Re: [PATCH v5 5/7] revocable: Add fops replacement

[Jason Gunthorpe <jgg@nvidia.com>]

On Fri, Nov 07, 2025 at 04:11:40AM +0000, Tzung-Bi Shih wrote:
> Realized the approach doesn't work for the issue I'm looking into.
> 
> - All misc devices share the same cdev[1].  If misc_deregister() calls
>   cdev_sync_revoke(), the misc stop working due to one of the miscdevice
>   deregistered.

> [1] https://elixir.bootlin.com/linux/v6.17/source/drivers/char/misc.c#L299

That's not a "cdev" in this context, but yes, misc doesn't use
struct cdev at all.. Instead you have a struct miscdevice which has a
similar lifecycle as cdev. Indeed you can't use what I showed above at
the cdev layer exactly as is, but there is not a fundamental issue
here.

> - The context (struct cdev_sync_data) should be the same lifecycle with
>   the opening file (e.g. struct file).  Otherwise, when accessing the
>   context in the fops wrappers, it results an UAF.  For example, the
>   sturct cdev is likely freed after cdev_sync_revoke().

Yes, it should be tied to the memory lifecycle of the struct device
under the cdev which would then by tied to the file lifecycle. It is
not hard.

Jason