Date:	Sat, 16 Jul 2016 17:41:56 +0800
From:	Shanker Wang <miao.wang@...a.tsinghua.edu.cn>
To:	Richard Weinberger <richard@....at>
Cc:	netdev@...r.kernel.org
Subject: Re: [PATCH] net:ppp: replace too strict capability restriction on opening /dev/ppp

Hi Richard,

Sorry for the late reply after dealing with my personal issues.

After reading the code, I find out that when /dev/ppp is opened, 
the instance is in “unattached” state and `file->private_data`
is null correspondingly. After creating a new instance, attaching
to a instance or attaching to a channel using ioctl 
(ppp_unattached_ioctl), the state is changed to “attached” and 
`file->private_data` becomes valid. My opinion is that only 
checking capability in `current->nsproxy->net_ns->user_ns` is 
enough. The reason is that to create a ppp interface or control a
 ppp interface (in short, to “modify” the network configuration),
we should always first create a new instance by opening `/dev/ppp` 
and attach it to a unit or so. 

As you mentioned, “an user can create a new user_ns followed by 
a new net_ns and has CAP_NET_ADMIN”. However, in that case, he can 
create new units by `ioctl PPPIOCNEWUNIT` but cannot attach to 
units existing in other net_ns by `ioctl PPPIOCATTACH`, because 
`ppp_find_unit` can only find unit in the current net_ns.

In spite of that, I have to admit that there is risk that lots 
of ppp devices are created filling all kernel memory. I think this
can be avoided by some other methods.

You have mentioned in another reply that “the use of nsproxy has
to be removed” but I don’t understand the reason. In comparison,
I found similar usage of nsproxy in `tun.c`. So could you please
give a more detailed explanation? Thanks!


> 在 2016年6月19日，18:40，Richard Weinberger <richard@....at> 写道：
> 
> Am 19.06.2016 um 12:36 schrieb Shanker Wang:
>> 
>>> 在 2016年6月19日，12:13，Richard Weinberger <richard@....at> 写道：
>>> 
>>> Am 19.06.2016 um 07:21 schrieb Shanker Wang:
>>>> This patch removes the check for CAP_NET_ADMIN in the initial namespace
>>>> when opening /dev/open. Instead, CAP_NET_ADMIN is checked in the user
>>>> namespace the net namespace was created so that /dev/ppp cat get opened
>>>> in a unprivileged container.
>>>> 
>>>> Cc: Hannes Frederic Sowa <hannes@...essinduktion.org>
>>>> Cc: Richard Weinberger <richard.weinberger@...il.com>
>>>> Cc: Guillaume Nault <g.nault@...halink.fr>
>>>> Cc: Miao Wang <shankerwangmiao@...il.com>
>>>> Signed-off-by: Miao Wang <miao.wang@...a.tsinghua.edu.cn>
>>>> ---
>>>> drivers/net/ppp/ppp_generic.c | 2 +-
>>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>> 
>>>> diff --git a/drivers/net/ppp/ppp_generic.c b/drivers/net/ppp/ppp_generic.c
>>>> index f572b31..4b3b2b5 100644
>>>> --- a/drivers/net/ppp/ppp_generic.c
>>>> +++ b/drivers/net/ppp/ppp_generic.c
>>>> @@ -380,7 +380,7 @@ static int ppp_open(struct inode *inode, struct file *file)
>>>> 	/*
>>>> 	 * This could (should?) be enforced by the permissions on /dev/ppp.
>>>> 	 */
>>>> -	if (!capable(CAP_NET_ADMIN))
>>>> +	if (!ns_capable(current->nsproxy->net_ns->user_ns, CAP_NET_ADMIN))
>>>> 		return -EPERM;
>>> 
>>> Shouldn't this be a ns_capable(net->user_ns, …?
>>> Otherwise an user can create a new user_ns followed by a new net_ns and has
>>> CAP_NET_ADMIN. We need to check whether he is allowed in the user_ns of the
>>> net_ns which belongs to the ppp net device which you want to open.
>> You are totally right. However, I wonder how can i get the “net” struct when 
>> opening /dev/ppp
> 
> I'm sure you can get it somehow via file->private_data.
> 
> Thanks,
> //richard

Powered by blists - more mailing lists