Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Feedback-ID: i31e841b0:Fastmail
Message-ID: <996e11bf-5f22-3ab7-2951-92109649195d@themaw.net>
Date:   Fri, 28 Jul 2023 08:00:39 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.12.0
Subject: Re: [PATCH 1/2] kernfs: dont take i_lock on inode attr read
To:     Imran Khan <imran.f.khan@oracle.com>,
        Anders Roxell <anders.roxell@linaro.org>
Cc:     Arnd Bergmann <arnd@arndb.de>, Tejun Heo <tj@kernel.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Minchan Kim <minchan@kernel.org>,
        Eric Sandeen <sandeen@sandeen.net>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Rick Lindsley <ricklind@linux.vnet.ibm.com>,
        David Howells <dhowells@redhat.com>,
        Miklos Szeredi <miklos@szeredi.hu>,
        Carlos Maiolino <cmaiolino@redhat.com>,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        Kernel Mailing List <linux-kernel@vger.kernel.org>,
        elver@google.com
References: <166606025456.13363.3829702374064563472.stgit@donald.themaw.net>
 <166606036215.13363.1288735296954908554.stgit@donald.themaw.net>
 <Y2BMonmS0SdOn5yh@slm.duckdns.org> <20221221133428.GE69385@mutt>
 <7815c8da-7d5f-c2c5-9dfd-7a77ac37c7f7@themaw.net>
 <e25ee08c-7692-4042-9961-a499600f0a49@app.fastmail.com>
 <9e35cf66-79ef-1f13-dc6b-b013c73a9fc6@themaw.net>
 <db933d76-1432-f671-8712-d94de35277d8@themaw.net> <20230718190009.GC411@mutt>
 <76fcd1fe-b5f5-dd6b-c74d-30c2300f3963@themaw.net>
 <ce407424e98bf5f2b186df5d28dd5749a6cbfa45.camel@themaw.net>
 <15eddad0-e73b-2686-b5ba-eaacc57b8947@themaw.net>
 <3505769d-9e7a-e76d-3aa7-286d689345b6@oracle.com>
Content-Language: en-US
From:   Ian Kent <raven@themaw.net>
In-Reply-To: <3505769d-9e7a-e76d-3aa7-286d689345b6@oracle.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 27/7/23 12:30, Imran Khan wrote:
> Hello Ian,
> Sorry for late reply. I was about to reply this week.
>
> On 27/7/2023 10:38 am, Ian Kent wrote:
>> On 20/7/23 10:03, Ian Kent wrote:
>>> On Wed, 2023-07-19 at 12:23 +0800, Ian Kent wrote:
> [...]
>>> I do see a problem with recent changes.
>>>
>>> I'll send this off to Greg after I've done some testing (primarily just
>>> compile and function).
>>>
>>> Here's a patch which describes what I found.
>>>
>>> Comments are, of course, welcome, ;)
>> Anders I was hoping you would check if/what lockdep trace
>>
>> you get with this patch.
>>
>>
>> Imran, I was hoping you would comment on my change as it
>>
>> relates to the kernfs_iattr_rwsem changes.
>>
>>
>> Ian
>>
>>> kernfs: fix missing kernfs_iattr_rwsem locking
>>>
>>> From: Ian Kent <raven@themaw.net>
>>>
>>> When the kernfs_iattr_rwsem was introduced a case was missed.
>>>
>>> The update of the kernfs directory node child count was also protected
>>> by the kernfs_rwsem and needs to be included in the change so that the
>>> child count (and so the inode n_link attribute) does not change while
>>> holding the rwsem for read.
>>>
> kernfs direcytory node's child count changes in kernfs_(un)link_sibling and
> these are getting invoked while adding (kernfs_add_one),
> removing(__kernfs_remove) or moving (kernfs_rename_ns)a node. Each of these
> operations proceed under kernfs_rwsem and I see each invocation of
> kernfs_link/unlink_sibling during the above mentioned operations is happening
> under kernfs_rwsem.
> So the child count should still be protected by kernfs_rwsem and we should not
> need to acquire kernfs_iattr_rwsem in kernfs_link/unlink_sibling.

Yes, that's exactly what I intended (assuming you mean write lock in 
those cases)

when I did it so now I wonder what I saw that lead to my patch, I'll 
need to look

again ...


>
> Kindly let me know your thoughts. I would still like to see new lockdep traces
> with this change.

Indeed, I hope Anders can find time to get the trace.


Ian

>
> Thanks,
> Imran
>
>>> Fixes: 9caf696142 (kernfs: Introduce separate rwsem to protect inode
>>> attributes)
>>>
>>> Signed-off-by: Ian Kent <raven@themaw.net>
>>> Cc: Anders Roxell <anders.roxell@linaro.org>
>>> Cc: Imran Khan <imran.f.khan@oracle.com>
>>> Cc: Arnd Bergmann <arnd@arndb.de>
>>> Cc: Minchan Kim <minchan@kernel.org>
>>> Cc: Eric Sandeen <sandeen@sandeen.net>
>>> ---
>>>    fs/kernfs/dir.c |    4 ++++
>>>    1 file changed, 4 insertions(+)
>>>
>>> diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c
>>> index 45b6919903e6..6e84bb69602e 100644
>>> --- a/fs/kernfs/dir.c
>>> +++ b/fs/kernfs/dir.c
>>> @@ -383,9 +383,11 @@ static int kernfs_link_sibling(struct kernfs_node
>>> *kn)
>>>        rb_insert_color(&kn->rb, &kn->parent->dir.children);
>>>          /* successfully added, account subdir number */
>>> +    down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
>>>        if (kernfs_type(kn) == KERNFS_DIR)
>>>            kn->parent->dir.subdirs++;
>>>        kernfs_inc_rev(kn->parent);
>>> +    up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
>>>          return 0;
>>>    }
>>> @@ -408,9 +410,11 @@ static bool kernfs_unlink_sibling(struct
>>> kernfs_node *kn)
>>>        if (RB_EMPTY_NODE(&kn->rb))
>>>            return false;
>>>    +    down_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
>>>        if (kernfs_type(kn) == KERNFS_DIR)
>>>            kn->parent->dir.subdirs--;
>>>        kernfs_inc_rev(kn->parent);
>>> +    up_write(&kernfs_root(kn)->kernfs_iattr_rwsem);
>>>          rb_erase(&kn->rb, &kn->parent->dir.children);
>>>        RB_CLEAR_NODE(&kn->rb);
>>>