Re: [PATCH] bpf: Fix recursion check in trampoline

On Tue, Apr 27, 2021 at 06:10:32PM -0700, Alexei Starovoitov wrote:

On Tue, Apr 27, 2021 at 3:42 PM Jiri Olsa [off-list ref] wrote:

quoted

The recursion check in __bpf_prog_enter and __bpf_prog_exit leaves
some (not inlined) functions unprotected:

In __bpf_prog_enter:
  - migrate_disable is called before prog->active is checked

In __bpf_prog_exit:
  - migrate_enable,rcu_read_unlock_strict are called after
    prog->active is decreased

When attaching trampoline to them we get panic like:

  traps: PANIC: double fault, error_code: 0x0
  double fault: 0000 [#1] SMP PTI
  RIP: 0010:__bpf_prog_enter+0x4/0x50
  ...
  Call Trace:
   <IRQ>
   bpf_trampoline_6442466513_0+0x18/0x1000
   migrate_disable+0x5/0x50
   __bpf_prog_enter+0x9/0x50
   bpf_trampoline_6442466513_0+0x18/0x1000
   migrate_disable+0x5/0x50
   __bpf_prog_enter+0x9/0x50
   bpf_trampoline_6442466513_0+0x18/0x1000
   migrate_disable+0x5/0x50
   __bpf_prog_enter+0x9/0x50
   bpf_trampoline_6442466513_0+0x18/0x1000
   migrate_disable+0x5/0x50
   ...

Making the recursion check before the rest of the calls
in __bpf_prog_enter and as last call in __bpf_prog_exit.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
 kernel/bpf/trampoline.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
index 4aa8b52adf25..301735f7e88e 100644
--- a/kernel/bpf/trampoline.c
+++ b/kernel/bpf/trampoline.c

@@ -558,12 +558,12 @@ static void notrace inc_misses_counter(struct bpf_prog *prog)
 u64 notrace __bpf_prog_enter(struct bpf_prog *prog)
        __acquires(RCU)
 {
-       rcu_read_lock();
-       migrate_disable();
        if (unlikely(__this_cpu_inc_return(*(prog->active)) != 1)) {
                inc_misses_counter(prog);
                return 0;
        }
+       rcu_read_lock();
+       migrate_disable();

That obviously doesn't work.
After cpu_inc the task can migrate and cpu_dec
will happen on a different cpu likely underflowing
the counter into negative.

ugh right

We can either mark migrate_disable as nokprobe/notrace or have bpf
trampoline specific denylist.

I was using notrace to disable that, but that would limit
other tracers.. I'll add bpf denylist

jirka