OpenZFS 6562 - Refquota on receive doesn't account for overage

Authored by: Dan McDonald <[email protected]>
Reviewed by: Matthew Ahrens <[email protected]>
Reviewed by: Yuri Pankov <[email protected]>
Reviewed by: Toomas Soome <[email protected]>
Approved by: Gordon Ross <[email protected]>
Ported-by: Brian Behlendorf <[email protected]>

OpenZFS-issue: https://www.illumos.org/issues/6562
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/5f7a8e6

1 files changed, 28 insertions, 3 deletions

diff --git a/module/zfs/dsl_dataset.c b/module/zfs/dsl_dataset.c
index e86e6cc73..3d1115e31 100644
--- a/module/zfs/dsl_dataset.c
+++ b/module/zfs/dsl_dataset.c
@@ -25,6 +25,7 @@
  * Copyright (c) 2014 RackTop Systems.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
+ * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
  */
 
 #include <sys/dmu_objset.h>
@@ -78,6 +79,8 @@ int zfs_max_recordsize = 1 * 1024 * 1024;
 
 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
 
+extern int spa_asize_inflation;
+
 /*
  * Figure out how much of this delta should be propogated to the dsl_dir
  * layer.  If there's a refreservation, that space has already been
@@ -2810,6 +2813,11 @@ int
 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
 {
+	/*
+	 * "slack" factor for received datasets with refquota set on them.
+	 * See the bottom of this function for details on its use.
+	 */
+	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
 	int64_t unused_refres_delta;
 
 	/* they should both be heads */
@@ -2852,10 +2860,22 @@ dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
 		return (SET_ERROR(ENOSPC));
 
-	/* clone can't be over the head's refquota */
+	/*
+	 * The clone can't be too much over the head's refquota.
+	 *
+	 * To ensure that the entire refquota can be used, we allow one
+	 * transaction to exceed the the refquota.  Therefore, this check
+	 * needs to also allow for the space referenced to be more than the
+	 * refquota.  The maximum amount of space that one transaction can use
+	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
+	 * overage ensures that we are able to receive a filesystem that
+	 * exceeds the refquota on the source system.
+	 *
+	 * So that overage is the refquota_slack we use below.
+	 */
 	if (origin_head->ds_quota != 0 &&
 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
-	    origin_head->ds_quota)
+	    origin_head->ds_quota + refquota_slack)
 		return (SET_ERROR(EDQUOT));
 
 	return (0);
@@ -2870,8 +2890,13 @@ dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
 	int64_t unused_refres_delta;
 
 	ASSERT(clone->ds_reserved == 0);
+	/*
+	 * NOTE: On DEBUG kernels there could be a race between this and
+	 * the check function if spa_asize_inflation is adjusted...
+	 */
 	ASSERT(origin_head->ds_quota == 0 ||
-	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
+	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
+	    DMU_MAX_ACCESS * spa_asize_inflation);
 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
 
 	/*