Hi Matt, thanks for your prompt answer. The testclient is using openssl-1.0.2m and my previous assumption that the IV was derived from the key_block as in TLS 1.0 was wrong. It seems that when initialising the cipher with the IV is ignored when the crypto is AES and the first 16 bytes of the encrypted payload is used as IV (as per TLS spec). This is a Java issue and has nothing to do with OpenSSL but assuming this is correct behaviour (I'll have to dig in to that) I use the same logic (in this respect) for TLS1.1 and TLS1.2 assuming that the client also does which it obviously does not.
Below is the logging from my server handling two authentications (the Finished message that is the first encrypted). The first (TLS1.1) succeeds and the second (1.2) fails. You can see that the decrypted text is correctly padded in the first case (TLS 1.1) indicating that the decryption succeeded but the second (1.2) decrypted buffer does not. The "Finished" message for this (AES; SHA) mac- and block size should be padded with 11 bytes which the first is. Also as seen I use the same provider and algoritm in the two cases, actually I handle the decryption in the same way for the two TLS-levels Also added the wireshark output for the two authentications where I've verified that the logged encrypted data really is whats coming in on the line. Basically when selecting 1.1 in the client everything works as expected but selecting 1.2 the prepended IV is not correct as far as I understand. The testclient vendor says they are using OpenSLL for TLS handling so I expect that they're not fiddling with any IV bytes at all. Given the version 1.0.2m, can this be explained? Best Curt
Description: Binary data
TLS 1.1 04:09:40.295 Cipher provider: SunJCE 04:09:40.295 Cipher algoritm: AES/CBC/NoPadding 04:09:40.295 IV from key_block used in cipher init: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] 04:09:40.295 Encrypted: [f5, 21, 77, 27, f7, bf, 31, 1f, b9, 74, 14, 50, be, 5f, 66, 21, 1b, 4e, b, 33, 78, 3b, b8, 31, a3, 7c, c2, 93, f6, ec, a, 8f, c4, 28, 71, cb, 82, b4, 12, c2, 6b, 56, f2, 9e, c8, b0, 2b, 64, 7c, 89, ef, bb, 68, 6b, 73, 6b, 80, 3a, 1b, 7, 33, 4e, 36, 6b] 04:09:40.296 Decrypted: [a6, c7, bc, 87, 7, 2, 85, c, c3, c7, 91, 73, b3, 85, 19, 21, 14, 0, 0, c, 1, 6f, 1e, 4e, 97, d0, c4, 10, a9, 35, 37, bc, 2a, 87, 79, 78, 4c, 4, 88, a1, fd, 35, 42, 26, 56, 51, b3, 36, d, 3b, be, 4a, b, b, b, b, b, b, b, b, b, b, b, b] TLS 1.2 04:09:47.779 Cipher provider: SunJCE 04:09:47.779 Cipher algoritm: AES/CBC/NoPadding 04:09:47.779 IV from key_block used in cipher init: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] 04:09:47.779 Encrypted: [80, 83, 98, 6b, 55, 98, 73, b4, f6, 22, 2d, 88, b, 4d, af, 6f, 2d, d6, 1e, 9, 1d, b1, 94, f4, 49, 20, 15, d2, a0, a0, 35, c5, 95, 3d, d3, 35, fd, 92, ef, a6, 0, 7d, 49, 3, 70, 5e, 5a, 57, e3, f9, 89, 3f, 83, 11, cf, 82, 4e, a3, 87, 6e, 9e, 97, 9c, 7c] 04:09:47.779 Decrypted: [f1, 4e, b3, 18, 26, d7, ae, 12, bf, 5d, fe, 2a, 18, a1, 6f, 19, 60, 73, 26, c8, 74, 7c, c6, a6, a1, 65, d3, ad, 45, f9, e3, 42, aa, e0, bd, 1d, a0, 18, b3, f3, 94, 28, 4e, 2a, ca, 24, 25, 70, 7c, a9, f6, 19, 17, f5, ef, ee, b, 30, 2f, ec, 4d, b3, 9c, 8b] > On 14 Feb 2018, at 0:26 , Matt Caswell <m...@openssl.org> wrote: > > > > On 13/02/18 22:02, Curt Johansson wrote: >> Hi all, >> >> I'm developing support for TLS 1.1 and 1.2 in a radius-server that until now >> only handles TLS 1.0. I'm testing with a testtool that the vendor says is >> using OpenSSL to implement the TLS support. It all seems to work except for >> the following: >> >> When all key exchange messages are sent from the server and back from the >> client the client sends the "Finished" message which >> is the first encrypted with the negotiated symmetric cipher suite. I use >> AES-128 in block mode and according to the spec (RFC4346 for TLS 1.1) >> the IV is prepended to the encrypted message (containing the payload, MAC >> and padding). The message size i right and when (in the server) I use the >> first 16 bytes >> of the message received from the client as IV the decryption fails but when >> I use the mechanism from TLS 1.0 to pick up the IV from >> the key_block the decryption is successful. I understand that this is one >> way to generate the IV that I suppose you use which is fine but >> shouldn't this 16 byte vector be prepended, unecrypted, to the encrypted >> data that is sent to the server? In TLS 1.2 there is not IV material >> generated >> at all in the key_block so in that case I don't even know where to find it. >> >> The simple question is, shouldn't the first 16 bytes (assuming AES) of the >> message (after the 5 byte header) be the unencrypted IV to be used in the >> decryption of the rest of the message? > > Yes, assuming you have negotiated an AES CBC ciphersuite in TLSv1.1 or > TLSv1.2, then that is what happens. You don't say what version of > OpenSSL you are using. Here is the code that does it for 1.1.1 (i.e. > master branch): > > https://github.com/openssl/openssl/blob/master/ssl/record/ssl3_record.c#L954 > > As can be seen on line 969 we just fill the IV with random bytes. > > Perhaps you could provide a wireshark trace of the handshake which might > provide some enlightenment as to what is happening. > > Matt > > >> >> I tried to dig in to the OpenSSL source but it's far too long ago I did some >> serious C coding so I hope someone with a working knowledge can enlighten >> me. I might have >> misunderstood the spec but in that case I would be grateful if someone could >> clarify this specific part of it. >> >> TIA >> Curt Johansson >> > -- > 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