Re: [PATCH] mm/memfd: Rename error labels for clarity
From: Andrew Morton
Date: Mon Jun 09 2025 - 20:27:03 EST
On Mon, 9 Jun 2025 11:18:58 +0800 Ye Liu <ye.liu@xxxxxxxxx> wrote:
> From: Ye Liu <liuye@xxxxxxxxxx>
>
> err_name --> err_fd (fd failure case)
> err_fd --> err_file (file failure case)
>
> ...
>
> --- a/mm/memfd.c
> +++ b/mm/memfd.c
> @@ -475,22 +475,22 @@ SYSCALL_DEFINE2(memfd_create,
> fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
> if (fd < 0) {
> error = fd;
> - goto err_name;
> + goto err_fd;
> }
>
> file = alloc_file(name, flags);
> if (IS_ERR(file)) {
> error = PTR_ERR(file);
> - goto err_fd;
> + goto err_file;
> }
>
> fd_install(fd, file);
> kfree(name);
> return fd;
>
> -err_fd:
> +err_file:
> put_unused_fd(fd);
> -err_name:
> +err_fd:
> kfree(name);
> return error;
> }
Not really, but I see what you mean.
"err_name" means "there was an error, so free the name".
"err_fd" means "there was a problem with the fd".
We tend to use the former convention. See
grep err_free mm/*.c
The memfd_create() code would be better if it used "err_free_name" and
"err_free_fd" to remove this ambiguity.
Someone who was feeling bored could go through
grep "goto err_" mm/*.c
and check that we use this convention uniformly ;)