[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5be1fab4-192d-8327-b541-bb92d53313cc@redhat.com>
Date: Thu, 17 Jul 2025 20:45:00 +0200 (CEST)
From: Mikulas Patocka <mpatocka@...hat.com>
To: Hou Tao <houtao1@...wei.com>
cc: Li Lingfeng <lilingfeng3@...wei.com>, dm-devel@...ts.linux.dev,
agk@...hat.com, snitzer@...nel.org, tusharsu@...ux.microsoft.com,
colin.i.king@...il.com, linux-kernel@...r.kernel.org, yi.zhang@...wei.com,
yangerkun@...wei.com, yukuai3@...wei.com, chengzhihao1@...wei.com,
lilingfeng@...weicloud.com
Subject: Re: [PATCH] dm: introduce spinlocks to prevent concurrent usage and
release of device_metadata and hash
On Thu, 17 Jul 2025, Hou Tao wrote:
> Hi,
>
> On 7/17/2025 10:07 PM, Li Lingfeng wrote:
> > There is a window between freeing md->ima.active_table.hash and setting
> > md->ima.active_table.hash to NULL in dm_ima_measure_on_device_resume().
> > If dm_ima_measure_on_device_remove() accesses md->ima.active_table.hash
> > concurrently during this window, it could lead to a double free or UAF,
> > as shown below:
>
> The UAF is a problem. However, I think the order of
> dm_ima_measure_on_device_remove() and dm_ima_measure_on_device_resume is
> bigger problem. Is it possible to order these two invocations properly
> by using suspend_lock ?
That's a good point. The patch just papers over the problem. It will fix
the UAF warning, but it won't fix the fact that
dm_ima_measure_on_device_resume can race with
dm_ima_measure_on_device_remove.
The real fix would be to make sure that dm_ima_measure_* calls are never
called concurrently on the same table - we should use existing device
mapper locks for that.
Mikukas
Powered by blists - more mailing lists