Re: [PATCH] Add skb_store_bytes() for BPF_PROG_TYPE_CGROUP_SKB

On Wed, Dec 22, 2021 at 10:49:45PM +0000, Tyler Wear wrote:

quoted

On 12/21/21 6:27 PM, Tyler Wear wrote:

quoted

Need to modify the ds field to support upcoming Wifi QoS Alliance
spec. Instead of adding generic function for just modifying the ds
field, add skb_store_bytes for BPF_PROG_TYPE_CGROUP_SKB. This allows
other fields in the network and transport header to be modified in the
future.

Could change tag from "[PATCH]" to "[PATCH bpf-next]"?
Please also indicate the version of the patch, so in this case, it should be "[PATCH bpf-next v2]".

I think you can add more contents in the commit message about why existing bpf_setsockopt() won't work and why
CGROUP_UDP[4|6]_SENDMSG is not preferred.
These have been discussed in v1 of this patch and they are valuable for people to understand full context and reasoning.

quoted

Signed-off-by: Tyler Wear <redacted>
---
  net/core/filter.c | 2 ++
  1 file changed, 2 insertions(+)

diff --git a/net/core/filter.c b/net/core/filter.c index

6102f093d59a..0c25aa2212a2 100644

--- a/net/core/filter.c
+++ b/net/core/filter.c

@@ -7289,6 +7289,8 @@ static const struct bpf_func_proto *
  cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
  {
      switch (func_id) {
+     case BPF_FUNC_skb_store_bytes:
+             return &bpf_skb_store_bytes_proto;

Typically different 'case's are added in chronological order to people can guess what is added earlier and what is added later. Maybe
add the new helper after BPF_FUNC_perf_event_output?

quoted

      case BPF_FUNC_get_local_storage:
              return &bpf_get_local_storage_proto;
      case BPF_FUNC_sk_fullsock:

Please add a test case to exercise the new usage of
bpf_skb_store_bytes() helper. You may piggy back on some existing cg_skb progs if it is easier to do.

Would it be sufficient to change the dscp value in tools/testing/selftests/bpf/progs/test_sock_fields.c via bpf_skb_store_bytes()

test_sock_fields focus on sk instead of skb, so it will not be a good fit.

load_bytes_relative.c may be a better fit.
The minimal is to write the dscp value by bpf_skb_store_bytes()
and be able to read it back at the receiver side (e.g.
by making a TCP connection like load_bytes_relative).