Re: [PATCH bpf-next v4 2/2] samples/bpf: Add xdp_sample_pkts example

["Toke Høiland-Jørgensen" <toke@toke.dk>]

Jakub Kicinski <jakub.kicinski@netronome.com> writes:

> On Tue, 05 Jun 2018 16:50:00 +0200, Toke Høiland-Jørgensen wrote:
>> Add an example program showing how to sample packets from XDP using the
>> perf event buffer. The example userspace program just prints the ethernet
>> header for every packet sampled.
>> 
>> Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
>
>> diff --git a/samples/bpf/xdp_sample_pkts_kern.c b/samples/bpf/xdp_sample_pkts_kern.c
>> new file mode 100644
>> index 000000000000..4560522ca015
>> --- /dev/null
>> +++ b/samples/bpf/xdp_sample_pkts_kern.c
>> @@ -0,0 +1,62 @@
>> +#include <linux/ptrace.h>
>> +#include <linux/version.h>
>> +#include <uapi/linux/bpf.h>
>> +#include "bpf_helpers.h"
>> +
>> +#define SAMPLE_SIZE 64ul
>> +#define MAX_CPUS 24
>
> That may be a lil' too few for modern HW with hyper-threading on ;)  
> My development machine says:
>
> $ ncpus
> 28
>
> 128, maybe?

What? 24 CPUs should be enough for everyone, surely? ;)

(I just took twice the number of CPUs in my largest machine; but fair
point, other people have access to more expensive toys I guess... will fix)

>> +#define bpf_printk(fmt, ...)					\
>> +({								\
>> +	       char ____fmt[] = fmt;				\
>> +	       bpf_trace_printk(____fmt, sizeof(____fmt),	\
>> +				##__VA_ARGS__);			\
>> +})
>> +
>> +struct bpf_map_def SEC("maps") my_map = {
>> +	.type = BPF_MAP_TYPE_PERF_EVENT_ARRAY,
>> +	.key_size = sizeof(int),
>> +	.value_size = sizeof(u32),
>> +	.max_entries = MAX_CPUS,
>> +};
>> +
>> +SEC("xdp_sample")
>> +int xdp_sample_prog(struct xdp_md *ctx)
>> +{
>> +	void *data_end = (void *)(long)ctx->data_end;
>> +	void *data = (void *)(long)ctx->data;
>> +
>> +        /* Metadata will be in the perf event before the packet data. */
>> +	struct S {
>> +		u16 cookie;
>> +		u16 pkt_len;
>> +	} __attribute__((packed)) metadata;
>> +
>> +	if (data + SAMPLE_SIZE < data_end) {
>> +		/* The XDP perf_event_output handler will use the upper 32 bits
>> +		 * of the flags argument as a number of bytes to include of the
>> +		 * packet payload in the event data. If the size is too big, the
>> +		 * call to bpf_perf_event_output will fail and return -EFAULT.
>> +		 *
>> +		 * See bpf_xdp_event_output in net/core/filter.c.
>> +		 *
>> +		 * The BPF_F_CURRENT_CPU flag means that the event output fd
>> +		 * will be indexed by the CPU number in the event map.
>> +		 */
>> +		u64 flags = (SAMPLE_SIZE << 32) | BPF_F_CURRENT_CPU;
>> +		int ret;
>> +
>> +		metadata.cookie = 0xdead;
>> +		metadata.pkt_len = (u16)(data_end - data);
>> +
>> +		ret = bpf_perf_event_output(ctx, &my_map, flags,
>> +				      &metadata, sizeof(metadata));
>> +		if(ret)
>
> Please run checkpatch --strict on the samples.

Will do!

>> +			bpf_printk("perf_event_output failed: %d\n", ret);
>> +	}
>> +
>> +	return XDP_PASS;
>> +}
>> +
>> +char _license[] SEC("license") = "GPL";
>> +u32 _version SEC("version") = LINUX_VERSION_CODE;