On Tue, Mar 16, 2010 at 01:35:54PM -0400, Trond Myklebust wrote:
> On Tue, 2010-03-16 at 10:17 -0700, Paul E. McKenney wrote:
> > On Tue, Mar 16, 2010 at 11:51:30AM +0000, David Howells wrote:
> > > diff --git a/fs/nfs/delegation.c b/fs/nfs/delegation.c
> > > index 2563beb..a77c735 100644
> > > --- a/fs/nfs/delegation.c
> > > +++ b/fs/nfs/delegation.c
> > > @@ -37,8 +37,10 @@ static void nfs_free_delegation(struct nfs_delegation *delegation)
> > > {
> > > struct rpc_cred *cred;
> > >
> > > + rcu_read_lock();
> > > cred = rcu_dereference(delegation->cred);
> > > rcu_assign_pointer(delegation->cred, NULL);
> >
> > The lock is probably held here, in which case something like the
> > following would work well without needing the artificial rcu_read_lock()
> > and rcu_read_unlock():
>
> No. The lock is not held here. At this point, the delegation has been
> detached from the inode that pointed to it, and so we can free up its
> contents.
OK. Is there some reference counter or pointer that can be checked to
verify that this data structure really is in a state that prevents
RCU readers from finding it?
> We still need the call_rcu() to free up the allocated memory in order to
> ensure that some process doing lockless traversal of the
> clp->cl_delegations list doesn't crash.
OK.
> > cred = rcu_dereference_check(delegation->cred,
> > lockdep_is_held(&delegation->lock));
> >
> > > + rcu_read_unlock();
> > > call_rcu(&delegation->rcu, nfs_free_delegation_callback);
> > > if (cred)
> > > put_rpccred(cred);
> > > @@ -212,10 +214,12 @@ int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct
> > > spin_lock_init(&delegation->lock);
> > >
> > > spin_lock(&clp->cl_lock);
> > > + rcu_read_lock();
> > > if (rcu_dereference(nfsi->delegation) != NULL) {
> >
> > Same here, though I am not sure whether clp->cl_lock or something in
> > nfs_inode should be used.
>
> Yes. As I indicated to David in another email, the clp->cl_lock protects
> us here, so it looks as if your suggestion above would be perfect.
Sounds good!
Thanx, Paul
> > > if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
> > > sizeof(delegation->stateid)) == 0 &&
> > > delegation->type == nfsi->delegation->type) {
> > > + rcu_read_unlock();
> > > goto out;
> > > }
> > > /*
> > > @@ -228,6 +232,7 @@ int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct
> > > if (delegation->type <= nfsi->delegation->type) {
> > > freeme = delegation;
> > > delegation = NULL;
> > > + rcu_read_lock();
> > > goto out;
> > > }
> > > freeme = nfs_detach_delegation_locked(nfsi, NULL);
>
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
