samples/bpf: xdp_redirect_cpu load balance like Suricata

This implement XDP CPU redirection load-balancing across available
CPUs, based on the hashing IP-pairs + L4-protocol.  This equivalent to
xdp-cpu-redirect feature in Suricata, which is inspired by the
Suricata 'ippair' hashing code.

An important property is that the hashing is flow symmetric, meaning
that if the source and destination gets swapped then the selected CPU
will remain the same.  This is helps locality by placing both directions
of a flows on the same CPU, in a forwarding/routing scenario.

The hashing INITVAL (15485863 the 10^6th prime number) was fairly
arbitrary choosen, but experiments with kernel tree pktgen scripts
(pktgen_sample04_many_flows.sh +pktgen_sample05_flow_per_thread.sh)
showed this improved the distribution.

This patch also change the default loaded XDP program to be this
load-balancer.  As based on different user feedback, this seems to be
the expected behavior of the sample xdp_redirect_cpu.

Link: https://github.com/OISF/suricata/commit/796ec08dd7a63
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

2 files changed, 105 insertions, 2 deletions

diff --git a/samples/bpf/xdp_redirect_cpu_kern.c b/samples/bpf/xdp_redirect_cpu_kern.c
index 8cb703671b04..081ef4bb4fe3 100644
--- a/samples/bpf/xdp_redirect_cpu_kern.c
+++ b/samples/bpf/xdp_redirect_cpu_kern.c
@@ -13,6 +13,7 @@
 
 #include <uapi/linux/bpf.h>
 #include "bpf_helpers.h"
+#include "hash_func01.h"
 
 #define MAX_CPUS 12 /* WARNING - sync with _user.c */
 
@@ -461,6 +462,108 @@ int  xdp_prognum4_ddos_filter_pktgen(struct xdp_md *ctx)
 	return bpf_redirect_map(&cpu_map, cpu_dest, 0);
 }
 
+/* Hashing initval */
+#define INITVAL 15485863
+
+static __always_inline
+u32 get_ipv4_hash_ip_pair(struct xdp_md *ctx, u64 nh_off)
+{
+	void *data_end = (void *)(long)ctx->data_end;
+	void *data     = (void *)(long)ctx->data;
+	struct iphdr *iph = data + nh_off;
+	u32 cpu_hash;
+
+	if (iph + 1 > data_end)
+		return 0;
+
+	cpu_hash = iph->saddr + iph->daddr;
+	cpu_hash = SuperFastHash((char *)&cpu_hash, 4, INITVAL + iph->protocol);
+
+	return cpu_hash;
+}
+
+static __always_inline
+u32 get_ipv6_hash_ip_pair(struct xdp_md *ctx, u64 nh_off)
+{
+	void *data_end = (void *)(long)ctx->data_end;
+	void *data     = (void *)(long)ctx->data;
+	struct ipv6hdr *ip6h = data + nh_off;
+	u32 cpu_hash;
+
+	if (ip6h + 1 > data_end)
+		return 0;
+
+	cpu_hash  = ip6h->saddr.s6_addr32[0] + ip6h->daddr.s6_addr32[0];
+	cpu_hash += ip6h->saddr.s6_addr32[1] + ip6h->daddr.s6_addr32[1];
+	cpu_hash += ip6h->saddr.s6_addr32[2] + ip6h->daddr.s6_addr32[2];
+	cpu_hash += ip6h->saddr.s6_addr32[3] + ip6h->daddr.s6_addr32[3];
+	cpu_hash = SuperFastHash((char *)&cpu_hash, 4, INITVAL + ip6h->nexthdr);
+
+	return cpu_hash;
+}
+
+/* Load-Balance traffic based on hashing IP-addrs + L4-proto.  The
+ * hashing scheme is symmetric, meaning swapping IP src/dest still hit
+ * same CPU.
+ */
+SEC("xdp_cpu_map5_lb_hash_ip_pairs")
+int  xdp_prognum5_lb_hash_ip_pairs(struct xdp_md *ctx)
+{
+	void *data_end = (void *)(long)ctx->data_end;
+	void *data     = (void *)(long)ctx->data;
+	struct ethhdr *eth = data;
+	u8 ip_proto = IPPROTO_UDP;
+	struct datarec *rec;
+	u16 eth_proto = 0;
+	u64 l3_offset = 0;
+	u32 cpu_dest = 0;
+	u32 cpu_idx = 0;
+	u32 *cpu_lookup;
+	u32 *cpu_max;
+	u32 cpu_hash;
+	u32 key = 0;
+
+	/* Count RX packet in map */
+	rec = bpf_map_lookup_elem(&rx_cnt, &key);
+	if (!rec)
+		return XDP_ABORTED;
+	rec->processed++;
+
+	cpu_max = bpf_map_lookup_elem(&cpus_count, &key);
+	if (!cpu_max)
+		return XDP_ABORTED;
+
+	if (!(parse_eth(eth, data_end, &eth_proto, &l3_offset)))
+		return XDP_PASS; /* Just skip */
+
+	/* Hash for IPv4 and IPv6 */
+	switch (eth_proto) {
+	case ETH_P_IP:
+		cpu_hash = get_ipv4_hash_ip_pair(ctx, l3_offset);
+		break;
+	case ETH_P_IPV6:
+		cpu_hash = get_ipv6_hash_ip_pair(ctx, l3_offset);
+		break;
+	case ETH_P_ARP: /* ARP packet handled on CPU idx 0 */
+	default:
+		cpu_hash = 0;
+	}
+
+	/* Choose CPU based on hash */
+	cpu_idx = cpu_hash % *cpu_max;
+
+	cpu_lookup = bpf_map_lookup_elem(&cpus_available, &cpu_idx);
+	if (!cpu_lookup)
+		return XDP_ABORTED;
+	cpu_dest = *cpu_lookup;
+
+	if (cpu_dest >= MAX_CPUS) {
+		rec->issue++;
+		return XDP_ABORTED;
+	}
+
+	return bpf_redirect_map(&cpu_map, cpu_dest, 0);
+}
 
 char _license[] SEC("license") = "GPL";
 
diff --git a/samples/bpf/xdp_redirect_cpu_user.c b/samples/bpf/xdp_redirect_cpu_user.c
index f6efaefd485b..007710d7c748 100644
--- a/samples/bpf/xdp_redirect_cpu_user.c
+++ b/samples/bpf/xdp_redirect_cpu_user.c
@@ -22,7 +22,7 @@ static const char *__doc__ =
 #define MAX_CPUS 12 /* WARNING - sync with _kern.c */
 
 /* How many xdp_progs are defined in _kern.c */
-#define MAX_PROG 5
+#define MAX_PROG 6
 
 /* Wanted to get rid of bpf_load.h and fake-"libbpf.h" (and instead
  * use bpf/libbpf.h), but cannot as (currently) needed for XDP
@@ -567,7 +567,7 @@ int main(int argc, char **argv)
 	int added_cpus = 0;
 	int longindex = 0;
 	int interval = 2;
-	int prog_num = 0;
+	int prog_num = 5;
 	int add_cpu = -1;
 	__u32 qsize;
 	int opt;