Re: [PATCH RFC] slub: do not drop slab_mutex for sysfs_slab_{add,remove}

[Vladimir Davydov <vdavydov@parallels.com>]

On 02/06/2014 08:22 PM, Christoph Lameter wrote:
> On Thu, 6 Feb 2014, Vladimir Davydov wrote:
>
>> When creating/destroying a kmem cache, we do a lot of work holding the
>> slab_mutex, but we drop it for sysfs_slab_{add,remove} for some reason.
>> Since __kmem_cache_create and __kmem_cache_shutdown are extremely rare,
>> I propose to simplify locking by calling sysfs_slab_{add,remove} w/o
>> dropping the slab_mutex.
> The problem is that sysfs does nasty things like spawning a process in
> user space that may lead to something wanting to create slabs too. The
> module may then hang waiting on the lock ...

Hmm... IIUC the only function of concern is kobject_uevent() -
everything else called from sysfs_slab_{add,remove} is a mix of kmalloc,
kfree, mutex_lock/unlock - in short, nothing dangerous. There we do
call_usermodehelper(), but we do it with UMH_WAIT_EXEC, which means
"wait for exec only, but not for the process to complete". An exec
shouldn't issue any slab-related stuff AFAIU. At least, I tried to run
the patched kernel with lockdep enabled and got no warnings at all when
getting uevents about adding/removing caches. That's why I started to
doubt whether we really need this lock...

Please correct me if I'm wrong.

> I would be very thankful, if you can get that actually working reliably
> without deadlock issues.

If there is no choice rather than moving sysfs_slab_{add,remove} out of
the slab_mutex critical section, I'll have to do it that way. But first
I'd like to make sure it cannot be done with less footprint.

Thanks.