Re: [PATCH v2] fuse: propagate async RELEASE queue errno in fuse_file_put()

[Joanne Koong]

On Mon, May 11, 2026 at 3:14 AM Li Wang <liwang@xxxxxxxxxx> wrote:
>
> fuse_file_put() queues asynchronous FUSE_RELEASE via fuse_simple_background().
> When queueing fails, the failure branch always passed -ENOTCONN into
> fuse_release_end(). fuse_simple_background() can however fail for multiple
> reasons with different return values, such as -ENOMEM or -EINTR. Pass the
> actual negative errno into fuse_release_end(). A similar pattern occurs
> when fuse_send_readpages() calls fuse_readpages_end(). In that case,
> if sending the request fails, fuse_readpages_end() is invoked
> directly using the error value returned by the sending function as an argument.
> Furthermore, the same applies to fuse_async_req_send() calling
> fuse_aio_complete_req(). This patch aligns with those cases.
>
> The parameter "error" is currently unused in fuse_release_end().
> While removing it seems like the better choice, however, while fuse_file_put()
> calls fuse_release_end() directly if sending the async request fails, it also
> registers fuse_release_end() as the success callback. Consequently,
> fuse_release_end() must strictly adhere to the callback signature; removing
> any parameter would trigger a compilation error. Add an explicit
> (void)error in fuse_release_end() so builds that enable
> -Wunused-parameter (for example make W=3) stay clean.
>
> Signed-off-by: Li Wang <liwang@xxxxxxxxxx>
> ---
> Changes since v1:
> - Amend commit message to clarify the reasoning behind the code changes.
>
>  fs/fuse/file.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/fs/fuse/file.c b/fs/fuse/file.c
> index 3bdab8d03373..8b7badb7721a 100644
> --- a/fs/fuse/file.c
> +++ b/fs/fuse/file.c
> @@ -96,6 +96,8 @@ static void fuse_release_end(struct fuse_args *args, int error)
>  {
>         struct fuse_release_args *ra = container_of(args, typeof(*ra), args);
>
> +       (void)error;
> +
>         iput(ra->inode);
>         kfree(ra);
>  }
> @@ -117,6 +119,8 @@ static void fuse_file_put(struct fuse_file *ff, bool sync)
>                         fuse_simple_request(ff->fm, args);
>                         fuse_release_end(args, 0);
>                 } else {
> +                       int err;
> +
>                         /*
>                          * DAX inodes may need to issue a number of synchronous
>                          * request for clearing the mappings.
> @@ -124,9 +128,10 @@ static void fuse_file_put(struct fuse_file *ff, bool sync)
>                         if (ra && ra->inode && FUSE_IS_DAX(ra->inode))
>                                 args->may_block = true;
>                         args->end = fuse_release_end;
> -                       if (fuse_simple_background(ff->fm, args,
> -                                                  GFP_KERNEL | __GFP_NOFAIL))
> -                               fuse_release_end(args, -ENOTCONN);
> +                       err = fuse_simple_background(ff->fm, args,
> +                                                    GFP_KERNEL | __GFP_NOFAIL);
> +                       if (err)
> +                               fuse_release_end(args, err);
>                 }

Hi Li,

I'm not sure I see how this change is needed since fuse_release_end()
does nothing with the error arg today. afaict the (void)error pattern
isn't typical in-kernel since W=3 isn't a standard build target. I
think it might be best to just keep the existing code as is to reduce
churn, and then if fuse_release_end() later in the future gets updated
to use the error arg, this would make more sense as part of that
change.

Thanks,
Joanne

>                 kfree(ff);
>         }
> --
> 2.34.1
>