Re: [PATCH 1/4] xfs: change xfs_isilocked() to always use lockdep()

[Dave Chinner <david@xxxxxxxxxxxxx>]

On Tue, Jan 28, 2020 at 03:55:25PM +0100, Pavel Reichl wrote:
> mr_writer is obsolete and the information it contains is accesible
> from mr_lock.
> 
> Signed-off-by: Pavel Reichl <preichl@xxxxxxxxxx>
> ---
>  fs/xfs/xfs_inode.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
> index c5077e6326c7..32fac6152dc3 100644
> --- a/fs/xfs/xfs_inode.c
> +++ b/fs/xfs/xfs_inode.c
> @@ -352,13 +352,17 @@ xfs_isilocked(
>  {
>  	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
>  		if (!(lock_flags & XFS_ILOCK_SHARED))
> -			return !!ip->i_lock.mr_writer;
> +			return !debug_locks ||
> +				lockdep_is_held_type(&ip->i_lock.mr_lock, 0);
>  		return rwsem_is_locked(&ip->i_lock.mr_lock);
>  	}
>  
>  	if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
>  		if (!(lock_flags & XFS_MMAPLOCK_SHARED))
> -			return !!ip->i_mmaplock.mr_writer;
> +			return !debug_locks ||
> +				lockdep_is_held_type(
> +					&ip->i_mmaplock.mr_lock,
> +					0);
>  		return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
>  	}

Ok, so this code is only called from ASSERT() statements, which
means this turns off write lock checking for XFS debug kernels if
lockdep is not enabled. Hence I think these checks need to be
restructured to be based around rwsem_is_locked() first and lockdep
second.

That is:

/* In all implementations count != 0 means locked */
static inline int rwsem_is_locked(struct rw_semaphore *sem)
{
        return atomic_long_read(&sem->count) != 0;
}

This captures both read and write locks on the rwsem, and doesn't
discriminate at all. Now we don't have explicit writer lock checking
in CONFIG_XFS_DEBUG=y kernels, I think we need to at least check
that the rwsem is locked in all cases to catch cases where we are
calling a function without the lock held. That will ctach most
programming mistakes, and then lockdep will provide the
read-vs-write discrimination to catch the "hold the wrong lock type"
mistakes.

Hence I think this code should end up looking like this:

	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
		bool locked = false;

		if (!rwsem_is_locked(&ip->i_lock))
			return false;
		if (!debug_locks)
			return true;
		if (lock_flags & XFS_ILOCK_EXCL)
			locked = lockdep_is_held_type(&ip->i_lock, 0);
		if (lock_flags & XFS_ILOCK_SHARED)
			locked |= lockdep_is_held_type(&ip->i_lock, 1);
		return locked;
	}

Thoughts?

-Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx