samples/bpf: Add xdp program on egress for xdp_redirect_map
authorHangbin Liu <liuhangbin@gmail.com>
Fri, 22 Jan 2021 02:50:07 +0000 (10:50 +0800)
committerDaniel Borkmann <daniel@iogearbox.net>
Fri, 22 Jan 2021 23:24:37 +0000 (00:24 +0100)
This patch add a xdp program on egress to show that we can modify
the packet on egress. In this sample we will set the pkt's src
mac to egress's mac address. The xdp_prog will be attached when
-X option supplied.

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Link: https://lore.kernel.org/bpf/20210122025007.2968381-1-liuhangbin@gmail.com
samples/bpf/xdp_redirect_map_kern.c
samples/bpf/xdp_redirect_map_user.c

index 6489352..a92b8e5 100644 (file)
 #include <linux/ipv6.h>
 #include <bpf/bpf_helpers.h>
 
+/* The 2nd xdp prog on egress does not support skb mode, so we define two
+ * maps, tx_port_general and tx_port_native.
+ */
 struct {
        __uint(type, BPF_MAP_TYPE_DEVMAP);
        __uint(key_size, sizeof(int));
        __uint(value_size, sizeof(int));
        __uint(max_entries, 100);
-} tx_port SEC(".maps");
+} tx_port_general SEC(".maps");
+
+struct {
+       __uint(type, BPF_MAP_TYPE_DEVMAP);
+       __uint(key_size, sizeof(int));
+       __uint(value_size, sizeof(struct bpf_devmap_val));
+       __uint(max_entries, 100);
+} tx_port_native SEC(".maps");
 
 /* Count RX packets, as XDP bpf_prog doesn't get direct TX-success
  * feedback.  Redirect TX errors can be caught via a tracepoint.
@@ -36,6 +46,14 @@ struct {
        __uint(max_entries, 1);
 } rxcnt SEC(".maps");
 
+/* map to store egress interface mac address */
+struct {
+       __uint(type, BPF_MAP_TYPE_ARRAY);
+       __type(key, u32);
+       __type(value, __be64);
+       __uint(max_entries, 1);
+} tx_mac SEC(".maps");
+
 static void swap_src_dst_mac(void *data)
 {
        unsigned short *p = data;
@@ -52,17 +70,16 @@ static void swap_src_dst_mac(void *data)
        p[5] = dst[2];
 }
 
-SEC("xdp_redirect_map")
-int xdp_redirect_map_prog(struct xdp_md *ctx)
+static __always_inline int xdp_redirect_map(struct xdp_md *ctx, void *redirect_map)
 {
        void *data_end = (void *)(long)ctx->data_end;
        void *data = (void *)(long)ctx->data;
        struct ethhdr *eth = data;
        int rc = XDP_DROP;
-       int vport, port = 0, m = 0;
        long *value;
        u32 key = 0;
        u64 nh_off;
+       int vport;
 
        nh_off = sizeof(*eth);
        if (data + nh_off > data_end)
@@ -79,7 +96,40 @@ int xdp_redirect_map_prog(struct xdp_md *ctx)
        swap_src_dst_mac(data);
 
        /* send packet out physical port */
-       return bpf_redirect_map(&tx_port, vport, 0);
+       return bpf_redirect_map(redirect_map, vport, 0);
+}
+
+SEC("xdp_redirect_general")
+int xdp_redirect_map_general(struct xdp_md *ctx)
+{
+       return xdp_redirect_map(ctx, &tx_port_general);
+}
+
+SEC("xdp_redirect_native")
+int xdp_redirect_map_native(struct xdp_md *ctx)
+{
+       return xdp_redirect_map(ctx, &tx_port_native);
+}
+
+SEC("xdp_devmap/map_prog")
+int xdp_redirect_map_egress(struct xdp_md *ctx)
+{
+       void *data_end = (void *)(long)ctx->data_end;
+       void *data = (void *)(long)ctx->data;
+       struct ethhdr *eth = data;
+       __be64 *mac;
+       u32 key = 0;
+       u64 nh_off;
+
+       nh_off = sizeof(*eth);
+       if (data + nh_off > data_end)
+               return XDP_DROP;
+
+       mac = bpf_map_lookup_elem(&tx_mac, &key);
+       if (mac)
+               __builtin_memcpy(eth->h_source, mac, ETH_ALEN);
+
+       return XDP_PASS;
 }
 
 /* Redirect require an XDP bpf_prog loaded on the TX device */
index 31131b6..0e81926 100644 (file)
 #include <unistd.h>
 #include <libgen.h>
 #include <sys/resource.h>
+#include <sys/ioctl.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netinet/in.h>
 
 #include "bpf_util.h"
 #include <bpf/bpf.h>
@@ -22,6 +26,7 @@
 static int ifindex_in;
 static int ifindex_out;
 static bool ifindex_out_xdp_dummy_attached = true;
+static bool xdp_devmap_attached;
 static __u32 prog_id;
 static __u32 dummy_prog_id;
 
@@ -83,6 +88,32 @@ static void poll_stats(int interval, int ifindex)
        }
 }
 
+static int get_mac_addr(unsigned int ifindex_out, void *mac_addr)
+{
+       char ifname[IF_NAMESIZE];
+       struct ifreq ifr;
+       int fd, ret = -1;
+
+       fd = socket(AF_INET, SOCK_DGRAM, 0);
+       if (fd < 0)
+               return ret;
+
+       if (!if_indextoname(ifindex_out, ifname))
+               goto err_out;
+
+       strcpy(ifr.ifr_name, ifname);
+
+       if (ioctl(fd, SIOCGIFHWADDR, &ifr) != 0)
+               goto err_out;
+
+       memcpy(mac_addr, ifr.ifr_hwaddr.sa_data, 6 * sizeof(char));
+       ret = 0;
+
+err_out:
+       close(fd);
+       return ret;
+}
+
 static void usage(const char *prog)
 {
        fprintf(stderr,
@@ -90,24 +121,26 @@ static void usage(const char *prog)
                "OPTS:\n"
                "    -S    use skb-mode\n"
                "    -N    enforce native mode\n"
-               "    -F    force loading prog\n",
+               "    -F    force loading prog\n"
+               "    -X    load xdp program on egress\n",
                prog);
 }
 
 int main(int argc, char **argv)
 {
        struct bpf_prog_load_attr prog_load_attr = {
-               .prog_type      = BPF_PROG_TYPE_XDP,
+               .prog_type      = BPF_PROG_TYPE_UNSPEC,
        };
-       struct bpf_program *prog, *dummy_prog;
+       struct bpf_program *prog, *dummy_prog, *devmap_prog;
+       int prog_fd, dummy_prog_fd, devmap_prog_fd = 0;
+       int tx_port_map_fd, tx_mac_map_fd;
+       struct bpf_devmap_val devmap_val;
        struct bpf_prog_info info = {};
        __u32 info_len = sizeof(info);
-       int prog_fd, dummy_prog_fd;
-       const char *optstr = "FSN";
+       const char *optstr = "FSNX";
        struct bpf_object *obj;
        int ret, opt, key = 0;
        char filename[256];
-       int tx_port_map_fd;
 
        while ((opt = getopt(argc, argv, optstr)) != -1) {
                switch (opt) {
@@ -120,14 +153,21 @@ int main(int argc, char **argv)
                case 'F':
                        xdp_flags &= ~XDP_FLAGS_UPDATE_IF_NOEXIST;
                        break;
+               case 'X':
+                       xdp_devmap_attached = true;
+                       break;
                default:
                        usage(basename(argv[0]));
                        return 1;
                }
        }
 
-       if (!(xdp_flags & XDP_FLAGS_SKB_MODE))
+       if (!(xdp_flags & XDP_FLAGS_SKB_MODE)) {
                xdp_flags |= XDP_FLAGS_DRV_MODE;
+       } else if (xdp_devmap_attached) {
+               printf("Load xdp program on egress with SKB mode not supported yet\n");
+               return 1;
+       }
 
        if (optind == argc) {
                printf("usage: %s <IFNAME|IFINDEX>_IN <IFNAME|IFINDEX>_OUT\n", argv[0]);
@@ -150,24 +190,28 @@ int main(int argc, char **argv)
        if (bpf_prog_load_xattr(&prog_load_attr, &obj, &prog_fd))
                return 1;
 
-       prog = bpf_program__next(NULL, obj);
-       dummy_prog = bpf_program__next(prog, obj);
-       if (!prog || !dummy_prog) {
-               printf("finding a prog in obj file failed\n");
-               return 1;
+       if (xdp_flags & XDP_FLAGS_SKB_MODE) {
+               prog = bpf_object__find_program_by_name(obj, "xdp_redirect_map_general");
+               tx_port_map_fd = bpf_object__find_map_fd_by_name(obj, "tx_port_general");
+       } else {
+               prog = bpf_object__find_program_by_name(obj, "xdp_redirect_map_native");
+               tx_port_map_fd = bpf_object__find_map_fd_by_name(obj, "tx_port_native");
+       }
+       dummy_prog = bpf_object__find_program_by_name(obj, "xdp_redirect_dummy_prog");
+       if (!prog || dummy_prog < 0 || tx_port_map_fd < 0) {
+               printf("finding prog/dummy_prog/tx_port_map in obj file failed\n");
+               goto out;
        }
-       /* bpf_prog_load_xattr gives us the pointer to first prog's fd,
-        * so we're missing only the fd for dummy prog
-        */
+       prog_fd = bpf_program__fd(prog);
        dummy_prog_fd = bpf_program__fd(dummy_prog);
-       if (prog_fd < 0 || dummy_prog_fd < 0) {
+       if (prog_fd < 0 || dummy_prog_fd < 0 || tx_port_map_fd < 0) {
                printf("bpf_prog_load_xattr: %s\n", strerror(errno));
                return 1;
        }
 
-       tx_port_map_fd = bpf_object__find_map_fd_by_name(obj, "tx_port");
+       tx_mac_map_fd = bpf_object__find_map_fd_by_name(obj, "tx_mac");
        rxcnt_map_fd = bpf_object__find_map_fd_by_name(obj, "rxcnt");
-       if (tx_port_map_fd < 0 || rxcnt_map_fd < 0) {
+       if (tx_mac_map_fd < 0 || rxcnt_map_fd < 0) {
                printf("bpf_object__find_map_fd_by_name failed\n");
                return 1;
        }
@@ -199,11 +243,39 @@ int main(int argc, char **argv)
        }
        dummy_prog_id = info.id;
 
+       /* Load 2nd xdp prog on egress. */
+       if (xdp_devmap_attached) {
+               unsigned char mac_addr[6];
+
+               devmap_prog = bpf_object__find_program_by_name(obj, "xdp_redirect_map_egress");
+               if (!devmap_prog) {
+                       printf("finding devmap_prog in obj file failed\n");
+                       goto out;
+               }
+               devmap_prog_fd = bpf_program__fd(devmap_prog);
+               if (devmap_prog_fd < 0) {
+                       printf("finding devmap_prog fd failed\n");
+                       goto out;
+               }
+
+               if (get_mac_addr(ifindex_out, mac_addr) < 0) {
+                       printf("get interface %d mac failed\n", ifindex_out);
+                       goto out;
+               }
+
+               ret = bpf_map_update_elem(tx_mac_map_fd, &key, mac_addr, 0);
+               if (ret) {
+                       perror("bpf_update_elem tx_mac_map_fd");
+                       goto out;
+               }
+       }
+
        signal(SIGINT, int_exit);
        signal(SIGTERM, int_exit);
 
-       /* populate virtual to physical port map */
-       ret = bpf_map_update_elem(tx_port_map_fd, &key, &ifindex_out, 0);
+       devmap_val.ifindex = ifindex_out;
+       devmap_val.bpf_prog.fd = devmap_prog_fd;
+       ret = bpf_map_update_elem(tx_port_map_fd, &key, &devmap_val, 0);
        if (ret) {
                perror("bpf_update_elem");
                goto out;