[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Yg0421B10PPwunI+@kroah.com>
Date: Wed, 16 Feb 2022 18:48:11 +0100
From: Greg KH <gregkh@...uxfoundation.org>
To: JeffleXu <jefflexu@...ux.alibaba.com>
Cc: dhowells@...hat.com, linux-cachefs@...hat.com, xiang@...nel.org,
chao@...nel.org, linux-erofs@...ts.ozlabs.org,
torvalds@...ux-foundation.org, willy@...radead.org,
linux-fsdevel@...r.kernel.org, joseph.qi@...ux.alibaba.com,
bo.liu@...ux.alibaba.com, tao.peng@...ux.alibaba.com,
gerry@...ux.alibaba.com, eguan@...ux.alibaba.com,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v4 05/23] cachefiles: introduce new devnode for on-demand
read mode
On Wed, Feb 16, 2022 at 08:49:35PM +0800, JeffleXu wrote:
> >> +struct cachefiles_req_in {
> >> + uint64_t id;
> >> + uint64_t off;
> >> + uint64_t len;
> >
> > For structures that cross the user/kernel boundry, you have to use the
> > correct types. For this it would be __u64.
>
> OK I will change to __xx style in the next version.
>
> By the way, I can't understand the disadvantage of uintxx_t style.
The "uint*" types are not valid kernel types. They are userspace types
and do not transfer properly in all arches and situations when crossing
the user/kernel boundry. They are also in a different C "namespace", so
should not even be used in kernel code, although a lot of people do
because they are used to writing userspace C code :(
thanks,
greg k-h
Powered by blists - more mailing lists