Re: [PATCH v3 bpf-next 3/5] [RFC] net: Skip taking lock in BPF context

[Stanislav Fomichev <sdf@google.com>]

On 03/21, Aditi Ghag wrote:
> When sockets are destroyed in the BPF iterator context, sock
> lock is already acquired, so skip taking the lock. This allows
> TCP listening sockets to be destroyed from BPF programs.

> Signed-off-by: Aditi Ghag <aditi.ghag@isovalent.com>
> ---
>   net/ipv4/inet_hashtables.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)

> diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
> index e41fdc38ce19..5543a3e0d1b4 100644
> --- a/net/ipv4/inet_hashtables.c
> +++ b/net/ipv4/inet_hashtables.c
> @@ -777,9 +777,11 @@ void inet_unhash(struct sock *sk)
>   		/* Don't disable bottom halves while acquiring the lock to
>   		 * avoid circular locking dependency on PREEMPT_RT.
>   		 */
> -		spin_lock(&ilb2->lock);
> +		if (!has_current_bpf_ctx())
> +			spin_lock(&ilb2->lock);
>   		if (sk_unhashed(sk)) {
> -			spin_unlock(&ilb2->lock);
> +			if (!has_current_bpf_ctx())
> +				spin_unlock(&ilb2->lock);

That's bucket lock, why do we have to skip it?

>   			return;
>   		}

> @@ -788,7 +790,8 @@ void inet_unhash(struct sock *sk)

>   		__sk_nulls_del_node_init_rcu(sk);
>   		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
> -		spin_unlock(&ilb2->lock);
> +		if (!has_current_bpf_ctx())
> +			spin_unlock(&ilb2->lock);
>   	} else {
>   		spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);

> --
> 2.34.1