Re: [PATCH v3 2/3] Btrfs: rescan for qgroups

[Wang Shilong <wangsl-fnst@cn.fujitsu.com>]

Hello Jan,

[..snip..]



>  	/*
>  	 * the delayed ref sequence number we pass depends on the direction of
>  	 * the operation. for add operations, we pass (node->seq - 1) to skip
> @@ -1401,7 +1428,17 @@ int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
>  	if (ret < 0)
>  		return ret;
>  
> +	mutex_lock(&fs_info->qgroup_rescan_lock);
>  	spin_lock(&fs_info->qgroup_lock);
> +	if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
> +		if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
> +			ret = 0;
> +			mutex_unlock(&fs_info->qgroup_rescan_lock);
> +			goto unlock;
> +		}
> +	}
> +	mutex_unlock(&fs_info->qgroup_rescan_lock);
> +
>  	quota_root = fs_info->quota_root;
>  	if (!quota_root)
>  		goto unlock;
> @@ -1820,3 +1857,250 @@ void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
>  		trans->delayed_ref_elem.seq);
>  	BUG();
>  }
> +
> +/*
> + * returns < 0 on error, 0 when more leafs are to be scanned.
> + * returns 1 when done, 2 when done and FLAG_INCONSISTENT was cleared.
> + */
> +static int
> +qgroup_rescan_leaf(struct qgroup_rescan *qscan, struct btrfs_path *path,
> +		   struct btrfs_trans_handle *trans, struct ulist *tmp,
> +		   struct extent_buffer *scratch_leaf)
> +{
> +	struct btrfs_key found;
> +	struct btrfs_fs_info *fs_info = qscan->fs_info;
> +	struct ulist *roots = NULL;
> +	struct ulist_node *unode;
> +	struct ulist_iterator uiter;
> +	struct seq_list tree_mod_seq_elem = {};
> +	u64 seq;
> +	int slot;
> +	int ret;
> +
> +	path->leave_spinning = 1;
> +	mutex_lock(&fs_info->qgroup_rescan_lock);
> +	ret = btrfs_search_slot_for_read(fs_info->extent_root,
> +					 &fs_info->qgroup_rescan_progress,
> +					 path, 1, 0);
> +
> +	pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
> +		 (unsigned long long)fs_info->qgroup_rescan_progress.objectid,
> +		 fs_info->qgroup_rescan_progress.type,
> +		 (unsigned long long)fs_info->qgroup_rescan_progress.offset,
> +		 ret);
> +
> +	if (ret) {
> +		/*
> +		 * The rescan is about to end, we will not be scanning any
> +		 * further blocks. We cannot unset the RESCAN flag here, because
> +		 * we want to commit the transaction if everything went well.
> +		 * To make the live accounting work in this phase, we set our
> +		 * scan progress pointer such that every real extent objectid
> +		 * will be smaller.
> +		 */
> +		fs_info->qgroup_rescan_progress.objectid = (u64)-1;
> +		btrfs_release_path(path);
> +		mutex_unlock(&fs_info->qgroup_rescan_lock);
> +		return ret;
> +	}
> +
> +	btrfs_item_key_to_cpu(path->nodes[0], &found,
> +			      btrfs_header_nritems(path->nodes[0]) - 1);
> +	fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
> +
> +	btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
> +	memcpy(scratch_leaf, path->nodes[0], sizeof(*scratch_leaf));
> +	slot = path->slots[0];
> +	btrfs_release_path(path);
> +	mutex_unlock(&fs_info->qgroup_rescan_lock);
> +
> +	for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
> +		btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
> +		if (found.type != BTRFS_EXTENT_ITEM_KEY)
> +			continue;
> +		ret = btrfs_find_all_roots(trans, fs_info, found.objectid,
> +					   tree_mod_seq_elem.seq, &roots);
> +		if (ret < 0)
> +			break;
> +		spin_lock(&fs_info->qgroup_lock);
> +		seq = fs_info->qgroup_seq;
> +		fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
> +
> +		ulist_reinit(tmp);
> +		ULIST_ITER_INIT(&uiter);
> +		while ((unode = ulist_next(roots, &uiter))) {
> +			struct btrfs_qgroup *qg;
> +
> +			qg = find_qgroup_rb(fs_info, unode->val);
> +			if (!qg)
> +				continue;
> +
> +			ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg,
> +					GFP_ATOMIC);


If ulist_add() fails, we still need to call ulist_free(roots)..

> +			if (ret < 0) {
> +				spin_unlock(&fs_info->qgroup_lock);

> +				goto out;
> +			}
> +		}
> +
> +		/* this is similar to step 2 of btrfs_qgroup_account_ref */
> +		ULIST_ITER_INIT(&uiter);
> +		while ((unode = ulist_next(tmp, &uiter))) {
> +			struct btrfs_qgroup *qg;
> +			struct btrfs_qgroup_list *glist;
> +
> +			qg = (struct btrfs_qgroup *)(uintptr_t) unode->aux;
> +			qg->rfer += found.offset;
> +			qg->rfer_cmpr += found.offset;
> +			WARN_ON(qg->tag >= seq);
> +			WARN_ON(qg->refcnt >= seq);
> +			if (qg->refcnt < seq)
> +				qg->refcnt = seq + 1;
> +			else
> +				qg->refcnt = qg->refcnt + 1;
> +			qgroup_dirty(fs_info, qg);
> +
> +			list_for_each_entry(glist, &qg->groups, next_group) {
> +				ret = ulist_add(tmp, glist->group->qgroupid,
> +						(uintptr_t)glist->group,
> +						GFP_ATOMIC);


If ulist_add() fails, we still need to call ulist_free(roots)..

> +				if (ret < 0) {
> +					spin_unlock(&fs_info->qgroup_lock);
> +					goto out;
> +				}
> +			}
> +		}
> +
> +		qgroup_account_ref_step3(fs_info, roots, tmp, seq, -1,
> +					 found.offset);

please check return value of qgroup_account_ref_step3()... As -ENOMEM may happen.

> +
> +		spin_unlock(&fs_info->qgroup_lock);
> +		ulist_free(roots);
> +		ret = 0;
> +	}
> +
> +out:
> +	btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
> +
> +	return ret;
> +}
> +
>

[..snip..]