Re: [PATCH v2 20/21] btrfs: skip LOOP_NO_EMPTY_SIZE if not clustered allocation

From: Josef Bacik <josef@toxicpanda.com>
To: Naohiro Aota <naohiro.aota@wdc.com>
Cc: linux-btrfs@vger.kernel.org, David Sterba <dsterba@suse.com>,
	Chris Mason <clm@fb.com>, Nikolay Borisov <nborisov@suse.com>,
	Damien Le Moal <damien.lemoal@wdc.com>,
	Johannes Thumshirn <Johannes.Thumshirn@wdc.com>,
	Hannes Reinecke <hare@suse.com>,
	Anand Jain <anand.jain@oracle.com>,
	linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH v2 20/21] btrfs: skip LOOP_NO_EMPTY_SIZE if not clustered allocation
Date: Thu, 20 Feb 2020 10:40:22 -0500	[thread overview]
Message-ID: <c0829c4d-17b1-ad99-c99e-87f859f290e4@toxicpanda.com> (raw)
In-Reply-To: <20200220095631.7rlk7lmnp7np4nvg@naota.dhcp.fujisawa.hgst.com>

On 2/20/20 4:56 AM, Naohiro Aota wrote:
> On Thu, Feb 13, 2020 at 02:55:30PM -0500, Josef Bacik wrote:
>> On 2/12/20 2:20 AM, Naohiro Aota wrote:
>>> LOOP_NO_EMPTY_SIZE is solely dedicated for clustered allocation. So,
>>> we can skip this stage and go to LOOP_GIVEUP stage to indicate we gave
>>> up the allocation. This commit also moves the scope of the "clustered"
>>> variable.
>>>
>>> Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
>>> ---
>>>  fs/btrfs/extent-tree.c | 6 ++++++
>>>  1 file changed, 6 insertions(+)
>>>
>>> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
>>> index 8f0d489f76fa..3ab0d2f5d718 100644
>>> --- a/fs/btrfs/extent-tree.c
>>> +++ b/fs/btrfs/extent-tree.c
>>> @@ -3373,6 +3373,7 @@ enum btrfs_loop_type {
>>>      LOOP_CACHING_WAIT,
>>>      LOOP_ALLOC_CHUNK,
>>>      LOOP_NO_EMPTY_SIZE,
>>> +    LOOP_GIVEUP,
>>
>> Why do we need a new loop definition here?  Can we just return ENOSPC and be 
>> done?  You don't appear to use it anywhere, so it doesn't seem like it's 
>> needed.  Thanks,
>>
>> Josef
> 
> This is for other allocation policy to skip unnecessary loop stages
> (e.g. LOOP_NO_EMPTY_SIZE) from an earlier stage. For example, zoned
> allocation policy can implement the code below in
> chunk_allocation_failed() to skip the following stages.
> 
> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> index 4badfae0c932..0a18c09b078b 100644
> --- a/fs/btrfs/extent-tree.c
> +++ b/fs/btrfs/extent-tree.c
> @@ -3775,6 +3854,10 @@ static int chunk_allocation_failed(struct 
> find_free_extent_ctl *ffe_ctl)
>                   */
>                  ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
>                  return 0;
> +       case BTRFS_EXTENT_ALLOC_ZONED:
> +               /* give up here */
> +               ffe_ctl->loop = LOOP_GIVEUP;
> +               return -ENOSPC;
>          default:
>                  BUG();
>          }
> 
> But, I can keep this LOOP_GIVEUP introduction patch later with this
> zoned allocator ones.
> 

Yes I'd rather they be with the real user, otherwise it's just confusing.  Thanks,

Josef