On Mon, Jan 11, 2021 at 1:17 PM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> On Mon, Jan 11, 2021 at 12:34:48PM -0800, Andrii Nakryiko wrote:
> > On Mon, Jan 11, 2021 at 11:18 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > >
> > > The bpf_tracing_prog_attach error path calls bpf_prog_put
> > > on prog, which causes refcount underflow when it's called
> > > from link_create function.
> > >
> > > link_create
> > > prog = bpf_prog_get <-- get
> > > ...
> > > tracing_bpf_link_attach(prog..
> > > bpf_tracing_prog_attach(prog..
> > > out_put_prog:
> > > bpf_prog_put(prog); <-- put
> > >
> > > if (ret < 0)
> > > bpf_prog_put(prog); <-- put
> > >
> > > Removing bpf_prog_put call from bpf_tracing_prog_attach
> > > and making sure its callers call it instead.
> > >
> > > Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> > > ---
> >
> > I also double-checked all other attach functions called from
> > link_create, they all seem to be fine and don't put prog on failure,
> > so this should be the only needed fix. Also, missing:
>
> it'd be easier to spot this if we use refcount_t instead of the atomic64_t,
> I replaced it for this refcount and got nice console warning for this bug
>
> then I saw:
> 85192dbf4de0 bpf: Convert bpf_prog refcnt to atomic64_t
>
> so I guess we need something like refcount64_t first
Having a non-failing refcount simplifies code quite a lot. I was
having a problem having to deal with potential refcount failure during
mmap()'ing, where it's impossible to communicate failure back to the
kernel. So if atomic64_t would never fail, but would generate a
warning on underflow, then yeah, it would be an improvement.
>
> jirka
>
> >
> > Fixes: 4a1e7c0c63e0 ("bpf: Support attaching freplace programs to
> > multiple attach points")
> >
> > Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> >
> > > kernel/bpf/syscall.c | 6 ++++--
> > > 1 file changed, 4 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> > > index c3bb03c8371f..e5999d86c76e 100644
> > > --- a/kernel/bpf/syscall.c
> > > +++ b/kernel/bpf/syscall.c
> > > @@ -2712,7 +2712,6 @@ static int bpf_tracing_prog_attach(struct bpf_prog *prog,
> > > out_put_prog:
> > > if (tgt_prog_fd && tgt_prog)
> > > bpf_prog_put(tgt_prog);
> > > - bpf_prog_put(prog);
> > > return err;
> > > }
> > >
> > > @@ -2825,7 +2824,10 @@ static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
> > > tp_name = prog->aux->attach_func_name;
> > > break;
> > > }
> > > - return bpf_tracing_prog_attach(prog, 0, 0);
> > > + err = bpf_tracing_prog_attach(prog, 0, 0);
> > > + if (err >= 0)
> > > + return err;
> > > + goto out_put_prog;
> > > case BPF_PROG_TYPE_RAW_TRACEPOINT:
> > > case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
> > > if (strncpy_from_user(buf,
> > > --
> > > 2.26.2
> > >
> >
>
