linux-kernel - Re: [PATCH 3/3] zonefs: convert to use kobject_is

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]

Message-ID: <003260d1-f1db-9d62-23fa-9acfba849782@fastmail.com>
Date:   Thu, 6 Apr 2023 20:23:57 +0900
From:   Damien Le Moal <dlemoal@...tmail.com>
To:     Greg KH <gregkh@...uxfoundation.org>
Cc:     Yangtao Li <frank.li@...o.com>, xiang@...nel.org, chao@...nel.org,
        huyue2@...lpad.com, jefflexu@...ux.alibaba.com,
        damien.lemoal@...nsource.wdc.com, naohiro.aota@....com,
        jth@...nel.org, rafael@...nel.org, linux-erofs@...ts.ozlabs.org,
        linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org
Subject: Re: [PATCH 3/3] zonefs: convert to use kobject_is_added()

On 4/6/23 20:18, Greg KH wrote:
> On Thu, Apr 06, 2023 at 07:58:38PM +0900, Damien Le Moal wrote:
>> On 4/6/23 19:26, Greg KH wrote:
>>> On Thu, Apr 06, 2023 at 07:13:38PM +0900, Damien Le Moal wrote:
>>>> On 4/6/23 19:05, Greg KH wrote:
>>>>> On Thu, Apr 06, 2023 at 05:30:56PM +0800, Yangtao Li wrote:
>>>>>> Use kobject_is_added() instead of local `s_sysfs_registered` variables.
>>>>>> BTW kill kobject_del() directly, because kobject_put() actually covers
>>>>>> kobject removal automatically.
>>>>>>
>>>>>> Signed-off-by: Yangtao Li <frank.li@...o.com>
>>>>>> ---
>>>>>>  fs/zonefs/sysfs.c  | 11 +++++------
>>>>>>  fs/zonefs/zonefs.h |  1 -
>>>>>>  2 files changed, 5 insertions(+), 7 deletions(-)
>>>>>>
>>>>>> diff --git a/fs/zonefs/sysfs.c b/fs/zonefs/sysfs.c
>>>>>> index 8ccb65c2b419..f0783bf7a25c 100644
>>>>>> --- a/fs/zonefs/sysfs.c
>>>>>> +++ b/fs/zonefs/sysfs.c
>>>>>> @@ -101,8 +101,6 @@ int zonefs_sysfs_register(struct super_block *sb)
>>>>>>  		return ret;
>>>>>>  	}
>>>>>>  
>>>>>> -	sbi->s_sysfs_registered = true;
>>>>>
>>>>> You know this, why do you need to have a variable tell you this or not?
>>>>
>>>> If kobject_init_and_add() fails, zonefs_sysfs_register() returns an error and
>>>> fill_super will also return that error. vfs will then call kill_super, which
>>>> calls zonefs_sysfs_unregister(). For that case, we need to know that we actually
>>>> added the kobj.
>>>
>>> Ok, but then why not just 0 out the kobject pointer here instead?  That
>>> way you will always know if it's a valid pointer or not and you don't
>>> have to rely on some other variable?  Use the one that you have already :)
>>
>> but sbi->s_kobj is the kobject itself, not a pointer.
> 
> Then it should not be there if the kobject is not valid as it should
> have been freed when the kobject_init_and_add() call failed, right?

What do you mean freed ? the kboject itself is a field of zonefs sbi. So the
kobject gets freed together with sbi.

>> I can still zero it out in
>> case of error to avoid using the added s_sysfs_registered bool. I would need to
>> check a field of s_kobj though, which is not super clean and makes the code
>> dependent on kobject internals. Not super nice in my opinion, unless I am
>> missing something.
> 
> See above, if a kobject fails to be registered, just remove the whole
> object as it's obviously "dead" now and you can not trust it.

Well yes, that is what s_sysfs_registered indicates, that the kobject is not
valid. I do not understand what you mean with "just remove the whole object".

>>> And you really don't even need to check anything, just pass in NULL to
>>> kobject_del() and friends, it should handle it.>
>>>>>> -
>>>>>>  	return 0;
>>>>>>  }
>>>>>>  
>>>>>> @@ -110,12 +108,13 @@ void zonefs_sysfs_unregister(struct super_block *sb)
>>>>>>  {
>>>>>>  	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
>>>>>>  
>>>>>> -	if (!sbi || !sbi->s_sysfs_registered)
>>>>>
>>>>> How can either of these ever be true?  Note, sbi should be passed here
>>>>> to this function, not the super block as that is now unregistered from
>>>>> the system.  Looks like no one has really tested this codepath that much
>>>>> :(
>>>>>
>>>>>> +	if (!sbi)
>>>>>>  		return;
>>>>>
>>>>> this can not ever be true, right?
>>>>
>>>> Yes it can, if someone attempt to mount a non zoned device. In that case,
>>>> fill_super returns early without setting sb->s_fs_info but vfs still calls
>>>> kill_super.
>>>
>>> But you already had a sbi pointer in the place that this was called, so
>>> you "know" if you need to even call into here or not.  You are having to
>>> look up the same pointer multiple times in this call chain, there's no
>>> need for that.
>>
>> I am not following here. Either we check that we have sbi here in
>> zonefs_sysfs_unregister(), or we conditionally call this function in
>> zonefs_kill_super() with a "if (sbi)". Either way, we need to check since sbi
>> can be NULL.
> 
> In zonefs_kill_super() you have get the spi at the top of the function,
> so use that, don't make zonefs_sysfs_unregister() have to compute it
> again.

That I can do, yes.

> 
> But again, if the kobject fails to be registered, you have to treat the
> memory contained there as not valid and get rid of it as soon as
> possible.

If the kobject add failed, we never touch it thanks to s_sysfs_registered. I
still do not see the issue here.

> 
> thanks,
> 
> greg k-h