Re: [patch] rewrite rd

[Rob Landley <rob@landley.net>]

On Tuesday 04 December 2007 03:29:40 Nick Piggin wrote:
> On Tue, Dec 04, 2007 at 01:55:17AM -0600, Rob Landley wrote:
> > On Monday 03 December 2007 22:26:28 Nick Piggin wrote:
> > > There is one slight downside -- direct block device access and
> > > filesystem metadata access goes through an extra copy and gets stored
> > > in RAM twice. However, this downside is only slight, because the real
> > > buffercache of the device is now reclaimable (because we're not playing
> > > crazy games with it), so under memory intensive situations, footprint
> > > should effectively be the same -- maybe even a slight advantage to the
> > > new driver because it can also reclaim buffer heads.
> >
> > For the embedded world, initramfs has pretty much rendered initrd
> > obsolete, and that was the biggest user of the ramdisk code I know of. 
> > Beyond that, loopback mounts give you more flexible transient block
> > devices than ramdisks do.  (In fact, ramdisks are such an amazing pain to
> > use/size/free that if I really needed something like that I'd just make a
> > loopback mount in a ramfs instance.)
>
> They are, although we could easily hook up a couple more ioctls for them
> now (if anybody is interested).

>From a usability perspective, an ioctl is no substitute for being able to 
resize a device with "dd".  (Or for that matter, make it sparse.)

> The rd driver can potentially be a _lot_ more scalable than the loop
> driver. It's completely lockless in the fastpath and doesn't even dirty any
> cachelines except for the actual destination memory. OK, this is only
> really important for testing purposes (eg. testing scalability of a
> filesystem etc.) But it is one reason why I (personally) want brd...

I wouldn't say not important: The "database in RAM" people will probably love 
you, at least if they insist on using Oracle.  (Filesystem schmilesystem, 
gimme a raw block device and let me implement my own filesystem from 
userspace...)

But I'm not personally likely to care. :)

> > Embedded users who still want a block interface for memory are generally
> > trying to use a cramfs or squashfs image out of ROM or flash, although
> > there are flash-specific filesystems for this and I dunno if they're
> > actually mounting /dev/mem at an offset or something (md?  losetup -o? 
> > Beats me, I haven't tried that myself yet...)
>
> OK, it would be interesting to hear from anyone using rd for things like
> cramfs.

I'd be interested in that too.  I've heard it proposed a lot but not actually 
seen anyone bother to implement it yet...

Rob
-- 
"One of my most productive days was throwing away 1000 lines of code."
  - Ken Thompson.