[PATCH] aarch64: Fix error messages for GCS and BTI incompatible modules

[Yury Khrustalev]

Thanks for your comments, Adhemerval!

On Fri, Dec 19, 2025 at 11:17:55AM -0300, Adhemerval Zanella Netto wrote:
> 
> On 12/12/25 09:14, Yury Khrustalev wrote:
> > When either program path of module name is empty, don't print an
> > empty string followed by a colon.
> 
> Afaik the module name will be always not empty.

OK, I'll fix the commit message.

> > ---
> >  sysdeps/aarch64/dl-bti.c | 15 ++++++++++++---
> >  sysdeps/aarch64/dl-gcs.c | 16 ++++++++++++----
> >  2 files changed, 24 insertions(+), 7 deletions(-)
> > 
> > diff --git a/sysdeps/aarch64/dl-bti.c b/sysdeps/aarch64/dl-bti.c
> > index e654dde484..c506090485 100644
> > --- a/sysdeps/aarch64/dl-bti.c
> > +++ b/sysdeps/aarch64/dl-bti.c
> > @@ -62,17 +62,26 @@ _dl_bti_protect (struct link_map *map, int fd)
> >        }
> >  }
> >  
> > +#define STR_NOT_EMPTY(x) ((x != NULL) && x[0])
> >  
> >  static void
> >  bti_failed (struct link_map *l, const char *program)
> >  {
> > -  if (program)
> > +  if (STR_NOT_EMPTY (program) && STR_NOT_EMPTY (l->l_name))
> > +    /* A program's dependency is not GCS compatible.  */
> >      _dl_fatal_printf ("%s: %s: failed to turn on BTI protection\n",
> >  		      program, l->l_name);
> > +  else if (STR_NOT_EMPTY (program))
> > +    /* The program itself is not GCS compatible.  */
> > +    _dl_fatal_printf ("%s: failed to turn on BTI protection\n", program);
> > +  else if (program)
> > +    /* For static binaries, program will be an empty string.  */
> > +    _dl_fatal_printf ("error: failed to turn on BTI protection\n");
> >    else
> > -    /* Note: the errno value is not available any more.  */
> > +    /* If program is NULL, we are processing a dlopen operation.
> > +       Note: the errno value is not available any more.  */
> >      _dl_signal_error (0, l->l_name, "dlopen",
> > -		      N_("failed to turn on BTI protection"));
> > +		      "failed to turn on BTI protection");
> >  }
> >  
> 
> We can assume that l->l_name is always non NULL (even though there is
> some code that assumes otherwise) and if is not empty then program
> is also non-null:
> 
>   elf/rtld.c
>   \_ _rtld_main_check (main_map, _dl_argv[0]);
>       \_ _dl_bti_check (m, program);
>          \_ _dl_bti_check
> 
> In this case _dl_argv is always non-null as well.
> 
> If program is NULL, we can then assume that l->l_name always contains
> a proper name (dlopen case)
> 
> So I think we can simplify to:
> 
>   static void
>   bti_failed (struct link_map *l, const char *program)
>   {
>     if (l->l_name[0] == '\0')
>       {
>         if (program[0] == '\0')
>           _dl_fatal_printf ("error: failed to turn on BTI protection\n");
>         else
>           _dl_fatal_printf ("%s: failed to turn on BTI protection\n",
>                            program);
>       }
>     else
>       /* Note: the errno value is not available any more.  */
>       _dl_signal_error (0, l->l_name, "dlopen",
>                         N_("failed to turn on BTI protection"));
>   }
> 
> The dynamic linked binaries will now issue:
> 
>   <program>: failed to turn on BTI protection
> 
> And static/static-pie will issue:
> 
>   error: failed to turn on BTI protection
> 
> Unfortunately we can safely enable an 'assert (program != NULL)' on the
> startup case because of [1]
> 
> [1] https://patchwork.sourceware.org/project/glibc/patch/20251117202613.2565803-5-adhemerval.zanella@linaro.org/

OK, I'll try that, thanks!