Slashdot Mirror
Should Vendors Close All Security Holes?
johnmeister writes to tell us that InfoWorld's Roger Grimes is finding it hard to completely discount a reader's argument to only patch minimum or low security bugs when they are publicly discovered. "The reader wrote to say that his company often sits on security bugs until they are publicly announced or until at least one customer complaint is made. Before you start disagreeing with this policy, hear out the rest of his argument. 'Our company spends significantly to root out security issues,' says the reader. 'We train all our programmers in secure coding, and we follow the basic tenets of secure programming design and management. When bugs are reported, we fix them. Any significant security bug that is likely to be high risk or widely used is also immediately fixed. But if we internally find a low- or medium-risk security bug, we often sit on the bug until it is reported publicly. We still research the bug and come up with tentative solutions, but we don't patch the problem.'"
I did not RTFA, but I did read the summary. I did not hear his argument, I heard his conclusion repeated with more words.
Exploit Chaining means that low risk holes can become high risk hole when combined. Patch them all. Patch them quickly.
If you mod me down, I shall become more powerful than you could possibly imagine.
Also, vendors should include a free pony with every software license they sell.
Closing all vulnerabilities is not practical. In any sufficiently complex piece of software, there will be bugs and security holes. Obviously, you need to close the nasty ones, but many of these exploits are not particularly high risk. In these cases, especially if the fix would involve a major redesign or other highly disruptive solution, it may be best to just leave them alone.
If, for example, the underlying design of your product allows for a minor, difficult to exploit security hole, it is probably not worth it to spend the time and money to redesign the product. More likely, your choices would be either a.) live with the (small) vulnerability, or b.) scrap the product entirely.
The decision to close a security hole should be dependent on the potential impact of the hole, the urgency of the issue (are there already exploits in the wild, for example), and how many resources (time and money) it will take to fix it.
As is too common, the ./ summary doesn't have the relevant portions of the article under discussion, so let me try to summarize the main points of their argument.
1. It is better to focus resources on high risk security bugs.
2. We get rated better (internally and externally) for fixing publically known problems.
3. Hackers usually find additional bugs by examining patches to existing bugs, so a patch could expose more bugs than fixes are available for.
4. If we disclose a bug and fix it, it just escalates the "arms race" with the hackers. Better to keep it hidden.
5. Not all customers immediately patch. So by announcing a patch to previously unknown to the public bug, we actually exponentially increase the chances of that bug being exploited by hackers.
Hax-fu?
Basically ...
.... what the fuck?
#1. If we spend time fixing those bugs, we won't have as much time to fix the important bugs.
Translation: we put in so many bugs that we don't have time to fix them all.
#2. We give priority to any bugs that someone's leaked to the press.
Translation: we only fix it if you force us to.
#3. "Third, the additional bugs that external hackers find are commonly found by examining the patches we apply to our software."
I had to post that verbatim. They're releasing new bugs in their patches.
#4. "Fourth, every disclosed bug increases the pace of battle against the hackers."
Yeah, that one too. The more bugs they fix, the faster the
#5. If we don't announce it, they won't know about it.
Great. So your customers are at risk, but don't know it.
Pre-emptive disclosure works against the typical closed source company.
Option 1:
Exploit is published, patch is delivered really quickly. Sysadmin thinks, "Those guys at company X are on top of it..." PHB will say the same.
Option 2:
Unilaterally announce fixes, make patches available. Sysadmin doesn't bother to read the whole announcement and whines because it makes work she doesn't understand or think is urgent. PHB thinks "Gee company X's software isn't very good, look at all the patches..."
The market for secure software is small even smaller if you add standards compliance. Microsoft is a shining example of how large the market is for insecure, non-standard software.
http://www.maxineudall.com/2010/02/should-economists-be-sued-for-malpractice.html
We do the same thing. Every company has limited resources, and every decision is a usiness priority decision. So the decision is always between new features and old bugs.
Outside of terribly serious security holes, security holes are only security holes when they become a problem. Until then, they are merely potential problems. Solving potential problems is rarely a good idea.
We're not talking about tiny functions that don't consider zero values. We're talking about complex systems where every piece of programming has the potential to add problems not only to the system logic, but also to add more security holes.
That's right, I said it -- security patches can cause security holes.
It is our standard practice not to touch things that are working. Not every application is a military application.
I'll say it again. Not every application is a military application.
Your front door has a key-lock. That's hardly secure -- they are easily picked. But it's secure enough for most neighbourhoods.
So the question with your software is: when does this security hole matter, and how does it matter. If it's only going to matter when a malicious person attacks, then the question comes down to how many attackers you have. And if those attackers are professional, you might as well make their life easier, because they'll get in eventually in one way or another -- I'd rather know how they got in and be able to recover.
How it matters. If it reveals sensitive information, it matters. If it causes your application to occasionally crash, when an operator is near-by, then it doesn't matter.
There are omre important things to work on -- and many of these minor security holes actually disappear with feature upgrades -- as code is replaced.
Low-risk does not mean easy to fix. Sometimes, a bug might be a very low-risk bug, but demand immense amounts of time to find and fix. For instance, sometimes I might be writing a program, and at some point, it begins crashing unpredictably, but very rarely. I know that there is a bug, but I have no idea what the trigger is, I have no idea which part of the code contains the bug, and I have no idea how to fix it. Since the MTBF is (say) 3 months, and (say) the code is not long-running (like a daemon or a kernel), it is probably not worth finding and fixing the bug.
Now, that's bugs, which is a wider category than security holes. So, suppose that instead of crashing, it very rarely and briefly enters a state in which a particular sequence of bytes sent to it via the net can cause it to execute arbitrary code. Furthermore, suppose the program should never be running as root, so the arbitrary code is nearly useless. This is a low risk security hole, and probably not worth patching.
Could take hundreds of man-hours to find the cause, and perhaps even longer to fix. Probability of ever seeing this exploited is very very low. Should it then be patched?
SIGSEGV caught, terminating
wait... not that kind of sig.
It was Ford and it was the Pinto. The problem is.
1. The Pinto even before the "fix" didn't have the highest death rate in it's class. Other small cars had the same death per mile or worse.
2. The NTSA had the dollars per rate figure in the national standards for safety and Ford referenced in in their internal documentation which the lawyer used in the case.
3. Had Ford not identified the risk that a bolt posed to the fuel tank and documented it they probably wouldn't have lost so big in court.
Just thought I would try and kill a myth with some truth. Probably will not work but it is worth a shot.
See my blog http://ilovecookes.blogspot.com/ for light hearted technical information.
for better or worth
Let me guess - you're a LISP programmer?
The place I worked for was a security company. They had no automated unit testing and never analyzed for intrusions. You'd be shocked to find out how many holes exist on devices people depend on to keep them safe. The employees took it upon themselves (subverted authority) to patch our product. Security problems, even on security hardware, were not "priority" issues.
We too "trained" our coders in the art of secure programming. The problem, of course, is that we were also training them in basic things like what a C pointer is and how to not leak memory. Advanced security topics were over their head. This is the case in 9 out of 10 places I've worked. The weak links, once identified, can't be fired. No... these places move them to critical projects to "get their feet wet."
At the security giant, training only covered the absolute basics: shell escaping and preventing buffer overflows with range checking. The real problem is that only half of our bugs were caused by those problems. The overwhelming majority were caused by poor design often enabled by poor encapsulation (or complete lack of it).
There were so many use cases for our product that hadn't been modeled. Strange combinations of end-user interaction had the potential to execute code our appliance. Surprisingly, our QA team's manual regression testing (click around our U.I.) never caught these issues, but did catch many typos.
I don't believe security problems are inevitable. I've been writing code for years and mine never has these problems (arrogant, but mostly true). I can say, with certainty, than any given minor-version release has had 1,000's of high-quality tests performed and verified. I use the computer, not people... so there's hardly any "cost" to do so repeatedly.
I run my code through the paces. I'm cautious whenever I access RAM directly. My permissions engines are always centralized and the most thoroughly tested. I use malformed data to ensure that my code can always gracefully handle garbage. I model my use cases. I profile my memory usage. I write automated test suites to get as close to 100% code coverage as possible. I don't stop there. I test each decision branch for every bit of functionality that modifies state.
Aside from my debug-time assertions, I use exception handling quite liberally. It helps keep my code from doing exceptional things. Buffer overflows are never a problem, because I assume that all incoming data from ANY source should be treated as if it were pure Ebola virus. I assume nothing, even if the protocol / file header is of a certain type.
Security problems exist because bad coders exist. If you code and you know your art form well, you don't build code that works in unintended ways. Proper planning, good design, code reviews, and disciplined testing is all you need.
The article and responses miss an important point: patches of any kind are risky! And not just because they might introduce a new security flaw, but more generally because they may break some feature or another. In applications with millions of lines of code, and where the cost of doing a patch release amortized over all customers is millions of dollars, it can make lots of sense to just roll a fix into the next planned upgrade release. That way you get a complete Q/A and customer beta-test cycle to increase the confidence level of the fix.
Your example is way to simplistic. I've seen core dump cases in which it was perfectly clear why it was crashing: the data structure got into a logically inconsistent state that it should never be in. The question is how and when. In case of big data structures (in some of the cases I've had to deal with: hundreds of thousands of objects, built gradually and modified heavily during several hours) finding the exact sequence that causes the inconsistency can be a nightmare. Plus, it might have been sitting around in this state for a long time before the program actually enters a path that leads to a crash.
Also, the worst type of bug is the Heisenbug: those that go away as soon as you enable debugging, or add even just a single line of extra code to monitor something while the stuff is running. I've seen my share of those as well. Sometimes persistance pays off and you find the root cause within hours or days, but sometimes reality forces you to give up. It's no use spending five weeks fruitlessly looking for a rare intermittent bug triggered by a convoluted internal unit test case if at the same time easily tracable bugs are being reported by paying users that need a solution within a week.
Linux user since early January 1992.