Hi,

On Thu, 2023-11-23 at 17:32 +0100, Eric Dumazet wrote:

On Thu, Nov 23, 2023 at 3:34 PM Valentin Schneider [off-list ref] wrote:

quoted

So AFAICT, after we go through the hashdance, there's a reference on
tw_refcnt held by the tw_timer.
inet_twsk_deschedule_put() can race with arming the timer, but it only
calls inet_twsk_kill() if the timer
was already armed & has been deleted, so there's no risk of calling it
twice... If I got it right :-)

Again, I think you missed some details.

I am OOO for a few days, I do not have time to elaborate.

You will need to properly track active timer by elevating
tw->tw_refcnt, or I guarantee something wrong will happen.

I'm sorry to bring this up again, but I tried to understand what is
missing in Valentin's patch and I could not find it.

Direct link to the patch, just in case the thread has been lost:
https://lore.kernel.org/lkml/20231115210509.481514-2-vschneid@redhat.com/ (local)

The patch raises the initial tw->tw_refcnt to 4, so it tracks (in
advance) the reference for the tw_timer. AFAICS the patch is still
prone to the race you mentioned on the RFC:

CPU0:

   allocates a tw, insert it in hash table

CPU1:
   finds the TW and removes it (timer cancel does nothing)

CPU0:
   arms a TW timer, lasting

but I understood such race is acceptable.

Could you please shed some light? 

Many thanks,

Paolo