Re: [PATCH] tls: don't abort the connection on signal-interrupted sends

[Maximilian Immanuel Brandtner <maxbr@linux.ibm.com>]

On Thu, 2026-07-23 at 09:26 -0700, Jakub Kicinski wrote:
> On Mon, 20 Jul 2026 11:08:47 +0200 Maximilian Immanuel Brandtner
> wrote:
> > When a signal interrupts a blocking send, tls_tx_records() treats
> > the
> > resulting -ERESTARTSYS as a transmission failure and marks the
> > socket
> > errored via tls_err_abort() with the raw error code. Later syscalls
> > return the kernel-internal errno 512 (ERESTARTSYS) to userspace, as
> > the
> > signal it stems from is no longer pending during syscall exit and
> > thus
> > never translated.
> 
> Can we just add ERESTARTSYS handling? I never heard of the other
> codes
> you're checking TBH, can they actually surface?

I don't know whether three are any drivers that call these error codes
in practice. However, I know that these other errnos also get handled,
by the signal re-entrancy logic (for example see `handle_signal()` in
`arch/x86/kernel/signal.c`). I think all these error codes should
propagate to the relevant signal logic to be converted to the right
error code and avoid leaking kernel-internal error codes to user-space.

Maybe EINTR could be removed from the patch. I just kept it in just in
case.

> 
> > An interrupted send is not a connection error: the partially sent
> > record
> > stays queued and is resent later. Interrupt error codes are
> > therefore
> > excluded from the abort in the same way as -EAGAIN.
> > 
> > Fixes: b341ca51d267 ("tls: Fix tls_sw_sendmsg error handling")
> > Signed-off-by: Maximilian Immanuel Brandtner <maxbr@linux.ibm.com>
> > ---
> > Tested on top of net commit 3f1f75536668 ("net: openvswitch: reject
> > oversized nested action attrs").
> > 
> > Several subsystems contain static helper functions to classify
> > these
> > interrupt errnos. It might be worthwhile to refactor these static
> > functions into a generic helper function.
> 
> Yes, please.

I'll prepare a seperate patch-set for that then. Note this helper will
only cover `ERESTARTSYS`, `ERESTARTNOINTR`, `ERESTARTNOHAND`, and
`ERESTART_RESTARTBLOCK`, because there are a bunch of places in the
kernel where these 4 error codes in particular are checked against.
> 
> > diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
> > index d4afc90fd796..3c9e94069e82 100644
> > --- a/net/tls/tls_sw.c
> > +++ b/net/tls/tls_sw.c
> > @@ -405,6 +405,24 @@ static void tls_free_open_rec(struct sock *sk)
> >  	}
> >  }
> >  
> > +static bool tls_is_non_restartable_err(int err)
> > +{
> > +	if (err >= 0)
> > +		return false;
> 
> This doesn't belong in a helper for classifying errors.

So you would prefer if (rc < 0 && !tls_is_restartable_err(rc))

> 
> > +	switch (err) {
> > +	case -EAGAIN:
> > +	case -EINTR:
> > +	case -ERESTARTSYS:
> > +	case -ERESTARTNOINTR:
> > +	case -ERESTARTNOHAND:
> > +	case -ERESTART_RESTARTBLOCK:
> > +		return false;
> > +	default:
> > +		return true;
> > +	}
> > +}
> > +
> >  int tls_tx_records(struct sock *sk, int flags)
> >  {
> >  	struct tls_context *tls_ctx = tls_get_ctx(sk);
> > @@ -458,7 +476,7 @@ int tls_tx_records(struct sock *sk, int flags)
> >  	}
> >  
> >  tx_err:
> > -	if (rc < 0 && rc != -EAGAIN)
> > +	if (tls_is_non_restartable_err(rc))
> >  		tls_err_abort(sk, rc);
> >  
> >  	return rc;