emacs-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Preventing stack overflows with alloca.


From: Kim F. Storm
Subject: Re: Preventing stack overflows with alloca.
Date: 18 Jun 2004 14:02:00 +0200
User-agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.3.50

Kenichi Handa <address@hidden> writes:

> > But there are other risky uses of alloca, so I suggest the
> > patch below which takes care of the problems in fns.c.
>
> Oops, I have completely forgotten about this problem.
> Actually I noticed it last year and sent the attached mail,
> but haven't worked on it any further.  I remember Richard
> agreed with having the macro SAFE_ALLOCA.  Of course we need
> the pairing SAFE_FREE which frees `address' and calls
> unbind_to if necessary.


Actually, now that you mention it, I remember your mail.

I think your approach is ok, but we can make it a bit
more generic with the following approach:


/* Define this in alloc.c */

Lisp_Object
safe_alloca_unwind (arg)
     Lisp_Object arg;
{
  xfree ((void *)arg);
  return Qnil;
}


/* Add this to lisp.h */

extern Lisp_Object safe_alloca_unwind (Lisp_Object *);


#define USE_SAFE_ALLOCA                                 \
  int sa_count = SPECPDL_INDEX ()

#define SAFE_ALLOCA(buf, size)                          \
  do {                                                  \
    if (size < MAX_ALLOCA)                              \
      buf = alloca (size);                              \
    else                                                \
      record_unwind_protect (safe_alloca_unwind,        \
              (Lisp_Object)(buf = xmalloc (size)));     \
  } while (0)

#define SAFE_FREE(size)                                 \
  do {                                                  \
    if (size >= MAX_ALLOCA)                             \
      unbind_to (sa_count, Qnil);                       \
  } while (0)


Of course, if a function already uses unwind protect,
it doesn't need USE_SAFE_ALLOCA and SAFE_FREE.


Since old_value is not a lisp object here,
we would need to fix (hack) GC like this:

  for (bind = specpdl; bind != specpdl_ptr; bind++)
    {
      mark_object (bind->symbol);
      if (bind->func != safe_alloca_unwind)
        mark_object (bind->old_value);
    }



A sample use would look like this:

Lisp_Object
string_make_multibyte (string)
     Lisp_Object string;
{
  unsigned char *buf;
  int nbytes;
  Lisp_Object ret;
  USE_SAFE_ALLOCA;

  if (STRING_MULTIBYTE (string))
    return string;

  nbytes = count_size_as_multibyte (SDATA (string),
                                    SCHARS (string));
  /* If all the chars are ASCII, they won't need any more bytes
     once converted.  In that case, we can return STRING itself.  */
  if (nbytes == SBYTES (string))
    return string;

  SAFE_ALLOCA (buf, nbytes);

  copy_text (SDATA (string), buf, SBYTES (string),
             0, 1);

  ret = make_multibyte_string (buf, SCHARS (string), nbytes);

  SAFE_FREE (nbytes);

  return ret;
}


--
Kim F. Storm <address@hidden> http://www.cua.dk





reply via email to

[Prev in Thread] Current Thread [Next in Thread]