mpls: flow-based multipath selection
Change the selection of a multipath route to use a flow-based
hash. This more suitable for traffic sensitive to reordering within a
flow (e.g. TCP, L2VPN) and whilst still allowing a good distribution
of traffic given enough flows.
Selection of the path for a multipath route is done using a hash of:
1. Label stack up to MAX_MP_SELECT_LABELS labels or up to and
including entropy label, whichever is first.
2. 3-tuple of (L3 src, L3 dst, proto) from IPv4/IPv6 header in MPLS
payload, if present.
Naturally, a 5-tuple hash using L4 information in addition would be
possible and be better in some scenarios, but there is a tradeoff
between looking deeper into the packet to achieve good distribution,
and packet forwarding performance, and I have erred on the side of the
latter as the default.
Signed-off-by: Robert Shearman <rshearma@brocade.com>
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/mpls/af_mpls.c b/net/mpls/af_mpls.c
index ee3097a..cc972e3 100644
--- a/net/mpls/af_mpls.c
+++ b/net/mpls/af_mpls.c
@@ -22,6 +22,11 @@
#include <net/nexthop.h>
#include "internal.h"
+/* Maximum number of labels to look ahead at when selecting a path of
+ * a multipath route
+ */
+#define MAX_MP_SELECT_LABELS 4
+
static int zero = 0;
static int label_limit = (1 << 20) - 1;
@@ -77,10 +82,78 @@
}
EXPORT_SYMBOL_GPL(mpls_pkt_too_big);
-static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt)
+static struct mpls_nh *mpls_select_multipath(struct mpls_route *rt,
+ struct sk_buff *skb, bool bos)
{
- /* assume single nexthop for now */
- return &rt->rt_nh[0];
+ struct mpls_entry_decoded dec;
+ struct mpls_shim_hdr *hdr;
+ bool eli_seen = false;
+ int label_index;
+ int nh_index = 0;
+ u32 hash = 0;
+
+ /* No need to look further into packet if there's only
+ * one path
+ */
+ if (rt->rt_nhn == 1)
+ goto out;
+
+ for (label_index = 0; label_index < MAX_MP_SELECT_LABELS && !bos;
+ label_index++) {
+ if (!pskb_may_pull(skb, sizeof(*hdr) * label_index))
+ break;
+
+ /* Read and decode the current label */
+ hdr = mpls_hdr(skb) + label_index;
+ dec = mpls_entry_decode(hdr);
+
+ /* RFC6790 - reserved labels MUST NOT be used as keys
+ * for the load-balancing function
+ */
+ if (likely(dec.label >= MPLS_LABEL_FIRST_UNRESERVED)) {
+ hash = jhash_1word(dec.label, hash);
+
+ /* The entropy label follows the entropy label
+ * indicator, so this means that the entropy
+ * label was just added to the hash - no need to
+ * go any deeper either in the label stack or in the
+ * payload
+ */
+ if (eli_seen)
+ break;
+ } else if (dec.label == MPLS_LABEL_ENTROPY) {
+ eli_seen = true;
+ }
+
+ bos = dec.bos;
+ if (bos && pskb_may_pull(skb, sizeof(*hdr) * label_index +
+ sizeof(struct iphdr))) {
+ const struct iphdr *v4hdr;
+
+ v4hdr = (const struct iphdr *)(mpls_hdr(skb) +
+ label_index);
+ if (v4hdr->version == 4) {
+ hash = jhash_3words(ntohl(v4hdr->saddr),
+ ntohl(v4hdr->daddr),
+ v4hdr->protocol, hash);
+ } else if (v4hdr->version == 6 &&
+ pskb_may_pull(skb, sizeof(*hdr) * label_index +
+ sizeof(struct ipv6hdr))) {
+ const struct ipv6hdr *v6hdr;
+
+ v6hdr = (const struct ipv6hdr *)(mpls_hdr(skb) +
+ label_index);
+
+ hash = __ipv6_addr_jhash(&v6hdr->saddr, hash);
+ hash = __ipv6_addr_jhash(&v6hdr->daddr, hash);
+ hash = jhash_1word(v6hdr->nexthdr, hash);
+ }
+ }
+ }
+
+ nh_index = hash % rt->rt_nhn;
+out:
+ return &rt->rt_nh[nh_index];
}
static bool mpls_egress(struct mpls_route *rt, struct sk_buff *skb,
@@ -175,7 +248,7 @@
if (!rt)
goto drop;
- nh = mpls_select_multipath(rt);
+ nh = mpls_select_multipath(rt, skb, dec.bos);
if (!nh)
goto drop;
@@ -541,6 +614,12 @@
if (!rtnh_ok(rtnh, remaining))
goto errout;
+ /* neither weighted multipath nor any flags
+ * are supported
+ */
+ if (rtnh->rtnh_hops || rtnh->rtnh_flags)
+ goto errout;
+
attrlen = rtnh_attrlen(rtnh);
if (attrlen > 0) {
struct nlattr *attrs = rtnh_attrs(rtnh);