Offsets are wrong for iOS 7.1+

[AurelienRibon]

The wipe function works fine on iOS 6 and 7.0, but on my iOS 7.1.1 phone, wiping a NSString crashes the phone. Actually, looking at the memory, I found that in IMSMemoryManager.h, the OFFSET_SSTRING should be 17. Long strings are ok, as well as NSData. However, Wiping NSNumber crashes the app too, so its offset should be wrong too.

I also found out that in your demo, when replacing the 16 byte string ("0123456789ABCDEF") by a 512 byte string, the string appears twice in memory (shown when selecting "2048" as memory view window), and the wipe function only clears the first occurence of the string. The resulting string is "" in the debugguer, but the original content can still be shown in the memory, which defeats the purpose of the library.

I'm not an expert, so I can't help you more on that, but you may find this useful to know.

Thanks for this project, this can become an awesome library for secure apps!

[zerpt]

@AurelienRibon
Hello! Thanks for this. After an initial inspection, the offsets issue looks to stem from two things: 1) a lack of an intelligent way to dynamically figure out the offsets, and 2) the current offsets work in a 32-bit environment, but I found the problems you describe on a 64-bit device. I assume your 7.1.1 phone is arm64? I'll do some experimenting and try to find the proper offsets for 64-bit archs.

I'm currently looking into the demo string appearing multiple times in memory. I found a similar issue a bit back with string literals and other code constants that get placed in the read-only data memory segment, which to avoid just requires NSObjects to be created in a certain way.

I'll update when I have more info. Thanks for trying the project!

[zerpt]

After some investigating, I found proper offsets for short and long NSStrings, NSData, and decimal-valued NSNumbers. Unfortunately, it looks like NSNumber on 64bit archs handles certain data types (int, long, etc) in a much different manner than the others (i.e. storing the value in the "pointer" itself). We're looking at a couple different avenues, and hopefully will have something pushed out soon to support both 32 and 64bit.

[zerpt]

I've merged a few working updates allowing MemSec on 64bit hardware/simulators. Unfortunately NSNumbers on 64bit with ints or longs is problematic, and I've replicated the appearing multiple times in memory issue, but these are deep seeded issues that may take some time to construct a resolution.