[ruby-core:109155] [Ruby master Bug#18893] Don't redefine memcpy(3)

Issue #18893 has been updated by alx (Alejandro Colomar).


shyouhei (Shyouhei Urabe) wrote in #note-8:
> This is my take:
> 
> ```patch
> From 505d00311d82fdbed80688e9db792c3b54b8ee27 Mon Sep 17 00:00:00 2001
> From: 卜部昌平
>  <[e-mail]>
> Date: Wed, 6 Jul 2022 09:18:28 +0900
> Subject: [PATCH] do not define our own version of memcpy
> 
> The (sole) use of memcpy in our public header is now replaced to
> directly call ruby_nonempty_memcpy, and the previous definition of
> memcpy is now internal-only.
> ---
>  include/ruby/internal/memory.h | 6 +-----
>  internal.h                     | 3 +++
>  2 files changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/include/ruby/internal/memory.h b/include/ruby/internal/memory.h
> index aa3464465d..0f59262a91 100644
> --- a/include/ruby/internal/memory.h
> +++ b/include/ruby/internal/memory.h
> @@ -363,7 +363,7 @@ typedef uint128_t DSIZE_T;
>   * @return  `p1`.
>   * @post    First `n` elements of `p2` are copied into `p1`.
>   */
> -#define MEMCPY(p1,p2,type,n) memcpy((p1), (p2), rbimpl_size_mul_or_raise(sizeof(type), (n)))
> +#define MEMCPY(p1,p2,type,n) ruby_nonempty_memcpy((p1), (p2), rbimpl_size_mul_or_raise(sizeof(type), (n)))
>  
>  /**
>   * Handy macro to call memmove.
> @@ -644,7 +644,6 @@ rb_alloc_tmp_buffer2(volatile VALUE *store, long count, size_t elsize)
>      return rb_alloc_tmp_buffer_with_count(store, total_size, cnt);
>  }
>  
> -#if ! defined(__MINGW32__) && ! defined(__DOXYGEN__)
>  RBIMPL_SYMBOL_EXPORT_BEGIN()
>  RBIMPL_ATTR_NOALIAS()
>  RBIMPL_ATTR_NONNULL((1))
> @@ -663,8 +662,5 @@ ruby_nonempty_memcpy(void *dest, const void *src, size_t n)
>      }
>  }
>  RBIMPL_SYMBOL_EXPORT_END()
> -#undef memcpy
> -#define memcpy ruby_nonempty_memcpy
> -#endif
>  
>  #endif /* RBIMPL_MEMORY_H */
> diff --git a/internal.h b/internal.h
> index 00a8295295..7e1abcbab8 100644
> --- a/internal.h
> +++ b/internal.h
> @@ -106,4 +106,7 @@ RUBY_SYMBOL_EXPORT_END
>  #define RBOOL(v) ((v) ? Qtrue : Qfalse)
>  #define RB_BIGNUM_TYPE_P(x) RB_TYPE_P((x), T_BIGNUM)
>  
> +#ifndef __MINGW32__
> +#define memcpy ruby_nonempty_memcpy

Is this header something that's not installed in the system?
Otherwise, the problem would remain I guess.
If it's not installed, it makes sense (it would not define memcyp(3) for others.

Thanks,

Alex

> +#endif
>  #endif /* RUBY_INTERNAL_H */
> -- 
> 2.17.1
> 
> ```



----------------------------------------
Bug #18893: Don't redefine memcpy(3)
https://bugs.ruby-lang.org/issues/18893#change-98298

* Author: alx (Alejandro Colomar)
* Status: Open
* Priority: Normal
* Assignee: shyouhei (Shyouhei Urabe)
* ruby -v: ruby 3.0.4p208 (2022-04-12 revision 3fa771dded) [x86_64-linux-gnu]
* Backport: 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN
----------------------------------------
It is Undefined Behavior, by any standard ever issued.


See what I have in my system right now:


```
alx@asus5775:/usr/include$ grepc memcpy
./string.h:43:
extern void *memcpy (void *__restrict __dest, const void *__restrict __src,
		     size_t __n) __THROW __nonnull ((1, 2));


./x86_64-linux-gnu/ruby-3.0.0/rb_mjit_min_header-3.0.4.h:1520:
extern void *memcpy (void *__restrict __dest, const void *__restrict __src,
       size_t __n) __attribute__ ((__nothrow__ , __leaf__)) __attribute__ ((__nonnull__ (1, 2)));


./x86_64-linux-gnu/ruby-3.0.0/rb_mjit_min_header-3.0.4.h:1670:
extern __inline __attribute__ ((__always_inline__)) __attribute__ ((__gnu_inline__)) __attribute__ ((__artificial__)) void *
__attribute__ ((__nothrow__ , __leaf__)) memcpy (void *__restrict __dest, const void *__restrict __src, size_t __len)
{
  return __builtin___memcpy_chk (__dest, __src, __len,
     __builtin_object_size (__dest, 0));
}


./ruby-3.0.0/ruby/internal/memory.h:278:
#define memcpy ruby_nonempty_memcpy


./x86_64-linux-gnu/ruby-3.0.0/rb_mjit_min_header-3.0.4.h:22679:
#define memcpy ruby_nonempty_memcpy


$ grepc ruby_nonempty_memcpy
./ruby-3.0.0/ruby/internal/memory.h:266:
static inline void *
ruby_nonempty_memcpy(void *dest, const void *src, size_t n)
{
    if (n) {
        return memcpy(dest, src, n);
    }
    else {
        return dest;
    }
}


./x86_64-linux-gnu/ruby-3.0.0/rb_mjit_min_header-3.0.4.h:5673:
__attribute__((__nonnull__ (1)))
__attribute__((__returns_nonnull__))
static inline void *
ruby_nonempty_memcpy(void *dest, const void *src, size_t n)
{
    if (n) {
        return memcpy(dest, src, n);
    }
    else {
        return dest;
    }
}
```

Some code that I maintain includes some ruby headers, which end up defining memcpy(3) to that thing.  Then, my code calls memcpy(3) from a function, which happens to be inline (yes, inline, not static inline, which is a horrible thing), and I get an error from the compiler, for using a static function within a non-static inline function.

So, I'd like you to please remove that definition from public headers, and refrain from redefining any ISO C functions.
If not, please define it to something not broken (and yes, static inline is broken, as it produces duplicated code when not inlined; you could use [[gnu::always_inline]] if you want to keep static, but don't).  Just C99 inline would be reasonable.  See also: <https://www.greenend.org.uk/rjk/tech/inline.html>

Thanks,

Alex



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>