On Fri, 2026-05-15 at 11:06 +0200, Tatrai, Peter via lists.openembedded.org 
wrote:
> From: Peter Tatrai <[email protected]>
> 
> When SDKMACHINE is set to "i686", nativesdk-ninja is compiled as a
> 32-bit binary. Without large file support (LFS), ninja's internal stat()
> call uses a 32-bit struct stat, which results in EOVERFLOW on filesystems
> that report inode numbers larger than 2^32 (e.g. container overlay
> filesystems).
> 
> Add -D_FILE_OFFSET_BITS=64 to CXXFLAGS for nativesdk builds to enable
> LFS, ensuring stat() uses 64-bit types regardless of host word size.
> 
> Signed-off-by: Peter Tatrai <[email protected]>
> ---
>  meta/recipes-devtools/ninja/ninja_1.13.2.bb | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/meta/recipes-devtools/ninja/ninja_1.13.2.bb 
> b/meta/recipes-devtools/ninja/ninja_1.13.2.bb
> index e750c78c46..4fd048efa6 100644
> --- a/meta/recipes-devtools/ninja/ninja_1.13.2.bb
> +++ b/meta/recipes-devtools/ninja/ninja_1.13.2.bb
> @@ -30,6 +30,11 @@ do_install() {
>       install -D -m 0755  ${S}/ninja ${D}${bindir}/ninja
>  }
>  
> +# Enable large file support to avoid EOVERFLOW when stat() is called on files
> +# with large inode numbers (e.g. on overlay filesystems). This is critical 
> when
> +# building nativesdk for a 32-bit SDKMACHINE (e.g. i686).
> +CXXFLAGS:append:class-nativesdk = " -D_FILE_OFFSET_BITS=64"
> +
>  BBCLASSEXTEND = "native nativesdk"
>  
>  CVE_STATUS[CVE-2021-4336] = "cpe-incorrect: This is a different Ninja"

I think we should set both -D_TIME_BITS=64 and -D_FILE_OFFSET_BITS=64
for i686 unconditionally in a similar way that time64.inc is doing for
the target else we're just going to run into this elsewhere...

Cheers,

Richard

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#237261): 
https://lists.openembedded.org/g/openembedded-core/message/237261
Mute This Topic: https://lists.openembedded.org/mt/119326878/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to