Allow userspace to end a TLS record without supplying any data by calling
send()/sendto()/sendmsg() with no data and no MSG_MORE flag.  This can be
used to flush a previous send/splice that had MSG_MORE or SPLICE_F_MORE set
or a sendfile() that was incomplete.

Without this, a zero-length send to tls-sw is just ignored.  I think
tls-device will do the right thing without modification.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Chuck Lever <chuck.lever@oracle.com>
cc: Boris Pismenny <borisp@nvidia.com>
cc: John Fastabend <john.fastabend@gmail.com>
cc: Jakub Kicinski <kuba@kernel.org>
cc: Eric Dumazet <edumazet@google.com>
cc: "David S. Miller" <davem@davemloft.net>
cc: Paolo Abeni <pabeni@redhat.com>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
cc: netdev@vger.kernel.org
---
 net/tls/tls_sw.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
index cac1adc968e8..6aa6d17888f5 100644
--- a/net/tls/tls_sw.c
+++ b/net/tls/tls_sw.c
@@ -945,7 +945,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 	struct tls_rec *rec;
 	int required_size;
 	int num_async = 0;
-	bool full_record;
+	bool full_record = false;
 	int record_room;
 	int num_zc = 0;
 	int orig_size;
@@ -971,6 +971,9 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 		}
 	}
 
+	if (!msg_data_left(msg) && eor)
+		goto just_flush;
+
 	while (msg_data_left(msg)) {
 		if (sk->sk_err) {
 			ret = -sk->sk_err;
@@ -1082,6 +1085,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
 		 */
 		tls_ctx->pending_open_record_frags = true;
 		copied += try_to_copy;
+just_flush:
 		if (full_record || eor) {
 			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
 						  record_type, &copied,

+ dan Carpenter

On Fri, Jun 02, 2023 at 04:07:44PM +0100, David Howells wrote:
> Allow userspace to end a TLS record without supplying any data by calling
> send()/sendto()/sendmsg() with no data and no MSG_MORE flag.  This can be
> used to flush a previous send/splice that had MSG_MORE or SPLICE_F_MORE set
> or a sendfile() that was incomplete.
> 
> Without this, a zero-length send to tls-sw is just ignored.  I think
> tls-device will do the right thing without modification.
> 
> Signed-off-by: David Howells <dhowells@redhat.com>
> cc: Chuck Lever <chuck.lever@oracle.com>
> cc: Boris Pismenny <borisp@nvidia.com>
> cc: John Fastabend <john.fastabend@gmail.com>
> cc: Jakub Kicinski <kuba@kernel.org>
> cc: Eric Dumazet <edumazet@google.com>
> cc: "David S. Miller" <davem@davemloft.net>
> cc: Paolo Abeni <pabeni@redhat.com>
> cc: Jens Axboe <axboe@kernel.dk>
> cc: Matthew Wilcox <willy@infradead.org>
> cc: netdev@vger.kernel.org
> ---
>  net/tls/tls_sw.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
> index cac1adc968e8..6aa6d17888f5 100644
> --- a/net/tls/tls_sw.c
> +++ b/net/tls/tls_sw.c
> @@ -945,7 +945,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
>  	struct tls_rec *rec;
>  	int required_size;
>  	int num_async = 0;
> -	bool full_record;
> +	bool full_record = false;
>  	int record_room;
>  	int num_zc = 0;
>  	int orig_size;
> @@ -971,6 +971,9 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
>  		}
>  	}
>  
> +	if (!msg_data_left(msg) && eor)
> +		goto just_flush;
> +

Hi David,

the flow of this function is not entirely simple, so it is not easy for me
to manually verify this. But in combination gcc-12 -Wmaybe-uninitialized
and Smatch report that the following may be used uninitialised as a result
of this change:

 * msg_pl
 * orig_size
 * msg_en
 * required_size
 * try_to_copy

>  	while (msg_data_left(msg)) {
>  		if (sk->sk_err) {
>  			ret = -sk->sk_err;
> @@ -1082,6 +1085,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
>  		 */
>  		tls_ctx->pending_open_record_frags = true;
>  		copied += try_to_copy;
> +just_flush:
>  		if (full_record || eor) {
>  			ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
>  						  record_type, &copied,
> 
>

On Fri, Jun 02, 2023 at 08:27:56PM +0200, Simon Horman wrote:
> + dan Carpenter
> 
> On Fri, Jun 02, 2023 at 04:07:44PM +0100, David Howells wrote:
> > Allow userspace to end a TLS record without supplying any data by calling
> > send()/sendto()/sendmsg() with no data and no MSG_MORE flag.  This can be
> > used to flush a previous send/splice that had MSG_MORE or SPLICE_F_MORE set
> > or a sendfile() that was incomplete.
> > 
> > Without this, a zero-length send to tls-sw is just ignored.  I think
> > tls-device will do the right thing without modification.
> > 
> > Signed-off-by: David Howells <dhowells@redhat.com>
> > cc: Chuck Lever <chuck.lever@oracle.com>
> > cc: Boris Pismenny <borisp@nvidia.com>
> > cc: John Fastabend <john.fastabend@gmail.com>
> > cc: Jakub Kicinski <kuba@kernel.org>
> > cc: Eric Dumazet <edumazet@google.com>
> > cc: "David S. Miller" <davem@davemloft.net>
> > cc: Paolo Abeni <pabeni@redhat.com>
> > cc: Jens Axboe <axboe@kernel.dk>
> > cc: Matthew Wilcox <willy@infradead.org>
> > cc: netdev@vger.kernel.org
> > ---
> >  net/tls/tls_sw.c | 6 +++++-
> >  1 file changed, 5 insertions(+), 1 deletion(-)
> > 
> > diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
> > index cac1adc968e8..6aa6d17888f5 100644
> > --- a/net/tls/tls_sw.c
> > +++ b/net/tls/tls_sw.c
> > @@ -945,7 +945,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
> >  	struct tls_rec *rec;
> >  	int required_size;
> >  	int num_async = 0;
> > -	bool full_record;
> > +	bool full_record = false;
> >  	int record_room;
> >  	int num_zc = 0;
> >  	int orig_size;
> > @@ -971,6 +971,9 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
> >  		}
> >  	}
> >  
> > +	if (!msg_data_left(msg) && eor)
> > +		goto just_flush;
> > +
> 
> Hi David,
> 
> the flow of this function is not entirely simple, so it is not easy for me
> to manually verify this. But in combination gcc-12 -Wmaybe-uninitialized
> and Smatch report that the following may be used uninitialised as a result
> of this change:
> 
>  * msg_pl

This warning seems correct to me.

>  * orig_size

This warning assumes we hit the first warning and then hit the goto
wait_for_memory;

>  * msg_en

I don't get this warning on my system but it's the same thing.  Hit the
first warning then the goto wait_for_memory.

>  * required_size

Same.

>  * try_to_copy

I don't really understand this warning and I can't reproduce it.
Strange.

regards,
dan carpenter

On Fri, Jun 02, 2023 at 10:00:45PM +0300, Dan Carpenter wrote:
> On Fri, Jun 02, 2023 at 08:27:56PM +0200, Simon Horman wrote:
> > + dan Carpenter
> > 
> > On Fri, Jun 02, 2023 at 04:07:44PM +0100, David Howells wrote:
> > > Allow userspace to end a TLS record without supplying any data by calling
> > > send()/sendto()/sendmsg() with no data and no MSG_MORE flag.  This can be
> > > used to flush a previous send/splice that had MSG_MORE or SPLICE_F_MORE set
> > > or a sendfile() that was incomplete.
> > > 
> > > Without this, a zero-length send to tls-sw is just ignored.  I think
> > > tls-device will do the right thing without modification.
> > > 
> > > Signed-off-by: David Howells <dhowells@redhat.com>
> > > cc: Chuck Lever <chuck.lever@oracle.com>
> > > cc: Boris Pismenny <borisp@nvidia.com>
> > > cc: John Fastabend <john.fastabend@gmail.com>
> > > cc: Jakub Kicinski <kuba@kernel.org>
> > > cc: Eric Dumazet <edumazet@google.com>
> > > cc: "David S. Miller" <davem@davemloft.net>
> > > cc: Paolo Abeni <pabeni@redhat.com>
> > > cc: Jens Axboe <axboe@kernel.dk>
> > > cc: Matthew Wilcox <willy@infradead.org>
> > > cc: netdev@vger.kernel.org
> > > ---
> > >  net/tls/tls_sw.c | 6 +++++-
> > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/net/tls/tls_sw.c b/net/tls/tls_sw.c
> > > index cac1adc968e8..6aa6d17888f5 100644
> > > --- a/net/tls/tls_sw.c
> > > +++ b/net/tls/tls_sw.c
> > > @@ -945,7 +945,7 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
> > >  	struct tls_rec *rec;
> > >  	int required_size;
> > >  	int num_async = 0;
> > > -	bool full_record;
> > > +	bool full_record = false;
> > >  	int record_room;
> > >  	int num_zc = 0;
> > >  	int orig_size;
> > > @@ -971,6 +971,9 @@ int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
> > >  		}
> > >  	}
> > >  
> > > +	if (!msg_data_left(msg) && eor)
> > > +		goto just_flush;
> > > +
> > 
> > Hi David,
> > 
> > the flow of this function is not entirely simple, so it is not easy for me
> > to manually verify this. But in combination gcc-12 -Wmaybe-uninitialized
> > and Smatch report that the following may be used uninitialised as a result
> > of this change:
> > 
> >  * msg_pl
> 
> This warning seems correct to me.
> 
> >  * orig_size
> 
> This warning assumes we hit the first warning and then hit the goto
> wait_for_memory;
> 
> >  * msg_en
> 
> I don't get this warning on my system but it's the same thing.  Hit the
> first warning then the goto wait_for_memory.
> 
> >  * required_size
> 
> Same.
> 
> >  * try_to_copy
> 
> I don't really understand this warning and I can't reproduce it.
> Strange.

Thanks Dan.

Of the above I think only the last one was flagged
by GCC but not Smatch. I can try investigating further if it is useful.