Re: [PATCH] Btrfs: Intialize btrfs_root->highest_objectid when loading tree root and subvolume roots

[Chandan Rajendra <chandan@linux.vnet.ibm.com>]

On Wednesday 07 Oct 2015 11:25:03 David Sterba wrote:
> On Mon, Oct 05, 2015 at 10:14:24PM +0530, Chandan Rajendra wrote:
> > +	if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) {
> > +		mutex_unlock(&root->objectid_mutex);
> > +		ret = -ENOSPC;
> 
> ENOSPC ... I don't think it's right as this could be with a normal
> enospc during subvolume creation. The problem is that theh inode number
> space is exhausted, the closest error code I see is EOVERFLOW. As this
> is an ioctl we can afford to define the meaning of this return value as
> such (unlike for eg. creat()/open()).
> 
> > +		goto free_root_dev;
> > +	}
> > +
> > +	mutex_unlock(&root->objectid_mutex);
> > +
> > 
> >  	return 0;

David, Are you suggesting that we return -EOVERFLOW from within
btrfs_init_fs_root() and continue returning -ENOSPC in case of error
(i.e. tree_root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID) from
open_ctree()?

If yes, btrfs_init_fs_root() gets invoked from open_ctree() via
btrfs_read_fs_root_no_name() and hence we may end up returning -EOVERFLOW when
servicing the mount() syscall.

-- 
chandan