Re: [RFC] [PATCH] Inverse locking order of page_lock and transaction start

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

On Thu, Mar 27, 2008 at 05:27:42PM +0100, Jan Kara wrote:
>   Hi,
> 
>   below is the first version of the patch that reverses locking order of
> page_lock and transaction start. I have tested it with fsx-linux, ltp DIO
> tests etc. and lockdep didn't complain so hopefully I got it mostly right
> but review is definitely needed. Especially I'd like to know what people
> think about the way I've implemented ext3_page_mkwrite() - ext4 has
> an incorrect code AFAICT because in ordered and journaled modes we should
> write block of zeros and properly journal it (and no, block_page_mkwrite()
> doesn't do it). We could implement ext3/4_page_mkwrite() in a similar way
> we currently implement writepage calls but calling write_begin + write_end
> does the job and should be only a tiny bit slower...
>   If nobody finds a serious flaw in the approach, I'll rediff the patch
> against ext4 (I'll also try to convert delayed-alloc path - from a quick
> look converting da_writepages path is going to be interesting).
>   I'm looking forward to your comments :)
> 

I guess we should make this change first to Ext4 and then make the
same to ext3.

-aneesh