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@xxxxxxxxxxxxxx>
>> Signed-off-by: Qu Wenruo <wqu@xxxxxxxx>
>> ---
>> 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().
So I think it's OK.
Thanks,
Qu
Attachment:
signature.asc
Description: OpenPGP digital signature
