Morpheus: A Vulnerability-Tolerant Secure Architecture [pdf](web.eecs.umich.edu)
web.eecs.umich.edu
Morpheus: A Vulnerability-Tolerant Secure Architecture [pdf]
https://web.eecs.umich.edu/~barisk/public/morpheus.pdf
7 comments
I wonder how this impacts debugging. It kind of rubs me the wrong way too. Typically, robust security is about keeping things simple so they can be audited or verified. Pretty cool results though.
I think this depends on what you think of ASLR. If ASLR rubs you the wrong way, this will too.
This looks like it goes way beyond ASLR.
There are a series of different protections they are doing. It's not clear if they intend them to be uniquely switchable or if they were just separating them to indicate which attack techniques require which protections.
For example it appears to combat ROP they are additionally encrypting the code segment of the program and decrypting on an as needed basis. I am absolutely sure this would make debugging more difficult. It wouldn't be used on a debug build though so I guess the question is how often would you find yourself debugging a production executable.
I admit I didn't read it in depth and may have missed something but I wonder if they are doing something similar for dynamic libraries or if it relies on the code being available at compile time. If not then I'm sure someone clever will still get around it with ROP.
There are a series of different protections they are doing. It's not clear if they intend them to be uniquely switchable or if they were just separating them to indicate which attack techniques require which protections.
For example it appears to combat ROP they are additionally encrypting the code segment of the program and decrypting on an as needed basis. I am absolutely sure this would make debugging more difficult. It wouldn't be used on a debug build though so I guess the question is how often would you find yourself debugging a production executable.
I admit I didn't read it in depth and may have missed something but I wonder if they are doing something similar for dynamic libraries or if it relies on the code being available at compile time. If not then I'm sure someone clever will still get around it with ROP.
The debugger doesn't really rely on the code segment being readable generally.
You are conflating the act of debugging with the operation of a "debugger".
Sure you can start at 0 and step forward one instruction at a time, but how often do you do that without at least looking at the code?
Not being able to look ahead or behind but only within what is actively on the stack is certainly a limiting factor in debugging.
It is of course a problem that has been solved before. Someone would develop an unpacker/decrypter for it at some point and make this trivial, but that would still be an added level of complexity.
Sure you can start at 0 and step forward one instruction at a time, but how often do you do that without at least looking at the code?
Not being able to look ahead or behind but only within what is actively on the stack is certainly a limiting factor in debugging.
It is of course a problem that has been solved before. Someone would develop an unpacker/decrypter for it at some point and make this trivial, but that would still be an added level of complexity.
You have access to the binary itself, you don't need to read it out of the debuggee process.
I agree. I'm not sure I understand your point though. This isn't contrary to anything I said. I mention unpacking which would be a pre-debugger process.
I believe you are just being pedantic about my use of the word "debugging" to include things like unpacking, static analysis, and a variety of tools and processes used.
I believe you are just being pedantic about my use of the word "debugging" to include things like unpacking, static analysis, and a variety of tools and processes used.
The binaries aren't preencrypted here, the dynamic nature is the whole point. There's nothing to unpack since you just reference the binary, and no added work for the debugger.
So when you say
> For example it appears to combat ROP they are additionally encrypting the code segment of the program and decrypting on an as needed basis. I am absolutely sure this would make debugging more difficult.
It's just not true.
So when you say
> For example it appears to combat ROP they are additionally encrypting the code segment of the program and decrypting on an as needed basis. I am absolutely sure this would make debugging more difficult.
It's just not true.
So yes, you are correct it's being encrypted dynamically at the hardware level.
You just debug in a system without these protections turned on then.
You just debug in a system without these protections turned on then.
No, I'm saying you don't have to turn off these protections.
Like, can you give an example of a debugging task you have carried out that requires reading the code segment of the running process?
Like, can you give an example of a debugging task you have carried out that requires reading the code segment of the running process?
Required? No, but I didn't say impossible. I said "more difficult".
Made easier by? Most everything ever loaded in a debugger.
Made easier by? Most everything ever loaded in a debugger.
Just for an example. How about debugging a non trivial issue between a program and a dynamically loaded library.
You can't just scan the IAT/PLT. You are going to have to find load points or manually step through, and maybe calculate function offsets in the library. Not impossible at all.
It is however something you could have just quickly glanced at and found otherwise.
You can't just scan the IAT/PLT. You are going to have to find load points or manually step through, and maybe calculate function offsets in the library. Not impossible at all.
It is however something you could have just quickly glanced at and found otherwise.
The article reads like a brute force approach to secure design, sprinkling randomization everywhere. Performance results are promising, though.
Hmm, like Shape Security did with web elements, this appears to do with hardware elements.
Seems promising, but not the level of enterprise architecture most at risk.
Seems promising, but not the level of enterprise architecture most at risk.
Wonder if the commercial Morpheus will have a challenge with this naming (https://www.morpheusdata.com/, "Next-gen multi-cloud management platform for Hybrid IT and DevOps automation")
Morpheus is the ancient greek god of sleep and dreams.
The etymology fits better in The Matrix than either of these projects.
The etymology fits better in The Matrix than either of these projects.
Impressive. I wonder why RISC-V only?
Morpheus modified hardware to accelerate the procedures which randomly alter the system stack parameters that attackers learn by probing the system. They used risc-v for this prototype.
Note that no hardware engineering (say, writing HDL, not to speak of tapeout) was done. It was implemented only in simulator.
I would like to hack this CPU..