Re: [PATCH 0/7] BKL removal follow-up

[Evgeniy Dushistov <dushistov@mail.ru>]

On Tue, Dec 21, 2010 at 11:54:06PM +0100, Arnd Bergmann wrote:
> On Monday 22 November 2010 16:17:23 Nick Bowler wrote:
> > On 2010-11-21 09:45 -0800, Linus Torvalds wrote:
> > > Yes, I'd be ok with UDF doing a "select BKL" along with a "default n"
> > > for BKL itself.
> > > 
> > > I think UDF currently is the only sane reason to have BKL enabled any
> > > more, and yes, it would probably make it easier to configure things.
> > 
> > UFS (which I use) also relies on BKL.
> 
> Would you mind running a kernel with this patch and lockdep enabled then?
> 
> It's quite likely that this doesn't work, but the easiest way to find
> out is to just try it if you don't understand the code. I can't see anything
> in the code that relies on the release-on-sleep semantics and there
> are no obvious recursive lock_kernel() calls.
> 

I see one without looking at code (am I missed something?). See below.

> @@ -922,22 +922,22 @@ void ufs_evict_inode(struct inode * inode)
>  	if (want_delete) {
>  		loff_t old_i_size;
>  		/*UFS_I(inode)->i_dtime = CURRENT_TIME;*/
> -		lock_kernel();
> +		lock_ufs();

lock
>  		mark_inode_dirty(inode);
>  		ufs_update_inode(inode, IS_SYNC(inode));
>  		old_i_size = inode->i_size;
>  		inode->i_size = 0;
>  		if (inode->i_blocks && ufs_truncate(inode, old_i_size))
call truncate with lock
>  			ufs_warning(inode->i_sb, __func__, "ufs_truncate failed\n");
> -		unlock_kernel();
> +		unlock_ufs();
>  	}

> diff --git a/fs/ufs/truncate.c b/fs/ufs/truncate.c
> index a58f915..9f22d00 100644
> --- a/fs/ufs/truncate.c
> +++ b/fs/ufs/truncate.c
> @@ -467,7 +467,7 @@ int ufs_truncate(struct inode *inode, loff_t old_i_size)
>  
>  	block_truncate_page(inode->i_mapping, inode->i_size, ufs_getfrag_block);
>  
> -	lock_kernel();
> +	lock_ufs();
call lock again

-- 
/Evgeniy