Add interface to interface spa_get_worst_case_min_alloc() function

Provide an interface to retrieve the lowest and highest minimum
allocation size for the normal allocation class.  This can be used
by external consumers of the DMU to estimate potential wasted
capacity when setting the recordsize for an object.

The new "min_alloc" and "max_alloc" keys are added to the pool
configuration and used by default_volblocksize() to warn when
an ineffecient block size is requested.  For older kmods which
don't yet include the new keys fallback to the previous logic.

Reviewed-by: Tony Hutter <[email protected]>
Reviewed-by: Alexander Motin <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes openzfs#17758

Author: behlendorf

cmd/zfs/zfs_main.c

@@ -930,19 +930,15 @@ zfs_do_clone(int argc, char **argv)
 }
 
 /*
- * Return a default volblocksize for the pool which always uses more than
- * half of the data sectors.  This primarily applies to dRAID which always
- * writes full stripe widths.
+ * Calculate the minimum allocation size based on the top-level vdevs.
  */
 static uint64_t
-default_volblocksize(zpool_handle_t *zhp, nvlist_t *props)
+calculate_volblocksize(nvlist_t *config)
 {
-	uint64_t volblocksize, asize = SPA_MINBLOCKSIZE;
+	uint64_t asize = SPA_MINBLOCKSIZE;
 	nvlist_t *tree, **vdevs;
 	uint_t nvdevs;
 
-	nvlist_t *config = zpool_get_config(zhp, NULL);
-
 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree) != 0 ||
 	    nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
 	    &vdevs, &nvdevs) != 0) {
@@ -973,6 +969,24 @@ default_volblocksize(zpool_handle_t *zhp, nvlist_t *props)
 		}
 	}
 
+	return (asize);
+}
+
+/*
+ * Return a default volblocksize for the pool which always uses more than
+ * half of the data sectors.  This primarily applies to dRAID which always
+ * writes full stripe widths.
+ */
+static uint64_t
+default_volblocksize(zpool_handle_t *zhp, nvlist_t *props)
+{
+	uint64_t volblocksize, asize = SPA_MINBLOCKSIZE;
+
+	nvlist_t *config = zpool_get_config(zhp, NULL);
+
+	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_MAX_ALLOC, &asize) != 0)
+		asize = calculate_volblocksize(config);
+
 	/*
 	 * Calculate the target volblocksize such that more than half
 	 * of the asize is used. The following table is for 4k sectors.

include/sys/fs/zfs.h

@@ -740,6 +740,8 @@ typedef struct zpool_load_policy {
 #define	ZPOOL_CONFIG_METASLAB_SHIFT	"metaslab_shift"
 #define	ZPOOL_CONFIG_ASHIFT		"ashift"
 #define	ZPOOL_CONFIG_ASIZE		"asize"
+#define	ZPOOL_CONFIG_MIN_ALLOC		"min_alloc"
+#define	ZPOOL_CONFIG_MAX_ALLOC		"max_alloc"
 #define	ZPOOL_CONFIG_DTL		"DTL"
 #define	ZPOOL_CONFIG_SCAN_STATS		"scan_stats"	/* not stored on disk */
 #define	ZPOOL_CONFIG_REMOVAL_STATS	"removal_stats"	/* not stored on disk */

include/sys/spa.h

@@ -1055,6 +1055,7 @@ extern pool_state_t spa_state(spa_t *spa);
 extern spa_load_state_t spa_load_state(spa_t *spa);
 extern uint64_t spa_freeze_txg(spa_t *spa);
 extern uint64_t spa_get_worst_case_asize(spa_t *spa, uint64_t lsize);
+extern void spa_get_min_alloc_range(spa_t *spa, uint64_t *min, uint64_t *max);
 extern uint64_t spa_get_dspace(spa_t *spa);
 extern uint64_t spa_get_checkpoint_space(spa_t *spa);
 extern uint64_t spa_get_slop_space(spa_t *spa);

include/sys/spa_impl.h

@@ -267,6 +267,7 @@ struct spa {
 	uint64_t	spa_min_ashift;		/* of vdevs in normal class */
 	uint64_t	spa_max_ashift;		/* of vdevs in normal class */
 	uint64_t	spa_min_alloc;		/* of vdevs in normal class */
+	uint64_t	spa_max_alloc;		/* of vdevs in normal class */
 	uint64_t	spa_gcd_alloc;		/* of vdevs in normal class */
 	uint64_t	spa_config_guid;	/* config pool guid */
 	uint64_t	spa_load_guid;		/* spa_load initialized guid */

module/zfs/spa_config.c

@@ -461,6 +461,8 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
 	fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_MIN_ALLOC, spa->spa_min_alloc);
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_MAX_ALLOC, spa->spa_max_alloc);
 	if (spa->spa_comment != NULL)
 		fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
 		    spa->spa_comment);

module/zfs/spa_misc.c

@@ -814,6 +814,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 	spa->spa_min_ashift = INT_MAX;
 	spa->spa_max_ashift = 0;
 	spa->spa_min_alloc = INT_MAX;
+	spa->spa_max_alloc = 0;
 	spa->spa_gcd_alloc = INT_MAX;
 
 	/* Reset cached value */
@@ -1876,6 +1877,19 @@ spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
 	return (MAX(lsize, 1 << spa->spa_max_ashift) * spa_asize_inflation);
 }
 
+/*
+ * Return the range of minimum allocation sizes for the normal allocation
+ * class. This can be used by external consumers of the DMU to estimate
+ * potential wasted capacity when setting the recordsize for an object.
+ * This is mainly for dRAID pools which always pad to a full stripe width.
+ */
+void
+spa_get_min_alloc_range(spa_t *spa, uint64_t *min_alloc, uint64_t *max_alloc)
+{
+	*min_alloc = spa->spa_min_alloc;
+	*max_alloc = spa->spa_max_alloc;
+}
+
 /*
  * Return the amount of slop space in bytes.  It is typically 1/32 of the pool
  * (3.2%), minus the embedded log space.  On very small pools, it may be
@@ -3083,6 +3097,7 @@ EXPORT_SYMBOL(spa_version);
 EXPORT_SYMBOL(spa_state);
 EXPORT_SYMBOL(spa_load_state);
 EXPORT_SYMBOL(spa_freeze_txg);
+EXPORT_SYMBOL(spa_get_min_alloc_range); /* for Lustre */
 EXPORT_SYMBOL(spa_get_dspace);
 EXPORT_SYMBOL(spa_update_dspace);
 EXPORT_SYMBOL(spa_deflate);

module/zfs/vdev.c

@@ -1474,12 +1474,14 @@ vdev_spa_set_alloc(spa_t *spa, uint64_t min_alloc)
 {
 	if (min_alloc < spa->spa_min_alloc)
 		spa->spa_min_alloc = min_alloc;
-	if (spa->spa_gcd_alloc == INT_MAX) {
+
+	if (min_alloc > spa->spa_max_alloc)
+		spa->spa_max_alloc = min_alloc;
+
+	if (spa->spa_gcd_alloc == INT_MAX)
 		spa->spa_gcd_alloc = min_alloc;
-	} else {
-		spa->spa_gcd_alloc = vdev_gcd(min_alloc,
-		    spa->spa_gcd_alloc);
-	}
+	else
+		spa->spa_gcd_alloc = vdev_gcd(min_alloc, spa->spa_gcd_alloc);
 }
 
 void
@@ -1533,8 +1535,7 @@ vdev_metaslab_group_create(vdev_t *vd)
 			if (vd->vdev_ashift < spa->spa_min_ashift)
 				spa->spa_min_ashift = vd->vdev_ashift;
 
-			uint64_t min_alloc = vdev_get_min_alloc(vd);
-			vdev_spa_set_alloc(spa, min_alloc);
+			vdev_spa_set_alloc(spa, vdev_get_min_alloc(vd));
 		}
 	}
 }

module/zfs/vdev_label.c

@@ -511,6 +511,8 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
 		    vd->vdev_asize);
+		fnvlist_add_uint64(nv, ZPOOL_CONFIG_MIN_ALLOC,
+		    vdev_get_min_alloc(vd));
 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
 		if (vd->vdev_noalloc) {
 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,