Message ID | 20221107145213.913178-3-jiri@resnulli.us (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | net: devlink: move netdev notifier block to dest namespace during reload | expand |
On Mon, Nov 07, 2022 at 03:52:13PM +0100, Jiri Pirko wrote: > From: Jiri Pirko <jiri@nvidia.com> > > The notifier block tracking netdev changes in devlink is registered > during devlink_alloc() per-net, it is then unregistered > in devlink_free(). When devlink moves from net namespace to another one, > the notifier block needs to move along. > > Fix this by adding forgotten call to move the block. > > Reported-by: Ido Schimmel <idosch@idosch.org> > Fixes: 02a68a47eade ("net: devlink: track netdev with devlink_port assigned") > Signed-off-by: Jiri Pirko <jiri@nvidia.com> Does not trigger with my reproducer. Will test the fix tonight in regression and report tomorrow morning. > --- > net/core/devlink.c | 5 ++++- > 1 file changed, 4 insertions(+), 1 deletion(-) > > diff --git a/net/core/devlink.c b/net/core/devlink.c > index 40fcdded57e6..ea0b319385fc 100644 > --- a/net/core/devlink.c > +++ b/net/core/devlink.c > @@ -4502,8 +4502,11 @@ static int devlink_reload(struct devlink *devlink, struct net *dest_net, > if (err) > return err; > > - if (dest_net && !net_eq(dest_net, curr_net)) > + if (dest_net && !net_eq(dest_net, curr_net)) { > + move_netdevice_notifier_net(curr_net, dest_net, > + &devlink->netdevice_nb); > write_pnet(&devlink->_net, dest_net); > + } I suggest adding this: diff --git a/net/core/devlink.c b/net/core/devlink.c index 83fd10aeddd5..3b5aedc93335 100644 --- a/net/core/devlink.c +++ b/net/core/devlink.c @@ -9843,8 +9843,8 @@ void devlink_free(struct devlink *devlink) xa_destroy(&devlink->snapshot_ids); - unregister_netdevice_notifier_net(devlink_net(devlink), - &devlink->netdevice_nb); + WARN_ON(unregister_netdevice_notifier_net(devlink_net(devlink), + &devlink->netdevice_nb)); xa_erase(&devlinks, devlink->index); This tells about the failure right away. Instead, we saw random memory corruptions in later tests. > > err = devlink->ops->reload_up(devlink, action, limit, actions_performed, extack); > devlink_reload_failed_set(devlink, !!err); > -- > 2.37.3 >
Mon, Nov 07, 2022 at 05:52:08PM CET, idosch@idosch.org wrote: >On Mon, Nov 07, 2022 at 03:52:13PM +0100, Jiri Pirko wrote: >> From: Jiri Pirko <jiri@nvidia.com> >> >> The notifier block tracking netdev changes in devlink is registered >> during devlink_alloc() per-net, it is then unregistered >> in devlink_free(). When devlink moves from net namespace to another one, >> the notifier block needs to move along. >> >> Fix this by adding forgotten call to move the block. >> >> Reported-by: Ido Schimmel <idosch@idosch.org> >> Fixes: 02a68a47eade ("net: devlink: track netdev with devlink_port assigned") >> Signed-off-by: Jiri Pirko <jiri@nvidia.com> > >Does not trigger with my reproducer. Will test the fix tonight in >regression and report tomorrow morning. Ok! > >> --- >> net/core/devlink.c | 5 ++++- >> 1 file changed, 4 insertions(+), 1 deletion(-) >> >> diff --git a/net/core/devlink.c b/net/core/devlink.c >> index 40fcdded57e6..ea0b319385fc 100644 >> --- a/net/core/devlink.c >> +++ b/net/core/devlink.c >> @@ -4502,8 +4502,11 @@ static int devlink_reload(struct devlink *devlink, struct net *dest_net, >> if (err) >> return err; >> >> - if (dest_net && !net_eq(dest_net, curr_net)) >> + if (dest_net && !net_eq(dest_net, curr_net)) { >> + move_netdevice_notifier_net(curr_net, dest_net, >> + &devlink->netdevice_nb); >> write_pnet(&devlink->_net, dest_net); >> + } > >I suggest adding this: > >diff --git a/net/core/devlink.c b/net/core/devlink.c >index 83fd10aeddd5..3b5aedc93335 100644 >--- a/net/core/devlink.c >+++ b/net/core/devlink.c >@@ -9843,8 +9843,8 @@ void devlink_free(struct devlink *devlink) > > xa_destroy(&devlink->snapshot_ids); > >- unregister_netdevice_notifier_net(devlink_net(devlink), >- &devlink->netdevice_nb); >+ WARN_ON(unregister_netdevice_notifier_net(devlink_net(devlink), >+ &devlink->netdevice_nb)); > > xa_erase(&devlinks, devlink->index); > >This tells about the failure right away. Instead, we saw random memory >corruptions in later tests. Should be a separate patch then. > >> >> err = devlink->ops->reload_up(devlink, action, limit, actions_performed, extack); >> devlink_reload_failed_set(devlink, !!err); >> -- >> 2.37.3 >>
On Mon, Nov 07, 2022 at 03:52:13PM +0100, Jiri Pirko wrote: > From: Jiri Pirko <jiri@nvidia.com> > > The notifier block tracking netdev changes in devlink is registered > during devlink_alloc() per-net, it is then unregistered > in devlink_free(). When devlink moves from net namespace to another one, > the notifier block needs to move along. > > Fix this by adding forgotten call to move the block. > > Reported-by: Ido Schimmel <idosch@idosch.org> > Fixes: 02a68a47eade ("net: devlink: track netdev with devlink_port assigned") > Signed-off-by: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Ido Schimmel <idosch@nvidia.com> Tested-by: Ido Schimmel <idosch@nvidia.com> Thanks!
Tue, Nov 08, 2022 at 09:11:49AM CET, idosch@idosch.org wrote: >On Mon, Nov 07, 2022 at 03:52:13PM +0100, Jiri Pirko wrote: >> From: Jiri Pirko <jiri@nvidia.com> >> >> The notifier block tracking netdev changes in devlink is registered >> during devlink_alloc() per-net, it is then unregistered >> in devlink_free(). When devlink moves from net namespace to another one, >> the notifier block needs to move along. >> >> Fix this by adding forgotten call to move the block. >> >> Reported-by: Ido Schimmel <idosch@idosch.org> >> Fixes: 02a68a47eade ("net: devlink: track netdev with devlink_port assigned") >> Signed-off-by: Jiri Pirko <jiri@nvidia.com> > >Reviewed-by: Ido Schimmel <idosch@nvidia.com> >Tested-by: Ido Schimmel <idosch@nvidia.com> Sending v2 with cosmetical changes. Please put your tags there again. Thanks! > >Thanks!
Tue, Nov 08, 2022 at 01:59:50PM CET, jiri@resnulli.us wrote: >Tue, Nov 08, 2022 at 09:11:49AM CET, idosch@idosch.org wrote: >>On Mon, Nov 07, 2022 at 03:52:13PM +0100, Jiri Pirko wrote: >>> From: Jiri Pirko <jiri@nvidia.com> >>> >>> The notifier block tracking netdev changes in devlink is registered >>> during devlink_alloc() per-net, it is then unregistered >>> in devlink_free(). When devlink moves from net namespace to another one, >>> the notifier block needs to move along. >>> >>> Fix this by adding forgotten call to move the block. >>> >>> Reported-by: Ido Schimmel <idosch@idosch.org> >>> Fixes: 02a68a47eade ("net: devlink: track netdev with devlink_port assigned") >>> Signed-off-by: Jiri Pirko <jiri@nvidia.com> >> >>Reviewed-by: Ido Schimmel <idosch@nvidia.com> >>Tested-by: Ido Schimmel <idosch@nvidia.com> > >Sending v2 with cosmetical changes. Please put your tags there again. Actually, this patch stays untouched. So I'll add it. >Thanks! > >> >>Thanks!
diff --git a/net/core/devlink.c b/net/core/devlink.c index 40fcdded57e6..ea0b319385fc 100644 --- a/net/core/devlink.c +++ b/net/core/devlink.c @@ -4502,8 +4502,11 @@ static int devlink_reload(struct devlink *devlink, struct net *dest_net, if (err) return err; - if (dest_net && !net_eq(dest_net, curr_net)) + if (dest_net && !net_eq(dest_net, curr_net)) { + move_netdevice_notifier_net(curr_net, dest_net, + &devlink->netdevice_nb); write_pnet(&devlink->_net, dest_net); + } err = devlink->ops->reload_up(devlink, action, limit, actions_performed, extack); devlink_reload_failed_set(devlink, !!err);