bpftool: use a local copy of perf_event to fix accessing :: Bpf_cookie
authorAlexander Lobakin <alobakin@pm.me>
Fri, 7 Jul 2023 09:54:22 +0000 (10:54 +0100)
committerAndrii Nakryiko <andrii@kernel.org>
Mon, 10 Jul 2023 22:29:20 +0000 (15:29 -0700)
commit4cbeeb0dc02f8ac7b975b2ab0080ace53d43d62a
treeb8402a4ba990cf86f9d253581071f6a1ac57213f
parentc628747cc8800cf6d33d09f7f42c8b6f91e64dc7
bpftool: use a local copy of perf_event to fix accessing :: Bpf_cookie

When CONFIG_PERF_EVENTS is not set, struct perf_event remains empty.
However, the structure is being used by bpftool indirectly via BTF.
This leads to:

skeleton/pid_iter.bpf.c:49:30: error: no member named 'bpf_cookie' in 'struct perf_event'
        return BPF_CORE_READ(event, bpf_cookie);
               ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~

...

skeleton/pid_iter.bpf.c:49:9: error: returning 'void' from a function with incompatible result type '__u64' (aka 'unsigned long long')
        return BPF_CORE_READ(event, bpf_cookie);
               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Tools and samples can't use any CONFIG_ definitions, so the fields
used there should always be present.
Define struct perf_event___local with the `preserve_access_index`
attribute inside the pid_iter BPF prog to allow compiling on any
configs. CO-RE will substitute it with the real struct perf_event
accesses later on.

Fixes: cbdaf71f7e65 ("bpftool: Add bpf_cookie to link output")
Suggested-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: Quentin Monnet <quentin@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20230707095425.168126-2-quentin@isovalent.com
tools/bpf/bpftool/skeleton/pid_iter.bpf.c