On Wed, Aug 31, 2022 at 4:36 PM Harsh Modi <harshmodi@xxxxxxxxxx> wrote:
>
> eBPF currently does not have a good way to support more advanced
> checksums like crc32c checksums. A bpf helper that allows users
> to hash packet data from eBPF will use this.
>
> Currently, it only supports crc32c, however, additional support for
> new hashes can be supported by adding an additional enum and
> implementing the corresponding code in net/core/filter.c.
>
> Signed-off-by: Harsh Modi <harshmodi@xxxxxxxxxx>
> ---
> include/net/xdp.h | 3 +
> include/uapi/linux/bpf.h | 33 ++++
> net/core/filter.c | 100 +++++++++++
> net/core/xdp.c | 51 ++++++
> scripts/bpf_doc.py | 2 +
> tools/include/uapi/linux/bpf.h | 33 ++++
> .../selftests/bpf/prog_tests/packet_hash.c | 159 ++++++++++++++++++
> .../testing/selftests/bpf/progs/packet_hash.c | 125 ++++++++++++++
> 8 files changed, 506 insertions(+)
> create mode 100644 tools/testing/selftests/bpf/prog_tests/packet_hash.c
> create mode 100644 tools/testing/selftests/bpf/progs/packet_hash.c
>
> diff --git a/include/net/xdp.h b/include/net/xdp.h
> index 04c852c7a77f..cbfec47e391d 100644
> --- a/include/net/xdp.h
> +++ b/include/net/xdp.h
> @@ -407,6 +407,9 @@ struct netdev_bpf;
> void xdp_attachment_setup(struct xdp_attachment_info *info,
> struct netdev_bpf *bpf);
>
> +__wsum __xdp_checksum(struct xdp_buff *xdp, int offset, int len,
> + __wsum csum, const struct skb_checksum_ops *ops);
> +
> #define DEV_MAP_BULK_SIZE XDP_BULK_QUEUE_SIZE
>
> #endif /* __LINUX_NET_XDP_H__ */
> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> index 962960a98835..c8313a13a948 100644
> --- a/include/uapi/linux/bpf.h
> +++ b/include/uapi/linux/bpf.h
> @@ -5386,6 +5386,25 @@ union bpf_attr {
> * Return
> * Current *ktime*.
> *
> + * int bpf_skb_packet_hash(struct sk_buff *skb, struct bpf_packet_hash_params *params, void *hash, u32 len)
> + * Description
> + * Hash the packet data based on the parameters set in *params*.
> + * The hash will be set in *hash*. The value of *len* will be
> + * dependent on the hash algorithm.
> + * Currently only crc32c is supported.
> + *
> + * Return
> + * 0 on success, or negative errno if there is an error.
> + *
> + * int bpf_xdp_packet_hash(struct xdp_buff *xdp, struct bpf_packet_hash_params *params, void *hash, u32 len)
> + * Description
> + * Hash the packet data based on the parameters set in *params*.
> + * The hash will be set in *hash*. The value of *len* will be
> + * dependent on the hash algorithm.
> + * Currently only crc32c is supported.
> + *
> + * Return
> + * 0 on success, or negative errno if there is an error.
> */
What are your thoughts on having a more generic hashing helper? I'm
thinking about something like
bpf_dynptr_hash(enum bpf_hash hash_type, struct bpf_dynptr src, struct
bpf_dynptr dst, u64 flags);
where this helper would extend easily to other cases as well (for
example, to malloc-type dynpts) instead of needing multiple helpers.
> #define __BPF_FUNC_MAPPER(FN) \
> FN(unspec), \
> @@ -5597,6 +5616,8 @@ union bpf_attr {
> FN(tcp_raw_check_syncookie_ipv4), \
> FN(tcp_raw_check_syncookie_ipv6), \
> FN(ktime_get_tai_ns), \
> + FN(skb_packet_hash), \
> + FN(xdp_packet_hash), \
> /* */
[...]
> --
> 2.37.2.672.g94769d06f0-goog
>
