diff options
| author |   Andrii Nakryiko <andriin@fb.com> | 2020-01-23 21:38:37 -0800 |
|---|---|---|
| committer |   Daniel Borkmann <daniel@iogearbox.net> | 2020-01-24 22:16:26 +0100 |
| commit | d7a252708dbc950ca2064310217e8b9f85846e2f (patch) | |
| tree | 22fc01fa40da05e5cb32a633298e6eef1bfac0ef /tools/lib/bpf/libbpf.h | |
| parent | selftests: bpf: Reset global state between reuseport test runs (diff) | |
| download | linux-dev-d7a252708dbc950ca2064310217e8b9f85846e2f.tar.xz linux-dev-d7a252708dbc950ca2064310217e8b9f85846e2f.zip | |
libbpf: Improve handling of failed CO-RE relocations
Previously, if libbpf failed to resolve CO-RE relocation for some
instructions, it would either return error immediately, or, if
.relaxed_core_relocs option was set, would replace relocatable offset/imm part
of an instruction with a bogus value (-1). Neither approach is good, because
there are many possible scenarios where relocation is expected to fail (e.g.,
when some field knowingly can be missing on specific kernel versions). On the
other hand, replacing offset with invalid one can hide programmer errors, if
this relocation failue wasn't anticipated.
This patch deprecates .relaxed_core_relocs option and changes the approach to
always replacing instruction, for which relocation failed, with invalid BPF
helper call instruction. For cases where this is expected, BPF program should
already ensure that that instruction is unreachable, in which case this
invalid instruction is going to be silently ignored. But if instruction wasn't
guarded, BPF program will be rejected at verification step with verifier log
pointing precisely to the place in assembly where the problem is.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200124053837.2434679-1-andriin@fb.com
Diffstat (limited to 'tools/lib/bpf/libbpf.h')
| -rw-r--r-- | tools/lib/bpf/libbpf.h | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h index 2a5e3b087002..3fe12c9d1f92 100644 --- a/tools/lib/bpf/libbpf.h +++ b/tools/lib/bpf/libbpf.h @@ -77,7 +77,11 @@ struct bpf_object_open_opts { const char *object_name; /* parse map definitions non-strictly, allowing extra attributes/data */ bool relaxed_maps; - /* process CO-RE relocations non-strictly, allowing them to fail */ + /* DEPRECATED: handle CO-RE relocations non-strictly, allowing failures. + * Value is ignored. Relocations always are processed non-strictly. + * Non-relocatable instructions are replaced with invalid ones to + * prevent accidental errors. + * */ bool relaxed_core_relocs; /* maps that set the 'pinning' attribute in their definition will have * their pin_path attribute set to a file in this directory, and be |