linuxcon na 2015:are today's foss security practices robust enough in the cloud era?
TRANSCRIPT
Lars Kurth Community Manger, Xen Project
Chairman, Xen Project Advisory Board
Lead CentOS Virtualization SIG
Director, Open Source Business Office, Citrix lars_kurth
Was a contributor to various projects
Worked in parallel computing, tools, mobile and now virtualization
Community guy at Symbian FoundationLearned how NOT to do stuff
Community guy for the Xen ProjectWorking for CitrixMember of OSS Business OfficeAccountable to Xen Project Advisory BoardChairman of Xen Project Advisory Board
Think of your vulnerability process as a team-effort to ensure that …
• All doors are closed
• All doors are locked
• All windows are boarded up
• Fences have no weaknesses
• …
Encourage discoverers to report security issues to security@yourproject
Discoverers are in control You can’t stop them from releasing/using informationA robust vulnerability process encourages discoverers to work with you
Ensure that your project fixes security issuesas quickly as possible
You don’t want unaddressed vulnerabilities
ResponsibleF A X
FullF XA
I: Vulnerability Introduced
D: Vulnerability Discovered
R: Vulnerability Reported
A: Vulnerability Announced
F: Fix Available
X: Fix Deployed
Vulnerability exists: we do not know whether it is exploited in the wild
Vulnerability is known about by a privileged and small group of users
Vulnerability is known publicly
R
R
I D
Linux Kernel/LXC/KVM if reported via OSS Security
Linux Kernel/LXC/KVM if reported via [email protected]
OpenStack for low impact issues
Full
Linux Kernel/LXC/KVM if reported via OSS Security Distros
Linux Distributions (both open source and commercial)
QEMU, Libvirt, oVirt, ...
Responsible
“Distro Model”
OpenStack for intermediate to high impact issues
OPNFV, OpenDayLight : process modeled on OpenStack
Xen Project for all issues (also handles 3rd party issues, e.g. QEMU)
Responsible
“Cloud Model”
Not clearly
stated
Docker : states responsible disclosure; but policy docs / CVE pages are empty
Cloud Foundry : no clearly stated process; no published CVE’s
CoreOS: just a mail to report issues
Kubernetes: no information (or could not find it, which is an issue in itself)
Approach Used by Projects
Open-source software projects are often well intended, but security can take a back seat to making the code work. OpenDaylight, the multivendor software-defined networking (SDN) project, learned that the hard way last August after a critical vulnerability was found in its platform. It took until December for the flaw, called Netdump, to get patched …
PC World, March 2015
Wide range of approachesNo consistent Best Practice across projectsNewer projects are lagging behind
Using the pre-dominant model as baselineApplies to Linux Distros, OSS Sec Distros, QEMU, …
Mike Licht @ Flickr
A X
Typically fixed time during which the security issue is handled secretly
Depends on discoverer’s wishes
R: Vulnerability Reported
T: Triage
P: Vulnerability Pre-disclosed
A: Vulnerability Announced
F: Fix Available
X: Fix Deployed
Vulnerability is known by the reporter and the security team
Note: It may also be known and used by black hats
Vulnerability is known about by a privileged and small group of users
Vulnerability is known publicly
Description, CVE
allocation, …
Pre-disclosure period
What can and can’t be done
with privileged information
can differ significantly between
projects
R
Patch/fix creation
and validation
FT P
I personally don't like embargoes. I don't think they work. That means that I want to fix things asap.
Linus Torvalds, 2008
People are less willing sometimes to brush the problem [of fixing security issues] under the mat, and leave it up to vendors that have disclosures, like infinitely long disclosure times.
Linus Torvalds, 2015
Long disclosure times discredit responsible disclosureFrom a few days to many months (recent example: Apple)
Long disclosure times create a disincentive for reporters to work with youIncreases the risk of 0 day exploits
Pre-defined disclosure times help manage vendorsExample later
Most successful projects have a 2-3 weeks disclosure period
The capability to fix issues within the recommended time
Larger and distributed projects can struggle to fix all issues in time
The capability to handle the entire process
in secret
Assigning CVE numbers is best practice in by
established projects and vendors in the
Linux/Cloud ecosystem
CVE databases (such as www.cvedetails.com) can be used to evaluate your project
This shows Xen Project CVE stats
Before 2012, we didn’t have fewer vulnerabilities than after
We just didn’t have a process requiring creation of CVEs
A fair comparison between projects/technologies using CVE data is not easily possible
Not all projects/products create CVEs for all their issues Example: Linux/QEMU only do so for severe onesPolicies are not always published
Some projects don’t assign CVEs at all
Some technologies/products cannot be easily identified in databases Example: KVM, LXC
Sometimes CVEs can affect several productsBut are counted only against oneOpen source product definitions on cvedetails are often sloppy
Description, CVE
allocation, …
A D
Pre-disclosure period
R
Patch/fix creation
and validation
FT P
What happens here depends
on your process goals
Make sure that a fix is available before disclosure
Make sure that downstream projects and products (e.g. distros) can package and test the fix in their environment
Allow service providers that use your Software to start planning an upgrade (at scale this can take a week)
Allow service providers that use your Software to deploy an upgrade before the embargo completes
What is allowed during pre-disclosure
Who is privileged and trusted to be on the pre-disclosure mailing list
Disclosure Time
Make sure that a fix is available before disclosure
Make sure that downstream projects and products (e.g. distros) can package and test the fix in their environment
Allow service providers that use your Software to start planning an upgrade (at scale this can take a week)
Allow service providers that use your Software to deploy an upgrade before the embargo completesCloud Model
Distro Model
Emerged recently!
Recognizes the needs of service providers
Pre-Cloud Computing!
Services and their users are vulnerable
during pre-disclosure period
More Cloud/Service users than direct users of your software
Example:
AWS stated in 2014 that they have > 1M users (and a lot more instances)AliCloud claims that they have > 1M users…
Just imagine what the reputation damage would have been, if Xen had put AWS, Rackspace, SoftLayer, … users at real risk of a vulnerability.
There were 100’s of
stories at the time,
despite the fact that
users were never put
at risk, but merely
inconvenienced !
Pre-disclosure list membership: more members, more risk of leakage
In the Distro Model, the number of privileged users is typically <10In the Cloud Model, the number could be an order of magnitude higher (50-100)This increases risk of information being accidentally released
Restricting pre-disclosure list membership
Restricting membership to large service providers to minimize riskThat creates issues of “fairness” Which may be incompatible with your communities' values
How the Xen Project got to its Vulnerability Process
xenproject.org/security-policy.html
Moyan Brenn @ Flickr
2011 2012 2013 2014 2015 2016
Goals: Allow fixing, packaging and testing; Allow service providers to prepare (but not deploy) during embargo
Pre-disclosure: Membership biased towards distros & large service providersNo predefined disclosure time
1.0
2011 2012 2013 2014 2015 2016
July 2012: CVE-2012-0217, Intel SYSRETAffected FreeBSD, NetBSD, Solaris, Xen and Microsoft Windows
A large pre-disclosure list member put pressure on
key members of the Xen Project Community to get an embargo
extension
They eventually convinced the discoverer to request an extension
1.0
2011 2012 2013 2014 2015 2016
Centered on:
Predetermined disclosure schedule: 1 week to fix, 2 weeks embargo
Who should be allowed on the pre-disclosure listFairness issues between small and large service providersDirect vs. indirect Xen consumersThe risk of larger pre-disclosure list membership
1.0
2011 2012 2013 2014 2015 2016
Strongly recommended disclosure scheduleInclusive pre-disclosure list membership Changes to application procedure (based on checkable criteria)
1.0 2.0
2011 2012 2013 2014 2015 2016
Sept 2014: CVE-2014-7118
Leading to the first Cloud Reboot
AWS pre-announced cloud reboot to their customers
Other vendors didn’t.
Policy was interpreted differently by vendors.
This highlighted ambiguities in the project’s security policy
(what can/can’t be said/done during an embargo)
1.0 2.0
2011 2012 2013 2014 2015 2016
Goals: Allow fixing, packaging and testingAllow service providers to prepare (and normally to deploy) during embargo
Pre-disclosure: Clearer application criteriaPublic application process (transparency) Clear information on what is/is not allowed during an embargo (per XSA)Means for pre-disclosure list members to collaborate
1.0 2.0 3.0
2011 2012 2013 2014 2015 2016
Conducted XSA-133 Retrospective upon requestProcess change: Earlier embargoed pre-disclosure without patches
May 2015: CVE-2015-3456
First time we were affected by a branded bug
QEMU bug, which was handled by several security teams: QEMU,
OSS Distro Security, Oracle Security & Xen Project
From a process perspective: were not able to provide a
fix 2 weeks before the embargo date ended
1.0 2.0 3.0
Larger pre-disclosure list has not caused a single issues in two years of operating an inclusive approach
We have not had a single 0-day vulnerability
A well run vulnerability process builds trust
Willingness to adapt to your stake-holders needs builds more trust
It creates collaboration and understanding of stake-holders
Fairness is a difficult issue
There will always be practical issues, e.g. “interpretations of policy”, etc.
The Xen Project’s process is the only example case, where this issue has been tackled through a community consultation.
To Contrast:
OpenStack does not publish who is on their pre-disclosure list
OpenStack does not have a formal application process
Avoids dealing with the “fairness” issue head-on
Security stories are “hot”
Xen is widely used, thus security stories “sell”
It’s too easy for reporters to write a story
Reporters just have to check our page,
and know when the next story comes
Very wide range of approaches vs.The reality that SW stacks contain many layers Consider the weakest link in your SW stack
Best Practice appears to be emergingOlder projects seem slow to changeNew projects, don’t build security management into their culture from the beginning
New Post-Snowden era pressuresHow to effectively deal with media Hype?