Prefer embedded blocks to dedup

Since embedded blocks introduction 11 years ago, their writing was
blocked if dedup is enabled.  After searching through the modern
code I see no reason for this restriction to exist.  Same time
embedded blocks are dramatically cheaper.  Even regular write of
so small blocks would likely be cheaper than deduplication, even
if the last is successful, not mentioning otherwise.

Reviewed-by: Allan Jude <[email protected]>
Reviewed-by: Tony Hutter <[email protected]>
Signed-off-by:	Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.
Closes openzfs#17113

Author: amotin

module/zfs/zio.c

@@ -2005,8 +2005,7 @@ zio_write_compress(zio_t *zio)
 			compress = ZIO_COMPRESS_OFF;
 			if (cabd != NULL)
 				abd_free(cabd);
-		} else if (!zp->zp_dedup && !zp->zp_encrypt &&
-		    psize <= BPE_PAYLOAD_SIZE &&
+		} else if (psize <= BPE_PAYLOAD_SIZE && !zp->zp_encrypt &&
 		    zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) &&
 		    spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) {
 			void *cbuf = abd_borrow_buf_copy(cabd, lsize);

tests/zfs-tests/tests/functional/cli_root/zpool_prefetch/zpool_prefetch_001_pos.ksh

@@ -75,7 +75,7 @@ log_must zpool prefetch -t ddt $TESTPOOL
 # to generate a reasonable size DDT for testing purposes.
 
 DATASET=$TESTPOOL/ddt
-log_must zfs create -o dedup=on $DATASET
+log_must zfs create -o compression=off -o dedup=on $DATASET
 MNTPOINT=$(get_prop mountpoint $TESTPOOL/ddt)
 
 log_note "Generating dataset ..."

tests/zfs-tests/tests/functional/dedup/dedup_quota.ksh

@@ -79,7 +79,7 @@ function do_setup
 	log_must truncate -s 5G $VDEV_GENERAL
 	# Use 'xattr=sa' to prevent selinux xattrs influencing our accounting
 	log_must zpool create -o ashift=12 -f -O xattr=sa -m $MOUNTDIR $POOL $VDEV_GENERAL
-	log_must zfs set dedup=on $POOL
+	log_must zfs set compression=off dedup=on $POOL
 	log_must set_tunable32 TXG_TIMEOUT 600
 }