How does watson use concurrency

Download preview PDF. Skip to main content. This service is more advanced with JavaScript available. Advertisement Hide. Conference paper. This process is experimental and the keywords may be updated as the learning algorithm improves. This is a preview of subscription content, log in to check access. Agarwal, S. Callahan, D. Cytron, R. ACM Trans. Duesterwald, E. Jeremiassen, T. Kamil, A. Trusting it as a security tool is the mistake same as trusting your car-repairer as a security expert from now on.

Something like that… But I am not a kernel expert. And this will obviously slow things down, but security tools rarely do not. In the paper, I argue that the problem here is not with a specific piece of software since identical vulnerabilities exist in a broad range of similar such systems , but rather that the system call wrapper approach is fundamentally flawed in the context of current operating system designs.

The paper documents several that have merit, including moving to a true message passing model offering argument atomicity guarantees or using an integrated security framework present on several OS platforms , etc. The solutions are known, and have been for several years, they just need to be adopted.

If I understand your proposal correctly, the wrapper would sign the arguments as checked before the kernel system call code runs, and then verify the signature later to try to detect modification. Unfortunately, this mechanism itself is also raceable — the attacker can replace the memory after the wrapper signs it, then restore the original value before the wrapper verifies it, in much the same way the attacker can replace arguments after they are checked and change them again before they are audited.

For this to be race-free, the actual kernel copyin would need to perform the signature verification on the argument used in the access—i. There are many hardening techniques that can be applied, and in several cases are already applied by current systems, but the underlying concern here is that the design fails to take into account concurrency, making these work-arounds that themselves are frequently vulnerable to very similar attacks.

What is required is an architecture that addresses concurrency directly. Yes — I originally reported these bugs to Niels in , and have posted about this general class of vulnerabilities, including those in systrace, on public mailing lists since that time.

Among other things, this lead to Niels documenting some aspects of the problem in the systrace man page although my recent work makes it clear that the caveat regarding clone is too narrow, as I successfully exploited these vulnerabilities using regular fork. The purpose of the current paper was to provide a more thorough analysis of the vulnerability class, and concretely explore exploit strategies that might be used by attackers. As it turned out, the concrete exploit strategies were remarkably straight forward.

LSM-style implementations have their design problems as well. Layered defense is always the best approach. Probably not ALL accesses would need to be stamped, only those used by system calls if this makes sense, which I am not certain. So to speak, given a system call. Posterior step: verify timestamps of syscall and memory at a,b,c.

If this is later, forgery has happened. I think this addresses your issue but is probably overkill. I agree that a better engineering is a better way.

And of course.. SQL injection is still the most attacked vulnerability, easy , remote, and still the world is not hanging neither broke, and each day appears a new vulnerable application. Nice presentation. In this presentation your examples are perfectly clear and easy to follow. They should be added to the man page as a warning not to rely on systrace as a security mechanism.

Improve this question. Chris Snow ItsJose ItsJose 41 6 6 bronze badges. What part of the message is unclear? How does the message relate to your use of concurrent sessions?

The question is not answerable given the information you've provided. JimGarrison I've updated the original post with a snippet of my code, what's unclear is that I don't know why I'm getting the error.

This is the first thing I'm trying to run when I run my app. IBM'sdocumentation for the java-sdk isn't thorough and I can't seem to find any reference for 'concurrency' limits anywhere. The issue will be where you instantiate service. I have no idea how Watson VR works, but you need to examine your code around establishing the connection to make sure you're not creating a new connection on each request. Add a comment. Active Oldest Votes. Improve this answer. That is correct - sorry for the inconvenience, ItsJose — Matt Hill.