Re: [PATCH bpf v2] bpf: Fix nested bpf_bprintf_prepare with more per-cpu buffers

[Florent Revest]

On Tue, May 11, 2021 at 11:07 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Tue, May 11, 2021 at 1:12 AM Florent Revest <revest@xxxxxxxxxxxx> wrote:
> >
> > The bpf_seq_printf, bpf_trace_printk and bpf_snprintf helpers share one
> > per-cpu buffer that they use to store temporary data (arguments to
> > bprintf). They "get" that buffer with try_get_fmt_tmp_buf and "put" it
> > by the end of their scope with bpf_bprintf_cleanup.
> >
> > If one of these helpers gets called within the scope of one of these
> > helpers, for example: a first bpf program gets called, uses
> > bpf_trace_printk which calls raw_spin_lock_irqsave which is traced by
> > another bpf program that calls bpf_snprintf, then the second "get"
> > fails. Essentially, these helpers are not re-entrant. They would return
> > -EBUSY and print a warning message once.
> >
> > This patch triples the number of bprintf buffers to allow three levels
> > of nesting. This is very similar to what was done for tracepoints in
> > "9594dc3c7e7 bpf: fix nested bpf tracepoints with per-cpu data"
> >
> > Fixes: d9c9e4db186a ("bpf: Factorize bpf_trace_printk and bpf_seq_printf")
> > Reported-by: syzbot+63122d0bc347f18c1884@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Signed-off-by: Florent Revest <revest@xxxxxxxxxxxx>
> > ---
> >  kernel/bpf/helpers.c | 27 ++++++++++++++-------------
> >  1 file changed, 14 insertions(+), 13 deletions(-)
> >
> > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > index 544773970dbc..ef658a9ea5c9 100644
> > --- a/kernel/bpf/helpers.c
> > +++ b/kernel/bpf/helpers.c
> > @@ -696,34 +696,35 @@ static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
> >   */
> >  #define MAX_PRINTF_BUF_LEN     512
> >
> > -struct bpf_printf_buf {
> > -       char tmp_buf[MAX_PRINTF_BUF_LEN];
> > +/* Support executing three nested bprintf helper calls on a given CPU */
> > +struct bpf_bprintf_buffers {
> > +       char tmp_bufs[3][MAX_PRINTF_BUF_LEN];
> >  };
> > -static DEFINE_PER_CPU(struct bpf_printf_buf, bpf_printf_buf);
> > -static DEFINE_PER_CPU(int, bpf_printf_buf_used);
> > +static DEFINE_PER_CPU(struct bpf_bprintf_buffers, bpf_bprintf_bufs);
> > +static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
> >
> >  static int try_get_fmt_tmp_buf(char **tmp_buf)
> >  {
> > -       struct bpf_printf_buf *bufs;
> > -       int used;
> > +       struct bpf_bprintf_buffers *bufs;
> > +       int nest_level;
> >
> >         preempt_disable();
> > -       used = this_cpu_inc_return(bpf_printf_buf_used);
> > -       if (WARN_ON_ONCE(used > 1)) {
> > -               this_cpu_dec(bpf_printf_buf_used);
> > +       nest_level = this_cpu_inc_return(bpf_bprintf_nest_level);
> > +       if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bufs->tmp_bufs))) {
> > +               this_cpu_dec(bpf_bprintf_nest_level);
>
> Applied to bpf tree.

Thanks Alexei!

> I think at the end the fix is simple enough and much better than an
> on-stack buffer.

Agree. :) I was skeptical at first but this turned out quite well in
the end, thank you for convincing me Daniel & Andrii. ;)