AppArmor leaks kernel memory during profile reloads

[Jamie Strandboge]

The kmemleak kernel debugging system indicates that AppArmor leaks kernel memory during profile reloads.

I've attached a script (test.sh) that I can consistently reproduce the kmemleak report with.

I've also attached the output of the script, which includes the kmemleak report.

These leaks can be reproduced with debug kernel builds of Ubuntu-3.16.0-16.22 and Ubuntu-3.16.0-19.26, meaning that it is present before and after the apparmor 3 RC 1 sync.

[Tyler Hicks (tyhicks)]

I've looked into this more and have determined that it is an old bug and was not introduced by apparmor 3 RC1. There may be new leaks in apparmor 3 RC1 but there is a basic, fundamental leak on profile reloading present in the kernel released just before apparmor 3 RC1 landed.

I'll adjust the title and description to include the details.

[Tyler Hicks (tyhicks)]

Test script to be used for reproducing. (I'm not sure why I have to kick off the kmemleak scan twice)

[Tyler Hicks (tyhicks)]

Output of test.sh script, which includes the kmemleak report.

[Jamie Strandboge (jdstrand)]

I'm curious how far back this goes. Does it exist in trusty? precise? lucid?

[John Johansen (jjohansen)]

possibly trusty, though I did do a kmemleak check at the time and didn't notice anything. There where leaks during its dev cycle but I got at least most of them.

Lucid and precise are very different in their code base, and while I won't claim there aren't leaks, they are much less likely. The list management and reference counting are simpler, nor do they have the rcu changes etc.

[Tyler Hicks (tyhicks)]

The reproducer triggers the leak on Trusty (Ubuntu-3.13.0-37.64) but not Precise (Ubuntu-3.2.0-70.105). I didn't test Lucid.

[Jamie Strandboge (jdstrand)]

I'm going to reduce the priority of this to Medium since it exists on trusty and noone has reported it yet. We still need to fix it, but there are likely others things we want to get to first.