linux-kernel - Re: [PATCH] fs: show locked and lock

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] [day] [month] [year] [list]

Message-ID: <87twx0x3co.fsf@x220.int.ebiederm.org>
Date:	Tue, 31 Mar 2015 16:29:43 -0500
From:	ebiederm@...ssion.com (Eric W. Biederman)
To:	Andrey Wagin <avagin@...il.com>
Cc:	Richard Weinberger <richard@....at>,
	linux-fsdevel <linux-fsdevel@...r.kernel.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Cyrill Gorcunov <gorcunov@...nvz.org>,
	Pavel Emelyanov <xemul@...allels.com>,
	Alexander Viro <viro@...iv.linux.org.uk>
Subject: Re: [PATCH] fs: show locked and lock_ro options in mountinfo

Andrey Wagin <avagin@...il.com> writes:

> 2015-03-28 1:47 GMT+03:00 Richard Weinberger <richard@....at>:
>> Hi!
>>
>> Am 27.03.2015 um 23:35 schrieb Andrey Wagin:
>>> 2015-03-28 0:42 GMT+03:00 Richard Weinberger <richard.weinberger@...il.com>:
>>>> On Fri, Mar 27, 2015 at 9:39 PM, Andrey Vagin <avagin@...nvz.org> wrote:
>>>>> I don't see any reasons to hide them. This information can help to
>>>>> understand errors.
>>>>
>>>> Because these flags are set/read only internally by the VFS. In contrast
>>>> to the other flags shown by mountinfo MNT_LOCKED is not a mount option.
>>>
>>> But this flag is set as a result of the specified user action, when he
>>> unshares userns and mntns. This flag affects visiable behaviour.
>>
>> It is a implicit result. Used by the VFS internally.
>> If you expose it it becomes ABI and changing the behavior will be
>> tricky or impossible.
>>
>>>>
>>>> Why does it help to debug errors?
>>>> How would a user know that mount() with MS_BIND returns EINVAL because
>>>> the mount source is MNT_LOCKED? This information is useless for her.
>>>
>>> If I see lock_ro, I can be sure that mount -o remount,bind,rw /XXX will fail.
>>> If I see locked, I  know that this mount can't be umounted or moved
>>> and can be bind-mounted only recursively.
>>>
>>> If a user see these flags, he can check that a mount namespace is
>>> configured correctly without security issues.
>>>
>>> Sorry but I don't understand why you think that this information is
>>> useless for users.
>>
>> You can only know if you know how the VFS works internally.
>> If know that EINVAL from mount(2) with MS_BIND can be caused by MNT_LOCKED
>> because I know the source. I bet you know the source too. But not Joey random
>> admin who looks into mountinfo to figure out why something does not work.
>>
>> If you expose MNT_LOCKED to userspace you'll have to update also the mount(2)
>> manpage with all glory details of that flag.
>>
>>>> If you argue like that you'd have to expose the whole VFS state to userland.
>>>
>>> I have not noticed other MNT_LOCK_* flags. I should think more about
>>> what information are a really required for dumping mount namespaces.
>>>
>>>>
>>>>> And this information is required for correct checkpoint/restore of mount
>>>>> namespaces.
>>>>
>>>> Why especially MNT_LOCKED and not all the other flags used by VFS?
>>>
>>> My goal is to dump enough information about a mount namespace to be
>>> able to restore it back later. I don't know how to do this without
>>> knowledge about locked mounts. I will think.
>>
>> How do you plan to restore a MNT_LOCKED mount?
>> IIRC we have currently no way to directly set MNT_LOCKED from userspace.
>
> It's the second question. The first question is how to check that we
> will be able to restore what we are dumping.
>
> If CRIU meets something what it doesn't know how to restore, it (must)
> refuses to dump this configuration.

As a practical matter if the underlying directory is empty, and will
remain empty MNT_LOCKED does not matter.


>>>> Say MNT_DOOMED?
>>>
>>> Mounts with MNT_DOOMED are never shown in mountinfo, are they?
>>
>> It was just an example. There are other flags too, did you double check
>> which ones you really need?
>>
>> To make the story short, my concern is that exposing such flags to userspace
>> has to be well thought. :-)
>> As long they are just internal we can change them as we like, as soon userspace
>> depends somehow on it the pain begins.
>
> I'm agree with you. I'm going to think more about this. Thank you for
> response.

A big question from me is do you have the ability to find the user
namespace of a mount namespace.  Without that these mount flags do not
matter.

I would think getting the user namespace of mount namespace and getting
the mount propgation tree correct would precede little things like
worrying if the mount propagation state is correct.

Eric

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/