Re: [PATCH v4l-utils v1] v4l2-tracer: fix 'symbol mmap64/open64 is already defined' compile failure

[Peter Seiderer <ps.report@gmx.net>]

Hello Hans,

On Fri, 31 Mar 2023 10:27:18 +0200, Hans Verkuil <hverkuil@xs4all.nl> wrote:

> Hi Peter,
>
> On 30/03/2023 17:06, Peter Seiderer wrote:
> > Compiling for RPi4 (64-bit) using buildroot failes with the following
> > error:
> >
> >   .../host/bin/aarch64-buildroot-linux-gnu-g++ -DHAVE_CONFIG_H -I. -I../.. -I../../utils/common -I.../aarch64-buildroot-linux-gnu/sysroot/usr/include/json-c -I../../lib/include -Wall -Wpointer-arith -D_GNU_SOURCE -I../../include -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Os -g0 -D_FORTIFY_SOURCE=1 -std=gnu++11 -c libv4l2tracer.cpp  -fPIC -DPIC -o .libs/libv4l2tracer_la-libv4l2tracer.o
> >   /tmp/ccfbectY.s: Assembler messages:
> >   /tmp/ccfbectY.s:208: Error: symbol `open64' is already defined
> >   /tmp/ccfbectY.s:762: Error: symbol `mmap64' is already defined
> >
> > The preprocessor output shows:
> >
> >   [...]
> >   extern "C" {
> >   # 61 ".../host/aarch64-buildroot-linux-gnu/sysroot/usr/include/sys/mman.h" 3 4
> >   extern void * mmap (void *__addr, size_t __len, int __prot, int __flags, int __fd, __off64_t __offset) noexcept (true) __asm__ ("" "mmap64");
> >   [...]
> >   extern void *mmap64 (void *__addr, size_t __len, int __prot,
> >          int __flags, int __fd, __off64_t __offset) noexcept (true);
> >
> > And host/aarch64-buildroot-linux-gnu/sysroot/usr/include/sys/mman.h:
> >
> >   56 #ifndef __USE_FILE_OFFSET64
> >   57 extern void *mmap (void *__addr, size_t __len, int __prot,
> >   58                    int __flags, int __fd, __off_t __offset) __THROW;
> >   59 #else
> >   60 # ifdef __REDIRECT_NTH
> >   61 extern void * __REDIRECT_NTH (mmap,
> >   62                               (void *__addr, size_t __len, int __prot,
> >   63                                int __flags, int __fd, __off64_t __offset),
> >   64                               mmap64);
> >   65 # else
> >   66 #  define mmap mmap64
> >   67 # endif
> >   68 #endif
> >   69 #ifdef __USE_LARGEFILE64
> >   70 extern void *mmap64 (void *__addr, size_t __len, int __prot,
> >   71                      int __flags, int __fd, __off64_t __offset) __THROW;
> >   72 #endif
> >
> > Fix it by applying the same undef _LARGEFILE_SOURCE/_FILE_OFFSET_BITS,
> > define _LARGEFILE64_SOURCE as in as in lib/libv4l1/v4l1compat.c
>
> If I look at 'man feature_test_macros', then that man page suggests that
> _LARGEFILE64_SOURCE is out of date and instead you should use
> '#define _FILE_OFFSET_BITS 64'.
>
> Can you test that? If this works, then it should be applied to v4l1compat.c
> as well.

Did you mean as utils/v4l2-tracer/meson.build all ready does:

 73 libv4l2_tracer_cpp_args = [
 74     # Meson enables large file support unconditionally, which redirects file
 75     # operations to 64-bit versions. This results in some symbols being
 76     # renamed, for instance open() being renamed to open64(). As the library
 77     # needs to provide both 32-bit and 64-bit versions of file operations,
 78     # disable transparent large file support.
 79     '-U_FILE_OFFSET_BITS',
 80     '-D_FILE_OFFSET_BITS=32',
 81     '-D_LARGEFILE64_SOURCE',
 82 ]

Did my previous test against the 1.24.1 release using the autoconf/automake
build system...

A quick (compile) test for lib/libv4l1/v4l1compat.c with

	#undef _FILE_OFFSET_BITS
	#define _FILE_OFFSET_BITS 32
	#define _LARGEFILE64_SOURCE 1

seems to work...

>
> >
> > Signed-off-by: Peter Seiderer <ps.report@gmx.net>
> > ---
> > Notes:
> >
> >   - maybe the 'defined(linux) && defined(__GLIBC__)' protection
> >     present in lib/libv4l1/v4l1compat.c for open64/mmap64 is needed
> >     for non glibc compiles of utils/v4l2-tracer/libv4l2tracer.cpp too?
>
> I think we need this too. If nothing else, it is consistent.

So this patch should be o.k. for the stable-1.24 branch?

Will send one adding the 'defined(linux) && defined(__GLIBC__)' to
v4l2-tracer and one changing the in-file undef/define to the
meson.build solution for lib/libv4l1/v4l1compat.c (in case this
is o.k. here)...

Regards,
Peter

>
> Regards,
>
> 	Hans
>
> > ---
> >  utils/v4l2-tracer/libv4l2tracer.cpp | 5 +++++
> >  1 file changed, 5 insertions(+)
> >
> > diff --git a/utils/v4l2-tracer/libv4l2tracer.cpp b/utils/v4l2-tracer/libv4l2tracer.cpp
> > index a9f039c7..1e3900db 100644
> > --- a/utils/v4l2-tracer/libv4l2tracer.cpp
> > +++ b/utils/v4l2-tracer/libv4l2tracer.cpp
> > @@ -3,6 +3,11 @@
> >   * Copyright 2022 Collabora Ltd.
> >   */
> >
> > +/* ensure we see *64 variants and they aren't transparently used */
> > +#undef _LARGEFILE_SOURCE
> > +#undef _FILE_OFFSET_BITS
> > +#define _LARGEFILE64_SOURCE 1
> > +
> >  #include "trace.h"
> >  #include <dlfcn.h>
> >  #include <stdarg.h>
>