Re: [PATCH 2/3] fs/locks: Remove fl_nspid

[Jeff Layton]

On Tue, 2017-06-06 at 14:57 -0400, Benjamin Coddington wrote:
> On 6 Jun 2017, at 14:25, Jeff Layton wrote:
> 
> > On Tue, 2017-06-06 at 14:00 -0400, Jeff Layton wrote:
> > > On Tue, 2017-06-06 at 13:19 -0400, Benjamin Coddington wrote:
> > > > Since commit c69899a17ca4 "NFSv4: Update of VFS byte range lock must 
> > > > be
> > > > atomic with the stateid update", NFSv4 has been inserting locks in 
> > > > rpciod
> > > > worker context.  The result is that the file_lock's fl_nspid is the
> > > > kworker's pid instead of the original userspace pid.
> > > > 
> > > > The fl_nspid is only used to represent the namespaced virtual pid 
> > > > number
> > > > when displaying locks or returning from F_GETLK.  There's no reason 
> > > > to set
> > > > it for every inserted lock, since we can usually just look it up 
> > > > from
> > > > fl_pid.  So, instead of looking up and holding struct pid for every 
> > > > lock,
> > > > let's just look up the virtual pid number from fl_pid when it is 
> > > > needed.
> > > > That means we can remove fl_nspid entirely.
> > > > 
> > > 
> > > With this set, I think we ought to codify that the stored pid must be
> > > relative
> > 
> > ...to the init_pid_ns. Let's make that clear in the comments for
> > filesystem authors.
> 
> OK, but I think you mean fl_pid should always be current->tgid or the 
> pid as
> it is in init_pid_ns.  We translate that pid into the virtual pid of the
> process doing F_GETLK or reading /proc/locks.
> 
> > > > Signed-off-by: Benjamin Coddington <bcodding@xxxxxxxxxx>
> > > > ---
> > > >  fs/locks.c         | 58 
> > > > ++++++++++++++++++++++++++++++++----------------------
> > > >  include/linux/fs.h |  1 -
> > > >  2 files changed, 35 insertions(+), 24 deletions(-)
> > > > 
> > > > diff --git a/fs/locks.c b/fs/locks.c
> > > > index d7daa6c8932f..104398ccc9b9 100644
> > > > --- a/fs/locks.c
> > > > +++ b/fs/locks.c
> > > > @@ -733,7 +733,6 @@ static void locks_wake_up_blocks(struct 
> > > > file_lock *blocker)
> > > >  static void
> > > >  locks_insert_lock_ctx(struct file_lock *fl, struct list_head 
> > > > *before)
> > > >  {
> > > > -	fl->fl_nspid = get_pid(task_tgid(current));
> > > >  	list_add_tail(&fl->fl_list, before);
> > > >  	locks_insert_global_locks(fl);
> > > >  }
> > > > @@ -743,10 +742,6 @@ locks_unlink_lock_ctx(struct file_lock *fl)
> > > >  {
> > > >  	locks_delete_global_locks(fl);
> > > >  	list_del_init(&fl->fl_list);
> > > > -	if (fl->fl_nspid) {
> > > > -		put_pid(fl->fl_nspid);
> > > > -		fl->fl_nspid = NULL;
> > > > -	}
> > > >  	locks_wake_up_blocks(fl);
> > > >  }
> > > > 
> > > > @@ -823,8 +818,6 @@ posix_test_lock(struct file *filp, struct 
> > > > file_lock *fl)
> > > >  	list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
> > > >  		if (posix_locks_conflict(fl, cfl)) {
> > > >  			locks_copy_conflock(fl, cfl);
> > > > -			if (cfl->fl_nspid)
> > > > -				fl->fl_pid = pid_vnr(cfl->fl_nspid);
> > > >  			goto out;
> > > >  		}
> > > >  	}
> > > > @@ -2048,6 +2041,31 @@ int vfs_test_lock(struct file *filp, struct 
> > > > file_lock *fl)
> > > >  }
> > > >  EXPORT_SYMBOL_GPL(vfs_test_lock);
> > > > 
> > > > +/**
> > > > + * locks_translate_pid - translate a pid number into a namespace
> > > > + * @nr: The pid number in the init_pid_ns
> > > > + * @ns: The namespace into which the pid should be translated
> > > > + *
> > > > + * Used to tranlate a fl_pid into a namespace virtual pid number
> > > > + */
> > > > +static pid_t locks_translate_pid(int init_nr, struct pid_namespace 
> > > > *ns)
> > > > +{
> > > > +	pid_t vnr = 0;
> > > > +	struct task_struct *task;
> > > > +
> > > > +	rcu_read_lock();
> > > > +	task = find_task_by_pid_ns(init_nr, &init_pid_ns);
> > > > +	if (task)
> > > > +		get_task_struct(task);
> > > > +	rcu_read_unlock();
> > > 
> > > Is that safe? What prevents get_task_struct from doing a 0->1 
> > > transition
> > > there after the task usage count has already gone 1->0 and is on its 
> > > way
> > > to being freed?
> 
> Uh, no -- seems not safe.  I copied that directly from fs/proc/base.c, 
> and
> seems a problem there too.
> 
> Changing this to the below avoids the race with the struct task being
> released:
> 
> rcu_read_lock();
> struct pid = find_pid_ns(init_nr, &init_pid_ns)
> vnr = pid_vnr(pid);
> rcu_read_unlock();
> 

Actually now that I've looked a little more closely, I think it may be
ok. call_rcu callback is what ends up putting the last reference when
the task is released, so if you can find the thing via
find_task_by_pid_ns, then you know that the refcount is at least 1 until
the rcu_read_lock is dropped.

Still, I think doing the pid assignment under rcu_read_lock and not
bothering with the refcount is better in this case...

Thanks,
-- 
Jeff Layton <jlayton@xxxxxxxxxx>