rxrpc: Fix ICMP/ICMP6 error handling
authorDavid Howells <dhowells@redhat.com>
Fri, 26 Aug 2022 14:39:28 +0000 (15:39 +0100)
committerDavid Howells <dhowells@redhat.com>
Thu, 1 Sep 2022 10:42:12 +0000 (11:42 +0100)
commitac56a0b48da86fd1b4389632fb7c4c8a5d86eefa
treea066ce6fd509bd85b3eefaf8086aee19a3799094
parentf612466ebecb12a00d9152344ddda6f6345f04dc
rxrpc: Fix ICMP/ICMP6 error handling

Because rxrpc pretends to be a tunnel on top of a UDP/UDP6 socket, allowing
it to siphon off UDP packets early in the handling of received UDP packets
thereby avoiding the packet going through the UDP receive queue, it doesn't
get ICMP packets through the UDP ->sk_error_report() callback.  In fact, it
doesn't appear that there's any usable option for getting hold of ICMP
packets.

Fix this by adding a new UDP encap hook to distribute error messages for
UDP tunnels.  If the hook is set, then the tunnel driver will be able to
see ICMP packets.  The hook provides the offset into the packet of the UDP
header of the original packet that caused the notification.

An alternative would be to call the ->error_handler() hook - but that
requires that the skbuff be cloned (as ip_icmp_error() or ipv6_cmp_error()
do, though isn't really necessary or desirable in rxrpc's case is we want
to parse them there and then, not queue them).

Changes
=======
ver #3)
 - Fixed an uninitialised variable.

ver #2)
 - Fixed some missing CONFIG_AF_RXRPC_IPV6 conditionals.

Fixes: 5271953cad31 ("rxrpc: Use the UDP encap_rcv hook")
Signed-off-by: David Howells <dhowells@redhat.com>
include/linux/udp.h
include/net/udp_tunnel.h
net/ipv4/udp.c
net/ipv4/udp_tunnel_core.c
net/ipv6/udp.c
net/rxrpc/ar-internal.h
net/rxrpc/local_object.c
net/rxrpc/peer_event.c