On Thu, Nov 06, 2014 at 02:47:29AM -0800, Pieter Wuille wrote:
> On Thu, Nov 6, 2014 at 2:38 AM, Peter Todd <pete@petertodd.org> wrote:
> > However the implementation of the STRICTENC flag simply makes pubkey
> > formats it doesn't recognize act as through the signature was invalid,
> > rather than failing the transaction. Similar to the invalid due to too
> > many sigops DoS attack I found before, this lets you fill up the mempool
> > with garbage transactions that will never be mined. OTOH I don't see any
> > way to exploit this in a v0.9.x IsStandard() transaction, so we haven't
> > shipped code that actually has this vulnerability. (dunno about
> > alt-implementations)
> 
> Yeah, there's even a comment in script/interpreter.h currently about
> how STRICTENC is not softfork safe.

Indeed.

I actually was thinking about SCRIPT_VERIFY_MINIMALDATA, CScript(), and
FindAndDelete() Specifically that if you were to change CScript() to
convert single-character PUSHDATA's to OP_<number> you'd be making a
consensus-critical change due to how FindAndDelete() is called with a a
CScript() signature. You didn't make that mistake, and I couldn't find a
way to exploit it anyway, but it reminded me of this STRICTENC stuff.

> I didn't realize that this would
> lead to the mempool accepting invalid transactions (I thought there
> was a second validity check with the actual consensus rules; if not,
> maybe we need to add that).

It should be enough to just duplicate the CheckInputs() call in
the AcceptToMemoryPool() function:

    if (!CheckInputs(tx, state, view, true, STANDARD_SCRIPT_VERIFY_FLAGS, true))
    {
        return error("AcceptToMemoryPool: : ConnectInputs failed %s", hash.ToString());
    }
    if (!CheckInputs(tx, state, view, true, MANDATORY_SCRIPT_VERIFY_FLAGS, true))
    {
        return error("AcceptToMemoryPool: : BUG FOUND Standard verify flags passed yet mandatory flags failed. %s", hash.ToString());
    }


> > I suggest we either change STRICTENC to simply fail unrecognized pubkeys
> > immediately - similar to how non-standard signatures are treated - or
> > fail the script if the pubkey is non-standard and signature verification
> > succeeds.
> 
> Sounds good to me, I disliked those semantics too.

Ok, then given we have to support hybrid encoding for awhile longer
anyway - I noticed your secp256k1 library supports it - lets do the
latter as a "least invasive" measure. I can't think of any case where
that'd be triggered other than delibrately. Doing that should make
STRICTENC a soft-fork-safe change, and we can decide at a later date if
we want to get rid of hybrid-encoded pubkeys in a further tightening of
the rules.

-- 
'peter'[:-1]@petertodd.org
000000000000000019b3c625f667bd0b93011c0a37950545a6a8fccf0b08ae73