[question] should 363b02dab09b3 be backported to stable 4.1+?

[question] should 363b02dab09b3 be backported to stable 4.1+?

[Sergey Senozhatsky]

Hello David, Eric,

please help me out.

I'm looking at 363b02dab09b ("KEYS: Fix race between updating and finding
a negative key") right now. So, I see that it has been backported to stable
4.4+. My question is -- do we have those test_bit(KEY_FLAG_INSTANTIATED)
and test_bit(KEY_FLAG_NEGATIVE) races in stable 4.1?

	-ss

Re: [question] should 363b02dab09b3 be backported to stable 4.1+?

[Eric Biggers]

Hi Sergey,

On Fri, Dec 15, 2017 at 11:47:06AM +0900, Sergey Senozhatsky wrote:
> Hello David, Eric,
> 
> please help me out.
> 
> I'm looking at 363b02dab09b ("KEYS: Fix race between updating and finding
> a negative key") right now. So, I see that it has been backported to stable
> 4.4+. My question is -- do we have those test_bit(KEY_FLAG_INSTANTIATED)
> and test_bit(KEY_FLAG_NEGATIVE) races in stable 4.1?
> 

Before 4.4 (146aa8b1453), ->reject_error was in union with ->type_data rather
than ->payload, and no key types that used ->type_data implemented ->update().
Therefore it was not possible to reproduce the crash.

I do see there was another possible race, only theoretically a problem on
architectures with weaker memory ordering than x86, where a key being negatively
instantiated could be momentarily observed to be positively instantiated.  But
even then I don't see where it could be a real problem.  (Note that most users
wait for KEY_FLAG_USER_CONSTRUCT rather than checking KEY_FLAG_INSTANTIATED
directly.)

You're free to backport the commit if you want to be absolutely sure, though I'd
personally be more worried about other backports that might have been missed,
and the bugs that haven't been found yet.

Eric

Re: [question] should 363b02dab09b3 be backported to stable 4.1+?

[Sergey Senozhatsky]

Hello Eric,

On (12/14/17 19:58), Eric Biggers wrote:
> Hi Sergey,
> 
> On Fri, Dec 15, 2017 at 11:47:06AM +0900, Sergey Senozhatsky wrote:
> > Hello David, Eric,
> > 
> > please help me out.
> > 
> > I'm looking at 363b02dab09b ("KEYS: Fix race between updating and finding
> > a negative key") right now. So, I see that it has been backported to stable
> > 4.4+. My question is -- do we have those test_bit(KEY_FLAG_INSTANTIATED)
> > and test_bit(KEY_FLAG_NEGATIVE) races in stable 4.1?
> > 
> 
> Before 4.4 (146aa8b1453), ->reject_error was in union with ->type_data rather
> than ->payload, and no key types that used ->type_data implemented ->update().
> Therefore it was not possible to reproduce the crash.
> 
> I do see there was another possible race, only theoretically a problem on
> architectures with weaker memory ordering than x86, where a key being negatively
> instantiated could be momentarily observed to be positively instantiated.  But
> even then I don't see where it could be a real problem.  (Note that most users
> wait for KEY_FLAG_USER_CONSTRUCT rather than checking KEY_FLAG_INSTANTIATED
> directly.)

thanks a ton. appreciate your help!

> You're free to backport the commit if you want to be absolutely sure, though I'd
> personally be more worried about other backports that might have been missed,
> and the bugs that haven't been found yet.

agreed.

	-ss