Re: [git patches] xfs and block fixes for virtually indexed arches

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Fri, 2009-12-18 at 13:44 +1100, Dave Chinner wrote:
> On Fri, Dec 18, 2009 at 10:00:21AM +0900, FUJITA Tomonori wrote:
> > On Fri, 18 Dec 2009 00:57:00 +0100
> > James Bottomley <James.Bottomley@suse.de> wrote:
> > 
> > > On Thu, 2009-12-17 at 20:36 +0100, Jens Axboe wrote:
> > > > On Thu, Dec 17 2009, Linus Torvalds wrote:
> > > > > 
> > > > > 
> > > > > On Thu, 17 Dec 2009, tytso@mit.edu wrote:
> > > > > > 
> > > > > > Sure, but there's some rumors/oral traditions going around that some
> > > > > > block devices want bio address which are page aligned, because they
> > > > > > want to play some kind of refcounting game,
> > > > > 
> > > > > Yeah, you might be right at that.
> > > > > 
> > > > > > And it's Weird Shit(tm) (aka iSCSI, AoE) type drivers, that most of us 
> > > > > > don't have access to, so just because it works Just Fine on SATA doesn't 
> > > > > > mean anything.
> > > > > > 
> > > > > > And none of this is documented anywhere, which is frustrating as hell.
> > > > > > Just rumors that "if you do this, AoE/iSCSI will corrupt your file
> > > > > > systems".
> > > > > 
> > > > > ACK. Jens? 
> > > > 
> > > > I've heard those rumours too, and I don't even know if they are true.
> > > > Who has a pointer to such a bug report and/or issue? The block layer
> > > > itself doesn't not have any such requirements, and the only places where
> > > > we play page games is for bio's that were explicitly mapped with pages
> > > > by itself (like mapping user data).o
> > > 
> > > OK, so what happened is that prior to the map single fix
> > > 
> > > commit df46b9a44ceb5af2ea2351ce8e28ae7bd840b00f
> > > Author: Mike Christie  <michaelc@cs.wisc.edu>
> > > Date:   Mon Jun 20 14:04:44 2005 +0200
> > > 
> > >     [PATCH] Add blk_rq_map_kern()
> > > 
> > > 
> > > bio could only accept user space buffers, so we had a special path for
> > > kernel allocated buffers.  That commit unified the path (with a separate
> > > block API) so we could now submit kmalloc'd buffers via block APIs.
> > > 
> > > So the rule now is we can accept any user mapped area via
> > > blk_rq_map_user and any kmalloc'd area via blk_rq_map_kern().  We might
> > > not be able to do a stack area (depending on how the arch maps the
> > > stack) and we definitely cannot do a vmalloc'd area.
> > > 
> > > So it sounds like we only need a blk_rq_map_vmalloc() using the same
> > > techniques as the patch set and we're good to go.
> > 
> > I'm not sure about it.
> > 
> > As I said before (when I was against this 'adding vmalloc support to
> > the block layer' stuff), are there potential users of this except for
> > XFS? Are there anyone who does such a thing now?
> 
> As Christoph already mentioned, XFS is not passing the vmalloc'd
> range to the block layer - it passes the underlying pages to the
> block layer. Hence I'm not sure there actually is anyone who is
> passing vmalloc'd addresses to the block layer. Perhaps we should
> put a WARN_ON() in the block layer to catch anyone doing such a
> thing before considering supporting vmalloc'd addresses in the block
> layer?

vmalloc is just an alias for vmap/vmalloc in the above statements
(basically anything with an additional kernel virtual mapping which
causes aliases).  If we support vmap, we naturally support vmalloc as
well.

> > This API might be useful for only journaling file systems using log
> > formats that need large contiguous buffer. Sound like only XFS?
> 
> FWIW, mapped buffers larger than PAGE_SIZE are used for more than just log
> recovery in XFS. e.g. filesystems with directory block size larger
> than page size uses mapped buffers.

However, XFS is the only fs that actually uses kernel virtual mapping to
solve this problem.

James