Merge tag 'zfs-2.1.14' into truenas/zfs-2.1-release

ZFS Version 2.1.14

META

@@ -1,7 +1,7 @@
 Meta:          1
 Name:          zfs
 Branch:        1.0
-Version:       2.1.13
+Version:       2.1.14
 Release:       1
 Release-Tags:  relext
 License:       CDDL

copy-builtin

@@ -43,32 +43,8 @@ config ZFS
 	  If unsure, say N.
 EOF
 
-add_after()
-{
-	FILE="$1"
-	MARKER="$2"
-	NEW="$3"
-
-	while IFS='' read -r LINE
-	do
-		printf "%s\n" "$LINE"
-
-		if [ -n "$MARKER" ] && [ "$LINE" = "$MARKER" ]
-		then
-			printf "%s\n" "$NEW"
-			MARKER=''
-			if IFS='' read -r LINE
-			then
-				[ "$LINE" != "$NEW" ] && printf "%s\n" "$LINE"
-			fi
-		fi
-	done < "$FILE" > "$FILE.new"
-
-	mv "$FILE.new" "$FILE"
-}
-
-add_after "$KERNEL_DIR/fs/Kconfig" 'if BLOCK' 'source "fs/zfs/Kconfig"'
-add_after "$KERNEL_DIR/fs/Makefile" 'endif' 'obj-$(CONFIG_ZFS) += zfs/'
+sed -i '/source "fs\/ext2\/Kconfig\"/i\source "fs/zfs/Kconfig"' "$KERNEL_DIR/fs/Kconfig"
+echo 'obj-$(CONFIG_ZFS) += zfs/' >> "$KERNEL_DIR/fs/Makefile"
 
 echo "$0: done. now you can build the kernel with ZFS support." >&2
 echo "$0: make sure you enable ZFS support (CONFIG_ZFS) before building." >&2

module/zfs/dnode.c

@@ -1773,15 +1773,23 @@ dnode_try_claim(objset_t *os, uint64_t object, int slots)
 }
 
 /*
- * Checks if the dnode contains any uncommitted dirty records.
+ * Checks if the dnode itself is dirty, or is carrying any uncommitted records.
+ * It is important to check both conditions, as some operations (eg appending
+ * to a file) can dirty both as a single logical unit, but they are not synced
+ * out atomically, so checking one and not the other can result in an object
+ * appearing to be clean mid-way through a commit.
+ *
+ * Do not change this lightly! If you get it wrong, dmu_offset_next() can
+ * detect a hole where there is really data, leading to silent corruption.
  */
 boolean_t
 dnode_is_dirty(dnode_t *dn)
 {
 	mutex_enter(&dn->dn_mtx);
 
 	for (int i = 0; i < TXG_SIZE; i++) {
-		if (multilist_link_active(&dn->dn_dirty_link[i])) {
+		if (multilist_link_active(&dn->dn_dirty_link[i]) ||
+		    !list_is_empty(&dn->dn_dirty_records[i])) {
 			mutex_exit(&dn->dn_mtx);
 			return (B_TRUE);
 		}

module/zfs/vdev_trim.c

@@ -23,6 +23,7 @@
  * Copyright (c) 2016 by Delphix. All rights reserved.
  * Copyright (c) 2019 by Lawrence Livermore National Security, LLC.
  * Copyright (c) 2021 Hewlett Packard Enterprise Development LP
+ * Copyright 2023 RackTop Systems, Inc.
  */
 
 #include <sys/spa.h>
@@ -572,6 +573,7 @@ vdev_trim_ranges(trim_args_t *ta)
 	uint64_t extent_bytes_max = ta->trim_extent_bytes_max;
 	uint64_t extent_bytes_min = ta->trim_extent_bytes_min;
 	spa_t *spa = vd->vdev_spa;
+	int error = 0;
 
 	ta->trim_start_time = gethrtime();
 	ta->trim_bytes_done = 0;
@@ -591,19 +593,32 @@ vdev_trim_ranges(trim_args_t *ta)
 		uint64_t writes_required = ((size - 1) / extent_bytes_max) + 1;
 
 		for (uint64_t w = 0; w < writes_required; w++) {
-			int error;
-
 			error = vdev_trim_range(ta, VDEV_LABEL_START_SIZE +
 			    rs_get_start(rs, ta->trim_tree) +
 			    (w *extent_bytes_max), MIN(size -
 			    (w * extent_bytes_max), extent_bytes_max));
 			if (error != 0) {
-				return (error);
+				goto done;
 			}
 		}
 	}
 
-	return (0);
+done:
+	/*
+	 * Make sure all TRIMs for this metaslab have completed before
+	 * returning. TRIM zios have lower priority over regular or syncing
+	 * zios, so all TRIM zios for this metaslab must complete before the
+	 * metaslab is re-enabled. Otherwise it's possible write zios to
+	 * this metaslab could cut ahead of still queued TRIM zios for this
+	 * metaslab causing corruption if the ranges overlap.
+	 */
+	mutex_enter(&vd->vdev_trim_io_lock);
+	while (vd->vdev_trim_inflight[0] > 0) {
+		cv_wait(&vd->vdev_trim_io_cv, &vd->vdev_trim_io_lock);
+	}
+	mutex_exit(&vd->vdev_trim_io_lock);
+
+	return (error);
 }
 
 static void
@@ -922,11 +937,6 @@ vdev_trim_thread(void *arg)
 	}
 
 	spa_config_exit(spa, SCL_CONFIG, FTAG);
-	mutex_enter(&vd->vdev_trim_io_lock);
-	while (vd->vdev_trim_inflight[0] > 0) {
-		cv_wait(&vd->vdev_trim_io_cv, &vd->vdev_trim_io_lock);
-	}
-	mutex_exit(&vd->vdev_trim_io_lock);
 
 	range_tree_destroy(ta.trim_tree);