Re: [Dazuko-devel] Bug in dazukofs_open()

[John Ogness]

On 2008-11-27, Lino Sanfilippo <address@hidden> wrote:
> while I was working with dazukoFS I encountered some strange
> problems with files that had been blocked by an application via
> dazuko before.  If such a file has been deleted it was not possible
> to create a file (i.e. by touching it) with the same name where it
> has been located before.
>
> The error message was :"File exists", though an "ls" did not show it
> anymore.
>
> It was also impossible to unmount the underlying device (i.e. a usb
> stick). The error message was: "device is busy".
>
> I found out that the reason for these problems is a bug in
> dazukofs_open():
>
> If dazuko_check_access() returns with an error code, the usage
> counters of lower_dentry and lower_mnt that had been incremented
> before, are not decremented again. This prevents the underlaying
> filesystem from removing the corresponding dentry and inode.  I
> created a patch (for 3.00-rc3 with linux-2.6.26 patch) that resolves
> that problem.

Thank you. This bug appeared in nullfs back in September after I
realized that dentry_open() already releases the resources. It was a
careless change on my part. I am glad that you found it.

I have made the fixes in the kernel submission patches I am preparing.

http://dazuko.dnsalias.org/wiki/index.php/Linux_Submission

The patches also include several other minor fixes. Once all 5 patches
are ready, I will post 3.0.0-rc4.

> Will dazukoFS support additional file access events in future,
> i.e. for "on close"?

I am not yet certain about this. Assuming we get DazukoFS into the
mainline kernel, we can start looking into adding new features. For
example, I really like how RedirFS handles the various VFS hooks. I
could imagine using the same API to allow DazukoFS to support pre/post
hooks for every VFS function.

But we will have to wait and see. Right now I am really spending every
free moment preparing the DazukoFS patches for kernel submission.

John Ogness

-- 
Dazuko Maintainer