Re: [PATCH] fs: WARN when f_count resurrection is attempted

From: Peter Zijlstra
Date: Mon May 06 2024 - 06:39:09 EST


On Fri, May 03, 2024 at 01:16:25PM -0700, Kees Cook wrote:
> It should never happen that get_file() is called on a file with
> f_count equal to zero. If this happens, a use-after-free condition
> has happened[1], and we need to attempt a best-effort reporting of
> the situation to help find the root cause more easily. Additionally,
> this serves as a data corruption indicator that system owners using
> warn_limit or panic_on_warn would like to have detected.
>
> Link: https://lore.kernel.org/lkml/7c41cf3c-2a71-4dbb-8f34-0337890906fc@xxxxxxxxx/ [1]
> Suggested-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> ---
> Cc: Christian Brauner <brauner@xxxxxxxxxx>
> Cc: Alexander Viro <viro@xxxxxxxxxxxxxxxxxx>
> Cc: Jens Axboe <axboe@xxxxxxxxx>
> Cc: Jann Horn <jannh@xxxxxxxxxx>
> Cc: Jan Kara <jack@xxxxxxx>
> Cc: linux-fsdevel@xxxxxxxxxxxxxxx
> ---
> include/linux/fs.h | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 00fc429b0af0..fa9ea5390f33 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -1038,7 +1038,8 @@ struct file_handle {
>
> static inline struct file *get_file(struct file *f)
> {
> - atomic_long_inc(&f->f_count);
> + long prior = atomic_long_fetch_inc_relaxed(&f->f_count);
> + WARN_ONCE(!prior, "struct file::f_count incremented from zero; use-after-free condition present!\n");

This reminds me, I should some day try and fix the horrible code-gen for
WARN() :/ WARN_ON_*() and friends turn into a single trap instruction,
but the WARN() and friends thing turns into a horrible piece of crap for
the printk().