Re: [PATCH net v2 1/2] tls: handle data disappearing from under the TLS ULP

[Paolo Abeni <pabeni@redhat.com>]

On 8/12/25 3:28 PM, Jakub Kicinski wrote:
> On Tue, 12 Aug 2025 12:45:56 +0200 Paolo Abeni wrote:
>> On 8/8/25 1:29 AM, Jakub Kicinski wrote:
>>> diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
>>> index 549d1ea01a72..51c98a007dda 100644
>>> --- a/net/tls/tls_sw.c
>>> +++ b/net/tls/tls_sw.c
>>> @@ -1384,7 +1384,8 @@ tls_rx_rec_wait(struct sock *sk, struct sk_psock *psock, bool nonblock,
>>>  			return sock_intr_errno(timeo);
>>>  	}
>>>  
>>> -	tls_strp_msg_load(&ctx->strp, released);
>>> +	if (unlikely(!tls_strp_msg_load(&ctx->strp, released)))
>>> +		return tls_rx_rec_wait(sk, psock, nonblock, false);  
>>
>> I'm probably missing something relevant, but I don't see anything
>> preventing the above recursion from going very deep and cause stack
>> overflow.
>>
>> Perhaps something alike:
>>
>> 		released = false;
>> 		goto <function start>
>>
>> would be safer?
> 
> It's a tail call to the same function, the compiler should do that for
> us automatically. Can we not trust the compiler to be sensible? Both
> clang and gcc get it right.

Sound reasonable, I dumbly did not consider it. I'm fine with the patch
in the current form.

/P