Make ganging redundancy respect redundant_metadata property

The redundant_metadata setting in ZFS allows users to trade resilience
for performance and space savings. This applies to all data and metadata
blocks in zfs, with one exception: gang blocks. Gang blocks currently
just take the copies property of the IO being ganged and, if it's 1,
sets it to 2. This means that we always make at least two copies of a
gang header, which is good for resilience. However, if the users care
more about performance than resilience, their gang blocks will be even
more of a penalty than usual.

We add logic to calculate the number of gang headers copies directly,
and store it as a separate IO property. This is stored in the IO
properties and not calculated when we decide to gang because by that
point we may not have easy access to the relevant information about what
kind of block is being stored. We also check the redundant_metadata
property when doing so, and use that to decide whether to store an extra
copy of the gang headers, compared to the underlying blocks.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Signed-off-by: Paul Dagnelie <[email protected]>

Author: Paul Dagnelie

cmd/zdb/zdb.c

@@ -2544,12 +2544,14 @@ snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
 
 	blkbuf[0] = '\0';
 
-	for (i = 0; i < ndvas; i++)
+	for (i = 0; i < ndvas; i++) {
 		(void) snprintf(blkbuf + strlen(blkbuf),
-		    buflen - strlen(blkbuf), "%llu:%llx:%llx ",
+		    buflen - strlen(blkbuf), "%llu:%llx:%llx%s ",
 		    (u_longlong_t)DVA_GET_VDEV(&dva[i]),
 		    (u_longlong_t)DVA_GET_OFFSET(&dva[i]),
-		    (u_longlong_t)DVA_GET_ASIZE(&dva[i]));
+		    (u_longlong_t)DVA_GET_ASIZE(&dva[i]),
+		    (DVA_GET_GANG(&dva[i]) ? "G" : ""));
+	}
 
 	if (BP_IS_HOLE(bp)) {
 		(void) snprintf(blkbuf + strlen(blkbuf),
@@ -8980,7 +8982,7 @@ zdb_read_block(char *thing, spa_t *spa)
 
 	DVA_SET_VDEV(&dva[0], vd->vdev_id);
 	DVA_SET_OFFSET(&dva[0], offset);
-	DVA_SET_GANG(&dva[0], !!(flags & ZDB_FLAG_GBH));
+	DVA_SET_GANG(&dva[0], 0);
 	DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
 
 	BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
@@ -8995,7 +8997,7 @@ zdb_read_block(char *thing, spa_t *spa)
 	BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
 
 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
-	zio = zio_root(spa, NULL, NULL, 0);
+	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
 
 	if (vd == vd->vdev_top) {
 		/*

include/sys/dbuf.h

@@ -173,6 +173,7 @@ typedef struct dbuf_dirty_record {
 			arc_buf_t *dr_data;
 			override_states_t dr_override_state;
 			uint8_t dr_copies;
+			uint8_t dr_gang_copies;
 			boolean_t dr_nopwrite;
 			boolean_t dr_brtwrite;
 			boolean_t dr_diowrite;

include/sys/zio.h

@@ -349,6 +349,7 @@ typedef struct zio_prop {
 	uint8_t			zp_complevel;
 	uint8_t			zp_level;
 	uint8_t			zp_copies;
+	uint8_t			zp_gang_copies;
 	dmu_object_type_t	zp_type;
 	boolean_t		zp_dedup;
 	boolean_t		zp_dedup_verify;
@@ -574,7 +575,7 @@ extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
     zio_priority_t priority, zio_flag_t flags, zbookmark_phys_t *zb);
 
 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
-    boolean_t nopwrite, boolean_t brtwrite);
+    int gang_copies, boolean_t nopwrite, boolean_t brtwrite);
 
 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 

module/zfs/arc.c

@@ -6907,6 +6907,8 @@ arc_write(zio_t *pio, spa_t *spa, uint64_t txg,
 			localprop.zp_nopwrite = B_FALSE;
 			localprop.zp_copies =
 			    MIN(localprop.zp_copies, SPA_DVAS_PER_BP - 1);
+			localprop.zp_gang_copies =
+			    MIN(localprop.zp_gang_copies, SPA_DVAS_PER_BP - 1);
 		}
 		zio_flags |= ZIO_FLAG_RAW;
 	} else if (ARC_BUF_COMPRESSED(buf)) {

module/zfs/dbuf.c

@@ -5351,8 +5351,8 @@ dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx)
 		mutex_enter(&db->db_mtx);
 		dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
 		zio_write_override(dr->dr_zio, &dr->dt.dl.dr_overridden_by,
-		    dr->dt.dl.dr_copies, dr->dt.dl.dr_nopwrite,
-		    dr->dt.dl.dr_brtwrite);
+		    dr->dt.dl.dr_copies, dr->dt.dl.dr_gang_copies,
+		    dr->dt.dl.dr_nopwrite, dr->dt.dl.dr_brtwrite);
 		mutex_exit(&db->db_mtx);
 	} else if (data == NULL) {
 		ASSERT(zp.zp_checksum == ZIO_CHECKSUM_OFF ||

module/zfs/dmu.c

@@ -1915,6 +1915,7 @@ dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
 		dr->dt.dl.dr_overridden_by = *zio->io_bp;
 		dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
 		dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
+		dr->dt.dl.dr_gang_copies = zio->io_prop.zp_gang_copies;
 
 		/*
 		 * Old style holes are filled with all zeros, whereas
@@ -2321,6 +2322,7 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 	boolean_t dedup_verify = os->os_dedup_verify;
 	boolean_t encrypt = B_FALSE;
 	int copies = os->os_copies;
+	int gang_copies = os->os_copies;
 
 	/*
 	 * We maintain different write policies for each of the following
@@ -2353,15 +2355,24 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 		switch (os->os_redundant_metadata) {
 		case ZFS_REDUNDANT_METADATA_ALL:
 			copies++;
+			gang_copies++;
 			break;
 		case ZFS_REDUNDANT_METADATA_MOST:
 			if (level >= zfs_redundant_metadata_most_ditto_level ||
 			    DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))
 				copies++;
+			if (level + 1 >=
+			    zfs_redundant_metadata_most_ditto_level ||
+			    DMU_OT_IS_METADATA(type) || (wp & WP_SPILL))
+				gang_copies++;
 			break;
 		case ZFS_REDUNDANT_METADATA_SOME:
-			if (DMU_OT_IS_CRITICAL(type))
+			if (DMU_OT_IS_CRITICAL(type)) {
 				copies++;
+				gang_copies++;
+			} else if (DMU_OT_IS_METADATA(type)) {
+				gang_copies++;
+			}
 			break;
 		case ZFS_REDUNDANT_METADATA_NONE:
 			break;
@@ -2435,6 +2446,12 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 		nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags &
 		    ZCHECKSUM_FLAG_NOPWRITE) &&
 		    compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled);
+
+		if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL ||
+		    (os->os_redundant_metadata ==
+		    ZFS_REDUNDANT_METADATA_MOST &&
+		    zfs_redundant_metadata_most_ditto_level <= 1))
+			gang_copies++;
 	}
 
 	/*
@@ -2451,6 +2468,7 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 
 		if (DMU_OT_IS_ENCRYPTED(type)) {
 			copies = MIN(copies, SPA_DVAS_PER_BP - 1);
+			gang_copies = MIN(gang_copies, SPA_DVAS_PER_BP - 1);
 			nopwrite = B_FALSE;
 		} else {
 			dedup = B_FALSE;
@@ -2468,6 +2486,7 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 	zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type;
 	zp->zp_level = level;
 	zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa));
+	zp->zp_gang_copies = MIN(gang_copies, spa_max_replication(os->os_spa));
 	zp->zp_dedup = dedup;
 	zp->zp_dedup_verify = dedup && dedup_verify;
 	zp->zp_nopwrite = nopwrite;

module/zfs/dmu_recv.c

@@ -2299,6 +2299,9 @@ flush_write_batch_impl(struct receive_writer_arg *rwa)
 					zp.zp_nopwrite = B_FALSE;
 					zp.zp_copies = MIN(zp.zp_copies,
 					    SPA_DVAS_PER_BP - 1);
+					zp.zp_gang_copies =
+					    MIN(zp.zp_gang_copies,
+					    SPA_DVAS_PER_BP - 1);
 				}
 				zio_flags |= ZIO_FLAG_RAW;
 			} else if (DRR_WRITE_COMPRESSED(drrw)) {

module/zfs/zio.c

@@ -1404,8 +1404,8 @@ zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, abd_t *data,
 }
 
 void
-zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite,
-    boolean_t brtwrite)
+zio_write_override(zio_t *zio, blkptr_t *bp, int copies, int gang_copies,
+    boolean_t nopwrite, boolean_t brtwrite)
 {
 	ASSERT(zio->io_type == ZIO_TYPE_WRITE);
 	ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
@@ -1422,6 +1422,7 @@ zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite,
 	zio->io_prop.zp_nopwrite = nopwrite;
 	zio->io_prop.zp_brtwrite = brtwrite;
 	zio->io_prop.zp_copies = copies;
+	zio->io_prop.zp_gang_copies = gang_copies;
 	zio->io_bp_override = bp;
 }
 
@@ -3130,15 +3131,13 @@ zio_write_gang_block(zio_t *pio, metaslab_class_t *mc)
 	boolean_t has_data = !(pio->io_flags & ZIO_FLAG_NODATA);
 
 	/*
-	 * If one copy was requested, store 2 copies of the GBH, so that we
-	 * can still traverse all the data (e.g. to free or scrub) even if a
-	 * block is damaged.  Note that we can't store 3 copies of the GBH in
-	 * all cases, e.g. with encryption, which uses DVA[2] for the IV+salt.
+	 * Store multiple copies of the GBH, so that we can still traverse
+	 * all the data (e.g. to free or scrub) even if a block is damaged.
+	 * This value respects the redundant_metadata property.
 	 */
-	int gbh_copies = copies;
-	if (gbh_copies == 1) {
-		gbh_copies = MIN(2, spa_max_replication(spa));
-	}
+	int gbh_copies = gio->io_prop.zp_gang_copies;
+	ASSERT3S(gbh_copies, >, 0);
+	ASSERT3S(gbh_copies, <=, SPA_DVAS_PER_BP);
 
 	ASSERT(ZIO_HAS_ALLOCATOR(pio));
 	int flags = METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER;
@@ -3158,6 +3157,7 @@ zio_write_gang_block(zio_t *pio, metaslab_class_t *mc)
 		 * since metaslab_class_throttle_reserve() always allows
 		 * additional reservations for gang blocks.
 		 */
+		ASSERT3U(gbh_copies, >=, copies);
 		VERIFY(metaslab_class_throttle_reserve(mc, gbh_copies - copies,
 		    pio->io_allocator, pio, flags));
 	}
@@ -3220,6 +3220,7 @@ zio_write_gang_block(zio_t *pio, metaslab_class_t *mc)
 		zp.zp_type = zp.zp_storage_type = DMU_OT_NONE;
 		zp.zp_level = 0;
 		zp.zp_copies = gio->io_prop.zp_copies;
+		zp.zp_gang_copies = gio->io_prop.zp_gang_copies;
 		zp.zp_dedup = B_FALSE;
 		zp.zp_dedup_verify = B_FALSE;
 		zp.zp_nopwrite = B_FALSE;
@@ -3934,7 +3935,7 @@ zio_ddt_write(zio_t *zio)
 	 * grow the DDT entry by to satisfy the request.
 	 */
 	zio_prop_t czp = *zp;
-	czp.zp_copies = need_dvas;
+	czp.zp_copies = czp.zp_gang_copies = need_dvas;
 	zio_t *cio = zio_write(zio, spa, txg, bp, zio->io_orig_abd,
 	    zio->io_orig_size, zio->io_orig_size, &czp,
 	    zio_ddt_child_write_ready, NULL,

tests/runfiles/common.run

@@ -723,6 +723,10 @@ tests = ['large_dnode_001_pos', 'large_dnode_003_pos', 'large_dnode_004_neg',
     'large_dnode_005_pos', 'large_dnode_007_neg', 'large_dnode_009_pos']
 tags = ['functional', 'features', 'large_dnode']
 
+[tests/functional/gang_blocks]
+tests = ['gang_blocks_redundant']
+tags = ['functional', 'gang_blocks']
+
 [tests/functional/grow]
 pre =
 post =

tests/zfs-tests/include/tunables.cfg

@@ -62,6 +62,7 @@ MAX_DATASET_NESTING		max_dataset_nesting		zfs_max_dataset_nesting
 MAX_MISSING_TVDS		max_missing_tvds		zfs_max_missing_tvds
 METASLAB_DEBUG_LOAD		metaslab.debug_load		metaslab_debug_load
 METASLAB_FORCE_GANGING		metaslab.force_ganging		metaslab_force_ganging
+METASLAB_FORCE_GANGING_PCT	metaslab.force_ganging_pct	metaslab_force_ganging_pct
 MULTIHOST_FAIL_INTERVALS	multihost.fail_intervals	zfs_multihost_fail_intervals
 MULTIHOST_HISTORY		multihost.history		zfs_multihost_history
 MULTIHOST_IMPORT_INTERVALS	multihost.import_intervals	zfs_multihost_import_intervals