1

Okay, so I was looking for something like answered here. The only problem is I've been looking at that zygote fork code but I have no idea what is going on. I'm trying to figure out where exactly permissions are enforced for native method calls in Android. More basically, I want to know how the Linux Kernel is enforcing permissions. Something like enforcePermission() which I've seen in some of the android source code, except I want to know where it is at the kernel level. If someone can point me in some sort of direction or make sense of what that zygote fork code is doing I'd appreciate the hell out of it.

If that doesn't make sense it's probably because I'm an idiot or something.

Thanks!

Community
  • 1
  • 1
Corey
  • 771
  • 3
  • 14
  • 32

1 Answers1

2

What that post says (which is true) is that there is no special enforcePermission()-style call for native code: each permission granted effectively translates into a specific supplementary group id. Individual permissions checks are performed either using the standard Linux permissions/capabilities model, using specific code in IPC routines (so for example when you bind to certain services the services can check the calling processes membership in the appropriate group) or using specific patches the the kernel/libraries (for example network permission is explained here).

Femi
  • 64,273
  • 8
  • 118
  • 148
  • Okay that is starting to make sense. So I guess my question now is **how** are the permission checks performed using the standard Linux permission/capabilities model? I don't really understand what you mean by "each permission granted effectively translates into a specific supplementary group id" – Corey Jun 22 '11 at 18:10
  • Say for instance I wanted to know which native methods would cause a SecurityException, if it was missing the corresponding permission. This is fairly easy to trace for java in the source code but I can't find where this is being done for native calls. Does that make sense? – Corey Jun 22 '11 at 18:23
  • Ah. That is because there is no one standard location that it happens. Unfortunately there is no one standard place to check for everything: the things that use the standard kernel mechanisms (such as file permissions) are probably in the same place as they are on stock Linux, but for things like network security and other services that are handled via a bind you'll have to go spelunking through the git code to try and figure out where a specific security exception will be propagated up. – Femi Jun 22 '11 at 19:29