Re: [Ocfs2-devel] [PATCH V3] ocfs2: fix dead lock caused by ocfs2_defrag_extent

[Larry Chen]

Hi Joseph,

On 11/1/18 7:52 PM, Changwei Ge wrote:
> Hello Joseph,
>
> On 2018/11/1 17:01, Joseph Qi wrote:
> > Hi Larry,
> >
> > On 18/11/1 15:14, Larry Chen wrote:
> > > ocfs2_defrag_extent may fall into deadlock.
> > >
> > > ocfs2_ioctl_move_extents
> > >       ocfs2_ioctl_move_extents
> > >         ocfs2_move_extents
> > >           ocfs2_defrag_extent
> > >             ocfs2_lock_allocators_move_extents
> > >
> > >               ocfs2_reserve_clusters
> > >                 inode_lock GLOBAL_BITMAP_SYSTEM_INODE
> > >
> > > 	  __ocfs2_flush_truncate_log
> > >                 inode_lock GLOBAL_BITMAP_SYSTEM_INODE
> > >
> > > As backtrace shows above, ocfs2_reserve_clusters() will call inode_lock
> > > against the global bitmap if local allocator has not sufficient cluters.
> > > Once global bitmap could meet the demand, ocfs2_reserve_cluster will
> > > return success with global bitmap locked.
> > >
> > > After ocfs2_reserve_cluster(), if truncate log is full,
> > > __ocfs2_flush_truncate_log() will definitely fall into deadlock because it
> > > needs to inode_lock global bitmap, which has already been locked.
> > >
> > > To fix this bug, we could remove from ocfs2_lock_allocators_move_extents()
> > > the code which intends to lock global allocator, and put the removed code
> > > after __ocfs2_flush_truncate_log().
> > >
> > > ocfs2_lock_allocators_move_extents() is referred by 2 places, one is here,
> > > the other does not need the data allocator context, which means this patch
> > > does not affect the caller so far.
> > >
> > > Change log:
> > > 1. Correct the function comment.
> > > 2. Remove unused argument from ocfs2_lock_meta_allocator_move_extents.
> > >
> > > Signed-off-by: Larry Chen <lchen@xxxxxxxx>
> > > ---
> > >    fs/ocfs2/move_extents.c | 47 ++++++++++++++++++++++++++---------------------
> > >    1 file changed, 26 insertions(+), 21 deletions(-)
>
> > IMO, here clusters_to_move is only for data_ac, since we change this
> > function to only handle meta_ac, I'm afraid clusters_to_move related
> > logic has to been changed correspondingly.
>
> I think we can't remove *clusters_to_move* from here as clusters can be reserved latter outsides this function, but we
> still have to reserve metadata(extents) in advance.
> So we need that argument.

Yeah, I think clusters_to_move should be reserved, in order to keep the 
original logic as it was.

But I'm curious about why max_recs_needed should be equal to
2 * extents_to_split + cluster_to_move?
Does that mean that each cluster might form an extent??

Thanks,
Larry


> Thanks,
> Changwei
>
> > Thanks,
> > Joseph
> > >    					u32 extents_to_split,
> > >    					struct ocfs2_alloc_context **meta_ac,
> > > -					struct ocfs2_alloc_context **data_ac,
> > >    					int extra_blocks,
> > >    					int *credits)
> > >    {
> > > @@ -192,13 +188,6 @@ static int ocfs2_lock_allocators_move_extents(struct inode *inode,
> > >    		goto out;
> > >    	}
> > >    
> > > -	if (data_ac) {
> > > -		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
> > > -		if (ret) {
> > > -			mlog_errno(ret);
> > > -			goto out;
> > > -		}
> > > -	}
> > >    
> > >    	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
> > >    
> > > @@ -257,10 +246,10 @@ static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
> > >    		}
> > >    	}
> > >    
> > > -	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
> > > -						 &context->meta_ac,
> > > -						 &context->data_ac,
> > > -						 extra_blocks, &credits);
> > > +	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
> > > +						*len, 1,
> > > +						&context->meta_ac,
> > > +						extra_blocks, &credits);
> > >    	if (ret) {
> > >    		mlog_errno(ret);
> > >    		goto out;
> > > @@ -283,6 +272,21 @@ static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
> > >    		}
> > >    	}
> > >    
> > > +	/*
> > > +	 * Make sure ocfs2_reserve_cluster is called after
> > > +	 * __ocfs2_flush_truncate_log, otherwise, dead lock may happen.
> > > +	 *
> > > +	 * If ocfs2_reserve_cluster is called
> > > +	 * before __ocfs2_flush_truncate_log, dead lock on global bitmap
> > > +	 * may happen.
> > > +	 *
> > > +	 */
> > > +	ret = ocfs2_reserve_clusters(osb, *len, &context->data_ac);
> > > +	if (ret) {
> > > +		mlog_errno(ret);
> > > +		goto out_unlock_mutex;
> > > +	}
> > > +
> > >    	handle = ocfs2_start_trans(osb, credits);
> > >    	if (IS_ERR(handle)) {
> > >    		ret = PTR_ERR(handle);
> > > @@ -600,9 +604,10 @@ static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
> > >    		}
> > >    	}
> > >    
> > > -	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
> > > -						 &context->meta_ac,
> > > -						 NULL, extra_blocks, &credits);
> > > +	ret = ocfs2_lock_meta_allocator_move_extents(inode, &context->et,
> > > +						len, 1,
> > > +						&context->meta_ac,
> > > +						extra_blocks, &credits);
> > >    	if (ret) {
> > >    		mlog_errno(ret);
> > >    		goto out;
> >
> > _______________________________________________
> > Ocfs2-devel mailing list
> > Ocfs2-devel@xxxxxxxxxxxxxx
> > https://oss.oracle.com/mailman/listinfo/ocfs2-devel