Re: [PATCH 3/4] btrfs: switch GFP_NOFS to GFP_KERNEL in scrub_setup_recheck_block

[Anand Jain <anand.jain@oracle.com>]

On 21/10/2022 00:35, David Sterba wrote:
> On Thu, Oct 20, 2022 at 03:27:31PM +0800, Anand Jain wrote:
>> On 18/10/2022 22:27, David Sterba wrote:
>>> There's only one caller that calls scrub_setup_recheck_block in the
>>> memalloc_nofs_save/_restore protection so it's effectively already
>>> GFP_NOFS and it's safe to use GFP_KERNEL.
>>>
>>> Signed-off-by: David Sterba <dsterba@suse.com>
>>> ---
>>>    fs/btrfs/scrub.c | 4 ++--
>>>    1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
>>> index 9e3b2e60e571..2fc70a2cc7fe 100644
>>> --- a/fs/btrfs/scrub.c
>>> +++ b/fs/btrfs/scrub.c
>>> @@ -1491,7 +1491,7 @@ static int scrub_setup_recheck_block(struct scrub_block *original_sblock,
>>>    			return -EIO;
>>>    		}
>>>    
>>> -		recover = kzalloc(sizeof(struct scrub_recover), GFP_NOFS);
>>> +		recover = kzalloc(sizeof(struct scrub_recover), GFP_KERNEL);
>>
>>
>> I didn't get why GFP_KERNEL is better here, or would it make any
>> difference, given that we are already (and rightly) in the
>> memalloc_nofs_save() scope.
> 
> You said what's the reason, so I can only repeat the patterns:
> 
> 1) plain GFP_NOFS, with notice that it does not work with kvalloc
> 2) memalloc_nofs_save/k.alloc(GFP_KERNEL)/memalloc_nofs_restore
> 
> I.e. GFP_NOFS in the memalloc_nofs_ protection is redundant because we
> get th NOFS protection and can safely use GFP_KERNEL. 



> Because we want to
> minimize use of GFP_NOFS or completely switch to scoped nofs, but this
> requires other changes and we do that incrementally.

  Ok. Got it.