Re: [PATCH 01/13] seqnum_ops: Introduce Sequence Number Ops

[Shuah Khan]

On 11/10/20 5:18 PM, Kees Cook wrote:
> On Tue, Nov 10, 2020 at 09:43:02PM +0100, Greg KH wrote:
> > On Tue, Nov 10, 2020 at 09:41:40PM +0100, Greg KH wrote:
> > > On Tue, Nov 10, 2020 at 12:53:27PM -0700, Shuah Khan wrote:
> > > > +Decrement interface
> > > > +-------------------
> > > > +
> > > > +Decrements sequence number and doesn't return the new value. ::
> > > > +
> > > > +        seqnum32_dec() --> atomic_dec()
> > > > +        seqnum64_dec() --> atomic64_dec()
> > >
> > > Why would you need to decrement a sequence number?  Shouldn't they just
> > > always go up?
> > >
> > > I see you use them in your patch 12/13, but I don't think that really is
> > > a sequence number there, but rather just some other odd value :)
>
> To that end, they should likely be internally cast to u32 and u64 (and
> why is seqnum64 ifdef on CONFIG_64BIT?).

I had a reason for doing this, can't recall. I will revisit and remove
it which is ideal. I have to look at CONFIG_GENERIC_ATOMIC64 as well.

Not seeing any errors with my config which has CONFIG_64BIT=y


> > Note, other than this, I like the idea.  It makes it obvious what these
> > atomic variables are being used for, and they can't be abused for other
> > things.  Nice work.

Thanks.

> Agreed: this is a clear wrapping sequence counter. It's only abuse would
> be using it in a place where wrapping actually is _not_ safe. (bikeshed:
> can we call it wrap_u32 and wrap_u64?)

Still like seqnum_ops.

There is seqcount_t in seqlock.h which is a totally different feature.

thanks,
-- Shuah