Extend the interface to ether_input(): a NULL eh pointer means that

the mbuf contains the ethernet header (eh) as well, which ether_input()
will strip off as needed.

This permits the removal (in a backward compatible way) of the
header removal code which right now is replicated in all drivers,
sometimes in an inconsistent way. Also, because many functions
called after ether_input() require the eh in the mbuf, eventually
we can propagate the interface and handle outdated drivers just
in ether_input().

Individual driver changes to use the new interface will follow as
we have a chance to touch them.

NOTE THAT THIS CHANGE IS FULLY BACKWARD COMPATIBLE AND DOES NOT BREAK
BINARY COMPATIBILITY FOR DRIVERS.

MFC after: 3 days

1 files changed, 26 insertions, 12 deletions

diff --git a/sys/net/if_ethersubr.c b/sys/net/if_ethersubr.c
index 693f184..53c744d 100644
--- a/sys/net/if_ethersubr.c
+++ b/sys/net/if_ethersubr.c
@@ -535,30 +535,44 @@ ether_ipfw_chk(struct mbuf **m0, struct ifnet *dst,
 }
 
 /*
- * Process a received Ethernet packet;
- * the packet is in the mbuf chain m without
- * the ether header, which is provided separately.
+ * Process a received Ethernet packet. We have two different interfaces:
+ * one (conventional) assumes the packet in the mbuf, with the ethernet
+ * header provided separately in *eh. The second one (new) has everything
+ * in the mbuf, and we can tell it because eh == NULL.
+ * The caller MUST MAKE SURE that there are at least
+ * sizeof(struct ether_header) bytes in the first mbuf.
+ *
+ * This allows us to concentrate in one place a bunch of code which
+ * is replicated in all device drivers. Also, many functions called
+ * from ether_input() try to put the eh back into the mbuf, so we
+ * can later propagate the 'contiguous packet' interface to them,
+ * and handle the old interface just here.
  *
  * NOTA BENE: for many drivers "eh" is a pointer into the first mbuf or
  * cluster, right before m_data. So be very careful when working on m,
  * as you could destroy *eh !!
- * A (probably) more convenient and efficient interface to ether_input
- * is to have the whole packet (with the ethernet header) into the mbuf:
- * modules which do not need the ethernet header can easily drop it, while
- * others (most noticeably bridge and ng_ether) do not need to do additional
- * work to put the ethernet header back into the mbuf.
  *
  * First we perform any link layer operations, then continue
  * to the upper layers with ether_demux().
  */
 void
-ether_input(ifp, eh, m)
-	struct ifnet *ifp;
-	struct ether_header *eh;
-	struct mbuf *m;
+ether_input(struct ifnet *ifp, struct ether_header *eh, struct mbuf *m)
 {
 	struct ether_header save_eh;
 
+	if (eh == NULL) {
+		if (m->m_len < sizeof(struct ether_header)) {
+			/* XXX error in the caller. */
+			m_freem(m);
+			return;
+		}
+		m->m_pkthdr.rcvif = ifp;
+		eh = mtod(m, struct ether_header *);
+		m->m_data += sizeof(struct ether_header);
+		m->m_len -= sizeof(struct ether_header);
+		m->m_pkthdr.len = m->m_len;
+	}
+
 #ifdef MAC
 	mac_create_mbuf_from_ifnet(ifp, m);
 #endif