The ethertype should always be updated on mpls_pop
as there may be a transition between MPLS unicast (0x8847) and
MPLS multicast (0x8848).
Ben Pfaff tells me that this is consistent with the
behaviour described in EXT-194 of the JIRA bug tracker.
Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: Ben Pfaff <blp@nicira.com>
size_t len;
mh = packet->l2_5;
len = (char*)packet->l2_5 - (char*)packet->l2;
- /* If bottom of the stack set ethertype. */
+ set_ethertype(packet, ethtype);
if (mh->mpls_lse & htonl(MPLS_BOS_MASK)) {
- set_ethertype(packet, ethtype);
packet->l2_5 = NULL;
} else {
packet->l2_5 = (char*)packet->l2_5 + MPLS_HLEN;
\fBpush_mpls\fR being discarded.
.
.IP \fBpop_mpls\fR:\fIethertype\fR
-Strips the outermost MPLS label stack entry. If the MPLS label
-stripped was the only one, changes the ethertype of a packet to
-\fIethertype\fR, which should not ordinarily be an MPLS Ethertype.
+Strips the outermost MPLS label stack entry.
+Currently the implementation restricts \fIethertype\fR to a non-MPLS Ethertype
+and thus \fBpop_mpls\fR should only be applied to packets with
+an MPLS label stack depth of one.
.
.IP
There are some limitations in the implementation. \fBpop_mpls\fR