答复: [PATCH v4,bpf-next] samples/bpf: check detach prog exist or not in xdp_fwd

[shaozhengchao]

-----邮件原件-----
发件人: Toke Høiland-Jørgensen [mailto:toke@xxxxxxxxxx] 
发送时间: 2022年6月1日 17:25
收件人: shaozhengchao <shaozhengchao@xxxxxxxxxx>; bpf@xxxxxxxxxxxxxxx; netdev@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; ast@xxxxxxxxxx; daniel@xxxxxxxxxxxxx; davem@xxxxxxxxxxxxx; kuba@xxxxxxxxxx; hawk@xxxxxxxxxx; john.fastabend@xxxxxxxxx; andrii@xxxxxxxxxx; kafai@xxxxxx; songliubraving@xxxxxx; yhs@xxxxxx; kpsingh@xxxxxxxxxx
抄送: weiyongjun (A) <weiyongjun1@xxxxxxxxxx>; shaozhengchao <shaozhengchao@xxxxxxxxxx>; yuehaibing <yuehaibing@xxxxxxxxxx>
主题: Re: [PATCH v4,bpf-next] samples/bpf: check detach prog exist or not in xdp_fwd

Zhengchao Shao <shaozhengchao@xxxxxxxxxx> writes:

> Before detach the prog, we should check detach prog exist or not.
>
> Signed-off-by: Zhengchao Shao <shaozhengchao@xxxxxxxxxx>

The logic looks good now, just spotted one issue with the use of 'return errno' (which I just realised I suggested on the last version, sorry about that).

>  samples/bpf/xdp_fwd_user.c | 62 
> ++++++++++++++++++++++++++++++++------
>  1 file changed, 53 insertions(+), 9 deletions(-)
>
> diff --git a/samples/bpf/xdp_fwd_user.c b/samples/bpf/xdp_fwd_user.c 
> index 1828487bae9a..a4ba53891653 100644
> --- a/samples/bpf/xdp_fwd_user.c
> +++ b/samples/bpf/xdp_fwd_user.c
> @@ -47,17 +47,61 @@ static int do_attach(int idx, int prog_fd, int map_fd, const char *name)
>  	return err;
>  }
>  
> -static int do_detach(int idx, const char *name)
> +static int do_detach(int ifindex, const char *ifname, const char 
> +*app_name)
>  {
> -	int err;
> +	LIBBPF_OPTS(bpf_xdp_attach_opts, opts);
> +	struct bpf_prog_info prog_info = {};
> +	char prog_name[BPF_OBJ_NAME_LEN];
> +	__u32 info_len, curr_prog_id;
> +	int prog_fd;
> +	int err = 1;
> +
> +	if (bpf_xdp_query_id(ifindex, xdp_flags, &curr_prog_id)) {
> +		printf("ERROR: bpf_xdp_query_id failed (%s)\n",
> +		       strerror(errno));
> +		return -errno;

The printf above may override errno, so this could return 0; the actual return value is not really used by the caller, so you could just always 'return -1' or 'return err'.

> +	}
> +
> +	if (!curr_prog_id) {
> +		printf("ERROR: flags(0x%x) xdp prog is not attached to %s\n",
> +		       xdp_flags, ifname);
> +		goto err_out;

Doesn't really need a label, just do a direct return here as well.

> +	}
>  
> -	err = bpf_xdp_detach(idx, xdp_flags, NULL);
> -	if (err < 0)
> -		printf("ERROR: failed to detach program from %s\n", name);
> +	info_len = sizeof(prog_info);
> +	prog_fd = bpf_prog_get_fd_by_id(curr_prog_id);
> +	if (prog_fd < 0) {
> +		printf("ERROR: bpf_prog_get_fd_by_id failed (%s)\n",
> +		       strerror(errno));
> +		return -errno;

Same issue as above with errno being overwritten.

> +	}
> +
> +	err = bpf_obj_get_info_by_fd(prog_fd, &prog_info, &info_len);
> +	if (err) {
> +		printf("ERROR: bpf_obj_get_info_by_fd failed (%s)\n",
> +		       strerror(errno));
> +		goto close_out;
> +	}
> +	snprintf(prog_name, sizeof(prog_name), "%s_prog", app_name);
> +	prog_name[BPF_OBJ_NAME_LEN - 1] = '\0';
> +
> +	if (strcmp(prog_info.name, prog_name)) {
> +		printf("ERROR: %s isn't attached to %s\n", app_name, ifname);
> +		err = 1;
> +	} else {

You can save a level of indentation by adding a 'goto close_out' after 'err=1' instead of using an 'else' block.

> +		opts.old_prog_fd = prog_fd;
> +		err = bpf_xdp_detach(ifindex, xdp_flags, &opts);
> +		if (err < 0)
> +			printf("ERROR: failed to detach program from %s (%s)\n",
> +			       ifname, strerror(errno));
> +		/* TODO: Remember to cleanup map, when adding use of shared map
> +		 *  bpf_map_delete_elem((map_fd, &idx);
> +		 */
> +	}
>  
> -	/* TODO: Remember to cleanup map, when adding use of shared map
> -	 *  bpf_map_delete_elem((map_fd, &idx);
> -	 */
> +close_out:
> +	close(prog_fd);
> +err_out:

don't need the err_out label, see above.

-Toke

Hi Toke:
	Thank you for your reply. I will fix it in next patch.

Zhengchao Shao