| Message ID | 20201130154143.292882-1-toke@redhat.com |
|---|---|
| State | New |
| Headers | show |
| Series | [bpf] libbpf: reset errno after probing kernel features | expand |
On Mon, Nov 30, 2020 at 2:41 PM Toke Høiland-Jørgensen <toke@redhat.com> wrote: > > Andrii Nakryiko <andrii.nakryiko@gmail.com> writes: > > > On Mon, Nov 30, 2020 at 7:42 AM Toke Høiland-Jørgensen <toke@redhat.com> wrote: > >> > >> The kernel feature probing results in 'errno' being set if the probing > >> fails (as is often the case). This can stick around and leak to the caller, > >> which can lead to confusion later. So let's make sure we always reset errno > >> after calling a probe function. > > > > What specifically is the problem and what sort of confusion we are > > talking about here? You are not supposed to check errno, unless the > > function returned -1 or other error result. > > > > In some cases, you have to reset errno manually just to avoid > > confusion (see how strtol() is used, as an example). > > > > I.e., I don't see the problem here, any printf() technically can set > > errno to <0, we don't reset errno after each printf call though, > > right? > > Well yeah, technically things work fine in the common case. But this It works fine in all cases. Assuming "errno != 0 means last libc/syscall failed" is just wrong. > errno thing sent me on quite the wild goose chase when trying to find > the root cause of the pinning issue I also sent a patch for... > > So since reseting errno doesn't hurt either I figured I'd save others > ending up in similar trouble. If it's not to your taste feel free to > just drop the patch :) Yep, let's just drop it, no need to create a bad precedent. > > -Toke >
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 28baee7ba1ca..8d05132e1945 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -4021,6 +4021,8 @@ static bool kernel_supports(enum kern_feature_id feat_id) pr_warn("Detection of kernel %s support failed: %d\n", feat->desc, ret); WRITE_ONCE(feat->res, FEAT_MISSING); } + /* reset errno after probing to prevent leaking it to caller */ + errno = 0; } return READ_ONCE(feat->res) == FEAT_SUPPORTED;
The kernel feature probing results in 'errno' being set if the probing fails (as is often the case). This can stick around and leak to the caller, which can lead to confusion later. So let's make sure we always reset errno after calling a probe function. Fixes: 47b6cb4d0add ("libbpf: Make kernel feature probing lazy") Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com> --- tools/lib/bpf/libbpf.c | 2 ++ 1 file changed, 2 insertions(+)