Re: [PATCH 1/3] dcache: add a new enum type for 'dentry_d_lock_class'

["yukuai (C)" <yukuai3@huawei.com>]

On 2019/11/15 15:20, Greg KH wrote:
> On Fri, Nov 15, 2019 at 04:12:43AM +0000, Al Viro wrote:
>> On Fri, Nov 15, 2019 at 11:27:59AM +0800, Greg KH wrote:
>>> On Fri, Nov 15, 2019 at 11:27:50AM +0800, yu kuai wrote:
>>>> 'dentry_d_lock_class' can be used for spin_lock_nested in case lockdep
>>>> confused about two different dentry take the 'd_lock'.
>>>>
>>>> However, a single 'DENTRY_D_LOCK_NESTED' may not be enough if more than
>>>> two dentry are involed. So, and in 'DENTRY_D_LOCK_NESTED_2'
>>>>
>>>> Signed-off-by: yu kuai <yukuai3@huawei.com>
>>>> ---
>>>>   include/linux/dcache.h | 3 ++-
>>>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/include/linux/dcache.h b/include/linux/dcache.h
>>>> index 10090f1..8eb84ef 100644
>>>> --- a/include/linux/dcache.h
>>>> +++ b/include/linux/dcache.h
>>>> @@ -129,7 +129,8 @@ struct dentry {
>>>>   enum dentry_d_lock_class
>>>>   {
>>>>   	DENTRY_D_LOCK_NORMAL, /* implicitly used by plain spin_lock() APIs. */
>>>> -	DENTRY_D_LOCK_NESTED
>>>> +	DENTRY_D_LOCK_NESTED,
>>>> +	DENTRY_D_LOCK_NESTED_2
>>>
>>> You should document this, as "_2" does not make much sense to anyone
>>> only looking at the code :(
>>>
>>> Or rename it better.
>>
>> FWIW, I'm not sure it's a good solution.  What are the rules for callers
>> of that thing, anyway?  If it can be called when somebody is creating
>> more files in that subtree, we almost certainly will have massive
>> problems with the lifetimes of underlying objects...
>>
>> Could somebody familiar with debugfs explain how is that thing actually
>> used and what is required from/promised to its callers?  I can try and
>> grep through the tree and guess what the rules are, but I've way too
>> much on my platter right now and I don't want to get sidetracked into yet
>> another tree-wide search and analysis session ;-/
> 
> Yu wants to use simple_empty() in debugfs_remove_recursive() instead of
> manually checking:
> 	if (!list_empty(&child->d_subdirs)) {
> 
> See patch 3 of this series for that change and why they feel it is
> needed:
> 	https://lore.kernel.org/lkml/1573788472-87426-4-git-send-email-yukuai3@huawei.com/
> 
> As to if patch 3 really is needed, I'll leave that up to Yu given that I
> thought we had resolved these types of issues already a year or so ago.
> 
> thanks,
> 
> greg k-h
> 
> .
> 
The main purpose of this patchset is to fix the infinite loop in
debugfs_remove_recursive. Steven point out that simple replace
list_empty with simple_empty will cause a splat with lockdep enabled.
We try to fix it with the first two patch, do you think it's appropriate?

Thanks,
Yu Kuai