Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
Feedback-ID: i31e841b0:Fastmail
Message-ID: <cbc065c0-1dc3-974f-6e48-483baaf750a3@themaw.net>
Date:   Mon, 30 Oct 2023 13:45:33 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.13.0
Subject: Re: [PATCH v4 2/6] mounts: keep list of mounts in an rbtree
Content-Language: en-US
From:   Ian Kent <raven@themaw.net>
To:     Miklos Szeredi <mszeredi@redhat.com>
Cc:     linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-api@vger.kernel.org, linux-man@vger.kernel.org,
        linux-security-module@vger.kernel.org, Karel Zak <kzak@redhat.com>,
        David Howells <dhowells@redhat.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Al Viro <viro@zeniv.linux.org.uk>,
        Christian Brauner <christian@brauner.io>,
        Amir Goldstein <amir73il@gmail.com>,
        Matthew House <mattlloydhouse@gmail.com>,
        Florian Weimer <fweimer@redhat.com>,
        Arnd Bergmann <arnd@arndb.de>
References: <20231025140205.3586473-1-mszeredi@redhat.com>
 <20231025140205.3586473-3-mszeredi@redhat.com>
 <b69c1c17-35f9-351e-79a9-ef3ef5481974@themaw.net>
 <CAOssrKe76uZ5t714=Ta7GMLnZdS4QGm-fOfT9q5hNFe1fsDMVg@mail.gmail.com>
 <c938a7d9-aa9e-a3ad-a001-fb9022d21475@themaw.net>
 <dfb5f3d5-8db8-34af-d605-14112e8cc485@themaw.net>
In-Reply-To: <dfb5f3d5-8db8-34af-d605-14112e8cc485@themaw.net>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 30/10/23 13:37, Ian Kent wrote:
> On 28/10/23 09:36, Ian Kent wrote:
>> On 27/10/23 16:17, Miklos Szeredi wrote:
>>> On Fri, Oct 27, 2023 at 5:12 AM Ian Kent <raven@themaw.net> wrote:
>>>> On 25/10/23 22:02, Miklos Szeredi wrote:
>>>>> The mnt.mnt_list is still used to set up the mount tree and for
>>>>> propagation, but not after the mount has been added to a 
>>>>> namespace.  Hence
>>>>> mnt_list can live in union with rb_node.  Use MNT_ONRB mount flag to
>>>>> validate that the mount is on the correct list.
>>>> Is that accurate, propagation occurs at mount and also at umount.
>>> When propagating a mount, the new mount's mnt_list is used as a head
>>> for the new propagated mounts.  These are then moved to the rb tree by
>>> commit_tree().
>>>
>>> When umounting there's a "to umount" list called tmp_list in
>>> umount_tree(), this list is used to collect direct umounts and then
>>> propagated umounts.  The direct umounts are added in umount_tree(),
>>> the propagated ones umount_one().
>>>
>>> Note: umount_tree() can be called on a not yet finished mount, in that
>>> case the mounts are still on mnt_list, so umount_tree() needs to deal
>>> with both.
>>>
>>>> IDG how the change to umount_one() works, it looks like umount_list()
>>>>
>>>> uses mnt_list. It looks like propagate_umount() is also using 
>>>> mnt_list.
>>>>
>>>>
>>>> Am I missing something obvious?
>>> So when a mount is part of a namespace (either anonymous or not) it is
>>> on the rb tree, when not then it can temporarily be on mnt_list.
>>> MNT_ONRB flag is used to validate that the mount is on the list that
>>> we expect it to be on, but also to detect the case of the mount setup
>>> being aborted.
>>>
>>> We could handle the second case differently, since we should be able
>>> to tell when we are removing the mount from a namespace and when we
>>> are aborting a mount, but this was the least invasive way to do this.
>>
>> Thanks for the explanation, what you've said is essentially what I
>>
>> understood reading the series.
>>
>>
>> But I still haven't quite got this so I'll need to spend more time
>>
>> on this part of the patch series.
>>
>>
>> That's not a problem, ;).
>
> After cloning your git tree and looking in there I don't see what
>
> I was concerned about so I think I was confused by obscurity by
>
> diff rather than seeing a real problem, ;)
>
>
> Still that union worries me a little bit so I'll keep looking at
>
> the code for a while.

Is fs/namespace.c:iterate_mounts() a problem?

It's called from:

1) ./kernel/audit_tree.c:709: if (iterate_mounts(compare_root,
2) ./kernel/audit_tree.c:839:    err = iterate_mounts(tag_mount, tree, mnt);
3) ./kernel/audit_tree.c:917:        failed = iterate_mounts(tag_mount, 
tree, tagged);


 From functions 1) audit_trim_trees(), 2) audit_add_tree_rule() and

3) audit_tag_tree().


>
>
>>
>>
>> Ian
>>
>>>
>>> Thanks,
>>> Miklos
>>>
>>
>