netdev - Re: [PATCH net] rtnetlink: ndo_dflt_fdb

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]

Date:	Fri, 31 May 2013 16:42:25 -0700 (PDT)
From:	David Miller <davem@...emloft.net>
To:	vyasevic@...hat.com
Cc:	alanr@...x.sh, netdev@...r.kernel.org, davem@...emloft.net,
	kuznet@....inr.ac.ru
Subject: Re: [PATCH net] rtnetlink: ndo_dflt_fdb_del() never works

From: Vlad Yasevich <vyasevic@...hat.com>
Date: Fri, 31 May 2013 13:11:47 -0400

> On 05/31/2013 12:43 PM, Alan Robertson wrote:
>> On 05/31/2013 02:46 AM, David Miller wrote:
>>> From: Alan Robertson <alanr@...x.sh>
>>> Date: Thu, 30 May 2013 16:01:55 -0600
>>>
>>>> ndo_dflt_fdb_del is checking for a condition which is opposite that
>>>> which ndo_dflt_fdb_add enforces.  ndo_dflt_fdb_add declares an error
>>>> if (ndm->ndm_state && !(ndm->ndm_state) & NUD_PERMANENT)) - that is,
>>>> if the
>>>> entry is static.  This is consistent with the failure error message.
>>>>
>>>> On the other hand, ndo_dflt_del() declares an error
>>>> if (ndm_state & NUD_PERMANENT) - which is inconsistent with the add
>>>> precondition, and inconsistent with its failure message text.
>>>> As it is now, you can't delete any entry which add allows to be added
>>>> -
>>>> so entry deletion always fails.
>>>>
>>>> Signed-off-by: Alan Robertson <alanr@...x.sh>
>>> What about the ->ndm_state part of the add() test?  Why not include
>>> that in the del() check?
>> I had three different thoughts about this:
>>    1) Replicated the add check in the delete
>>    2) Do what I did - make it where you can only delete those really
>> marked as static
>>    3) Eliminate all delete checks
>>
>> The problem is -- I'm not sure why the ndm->ndm_state check is in
>> there
>> -- I don't know how to make that condition occur.  It has the feel of a
>> check to be made before things are fully initialized - or maybe even
>> just leftover cruft.
>>
> 
> The test is there to support simultaneous master and self
> operations. The operation on a master may not always require a
> NUD_PERMANENT state (ex: bridge) and we don't want to perform self
> operations in that instance.

I still don't understand the ndm_state check.  Please use different
words to explain it so that even an idiot like me can understand.

Once we define what the check should exactly be I propose:

1) Keeping the check only in add()

2) Removing the state checks completely in del()

3) Validating at netdevice registry time or elsewhere that these
   default fdb ops are always used together.  That wraps up everything
   to ensure that only doing the check in add() is provably correct.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html