Re: btrfs: qgroup scan failed with -12

[Josef Bacik <jbacik@fusionio.com>]

On Mon, Sep 23, 2013 at 07:19:06PM +0200, David Sterba wrote:
> On Mon, Sep 23, 2013 at 07:43:44AM +0700, Tomasz Chmielewski wrote:
> > Not sure if it's anything interesting - I had the following entry in
> > dmesg a few days ago, on a server with 32 GB RAM. The system is still working fine.
> 
> Yes this is interesting of course.
> > 
> > [1878432.675210] btrfs-qgroup-re: page allocation failure: order:5, mode:0x104050
> 
> Order 5 allocation, not guaranteed to succeed.
> 
> > [1878432.675319] CPU: 5 PID: 22251 Comm: btrfs-qgroup-re Not tainted 3.11.0-rc7 #2
> > [1878432.676204]  [<ffffffff810bde12>] krealloc+0x52/0x8c
> > [1878432.676324]  [<ffffffffa036bcf0>] find_parent_nodes+0x49c/0x5a5 [btrfs]
> > [1878432.676383]  [<ffffffffa036be75>] btrfs_find_all_roots+0x7c/0xd7 [btrfs]
> > [1878432.676441]  [<ffffffffa036d6e1>] ? qgroup_account_ref_step1+0xea/0x102 [btrfs]
> > [1878432.676542]  [<ffffffffa036d915>] btrfs_qgroup_rescan_worker+0x21c/0x516 [btrfs]
> 
> 220                 new_nodes = krealloc(old, sizeof(*new_nodes) * new_alloced,
> 221                                      gfp_mask);
> 222                 if (!new_nodes)
> 223                         return -ENOMEM;
> 
> The requested size is between 64k and 128k, with 40 bytes of ulist_node
> it's 1638 to 3276 elements. So, lots of things going on during the
> rescan, quite expectable.
> 
> I don't know if krealloc can be replaced with something more friendly to
> allocator, eg. a list of page-sized blocks instead of one contiguous
> array.
> 

I've done that with a patch in bugzilla, hopefully that will fix it.  I've not
had time to try and reproduce myself, but I assume if you do something like
create a random file, then create 100000 snapshots and then defrag it will
probably hit the same problem.  Thanks,

Josef