On 05.02.2026 21:41, Willem de Bruijn wrote:
Vadim Fedorenko wrote:
On 05/02/2026 16:43, Sebastian Andrzej Siewior wrote:
On 2026-02-05 16:27:03 [+0000], Vadim Fedorenko wrote:
So the only thing that bothers me is the read_lock_bh() in
skb_may_tx_timestamp() which deadlocks if the socket is write-locked on
the same CPU.
Alright. Now you make me think whether we should enforce OPT_TSONLY
option on socket which doesn't have CAP_NET_RAW? Then we can get rid of this
check, and in case sysctl was flipped off - drop TX timestamps as
it's done now?
This would "fix" this problem for all users which do deliver the
timestamp from their IRQ handler instead of napi. There are a few of
those…
This would be considered stable material, right? (despite the fact that
we have it for quite some time and nobody complained so far).
cc: Willem as he is the author of the check introduced back in 2015.
But it's more like a question to maintainers whether it is acceptable
way of "fixing" drivers or it's no-go solution
Requiring OPT_TSONLY unless CAP_NET_RAW would break legacy users.
Well, they are kinda broken already. Without OPT_TSONLY and CAP_NET_RAW all TX
timestamps are silently dropped. To receive these timestamps users have to get
CAP_NET_RAW permission, and it will work with the updated logic as well...
