Re: [PATCHv6 bpf-next] samples/bpf: add xdp program on egress for xdp_redirect_map

[Yonghong Song <yhs@fb.com>]

On 1/14/21 6:27 AM, Hangbin Liu wrote:
> 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>
> 
> ---
> v6: no code update, only rebase the code on latest bpf-next
> 
> v5:
> a) close fd when err out in get_mac_addr()
> b) exit program when both -S and -X supplied.
> 
> v4:
> a) Update get_mac_addr socket create
> b) Load dummy prog regardless of 2nd xdp prog on egress
> 
> v3:
> a) modify the src mac address based on egress mac
> 
> v2:
> a) use pkt counter instead of IP ttl modification on egress program
> b) make the egress program selectable by option -X
> ---
>   samples/bpf/xdp_redirect_map_kern.c |  75 ++++++++++++++--
>   samples/bpf/xdp_redirect_map_user.c | 135 +++++++++++++++++++++++-----
>   2 files changed, 184 insertions(+), 26 deletions(-)
> 
> diff --git a/samples/bpf/xdp_redirect_map_kern.c b/samples/bpf/xdp_redirect_map_kern.c
> index 6489352ab7a4..8b8e73d25ad6 100644
> --- a/samples/bpf/xdp_redirect_map_kern.c
> +++ b/samples/bpf/xdp_redirect_map_kern.c
> @@ -19,12 +19,22 @@
>   #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 stroe egress interface mac address */

s/stroe/store

> +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];
>   }
>   
[...]
>   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 devmap_prog_fd_0 = -1, devmap_prog_fd_1 = -1;

The default value is -1 here. I remembered there was a discussion
about the default value here, does default value 0 work here?

> +	int prog_fd, dummy_prog_fd;
> +	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 +154,21 @@ 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 +191,28 @@ 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_title(obj, "xdp_redirect_general");

libbpf supports each section having multiple programs, so 
bpf_object__find_program_by_title() is not recommended.
Could you change to bpf_object__find_program_by_name()?

> +		tx_port_map_fd = bpf_object__find_map_fd_by_name(obj, "tx_port_general");
> +	} else {
> +		prog = bpf_object__find_program_by_title(obj, "xdp_redirect_native");
> +		tx_port_map_fd = bpf_object__find_map_fd_by_name(obj, "tx_port_native");
> +	}
> +	dummy_prog = bpf_object__find_program_by_title(obj, "xdp_redirect_dummy");
> +	if (!prog || dummy_prog < 0 || tx_port_map_fd < 0) {
> +		printf("finding 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;
>   	}
[...]