Re: [deadlock or dead code] ceph_encode_dentry_release() misuse of dget()

[Xiubo Li <xiubli@redhat.com>]

On 11/17/23 00:50, Jeff Layton wrote:
> On Thu, 2023-11-16 at 16:28 +0000, Al Viro wrote:
>> On Thu, Nov 16, 2023 at 07:50:03AM -0500, Jeff Layton wrote:
>>
>>>> Am I missing something subtle here?  Looks like that dget() is never
>>>> reached...
>>> No, I think you're correct. That looks like dead code to me too.
>>> Probably we can just remove that "if (!dir)" condition altogether.
>>>
>>> Did you want to send a patch, or would you rather Xiubo or I do it?
>> Up to you...  AFAICS, it had been dead code since ca6c8ae0f793 "ceph: pass
>> parent inode info to ceph_encode_dentry_release if we have it".  In other
>> words, that "if we have it" had already been true at that point.  Prior
>> to that commit dget() in there had been unconditional (and really a deadlock
>> fodder); making it conditional had made it actually unreachable and that
>> fixed the actual bug.
> That makes sense.
>
> Xiubo, would you mind spinning up a patch for this? You're probably in
> the best position to make sure it gets tested these days.

Hi Jeff, Al

Sure, I will fix this. Thanks very much.

- Xiubo


> Thanks!