Re: XFS and Kernel intregration and other things

[Stephen Lord <lord@xxxxxxx>]

Shawn Starr wrote:

> Does anyone know when/how XFS will go into 2.4/2.5?

When I have the time to attempt it - which is not now, nor
probably for a while. Linux XFS represents about 20% of
what I am working on right now.

> A lot of kernel developers tell me that XFS needs to dump the pagebuf
> and other layers on top of the VM and VFS layers.
Well, that might be their opinion, unfortunately, the linux buffer cache 
does
not do anything near what xfs requires when it comes to data caching. 
And one
of the reasons I am not submitting code to Linus right now is that I do not
have the time to respond to all of the pushback there will be from people.
XFS represents many man years of development, and you do not just turn it
around and make it use new caches or vfs interfaces overnight.

> I've gotten XFS to work with 2.4.18-pre7 w/ rml's preempt and lockbreak
> patch and the O(1) new scheduler + wli's patch cluster. XFS runs fine
> without errors. I have not seen corruption yet.  I removed riel's rmap
> patch because it seems to break XFS somehow. rmap12a completely changes
> vmscan.c which blows up XFS ;/
There always needs to be a chunk of code in vmscan.c for dealing with 
delayed
allocate data. Without this functioning correctly you will almost certainly
trash filesystems. Look at the diff against a baseline tree, somewhere in
Rick's code there will be something for dealing with dirty pages, this
also needs to apply to Delalloc pages.

> I'm going to be using XFS's testing programs from the CVS tree and
> report the results.
>
> Shawn

Steve