On Tue, Jun 20, 2023 at 07:50:19PM +0200, Theo Buehler wrote: > X509_get_ext_d2i() is one of those very special OpenSSL interfaces... > > It can return NULL for various reasons. If it returns NULL and crit is > not -1, something bad happened. If crit is -2, multiple extensions with > the same OID as the one corresponding to the nid were found (this is not > allowed per RFC 5280, 4.2). It returns NULL if it failed to deserialize > the extension, be it due to bad DER or an allocation failure. In these > cases crit will be 1 or 0, depending on whether the extension was marked > critical. > > So instead of accepting an object in the situation that crit != -1, we > should warn and throw an error. We can't errx() since we can't really > tell allocation failure from failure due to a malformed extension. > > I also added a check for NULL and criticality of basic constraints, > which were missing (per RFC 5280 criticality is optional, so libcrypto > doesn't check that, but RFC 6487 is clear here). > > The warnings in x509*inherits() are minimal. The callers of > x509_inherits() warn. For x509_any_inherits() this is not so. The > annoying bit is that it is used in auth_insert(). I know how I want > to deal with that, but that is largely independent of this diff.
OK job@
