Closed akoeplinger closed 8 months ago
Tagging subscribers to this area: @dotnet/area-system-security, @bartonjs, @vcsjones See info in area-owners.md if you want to be subscribed.
Author: | akoeplinger |
---|---|
Assignees: | - |
Labels: | `area-System.Security`, `os-linux` |
Milestone: | - |
Weird. This is about as deterministic as a crypto test gets (it uses a known key to decrypt a known value), and it clearly succeeded on retry since the log linked above shows success.
This feels like one of those "the computer said 2 + 2 = 5" never-reproducible bugs... but let's see if we get some more hits to try and force a repro.
@bartonjs where are you seeing this succeed? It fails on every outerloop run on Fedora 38 across both mono and coreclr.
where are you seeing this succeed?
The log you linked to says no failures...
Ah. The log you originally linked to, which is what I got in the email notification. I see the current link has a failure.
Yeah sorry, that was a copy-paste mistake :D
I can reproduce it on Fedora 39 but not Ubuntu 23.04. Odd.
Okay. @bartonjs and I tracked it down to https://github.com/openssl/openssl/pull/13817. This is a change in OpenSSL 3.2 that Fedora / Red Hat back ported.
We are going to disable the implicit rejection so the APIs work as they have always worked.
Re-opening because we probably need to backport this.
@vcsjones We have implemented it in OpenSSL specifically to protect users of such broken API as the one provided by C# and VB.NET.
By disabling implicit rejection you are MAKING them vulnerable
Please, see https://people.redhat.com/~hkario/marvin/ and read it whole, including both of the papers.
Outerloop runs fail on Fedora 38:
https://helix.dot.net/api/2019-06-17/jobs/037f28b7-637f-4634-8ab1-3b4527ffc077/workitems/System.Security.Cryptography.Pkcs.Tests/console