Knowing now that RSA_new_method() copies the RSA_METHOD flags field to the RSA flags field I modified my instance of wpa_supplicant such that it works even without changing OpenSSL. However, I still say that this is a bug with OpenSSL. Let me tell you the short story that leads me to that conclusion.
wpa_supplicant has a bit of code that 1. Allocates and defines a RSA_METHOD structure. 2. calls RSA_new(); 3. calls RSA_set_method(). In that code the rsa_sign and rsa_verify members of the RSA_METHOD structure are not defined, thus making it compatible with the really old versions of OpenSSL. But I need to change that and use the rsa_sign method, so I set the RSA_FLAG_SIGN_VER bit of the RSA_METHOD structure to indicate that I need to use that. But my rsa_sign method was never called. My "fix" is to set that same bit in the RSA flags field before calling RSA_set_method. rsa->flags |= RSA_FLAG_SIGN_VER; But shouldn't OpenSSL work without me having to do that? Perhaps RSA_set_method() should also copy the flags field to match what RSA_new_method does? Either that, or the meth->flags field should be tested. Which is it? -Ike- John Eichenberger Intermec by Honeywell Principal Engineer: Sustaining Engineering 425.921.4507 -----Original Message----- From: openssl-users [mailto:openssl-users-boun...@openssl.org] On Behalf Of Eichenberger, John Sent: Wednesday, April 04, 2018 9:07 AM To: firstname.lastname@example.org Subject: Re: [openssl-users] [External] Re: Correct the check of RSA_FLAG_SIGN_VER Honeywell Internal Your answer #1 below presumes that RSA_new_method() is called AFTER RSA_set_method(). Is that a valid presumption? How is that documented as a requirement? When the flag is set in a call to RSA_set_method() after a call to RSA_new(), the flag gets ignored with the current implementation. -Ike- John Eichenberger Intermec by Honeywell Principal Engineer: Sustaining Engineering 425.921.4507 -----Original Message----- From: openssl-users [mailto:openssl-users-boun...@openssl.org] On Behalf Of Richard Levitte Sent: Tuesday, April 03, 2018 7:43 PM To: email@example.com Subject: [External] Re: [openssl-users] Correct the check of RSA_FLAG_SIGN_VER In message <bn6pr07mb3553b5b834436f685450d434f6...@bn6pr07mb3553.namprd07.prod.outlook.com> on Tue, 3 Apr 2018 22:55:06 +0000, "Eichenberger, John" <john.eichenber...@honeywell.com> said: John.Eichenberger> Honeywell Internal John.Eichenberger> John.Eichenberger> I think I have an answer for why this commit needed to be reverted: John.Eichenberger> John.Eichenberger> Author: Dr. Stephen Henson <st...@openssl.org> John.Eichenberger> Author date: 2 years ago (12/20/2015 10:18:43 AM) John.Eichenberger> Commit date: 2 years ago (12/20/2015 11:27:03 AM) John.Eichenberger> Commit hash: 6656ba7152dfe4bba865e327dd362ea08544aa80 John.Eichenberger> Children: 1c7de36f62 John.Eichenberger> Parent(s): 17592f323a John.Eichenberger> John.Eichenberger> Don't check RSA_FLAG_SIGN_VER. John.Eichenberger> John.Eichenberger> Reviewed-by: Richard Levitte <levi...@openssl.org> John.Eichenberger> John.Eichenberger> The change made in that commit was to simply remove John.Eichenberger> the attempt to check for the RSA_FLAG_SIGN_VER flag. John.Eichenberger> But that's not what is wrong with this code that John.Eichenberger> required changing. The change should be to add John.Eichenberger> "meth" prior to flags: Well, not quite, actually. We can easily study the code prior to this change by looking at the 1.0.2 source. 1. in RSA_new_method(), which is used to create new instances of the RSA structure, there's this line: ret->flags = ret->meth->flags & ~RSA_FLAG_NON_FIPS_ALLOW; So that makes the check of rsa->flags valid, no need to go via rsa->meth 2. In rsa.h (crypto/rsa/rsa.h in 1.0.2), you'll find this comment in the middle of the definition of rsa_meth_st /* * New sign and verify functions: some libraries don't allow arbitrary * data to be signed/verified: this allows them to be used. Note: for * this to work the RSA_public_decrypt() and RSA_private_encrypt() should * *NOT* be used RSA_sign(), RSA_verify() should be used instead. Note: * for backwards compatibility this functionality is only enabled if the * RSA_FLAG_SIGN_VER option is set in 'flags'. */ Do note the note on backward compatibility... you see, there were versions of OpenSSL where the fields 'rsa_sign' and 'rsa_verify' didn't exist (they appeared in OpenSSL 0.9.5), so for the sake of allowing older applications to work with the newer OpenSSL without recompilation, we required all new RSA method implementations to use that flag to have the 'rsa_sign' and 'rsa_verify' functions used. Without that flag, those functions were assumed not to exist, that the RSA method structure was pre-0.9.5. However, this was somewhere in 2000. Fast forward to 2015, when we were starting to make certain types opaque, and someone noticed this flag still hanging around, and we figured that 0.9.5 is long gone, and 1.0.1 was a year away from its end of life, and we figured that the reason to have this flag at all was a matter of years long past, it was time to simply drop its use. It had grown to become irrelevant a long time ago. Cheers, Richard -- Richard Levitte levi...@openssl.org OpenSSL Project http://www.openssl.org/~levitte/ -- openssl-users mailing list To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users -- openssl-users mailing list To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users -- openssl-users mailing list To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users