Re: [PATCH v4] btrfs: qgroup: Remove qgroup items along with subvolume deletion

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

[-- Attachment #1.1: Type: text/plain, Size: 5274 bytes --]



On 8/9/18 2:05 PM, Misono Tomohiro wrote:
> On 2018/08/09 14:47, Qu Wenruo wrote:
>>
>>
>> On 8/9/18 12:12 PM, Misono Tomohiro wrote:
>>> When qgroup is on, subvolume deletion does not remove qgroup items
>>> of the subvolume (qgroup info, limit, relation) from quota tree and
>>> they need to get removed manually by "btrfs qgroup destroy".
>>>
>>> Since level 0 qgroup cannot be used/inherited by any other subvolume,
>>> let's remove them automatically when subvolume is deleted
>>> (to be precise, when the subvolume root is dropped).
>>>
>>> Note that qgroup becomes inconsistent in following case:
>>>   1. qgroup relation exists
>>>   2. and subvolume's excl != rref
>>
>> That's a little strange.
>>
>> If a subvolume is completely dropped, its excl should be the same rfer,
>> all 0, and removing its relationship should not mark qgroup inconsistent.
>>
>> So the problem is the timing when btrfs_remove_qgroup() is called.
>>
>> Since qgroup accounting is only called at transaction commit time, and
>> we're holding a trans handler, it's almost ensured we can't commit this
>> transaction, thus the number is not updated yet (still not 0)
>>
>> So that's why qgroup is inconsistent.
>>
>> What about commit current transaction and then call btrfs_remove_qgroup()?
>>
>> (Sorry I didn't catch this problem last time I reviewed this patch)
> 
> well, I'm little confusing about flow of transaction commit.
> btrfs_drop_snapshot() is called from cleaner_kthread and
> is it ok to commit transaction in it?

Not completely clear of the cleaner_kthread(), but from what I see in
btrfs_drop_snapshot(), btrfs_end_transaction_throttle() itself could
commit current transaction.

So in theory we should be OK to finish all the original work of
btrfs_drop_snapshot(), and then commit current transaction, and finally
do the qgroup cleanup work.

But I could totally be wrong, and feel free to point what I'm missing.

Thanks,
Qu

> 
>>
>> Thanks,
>> Qu
>>
>>> In this case manual qgroup rescan is needed.
>>>
>>> Reviewed-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
>>> Reviewed-by: Qu Wenruo <wqu@suse.com>
>>> Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
>>> ---
>>> Hi David,
>>> It turned out that this patch may cause qgroup inconsistency in case
>>> described above and need manual rescan. Since current code will keep 
>>> qgroup items but not break qgroup consistency when deleting subvolume,
>>> I cannot clearly say which behavior is better for qgroup usability.
>>> Can I ask your opinion?
>>>
>>> v3 -> v4:
>>>   Check return value of btrfs_remove_qgroup() and if it is 1,
>>>   print message in syslog that fs needs qgroup rescan
>>>
>>>  fs/btrfs/extent-tree.c | 22 ++++++++++++++++++----
>>>  1 file changed, 18 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
>>> index 9e7b237b9547..828d9e68047d 100644
>>> --- a/fs/btrfs/extent-tree.c
>>> +++ b/fs/btrfs/extent-tree.c
>>> @@ -8871,12 +8871,13 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
>>>  	struct btrfs_root_item *root_item = &root->root_item;
>>>  	struct walk_control *wc;
>>>  	struct btrfs_key key;
>>> +	u64 objectid = root->root_key.objectid;
>>>  	int err = 0;
>>>  	int ret;
>>>  	int level;
>>>  	bool root_dropped = false;
>>>  
>>> -	btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
>>> +	btrfs_debug(fs_info, "Drop subvolume %llu", objectid);
>>>  
>>>  	path = btrfs_alloc_path();
>>>  	if (!path) {
>>> @@ -9030,7 +9031,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
>>>  		goto out_end_trans;
>>>  	}
>>>  
>>> -	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
>>> +	if (objectid != BTRFS_TREE_RELOC_OBJECTID) {
>>>  		ret = btrfs_find_root(tree_root, &root->root_key, path,
>>>  				      NULL, NULL);
>>>  		if (ret < 0) {
>>> @@ -9043,8 +9044,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
>>>  			 *
>>>  			 * The most common failure here is just -ENOENT.
>>>  			 */
>>> -			btrfs_del_orphan_item(trans, tree_root,
>>> -					      root->root_key.objectid);
>>> +			btrfs_del_orphan_item(trans, tree_root, objectid);
>>>  		}
>>>  	}
>>>  
>>> @@ -9056,6 +9056,20 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
>>>  		btrfs_put_fs_root(root);
>>>  	}
>>>  	root_dropped = true;
>>> +
>>> +	 /* Remove level-0 qgroup items since no other subvolume can use them */
>>> +	ret = btrfs_remove_qgroup(trans, objectid);
>>> +	if (ret == 1) {
>>> +		/* This means qgroup becomes inconsistent by removing items */
>>> +		btrfs_info(fs_info,
>>> +		    "qgroup inconsistency found, need qgroup rescan");
>>> +	} else if (ret == -EINVAL || ret == -ENOENT) {
>>> +		/* qgroup is not enabled or already removed, just ignore this */
>>> +	} else if (ret) {
>>> +		btrfs_abort_transaction(trans, ret);
>>> +		err = ret;
>>> +	}
>>> +
>>>  out_end_trans:
>>>  	btrfs_end_transaction_throttle(trans);
>>>  out_free:
>>>
>>
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 


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