[PATCH v3] fakeroot: fix to work with glibc 2.33

[Felix Fietkau]

On 2021-02-14 05:41, Ilya Lipnitskiy wrote:
> The following commit removed _STAT_VER definitions from glibc:
> https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=8ed005daf0ab03e142500324a34087ce179ae78e
> 
> That subsequently broke fakeroot:
> https://bugs.archlinux.org/task/69572
> https://bugzilla.redhat.com/show_bug.cgi?id=1889862#c13
> https://forum.openwrt.org/t/unable-to-build-toolchain-fakeroot-fails-perhaps-others-after-it/87966
> 
> Make the patch based on Jan Pazdziora's suggestion from here:
> https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/message/SMQ3RYXEYTVZH6PLQMKNB3NM4XLPMNZO/
> 
> Add wrappers for newly exported symbols in glibc.
> 
> Apply patch from Debian to fix warnings in fts_read and fts_children:
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=676428
> https://sources.debian.org/patches/fakeroot/1.25.3-1.1/eglibc-fts-without-LFS/
> 
> Fix __xmknod{,at} dev pointer argument. Switch default to assume * and
> not the absence of *. On glibc 2.33+, there is no definition for these
> functions in header files, so the compile test doesn't work. But, we
> can default to using the pointer (as is the case with newer glibc), and
> use the header file on older platforms to fail the test and use no pointer.
> 
> Tested on my x86_64 Arch Linux machine, fakeroot unit tests pass.
> Also tested by building various .ipks and examining the tar contents, to
> ensure that the owner uid/gid was 0/0.
> 
> Signed-off-by: Ilya Lipnitskiy <ilya.lipnitskiy at gmail.com>
> ---
>  .../300-glibc-2.33-compatibility.patch        | 145 ++++++++++++++++++
>  1 file changed, 145 insertions(+)
>  create mode 100644 tools/fakeroot/patches/300-glibc-2.33-compatibility.patch
> 
> diff --git a/tools/fakeroot/patches/300-glibc-2.33-compatibility.patch b/tools/fakeroot/patches/300-glibc-2.33-compatibility.patch
> new file mode 100644
> index 0000000000..a460cace0c
> --- /dev/null
> +++ b/tools/fakeroot/patches/300-glibc-2.33-compatibility.patch
> @@ -0,0 +1,145 @@
> +--- a/libfakeroot.c
> ++++ b/libfakeroot.c
> +@@ -1358,6 +1368,54 @@ int renameat(int olddir_fd, const char *
> + #endif /* HAVE_FSTATAT */
> + 
> + 
> ++#if defined(__GLIBC__) && __GLIBC_PREREQ(2,33)
Turns out combining those two into a single #if was not such a good
idea. It breaks on macOS, because __GLIBC_PREREQ is evaluated though
__GLIBC__ is not defined.
Splitting it into two #if lines makes things work again.
The change to the __xmknod dev pointer argument checks also broke.
I've pushed a fix for both, please check it and submit it upstream.

Thanks,

- Felix