When generating an NX crash, previously the code would explicitly jump... #6
Conversation
…p to NULL, which modern versions of Clang correctly optimize out as provable undefined behavior (the compiler is free to do whatever it wants if it can prove that the code will always dereference NULL). Instead, map a valid memory space without the execute permission and jump to that pointer.
|
@gwynne, |
|
I just "stole" this for the Mobile Center SDK =) |
1) Because the jump to a bad address prevents the last “correct” frame from showing up, act more like real code and call another function that definitively isn’t inlined. 2) munmap() the pointer after the crash for good form.
|
Since Xcode 8.3 this test case didn't cause a crash any longer due to the reasons @gwynne explained. But the new test code will return different results, e.g. it is more likely now to get the proper filename and line numbers. Here is what HockeyApp would (correctly) return, and therefor the result not being To publish this test change properly, it would only be right that every contributing service could update their test results and we'll publish all changes (ideally) in one update. What do you think? Calling out the previous contributors on https://github.com/bitstadium/CrashProbe-Data: @QuantumNightmare @kattrali @finik @HazAT |
ghost
mentioned this pull request
|
Makes sense, we are getting similar results. Added a commit with an update to our pull request. |
|
We are in for that 👍
-> #7 |
|
@HazAT Please open an extra ticket for that discussion. Thanks :) |
ghost
merged commit 
to NULL, which modern versions of Clang correctly optimize out as provable undefined behavior (the compiler is free to do whatever it wants if it can prove that the code will always dereference NULL). Instead, map a valid memory space without the execute permission and jump to that pointer.