Received-SPF: pass (google.com: domain of linux-kernel+bounces-91021-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Mon, 4 Mar 2024 12:47:21 -0500
From: Benjamin LaHaise <ben@communityfibre.ca>
To: Bart Van Assche <bvanassche@acm.org>
Cc: Edward Adam Davis <eadavis@qq.com>,
	syzbot+b91eb2ed18f599dd3c31@syzkaller.appspotmail.com,
	brauner@kernel.org, jack@suse.cz, linux-aio@kvack.org,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	syzkaller-bugs@googlegroups.com, viro@zeniv.linux.org.uk
Subject: Re: [PATCH] fs/aio: fix uaf in sys_io_cancel
Message-ID: <20240304174721.GQ20455@kvack.org>
References: <0000000000006945730612bc9173@google.com> <tencent_DC4C9767C786D2D2FDC64F099FAEFDEEC106@qq.com> <14f85d0c-8303-4710-b8b1-248ce27a6e1f@acm.org> <20240304170343.GO20455@kvack.org> <73949a4d-6087-4d8c-bae0-cda60e733442@acm.org> <20240304173120.GP20455@kvack.org> <5ee4df86-458f-4544-85db-81dc82c2df4c@acm.org>
Precedence: bulk
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <5ee4df86-458f-4544-85db-81dc82c2df4c@acm.org>
User-Agent: Mutt/1.4.2.2i

On Mon, Mar 04, 2024 at 09:40:35AM -0800, Bart Van Assche wrote:
> On 3/4/24 09:31, Benjamin LaHaise wrote:
> >A revert is justified when a series of patches is buggy and had
> >insufficient review prior to merging.
> 
> That's not how Linux kernel development works. If a bug can get fixed
> easily, a fix is preferred instead of reverting + reapplying a patch.

Your original "fix" is not right, and it wasn't properly tested.  Commit
54cbc058d86beca3515c994039b5c0f0a34f53dd needs to be reverted.

> >Using the "a kernel warning hit" approach for work on cancellation is
> >very much a sign that the patches were half baked.
> Is there perhaps a misunderstanding? My patches fix a kernel warning and
> did not introduce any new WARN*() statements.

The change that introduced that callback by you was incorrect and should
be reverted.

> >Why are you touching the kiocb after ownership has already been
> >passed on to another entity?
> Touching the kiocb after ownership has been passed is the result of an
> oversight. Whether or not kiocb->ki_cancel() transfers ownership depends
> on the I/O type. The use-after-free was not introduced on purpose.

Your fix is still incorrect.  You're still touching memory that you don't
own.  The event should be generated via the ->ki_cancel method, not in the
io_cancel() syscall.

		-ben

> Bart.
> 
> 

-- 
"Thought is the essence of where you are now."