Backport of Fix available space accounting for special/dedup (#18222)

Sponsored-by: Klara, Inc.
Sponsored-by: OSNexus

Author: amotin

include/sys/metaslab.h

@@ -112,12 +112,12 @@ boolean_t metaslab_class_throttle_reserve(metaslab_class_t *, int, int,
 void metaslab_class_throttle_unreserve(metaslab_class_t *, int, int, zio_t *);
 void metaslab_class_evict_old(metaslab_class_t *, uint64_t);
 uint64_t metaslab_class_get_alloc(metaslab_class_t *);
+uint64_t metaslab_class_get_dalloc(metaslab_class_t *);
 uint64_t metaslab_class_get_space(metaslab_class_t *);
 uint64_t metaslab_class_get_dspace(metaslab_class_t *);
 uint64_t metaslab_class_get_deferred(metaslab_class_t *);
 
-void metaslab_space_update(vdev_t *, metaslab_class_t *,
-    int64_t, int64_t, int64_t);
+void metaslab_space_update(metaslab_group_t *, int64_t, int64_t, int64_t);
 
 metaslab_group_t *metaslab_group_create(metaslab_class_t *, vdev_t *, int);
 void metaslab_group_destroy(metaslab_group_t *);

include/sys/metaslab_impl.h

@@ -196,10 +196,12 @@ struct metaslab_class {
 
 	uint64_t		mc_alloc_groups; /* # of allocatable groups */
 
-	uint64_t		mc_alloc;	/* total allocated space */
-	uint64_t		mc_deferred;	/* total deferred frees */
+	uint64_t		mc_alloc;	/* allocated space */
+	uint64_t		mc_dalloc;	/* deflated allocated space */
+	uint64_t		mc_deferred;	/* deferred frees */
+	uint64_t		mc_ddeferred;	/* deflated deferred frees */
 	uint64_t		mc_space;	/* total space (alloc + free) */
-	uint64_t		mc_dspace;	/* total deflated space */
+	uint64_t		mc_dspace;	/* deflated total space */
 	uint64_t		mc_histogram[RANGE_TREE_HISTOGRAM_SIZE];
 
 	/*

module/zfs/metaslab.c

@@ -433,10 +433,12 @@ metaslab_class_destroy(metaslab_class_t *mc)
 {
 	spa_t *spa = mc->mc_spa;
 
-	ASSERT(mc->mc_alloc == 0);
-	ASSERT(mc->mc_deferred == 0);
-	ASSERT(mc->mc_space == 0);
-	ASSERT(mc->mc_dspace == 0);
+	ASSERT0(mc->mc_alloc);
+	ASSERT0(mc->mc_dalloc);
+	ASSERT0(mc->mc_deferred);
+	ASSERT0(mc->mc_ddeferred);
+	ASSERT0(mc->mc_space);
+	ASSERT0(mc->mc_dspace);
 
 	for (int i = 0; i < spa->spa_alloc_count; i++) {
 		metaslab_class_allocator_t *mca = &mc->mc_allocator[i];
@@ -477,36 +479,48 @@ metaslab_class_validate(metaslab_class_t *mc)
 
 static void
 metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta,
-    int64_t defer_delta, int64_t space_delta, int64_t dspace_delta)
+    int64_t dalloc_delta, int64_t deferred_delta, int64_t ddeferred_delta,
+    int64_t space_delta, int64_t dspace_delta)
 {
 	atomic_add_64(&mc->mc_alloc, alloc_delta);
-	atomic_add_64(&mc->mc_deferred, defer_delta);
+	atomic_add_64(&mc->mc_dalloc, dalloc_delta);
+	atomic_add_64(&mc->mc_deferred, deferred_delta);
+	atomic_add_64(&mc->mc_ddeferred, ddeferred_delta);
 	atomic_add_64(&mc->mc_space, space_delta);
 	atomic_add_64(&mc->mc_dspace, dspace_delta);
 }
 
 uint64_t
 metaslab_class_get_alloc(metaslab_class_t *mc)
 {
-	return (mc->mc_alloc);
+	return (atomic_load_64(&mc->mc_alloc));
+}
+
+uint64_t
+metaslab_class_get_dalloc(metaslab_class_t *mc)
+{
+	return (spa_deflate(mc->mc_spa) ? atomic_load_64(&mc->mc_dalloc) :
+	    atomic_load_64(&mc->mc_alloc));
 }
 
 uint64_t
 metaslab_class_get_deferred(metaslab_class_t *mc)
 {
-	return (mc->mc_deferred);
+	return (spa_deflate(mc->mc_spa) ? atomic_load_64(&mc->mc_ddeferred) :
+	    atomic_load_64(&mc->mc_deferred));
 }
 
 uint64_t
 metaslab_class_get_space(metaslab_class_t *mc)
 {
-	return (mc->mc_space);
+	return (atomic_load_64(&mc->mc_space));
 }
 
 uint64_t
 metaslab_class_get_dspace(metaslab_class_t *mc)
 {
-	return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space);
+	return (spa_deflate(mc->mc_spa) ? atomic_load_64(&mc->mc_dspace) :
+	    atomic_load_64(&mc->mc_space));
 }
 
 void
@@ -2631,16 +2645,21 @@ metaslab_set_selected_txg(metaslab_t *msp, uint64_t txg)
 }
 
 void
-metaslab_space_update(vdev_t *vd, metaslab_class_t *mc, int64_t alloc_delta,
+metaslab_space_update(metaslab_group_t *mg, int64_t alloc_delta,
     int64_t defer_delta, int64_t space_delta)
 {
+	vdev_t *vd = mg->mg_vd;
+	int64_t dalloc_delta = vdev_deflated_space(vd, alloc_delta);
+	int64_t ddefer_delta = vdev_deflated_space(vd, defer_delta);
+	int64_t dspace_delta = vdev_deflated_space(vd, space_delta);
+
 	vdev_space_update(vd, alloc_delta, defer_delta, space_delta);
 
 	ASSERT3P(vd->vdev_spa->spa_root_vdev, ==, vd->vdev_parent);
 	ASSERT(vd->vdev_ms_count != 0);
 
-	metaslab_class_space_update(mc, alloc_delta, defer_delta, space_delta,
-	    vdev_deflated_space(vd, space_delta));
+	metaslab_class_space_update(mg->mg_class, alloc_delta, dalloc_delta,
+	    defer_delta, ddefer_delta, space_delta, dspace_delta);
 }
 
 int
@@ -2738,8 +2757,7 @@ metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object,
 	 */
 	if (txg <= TXG_INITIAL) {
 		metaslab_sync_done(ms, 0);
-		metaslab_space_update(vd, mg->mg_class,
-		    metaslab_allocated_space(ms), 0, 0);
+		metaslab_space_update(mg, metaslab_allocated_space(ms), 0, 0);
 	}
 
 	if (txg != 0) {
@@ -2801,9 +2819,8 @@ metaslab_fini(metaslab_t *msp)
 	 * subtracted.
 	 */
 	if (!msp->ms_new) {
-		metaslab_space_update(vd, mg->mg_class,
-		    -metaslab_allocated_space(msp), 0, -msp->ms_size);
-
+		metaslab_space_update(mg, -metaslab_allocated_space(msp), 0,
+		    -msp->ms_size);
 	}
 	space_map_close(msp->ms_sm);
 	msp->ms_sm = NULL;
@@ -4290,7 +4307,7 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
 
 	if (msp->ms_new) {
 		/* this is a new metaslab, add its capacity to the vdev */
-		metaslab_space_update(vd, mg->mg_class, 0, 0, msp->ms_size);
+		metaslab_space_update(mg, 0, 0, msp->ms_size);
 
 		/* there should be no allocations nor frees at this point */
 		VERIFY0(msp->ms_allocated_this_txg);
@@ -4318,8 +4335,7 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
 	} else {
 		defer_delta -= range_tree_space(*defer_tree);
 	}
-	metaslab_space_update(vd, mg->mg_class, alloc_delta + defer_delta,
-	    defer_delta, 0);
+	metaslab_space_update(mg, alloc_delta + defer_delta, defer_delta, 0);
 
 	if (spa_syncing_log_sm(spa) == NULL) {
 		/*
@@ -5253,8 +5269,16 @@ metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
 			 */
 			if (mca->mca_aliquot == 0 && metaslab_bias_enabled) {
 				vdev_stat_t *vs = &vd->vdev_stat;
-				int64_t vs_free = vs->vs_space - vs->vs_alloc;
-				int64_t mc_free = mc->mc_space - mc->mc_alloc;
+				uint64_t vs_space =
+				    atomic_load_64(&vs->vs_space);
+				uint64_t mc_space =
+				    atomic_load_64(&mc->mc_space);
+				uint64_t mc_alloc =
+				    atomic_load_64(&mc->mc_alloc);
+				uint64_t vs_alloc =
+				    atomic_load_64(&vs->vs_alloc);
+				int64_t vs_free = vs_space - vs_alloc;
+				int64_t mc_free = mc_space - mc_alloc;
 				int64_t ratio;
 
 				/*

module/zfs/spa.c

@@ -6349,6 +6349,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 	spa->spa_removing_phys.sr_removing_vdev = -1;
 	spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
 	spa->spa_indirect_vdevs_loaded = B_TRUE;
+	spa->spa_deflate = (version >= SPA_VERSION_RAIDZ_DEFLATE);
 
 	/*
 	 * Create "The Godfather" zio to hold all async IOs
@@ -6478,7 +6479,6 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
 	/* Newly created pools with the right version are always deflated. */
 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
-		spa->spa_deflate = TRUE;
 		if (zap_add(spa->spa_meta_objset,
 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {

module/zfs/spa_log_spacemap.c

@@ -1253,10 +1253,9 @@ spa_ld_log_sm_data(spa_t *spa)
 		    range_tree_space(m->ms_unflushed_allocs) -
 		    range_tree_space(m->ms_unflushed_frees);
 
-		vdev_t *vd = m->ms_group->mg_vd;
-		metaslab_space_update(vd, m->ms_group->mg_class,
+		metaslab_space_update(m->ms_group,
 		    range_tree_space(m->ms_unflushed_allocs), 0, 0);
-		metaslab_space_update(vd, m->ms_group->mg_class,
+		metaslab_space_update(m->ms_group,
 		    -range_tree_space(m->ms_unflushed_frees), 0, 0);
 
 		ASSERT0(m->ms_weight & METASLAB_ACTIVE_MASK);

module/zfs/spa_misc.c

@@ -1899,6 +1899,11 @@ spa_update_dspace(spa_t *spa)
 		ASSERT3U(spa->spa_dspace, >=, spa->spa_nonallocating_dspace);
 		spa->spa_dspace -= spa->spa_nonallocating_dspace;
 	}
+	spa->spa_dspace = spa->spa_dspace +
+	    metaslab_class_get_dalloc(spa_special_class(spa)) +
+	    metaslab_class_get_dalloc(spa_dedup_class(spa)) +
+	    ddt_get_dedup_dspace(spa) +
+	    brt_get_dspace(spa);
 }
 
 /*