Re: [PATCH] proc: add locking checks in proc_inode_is_dead

[Eric W. Biederman]

Oleg Nesterov <oleg@xxxxxxxxxx> writes:

> On 11/30, Eric W. Biederman wrote:
>>
>> Ouch!!!!  Oleg I just looked the introduction of proc_inode_is_dead in
>> d855a4b79f49 ("proc: don't (ab)use ->group_leader in proc_task_readdir()
>> paths") introduced a ``regression''.
>>
>> Breaking the logic introduced in 7d8952440f40 ("[PATCH] procfs: Fix
>> listing of /proc/NOT_A_TGID/task") to keep those directory listings not
>> showing up.
>
> Sorry, I don't understand...
>
> Do you mean that "ls /proc/pid/task" can see an empty dir? Afaics this
> was possible before d855a4b79f49 too.
>
> Or what?

Bah. Brain fart on my part.

I read 7d8952440f40 too fast.  I thought it was attempting to make
it so that "ls /proc/tid/task/" would see an empty dir while "ls
/proc/tgid/task/" would see the complete set of threads.

Where tgid is the pid of the thread group leader and tid
is the pid of some thread in the thread group.


But 7d8952440f40 was just attempting to ensure that no thread was
listed more than once in "/proc/xxx/task".

My apologies for the confusion.

Eric