Re: [PATCH net-next 2/6] neigh: fix a possible leak issue of neigh entry

[Ying Xue <ying.xue@windriver.com>]

On 05/18/2015 12:58 PM, Eric Dumazet wrote:
> On Mon, 2015-05-18 at 11:24 +0800, Ying Xue wrote:
> 
>> If the issue of "neigh use-after-free" is fixed by the first patch although you
>> said the atomic_read() was not safe for us, is the patch still wrong? If it's
>> really wrong, can you please give more detailed explanation to help me
>> understand why the change is wrong and but why a similar timer usage in
>> sk_reset_timer() is not wrong?
> 
> Why do you believe sk_reset_timer() would be wrong ?
> 

Sorry, I don't think sk_reset_timer() is wrong, instead I suppose
neigh_add_timer() is wrong :)

> Difference between neigh_add_timer() and sk_reset_timer() is very
> simple :
> 
> neigh_add_timer() must be called while the timer is not yet armed.
> 

In case that the caller of neigh_add_timer() attempts to modify an active timer
due to a bug or something wrong else, why not prevent neigh_add_timer() from
taking neigh refcnt beside posting a warning message?

So, exactly speaking, we cannot say neigh_add_timer() is completely wrong
regarding your mentioned above assumption that neigh timer is absolutely not
armed when neigh_add_timer() is called. But we can say its behaviour is not
designed very well. From this point, the patch seems still a bit valuable for us.

Regards,
Ying

> sk_reset_timer() can be called while timer is already armed.
> 
> You are changing neigh_add_timer() for no good reason, just because you
> want it to be 'like sk_reset_timer()' ?
> 
> 
> 
> 
>