OSDN Git Service

bpf: disallow attaching modify_return tracing functions to other BPF programs
authorToke Høiland-Jørgensen <toke@redhat.com>
Fri, 25 Sep 2020 21:25:00 +0000 (23:25 +0200)
committerAlexei Starovoitov <ast@kernel.org>
Tue, 29 Sep 2020 00:08:07 +0000 (17:08 -0700)
commit1af9270e908cd50a4f5d815c9b6f794c7d57ed07
tree74df99b5438c997caf6158e6a7e8afcc1e903e46
parent84a20d8e8795b45e3f1673648d43b4cf0f594359
bpf: disallow attaching modify_return tracing functions to other BPF programs

From the checks and commit messages for modify_return, it seems it was
never the intention that it should be possible to attach a tracing program
with expected_attach_type == BPF_MODIFY_RETURN to another BPF program.
However, check_attach_modify_return() will only look at the function name,
so if the target function starts with "security_", the attach will be
allowed even for bpf2bpf attachment.

Fix this oversight by also blocking the modification if a target program is
supplied.

Fixes: 18644cec714a ("bpf: Fix use-after-free in fmod_ret check")
Fixes: 6ba43b761c41 ("bpf: Attachment verification for BPF_MODIFY_RETURN")
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
kernel/bpf/verifier.c