[v2,6/7] quota: simplify drop_dquot_ref()

  Now when dqput() drops the last reference count, it will call
synchronize_srcu(&dquot_srcu) in quota_release_workfn() to ensure that
no other user will use the dquot after the last reference count is dropped,
so we don't need to call synchronize_srcu(&dquot_srcu) in drop_dquot_ref()
and remove the corresponding logic directly to simplify the code.

Signed-off-by: Baokun Li <libaokun1@huawei.com>
---
 fs/quota/dquot.c | 33 ++++++---------------------------
 1 file changed, 6 insertions(+), 27 deletions(-)
  

On Wed 28-06-23 21:21:54, Baokun Li wrote:
> Now when dqput() drops the last reference count, it will call
> synchronize_srcu(&dquot_srcu) in quota_release_workfn() to ensure that
> no other user will use the dquot after the last reference count is dropped,
> so we don't need to call synchronize_srcu(&dquot_srcu) in drop_dquot_ref()
> and remove the corresponding logic directly to simplify the code.

Nice simplification!  It is also important that dqput() now cannot sleep
which was another reason for the logic with tofree_head in
remove_inode_dquot_ref(). Probably this is good to mention in the
changelog.

> Signed-off-by: Baokun Li <libaokun1@huawei.com>
> ---
>  fs/quota/dquot.c | 33 ++++++---------------------------
>  1 file changed, 6 insertions(+), 27 deletions(-)
> 
> diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
> index e8e702ac64e5..df028fb2ce72 100644
> --- a/fs/quota/dquot.c
> +++ b/fs/quota/dquot.c
> @@ -1090,8 +1090,7 @@ static int add_dquot_ref(struct super_block *sb, int type)
>   * Remove references to dquots from inode and add dquot to list for freeing
>   * if we have the last reference to dquot
>   */
> -static void remove_inode_dquot_ref(struct inode *inode, int type,
> -				   struct list_head *tofree_head)
> +static void remove_inode_dquot_ref(struct inode *inode, int type)
>  {
>  	struct dquot **dquots = i_dquot(inode);
>  	struct dquot *dquot = dquots[type];
> @@ -1100,21 +1099,7 @@ static void remove_inode_dquot_ref(struct inode *inode, int type,
>  		return;
>  
>  	dquots[type] = NULL;
> -	if (list_empty(&dquot->dq_free)) {
> -		/*
> -		 * The inode still has reference to dquot so it can't be in the
> -		 * free list
> -		 */
> -		spin_lock(&dq_list_lock);
> -		list_add(&dquot->dq_free, tofree_head);
> -		spin_unlock(&dq_list_lock);
> -	} else {
> -		/*
> -		 * Dquot is already in a list to put so we won't drop the last
> -		 * reference here.
> -		 */
> -		dqput(dquot);
> -	}
> +	dqput(dquot);
>  }

I think you can also just drop remove_inode_dquot_ref() as it is trivial
now and inline it at its only callsite...

								Honza
  

On 2023/6/29 19:08, Jan Kara wrote:
> On Wed 28-06-23 21:21:54, Baokun Li wrote:
>> Now when dqput() drops the last reference count, it will call
>> synchronize_srcu(&dquot_srcu) in quota_release_workfn() to ensure that
>> no other user will use the dquot after the last reference count is dropped,
>> so we don't need to call synchronize_srcu(&dquot_srcu) in drop_dquot_ref()
>> and remove the corresponding logic directly to simplify the code.
> Nice simplification!  It is also important that dqput() now cannot sleep
> which was another reason for the logic with tofree_head in
> remove_inode_dquot_ref().

I don't understand this sentence very well, so I would appreciate it

if you could explain it in detail. 🤔

> Probably this is good to mention in the
> changelog.
>
>> Signed-off-by: Baokun Li <libaokun1@huawei.com>
>> ---
>>   fs/quota/dquot.c | 33 ++++++---------------------------
>>   1 file changed, 6 insertions(+), 27 deletions(-)
>>
>> diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
>> index e8e702ac64e5..df028fb2ce72 100644
>> --- a/fs/quota/dquot.c
>> +++ b/fs/quota/dquot.c
>> @@ -1090,8 +1090,7 @@ static int add_dquot_ref(struct super_block *sb, int type)
>>    * Remove references to dquots from inode and add dquot to list for freeing
>>    * if we have the last reference to dquot
>>    */
>> -static void remove_inode_dquot_ref(struct inode *inode, int type,
>> -				   struct list_head *tofree_head)
>> +static void remove_inode_dquot_ref(struct inode *inode, int type)
>>   {
>>   	struct dquot **dquots = i_dquot(inode);
>>   	struct dquot *dquot = dquots[type];
>> @@ -1100,21 +1099,7 @@ static void remove_inode_dquot_ref(struct inode *inode, int type,
>>   		return;
>>   
>>   	dquots[type] = NULL;
>> -	if (list_empty(&dquot->dq_free)) {
>> -		/*
>> -		 * The inode still has reference to dquot so it can't be in the
>> -		 * free list
>> -		 */
>> -		spin_lock(&dq_list_lock);
>> -		list_add(&dquot->dq_free, tofree_head);
>> -		spin_unlock(&dq_list_lock);
>> -	} else {
>> -		/*
>> -		 * Dquot is already in a list to put so we won't drop the last
>> -		 * reference here.
>> -		 */
>> -		dqput(dquot);
>> -	}
>> +	dqput(dquot);
>>   }
> I think you can also just drop remove_inode_dquot_ref() as it is trivial
> now and inline it at its only callsite...
>
> 								Honza
Sure, I'll just remove it in the next version and inline the only 
remaining code
into remove_dquot_ref().

Thanks!
  

On Thu 29-06-23 20:13:05, Baokun Li wrote:
> On 2023/6/29 19:08, Jan Kara wrote:
> > On Wed 28-06-23 21:21:54, Baokun Li wrote:
> > > Now when dqput() drops the last reference count, it will call
> > > synchronize_srcu(&dquot_srcu) in quota_release_workfn() to ensure that
> > > no other user will use the dquot after the last reference count is dropped,
> > > so we don't need to call synchronize_srcu(&dquot_srcu) in drop_dquot_ref()
> > > and remove the corresponding logic directly to simplify the code.
> > Nice simplification!  It is also important that dqput() now cannot sleep
> > which was another reason for the logic with tofree_head in
> > remove_inode_dquot_ref().
> 
> I don't understand this sentence very well, so I would appreciate it
> 
> if you could explain it in detail. 🤔

OK, let me phrase it in a "changelog" way :):

remove_inode_dquot_ref() currently does not release the last dquot
reference but instead adds the dquot to tofree_head list. This is because
dqput() can sleep while dropping of the last dquot reference (writing back
the dquot and calling ->release_dquot()) and that must not happen under
dq_list_lock. Now that dqput() queues the final dquot cleanup into a
workqueue, remove_inode_dquot_ref() can call dqput() unconditionally
and we can significantly simplify it.

								Honza
  

On 2023/6/29 22:09, Jan Kara wrote:
> On Thu 29-06-23 20:13:05, Baokun Li wrote:
>> On 2023/6/29 19:08, Jan Kara wrote:
>>> On Wed 28-06-23 21:21:54, Baokun Li wrote:
>>>> Now when dqput() drops the last reference count, it will call
>>>> synchronize_srcu(&dquot_srcu) in quota_release_workfn() to ensure that
>>>> no other user will use the dquot after the last reference count is dropped,
>>>> so we don't need to call synchronize_srcu(&dquot_srcu) in drop_dquot_ref()
>>>> and remove the corresponding logic directly to simplify the code.
>>> Nice simplification!  It is also important that dqput() now cannot sleep
>>> which was another reason for the logic with tofree_head in
>>> remove_inode_dquot_ref().
>> I don't understand this sentence very well, so I would appreciate it
>>
>> if you could explain it in detail. 🤔
> OK, let me phrase it in a "changelog" way :):
>
> remove_inode_dquot_ref() currently does not release the last dquot
> reference but instead adds the dquot to tofree_head list. This is because
> dqput() can sleep while dropping of the last dquot reference (writing back
> the dquot and calling ->release_dquot()) and that must not happen under
> dq_list_lock. Now that dqput() queues the final dquot cleanup into a
> workqueue, remove_inode_dquot_ref() can call dqput() unconditionally
> and we can significantly simplify it.
>
> 								Honza
I suddenly understand what you mean, you mean that now dqput() doesn't have
any possible sleep operation. So it can be called in spin_lock at will.

I was confused because I understood that dqput() cannot be called in the 
context
of possible sleep.

Thank you very much for the detailed explanation!
Now there is only the problem in patch 5.

Thanks!