On 4/9/25 4:14 PM, Andrii Nakryiko wrote: > On Tue, Apr 8, 2025 at 11:41 AM Ihor Solodrai <ihor.solodrai@xxxxxxxxx> wrote: >> >> A valid ELF file may contain a SHT_NOBITS .BTF section. This case is >> not handled correctly in btf_parse_elf, which leads to a segfault. >> >> Add a null check for a buffer returned by elf_getdata() before >> proceeding with its processing. >> >> Bug report: https://github.com/libbpf/libbpf/issues/894 >> >> Signed-off-by: Ihor Solodrai <ihor.solodrai@xxxxxxxxx> >> --- >> tools/lib/bpf/btf.c | 6 ++++++ >> 1 file changed, 6 insertions(+) >> >> diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c >> index 38bc6b14b066..90599f0311bd 100644 >> --- a/tools/lib/bpf/btf.c >> +++ b/tools/lib/bpf/btf.c >> @@ -1201,6 +1201,12 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf, >> goto done; >> } >> >> + if (!secs.btf_data->d_buf) { >> + pr_warn("BTF data is empty in %s\n", path); >> + err = -ENODATA; >> + goto done; >> + } >> + > > let's handle this more generally for all BTF data sections in > btf_find_elf_sections()? Sure. I think it makes sense to check for the section type before attempting to load a buffer like this: @@ -1148,6 +1148,12 @@ static int btf_find_elf_sections(Elf *elf, const char *path, struct btf_elf_secs else continue; + if (sh.sh_type == SHT_NOBITS) { + pr_warn("failed to get section(%d, %s) data from %s\n", + idx, name, path); + goto err; + } + But then we might as well test for the expected section type, which is supposed to be SHT_PROGBITS, if I understand correctly. What I don't know is whether this is *the only* possible expected type (for ".BTF", ".BTF.ext" and ".BTF.base"), or are there others? Andrii, do you know if that's the case? > > let's also use similar style of warning messaging to others, maybe > something like > > "failed to get section(%s, %d) data from %s\n" ? > > > pw-bot: cr > >> if (secs.btf_base_data) { >> dist_base_btf = btf_new(secs.btf_base_data->d_buf, secs.btf_base_data->d_size, >> NULL); >> -- >> 2.49.0 >> >>
