > > me know and I'll return the second set of NS checks in patch v2. > > > Tested on kernel v5.14 with Neil's backports. > > > [1] https://lore.kernel.org/linux-nfs/cover.1708026931.git.josef@toxicpanda.com/ > > > testcases/network/nfs/nfsstat01/nfsstat01.sh | 18 ++++++++++++++++-- > > > 1 file changed, 16 insertions(+), 2 deletions(-) > > > diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh > > > index 8d7202cf3..3379c4d46 100755 > > > --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh > > > +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh > > > @@ -22,6 +22,7 @@ get_calls() > > > local name=$1 > > > local field=$2 > > > local nfs_f=$3 > > > + local netns=${4:-rhost} > > > local type="lhost" > > > local calls opt > > > @@ -30,7 +31,8 @@ get_calls() > > > if tst_net_use_netns; then > > > # In netns setup, rhost is the client > > > - [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost" > > > + [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && \ > > > + type="$netns" > > > else > > > [ "$nfs_f" != "nfs" ] && type="rhost" > > > fi > > > @@ -64,13 +66,14 @@ get_calls() > > > do_test() > > > { > > > local client_calls server_calls new_server_calls new_client_calls > > > - local client_field server_field > > > + local client_field server_field root_calls new_root_calls > > > local client_v=$VERSION server_v=$VERSION > > > tst_res TINFO "checking RPC calls for server/client" > > > server_calls="$(get_calls rpc 2 nfsd)" > > > client_calls="$(get_calls rpc 2 nfs)" > > > + root_calls="$(get_calls rpc 2 nfs lhost)" > > > tst_res TINFO "calls $server_calls/$client_calls" > > > @@ -79,6 +82,7 @@ do_test() > > > new_server_calls="$(get_calls rpc 2 nfsd)" > > > new_client_calls="$(get_calls rpc 2 nfs)" > > > + new_root_calls="$(get_calls rpc 2 nfs lhost)" > > > tst_res TINFO "new calls $new_server_calls/$new_client_calls" > > > if [ "$new_server_calls" -le "$server_calls" ]; then > > > @@ -93,6 +97,16 @@ do_test() > > > tst_res TPASS "client RPC calls increased" > > > fi > > > + if [ $NS_STAT_RHOST -ne 0 ]; then > > > + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" > > > + > > > + if [ $root_calls -ne $new_root_calls ]; then > > > + tst_res TFAIL "RPC stats leaked between net namespaces" > > > + else > > > + tst_res TPASS "RPC stats stay within net namespaces" > > > + fi > > > + fi > > > + > > > tst_res TINFO "checking NFS calls for server/client" > > > case $VERSION in > > > 2) client_field=13 server_field=13 > > > -- > > > 2.46.0
On 30. 08. 24 20:10, Chuck Lever wrote: > On Fri, Aug 30, 2024 at 04:13:40PM +0200, Martin Doucha wrote: >> When the NFS server and client run on the same host in different net >> namespaces, check that RPC calls from the client namespace don't >> change RPC statistics in the root namespace. >> >> Signed-off-by: Martin Doucha <mdoucha@suse.cz> >> --- >> >> I've initially tried to test both NFS and RPC client stats but it appears >> that NFS client stats are still shared across all namespaces. Only RPC >> client stats are separate for each net namespace. The kernel patchset[1] >> which introduced per-NS stats confirms that only RPC stats have been changed. > > I believe that is correct, Josef changed only RPC counters. Which > counters did you expect also would be containerized, exactly? > Perhaps this issue should be raised on linux-nfs@vger, it could be > considered to be another information leak. I tried to test the NFS client call counters, fields 13, 15 or 24 (depending on NFS version) in the "procX" line of /proc/net/rpc/nfs. These are the counters that the test already checks after RPC. Although when I think about it some more, I'm not sure whether the NFS/RPC client statistics should be attached to network namespaces in the first place. AFAICT, processes from any network namespace can trigger client calls for both RPC and NFS as long as they can access the NFS mountpoint. Perhaps a mount namespace would be the more logical domain for counting per-NS statistics instead?
Hi, On 30. 08. 24 22:15, Petr Vorel wrote: >> @@ -93,6 +97,16 @@ do_test() >> tst_res TPASS "client RPC calls increased" >> fi > >> + if [ $NS_STAT_RHOST -ne 0 ]; then >> + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" >> + >> + if [ $root_calls -ne $new_root_calls ]; then >> + tst_res TFAIL "RPC stats leaked between net namespaces" >> + else >> + tst_res TPASS "RPC stats stay within net namespaces" >> + fi > > Maybe also add TCONF message? (can be added before merge) > > else > tst_res TCONF "Not testing leak between root NS and net NS due old kernel" I think the TCONF message doesn't make sense here. There are several cases where the new check will be skipped: 1) the NFS server runs on another machine ($RHOST is not empty) 2) the test is configure to ignore namespaces ($LTP_NFS_NETNS_USE_LO is not empty) 3) /proc/net/rpc/nfs doesn't exist in nested net namespaces You want to print the TCONF message only in case #3. Let's keep the condition above simple.
> On Sep 2, 2024, at 7:49 AM, Martin Doucha <mdoucha@suse.cz> wrote: > > On 30. 08. 24 20:10, Chuck Lever wrote: >> On Fri, Aug 30, 2024 at 04:13:40PM +0200, Martin Doucha wrote: >>> When the NFS server and client run on the same host in different net >>> namespaces, check that RPC calls from the client namespace don't >>> change RPC statistics in the root namespace. >>> >>> Signed-off-by: Martin Doucha <mdoucha@suse.cz> >>> --- >>> >>> I've initially tried to test both NFS and RPC client stats but it appears >>> that NFS client stats are still shared across all namespaces. Only RPC >>> client stats are separate for each net namespace. The kernel patchset[1] >>> which introduced per-NS stats confirms that only RPC stats have been changed. >> I believe that is correct, Josef changed only RPC counters. Which >> counters did you expect also would be containerized, exactly? >> Perhaps this issue should be raised on linux-nfs@vger, it could be >> considered to be another information leak. > > I tried to test the NFS client call counters, fields 13, 15 or 24 (depending on NFS version) in the "procX" line of /proc/net/rpc/nfs. These are the counters that the test already checks after RPC. > > Although when I think about it some more, I'm not sure whether the NFS/RPC client statistics should be attached to network namespaces in the first place. AFAICT, processes from any network namespace can trigger client calls for both RPC and NFS as long as they can access the NFS mountpoint. Perhaps a mount namespace would be the more logical domain for counting per-NS statistics instead? Disclaimer: I'm not one of the NFS client maintainers, but only a very long time contributor to the Linux NFS implementation, so I can offer only a somewhat-educated opinion on this topic. IIRC in a container, the RPC client is bound to the network namespace. The statistics in /proc/self/mountstats are accrued to an individual mount. I think those are associated with the mount namespace. I could very well be wrong about that. This is another topic that would be appropriate to bring to linux-nfs@ . -- Chuck Lever
Hi Martin, > Hi, > On 30. 08. 24 22:15, Petr Vorel wrote: > > > @@ -93,6 +97,16 @@ do_test() > > > tst_res TPASS "client RPC calls increased" > > > fi > > > + if [ $NS_STAT_RHOST -ne 0 ]; then > > > + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" > > > + > > > + if [ $root_calls -ne $new_root_calls ]; then > > > + tst_res TFAIL "RPC stats leaked between net namespaces" > > > + else > > > + tst_res TPASS "RPC stats stay within net namespaces" > > > + fi > > Maybe also add TCONF message? (can be added before merge) > > else > > tst_res TCONF "Not testing leak between root NS and net NS due old kernel" > I think the TCONF message doesn't make sense here. There are several cases > where the new check will be skipped: > 1) the NFS server runs on another machine ($RHOST is not empty) > 2) the test is configure to ignore namespaces ($LTP_NFS_NETNS_USE_LO is not > empty) > 3) /proc/net/rpc/nfs doesn't exist in nested net namespaces > You want to print the TCONF message only in case #3. Let's keep the > condition above simple. Fair enough. I merged just with adding kernel commit and patchset link in the first commit. Thanks! NOTE: I noticed your point about mount namespaces (interesting I wonder if there will be some change), but merged as it reflects the current kernel code. Kind regards, Petr
On 02. 09. 24 20:13, Chuck Lever III wrote: > Disclaimer: I'm not one of the NFS client maintainers, but only > a very long time contributor to the Linux NFS implementation, > so I can offer only a somewhat-educated opinion on this topic. > > IIRC in a container, the RPC client is bound to the network > namespace. I'm not a NFS expert either. Just thinking out loud because I've noticed that the nfsstat01 test triggers RPC calls from outside the client net namespace.
diff --git a/testcases/network/nfs/nfsstat01/nfsstat01.sh b/testcases/network/nfs/nfsstat01/nfsstat01.sh index 8d7202cf3..3379c4d46 100755 --- a/testcases/network/nfs/nfsstat01/nfsstat01.sh +++ b/testcases/network/nfs/nfsstat01/nfsstat01.sh @@ -22,6 +22,7 @@ get_calls() local name=$1 local field=$2 local nfs_f=$3 + local netns=${4:-rhost} local type="lhost" local calls opt @@ -30,7 +31,8 @@ get_calls() if tst_net_use_netns; then # In netns setup, rhost is the client - [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && type="rhost" + [ "$nfs_f" = "nfs" ] && [ $NS_STAT_RHOST -ne 0 ] && \ + type="$netns" else [ "$nfs_f" != "nfs" ] && type="rhost" fi @@ -64,13 +66,14 @@ get_calls() do_test() { local client_calls server_calls new_server_calls new_client_calls - local client_field server_field + local client_field server_field root_calls new_root_calls local client_v=$VERSION server_v=$VERSION tst_res TINFO "checking RPC calls for server/client" server_calls="$(get_calls rpc 2 nfsd)" client_calls="$(get_calls rpc 2 nfs)" + root_calls="$(get_calls rpc 2 nfs lhost)" tst_res TINFO "calls $server_calls/$client_calls" @@ -79,6 +82,7 @@ do_test() new_server_calls="$(get_calls rpc 2 nfsd)" new_client_calls="$(get_calls rpc 2 nfs)" + new_root_calls="$(get_calls rpc 2 nfs lhost)" tst_res TINFO "new calls $new_server_calls/$new_client_calls" if [ "$new_server_calls" -le "$server_calls" ]; then @@ -93,6 +97,16 @@ do_test() tst_res TPASS "client RPC calls increased" fi + if [ $NS_STAT_RHOST -ne 0 ]; then + tst_res TINFO "Root NS client RPC calls: $root_calls => $new_root_calls" + + if [ $root_calls -ne $new_root_calls ]; then + tst_res TFAIL "RPC stats leaked between net namespaces" + else + tst_res TPASS "RPC stats stay within net namespaces" + fi + fi + tst_res TINFO "checking NFS calls for server/client" case $VERSION in 2) client_field=13 server_field=13
When the NFS server and client run on the same host in different net namespaces, check that RPC calls from the client namespace don't change RPC statistics in the root namespace. Signed-off-by: Martin Doucha <mdoucha@suse.cz> --- I've initially tried to test both NFS and RPC client stats but it appears that NFS client stats are still shared across all namespaces. Only RPC client stats are separate for each net namespace. The kernel patchset[1] which introduced per-NS stats confirms that only RPC stats have been changed. If NFS client stats should be separate for each namespace as well, let me know and I'll return the second set of NS checks in patch v2. Tested on kernel v5.14 with Neil's backports. [1] https://lore.kernel.org/linux-nfs/cover.1708026931.git.josef@toxicpanda.com/ testcases/network/nfs/nfsstat01/nfsstat01.sh | 18 ++++++++++++++++-- 1 file changed, 16 insertions(+), 2 deletions(-)