Re: use-after-free in srpt_enable_tpg()

[Bart Van Assche <bvanassche@acm.org>]

On 6/30/22 09:40, Mike Christie wrote:
> On 6/27/22 11:37 AM, Bart Van Assche wrote:
>> On 6/27/22 00:09, lizhijian@fujitsu.com wrote:
>>> So far, I doubt if the previous defect of configfs mentioned in
>>> 9b64f7d0b: "(RDMA/srpt: Postpone HCA removal until after configfs
>>> directory removal)" has got a better solution. if not, i have no a
>>> clear mechanism to avoid it yet.
>>>
>>> feedbacks are very welcome.
>> Mike, are you perhaps aware of any plans to add functions to the LIO core for removing tpg and wwn objects?
>>
> 
> I don't know any work being done in this area. I was only working
> on the refcounting/configfs parts for sessions in those configfs/sysfs
> patchsets. However, I think I hit similar issues with the session.
> I went around the world with solutions but didn't really like them
> so I never pushed the patches for inclusion.
> 
> What was the hang caused by 9b64f7d0bb0a?

Hi Mike,

I have not been able to find Honggang's bug report that led to the 
revert of that commit. I guess that the hang happened in the while-loop 
in srpt_release_sport() that was modified by commit 9b64f7d0bb0a.

Thanks,

Bart.