Re: [PATCH bpf] bpf,tcp: avoid infinite recursion in BPF_SOCK_OPS_HDR_OPT_LEN_CB

[Jiayuan Chen <jiayuan.chen@linux.dev>]

On 4/14/26 11:37 PM, mkf wrote:
> On Tue, 2026-04-14 at 18:57 +0800, Jiayuan Chen wrote:


[...]

> --- a/include/linux/tcp.h
> +++ b/include/linux/tcp.h
> @@ -475,12 +475,21 @@ struct tcp_sock {
>   	u8	bpf_sock_ops_cb_flags;  /* Control calling BPF programs
>   					 * values defined in uapi/linux/tcp.h
>   					 */
> -	u8	bpf_chg_cc_inprogress:1; /* In the middle of
> +	u8	bpf_chg_cc_inprogress:1, /* In the middle of
>   					  * bpf_setsockopt(TCP_CONGESTION),
>   					  * it is to avoid the bpf_tcp_cc->init()
>   					  * to recur itself by calling
>   					  * bpf_setsockopt(TCP_CONGESTION, "itself").
>   					  */
> +		bpf_hdr_opt_len_cb_inprogress:1; /* It is set before invoking the
> +						  * callback so that a nested
> +						  * bpf_setsockopt(TCP_NODELAY) or
> +						  * bpf_setsockopt(TCP_CORK) cannot
> +						  * trigger tcp_push_pending_frames(),
> +						  * which would call tcp_current_mss()
> +						  * -> bpf_skops_hdr_opt_len(), causing
> +						  * infinite recursion.
> +						  */
>   #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
>   #else
>   #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
> diff --git a/net/core/filter.c b/net/core/filter.c
> index 78b548158fb0..518699429a7a 100644
> --- a/net/core/filter.c
> +++ b/net/core/filter.c
> @@ -5483,6 +5483,10 @@ static int sol_tcp_sockopt(struct sock *sk, int optname,
>   	if (sk->sk_protocol != IPPROTO_TCP)
>   		return -EINVAL;
>   
> +	if ((optname == TCP_NODELAY || optname == TCP_CORK) &&
> +	    tcp_sk(sk)->bpf_hdr_opt_len_cb_inprogress)
> +		return -EBUSY;
> +
> TCP_CORK is not support in sol_tcp_sockopt(), return -EINVAL by default. and put the check here
> could also prevent us from calling getsockopt(TCP_NODELAY) below.
>
>>   	switch (optname) {
>>   	case TCP_NODELAY:
>>   	case TCP_MAXSEG:
>> diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
>> index dafb63b923d0..fb06c464ac16 100644
>> --- a/net/ipv4/tcp_minisocks.c
>> +++ b/net/ipv4/tcp_minisocks.c
>> @@ -663,6 +663,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
>>   	RCU_INIT_POINTER(newtp->fastopen_rsk, NULL);
>>   
>>   	newtp->bpf_chg_cc_inprogress = 0;
>> +	newtp->bpf_hdr_opt_len_cb_inprogress = 0;
>>   	tcp_bpf_clone(sk, newsk);
>>   
>>   	__TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
>> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
>> index 326b58ff1118..c9654e690e1a 100644
>> --- a/net/ipv4/tcp_output.c
>> +++ b/net/ipv4/tcp_output.c
>> @@ -475,6 +475,7 @@ static void bpf_skops_hdr_opt_len(struct sock *sk, struct sk_buff *skb,
>>   				  unsigned int *remaining)
>>   {
>>   	struct bpf_sock_ops_kern sock_ops;
>> +	struct tcp_sock *tp = tcp_sk(sk);
>>   	int err;
>>   
>>   	if (likely(!BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk),
>> @@ -519,7 +520,9 @@ static void bpf_skops_hdr_opt_len(struct sock *sk, struct sk_buff *skb,
>>   	if (skb)
>>   		bpf_skops_init_skb(&sock_ops, skb, 0);
>>   
>> +	tp->bpf_hdr_opt_len_cb_inprogress = 1;
> we check the BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG before calling BPF_CGROUP_RUN_PROG_SOCK_OPS_SK,
> could this flag use for the same purpose? so we don't need to add an extra field.
>
> 	if (likely(!BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk),
> 					   BPF_SOCK_OPS_WRITE_HDR_OPT_CB_FLAG)) ||
> 	    !*remaining)
> 		return;


Hi Martin, I saw your patch. Your solution is better, please ignore mine :)