Re: [PATCH v2] hfs/hfsplus: validate uid and gid mount options

From: Viacheslav Dubeyko

Date: Wed Nov 18 2026 - 13:50:06 EST


On Wed, 2026-11-18 at 16:28 +0800, Lei Qiu wrote:
> The conversions to the new mount API declare uid and gid as
> fsparam_u32
> but consume result.uid and result.gid. This bypasses make_kuid() and
> make_kgid() as well as the validity checks in the old parsers. As a
> result, both filesystems accept the invalid ID 4294967295, and IDs
> supplied through fsconfig() in a non-initial user namespace are
> treated
> as raw kernel IDs instead of being translated through that namespace.
>
> Use fsparam_uid and fsparam_gid so the shared parser translates the
> IDs
> and rejects invalid or unmapped values before storing the typed
> result.
>
> Verify both modules with W=1 builds and a before/after QEMU test
> covering
> fsconfig(), legacy mounts, and a filesystem context configured from a
> user namespace mapping ID 0 to ID 1000. The fix rejects invalid IDs
> with
> EINVAL and preserves the mapped ownership and valid mount options.
>
> OpenAI Codex assisted with identifying the issue, implementing the
> fix
> and reproducer, and running the validation.
>
> Fixes: ffcd06b6d13b ("hfs: convert hfs to use the new mount api")
> Fixes: 432f7c78cb00 ("hfsplus: convert hfsplus to use the new mount
> api")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Lei Qiu <750592120@xxxxxx>
> ---
> Changes since v1: realigned the new fsparam_uid/gid entries with the
> existing column style in both option tables (no functional change).
> ---
>  build.log            | 35380
> +++++++++++++++++++++++++++++++++++++++++
>  fs/hfs/super.c       |     4 +-
>  fs/hfsplus/options.c |     4 +-
>  3 files changed, 35384 insertions(+), 4 deletions(-)
>  create mode 100644 build.log
>
> diff --git a/build.log b/build.log
> new file mode 100644
> index 000000000..abd0bdf55
> --- /dev/null
> +++ b/build.log
> @@ -0,0 +1,35380 @@
> +  UPD     arch/x86/include/generated/asm/cpufeaturemasks.h
> +  DESCEND objtool
> +  CC      arch/x86/kernel/asm-offsets.s

<skipped>

Why build.log has been included in the patch?

Thanks,
Slava.