On Fri, 28 Jun 2024, Jeff Layton wrote:
> On Thu, 2024-06-27 at 13:14 -0400, Mike Snitzer wrote:
> > On Thu, Jun 27, 2024 at 07:24:54AM -0400, Jeff Layton wrote:
> > > On Wed, 2024-06-26 at 14:24 -0400, Mike Snitzer wrote:
> > > > Introduce nfsd_serv_get and nfsd_serv_put and update the nfsd
> > > > code to
> > > > prevent nfsd_destroy_serv from destroying nn->nfsd_serv until any
> > > > client initiated localio calls to nfsd (that are _not_ in the
> > > > context
> > > > of nfsd) are complete.
> > > >
> > > > nfsd_open_local_fh is updated to nfsd_serv_get before opening its
> > > > file
> > > > handle and then drop the reference using nfsd_serv_put at the end
> > > > of
> > > > nfsd_open_local_fh.
> > > >
> > > > This "interlock" working relies heavily on nfsd_open_local_fh()'s
> > > > maybe_get_net() safely dealing with the possibility that the
> > > > struct
> > > > net (and nfsd_net by association) may have been destroyed by
> > > > nfsd_destroy_serv() via nfsd_shutdown_net().
> > > >
> > > > Verified to fix an easy to hit crash that would occur if an nfsd
> > > > instance running in a container, with a localio client mounted,
> > > > is
> > > > shutdown. Upon restart of the container and associated nfsd the
> > > > client
> > > > would go on to crash due to NULL pointer dereference that
> > > > occuured due
> > > > to the nfs client's localio attempting to nfsd_open_local_fh(),
> > > > using
> > > > nn->nfsd_serv, without having a proper reference on nn-
> > > > >nfsd_serv.
> > > >
> > > > Signed-off-by: Mike Snitzer <snitzer@xxxxxxxxxx>
> > > > ---
> > > > fs/nfsd/localio.c | 2 ++
> > > > fs/nfsd/netns.h | 8 +++++++-
> > > > fs/nfsd/nfssvc.c | 39 +++++++++++++++++++++++++++++++++++++++
> > > > 3 files changed, 48 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/fs/nfsd/localio.c b/fs/nfsd/localio.c
> > > > index 48118db801b5..819589ae2008 100644
> > > > --- a/fs/nfsd/localio.c
> > > > +++ b/fs/nfsd/localio.c
> > > > @@ -204,6 +204,7 @@ int nfsd_open_local_fh(struct net *net,
> > > > }
> > > > nn = net_generic(net, nfsd_net_id);
> > > >
> > > > + nfsd_serv_get(nn);
> > > > serv = READ_ONCE(nn->nfsd_serv);
> > > > if (unlikely(!serv)) {
> > > > dprintk("%s: localio denied. Server not
> > > > running\n", __func__);
> > > > @@ -244,6 +245,7 @@ int nfsd_open_local_fh(struct net *net,
> > > > out_revertcred:
> > > > revert_creds(save_cred);
> > > > out_net:
> > > > + nfsd_serv_put(nn);
> > > > put_net(net);
> > > > return status;
> > > > }
> > > > diff --git a/fs/nfsd/netns.h b/fs/nfsd/netns.h
> > > > index 0c5a1d97e4ac..17a73c780ca1 100644
> > > > --- a/fs/nfsd/netns.h
> > > > +++ b/fs/nfsd/netns.h
> > > > @@ -13,6 +13,7 @@
> > > > #include <linux/filelock.h>
> > > > #include <linux/nfs4.h>
> > > > #include <linux/percpu_counter.h>
> > > > +#include <linux/percpu-refcount.h>
> > > > #include <linux/siphash.h>
> > > > #include <linux/sunrpc/stats.h>
> > > > #include <linux/nfslocalio.h>
> > > > @@ -140,7 +141,9 @@ struct nfsd_net {
> > > >
> > > > struct svc_info nfsd_info;
> > > > #define nfsd_serv nfsd_info.serv
> > > > -
> > > > + struct percpu_ref nfsd_serv_ref;
> > > > + struct completion nfsd_serv_confirm_done;
> > > > + struct completion nfsd_serv_free_done;
> > > >
> > > > /*
> > > > * clientid and stateid data for construction of net
> > > > unique COPY
> > > > @@ -225,6 +228,9 @@ struct nfsd_net {
> > > > extern bool nfsd_support_version(int vers);
> > > > extern void nfsd_netns_free_versions(struct nfsd_net *nn);
> > > >
> > > > +void nfsd_serv_get(struct nfsd_net *nn);
> > > > +void nfsd_serv_put(struct nfsd_net *nn);
> > > > +
> > > > extern unsigned int nfsd_net_id;
> > > >
> > > > void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net
> > > > *nn);
> > > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > > > index a477d2c5088a..be5acb7a4057 100644
> > > > --- a/fs/nfsd/nfssvc.c
> > > > +++ b/fs/nfsd/nfssvc.c
> > > > @@ -258,6 +258,30 @@ int nfsd_minorversion(struct nfsd_net *nn,
> > > > u32 minorversion, enum vers_op change
> > > > return 0;
> > > > }
> > > >
> > > > +void nfsd_serv_get(struct nfsd_net *nn)
> > > > +{
> > > > + percpu_ref_get(&nn->nfsd_serv_ref);
> > > > +}
> > > > +
> > > > +void nfsd_serv_put(struct nfsd_net *nn)
> > > > +{
> > > > + percpu_ref_put(&nn->nfsd_serv_ref);
> > > > +}
> > > > +
> > > > +static void nfsd_serv_done(struct percpu_ref *ref)
> > > > +{
> > > > + struct nfsd_net *nn = container_of(ref, struct nfsd_net,
> > > > nfsd_serv_ref);
> > > > +
> > > > + complete(&nn->nfsd_serv_confirm_done);
> > > > +}
> > > > +
> > > > +static void nfsd_serv_free(struct percpu_ref *ref)
> > > > +{
> > > > + struct nfsd_net *nn = container_of(ref, struct nfsd_net,
> > > > nfsd_serv_ref);
> > > > +
> > > > + complete(&nn->nfsd_serv_free_done);
> > > > +}
> > > > +
> > > > /*
> > > > * Maximum number of nfsd processes
> > > > */
> > > > @@ -462,6 +486,7 @@ static void nfsd_shutdown_net(struct net
> > > > *net)
> > > > lockd_down(net);
> > > > nn->lockd_up = false;
> > > > }
> > > > + percpu_ref_exit(&nn->nfsd_serv_ref);
> > > > #if IS_ENABLED(CONFIG_NFSD_LOCALIO)
> > > > list_del_rcu(&nn->nfsd_uuid.list);
> > > > #endif
> > > > @@ -544,6 +569,13 @@ void nfsd_destroy_serv(struct net *net)
> > > > struct nfsd_net *nn = net_generic(net, nfsd_net_id);
> > > > struct svc_serv *serv = nn->nfsd_serv;
> > > >
> > > > + lockdep_assert_held(&nfsd_mutex);
> > > > +
> > > > + percpu_ref_kill_and_confirm(&nn->nfsd_serv_ref,
> > > > nfsd_serv_done);
> > > > + wait_for_completion(&nn->nfsd_serv_confirm_done);
> > > > + wait_for_completion(&nn->nfsd_serv_free_done);
> > > > + /* percpu_ref_exit is called in nfsd_shutdown_net */
> > > > +
> > >
> > > At this point where you're waiting on these completion vars, what
> > > stops
> > > the client from issuing new localio requests?
> >
> > The server going down. The interlock is with nfsd_open_local_fh, so
> > once any active file opens complete the server will get these
> > completions and the server will go down. The client will reconnect
> > if/when the server comes back up.
> >
> > Maybe I'm not imaginative enough, but is there risk of never-ending
> > overlapping file opens that prevent nfsd_destroy_serv() from
> > progressing?
> >
>
> Yeah, I think it's a possibility, especially if the underlying
> filesystem is particularly slow. You can probably solve that by just
> not handing out new references somehow once you're going to call
> percpu_ref_kill_and_confirm.
percpu-refcount has just to tool you need.
Use percpu_ref_tryget_live(). nfsd_serv_get() must be able to fail, and
so it must use this. If it does then it can safely read ->nfsd_serv
without checking for NULL.
NeilBrown
