On Sat, Sep 19, 2020 at 4:50 AM Toke Høiland-Jørgensen <toke@xxxxxxxxxx> wrote:
>
> From: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>
>
> The check_attach_btf_id() function really does three things:
>
> 1. It performs a bunch of checks on the program to ensure that the
> attachment is valid.
>
> 2. It stores a bunch of state about the attachment being requested in
> the verifier environment and struct bpf_prog objects.
>
> 3. It allocates a trampoline for the attachment.
>
> This patch splits out (1.) and (3.) into separate functions in preparation
> for reusing them when the actual attachment is happening (in the
> raw_tracepoint_open syscall operation), which will allow tracing programs
> to have multiple (compatible) attachments.
>
> No functional change is intended with this patch.
>
> Acked-by: Andrii Nakryiko <andriin@xxxxxx>
> Signed-off-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx>
> ---
Ok, so bad news: you broke another selftest (test_overhead). Please,
do run test_progs and make sure everything succeeds, every time before
you post a new version.
Good news, though, is that this refactoring actually fixed a pretty
nasty bug in check_attach_btf_id logic: whenever bpf_trampoline
already existed (e.g., due to successful fentry to function X), all
subsequent fentry/fexit/fmod_ret and all their sleepable variants
would skip a bunch of check. So please attach the following Fixes
tags:
Fixes: 6ba43b761c41 ("bpf: Attachment verification for BPF_MODIFY_RETURN")
Fixes: 1e6c62a88215 ("bpf: Introduce sleepable BPF programs")
As for selftests, feel free to just drop the fmod_ret program, it was
never supposed to be possible, I just never realized that.
> include/linux/bpf.h | 7 +
> include/linux/bpf_verifier.h | 9 ++
> kernel/bpf/trampoline.c | 20 ++++
> kernel/bpf/verifier.c | 197 ++++++++++++++++++++++++------------------
> 4 files changed, 149 insertions(+), 84 deletions(-)
>
[...]
