Re: [patch 3/5] fs: introduce new truncate sequence

[Christoph Hellwig <hch@lst.de>]

On Mon, Aug 17, 2009 at 06:41:44PM +0200, Nick Piggin wrote:
> Just wonder what you think of this for an updated patch. Main changes
> since last time are changelog changed, more comments, and addition of 
> simple_setattr and inode_set_attributes. I was going to leave
> simple_setattr for later, but now I found when converting the simple
> filesystem ramfs that we basically need it anyway.
> 
> inode_set_attributes is also following your suggestion and if we use
> it then we don't have to do that masking away ATTR_SIZE then calling
> inode_setattr that you didn't like (and I agree with).

I like this a lot.  But please change the inode_set_attributes name,
it's awkward and totally falls out of the scheme.  It defintively should
be <something>_setattr.  Not sure about what to use for the <something>.
Maybe just generic_setattr, mirroring generic_getattr?

Btw, one idea on how to avoid having to touch all the
begin_write/end_write/direct_IO instances:  What about passing another
callback to them, in addition to the get_blocks also a trim_blocks
which we call to trim blocks over i_size.  That would be the old
->truncate minus the block_truncate_page.