[lxc-devel] [RFC PATCH] lxclock: Replace named sempahore with flock

[Serge Hallyn]

Quoting Dwight Engen (dwight.engen at oracle.com):
> On Fri, 24 May 2013 08:09:21 -0500
> > -sem_t *lxc_newlock(const char *name)
> > +struct lxc_lock *lxc_newlock(const char *lxcpath, const char *name)
> >  {
> > -	char *lname;
> > -	sem_t *lock;
> > +	struct lxc_lock *l;
> > +	int ret = pthread_mutex_lock(&thread_mutex);
> 
> What is taking this lock here protecting? It looks like its all thread
> local allocations? Maybe I'm missing something but I'm confused why we
> need the pthread mutex around any of these locking primitives. Won't
> this potentially cause a threadA trying to operate on unlocked
> container1 to block when threadB is in a critical section with
> unrelated container2?

Yeah, we need it around the fopen() and flock(), you're right
we shouldn't need it here.

-serge