Hacker News Reader

> This is an architectural decision

It's also a business decision. You need to invest staff time in testing and automating said updates, and will inevitably lead to more outages. Thus less profits, directly contrary to your assertion. Managers have trade-off between a small but relatively predictable cost, and a large but presumed highly improbable cost. This is something that neither data nor human intuition are particularly good with.

Best case scenario insurance lets you convert your large risk into a small predictable cost, and the premium difference decides whether you stay on the CVE treadmill or not. But its not like actuaries have a lot of data to model to the point of accurately imposing a cost per day of unpatched CVE.

Worse, many places have competitors, and if they're choosing to take on risks in the name of lower prices and higher market share, the customers of your secured app walk out the door. Not only do you need to invest in security, you need the competition to as well. I have a few ideas on how to achieve this, but they're not historically palatable to the HN crowd.

I think it's worth it when the alternative are multi billion dollar lawsuits and government regulation.

> It's also a business decision. You need to invest staff time in testing and automating said updates, and will inevitably lead to more outages.

There are many vendors attacking this exact problem from many directions, including my employers[0], but pretty much every platform vendor has something here.

I do agree that automation and risk are at the root of the tension between security and stability. It's a feedback loop that can run in two directions. If you embrace stability, the loop becomes "no changes ever" and security becomes much more difficult to achieve in practice. If you embrace security, then at first you face instability -- but embracing rapidity and building out tooling for it makes you both more secure and more stable in the long run.

> But its not like actuaries have a lot of data to model to the point of accurately imposing a cost per day of unpatched CVE.

I'd be interested to know if this is actually true.

I recently got interested in the FAIR risk modelling approach. It requires a bunch of legwork, so in practice it won't catch on without determination to use it or something like it.

Usefully it breaks risk down into more easily-estimated, easily-tracked subcomponents[1]. "Probability" becomes "Loss Event Frequency", which further breaks down into Threat Event Frequency and Vulnerability, these break down further in turn. Similarly, the usually-vague "Impact" becomes Primary Loss and Secondary Loss, each of which has 6 specific types that can be estimated. Those estimates can be based on available data, whether your own or industry statistics.

> Worse, many places have competitors, and if they're choosing to take on risks in the name of lower prices and higher market share, the customers of your secured app walk out the door.

[0] I work for Pivotal. We've been building towards the vision of "Rotate, Repair, Repave" in Cloud Foundry: https://builttoadapt.io/the-three-r-s-of-enterprise-security...

[1] https://cdn2.hubspot.net/hubfs/1616664/The%20FAIR%20Model_FI...

Interesting points, but you seem to have missed responding to your last > quote. The "> Worse, many...". I'd be interested to see what your remarks are there.

Ah, good catch, and this is why I should wait for caffeine levels to peak before screeding.

At work we have a lot of customers now, many of them with large security orgs. Their approaches vary a lot, presumably because of their experiences and security maturity.

Beyond a certain level, there is always a security org and quite typically it has veto powers that override any local manager's budgetary concerns. Whether that veto is applied well is quite distinct from the way it creates a ceiling over potential spending on features.

That all said, I agree that there are fundamental properties of security that will cause it to be consistently under-attended relative to what a more actuarial or financial model would suggest is rational. It is only observable in its absence, but most feedback systems -- formal or informal -- rely on observable signals. Market releases may lead to a juicy bonus at many companies. Invisible improvements won't.

I've thought about this a lot in the past few months and published a bunch about it internally at Pivotal. I'm not alone, of course. We're streets ahead of many vendors, but I also feel we're only getting started on the deeper and more difficult process of blending it deeply into our cultural DNA. It's a culture which believes in doing the right thing, but also a culture which detects vetoes as faults and routes around them.

Interesting times.

> You need to invest staff time in testing and automating said updates, and will inevitably lead to more outages.

It should follow the same processes as any other application update. Are you suggesting it's impossible to push updates to software without incurring outages?

Judging by the number of banking and government institutions that have weekly or nightly planned outages, there's appears to be a number of legacy systems where companies certainly think upgrades mandate outages. Even a non-zero probability of an outage is a cost.

Obviously this is fixable, but it's no free lunch.

The problem is that a lot of these systems aren't legacy. They've been built within the past 5 years, but the customers would almost rather continue on with their weekly outages than apply basic modern architectural designs that allow for functionality like round-robining updates behind a LB.

> PATCHING OF KNOWN CRITICAL SECURITY VULNERABILITIES IMMEDIATELY.

I worked in system admin for many years. If anyone was actually foolish enough to do this, they would be dealing with multiple major outages every year.

> for every complex problem there is a solution that is simple, obvious, and wrong.

"There might be an outage" is the stupidest reason I have ever heard to advocate not applying security patches.

There's plenty of companies that are able to regularly patch and upgrade software, incident free. If you team isn't able to do this, they need to fix their process instead of putting their head in the sand.

> stupid

I am not defending the totality of EFX's process by any means. I am just pointing out that "install every high severity patch right away" is not as easy as it looks.

Others have pointed out this patch was not available for all current versions, for example. So, to install the patch you need to upgrade. Oh, that breaks <dependency>. So we need to upgrade <other thing>. But that regresses a feature we use ....

The major issue with EFX was the fragility of their overall architecture. A single weakness should not be enough to lay the whole DBMS open.

The tech industry has to move more towards declarative systems, rather than procedural ones. Docker, Kubernetes and similar tech can auto-update in the background without any downtime. It's impossible to maintain thousands of VMs without some sort of declarative system.

Each thing you are listing increases complexity and attack surface.

It actually decreases complexity. Updating 1000 VM servers using these technologies can be made automatic. True, it provides an additional attack surface, but the gains of having everything always patched are well worth it.

Everything does not include your actual app, so yes it's convenient but doesn't really help that much

I'd argue that convenience is a security feature. The easier something is to do, the fewer mistakes you or your team will make.

Stand by my initial post, declarative systems will become more and more popular and will improve security.

What is the non-tech industry supposed to do?

There are great reasons for not installing patches right away.

"There might be an outage" is not one of them.

I don't recall being at, or working with, a company where this wasn't _the_ reason patches were delayed at least some period if time.

Companies are driven by profit; unknown patches can negatively affect profit and are usually treated as such.

Alternatively, there are sometimes political reasons. A director may not want to risk an outage for fear of being dressed down by their VP.

Any change to a production environment comes with risks, sometimes having a security risk is judged to be lower risk than making a change.

I've seen this from small startups to multi-billion dollar companies.

Ignoring these factors is also putting your head in the sand.

> Alternatively, there are sometimes political reasons. A director may not want to risk an outage for fear of being dressed down by their VP.

This is such a weird thing to me. I've never worked anywhere like this and so I have a hard time imagining it really. Wouldn't that same director get more than dressed down by the VP if on their watch a _known security vulnerability_ was ignored and the system was breached? I just can't understand this way of thinking.

It's easy to understand once you get just how simplistic it is. The VP will deliver a dressing down if the company loses money and the blame can be placed on the director.

The cost of an outage is very easy to quantify (revenue per minute the system is down), and the probability that something will go wrong while applying the patch is also somewhat easy to predict, and usually greater than zero. The director will be blamed with certainty for the outage, since he approved it.

The cost of a security breach is difficult to quantify; it depends on what gets breached and how bad. Note here, I say breach, not vulnerability. Even if there is a known security vulnerability, it's not immediately obvious in all systems what the consequence will be; there may be other mitigations in place outside of software that reduce the potential damage, or there may be unknown vulnerabilities that are exploitable due to the known vulnerability that make would make a breach worse. The lack of certainty about the consequences means it's also possible for the director to avoid blame if the breach is minor ("how was I supposed to know that other team is still using MD5?"). If there is no breach, then there is nothing for the director to be blamed for.

Given that the director would like to avoid being dressed down, director will be more inclined to delay patching over possibly causing an outage, because the costs of an outage are easy to predict and he will take all blame for it. The breach may never happen and even if it does, it may cost him personally less than the outage.

If this still seems weird, it might be because you are someone who views patching as an easy thing to do, because you probably work for a software company. Software companies are used to managing changing software, and have all kinds of practices around minimizing the risks of doing so. Non-software companies typically find patching to be hard and costly because their core business is something else; changes can disturb the "something else."

While you may sort of have a point for low-severity issues (if you patch every one ASAP, you are constantly restarting; should be tested in a group in stage and released to prod on a regular timeline), this was a 10.0 severity CVE and it is _definitely_ the kind of issue that you tolerate some downtime for. Check out the CVE, this could be exploited by anyone with a basic knowledge of cURL in 60 seconds or less. The vulnerability was just a matter of sending an arbitrary command in an HTTP header.

You need to do some testing of course. "IMMEDIATELY" doesn't mean within 5 seconds. If you can't properly apply security updates in a reasonable time frame without it leading to major outages, something is extremely wrong with either the architecture, the chosen software or the QA and deployment strategy.

> without it leading to major outages

I think I would have preferred an outage over leaking millions of american's private information.

> I think I would have preferred an outage over leaking millions of american's private information.

With full 100% hindsight vision, of course.

It's not the first time this has happened to a company, and it surely won't be the last. We've already had full 100% hindsight vision a dozen times over, and plenty of companies have chosen to ignore it.

- PATCHING OF KNOWN CRITICAL SECURITY VULNERABILITIES IMMEDIATELY.

Immediately? No real company can do this. 1-2 weeks after the announcement is the best you'll see, even with something of the magnitude and publicity of Heartbleed. I'm not saying that's how it should be but that's how it is. Most organisation's it's quarterly!

All of these can be designed into a system and still maintain 100% uptime

Sure. If you were designing from a clean slate right now. The vast majority of organisations in the world aren't.

And believe me: in 10 or 20 years your new design would have revealed flaws too, and look laughably primitive to the practitioners of the day.

What's the solution then? An architecture in which no single compromise can compromise the entire system. That's getting harder and harder to do as the landscape homogenises around Lintel.

> No real company can do this. 1-2 weeks after the announcement is the best you'll see

You don't need to patch your software right away, but you should have other strategies for mitigation such as a WAP in front of your services that you can configure to block malicious traffic when you identify a remote code execution vulnerability.

Or two, by two different vendors, ideally sharing no code whatsoever.

release: https://nvd.nist.gov/vuln/detail/CVE-2014-0160 (04/07/2014)

patch: https://aws.amazon.com/security/security-bulletins/heartblee... (4/8/2914)

IIRC: 12 or 18 hours rolled out to all of aws ELB (main part affected).

https://usn.ubuntu.com/usn/usn-2165-1/

Patch was available at time of vulnerability release for current and LTS releases of ubuntu. apt-get update.

But you know... That's just too hard.

Impressive. Let me amend my original statement from "no" to "a tiny minority" then but I believe my point stands - frantic patching to keep up is no substitute for an architecture with no single point of vulnerability.

If you're on an computer there is always going to be a single point of failure with a window of venerability. Any piece of software or combination thereof will possibly leave a RCOE veunerability, which means you always have a single point of veunerability. I think what you mean is, is the unauthorized access isolated enough such that by the time an attacker finds the next thing to attack you've noticed the odd network behavior, eg defense in depth? If so, yes. But theres a difference between frantic patches and a well controlled, tested and easy to deploy set of software which makes pushing updates much less "frantic" and much more controlled. No matter what your defense in depth is you're always going to get an "oh fuck we are totally uncovered" moment, this is where the agility helps out. Your other option in this case is "the emergency shut-down for the oasis", well or the equifax option of sticking your head in the sand and running the servers open for months. I assume we're all responsible adults here though so that last one is off the table, it all comes down to risk assesment. ef seems to have miscalculated on that one.

We need to differentiate then between "patch for this one specific thing" and "upgrade to the next version" which are often conflated. The former could literally be a one-line change to correct an off-by-one error say. The latter will have to go through the full regression and release process. Often it's a grey area.

Companies like Red Hat, Suse and Ubuntu provide security backporting for exactly this reason. It's one of the major reasons that folk subscribe to them.

Unless you are part of that "tiny minority", you have no business handling information that can substantially negatively impact people's lives.

Isn't it interesting that one needs a license to drive an automobile and yet companies that handle this type of information get to do so without demonstrating the required level of competence.

Isn't it interesting that one needs a license to drive an automobile and yet companies that handle this type of information get to do so without demonstrating the required level of competence

Oh, it's the whole industry tho'. VMS running on DEC hardware was at a level of security and reliability that we can only dream of today, but it was abandoned on the scrapheap and replaced by glorified PCs running a hobbyist OS. If you wanted to build an infrastructure on old-skool principles now you couldn't, you can't get the components anymore. The truth is we're fighting a losing battle and the real enemy is ourselves.

> No real company can do this.

They can, but it requires automation. I've worked on and used tooling for just such things. We can see an upstream CVE and have it patched in production inside 48hrs. Usually the same day.

Disclosure: I work for Pivotal. All this stuff (Cloud Foundry, BOSH etc) is opensource, but we sell commercial versions too.

seriously, FFS these are guidelines that we use for not-remotely-that-sensitive-data and equifax that literally... "You had one job to do, protect our data" Can't follow?

This is just business being a) ignorant b) incompetent c) careless one or more of the above applied for a hack like this to happen. It's really sad that even ycombinator crowd doesn't get it, given that the crowd here is significantly more technical and sensitive to these issues than the average Equifax Joe..

No, we get cases A, B, and C, because those cases are really easy for anyone to understand. What you're missing is that some of the "ycombinator crowd" have worked with enough "average Equifax Joes" enough to know that the problems faced by "Equifax Joe" are not always solvable by merely being knowledgeable, competent, and careful.