[lvc-project] [PATCH bpf] bpf: fix %p% runtime check in bpf_bprintf_prepare

[Florent Revest]

On Wed, Oct 9, 2024 at 12:58 PM Ilya Shchipletsov <rabbelkin at mail.ru> wrote:
>
> Fuzzing reports a warning in format_decode()
>
> Please remove unsupported %� in format string
> WARNING: CPU: 0 PID: 5091 at lib/vsprintf.c:2680 format_decode+0x1193/0x1bb0 lib/vsprintf.c:2680
> Modules linked in:
> CPU: 0 PID: 5091 Comm: syz-executor879 Not tainted 6.10.0-rc1-syzkaller-00021-ge0cce98fe279 #0
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 04/02/2024
> RIP: 0010:format_decode+0x1193/0x1bb0 lib/vsprintf.c:2680
> Call Trace:
>  <TASK>
>  bstr_printf+0x137/0x1210 lib/vsprintf.c:3253
>  ____bpf_trace_printk kernel/trace/bpf_trace.c:390 [inline]
>  bpf_trace_printk+0x1a1/0x230 kernel/trace/bpf_trace.c:375
>  bpf_prog_21da1b68f62e1237+0x36/0x41
>  bpf_dispatcher_nop_func include/linux/bpf.h:1243 [inline]
>  __bpf_prog_run include/linux/filter.h:691 [inline]
>  bpf_prog_run include/linux/filter.h:698 [inline]
>  bpf_test_run+0x40b/0x910 net/bpf/test_run.c:425
>  bpf_prog_test_run_skb+0xafa/0x13a0 net/bpf/test_run.c:1066
>  bpf_prog_test_run+0x33c/0x3b0 kernel/bpf/syscall.c:4291
>  __sys_bpf+0x48d/0x810 kernel/bpf/syscall.c:5705
>  __do_sys_bpf kernel/bpf/syscall.c:5794 [inline]
>  __se_sys_bpf kernel/bpf/syscall.c:5792 [inline]
>  __x64_sys_bpf+0x7c/0x90 kernel/bpf/syscall.c:5792
>  do_syscall_x64 arch/x86/entry/common.c:52 [inline]
>  do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
>  entry_SYSCALL_64_after_hwframe+0x77/0x7f
>
> The problem occurs when trying to pass %p% at the end of format string,
> which would result in skipping last % and passing invalid format string
> down to format_decode() that would cause warning because of invalid
> character after %.
>
> Fix issue by advancing pointer only if next char is format modifier.
> If next char is null/space/punct, then just accept formatting as is,
> without advancing the pointer.
>
> Fixes: 48cac3f4a96d ("bpf: Implement formatted output helpers with bstr_printf")
> Co-developed-by: Nikita Marushkin <hfggklm at gmail.com>
> Signed-off-by: Nikita Marushkin <hfggklm at gmail.com>
> Signed-off-by: Ilya Shchipletsov <rabbelkin at mail.ru>

This looks like
https://syzkaller.appspot.com/bug?extid=e2c932aec5c8a6e1d31c could you
add:

Reported-by: syzbot+e2c932aec5c8a6e1d31c at syzkaller.appspotmail.com

> ---
>  kernel/bpf/helpers.c | 13 +++++++++----
>  1 file changed, 9 insertions(+), 4 deletions(-)
>
> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> index c9e235807cac..bd771d6aacdb 100644
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
> @@ -892,14 +892,19 @@ int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
>                                 goto fmt_str;
>                         }
>
> +                       if (fmt[i + 1] == 'K' || fmt[i + 1] == 'x' ||
> +                           fmt[i + 1] == 's' || fmt[i + 1] == 'S') {
> +                               if (tmp_buf)
> +                                       cur_arg = raw_args[num_spec];
> +                               i++;
> +                               goto nocopy_fmt;
> +                       }
> +
>                         if (fmt[i + 1] == 0 || isspace(fmt[i + 1]) ||
> -                           ispunct(fmt[i + 1]) || fmt[i + 1] == 'K' ||
> -                           fmt[i + 1] == 'x' || fmt[i + 1] == 's' ||
> -                           fmt[i + 1] == 'S') {
> +                           ispunct(fmt[i + 1])) {
>                                 /* just kernel pointers */

Maybe we should duplicate or drop this comment ? The intent there was
to say "we only have to copy from raw_args" which apply to both blocks
now. In hindsight it doesn't seem to be a very useful comment though
so maybe it's not worth keeping around.

>                                 if (tmp_buf)
>                                         cur_arg = raw_args[num_spec];
> -                               i++;
>                                 goto nocopy_fmt;
>                         }
>
> --
> 2.43.0
>

Could you extend test_snprintf_negative() in
tools/testing/selftests/bpf/prog_tests/snprintf.c to cover %p% ? FWIW:
This exact same problem already happened in a previous life of this
code https://lkml.kernel.org/netdev/85a08645-453b-78ad-e401-55d2894fa64a@iogearbox.net/T/
so it would be interesting to add more thorough test cases to convince
ourselves that everything works well now, like %pB% too or others
maybe ?

Thanks!