| opendevreview | Merged openstack/governance master: Mark SRBAC goal completed https://review.opendev.org/c/openstack/governance/+/1009912 | 00:23 |
|---|---|---|
| mikal | gouthamr: noting that the broad capabilities was a deliberate design decision taken at the time by Nova. It was not the originally proposed implementation. I assume that has been discussed and I just didn't notice? | 18:33 |
| gouthamr | mikal: i am unsure if that was discussed (or evolved over time).. the team was at some point looking at modifying nova to use multiple privsep contexts: https://blueprints.launchpad.net/nova/+spec/privsep-usage-review | 18:40 |
| gouthamr | work's stalled afaics.. | 18:40 |
| gouthamr | https://review.opendev.org/c/openstack/nova/+/871729 | 18:40 |
| mikal | gouthamr: the initial proposal was more granular, but was walked back at a PTG because of concerns about the stability of the set of capabilities required for given calls in the kernel. IIRC it wasn't a documented / stable interface at the time. Honestly, if not everyone has transitioned at all surely getting that done is more important than tuning | 18:52 |
| mikal | the exact capabilities in projects which have adopted it? | 18:52 |
| gouthamr | I agree; we’ll get the stragglers addressed this release | 18:54 |
| gouthamr | ill nudge Oslo folks signal oslo.rootwrap’s deprecation - it’s informally spoken about so far | 18:55 |
| sean-k-mooney | gouthamr: yes at some pont btu that kind of stalled out | 20:15 |
| sean-k-mooney | but we are also usign some legacy capabity defeingion that are semi deprecated in the kernel too | 20:15 |
| sean-k-mooney | i tried to remove CAP_DAC_OVERRIDE in the past, nova only uses it for one lvm interaction at elast wehn i last tried to remove it an di had a plan to do that but never got time to finish it | 20:17 |
| sean-k-mooney | https://www.man7.org/linux/man-pages/man7/capabilities.7.html | 20:17 |
| sean-k-mooney | alsome all project that have adopted it have some combination of CAP_DAC_OVERRIDE CAP_SYS_ADMIN and CAP_NET_ADMIN | 20:18 |
| sean-k-mooney | wich is almost but not quite full root | 20:19 |
| sean-k-mooney | https://github.com/openstack/nova/blob/master/nova/privsep/__init__.py#L21 | 20:21 |
| sean-k-mooney | https://github.com/openstack/neutron/blob/7157cab2caf48d0b3a7fe0ab9f0f09fa7e109c2c/neutron/privileged/__init__.py#L18-L73 | 20:21 |
| sean-k-mooney | neutro is slightly better. | 20:21 |
| sean-k-mooney | but cidner also has the same permisions as nova https://github.com/openstack/cinder/blob/master/cinder/privsep/__init__.py | 20:22 |
| sean-k-mooney | i will try and eventually audit cybrogs usage and see if we can drop https://github.com/openstack/cyborg/blob/8f58f18b90de98a047dabc93a7c5f41b0a4ae656/cyborg/privsep/__init__.py#L20-L34 | 20:24 |
| sean-k-mooney | but ya movign to privesep by itself is not more secure the is not nessiarly any more secure then just using raw sudo without a lot of work on the project to carfully limit scope | 20:26 |
| sean-k-mooney | the problem is that each privsep context has its own deamon and need to load the prouces code into memroy in a new python interpertry to service it request | 20:26 |
| sean-k-mooney | so its a memory vs secuirty vs performace trade | 20:27 |
| opendevreview | Merged openstack/openstack-manuals master: Fix www/project-data/latest.yaml https://review.opendev.org/c/openstack/openstack-manuals/+/1011328 | 20:48 |
| opendevreview | Merged openstack/openstack-manuals master: Update 2026.2 docs page links https://review.opendev.org/c/openstack/openstack-manuals/+/1011329 | 20:49 |
| opendevreview | Merged openstack/openstack-manuals master: [www] Set 2025.1 Epoxy state to Unmaintained https://review.opendev.org/c/openstack/openstack-manuals/+/1013916 | 21:12 |
| gouthamr | sean-k-mooney: yeah, well put.. i think there's a bunch of compromises/design decisions that project teams took considering that trade-off (or sometimes not)... a couple of thoughts around this, i would love to have some guidelines in the docs themselves for developers - and maybe the per-project threat model that we sought will surface what the project decisions looked like, and any pending tasks they imagined.. | 21:23 |
| gouthamr | right now, the old spec https://specs.openstack.org/openstack/oslo-specs/specs/liberty/privsep.html is more comprehensive than the official docs | 21:24 |
| fungi | JayF: worth noting, enisa is making references to "security.txt" files in https://www.enisa.europa.eu/sites/default/files/2026-10/Technical%20Advisory%20on%20security.txt%20CVD%20Policies%20and%20Security%20Advisories%20Public%20Consultation.pdf which might be worth pointing out to your github contact too | 21:56 |
| JayF | ack, ty | 21:57 |
| fungi | though it's more in context of web sites/applications e.g. https://yourdomain.example/.well-known/security.txt rather than in git repositories specifically | 22:00 |
| gouthamr | seems like they are standardizing on https://www.cisa.gov/news-events/news/securitytxt-simple-file-big-value ? | 22:00 |
| gouthamr | https://www.rfc-editor.org/info/rfc9116/ | 22:00 |
| fungi | gouthamr: yes, it makes references to ietf rfc 9116 for that | 22:02 |
| fungi | i don't think enisa is going to directly recommend a cisa directive, that's from a completely different (and potentially antagonistic) jurisdiction | 22:03 |
| gouthamr | yeah :P i was just curious | 22:03 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!