Gmail smartphone app hacked by researchers

Gmail smartphone app hacked by researchers(bbc.com)

30 points by erlendfh 11 years ago | 13 comments

Someone1234 11 years ago |

> A Google spokeswoman said the technology giant welcomed the research. "Third-party research is one of the ways Android is made stronger and more secure," she said.

That's a really great response and attitude. Nice to see and gives me hope that Android will be something we can look to for security and privacy at some stage in the not too distant future.

I wonder if a permissions overhaul is in the works? The recent change to categorized permissions seems both like a step backwards but also like it is showing off how flawed Android's permissions system really is. If Google really want Android to be "stronger and more secure" this would be a good place to start.

sjf 11 years ago |

There's not many details, it looks a failure of JVM sandboxing?

Ok, the actual paper describes the problem "popular GUI frameworks by design can potentially reveal every UI state change through a newly-discovered public side channel — shared memory".

The attack works by watching the size of shared memory in use in /proc, detecting UI events by allocations of shared memory, and then pre-empting the new screen with a phishing screen.

giovannibajo1 11 years ago | |

After some reading, the attack is basically threefold:

1) A service (with no permissions) can monitor /proc and come up with a way to identify Activity transitions, and pinpoint specific activities in specific apps. This is brilliant IMO.

2) With good timing, a service (with no permission) can inject a "phishing" Activity in foreground, just when the app is supposed to show its own Activity. This can be used to steal credentials, etc. Videos show that it works flawlessly, with perfect timing and no graphic glitches.

3) A service (with no permissions) can take camera pictures in background with no permissions, bypassing the permissions because of a bug (the camera preview into an OpenGL texture seems not guarded by app permission checks).

My comments:

#1 shows a security issue because of inheritance of bad security practices of desktop operating systems. Since applications don't need /proc (especially if they're supposed to always go through the framework, for security reasons), they shouldn't be given access.

#2 shows a glaring design bug in Android backgrounding system. The fact that a service can pop up an activity in front of an user while they are doing something else, is not only a UX mistake, but a clear security issue.

#3 seems like "just a bug", but the idea that any background service (with any permission level) could be taking pictures in background on any Android device is terrifying.

fidotron 11 years ago | |

Sounds to me like it detects the activity to activity transition (which on Android creates an Intent which is passed to the system and back via shared memory) and then does something else, i.e. display a fake password entry screen when the transition is detected.

It sounds like the actual Intent contents remain secure, however, the creation of an Intent while a particular app is in the foreground and using a particular amount of memory might be the kind of trigger they'd be looking for.

An OS level patch to get around this could involve preallocating shared memory for each application so that differences in it during the app lifecycle do not actually occur. Modern systems have enough RAM, and there are low enough limits on what is flying around, that this shouldn't be a problem.

giovannibajo1 11 years ago | | |

or just remove /proc access from apps, which is totally useless.

peterwwillis 11 years ago | |

They are calling the main attack UI state inference attack (looking at shared memory for window manager state changes) from which they begin an Activity hijacking attack (phishing). The camera attack is merely taking a picture right after a secured app (e.g. Banking app) has taken a picture.

The paper is linked in the article. Here are demos linked from their paper: https://sites.google.com/site/uistateinferenceattack/demos

pritambaral 11 years ago | |

TA mentions the researchers claiming similar attacks could be carried out on Windows and iOS too. Without more info, I suspect the failure isn't anything platform-specific like JVM sandboxing.