2024-08-08 Meeting notes

Craig McNally 

Axel Dörrer 

Chris Rutledge 

Jakub Skoczen 

Skott Klebe 

• Production systems should be setup with SameSite = Strict or Lax
  
  • Bugfest should be setup in this way, but it's currently set to none.
  • Julian will continue to try to get Kitfox to move this back to Strict/Lax
• Hosted reference envs, and other envs (e.g. rancher envs, etc.) which UI devs might develop against (point a local UI build at) should either:
  • Use SameSite = none OR
  • UI devs should use a proxy like NGINX (running locally)
    • Action: Julian Ladisch to work on a PoC of this.
• Generally speaking it's better to user SameSite=Strict/Lax when possible to help identify issues earlier
• Consider setting the default to Strict instead of Lax.  See https://github.com/folio-org/mod-login
• Zak:  Lax doesn't really buy us anything, we should use either none or strict.  This is because stripes is an SPA.  It's the backend that cares/uses the cookies, the frontend doesn't know anything about these since they're http-only cookies.
  • This comes up when using deep links, e.g. to other domains

• Update from Zak (6/13):
  • WRT LOGIN_COOKIE_SAMESITE env variable in mod-login, Steve Ellis and I chatted and confirmed we could default to Strict instead of Lax without any downside.
  • He will create JIRAs for mod-login and mod-login-keycloak to adjust the default, and if needed we can create another story for devops if they're overriding the default for hosted reference envs.

• Action:  Craig McNally to touch base with Zak on progress of the Jira creation.
• Some feel that the hosted reference envs should be setup like production (e.g. SAMESITE=Strict).  However that interferes with UI developer's ability to point a local build of the UI at these envs.
  • Potential compromise:  configure one of the snapshot envs to have Strict, and the other to have None.

• A new PR ( STCLI-246 - Getting issue details... STATUS , https://github.com/folio-org/stripes-cli/pull/350) introduced a proxy server to assist front-end developers with the samesite issue should they need to work against the Bugfest environment.
• Action: John to try out the new functionality – check on this 25 Jul 2024 

• John Coburn tried this out in stripes-cli and it worked well.  Another developer is going to give this a go as well (Jason S.). 
• What are the next steps?
  • This is documented in the stripes-cli, but not in onboarding
  • Maybe we should raise awareness amongst UI devs
  • Still outstanding.  This will be communicated along with other things.

Today:

• This is looking good, can probably remove from future agendas.
• Only remaining gotcha is how this will work with STSH

We're on track here.  Ian H. applied mitigation.  Upgrade will happen when John Malconian returns from vacation.

• Action:
  • Jakub Skoczen to circle back and check in with John M.
  • Craig posted a reminder in #devops
  • Check in on this next week 01 Aug 2024 
  • Jakub Skoczen informed that the devops team will pick this up soon.

Today

• Craig McNally reached out to John Malconian and Jakub Skoczen for an update today.  
  
  • Response from John: 'I’ll be starting this soon.  Typically will need to be done during “off hours” so I just need to find a good time.'

From Craig in slack:

I've been in communication with David Crossley, Wayne Schneider, John Malconian and Peter Murray about the issue above.  They apparently didn't have access to these embargoed issues (SysOps and Core Team).  Peter shared this screenshot with me, which doesn't look right.  I'd like to review this at one of our meetings and come up with a list of changes/improvements for Peter to make.  A few ideas off the top of my head:

1. Add descriptions to each of the security groups, like we have for "FOLIO Security Group"
2. Maybe add a new security group and level for FOLIO devops
3. Review membership of each of these groups and remove users no longer on the project
4. Review the Security Level -> Group mappings.  Some of these don't look quite right to me.

https://files.slack.com/files-pri/T1EPYQDAQ-F075P5YMBJQ/image.png

• If it makes this easier, we could invite Peter to a meeting so we can see the groups/levels interactively and makes adjustments as we go

TODO:

• Create a new group:  devops, and assign Wayne, John, Ian, David, Peter.  Also create a new security level for "folio devops"
• Remove Johannes Drexl from the SysOps group, he's not longer involved in Folio.
• Action: Craig to look into how security level configuration works to gain a better understanding of why it behaves the way it does.  
  • Why do we have All Folio Developers in the "Sys Ops and Core Team" security level?

Today:

• No update today- Check in on this 22 Aug 2024 

• We need to determine if DevOps can work on  FOLIO-3896 - Getting issue details... STATUS  or if they don't have bandwidth, and the debian packages aren't used anymore, archive them.  It sounds like some SysOps do use this, but we know that DevOps has very little bandwidth these days.
• Action:  Create a Jira for folio devops to set Strict Transport Security header in hosted reference environments.  Then link the story to SECURITY-170 - Getting issue details... STATUS
  • Craig McNally will create the story and link it up.
• There is potential for a conflict between Strict Transport Security Header and the proxy approach used for cookies with SameSite=Strict.  We need to think through how this will work.
  • Problems stem from UI devs hosting dev builds of the UI locally (e.g. http://localhost:3000), and pointing to a hosted backend (e.g. folio-snapshot).
  • Maybe we can make the proxy listen on https (using a self-signed cert)
  • Jens proposed the idea of rewriting the cookie domain in the proxy
  • What about CORS though?
  • Action: John Coburn will discuss this with the dev who implemented the proxy stuff in stripes-cli.
• Discussed  CIRCSTORE-487 - Getting issue details... STATUS .  
  • This isn't a problem for master branch (ramsons) nor quesnelia (v17.2)
  • The remaining work here is:
    • Get the fix applied to the 16.0 branch of mod-circulation-storage and cut a release
    • See approval for inclusion of this fix into Poppy CSP7
      • Question:  Should this necessitate a Poppy CSP7?  Alternatively, maybe we get this in there and if there's something else which requires another CSP this will go out along with it.
      • Needs to be discussed with Julian Ladisch.  There's already a slack thread in our channel.

Today:

• TODO: Need to review/triage new SECURITY issues reported (found via PRISMA scan):
  • SECURITY-171 - Getting issue details... STATUS   -  SECURITY-179 - Getting issue details... STATUS
• Stephanie Buck has reached out about  CIRCSTORE-520 - Getting issue details... STATUS  asking what to do since it's missing the "security-reviewed" label

From slack conversation, I think I've gathered the following:

• In this case (bank account and transit numbers), the information is highly sensitive.  
  • Highly sensitive information should:
    • Be stored in it's own table
    • Accessed via a dedicated API
    • Protected by a dedicated permission
    • Encrypted in the database, not only on disk.  
      • This could mean either:
        • Explicitly encrypting/decrypting in the application layer and storing the encrypted data in postgres
        • Utilizing postgres extensions/modules to encypt certain columns, e.g. via pgcrypto (see https://www.postgresql.org/docs/current/encryption-options.html)

Let's review and discuss before providing this feedback to Raman.

Axel Dörrer also suggested that defining classes of sensitivity could help teams determine which techniques are applicable in various situations.  I agree having some general guidelines on this would be helpful.

• regular data
  • low sensitive - permission based on same API
  • high sensitive - permission based on dedicated API

It would probably help to provide concrete examples of data in each class.  This can be a longer term effort, we don't need to sort out all the details today.

Axel Dörrer 

Due to some absences on different reasons the group stalled. Axel will try to reactivate the group.