search

jovanbulck / sgx-step

A practical attack framework for precise enclave execution control
GNU General Public License v3.0
441 stars 83 forks source link

Make aep_trampoline.S relocatable #76

Open Danacus opened 8 months ago

Danacus commented 8 months ago

Upstreaming some changes in my personal fork. If I recall correctly, this change to aep_trampoline.S was necessary to make libsgxstep relocatable. Without these changes, it is not possible to link libsgxstep in a shared object.

jovanbulck commented 7 months ago

thanks for opening this PR! Not sure in which exact scenario we need this? Is this related to the Rust bindings (#78 )?

Also, don't we have to do this as well for other .S files like irq_entry.S?

Danacus commented 7 months ago

I don't think it is related to #78, but it's something I had to do when trying to build libsgxstep with -fPIC. If I recall correctly, I needed this because I was then linking libsgxstep into a shared object, which requires -fPIC.

I'm not sure if we should do this in other .S files as well.

jovanbulck commented 7 months ago

I see, is there a reason to compile libsgxstep with -fPIC though?

Danacus commented 7 months ago

Perhaps my use case is a bit niche, but I was using a program that loads a .so at runtime depending on some command line argument, and that .so then loads a certain enclave and uses some libsgxstep features. In order to link code into a shared object, it needs to be compiled with -fPIC.

jovanbulck commented 7 months ago

makes sense, thanks for clarifying.

In fact, I've already longer wanted to enable LD_PRELOADing libsgxstep into a standard application (cf #28 ) to dynamically hook into enclave entries.

So putting this on the list to figure out and upstream!

Danacus commented 7 months ago

Also, don't we have to do this as well for other .S files like irq_entry.S?

It turns out that we do. I was not running into this before, because I did not include these files when linking because these files were not used. When linking with --whole-archive, I do run into issues with irq_entry.S.

jovanbulck commented 1 month ago

I found this very helpful: https://stackoverflow.com/a/66713969

TL;DR

mov foo@GOTPCREL(%rip), %rax loads &foo into RAX. Often this is followed by mov (%rax), %eax or similar to actually get the value of a global variable like int foo;

so in your patch above, lea should be mov followed by another mov if the value is needed (tested this on another branch)