DOSE-768 zoa panicked at assertion failed: (left < right) during die_…

…mtbf stress test (openzfs#28)
sdimitro · Nov 25, 2021 · 7890c98 · 7890c98
1 parent 5929d24
commit 7890c98
Show file tree

Hide file tree

Showing 6 changed files with 62 additions and 10 deletions.
diff --git a/include/sys/zil.h b/include/sys/zil.h
@@ -454,6 +454,19 @@ typedef struct zil_stats {
 	 */
 	kstat_named_t zil_itx_metaslab_slog_count;
 	kstat_named_t zil_itx_metaslab_slog_bytes;
+
+	/*
+	 * Number of allocation failures to slog device when
+	 * using object-based pools.
+	 */
+	kstat_named_t zil_slog_alloc_failures;
+
+	/*
+	 * Object-based pools don't support sync write semantics unless
+	 * a slog exits. If a slog does not exist, then we skip the
+	 * zil_commit call. Keep a count of how often that occurs.
+	 */
+	kstat_named_t zil_skip_zil_commit;
 } zil_stats_t;
 
 extern zil_stats_t zil_stats;

diff --git a/lib/libzfs/libzfs_dataset.c b/lib/libzfs/libzfs_dataset.c
@@ -1492,6 +1492,21 @@ zfs_valid_proplist(libzfs_handle_t *hdl, zfs_type_t type, nvlist_t *nvl,
 			chosen_normal = (int)intval;
 			break;
 
+		case ZFS_PROP_LOGBIAS:
+			if (zpool_hdl != NULL &&
+			    zpool_is_object_based(zpool_hdl) &&
+			    intval == ZFS_LOGBIAS_THROUGHPUT) {
+				const char *valname;
+				verify(zfs_prop_index_to_string(prop,
+				    intval, &valname) == 0);
+				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
+				    "object based pools cannot set '%s' to "
+				    "'%s'"), propname, valname);
+				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
+				goto error;
+			}
+			break;
+
 		default:
 			break;
 		}

diff --git a/module/zfs/zfs_log.c b/module/zfs/zfs_log.c
@@ -551,10 +551,12 @@ zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 		return;
 	}
 
-	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
+	boolean_t object_based = spa_is_object_based(zilog->zl_spa);
+
+	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT && !object_based)
 		write_state = WR_INDIRECT;
 	else if (!spa_has_slogs(zilog->zl_spa) &&
-	    resid >= zfs_immediate_write_sz)
+	    resid >= zfs_immediate_write_sz && !object_based)
 		write_state = WR_INDIRECT;
 	else if (ioflag & (O_SYNC | O_DSYNC))
 		write_state = WR_COPIED;

diff --git a/module/zfs/zil.c b/module/zfs/zil.c
@@ -108,6 +108,8 @@ zil_stats_t zil_stats = {
 	{ "zil_itx_metaslab_normal_bytes",	KSTAT_DATA_UINT64 },
 	{ "zil_itx_metaslab_slog_count",	KSTAT_DATA_UINT64 },
 	{ "zil_itx_metaslab_slog_bytes",	KSTAT_DATA_UINT64 },
+	{ "zil_slog_alloc_failures",		KSTAT_DATA_UINT64 },
+	{ "zil_skip_zil_commit",		KSTAT_DATA_UINT64 },
 };
 
 static kstat_t *zil_ksp;
@@ -709,6 +711,8 @@ zil_create(zilog_t *zilog)
 
 		if (error == 0)
 			zil_init_log_chain(zilog, &blk);
+		else if (slog)
+			ZIL_STAT_BUMP(zil_slog_alloc_failures);
 	}
 
 	/*
@@ -1557,6 +1561,8 @@ zil_lwb_write_issue(zilog_t *zilog, lwb_t *lwb)
 	if (slog) {
 		ZIL_STAT_BUMP(zil_itx_metaslab_slog_count);
 		ZIL_STAT_INCR(zil_itx_metaslab_slog_bytes, lwb->lwb_nused);
+		if (error)
+			ZIL_STAT_BUMP(zil_slog_alloc_failures);
 	} else {
 		ZIL_STAT_BUMP(zil_itx_metaslab_normal_count);
 		ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused);
@@ -1753,6 +1759,7 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 				ZIL_STAT_INCR(zil_itx_needcopy_bytes, dnow);
 			} else {
 				ASSERT3S(itx->itx_wr_state, ==, WR_INDIRECT);
+				ASSERT(!spa_is_object_based(zilog->zl_spa));
 				dbuf = NULL;
 				ZIL_STAT_BUMP(zil_itx_indirect_count);
 				ZIL_STAT_INCR(zil_itx_indirect_bytes,
@@ -2956,10 +2963,14 @@ zil_commit(zilog_t *zilog, uint64_t foid)
 	 */
 	ASSERT3B(dmu_objset_is_snapshot(zilog->zl_os), ==, B_FALSE);
 
-	if (zilog->zl_sync == ZFS_SYNC_DISABLED ||
-	    (spa_is_object_based(zilog->zl_spa) &&
-	    !spa_has_slogs(zilog->zl_spa)))
+	if (zilog->zl_sync == ZFS_SYNC_DISABLED)
+		return;
+
+	if (spa_is_object_based(zilog->zl_spa) &&
+	    !spa_has_slogs(zilog->zl_spa)) {
+		ZIL_STAT_BUMP(zil_skip_zil_commit);
 		return;
+	}
 
 	if (!spa_writeable(zilog->zl_spa)) {
 		/*

diff --git a/module/zfs/zio.c b/module/zfs/zio.c
@@ -3741,15 +3741,23 @@ zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp,
 	int flags = METASLAB_FASTWRITE | METASLAB_ZIL;
 	int allocator = (uint_t)cityhash4(0, 0, 0,
 	    os->os_dsl_dataset->ds_object) % spa->spa_alloc_count;
+
+	/*
+	 * Object-based pools can only allocate zil block on the slog,
+	 * so if we failed to allocate from the slog we need to return
+	 * an error and not fallback to the normal class.
+	 */
+	boolean_t object_based = spa_is_object_based(spa);
 	error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1,
 	    txg, NULL, flags, &io_alloc_list, NULL, allocator);
-	*slog = (error == 0);
-	if (error != 0) {
+	*slog = (error == 0 || object_based);
+
+	if (error != 0 && !object_based) {
 		error = metaslab_alloc(spa, spa_embedded_log_class(spa), size,
 		    new_bp, 1, txg, NULL, flags,
 		    &io_alloc_list, NULL, allocator);
 	}
-	if (error != 0) {
+	if (error != 0 && !object_based) {
 		error = metaslab_alloc(spa, spa_normal_class(spa), size,
 		    new_bp, 1, txg, NULL, flags,
 		    &io_alloc_list, NULL, allocator);

diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c
@@ -580,10 +580,13 @@ zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
 	if (zil_replaying(zilog, tx))
 		return;
 
-	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
+	boolean_t object_based = spa_is_object_based(zilog->zl_spa);
+
+	if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT && !object_based)
 		write_state = WR_INDIRECT;
 	else if (!spa_has_slogs(zilog->zl_spa) &&
-	    size >= blocksize && blocksize > zvol_immediate_write_sz)
+	    size >= blocksize && blocksize > zvol_immediate_write_sz &&
+	    !object_based)
 		write_state = WR_INDIRECT;
 	else if (sync)
 		write_state = WR_COPIED;