Episode 122

Posted on Friday, Jul 2, 2021
This week we look at some new Linux kernel security features including the Landlock LSM and Core Scheduling plus we cover security updates for RabbitMQ, Ceph, Thunderbird and more.

Show Notes

Overview

This week we look at some new Linux kernel security features including the Landlock LSM and Core Scheduling plus we cover security updates for RabbitMQ, Ceph, Thunderbird and more.

This week in Ubuntu Security Updates

46 unique CVEs addressed

[USN-5004-1] RabbitMQ vulnerabilities [00:44]

  • 2 CVEs addressed in Xenial ESM (16.04 ESM), Bionic (18.04 LTS), Focal (20.04 LTS), Groovy (20.10), Hirsute (21.04)
  • AMQP server written in Erlang
  • Possible to cause the server to consume excessive memory by sending large values in the X-Reason HTTP header - resource exhaustion - DoS
  • Possible infinite loop - failed to perform sufficient validation - DoS

[USN-4998-1] Ceph vulnerabilities [01:38]

[USN-4995-2] Thunderbird vulnerabilities [02:22]

[USN-5000-2] Linux kernel (KVM) vulnerabilities [02:48]

[USN-4997-2] Linux kernel (KVM) vulnerabilities

Goings on in Ubuntu Security Community

Landlock released in 5.13 kernel [03:49]

  • Allows unprivileged processes to sandbox themselves - currently only supports file paths - so can specify read/write of files/dirs etc
  • Took 34 revisions of the patch set and it evolved significantly over time - was originally based on attaching BPF programs to LSM hooks but given how fraught unprivileged BPF has been this was NACKd and instead went with a new approach based on a custom API with brand new system calls to support it
  • API is quite low-level compared to say how AppArmor policy is specified so will be interesting to see if there becomes a liblandlock in the future to make this kind of thing easier (cf. libseccomp for doing seccomp BPF programs etc)
  • https://lwn.net/Articles/859908/
  • https://landlock.io/

Core Scheduling merged for 5.14 kernel [06:43]

  • SMT siblings share lots of microarchitectural state like L1D cache etc - various micro-arch attacks could only be mitigated across different SMT cores - so processes which shared the same core could snoop on each other (eg. L1TF - in the context of virtualisation, a malicious guest VM could snoop on the L1D contents of another VM on the same SMT core) - so the only option was to disable SMT which brings a big performance hit
  • Solution is core scheduling - ie. make the schedular aware of and respect SMT threads on the same core
  • Tag processes via cgroups - this defines the trust boundaries - processes in the same tagged cgroup share a trust boundary and can be scheduled on sibling SMT cores - and by default all processes are in the same group
  • Uses prctl() to allow setting / copying these - and can only set these on processes which you can ptrace
  • https://lwn.net/Articles/820321/
  • https://www.phoronix.com/scan.php?page=news%5Fitem&px=Core-Scheduling-Linux-Close

Get in contact