lkml.org 
[lkml]   [2015]   [Mar]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v9 2/2] uio: Fix uio driver to refcount device


On 23/03/15 20:41, Greg Kroah-Hartman wrote:
> On Fri, Mar 20, 2015 at 02:54:44PM +0000, Brian Russell wrote:
>> Protect uio driver from its owner being unplugged while there are open fds.
>> Embed struct device in struct uio_device, use refcounting on device, free
>> uio_device on release.
>> info struct passed in uio_register_device can be freed on unregister, so null
>> out the field in uio_unregister_device and check accesses.
>
> That's really not protecting anything except heavy-handed problems...
>
> Look at the code:
>
>> @@ -493,7 +499,7 @@ static unsigned int uio_poll(struct file *filep, poll_table *wait)
>> struct uio_listener *listener = filep->private_data;
>> struct uio_device *idev = listener->dev;
>>
>> - if (!idev->info->irq)
>> + if (!idev->info || !idev->info->irq)
>> return -EIO;
>>
>
> Great, you checked the irq value, but what if it changes the very next
> line:
>
>> poll_wait(filep, &idev->wait, wait);
>
> Or any other line within this function? Or any other function that you
> try to check the value for in the beginning...
>
> This really isn't protecting anything "properly", sorry. Either we
> don't care about it (hint, I don't think we really do), or we need to
> properly lock things and check, and protect, things that way.
>

The checks for irq value are already there. I added the checks for the
idev->info ptr and deliberately nulled it in uio_unregister_device as
the caller module may free uio_info after unregistering (dpdk's igb_uio
does anyway) and then release will be called later when fds are closed.

So I think I definitely need the check in uio_release. I didn't think
it hurt to return early from poll/read/write if we know the device
has been unregistered?

Thanks,

Brian

> Please do the first one, as the reference count should be all that we
> need to care about here.
>
> Sorry I missed this on the previous review, just realized it now this
> time around.
>
> thanks,
>
> greg k-h
>


\
 
 \ /
  Last update: 2015-03-24 14:21    [W:0.063 / U:0.428 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site