Re: 3.9-rc1 NULL pointer crash at find_pid_ns

[Sasha Levin <sasha.levin@oracle.com>]

On 03/07/2013 04:59 AM, ebiederm@xmission.com wrote:
> Li Zefan <lizefan@huawei.com> writes:
> 
>> Cc: sasha.levin@oracle.com
>> Cc: "Eric W. Biederman" <ebiederm@xmission.com>
>> Cc: container
>>
>> This is a second report... and the same address: 0xfffffffffffffff0 
> 
> Actually this is the third report I have seen with that address, and the
> others were on x86_64.
> 
> The obvious answer is that there is something subtlely wrong with:
> 
> commit b67bfe0d42cac56c512dd5da4b1b347a23f4b70a
> Author: Sasha Levin <sasha.levin@oracle.com>
> Date:   Wed Feb 27 17:06:00 2013 -0800
> 
>     hlist: drop the node parameter from iterators
> 
> 
> This is the only change the pid namespace that I am aware of in 3.9-rc1.
> 
> If you can reproduce this somewhat readily can you please revert the
> hlist change and see if this continues to happen.  Right now there are
> no other code changes that I can see.  And the address
> 0xfffffffffffffff0 is consistent with a bug in hlist_for_each_entry_rcu.

Looks like the hlist change is probably the issue, though it specifically
uses:

	#define hlist_entry_safe(ptr, type, member) \
        	(ptr) ? hlist_entry(ptr, type, member) : NULL

I'm still looking at the code in question and it's assembly, but I can't
figure out what's going wrong. I was also trying to see what's so special
about this loop in find_pid_ns as opposed to the rest of the kernel code
that uses hlist_for_each_entry_rcu() but couldn't find out why.

Is it somehow possible that if we rcu_dereference_raw() the same thing twice
inside the same rcu_read_lock() section we'll get different results? That's
really the only reason for this crash that comes to mind at the moment, very
unlikely - but that's all I have right now.

Is this bug reproducible easily on your setup? I've managed to reproduce it
exactly 3 times in the past month or so, twice when I reported it and only
once since then - at some point I thought that it was a freak compiler issue
that went away when the code changed but since you're reporting it again
I guess that it isn't the case.


Paul, any chance you can give hlist_for_each_entry_rcu() a second look please?
I know you've already acked it before, but is it possible I missed a subtle
detail with RCU that causes this?


Thanks,
Sasha