[PATCH v2] fs: Kill DCACHE_DONTCACHE dentry even if DCACHE_REFERENCED is set

[PATCH v2] fs: Kill DCACHE_DONTCACHE dentry even if DCACHE_REFERENCED is set

[Hao Li]

If DCACHE_REFERENCED is set, fast_dput() will return true, and then
retain_dentry() have no chance to check DCACHE_DONTCACHE. As a result,
the dentry won't be killed and the corresponding inode can't be evicted.
In the following example, the DAX policy can't take effects unless we
do a drop_caches manually.

  # DCACHE_LRU_LIST will be set
  echo abcdefg > test.txt

  # DCACHE_REFERENCED will be set and DCACHE_DONTCACHE can't do anything
  xfs_io -c 'chattr +x' test.txt

  # Drop caches to make DAX changing take effects
  echo 2 > /proc/sys/vm/drop_caches

What this patch does is preventing fast_dput() from returning true if
DCACHE_DONTCACHE is set. Then retain_dentry() will detect the
DCACHE_DONTCACHE and will return false. As a result, the dentry will be
killed and the inode will be evicted. In this way, if we change per-file
DAX policy, it will take effects automatically after this file is closed
by all processes.

I also add some comments to make the code more clear.

Signed-off-by: Hao Li <lihao2018.fnst@cn.fujitsu.com>
---
v1 is split into two standalone patch as discussed in [1], and the first
patch has been reviewed in [2]. This is the second patch.

[1]: https://lore.kernel.org/linux-fsdevel/20200831003407.GE12096@dread.disaster.area/
[2]: https://lore.kernel.org/linux-fsdevel/20200906214002.GI12131@dread.disaster.area/

 fs/dcache.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/fs/dcache.c b/fs/dcache.c
index ea0485861d93..97e81a844a96 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -793,10 +793,17 @@ static inline bool fast_dput(struct dentry *dentry)
 	 * a reference to the dentry and change that, but
 	 * our work is done - we can leave the dentry
 	 * around with a zero refcount.
+	 *
+	 * Nevertheless, there are two cases that we should kill
+	 * the dentry anyway.
+	 * 1. free disconnected dentries as soon as their refcount
+	 *    reached zero.
+	 * 2. free dentries if they should not be cached.
 	 */
 	smp_rmb();
 	d_flags = READ_ONCE(dentry->d_flags);
-	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
+	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
+			DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
 
 	/* Nothing to do? Dropping the reference was all we needed? */
 	if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
-- 
2.28.0

Re: [PATCH v2] fs: Kill DCACHE_DONTCACHE dentry even if DCACHE_REFERENCED is set

[Jan Kara]

On Thu 24-09-20 13:59:58, Hao Li wrote:
> If DCACHE_REFERENCED is set, fast_dput() will return true, and then
> retain_dentry() have no chance to check DCACHE_DONTCACHE. As a result,
> the dentry won't be killed and the corresponding inode can't be evicted.
> In the following example, the DAX policy can't take effects unless we
> do a drop_caches manually.
> 
>   # DCACHE_LRU_LIST will be set
>   echo abcdefg > test.txt
> 
>   # DCACHE_REFERENCED will be set and DCACHE_DONTCACHE can't do anything
>   xfs_io -c 'chattr +x' test.txt
> 
>   # Drop caches to make DAX changing take effects
>   echo 2 > /proc/sys/vm/drop_caches
> 
> What this patch does is preventing fast_dput() from returning true if
> DCACHE_DONTCACHE is set. Then retain_dentry() will detect the
> DCACHE_DONTCACHE and will return false. As a result, the dentry will be
> killed and the inode will be evicted. In this way, if we change per-file
> DAX policy, it will take effects automatically after this file is closed
> by all processes.
> 
> I also add some comments to make the code more clear.
> 
> Signed-off-by: Hao Li <lihao2018.fnst@cn.fujitsu.com>

The patch looks good to me. You can add:

Reviewed-by: Jan Kara <jack@suse.cz>

								Honza

> ---
> v1 is split into two standalone patch as discussed in [1], and the first
> patch has been reviewed in [2]. This is the second patch.
> 
> [1]: https://lore.kernel.org/linux-fsdevel/20200831003407.GE12096@dread.disaster.area/
> [2]: https://lore.kernel.org/linux-fsdevel/20200906214002.GI12131@dread.disaster.area/
> 
>  fs/dcache.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/dcache.c b/fs/dcache.c
> index ea0485861d93..97e81a844a96 100644
> --- a/fs/dcache.c
> +++ b/fs/dcache.c
> @@ -793,10 +793,17 @@ static inline bool fast_dput(struct dentry *dentry)
>  	 * a reference to the dentry and change that, but
>  	 * our work is done - we can leave the dentry
>  	 * around with a zero refcount.
> +	 *
> +	 * Nevertheless, there are two cases that we should kill
> +	 * the dentry anyway.
> +	 * 1. free disconnected dentries as soon as their refcount
> +	 *    reached zero.
> +	 * 2. free dentries if they should not be cached.
>  	 */
>  	smp_rmb();
>  	d_flags = READ_ONCE(dentry->d_flags);
> -	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
> +	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
> +			DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
>  
>  	/* Nothing to do? Dropping the reference was all we needed? */
>  	if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
> -- 
> 2.28.0
> 
> 
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH v2] fs: Kill DCACHE_DONTCACHE dentry even if DCACHE_REFERENCED is set

[Ira Weiny]

On Thu, Sep 24, 2020 at 01:59:58PM +0800, Hao Li wrote:
> If DCACHE_REFERENCED is set, fast_dput() will return true, and then
> retain_dentry() have no chance to check DCACHE_DONTCACHE. As a result,
> the dentry won't be killed and the corresponding inode can't be evicted.
> In the following example, the DAX policy can't take effects unless we
> do a drop_caches manually.
> 
>   # DCACHE_LRU_LIST will be set
>   echo abcdefg > test.txt
> 
>   # DCACHE_REFERENCED will be set and DCACHE_DONTCACHE can't do anything
>   xfs_io -c 'chattr +x' test.txt
> 
>   # Drop caches to make DAX changing take effects
>   echo 2 > /proc/sys/vm/drop_caches
> 
> What this patch does is preventing fast_dput() from returning true if
> DCACHE_DONTCACHE is set. Then retain_dentry() will detect the
> DCACHE_DONTCACHE and will return false. As a result, the dentry will be
> killed and the inode will be evicted. In this way, if we change per-file
> DAX policy, it will take effects automatically after this file is closed
> by all processes.
> 
> I also add some comments to make the code more clear.
> 
> Signed-off-by: Hao Li <lihao2018.fnst@cn.fujitsu.com>

Reviewed-by: Ira Weiny <ira.weiny@intel.com>

> ---
> v1 is split into two standalone patch as discussed in [1], and the first
> patch has been reviewed in [2]. This is the second patch.
> 
> [1]: https://lore.kernel.org/linux-fsdevel/20200831003407.GE12096@dread.disaster.area/
> [2]: https://lore.kernel.org/linux-fsdevel/20200906214002.GI12131@dread.disaster.area/
> 
>  fs/dcache.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/dcache.c b/fs/dcache.c
> index ea0485861d93..97e81a844a96 100644
> --- a/fs/dcache.c
> +++ b/fs/dcache.c
> @@ -793,10 +793,17 @@ static inline bool fast_dput(struct dentry *dentry)
>  	 * a reference to the dentry and change that, but
>  	 * our work is done - we can leave the dentry
>  	 * around with a zero refcount.
> +	 *
> +	 * Nevertheless, there are two cases that we should kill
> +	 * the dentry anyway.
> +	 * 1. free disconnected dentries as soon as their refcount
> +	 *    reached zero.
> +	 * 2. free dentries if they should not be cached.
>  	 */
>  	smp_rmb();
>  	d_flags = READ_ONCE(dentry->d_flags);
> -	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
> +	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
> +			DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
>  
>  	/* Nothing to do? Dropping the reference was all we needed? */
>  	if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
> -- 
> 2.28.0
> 
> 
> 

Re: [PATCH v2] fs: Kill DCACHE_DONTCACHE dentry even if DCACHE_REFERENCED is set

[Jan Kara]

Hum, Al, did this patch get lost?

								Honza

On Thu 24-09-20 16:58:56, Jan Kara wrote:
> On Thu 24-09-20 13:59:58, Hao Li wrote:
> > If DCACHE_REFERENCED is set, fast_dput() will return true, and then
> > retain_dentry() have no chance to check DCACHE_DONTCACHE. As a result,
> > the dentry won't be killed and the corresponding inode can't be evicted.
> > In the following example, the DAX policy can't take effects unless we
> > do a drop_caches manually.
> > 
> >   # DCACHE_LRU_LIST will be set
> >   echo abcdefg > test.txt
> > 
> >   # DCACHE_REFERENCED will be set and DCACHE_DONTCACHE can't do anything
> >   xfs_io -c 'chattr +x' test.txt
> > 
> >   # Drop caches to make DAX changing take effects
> >   echo 2 > /proc/sys/vm/drop_caches
> > 
> > What this patch does is preventing fast_dput() from returning true if
> > DCACHE_DONTCACHE is set. Then retain_dentry() will detect the
> > DCACHE_DONTCACHE and will return false. As a result, the dentry will be
> > killed and the inode will be evicted. In this way, if we change per-file
> > DAX policy, it will take effects automatically after this file is closed
> > by all processes.
> > 
> > I also add some comments to make the code more clear.
> > 
> > Signed-off-by: Hao Li <lihao2018.fnst@cn.fujitsu.com>
> 
> The patch looks good to me. You can add:
> 
> Reviewed-by: Jan Kara <jack@suse.cz>
> 
> 								Honza
> 
> > ---
> > v1 is split into two standalone patch as discussed in [1], and the first
> > patch has been reviewed in [2]. This is the second patch.
> > 
> > [1]: https://lore.kernel.org/linux-fsdevel/20200831003407.GE12096@dread.disaster.area/
> > [2]: https://lore.kernel.org/linux-fsdevel/20200906214002.GI12131@dread.disaster.area/
> > 
> >  fs/dcache.c | 9 ++++++++-
> >  1 file changed, 8 insertions(+), 1 deletion(-)
> > 
> > diff --git a/fs/dcache.c b/fs/dcache.c
> > index ea0485861d93..97e81a844a96 100644
> > --- a/fs/dcache.c
> > +++ b/fs/dcache.c
> > @@ -793,10 +793,17 @@ static inline bool fast_dput(struct dentry *dentry)
> >  	 * a reference to the dentry and change that, but
> >  	 * our work is done - we can leave the dentry
> >  	 * around with a zero refcount.
> > +	 *
> > +	 * Nevertheless, there are two cases that we should kill
> > +	 * the dentry anyway.
> > +	 * 1. free disconnected dentries as soon as their refcount
> > +	 *    reached zero.
> > +	 * 2. free dentries if they should not be cached.
> >  	 */
> >  	smp_rmb();
> >  	d_flags = READ_ONCE(dentry->d_flags);
> > -	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
> > +	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
> > +			DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
> >  
> >  	/* Nothing to do? Dropping the reference was all we needed? */
> >  	if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
> > -- 
> > 2.28.0
> > 
> > 
> > 
> -- 
> Jan Kara <jack@suse.com>
> SUSE Labs, CR
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH v2] fs: Kill DCACHE_DONTCACHE dentry even if DCACHE_REFERENCED is set

[Li, Hao]

Ping :)

On 2020/10/21 21:51, Jan Kara wrote:
> Hum, Al, did this patch get lost?
>
> 								Honza
>
> On Thu 24-09-20 16:58:56, Jan Kara wrote:
>> On Thu 24-09-20 13:59:58, Hao Li wrote:
>>> If DCACHE_REFERENCED is set, fast_dput() will return true, and then
>>> retain_dentry() have no chance to check DCACHE_DONTCACHE. As a result,
>>> the dentry won't be killed and the corresponding inode can't be evicted.
>>> In the following example, the DAX policy can't take effects unless we
>>> do a drop_caches manually.
>>>
>>>   # DCACHE_LRU_LIST will be set
>>>   echo abcdefg > test.txt
>>>
>>>   # DCACHE_REFERENCED will be set and DCACHE_DONTCACHE can't do anything
>>>   xfs_io -c 'chattr +x' test.txt
>>>
>>>   # Drop caches to make DAX changing take effects
>>>   echo 2 > /proc/sys/vm/drop_caches
>>>
>>> What this patch does is preventing fast_dput() from returning true if
>>> DCACHE_DONTCACHE is set. Then retain_dentry() will detect the
>>> DCACHE_DONTCACHE and will return false. As a result, the dentry will be
>>> killed and the inode will be evicted. In this way, if we change per-file
>>> DAX policy, it will take effects automatically after this file is closed
>>> by all processes.
>>>
>>> I also add some comments to make the code more clear.
>>>
>>> Signed-off-by: Hao Li <lihao2018.fnst@cn.fujitsu.com>
>> The patch looks good to me. You can add:
>>
>> Reviewed-by: Jan Kara <jack@suse.cz>
>>
>> 								Honza
>>
>>> ---
>>> v1 is split into two standalone patch as discussed in [1], and the first
>>> patch has been reviewed in [2]. This is the second patch.
>>>
>>> [1]: https://lore.kernel.org/linux-fsdevel/20200831003407.GE12096@dread.disaster.area/
>>> [2]: https://lore.kernel.org/linux-fsdevel/20200906214002.GI12131@dread.disaster.area/
>>>
>>>  fs/dcache.c | 9 ++++++++-
>>>  1 file changed, 8 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/fs/dcache.c b/fs/dcache.c
>>> index ea0485861d93..97e81a844a96 100644
>>> --- a/fs/dcache.c
>>> +++ b/fs/dcache.c
>>> @@ -793,10 +793,17 @@ static inline bool fast_dput(struct dentry *dentry)
>>>  	 * a reference to the dentry and change that, but
>>>  	 * our work is done - we can leave the dentry
>>>  	 * around with a zero refcount.
>>> +	 *
>>> +	 * Nevertheless, there are two cases that we should kill
>>> +	 * the dentry anyway.
>>> +	 * 1. free disconnected dentries as soon as their refcount
>>> +	 *    reached zero.
>>> +	 * 2. free dentries if they should not be cached.
>>>  	 */
>>>  	smp_rmb();
>>>  	d_flags = READ_ONCE(dentry->d_flags);
>>> -	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
>>> +	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
>>> +			DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
>>>  
>>>  	/* Nothing to do? Dropping the reference was all we needed? */
>>>  	if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
>>> -- 
>>> 2.28.0
>>>
>>>
>>>
>> -- 
>> Jan Kara <jack@suse.com>
>> SUSE Labs, CR