Re: [PATCH v3] vfprintf-internal: Get rid of alloca.

[Joe Simmons-Talbott <josimmon@redhat.com>]

On Mon, Jun 12, 2023 at 05:15:46PM -0300, Adhemerval Zanella Netto wrote:
> 
> 
> On 22/05/23 14:06, Joe Simmons-Talbott via Libc-alpha wrote:
> > Avoid potential stack overflow from unbounded alloca.  Use the existing
> > scratch_buffer instead.
> > ---
> > Changes to v2:
> > * Don't assume the first scratch buffer resize adds enough space.
> > 
> >  stdio-common/vfprintf-internal.c | 23 ++++++++++++++++++++++-
> >  1 file changed, 22 insertions(+), 1 deletion(-)
> > 
> > diff --git a/stdio-common/vfprintf-internal.c b/stdio-common/vfprintf-internal.c
> > index c76c06e49b..20fd3c3043 100644
> > --- a/stdio-common/vfprintf-internal.c
> > +++ b/stdio-common/vfprintf-internal.c
> > @@ -1066,6 +1066,8 @@ printf_positional (struct Xprintf_buffer * buf, const CHAR_T *format,
> >    union printf_arg *args_value;
> >    int *args_size;
> >    int *args_type;
> > +  int *args_pa_user;
> > +  size_t args_pa_user_offset;
> >    {
> >      /* Calculate total size needed to represent a single argument
> >         across all three argument-related arrays.  */
> > @@ -1082,6 +1084,7 @@ printf_positional (struct Xprintf_buffer * buf, const CHAR_T *format,
> >         now.  */
> >      args_size = &args_value[nargs].pa_int;
> >      args_type = &args_size[nargs];
> > +    args_pa_user = &args_type[nargs];
> >      memset (args_type, (mode_flags & PRINTF_FORTIFY) != 0 ? '\xff' : '\0',
> >  	    nargs * sizeof (*args_type));
> >    }
> > @@ -1171,7 +1174,25 @@ printf_positional (struct Xprintf_buffer * buf, const CHAR_T *format,
> >  	else if (__glibc_unlikely (__printf_va_arg_table != NULL)
> >  		 && __printf_va_arg_table[args_type[cnt] - PA_LAST] != NULL)
> >  	  {
> > -	    args_value[cnt].pa_user = alloca (args_size[cnt]);
> > +	    while (args_pa_user + args_size[cnt] >
> > +	        (int *) &argsbuf + argsbuf.length)
> > +	      {
> > +		args_pa_user_offset = args_pa_user - &args_type[nargs];
> > +	        if (!scratch_buffer_grow_preserve (&argsbuf))
> > +	          {
> > +	            Xprintf_buffer_mark_failed (buf);
> > +	            goto all_done;
> > +	          }
> > +                args_value = argsbuf.data;
> > +                /* Set up the remaining two arrays to each point past the end of
> > +                   the prior array, since space for all three has been allocated
> > +                   now.  */
> > +                args_size = &args_value[nargs].pa_int;
> > +                args_type = &args_size[nargs];
> > +                args_pa_user = &args_type[nargs] + args_pa_user_offset;
> > +	      }
> > +	    args_value[cnt].pa_user = args_pa_user;
> > +	    args_pa_user += args_size[cnt];
> >  	    (*__printf_va_arg_table[args_type[cnt] - PA_LAST])
> >  	      (args_value[cnt].pa_user, ap_savep);
> >  	  }
> 
> This does not seem to be fully correct, trying to come up with a testcase
> to exercise this code path show a failure:
> 
> diff --git a/stdio-common/tst-vfprintf-user-type.c b/stdio-common/tst-vfprintf-user-type.c
> index 7cc69dc716..9058bc1978 100644
> --- a/stdio-common/tst-vfprintf-user-type.c
> +++ b/stdio-common/tst-vfprintf-user-type.c
> @@ -189,6 +189,52 @@ do_test (void)
>    TEST_COMPARE_STRING (str, "[[(123, 456.000000)]]");
>    free (str);
> 
> +  str = NULL;
> +  TEST_VERIFY (asprintf_alias (&str, "[[%1$P %1$P]]", 123L, 457.0) >= 0);
> +  TEST_COMPARE_STRING (str, "[[(123, 457.000000) (123, 457.000000)]]");
> +  free (str);
> +
> +  str = NULL;
> +  TEST_VERIFY (asprintf_alias (&str, "%1$P %2$P %3$P %4$P %5$P %6$P"
> +                                    "%7$P %8$P %9$P %10$P %11$P %12$P"
> +                                    "%13$P %14$P %15$P %16$P %17$P %18$P"
> +                                    "%19$P %20$P %21$P %22$P %23$P %24$P"
> +                                    "%25$P %26$P %27$P %28$P", /*%*29$P %30$P",*/
> +                              1L, 1.0,
> +                              2L, 2.0,
> +                              3L, 3.0,
> +                              4L, 4.0,
> +                              5L, 6.0,
> +                              6L, 6.0,
> +                              7L, 7.0,
> +                              8L, 8.0,
> +                              9L, 9.0,
> +                              10L, 10.0,
> +                              11L, 11.0,
> +                              12L, 12.0,
> +                              13L, 13.0,
> +                              14L, 14.0,
> +                              15L, 15.0,
> +                              16L, 16.0,
> +                              17L, 17.0,
> +                              18L, 18.0,
> +                              19L, 19.0,
> +                              20L, 20.0,
> +                              21L, 21.0,
> +                              22L, 22.0,
> +                              23L, 23.0,
> +                              24L, 34.0,
> +                              25L, 25.0,
> +                              26L, 26.0,
> +                              27L, 27.0,
> +                              28L, 28.0,
> +                              29L, 29.0,
> +                              30L, 30.0)
> +              >= 0);
> +  printf ("str=%s\n", str);
> +  free (str);
> +
> +
>    str = NULL;
>    TEST_VERIFY (asprintf_alias (&str, "[[%1$P %1$P]]", 123L, 457.0) >= 0);
>    TEST_COMPARE_STRING (str, "[[(123, 457.000000) (123, 457.000000)]]");
> 
> 
> $ ./testrun.sh stdio-common/tst-vfprintf-user-type
> Didn't expect signal from child: got `Segmentation fault'
> 
> While I would expect to:
> 
> $ ./testrun.sh stdio-common/tst-vfprintf-user-type
> str=(1, 1.000000) (2, 2.000000) (3, 3.000000) (4, 4.000000) (5, 6.000000) (6, 6.000000)(7, 7.000000) (8, 8.000000) (9, 9.000000) (10, 10.000000) (11, 11.000000) (12, 12.000000)(13, 13.000000) (14, 14.000000) (15, 15.000000) (16, 16.000000) (17, 17.000000) (18, 18.000000)(19, 19.000000) (20, 20.000000) (21, 21.000000) (22, 22.000000) (23, 23.000000) (24, 34.000000)(25, 25.000000) (26, 26.000000) (27, 27.000000) (28, 28.000000)
> 
> So I think it would be better to extend stdio-common/tst-vfprintf-user-type
> to use a large number of input arguments tests, and check why this code
> snippet is doing wrong.

Thanks for the testcase.  I added it to v4 of my patch and fixed the code to
properly use (void *) rather than (int *) which was precenting the while
loop from triggering the call to grow the scratch_buffer.

Thanks,
Joe