Re: [PATCH bpf] bpf: Set run context for rawtp test_run callback

[Jiri Olsa <olsajiri@gmail.com>]

On Mon, Jun 03, 2024 at 09:25:47AM -0700, Alexei Starovoitov wrote:
> On Mon, Jun 3, 2024 at 4:14 AM Jiri Olsa <jolsa@kernel.org> wrote:
> >
> > syzbot reported crash when rawtp program executed through the
> > test_run interface calls bpf_get_attach_cookie helper or any
> > other helper that touches task->bpf_ctx pointer.
> >
> > We need to setup bpf_ctx pointer in rawtp test_run as well,
> > so fixing this by moving __bpf_trace_run in header file and
> > using it in test_run callback.
> >
> > Also renaming __bpf_trace_run to bpf_prog_run_trace.
> >
> > Fixes: 7adfc6c9b315 ("bpf: Add bpf_get_attach_cookie() BPF helper to access bpf_cookie value")
> > Reported-by: syzbot+3ab78ff125b7979e45f9@syzkaller.appspotmail.com
> > Closes: https://syzkaller.appspot.com/bug?extid=3ab78ff125b7979e45f9
> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > ---
> >  include/linux/bpf.h      | 27 +++++++++++++++++++++++++++
> >  kernel/trace/bpf_trace.c | 28 ++--------------------------
> >  net/bpf/test_run.c       |  4 +---
> >  3 files changed, 30 insertions(+), 29 deletions(-)
> >
> > diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> > index 5e694a308081..4eb803b1d308 100644
> > --- a/include/linux/bpf.h
> > +++ b/include/linux/bpf.h
> > @@ -2914,6 +2914,33 @@ static inline void bpf_dynptr_set_rdonly(struct bpf_dynptr_kern *ptr)
> >  }
> >  #endif /* CONFIG_BPF_SYSCALL */
> >
> > +static __always_inline int
> > +bpf_prog_run_trace(struct bpf_prog *prog, u64 cookie, u64 *ctx,
> > +                  bpf_prog_run_fn run_prog)
> > +{
> > +       struct bpf_run_ctx *old_run_ctx;
> > +       struct bpf_trace_run_ctx run_ctx;
> > +       int ret = -1;
> > +
> > +       cant_sleep();
> 
> I suspect you should see a splat with that.

hum, __bpf_prog_test_run_raw_tp is called with preempt_disable,
so I think it should be fine

> 
> Overall I think it's better to add empty run_ctx to
> __bpf_prog_test_run_raw_tp()
> instead of moving such a big function to .h
> 
> No need for prog->active increments. test_run is running
> from syscall. If the same prog is attached somewhere as well
> it may recurse once and it's fine imo.

heh, it was my first change, then I was thinking let's not duplicate the
code and re-use the existing function.. but it's true that there's no
use for the prog->active intest_run interface

jirka

> 
> pw-bot: cr
> 
> > +       if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
> > +               bpf_prog_inc_misses_counter(prog);
> > +               goto out;
> > +       }
> > +
> > +       run_ctx.bpf_cookie = cookie;
> > +       old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
> > +
> > +       rcu_read_lock();
> > +       ret = run_prog(prog, ctx);
> > +       rcu_read_unlock();
> > +
> > +       bpf_reset_run_ctx(old_run_ctx);
> > +out:
> > +       this_cpu_dec(*(prog->active));
> > +       return ret;
> > +}
> > +
> >  static __always_inline int
> >  bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
> >  {
> > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > index d1daeab1bbc1..8a23ef42b76b 100644
> > --- a/kernel/trace/bpf_trace.c
> > +++ b/kernel/trace/bpf_trace.c
> > @@ -2383,31 +2383,6 @@ void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
> >         preempt_enable();
> >  }
> >
> > -static __always_inline
> > -void __bpf_trace_run(struct bpf_raw_tp_link *link, u64 *args)
> > -{
> > -       struct bpf_prog *prog = link->link.prog;
> > -       struct bpf_run_ctx *old_run_ctx;
> > -       struct bpf_trace_run_ctx run_ctx;
> > -
> > -       cant_sleep();
> > -       if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
> > -               bpf_prog_inc_misses_counter(prog);
> > -               goto out;
> > -       }
> > -
> > -       run_ctx.bpf_cookie = link->cookie;
> > -       old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
> > -
> > -       rcu_read_lock();
> > -       (void) bpf_prog_run(prog, args);
> > -       rcu_read_unlock();
> > -
> > -       bpf_reset_run_ctx(old_run_ctx);
> > -out:
> > -       this_cpu_dec(*(prog->active));
> > -}
> > -
> >  #define UNPACK(...)                    __VA_ARGS__
> >  #define REPEAT_1(FN, DL, X, ...)       FN(X)
> >  #define REPEAT_2(FN, DL, X, ...)       FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
> > @@ -2437,7 +2412,8 @@ void __bpf_trace_run(struct bpf_raw_tp_link *link, u64 *args)
> >         {                                                               \
> >                 u64 args[x];                                            \
> >                 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);                  \
> > -               __bpf_trace_run(link, args);                            \
> > +               (void) bpf_prog_run_trace(link->link.prog, link->cookie,\
> > +                                         args, bpf_prog_run);          \
> >         }                                                               \
> >         EXPORT_SYMBOL_GPL(bpf_trace_run##x)
> >  BPF_TRACE_DEFN_x(1);
> > diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
> > index f6aad4ed2ab2..84d1c91b01ab 100644
> > --- a/net/bpf/test_run.c
> > +++ b/net/bpf/test_run.c
> > @@ -728,9 +728,7 @@ __bpf_prog_test_run_raw_tp(void *data)
> >  {
> >         struct bpf_raw_tp_test_run_info *info = data;
> >
> > -       rcu_read_lock();
> > -       info->retval = bpf_prog_run(info->prog, info->ctx);
> > -       rcu_read_unlock();
> > +       info->retval = bpf_prog_run_trace(info->prog, 0, info->ctx, bpf_prog_run);
> >  }
> >
> >  int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
> > --
> > 2.45.1
> >