Re: [PATCH] btrfs: qgroup: Init flags with RESCAN bit at quota enable time

[Qu Wenruo <wqu@suse.de>]

[-- Attachment #1.1: Type: text/plain, Size: 3477 bytes --]



On 2018年07月27日 09:43, Misono Tomohiro wrote:
> On 2018/07/27 10:19, Qu Wenruo wrote:
>>
>>
>> On 2018年07月27日 09:10, Misono Tomohiro wrote:
>>> On 2018/07/26 18:15, Qu Wenruo wrote:
>>>> Between btrfs_quota_enable() finished and rescan kicked in, there is a
>>>> small window that quota status has (ON | INCONSISTENT) bits set but
>>>> without RESCAN bits set.
>>>>
>>>> And transaction is committed inside the window and then power loss
>>>> happens, we will have a quota tree with all qgroup numbers set to 0, and
>>>> not RESCAN bit set.
>>>>
>>>> At next mount time, qgroup rescan will not kick in due to the missing of
>>>> RESCAN bit, user needs to kick in rescan manually.
>>>>
>>>> This patch will fix it by setting RESCAN bit at btrfs_quota_enable(),
>>>> so even after power loss we will still kick in rescan automatically.
>>>>
>>>> Suggested-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
>>>> Signed-off-by: Qu Wenruo <wqu@suse.com>
>>>> ---
>>>>  fs/btrfs/qgroup.c | 5 +++--
>>>>  1 file changed, 3 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
>>>> index c25dc47210a3..13c1c7dd278d 100644
>>>> --- a/fs/btrfs/qgroup.c
>>>> +++ b/fs/btrfs/qgroup.c
>>>> @@ -930,7 +930,8 @@ int btrfs_quota_enable(struct btrfs_trans_handle *trans,
>>>>  	btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
>>>>  	btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
>>>>  	fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
>>>> -				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
>>>> +				BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
>>>> +				BTRFS_QGROUP_STATUS_FLAG_RESCAN;
>>>>  	btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
>>>>  	btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
>>>>  
>>>> @@ -987,7 +988,7 @@ int btrfs_quota_enable(struct btrfs_trans_handle *trans,
>>>>  	fs_info->quota_root = quota_root;
>>>>  	set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
>>>>  	spin_unlock(&fs_info->qgroup_lock);
>>>> -	ret = qgroup_rescan_init(fs_info, 0, 1);
>>>> +	ret = qgroup_rescan_init(fs_info, 0, 0);
>>>>  	if (!ret) {
>>>>  	        qgroup_rescan_zero_tracking(fs_info);
>>>>  	        btrfs_queue_work(fs_info->qgroup_rescan_workers,
>>>>
>>>
>>> This is what I think at first, but is it ok not holding fs_info->qgroup_ioctl_lock
>>> in brfs_qgroup_rescan() as you concerned in previous thread?
>>
>> I think it's OK, since we have larger mutex (subvol_sem) for
>> quota_enable/disable() so there will be no concurrency modifying flags.
>> And we're holding trans handler from btrfs_ioctl_quota_ctl(),
>> transaction won't be committed in btrfs_quota_enable().
> 
> Ok, but nikolay's patch in misc-next moves transaction commit in btrfs_quota_enable():
>   https://patchwork.kernel.org/patch/10508819/
>   ("btrfs: qgroups: Move transaction management inside btrfs_quota_enable/disable")

Since qgroup_rescan_init() has nothing do to with transaction, it looks
OK even with Nikolay's patch.

> 
> This is related to https://marc.info/?l=linux-btrfs&m=152999289017582.
> However, it seems that other people does not see the problem,
> so I'm not sure how the above patch ends up...

IIRC I also failed to reproduce it, thus can't provide much help for
that thread.

Thanks,
Qu



> 
> Thanks,
> Tomohiro Misono
> 
>>
>> So I think it's OK.
>>
>> Thanks,
>> Qu
>>
>>
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]