linux-kernel - Re: [PATCH] csiostor: Fix backwards locking in the function __csio_unreg

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <57051913.4050304@gmail.com>
Date:	Wed, 6 Apr 2016 10:11:31 -0400
From:	Bastien Philbert <bastienphilbert@...il.com>
To:	James Bottomley <jejb@...ux.vnet.ibm.com>,
	Julian Calaby <julian.calaby@...il.com>
Cc:	"Martin K. Petersen" <martin.petersen@...cle.com>,
	linux-scsi <linux-scsi@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] csiostor: Fix backwards locking in the function
 __csio_unreg_rnode



On 2016-04-06 09:38 AM, James Bottomley wrote:
> On Wed, 2016-04-06 at 09:21 -0400, Bastien Philbert wrote:
>>
>> On 2016-04-06 03:48 AM, Julian Calaby wrote:
>>> Hi Bastien,
>>>
>>> On Wed, Apr 6, 2016 at 7:19 AM, Bastien Philbert
>>> <bastienphilbert@...il.com> wrote:
>>>> This fixes backwards locking in the function __csio_unreg_rnode
>>>> to
>>>> properly lock before the call to the function csio_unreg_rnode
>>>> and
>>>> not unlock with spin_unlock_irq as this would not allow the
>>>> proper
>>>> protection for concurrent access on the shared csio_hw structure
>>>> pointer hw. In addition switch the locking after the critical
>>>> region
>>>> function call to properly unlock instead with spin_unlock_irq on
>>>>
>>>> Signed-off-by: Bastien Philbert <bastienphilbert@...il.com>
>>>> ---
>>>>  drivers/scsi/csiostor/csio_rnode.c | 4 ++--
>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/scsi/csiostor/csio_rnode.c
>>>> b/drivers/scsi/csiostor/csio_rnode.c
>>>> index e9c3b04..029a09e 100644
>>>> --- a/drivers/scsi/csiostor/csio_rnode.c
>>>> +++ b/drivers/scsi/csiostor/csio_rnode.c
>>>> @@ -580,9 +580,9 @@ __csio_unreg_rnode(struct csio_rnode *rn)
>>>>                 ln->last_scan_ntgts--;
>>>>         }
>>>>
>>>> -       spin_unlock_irq(&hw->lock);
>>>> -       csio_unreg_rnode(rn);
>>>>         spin_lock_irq(&hw->lock);
>>>> +       csio_unreg_rnode(rn);
>>>> +       spin_unlock_irq(&hw->lock);
>>>
>>> Are you _certain_ this is correct? This construct usually appears
>>> when
>>> a function has a particular lock held, then needs to unlock it to
>>> call
>>> some other function. Are you _certain_ that this isn't the case?
>>>
>>> Thanks,
>>>
>> Yes I am pretty certain this is correct. I checked the paths that
>> called this function
>> and it was weired that none of them gradded the spinlock before hand.
> 
> That's not good enough.  If your theory is correct, lockdep should be
> dropping an already unlocked assertion in this codepath ... do you see
> this?
> 
> James
> 
> 
Yes I do. For now just drop the patch but I am still concerned that we are double unlocking here.
Bastien
> 
>