if (port_includes_vlan(port, m->out_vlan)
&& set_dst(dst, flow, in_port, port, tags))
{
+ int flow_vlan;
+
if (port->vlan < 0) {
dst->vlan = m->out_vlan;
}
if (dst_is_duplicate(dsts, dst - dsts, dst)) {
continue;
}
- if (dst->dp_ifidx == flow->in_port
- && dst->vlan == vlan) {
+
+ /* Use the vlan tag on the original flow instead of
+ * the one passed in the vlan parameter. This ensures
+ * that we compare the vlan from before any implicit
+ * tagging tags place. This is necessary because
+ * dst->vlan is the final vlan, after removing implicit
+ * tags. */
+ flow_vlan = ntohs(flow->dl_vlan);
+ if (flow_vlan == 0) {
+ flow_vlan = OFP_VLAN_NONE;
+ }
+ if (port == in_port && dst->vlan == flow_vlan) {
/* Don't send out input port on same VLAN. */
continue;
}
goto done;
}
- /* Multicast (and broadcast) packets on bonds need special attention, to
- * avoid receiving duplicates. */
- if (in_port->n_ifaces > 1 && eth_addr_is_multicast(flow->dl_dst)) {
- *tags |= in_port->active_iface_tag;
- if (in_port->active_iface != in_iface->port_ifidx) {
- /* Drop all multicast packets on inactive slaves. */
- goto done;
- } else {
- /* Drop all multicast packets for which we have learned a different
- * input port, because we probably sent the packet on one slave
- * and got it back on the active slave. Broadcast ARP replies are
- * an exception to this rule: the host has moved to another
- * switch. */
- int src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
- if (src_idx != -1 && src_idx != in_port->port_idx) {
- if (packet) {
- if (!is_bcast_arp_reply(flow, packet)) {
- goto done;
- }
- } else {
- /* No way to know whether it's an ARP reply, because the
- * flow entry doesn't include enough information and we
- * don't have a packet. Punt. */
- return false;
- }
+ /* Packets received on bonds need special attention to avoid duplicates. */
+ if (in_port->n_ifaces > 1) {
+ int src_idx;
+
+ if (eth_addr_is_multicast(flow->dl_dst)) {
+ *tags |= in_port->active_iface_tag;
+ if (in_port->active_iface != in_iface->port_ifidx) {
+ /* Drop all multicast packets on inactive slaves. */
+ goto done;
}
}
+
+ /* Drop all packets for which we have learned a different input
+ * port, because we probably sent the packet on one slave and got
+ * it back on the other. Broadcast ARP replies are an exception
+ * to this rule: the host has moved to another switch. */
+ src_idx = mac_learning_lookup(br->ml, flow->dl_src, vlan);
+ if (src_idx != -1 && src_idx != in_port->port_idx &&
+ (!packet || !is_bcast_arp_reply(flow, packet))) {
+ goto done;
+ }
}
/* MAC learning. */
tags);
if (out_port_idx >= 0 && out_port_idx < br->n_ports) {
out_port = br->ports[out_port_idx];
+ } else if (!packet) {
+ /* If we are revalidating but don't have a learning entry then
+ * eject the flow. Installing a flow that floods packets will
+ * prevent us from seeing future packets and learning properly. */
+ return false;
}
}
/* Shifts 'hash' from 'from' to 'to' within 'port'. */
static void
bond_shift_load(struct slave_balance *from, struct slave_balance *to,
- struct bond_entry *hash)
+ int hash_idx)
{
+ struct bond_entry *hash = from->hashes[hash_idx];
struct port *port = from->iface->port;
uint64_t delta = hash->tx_bytes;
* it require more work, the only purpose it would be to allow that hash to
* be migrated to another slave in this rebalancing run, and there is no
* point in doing that. */
- if (from->hashes[0] == hash) {
+ if (hash_idx == 0) {
from->hashes++;
} else {
- int i = hash - from->hashes[0];
- memmove(from->hashes + i, from->hashes + i + 1,
- (from->n_hashes - (i + 1)) * sizeof *from->hashes);
+ memmove(from->hashes + hash_idx, from->hashes + hash_idx + 1,
+ (from->n_hashes - (hash_idx + 1)) * sizeof *from->hashes);
}
from->n_hashes--;
/* 'from' is carrying significantly more load than 'to', and that
* load is split across at least two different hashes. Pick a hash
* to migrate to 'to' (the least-loaded slave), given that doing so
- * must not cause 'to''s load to exceed 'from''s load.
+ * must decrease the ratio of the load on the two slaves by at
+ * least 0.1.
*
* The sort order we use means that we prefer to shift away the
* smallest hashes instead of the biggest ones. There is little
* reason behind this decision; we could use the opposite sort
* order to shift away big hashes ahead of small ones. */
size_t i;
+ bool order_swapped;
for (i = 0; i < from->n_hashes; i++) {
+ double old_ratio, new_ratio;
uint64_t delta = from->hashes[i]->tx_bytes;
- if (to->tx_bytes + delta < from->tx_bytes - delta) {
+
+ if (delta == 0 || from->tx_bytes - delta == 0) {
+ /* Pointless move. */
+ continue;
+ }
+
+ order_swapped = from->tx_bytes - delta < to->tx_bytes + delta;
+
+ if (to->tx_bytes == 0) {
+ /* Nothing on the new slave, move it. */
+ break;
+ }
+
+ old_ratio = (double)from->tx_bytes / to->tx_bytes;
+ new_ratio = (double)(from->tx_bytes - delta) /
+ (to->tx_bytes + delta);
+
+ if (new_ratio == 0) {
+ /* Should already be covered but check to prevent division
+ * by zero. */
+ continue;
+ }
+
+ if (new_ratio < 1) {
+ new_ratio = 1 / new_ratio;
+ }
+
+ if (old_ratio - new_ratio > 0.1) {
+ /* Would decrease the ratio, move it. */
break;
}
}
if (i < from->n_hashes) {
- bond_shift_load(from, to, from->hashes[i]);
+ bond_shift_load(from, to, i);
+ port->bond_compat_is_stale = true;
+
+ /* If the result of the migration changed the relative order of
+ * 'from' and 'to' swap them back to maintain invariants. */
+ if (order_swapped) {
+ swap_bals(from, to);
+ }
/* Re-sort 'bals'. Note that this may make 'from' and 'to'
* point to different slave_balance structures. It is only
} else {
from++;
}
- port->bond_compat_is_stale = true;
}
}
git://git.onelab.eu / sliver-openvswitch.git / blobdiff