On Tue, Apr 17, 2018 at 07:01:10AM -0700, Matthew Wilcox wrote:
> On Thu, Apr 05, 2018 at 02:58:06PM +0300, Kirill Tkhai wrote:
> > I observed the following deadlock between them:
> > 
> > [task 1]                          [task 2]                         [task 3]
> > kill_fasync()                     mm_update_next_owner()           copy_process()
> >  spin_lock_irqsave(&fa->fa_lock)   read_lock(&tasklist_lock)        write_lock_irq(&tasklist_lock)
> >   send_sigio()                    <IRQ>                             ...
> >    read_lock(&fown->lock)         kill_fasync()                     ...
> >     read_lock(&tasklist_lock)      spin_lock_irqsave(&fa->fa_lock)  ...
> > 
> > Task 1 can't acquire read locked tasklist_lock, since there is
> > already task 3 expressed its wish to take the lock exclusive.
> > Task 2 holds the read locked lock, but it can't take the spin lock.
> 
> I think the important question is to Peter ... why didn't lockdep catch this?
> 

I think the following will help lockdep to catch this:


@@ -570,7 +588,14 @@ static inline void print_irqtrace_events(struct task_struct *curr)
 #define spin_release(l, n, i)			lock_release(l, n, i)
 
 #define rwlock_acquire(l, s, t, i)		lock_acquire_exclusive(l, s, t, NULL, i)
-#define rwlock_acquire_read(l, s, t, i)		lock_acquire_shared_recursive(l, s, t, NULL, i)
+#define rwlock_acquire_read(l, s, t, i)				\
+do {									\
+	if (!IS_ENABLED(CONFIG_QUEUED_RWLOCKS) || in_interrupt())	\
+		lock_acquire_shared_recursive(l, s, t, NULL, i);	\
+	else								\
+		lock_acquire_shared(l, s, t, NULL, i);			\
+} while (0)
+

However, this will break several self tests in lib/locking-selftest.c,
because we used to treat read_lock() as recursive read locks for all
callsites from lockdep's viewpoint.

Besides, the above change will bring an interesting challenge for the
recursive read lock deadlock detection work that I'm doing:

	https://marc.info/?l=linux-kernel&m=152345444100825

I will explain that in the thread of that patchset and add you and
others Cced in case that you're interested.

Regards,
Boqun

> > -		spin_lock_irq(&fa->fa_lock);
> > +		write_lock_irq(&fa->fa_lock);
> >  		fa->fa_file = NULL;
> > -		spin_unlock_irq(&fa->fa_lock);
> > +		write_unlock_irq(&fa->fa_lock);
> ...
> > -		spin_lock_irq(&fa->fa_lock);
> > +		write_lock_irq(&fa->fa_lock);
> >  		fa->fa_fd = fd;
> > -		spin_unlock_irq(&fa->fa_lock);
> > +		write_unlock_irq(&fa->fa_lock);
> 
> Do we really need a lock here?  If we convert each of these into WRITE_ONCE,
> then 
> 
> ...
> > -		spin_lock_irqsave(&fa->fa_lock, flags);
> > +		read_lock(&fa->fa_lock);
> >  		if (fa->fa_file) {
> 
> file = READ_ONCE(fa->fa_file)
> 
> then we're not opening any new races, are we?
> 
> >  			fown = &fa->fa_file->f_owner;
> >  			/* Don't send SIGURG to processes which have not set a
> > @@ -997,7 +996,7 @@ static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
> >  			if (!(sig == SIGURG && fown->signum == 0))
> >  				send_sigio(fown, fa->fa_fd, band);
> >  		}
> > -		spin_unlock_irqrestore(&fa->fa_lock, flags);
> > +		read_unlock(&fa->fa_lock);
> >  		fa = rcu_dereference(fa->fa_next);
> >  	}
> >  }
> > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > index c6baf767619e..297e2dcd9dd2 100644
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -1250,7 +1250,7 @@ static inline int locks_lock_file_wait(struct file *filp, struct file_lock *fl)
> >  }
> >  
> >  struct fasync_struct {
> > -	spinlock_t		fa_lock;
> > +	rwlock_t		fa_lock;
> >  	int			magic;
> >  	int			fa_fd;
> >  	struct fasync_struct	*fa_next; /* singly linked list */
> >