Re: [PATCH] KEYS: X.509: Fix Basic Constraints CA flag parsing

[Lukas Wunner <lukas@wunner.de>]

On Fri, Sep 12, 2025 at 02:14:49PM -0700, Fan Wu wrote:
> On Fri, Sep 12, 2025 at 6:14 AM Lukas Wunner <lukas@wunner.de> wrote:
> > On Thu, Sep 11, 2025 at 10:53:56PM +0000, wufan@kernel.org wrote:
> > > +++ b/crypto/asymmetric_keys/x509_cert_parser.c
> > > @@ -623,7 +625,7 @@ int x509_process_extension(void *context, size_t hdrlen,
> > >               if (v[0] != (ASN1_CONS_BIT | ASN1_SEQ))
> > >                       return -EBADMSG;
> > >               if (vlen < 2)
> > >                       return -EBADMSG;
> > >               if (v[1] != vlen - 2)
> > >                       return -EBADMSG;
> > > -             if (vlen >= 4 && v[1] != 0 && v[2] == ASN1_BOOL && v[3] == 1)
> > > +             if (vlen >= 5 && v[1] != 0 && v[2] == ASN1_BOOL && v[3] == 1 && v[4] != 0)
> > >                       ctx->cert->pub->key_eflags |= 1 << KEY_EFLAG_CA;
> > >               return 0;
> > >       }
> >
> > Your patch is correct, however the conditions ...
> >
> >   vlen >= 5 && v[1] != 0 && v[2] == ASN1_BOOL && v[3] == 1
> >
> > ... all check well-formedness of the BasicConstraints object,
> > so it seems if any of those checks fails, -EBADMSG should be returned.
> >
> > The check "if (vlen < 2)" could be changed to "if (vlen < 5)" because
> > 5 bytes seems to be the minimum size of a well-formed BasicConstraints
> > object.  Then the "vlen >= 5" and "v[1] != 0" checks can be dropped.
> 
> Actually, we need to be careful here. OpenSSL produces
> BasicConstraints with CA:FALSE as just an empty SEQUENCE:
> 
> 06 03 55 1d 13 | 01 01 ff | 04 02 | 30 00
> [----OID------] [critical] [OCTET] [empty SEQ]

I see, thanks for the explanation.

This behavior of OpenSSL doesn't seem spec-compliant, or is it?
RFC 5280 sec 4.2.1.9 says the pathLenConstraint is optional,
but the cA boolean is not optional.  Is there a rule that booleans
need not be rendered if they are false?

BTW, I note that X.690 sec 11.1 says that for DER encoding,
all bits of a "true" boolean must be set, hence the 0xff value.
But I'm fine with your more permissive approach which checks for
a non-zero value, hence also allows BER encoding per X.690 sec 8.2.2.

Thanks!

Lukas