Add option none to zfs redundant_metadata property

Currently, additional/extra copies are created for metadata in
addition to the redundancy provided by the pool(mirror/raidz/draid),
due to this 2 times more space is utilized per inode and this decreases
the total number of inodes that can be created in the filesystem. By
setting redundant_metadata to none, no additional copies of metadata
are created, hence can reduce the space consumed by the additional
metadata copies and increase the total number of inodes that can be
created in the filesystem.

Reviewed-by: Dipak Ghosh <[email protected]>
Signed-off-by: Akash B <[email protected]>

include/sys/dsl_prop.h

@@ -106,6 +106,9 @@ int dsl_prop_set_string(const char *dsname, const char *propname,
     zprop_source_t source, const char *value);
 int dsl_prop_inherit(const char *dsname, const char *propname,
     zprop_source_t source);
+boolean_t dsl_prop_val_understood(zfs_prop_t prop, uint64_t value);
+void dsl_prop_set_iuv(objset_t *mos, uint64_t zapobj, const char *propname,
+    int intsz, int numints, const void *value, dmu_tx_t *tx);
 
 int dsl_prop_predict(dsl_dir_t *dd, const char *propname,
     zprop_source_t source, uint64_t value, uint64_t *newvalp);

include/sys/fs/zfs.h

@@ -501,7 +501,8 @@ typedef enum {
 
 typedef enum {
 	ZFS_REDUNDANT_METADATA_ALL,
-	ZFS_REDUNDANT_METADATA_MOST
+	ZFS_REDUNDANT_METADATA_MOST,
+	ZFS_REDUNDANT_METADATA_NONE
 } zfs_redundant_metadata_type_t;
 
 typedef enum {

man/man7/zfsprops.7

@@ -37,7 +37,7 @@
 .\" Copyright 2019 Joyent, Inc.
 .\" Copyright (c) 2019, Kjeld Schouten-Lebbing
 .\"
-.Dd May 24, 2021
+.Dd July 21, 2022
 .Dt ZFSPROPS 7
 .Os
 .
@@ -1454,7 +1454,7 @@ affects only files created afterward; existing files are unaffected.
 .Pp
 This property can also be referred to by its shortened column name,
 .Sy recsize .
-.It Sy redundant_metadata Ns = Ns Sy all Ns | Ns Sy most
+.It Sy redundant_metadata Ns = Ns Sy all Ns | Ns Sy most Ns | Ns Sy none
 Controls what types of metadata are stored redundantly.
 ZFS stores an extra copy of metadata, so that if a single block is corrupted,
 the amount of user data lost is limited.
@@ -1495,6 +1495,10 @@ of user data can be lost if a single on-disk block is corrupt.
 The exact behavior of which metadata blocks are stored redundantly may change in
 future releases.
 .Pp
+When set to
+.Sy none ,
+ZFS does not store any copies of metadata redundantly through this property.
+.Pp
 The default value is
 .Sy all .
 .It Sy refquota Ns = Ns Ar size Ns | Ns Sy none

module/zcommon/zfs_prop.c

@@ -369,6 +369,7 @@ zfs_prop_init(void)
 	static const zprop_index_t redundant_metadata_table[] = {
 		{ "all",	ZFS_REDUNDANT_METADATA_ALL },
 		{ "most",	ZFS_REDUNDANT_METADATA_MOST },
+		{ "none",	ZFS_REDUNDANT_METADATA_NONE },
 		{ NULL }
 	};
 
@@ -388,7 +389,7 @@ zfs_prop_init(void)
 	zprop_register_index(ZFS_PROP_REDUNDANT_METADATA, "redundant_metadata",
 	    ZFS_REDUNDANT_METADATA_ALL,
 	    PROP_INHERIT, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME,
-	    "all | most", "REDUND_MD",
+	    "all | most | none", "REDUND_MD",
 	    redundant_metadata_table, sfeatures);
 	zprop_register_index(ZFS_PROP_SYNC, "sync", ZFS_SYNC_STANDARD,
 	    PROP_INHERIT, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME,

module/zfs/dmu_objset.c

@@ -287,7 +287,8 @@ redundant_metadata_changed_cb(void *arg, uint64_t newval)
 	 * Inheritance and range checking should have been done by now.
 	 */
 	ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
-	    newval == ZFS_REDUNDANT_METADATA_MOST);
+	    newval == ZFS_REDUNDANT_METADATA_MOST ||
+	    newval == ZFS_REDUNDANT_METADATA_NONE);
 
 	os->os_redundant_metadata = newval;
 }

module/zfs/dsl_prop.c

@@ -41,6 +41,7 @@
 
 #define	ZPROP_INHERIT_SUFFIX "$inherit"
 #define	ZPROP_RECVD_SUFFIX "$recvd"
+#define	ZPROP_IUV_SUFFIX "$iuv"
 
 static int
 dodefault(zfs_prop_t prop, int intsz, int numints, void *buf)
@@ -81,6 +82,7 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
 	boolean_t inheriting = B_FALSE;
 	char *inheritstr;
 	char *recvdstr;
+	char *iuvstr;
 
 	ASSERT(dsl_pool_config_held(dd->dd_pool));
 
@@ -91,6 +93,7 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
 	inheritable = (prop == ZPROP_USERPROP || zfs_prop_inheritable(prop));
 	inheritstr = kmem_asprintf("%s%s", propname, ZPROP_INHERIT_SUFFIX);
 	recvdstr = kmem_asprintf("%s%s", propname, ZPROP_RECVD_SUFFIX);
+	iuvstr = kmem_asprintf("%s%s", propname, ZPROP_IUV_SUFFIX);
 
 	/*
 	 * Note: dd may become NULL, therefore we shouldn't dereference it
@@ -105,6 +108,18 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
 			inheriting = B_TRUE;
 		}
 
+		/* Check for a IUV value. */
+		err = zap_lookup(mos, dsl_dir_phys(dd)->dd_props_zapobj,
+		    iuvstr, intsz, numints, buf);
+		if (!dsl_prop_val_understood(zfs_name_to_prop(propname),
+		    *(uint64_t *)buf))
+			err = ENOENT;
+		if (err != ENOENT) {
+			if (setpoint != NULL && err == 0)
+				dsl_dir_name(dd, setpoint);
+			break;
+		}
+
 		/* Check for a local value. */
 		err = zap_lookup(mos, dsl_dir_phys(dd)->dd_props_zapobj,
 		    propname, intsz, numints, buf);
@@ -155,6 +170,7 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
 
 	kmem_strfree(inheritstr);
 	kmem_strfree(recvdstr);
+	kmem_strfree(iuvstr);
 
 	return (err);
 }
@@ -659,6 +675,7 @@ dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
 	const char *valstr = NULL;
 	char *inheritstr;
 	char *recvdstr;
+	char *iuvstr;
 	char *tbuf = NULL;
 	int err;
 	uint64_t version = spa_version(ds->ds_dir->dd_pool->dp_spa);
@@ -692,6 +709,7 @@ dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
 
 	inheritstr = kmem_asprintf("%s%s", propname, ZPROP_INHERIT_SUFFIX);
 	recvdstr = kmem_asprintf("%s%s", propname, ZPROP_RECVD_SUFFIX);
+	iuvstr = kmem_asprintf("%s%s", propname, ZPROP_IUV_SUFFIX);
 
 	switch ((int)source) {
 	case ZPROP_SRC_NONE:
@@ -709,11 +727,14 @@ dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
 		/*
 		 * remove propname$inherit
 		 * set propname -> value
+		 * set propname$iuv -> new property value
 		 */
 		err = zap_remove(mos, zapobj, inheritstr, tx);
 		ASSERT(err == 0 || err == ENOENT);
 		VERIFY0(zap_update(mos, zapobj, propname,
 		    intsz, numints, value, tx));
+		(void) dsl_prop_set_iuv(mos, zapobj, propname, intsz,
+		    numints, value, tx);
 		break;
 	case ZPROP_SRC_INHERITED:
 		/*
@@ -763,6 +784,7 @@ dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
 
 	kmem_strfree(inheritstr);
 	kmem_strfree(recvdstr);
+	kmem_strfree(iuvstr);
 
 	/*
 	 * If we are left with an empty snap zap we can destroy it.
@@ -860,6 +882,49 @@ dsl_prop_inherit(const char *dsname, const char *propname,
 	return (error);
 }
 
+boolean_t
+dsl_prop_val_understood(zfs_prop_t prop, uint64_t value)
+{
+	const char *str;
+	if (zfs_prop_get_type(prop) == PROP_TYPE_INDEX) {
+		if (!zfs_prop_index_to_string(prop, value, &str))
+			return 1;
+		else
+			return 0;
+		} else {
+			return 1;
+		}
+}
+
+void
+dsl_prop_set_iuv(objset_t *mos, uint64_t zapobj, const char *propname,
+    int intsz, int numints, const void *value, dmu_tx_t *tx)
+{
+	char *iuvstr = kmem_asprintf("%s%s", propname, ZPROP_IUV_SUFFIX);
+	uint64_t val;
+	int iuv_flag = 0;
+	zfs_prop_t prop = zfs_name_to_prop(propname);
+	switch (prop) {
+	case ZFS_PROP_REDUNDANT_METADATA:
+		if (*(uint64_t *)value >= ZFS_REDUNDANT_METADATA_NONE)
+			iuv_flag = 1;
+                break;
+        default:
+                break;
+        }
+
+	if (iuv_flag) {
+		VERIFY0(zap_update(mos, zapobj, iuvstr, intsz, numints,
+		    value, tx));
+		val = zfs_prop_default_numeric(prop);
+		VERIFY0(zap_update(mos, zapobj, propname, intsz, numints,
+		    &val, tx));
+	} else if (0 == zap_contains(mos, zapobj, iuvstr)) {
+		zap_remove(mos, zapobj, iuvstr, tx);
+	}
+	kmem_strfree(iuvstr);
+}
+
 int
 dsl_props_set_check(void *arg, dmu_tx_t *tx)
 {
@@ -1044,6 +1109,15 @@ dsl_prop_get_all_impl(objset_t *mos, uint64_t propobj,
 
 			source = ((flags & DSL_PROP_GET_INHERITING) ?
 			    setpoint : ZPROP_SOURCE_VAL_RECVD);
+		} else if (strcmp(suffix, ZPROP_IUV_SUFFIX) == 0) {
+			(void) strncpy(buf, za.za_name, (suffix - za.za_name));
+			buf[suffix - za.za_name] = '\0';
+			propname = buf;
+			source = setpoint;
+			prop = zfs_name_to_prop(propname);
+			if (!dsl_prop_val_understood(prop,
+			    za.za_first_integer))
+				continue;
 		} else {
 			/*
 			 * For backward compatibility, skip suffixes we don't
@@ -1064,6 +1138,10 @@ dsl_prop_get_all_impl(objset_t *mos, uint64_t propobj,
 		    !zfs_prop_valid_for_type(prop, ZFS_TYPE_SNAPSHOT, B_FALSE))
 			continue;
 
+		/* We only want propname$iuv for new property values */
+		if (suffix && (strcmp(suffix, ZPROP_IUV_SUFFIX) == 0))
+			nvlist_remove_all(nv, propname);
+
 		/* Skip properties already defined. */
 		if (nvlist_exists(nv, propname))
 			continue;

tests/zfs-tests/include/properties.shlib

@@ -27,7 +27,7 @@ typeset -a canmount_prop_vals=('on' 'off' 'noauto')
 typeset -a copies_prop_vals=('1' '2' '3')
 typeset -a logbias_prop_vals=('latency' 'throughput')
 typeset -a primarycache_prop_vals=('all' 'none' 'metadata')
-typeset -a redundant_metadata_prop_vals=('all' 'most')
+typeset -a redundant_metadata_prop_vals=('all' 'most' 'none')
 typeset -a secondarycache_prop_vals=('all' 'none' 'metadata')
 typeset -a snapdir_prop_vals=('hidden' 'visible')
 typeset -a sync_prop_vals=('standard' 'always' 'disabled')