BUG: cgroup_task_counter subsys may crash with whole-threadgroup move

BUG: cgroup_task_counter subsys may crash with whole-threadgroup move

[Ben Blum]

I was testing some patches for cgroup_attach_proc and managed to cause a
crash with the following usage pattern:

    mount -t cgroup none -o tasks /dev/cgroup
    cd /dev/cgroup
    mkdir foo
    echo $PID > foo/cgroup.procs
    echo $PID > tasks
    echo $PID > foo/cgroup.procs

Where $PID is the thread ID of a member of a multithreaded process (my
test program just does CLONE_THREAD 8 times and then all threads sleep).
(It doesn't matter if the thread is the group leader or not, but a
single-threaded process doesn't crash.)

And get the following kernel panic:
http://maximegalon.andrew.cmu.edu/cgroup-taskstats/panic.txt

It's deterministic, and happens only when the "tasks" subsystem is
mounted.

I'm using user-mode linux to test, with the following config:
http://maximegalon.andrew.cmu.edu/cgroup-taskstats/config.txt

and I ran it in GDB to get the following backtrace:
http://maximegalon.andrew.cmu.edu/cgroup-taskstats/bt.txt

Thanks,
Ben

Re: BUG: cgroup_task_counter subsys may crash with whole-threadgroup move

[Li Zefan]

08:09, Ben Blum wrote:
> I was testing some patches for cgroup_attach_proc and managed to cause a
> crash with the following usage pattern:
> 
>     mount -t cgroup none -o tasks /dev/cgroup
>     cd /dev/cgroup
>     mkdir foo
>     echo $PID > foo/cgroup.procs
>     echo $PID > tasks
>     echo $PID > foo/cgroup.procs
> 
> Where $PID is the thread ID of a member of a multithreaded process (my
> test program just does CLONE_THREAD 8 times and then all threads sleep).
> (It doesn't matter if the thread is the group leader or not, but a
> single-threaded process doesn't crash.)
> 
> And get the following kernel panic:
> http://maximegalon.andrew.cmu.edu/cgroup-taskstats/panic.txt
> 
> It's deterministic, and happens only when the "tasks" subsystem is
> mounted.
> 
> I'm using user-mode linux to test, with the following config:
> http://maximegalon.andrew.cmu.edu/cgroup-taskstats/config.txt
> 
> and I ran it in GDB to get the following backtrace:
> http://maximegalon.andrew.cmu.edu/cgroup-taskstats/bt.txt
> 

I've figured out what's wrong. Patch will be sent out soon.

Thanks for reporting this!

--
Li Zefan