2023-09-04 18:49:08

by Florian Westphal

[permalink] [raw]
Subject: Re: [PATCH] samples/bpf: Add sample usage for BPF_PROG_TYPE_NETFILTER

David Wang <[email protected]> wrote:
> This sample code implements a simple ipv4
> blacklist via the new bpf type BPF_PROG_TYPE_NETFILTER,
> which was introduced in 6.4.
>
> The bpf program drops package if destination ip address
> hits a match in the map of type BPF_MAP_TYPE_LPM_TRIE,
>
> The userspace code would load the bpf program,
> attach it to netfilter's FORWARD/OUTPUT hook,
> and then write ip patterns into the bpf map.

Thanks, I think its good to have this.

> diff --git a/samples/bpf/netfilter_ip4_blacklist.bpf.c b/samples/bpf/netfilter_ip4_blacklist.bpf.c
> new file mode 100644
> index 000000000000..d315d64fda7f
> --- /dev/null
> +++ b/samples/bpf/netfilter_ip4_blacklist.bpf.c
> @@ -0,0 +1,62 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#include "vmlinux.h"
> +#include <bpf/bpf_helpers.h>
> +
> +
> +#define NF_DROP 0
> +#define NF_ACCEPT 1

If you are interested, you could send a patch for nf-next that
makes the uapi headers expose this as enum, AFAIU that would make
the verdict nanes available via vmlinux.h.

> + /* search p->daddr in trie */
> + key.prefixlen = 32;
> + key.data = p->daddr;
> + pvalue = bpf_map_lookup_elem(&ipv4_lpm_map, &key);
> + if (pvalue) {
> + /* cat /sys/kernel/debug/tracing/trace_pipe */
> + bpf_printk("rule matched with %d...\n", *pvalue);

If you are interested you could send a patch that adds a kfunc to
nf_bpf_link that exposes nf_log_packet() to bpf.

nf_log_packet has a terrible api, I suggest to have the kfunc take
'struct nf_hook_state *' instead of 6+ members of that struct as
argument.

Thanks for the example.