Re: 3.0.0-rc2-git1 -- BUG: sleeping function called from invalidcontext at mm/slub.c:847

[John Johansen]

On 06/08/2011 01:09 PM, David Rientjes wrote:
> On Wed, 8 Jun 2011, Miles Lane wrote:
> 
>> BUG: sleeping function called from invalid context at mm/slub.c:847
>>  in_atomic(): 1, irqs_disabled(): 0, pid: 1583, name: cupsd
>>  2 locks held by cupsd/1583:
>>   #0:  (tasklist_lock){.+.+.+}, at: [<ffffffff8104dafa>] do_prlimit+0x61/0x189
>>   #1:  (&(&p->alloc_lock)->rlock){+.+.+.}, at: [<ffffffff8104db2d>]
>> do_prlimit+0x94/0x189
>>  Pid: 1583, comm: cupsd Not tainted 3.0.0-rc2-git1 #7
>>  Call Trace:
>>   [<ffffffff8102ebf2>] __might_sleep+0x10d/0x112
>>   [<ffffffff810e6f46>] slab_pre_alloc_hook.isra.49+0x2d/0x33
>>   [<ffffffff810e7bc4>] kmem_cache_alloc+0x22/0x132
>>   [<ffffffff8105b6e6>] prepare_creds+0x35/0xe4
>>   [<ffffffff811c0675>] aa_replace_current_profile+0x35/0xb2
>>   [<ffffffff811c4d2d>] aa_current_profile+0x45/0x4c
>>   [<ffffffff811c4d4d>] apparmor_task_setrlimit+0x19/0x3a
>>   [<ffffffff811beaa5>] security_task_setrlimit+0x11/0x13
>>   [<ffffffff8104db6b>] do_prlimit+0xd2/0x189
>>   [<ffffffff8104dea9>] sys_setrlimit+0x3b/0x48
>>   [<ffffffff814062bb>] system_call_fastpath+0x16/0x1b
>>
> 
> Not sure why this ever actually worked with apparmor if prepare_creds() 
> does an unconditional GFP_KERNEL allocation since this codepath hasn't 
> changed in at least a year and we're holding a spinlock from setrlimit.  
> John?

Hrmm, I am assuming it was missed because the profile replacement that
causes the prepare_creds() to be called is a rare condition when apparmor
is in use.  It will only occur when policy has been replaced and only if
the task doing the task_setrlimit() hasn't already updated its cred else
where.  It is clearly wrong, patch included below.

---