Re: buffer overflow check bug

Re: buffer overflow check bug

[Dan Carpenter]

Gar...  I have no idea.  That seems like an issue in sparse.

Does anyone know why MOD_NORETURN gets set for:
extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));

but not for:
extern __attribute__ ((__noreturn__)) void my_exit(const char*, ...);

GCC seems to accept both formats.

regards,
dan carpenter

Re: buffer overflow check bug

[Josh Triplett]

On Mon, Jun 18, 2012 at 10:53:44PM +0300, Dan Carpenter wrote:
> Gar...  I have no idea.  That seems like an issue in sparse.
> 
> Does anyone know why MOD_NORETURN gets set for:
> extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));
> 
> but not for:
> extern __attribute__ ((__noreturn__)) void my_exit(const char*, ...);
> 
> GCC seems to accept both formats.

I wonder if the latter gets incorrectly parsed as though the attribute
attached to "void" rather than "my_exit"?

- Josh Triplett

Re: buffer overflow check bug

[Dan Carpenter]

On Mon, Jun 18, 2012 at 01:04:07PM -0700, Josh Triplett wrote:
> On Mon, Jun 18, 2012 at 10:53:44PM +0300, Dan Carpenter wrote:
> > Gar...  I have no idea.  That seems like an issue in sparse.
> > 
> > Does anyone know why MOD_NORETURN gets set for:
> > extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));
> > 
> > but not for:
> > extern __attribute__ ((__noreturn__)) void my_exit(const char*, ...);
> > 
> > GCC seems to accept both formats.
> 
> I wonder if the latter gets incorrectly parsed as though the attribute
> attached to "void" rather than "my_exit"?
> 

I put a:
	printf("%d %d token = %s\n", token->pos.line, token->pos.pos, token->ident->name);
in handle_attributes() and it only prints when the attribute is at
the end.

regards,
dan carpenter

Re: buffer overflow check bug

[Xi Wang]

On Jun 18, 2012, at 3:53 PM, Dan Carpenter wrote:

> Gar...  I have no idea.  That seems like an issue in sparse.
> 
> Does anyone know why MOD_NORETURN gets set for:
> extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));
> 
> but not for:
> extern __attribute__ ((__noreturn__)) void my_exit(const char*, ...);
> 
> GCC seems to accept both formats.

Which version of sparse are you using?  Everything looks good to me here.

  extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));
  extern __attribute__ ((__noreturn__)) void another_exit(const char*, ...);
  void foo(void) { my_exit(""); }
  void bar(void) { another_exit(""); }

My backend "splay" emits:

  declare void @my_exit(i8*, ...) noreturn
  declare void @another_exit(i8*, ...) noreturn
  ...

The noreturn attribute is set simply by:

  if (sym->ctype.modifiers & MOD_NORETURN)
      LLVMAddFunctionAttr(func, LLVMNoReturnAttribute);

- xi

Re: buffer overflow check bug

[Dan Carpenter]

On Tue, Jun 19, 2012 at 04:37:20PM -0400, Xi Wang wrote:
> On Jun 18, 2012, at 3:53 PM, Dan Carpenter wrote:
> 
> > Gar...  I have no idea.  That seems like an issue in sparse.
> > 
> > Does anyone know why MOD_NORETURN gets set for:
> > extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));
> > 
> > but not for:
> > extern __attribute__ ((__noreturn__)) void my_exit(const char*, ...);
> > 
> > GCC seems to accept both formats.
> 
> Which version of sparse are you using?  Everything looks good to me here.

I'm on sparse 0.4.4.  Which version are you on?

> 
>   extern void my_exit(const char*, ...) __attribute__ ((__noreturn__));
>   extern __attribute__ ((__noreturn__)) void another_exit(const char*, ...);
>   void foo(void) { my_exit(""); }
>   void bar(void) { another_exit(""); }
> 
> My backend "splay" emits:
> 
>   declare void @my_exit(i8*, ...) noreturn
>   declare void @another_exit(i8*, ...) noreturn
>   ...
> 
> The noreturn attribute is set simply by:
> 
>   if (sym->ctype.modifiers & MOD_NORETURN)
>       LLVMAddFunctionAttr(func, LLVMNoReturnAttribute);

Yep.  I use the same test, but sym->ctype.modifiers is not set for
me. The debug printf I added in handle_attributes() should have
printed.

regards,
dan carpenter