Daniel Borkmann wrote:

On 3/24/21 10:07 PM, John Fastabend wrote:

quoted

With a relatively recent clang master branch test_map skips a section,

  libbpf: elf: skipping unrecognized data section(5) .rodata.str1.1

the cause is some pointless strings from bpf_printks in the BPF program
loaded during testing. Remove them so we stop tripping our test bots.

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
---
  .../selftests/bpf/progs/sockmap_tcp_msg_prog.c     |    3 ---
  1 file changed, 3 deletions(-)

diff --git a/tools/testing/selftests/bpf/progs/sockmap_tcp_msg_prog.c b/tools/testing/selftests/bpf/progs/sockmap_tcp_msg_prog.c
index fdb4bf4408fa..0f603253f4ed 100644
--- a/tools/testing/selftests/bpf/progs/sockmap_tcp_msg_prog.c
+++ b/tools/testing/selftests/bpf/progs/sockmap_tcp_msg_prog.c

@@ -16,10 +16,7 @@ int bpf_prog1(struct sk_msg_md *msg)
  	if (data + 8 > data_end)
  		return SK_DROP;
  
-	bpf_printk("data length %i\n", (__u64)msg->data_end - (__u64)msg->data);
  	d = (char *)data;

Do we still need 'd' as well in that case, or the data + 8 > data_end test if we don't
read any of the data? I'm not sure what was the original purpose of the prog, perhaps
just to test that we can attach /something/ in general? Maybe in that case empty prog
is sufficient if we don't do anything useful with the rest?

This program and test existed before test_sockmap was running and doing a
more complete test set. At that time it was the only thing verifying that
we could read the d[] and check lengths.

At this point these cases are covered there so it should be OK to just
make it an empty program. Then it is _just_ testing the map attach/detach
logic not that the programs themselves work correctly.

By the way without d marked violatile my compiler removes the load there
so its pointless as you note.

Because this is used for test maps I'll just make this an empty program.

quoted

-	bpf_printk("hello sendmsg hook %i %i\n", d[0], d[1]);
-
  	return SK_PASS;
  }