Skip to contentSkip to conversation

Security Working Group - Wednesday May 13

DISCUSSION Joseph Reynolds 1 participant 2 messages May 12, 2020 – May 13, 2020

lore
Share & tools
lore
#0 lore raw

This is a reminder of the OpenBMC Security Working Group meeting scheduled for this Wednesday May 13 at 10:00am PDT.

We'll discuss current development items, and anything else that comes up.

The current topics:

1. Note: concurrent OCP virtual summit.

2. Discuss SELinux email use cases (email). 3. Experimental bmcweb prototype for authentication rate-limiting. 4. The security wiki now links to OpenBMC's threat models. 5. Are there security considerations in using the fwupd tool? 6. Do we have requirements for BMC admins to disable specific crypto ciphers?

Access, agenda, and notes are in the wiki:

https://github.com/openbmc/openbmc/wiki/Security-working-group

- Joseph

#1 lore raw

Security Working Group - Wednesday May 13 - results

Quoted 7 lines · Joseph Reynolds

This is a reminder of the OpenBMC Security Working Group meeting scheduled for this Wednesday May 13 at 10:00am PDT.

We'll discuss current development items, and anything else that comes up.

The current topics:

1. Note: concurrent OCP virtual summit.

Decided to hold the security working group meeting regardless of concurrent OCP virtual summit.

Quoted 2 lines

2. Discuss SELinux email use cases (email).

Discuss SELinux email use cases - https://lists.ozlabs.org/pipermail/openbmc/2020-April/021477.html

Manoj led us through SELinux email use-cases.  SELinux can help control access so we (OpenBMC contributors) restrict access to those processes that should have access:

(Joseph disclaimer): I tried to summarize the use cases as Manoj walked through them.  This is a summary only.  Please refer to the email chain.

Use cases:

2-1 NBD (network block device) is used to offload dumps on POWER systems.  It reads from PLDM source and writes to NBD device.  Control who has write access to the NBD device.

2-2 All processes running root can write to config files under the /etc directory.

2-3 All processes can control use of systemctl command.

2-4 All processes can write to the journal (journald).

2-5 All processes can use network tcp and udp ports - port-based firewall capability.

2-6 D-Bus service names (paths under /xyz/openbmc_project/*)

2-7 D-Bus message passing - hardening D-Bus communication

2-8 Map existing roles (Administrator, Operator, etc.) to SELinux.  Assign SELinux label based on ingress vector (serial line vs SSH vs REST login).  Uses PAM-selinux config files.

2-9 System calls. For example, disable ptrace.

2-10 Run user executing capabilities.  executables only from specific paths, for example, don’t allow running programs from /tmp

Next steps: Compare AppArmor vs SELinux.  Consider Cost/benefit.

- Prioritize the benefits (as listed above).

- Consider costs including: developer time (person-months), BMC resources (flash size and memory requirements, BMC performance).

- Also consider the cost of future development, for example consider the complications for a developer who is not skilled in security practice: adding a security-relevant function (or entirely new service) or adapting a service for different use cases.  They can either struggle with getting the security controls correct, or accidentally create a security hole.

U-Boot focus: Interaction between SELinux and U-Boot (commands like fw_setenv)?  For example: functions for factory reset and select BMC boot source (such as for BMC firmware update) use the fw_setenv command (or underlying mechanism).  Are these functions the only ones who should be allowed to access the fw_setenv function?

Related side topic: Protect unattended BMC boot.  Can we harden U-Boot itself?

Anton is considering an alternative to apparmor and selinux: KRSI: Kernel Runtime Security Instrumentation

Next steps: prioritize which benefits we want.  Followup in email.

Quoted 1 line

3. Experimental bmcweb prototype for authentication rate-limiting.

https://gerrit.openbmc-project.xyz/c/openbmc/bmcweb/+/31841

Joseph briefly walked through the review and talked about Linux-PAM issues and alternatives like pam_abl.

Feedback was: Looks right.

Quoted 1 line

4. The security wiki now links to OpenBMC's threat models.

No discussion.

Quoted 1 line

5. Are there security considerations in using the fwupd tool?

No discussion.

Quoted 1 line

6. Do we have requirements for BMC admins to disable specific crypto ciphers?

That is, allowing the BMC admin to disable ciphers https://lists.ozlabs.org/pipermail/openbmc/2020-May/021619.htmlThis seems useful, but nobody jumped up to do it.

Drawbacks: implementation cost.  For example: do we try to prevent the admin from removing all supported algorithms so that are none left to connect to?

BONUS AGENDA ITEMS:

7 Requirements for security audit log - https://lists.ozlabs.org/pipermail/openbmc/2020-May/021640.html

Joseph walked through the email.

8 (Bonus topic) We need a better meeting time for India, Zurich, and others.

Joseph revived the email thread https://lists.ozlabs.org/pipermail/openbmc/2020-May/021641.html

Next step is for two people to agree on a meeting time.

- Joseph

Quoted 6 lines

Access, agenda, and notes are in the wiki:

https://github.com/openbmc/openbmc/wiki/Security-working-group

- Joseph