MAC integrity failure
rapier1 opened this issue · 5 comments
mac hmac-ripemd160-etm@openssh.com is failing to pass the regression tests for the KitchenSink version of the code. There seems to be some sort of odd interaction that wasn't happening in 7.1p2. I believe it's some issue between the code for the none cipher switch and the aes_mt_ctr cipher. The aes-mt-ctr cipher works fine on it's own, it's only when that code set is incorporated with the null switch that we seem to have problems. As a work around please do not use ripemd160-etm until this issue is fixed.
Well, it's not an interaction issue with the None cipher code as I see the same problem when I merge the AES-MT-CTR into the DynamicWindow tag. This is not an issue when only the AES_MT_CTR code set is applied.
More data: If the integrity test is forced to use other ciphers then it passes as expected so it's specific to the way that the aes-ctr-mt cipher is being used. The problem is that I'm not sure how the integrity tests are being run so I need to look into this more deeply. If anyone has any suggestions please speak up.
I forward ported the patch to openssh-7.3p1 and this failure does not happen for me (dropping the server logging stuff). I suspect that it must be an upstream issue rather than something in the hpn patchset.
I'll have to check that out. Also, thank you for the patches for CTR MT.
i've pushed updates for 7.3p1 to the repo and the tests are passing for me now