Sync with upstream zfs-2.2-release branch

cmd/zdb/zdb.c

@@ -79,6 +79,7 @@
 #include <sys/dsl_crypt.h>
 #include <sys/dsl_scan.h>
 #include <sys/btree.h>
+#include <sys/brt.h>
 #include <zfs_comutil.h>
 #include <sys/zstd/zstd.h>
 
@@ -5342,12 +5343,20 @@ static const char *zdb_ot_extname[] = {
 #define	ZB_TOTAL	DN_MAX_LEVELS
 #define	SPA_MAX_FOR_16M	(SPA_MAXBLOCKSHIFT+1)
 
+typedef struct zdb_brt_entry {
+	dva_t		zbre_dva;
+	uint64_t	zbre_refcount;
+	avl_node_t	zbre_node;
+} zdb_brt_entry_t;
+
 typedef struct zdb_cb {
 	zdb_blkstats_t	zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
 	uint64_t	zcb_removing_size;
 	uint64_t	zcb_checkpoint_size;
 	uint64_t	zcb_dedup_asize;
 	uint64_t	zcb_dedup_blocks;
+	uint64_t	zcb_clone_asize;
+	uint64_t	zcb_clone_blocks;
 	uint64_t	zcb_psize_count[SPA_MAX_FOR_16M];
 	uint64_t	zcb_lsize_count[SPA_MAX_FOR_16M];
 	uint64_t	zcb_asize_count[SPA_MAX_FOR_16M];
@@ -5368,6 +5377,8 @@ typedef struct zdb_cb {
 	int		zcb_haderrors;
 	spa_t		*zcb_spa;
 	uint32_t	**zcb_vd_obsolete_counts;
+	avl_tree_t	zcb_brt;
+	boolean_t	zcb_brt_is_active;
 } zdb_cb_t;
 
 /* test if two DVA offsets from same vdev are within the same metaslab */
@@ -5662,6 +5673,45 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
 	zcb->zcb_asize_len[bin] += BP_GET_ASIZE(bp);
 	zcb->zcb_asize_total += BP_GET_ASIZE(bp);
 
+	if (zcb->zcb_brt_is_active && brt_maybe_exists(zcb->zcb_spa, bp)) {
+		/*
+		 * Cloned blocks are special. We need to count them, so we can
+		 * later uncount them when reporting leaked space, and we must
+		 * only claim them them once.
+		 *
+		 * To do this, we keep our own in-memory BRT. For each block
+		 * we haven't seen before, we look it up in the real BRT and
+		 * if its there, we note it and its refcount then proceed as
+		 * normal. If we see the block again, we count it as a clone
+		 * and then give it no further consideration.
+		 */
+		zdb_brt_entry_t zbre_search, *zbre;
+		avl_index_t where;
+
+		zbre_search.zbre_dva = bp->blk_dva[0];
+		zbre = avl_find(&zcb->zcb_brt, &zbre_search, &where);
+		if (zbre != NULL) {
+			zcb->zcb_clone_asize += BP_GET_ASIZE(bp);
+			zcb->zcb_clone_blocks++;
+
+			zbre->zbre_refcount--;
+			if (zbre->zbre_refcount == 0) {
+				avl_remove(&zcb->zcb_brt, zbre);
+				umem_free(zbre, sizeof (zdb_brt_entry_t));
+			}
+			return;
+		}
+
+		uint64_t crefcnt = brt_entry_get_refcount(zcb->zcb_spa, bp);
+		if (crefcnt > 0) {
+			zbre = umem_zalloc(sizeof (zdb_brt_entry_t),
+			    UMEM_NOFAIL);
+			zbre->zbre_dva = bp->blk_dva[0];
+			zbre->zbre_refcount = crefcnt;
+			avl_insert(&zcb->zcb_brt, zbre, where);
+		}
+	}
+
 	if (dump_opt['L'])
 		return;
 
@@ -6664,6 +6714,20 @@ deleted_livelists_dump_mos(spa_t *spa)
 	iterate_deleted_livelists(spa, dump_livelist_cb, NULL);
 }
 
+static int
+zdb_brt_entry_compare(const void *zcn1, const void *zcn2)
+{
+	const dva_t *dva1 = &((const zdb_brt_entry_t *)zcn1)->zbre_dva;
+	const dva_t *dva2 = &((const zdb_brt_entry_t *)zcn2)->zbre_dva;
+	int cmp;
+
+	cmp = TREE_CMP(DVA_GET_VDEV(dva1), DVA_GET_VDEV(dva2));
+	if (cmp == 0)
+		cmp = TREE_CMP(DVA_GET_OFFSET(dva1), DVA_GET_OFFSET(dva2));
+
+	return (cmp);
+}
+
 static int
 dump_block_stats(spa_t *spa)
 {
@@ -6678,6 +6742,13 @@ dump_block_stats(spa_t *spa)
 
 	zcb = umem_zalloc(sizeof (zdb_cb_t), UMEM_NOFAIL);
 
+	if (spa_feature_is_active(spa, SPA_FEATURE_BLOCK_CLONING)) {
+		avl_create(&zcb->zcb_brt, zdb_brt_entry_compare,
+		    sizeof (zdb_brt_entry_t),
+		    offsetof(zdb_brt_entry_t, zbre_node));
+		zcb->zcb_brt_is_active = B_TRUE;
+	}
+
 	(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
 	    (dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
 	    (dump_opt['c'] == 1) ? "metadata " : "",
@@ -6779,7 +6850,8 @@ dump_block_stats(spa_t *spa)
 	    metaslab_class_get_alloc(spa_special_class(spa)) +
 	    metaslab_class_get_alloc(spa_dedup_class(spa)) +
 	    get_unflushed_alloc_space(spa);
-	total_found = tzb->zb_asize - zcb->zcb_dedup_asize +
+	total_found =
+	    tzb->zb_asize - zcb->zcb_dedup_asize - zcb->zcb_clone_asize +
 	    zcb->zcb_removing_size + zcb->zcb_checkpoint_size;
 
 	if (total_found == total_alloc && !dump_opt['L']) {
@@ -6820,6 +6892,9 @@ dump_block_stats(spa_t *spa)
 	    "bp deduped:", (u_longlong_t)zcb->zcb_dedup_asize,
 	    (u_longlong_t)zcb->zcb_dedup_blocks,
 	    (double)zcb->zcb_dedup_asize / tzb->zb_asize + 1.0);
+	(void) printf("\t%-16s %14llu    count: %6llu\n",
+	    "bp cloned:", (u_longlong_t)zcb->zcb_clone_asize,
+	    (u_longlong_t)zcb->zcb_clone_blocks);
 	(void) printf("\t%-16s %14llu     used: %5.2f%%\n", "Normal class:",
 	    (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
 

cmd/zed/zed.d/statechange-slot_off.sh

@@ -0,0 +1,61 @@
+#!/bin/sh
+#
+# Turn off disk's enclosure slot if it becomes FAULTED.
+#
+# Bad SCSI disks can often "disappear and reappear" causing all sorts of chaos
+# as they flip between FAULTED and ONLINE.  If
+# ZED_POWER_OFF_ENCLOUSRE_SLOT_ON_FAULT is set in zed.rc, and the disk gets
+# FAULTED, then power down the slot via sysfs:
+#
+# /sys/class/enclosure/<enclosure>/<slot>/power_status
+#
+# We assume the user will be responsible for turning the slot back on again.
+#
+# Note that this script requires that your enclosure be supported by the
+# Linux SCSI Enclosure services (SES) driver.  The script will do nothing
+# if you have no enclosure, or if your enclosure isn't supported.
+#
+# Exit codes:
+#   0: slot successfully powered off
+#   1: enclosure not available
+#   2: ZED_POWER_OFF_ENCLOUSRE_SLOT_ON_FAULT disabled
+#   3: vdev was not FAULTED
+#   4: The enclosure sysfs path passed from ZFS does not exist
+#   5: Enclosure slot didn't actually turn off after we told it to
+
+[ -f "${ZED_ZEDLET_DIR}/zed.rc" ] && . "${ZED_ZEDLET_DIR}/zed.rc"
+. "${ZED_ZEDLET_DIR}/zed-functions.sh"
+
+if [ ! -d /sys/class/enclosure ] ; then
+	# No JBOD enclosure or NVMe slots
+	exit 1
+fi
+
+if [ "${ZED_POWER_OFF_ENCLOUSRE_SLOT_ON_FAULT}" != "1" ] ; then
+	exit 2
+fi
+
+if [ "$ZEVENT_VDEV_STATE_STR" != "FAULTED" ] ; then
+	exit 3
+fi
+
+if [ ! -f "$ZEVENT_VDEV_ENC_SYSFS_PATH/power_status" ] ; then
+	exit 4
+fi
+
+echo "off" | tee "$ZEVENT_VDEV_ENC_SYSFS_PATH/power_status"
+
+# Wait for sysfs for report that the slot is off.  It can take ~400ms on some
+# enclosures.
+for i in $(seq 1 20) ; do
+	if [ "$(cat $ZEVENT_VDEV_ENC_SYSFS_PATH/power_status)" == "off" ] ; then
+		break
+	fi
+	sleep 0.1
+done
+
+if [ "$(cat $ZEVENT_VDEV_ENC_SYSFS_PATH/power_status)" != "off" ] ; then
+	exit 5
+fi
+
+zed_log_msg "powered down slot $ZEVENT_VDEV_ENC_SYSFS_PATH for $ZEVENT_VDEV_PATH"

cmd/zed/zed.d/zed.rc

@@ -142,3 +142,8 @@ ZED_SYSLOG_SUBCLASS_EXCLUDE="history_event"
 # Disabled by default, 1 to enable and 0 to disable.
 #ZED_SYSLOG_DISPLAY_GUIDS=1
 
+##
+# Power off the drive's slot in the enclosure if it becomes FAULTED.  This can
+# help silence misbehaving drives.  This assumes your drive enclosure fully
+# supports slot power control via sysfs.
+#ZED_POWER_OFF_ENCLOUSRE_SLOT_ON_FAULT=1

cmd/ztest.c

@@ -2412,7 +2412,6 @@ ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
 	int error;
 
 	ASSERT3P(lwb, !=, NULL);
-	ASSERT3P(zio, !=, NULL);
 	ASSERT3U(size, !=, 0);
 
 	ztest_object_lock(zd, object, RL_READER);
@@ -2446,6 +2445,7 @@ ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
 		    DMU_READ_NO_PREFETCH);
 		ASSERT0(error);
 	} else {
+		ASSERT3P(zio, !=, NULL);
 		size = doi.doi_data_block_size;
 		if (ISP2(size)) {
 			offset = P2ALIGN(offset, size);

contrib/dracut/90zfs/zfs-env-bootfs.service.in

@@ -12,11 +12,12 @@ ExecStart=/bin/sh -c '
     decode_root_args || exit 0;                                                                \
     [ "$root" = "zfs:AUTO" ] && root="$(@sbindir@/zpool list -H -o bootfs | grep -m1 -vFx -)"; \
     rootflags="$(getarg rootflags=)";                                                          \
-    case ",$rootflags," in                                                                     \
-        *,zfsutil,*) ;;                                                                        \
-        ,,) rootflags=zfsutil ;;                                                               \
-        *)  rootflags="zfsutil,$rootflags" ;;                                                  \
-    esac;                                                                                      \
+    [ "$(@sbindir@/zfs get -H -o value mountpoint "$root")" = legacy ] ||                      \
+        case ",$rootflags," in                                                                 \
+            *,zfsutil,*) ;;                                                                    \
+            ,,) rootflags=zfsutil ;;                                                           \
+            *)  rootflags="zfsutil,$rootflags" ;;                                              \
+        esac;                                                                                  \
     exec systemctl set-environment BOOTFS="$root" BOOTFSFLAGS="$rootflags"'
 
 [Install]

include/os/linux/spl/sys/kmem_cache.h

@@ -198,6 +198,14 @@ extern uint64_t spl_kmem_cache_entry_size(kmem_cache_t *cache);
     spl_kmem_cache_create(name, size, align, ctor, dtor, rclm, priv, vmp, fl)
 #define	kmem_cache_set_move(skc, move)	spl_kmem_cache_set_move(skc, move)
 #define	kmem_cache_destroy(skc)		spl_kmem_cache_destroy(skc)
+/*
+ * This is necessary to be compatible with other kernel modules
+ * or in-tree filesystem that may define kmem_cache_alloc,
+ * like bcachefs does it now.
+ */
+#ifdef kmem_cache_alloc
+#undef kmem_cache_alloc
+#endif
 #define	kmem_cache_alloc(skc, flags)	spl_kmem_cache_alloc(skc, flags)
 #define	kmem_cache_free(skc, obj)	spl_kmem_cache_free(skc, obj)
 #define	kmem_cache_reap_now(skc)	spl_kmem_cache_reap_now(skc)

include/sys/brt.h

@@ -36,6 +36,7 @@ extern "C" {
 #endif
 
 extern boolean_t brt_entry_decref(spa_t *spa, const blkptr_t *bp);
+extern uint64_t brt_entry_get_refcount(spa_t *spa, const blkptr_t *bp);
 
 extern uint64_t brt_get_dspace(spa_t *spa);
 extern uint64_t brt_get_used(spa_t *spa);

include/sys/dmu.h

@@ -572,11 +572,15 @@ int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
 int dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
     uint64_t length, int read, const void *tag, int *numbufsp,
     dmu_buf_t ***dbpp);
+int dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset,
+    const void *tag, dmu_buf_t **dbp);
 int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset,
     const void *tag, dmu_buf_t **dbp, int flags);
 int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset,
     uint64_t length, boolean_t read, const void *tag, int *numbufsp,
     dmu_buf_t ***dbpp, uint32_t flags);
+int dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, const void *tag,
+    dmu_buf_t **dbp);
 /*
  * Add a reference to a dmu buffer that has already been held via
  * dmu_buf_hold() in the current context.

include/sys/dmu_impl.h

@@ -247,8 +247,6 @@ typedef struct dmu_sendstatus {
 
 void dmu_object_zapify(objset_t *, uint64_t, dmu_object_type_t, dmu_tx_t *);
 void dmu_object_free_zapified(objset_t *, uint64_t, dmu_tx_t *);
-int dmu_buf_hold_noread(objset_t *, uint64_t, uint64_t,
-    const void *, dmu_buf_t **);
 
 #ifdef	__cplusplus
 }

include/sys/metaslab.h

@@ -80,7 +80,6 @@ uint64_t metaslab_largest_allocatable(metaslab_t *);
 #define	METASLAB_ASYNC_ALLOC		0x8
 #define	METASLAB_DONT_THROTTLE		0x10
 #define	METASLAB_MUST_RESERVE		0x20
-#define	METASLAB_FASTWRITE		0x40
 #define	METASLAB_ZIL			0x80
 
 int metaslab_alloc(spa_t *, metaslab_class_t *, uint64_t,
@@ -96,8 +95,6 @@ void metaslab_unalloc_dva(spa_t *, const dva_t *, uint64_t);
 int metaslab_claim(spa_t *, const blkptr_t *, uint64_t);
 int metaslab_claim_impl(vdev_t *, uint64_t, uint64_t, uint64_t);
 void metaslab_check_free(spa_t *, const blkptr_t *);
-void metaslab_fastwrite_mark(spa_t *, const blkptr_t *);
-void metaslab_fastwrite_unmark(spa_t *, const blkptr_t *);
 
 void metaslab_stat_init(void);
 void metaslab_stat_fini(void);

include/sys/metaslab_impl.h

@@ -313,7 +313,7 @@ struct metaslab_group {
  * Each metaslab maintains a set of in-core trees to track metaslab
  * operations.  The in-core free tree (ms_allocatable) contains the list of
  * free segments which are eligible for allocation.  As blocks are
- * allocated, the allocated segment are removed from the ms_allocatable and
+ * allocated, the allocated segments are removed from the ms_allocatable and
  * added to a per txg allocation tree (ms_allocating).  As blocks are
  * freed, they are added to the free tree (ms_freeing).  These trees
  * allow us to process all allocations and frees in syncing context
@@ -366,9 +366,9 @@ struct metaslab_group {
 struct metaslab {
 	/*
 	 * This is the main lock of the metaslab and its purpose is to
-	 * coordinate our allocations and frees [e.g metaslab_block_alloc(),
+	 * coordinate our allocations and frees [e.g., metaslab_block_alloc(),
 	 * metaslab_free_concrete(), ..etc] with our various syncing
-	 * procedures [e.g. metaslab_sync(), metaslab_sync_done(), ..etc].
+	 * procedures [e.g., metaslab_sync(), metaslab_sync_done(), ..etc].
 	 *
 	 * The lock is also used during some miscellaneous operations like
 	 * using the metaslab's histogram for the metaslab group's histogram

include/sys/vdev_impl.h

@@ -266,7 +266,6 @@ struct vdev {
 	metaslab_group_t *vdev_mg;	/* metaslab group		*/
 	metaslab_group_t *vdev_log_mg;	/* embedded slog metaslab group	*/
 	metaslab_t	**vdev_ms;	/* metaslab array		*/
-	uint64_t	vdev_pending_fastwrite; /* allocated fastwrites */
 	txg_list_t	vdev_ms_list;	/* per-txg dirty metaslab lists	*/
 	txg_list_t	vdev_dtl_list;	/* per-txg dirty DTL lists	*/
 	txg_node_t	vdev_txg_node;	/* per-txg dirty vdev linkage	*/