On Wed, Nov 29, 2023 at 3:20 AM Song Liu [off-list ref] wrote:

On Tue, Nov 28, 2023 at 10:47 PM Alexei Starovoitov
[off-list ref] wrote:

quoted

On Tue, Nov 28, 2023 at 4:37 PM Song Liu [off-list ref] wrote:

quoted

+char digest[MAGIC_SIZE + sizeof(struct fsverity_digest) + SHA256_DIGEST_SIZE];

when vmlinux is built without CONFIG_FS_VERITY the above fails
in a weird way:
  CLNG-BPF [test_maps] test_sig_in_xattr.bpf.o
progs/test_sig_in_xattr.c:36:26: error: invalid application of
'sizeof' to an incomplete type 'struct fsverity_digest'
   36 | char digest[MAGIC_SIZE + sizeof(struct fsverity_digest) +
SHA256_DIGEST_SIZE];
      |                          ^     ~~~~~~~~~~~~~~~~~~~~~~~~

Is there a way to somehow print a hint during the build what
configs users need to enable to pass the build ?

Patch 5/6 added CONFIG_FS_VERITY to tools/testing/selftests/bpf/config.
This is a more general question for all required CONFIG_* specified in the
file (and the config files for other selftests).

In selftests/bpf/Makefile, we have logic to find vmlinux. We can add similar
logic to find .config used to build the vmlinux, and grep for each required
CONFIG_* from the .config file. Does this sound like a viable solution?

No need for new logic to parse .config.
libbpf does it already and
extern bool CONFIG_FS_VERITY __kconfig __weak;
works.

Since you hard code MAGIC_SIZE anyway I'm asking
to hard code sizeof(struct fsverity_digest) as well, since the bpf prog
doesn't access it directly. It only needs to know its size.

While inside:
int BPF_PROG(test_file_open, struct file *f)
{
  if (!CONFIG_FS_VERITY) {
     skip_fs_verity_test = true;
     return 0;
  }

and report it as a clean error message in test_progs.

We keep adding new config requirements selftests/bpf/config which
forces all developers to keep adding new configs to their builds.
In the past, when we didn't have BPF CI, that was necessary, but now
BPF CI does it for us.
With clean error message from test_progs the developers can either
ignore the error and proceed with their work or adjust their .config
eventually. While hard selftest build error forces all devs to
update .config right away and build error has no info of what needs
to be done which is not developer friendly.

pw-bot: cr