Merge branch 'bpf-xdp-fwd-sample-improvements'

Jesper Dangaard Brouer says:

====================
V3: Hopefully fixed all issues point out by Yonghong Song

V2: Addressed issues point out by Yonghong Song
 - Please ACK patch 2/3 again
 - Added ACKs and reviewed-by to other patches

This patchset is focused on improvements for XDP forwarding sample
named xdp_fwd, which leverage the existing FIB routing tables as
described in LPC2018[1] talk by David Ahern.

The primary motivation is to illustrate how Toke's recent work
improves usability of XDP_REDIRECT via lookups in devmap. The other
patches are to help users understand the sample.

I have more improvements to xdp_fwd, but those might requires changes
to libbpf.  Thus, sending these patches first as they are isolated.

[1] http://vger.kernel.org/lpc-networking2018.html#session-1
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

samples/bpf/xdp_fwd_kern.c

@@ -23,7 +23,8 @@
 
 #define IPV6_FLOWINFO_MASK              cpu_to_be32(0x0FFFFFFF)
 
-struct bpf_map_def SEC("maps") tx_port = {
+/* For TX-traffic redirect requires net_device ifindex to be in this devmap */
+struct bpf_map_def SEC("maps") xdp_tx_ports = {
 	.type = BPF_MAP_TYPE_DEVMAP,
 	.key_size = sizeof(int),
 	.value_size = sizeof(int),
@@ -102,22 +103,42 @@ static __always_inline int xdp_fwd_flags(struct xdp_md *ctx, u32 flags)
 	fib_params.ifindex = ctx->ingress_ifindex;
 
 	rc = bpf_fib_lookup(ctx, &fib_params, sizeof(fib_params), flags);
-
-	/* verify egress index has xdp support
-	 * TO-DO bpf_map_lookup_elem(&tx_port, &key) fails with
-	 *       cannot pass map_type 14 into func bpf_map_lookup_elem#1:
-	 * NOTE: without verification that egress index supports XDP
-	 *       forwarding packets are dropped.
+	/*
+	 * Some rc (return codes) from bpf_fib_lookup() are important,
+	 * to understand how this XDP-prog interacts with network stack.
+	 *
+	 * BPF_FIB_LKUP_RET_NO_NEIGH:
+	 *  Even if route lookup was a success, then the MAC-addresses are also
+	 *  needed.  This is obtained from arp/neighbour table, but if table is
+	 *  (still) empty then BPF_FIB_LKUP_RET_NO_NEIGH is returned.  To avoid
+	 *  doing ARP lookup directly from XDP, then send packet to normal
+	 *  network stack via XDP_PASS and expect it will do ARP resolution.
+	 *
+	 * BPF_FIB_LKUP_RET_FWD_DISABLED:
+	 *  The bpf_fib_lookup respect sysctl net.ipv{4,6}.conf.all.forwarding
+	 *  setting, and will return BPF_FIB_LKUP_RET_FWD_DISABLED if not
+	 *  enabled this on ingress device.
 	 */
-	if (rc == 0) {
+	if (rc == BPF_FIB_LKUP_RET_SUCCESS) {
+		/* Verify egress index has been configured as TX-port.
+		 * (Note: User can still have inserted an egress ifindex that
+		 * doesn't support XDP xmit, which will result in packet drops).
+		 *
+		 * Note: lookup in devmap supported since 0cdbb4b09a0.
+		 * If not supported will fail with:
+		 *  cannot pass map_type 14 into func bpf_map_lookup_elem#1:
+		 */
+		if (!bpf_map_lookup_elem(&xdp_tx_ports, &fib_params.ifindex))
+			return XDP_PASS;
+
 		if (h_proto == htons(ETH_P_IP))
 			ip_decrease_ttl(iph);
 		else if (h_proto == htons(ETH_P_IPV6))
 			ip6h->hop_limit--;
 
 		memcpy(eth->h_dest, fib_params.dmac, ETH_ALEN);
 		memcpy(eth->h_source, fib_params.smac, ETH_ALEN);
-		return bpf_redirect_map(&tx_port, fib_params.ifindex, 0);
+		return bpf_redirect_map(&xdp_tx_ports, fib_params.ifindex, 0);
 	}
 
 	return XDP_PASS;

samples/bpf/xdp_fwd_user.c

@@ -27,14 +27,20 @@
 #include "libbpf.h"
 #include <bpf/bpf.h>
 
-
-static int do_attach(int idx, int fd, const char *name)
+static int do_attach(int idx, int prog_fd, int map_fd, const char *name)
 {
 	int err;
 
-	err = bpf_set_link_xdp_fd(idx, fd, 0);
-	if (err < 0)
+	err = bpf_set_link_xdp_fd(idx, prog_fd, 0);
+	if (err < 0) {
 		printf("ERROR: failed to attach program to %s\n", name);
+		return err;
+	}
+
+	/* Adding ifindex as a possible egress TX port */
+	err = bpf_map_update_elem(map_fd, &idx, &idx, 0);
+	if (err)
+		printf("ERROR: failed using device %s as TX-port\n", name);
 
 	return err;
 }
@@ -47,6 +53,9 @@ static int do_detach(int idx, const char *name)
 	if (err < 0)
 		printf("ERROR: failed to detach program from %s\n", name);
 
+	/* TODO: Remember to cleanup map, when adding use of shared map
+	 *  bpf_map_delete_elem((map_fd, &idx);
+	 */
 	return err;
 }
 
@@ -67,10 +76,10 @@ int main(int argc, char **argv)
 	};
 	const char *prog_name = "xdp_fwd";
 	struct bpf_program *prog;
+	int prog_fd, map_fd = -1;
 	char filename[PATH_MAX];
 	struct bpf_object *obj;
 	int opt, i, idx, err;
-	int prog_fd, map_fd;
 	int attach = 1;
 	int ret = 0;
 
@@ -103,8 +112,14 @@ int main(int argc, char **argv)
 			return 1;
 		}
 
-		if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
+		err = bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd);
+		if (err) {
+			printf("Does kernel support devmap lookup?\n");
+			/* If not, the error message will be:
+			 *  "cannot pass map_type 14 into func bpf_map_lookup_elem#1"
+			 */
 			return 1;
+		}
 
 		prog = bpf_object__find_program_by_title(obj, prog_name);
 		prog_fd = bpf_program__fd(prog);
@@ -113,16 +128,12 @@ int main(int argc, char **argv)
 			return 1;
 		}
 		map_fd = bpf_map__fd(bpf_object__find_map_by_name(obj,
-								  "tx_port"));
+							"xdp_tx_ports"));
 		if (map_fd < 0) {
 			printf("map not found: %s\n", strerror(map_fd));
 			return 1;
 		}
 	}
-	if (attach) {
-		for (i = 1; i < 64; ++i)
-			bpf_map_update_elem(map_fd, &i, &i, 0);
-	}
 
 	for (i = optind; i < argc; ++i) {
 		idx = if_nametoindex(argv[i]);
@@ -138,7 +149,7 @@ int main(int argc, char **argv)
 			if (err)
 				ret = err;
 		} else {
-			err = do_attach(idx, prog_fd, argv[i]);
+			err = do_attach(idx, prog_fd, map_fd, argv[i]);
 			if (err)
 				ret = err;
 		}