Re: [RFC 0/13] extents and 48bit ext3

On Sat, Jun 10, 2006 at 01:20:31AM +0400, Alex Tomas wrote:
> two point here:
> a) warnings should be made visible at mount time,
>    something like printk(KERN_CRIT ...)

	Too late, they're already broken!

> b) I don't think you're going to fight all crazy people in the world,
>    they'll definitely find a way to break something:
>    data or something else.

	Certainly not the crazy people.  But the random person who's
just humming along?  We should be nice to them.
 
> PS. in the end, "extents" option affects *new* files only. and one
>     can boot extents-enabled kernel and convert fs back.

	I just mentioned to Ted in another mail, since this is a
"permanent" change to the on-disk structure, why is this a mount option?
Shouldn't it rather be a tunefs(8)/mkfs(8) option?
	In general, anything you pass to "mount -o" is optional.  You
can mount with option X, then unmount and mount without option X.  Most
people "expect" this to work (Principle of Least Surprise).  So, when
you do:

    # mount -o extents /fs1
    # create_file /fs1/newfile
    # umount /fs1
    # mount /fs1

it breaks.  Lease Surprise expects it to work.
	However, tunefs(8) and mkfs(8) is generally understood to make
physical changes.  Why not "tunefs -extents" to turn them on?  It's
completely analogous to "tunefs -J", will fit everyone's expectation,
and won't surprise people.  "mkfs -extents" does the same thing.

Joel

-- 

Life's Little Instruction Book #232

	"Keep your promises."

Joel Becker
Principal Software Developer
Oracle
E-mail: [email protected]
Phone: (650) 506-8127