Re: [PATCH v2] unshare: Unsharing a thread does not require unsharing a vm

[Eric W. Biederman]

Oleg Nesterov <oleg@xxxxxxxxxx> writes:

> On 08/13, Eric W. Biederman wrote:
>>
>> Oleg Nesterov <oleg@xxxxxxxxxx> writes:
>>
>> > On 08/12, Eric W. Biederman wrote:
>> >>
>> >> +	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
>> >> +		if (atomic_read(&current->sighand->count) > 1)
>> >> +			return -EINVAL;
>> >> +	}
>> >
>> > I am still not sure we want this... please the the previous email.
>>
>> Reading your other email I did not see why you thought this check was
>> unnecessary.
>>
>> > But perhaps I missed something.
>>
>> In short:
>> clone(VM) --> mm_users > 1 && sighand_struct->count == 1
>> followed by:
>> unshare(SIGHAND)
>> the unshare should succeed.
>>
>> Meanwhile:
>> clone(VM|SIGHAND) --> mm_users > 1 && sighand_struct->count > 1
>> followed by:
>> unshare(SIGHAND)
>> the unshare should fail.
>
> Yes, yes, yes.
>
> But once again, I meant we can remove this sighand->count check
> if unshare(SIGHAND) checks current_is_single_threaded(). That is
> why I suggested to do
>
> 	if (unshare_flags & CLONE_SIGHAND)
> 		unshare_flags |= CLONE_VM;
>
> in sys_unshare(), or change check_unshare_flags() to check
> "unshare_flags & (CLONE_VM | CLONE_SIGHAND)" before
> current_is_single_threaded().

See the two cases above that change to unshare_flags will make
unshare(SIGHAND) fail when sighand_struct->count == 1.

Which is fundamentally wrong.

> Damn. And this discussion makes me think that another cleanup makes
> sense too. Can't we move all these unshare_flags manipulations into
> check_unshare_flags? So that sys_unshare() will only do
>
> 	err = check_unshare_flags(&unshare_flags);
>
> and the reader of this code won't need to read 2 functions to understand
> whats going on.

Eric

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/