On Mon, Sep 19, 2011 at 12:33:20PM -0500, Alex Elder wrote:
> On Wed, 2011-09-14 at 10:08 -0400, Christoph Hellwig wrote:
> > The code to flush buffers in the umount code is a bit iffy: we first flush
> > all delwri buffers out, but then might be able to queue up a new one when
> > logging the sb counts. On a normal shutdown that one would get flushed
> > out when doing the synchronous superblock write in xfs_unmountfs_writesb,
> > but we skip that one if the filesystem has been shut down.
> > Fix this by moving the delwri list flushing until just before unmounting
> > the log, and while we're at it also remove the superflous delwri list
> > and buffer lru flusing for the rt and log device that can never have
> > cached or delwri buffers.
> What prevents blocks on a realtime device from being cached?
Note that cached here means using buffers that end up in the rbtree
and/or on the lru. We simply never do that for the rt and log device,
as that would mean complicating the code for no reason - the only
time we ever use buffer on the rt device is for the size check during
mount and the _uncached variant where we free the buffer after the
refcount hits zero is just fine for that, and xfs_zero_remaining_bytes,
which actually operates on a regular file and thus must not use the
rbtree or the lru (and should eventually stop using buffer routines at
all). All actual metadata which the buffer cache is used for resides
on the main device. Log buffers stay alive from mount to unmount so
we're fine with that interface as well.
> I haven't looked exhaustively, but it looks like the summary
> and bitmap data are operated on without caching, but the
> realtime data itself is treated no different from "normal"
> data blocks.
The buffer cache is never used for actual file data.