ZFS/MFV: Use cached feature info in spa_add_feature_stats()

commit 417104bdd3
Author: Ned Bass <bass6@llnl.gov>
Date:   Thu Feb 26 12:24:11 2015 -0800

    Use cached feature info in spa_add_feature_stats()

    Avoid issuing I/O to the pool when retrieving feature flags information.
    Trying to read the ZAPs from disk means that zpool clear would hang if
    the pool is suspended and recovery would require a reboot. To keep the
    feature stats resident in memory, we hang a cached nvlist off of the
    spa.  It is built up from disk the first time spa_add_feature_stats() is
    called, and refreshed thereafter using the cached feature reference
    counts. spa_add_feature_stats() gets called at pool import time so we
    can be sure the cached nvlist will be available if the pool is later
    suspended.

    Signed-off-by: Ned Bass <bass6@llnl.gov>
    Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
    Closes #3082

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c

@@ -4371,18 +4371,14 @@ spa_add_l2cache(spa_t *spa, nvlist_t *config)
 }
 
 static void
-spa_add_feature_stats(spa_t *spa, nvlist_t *config)
+spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
 {
-	nvlist_t *features;
 	zap_cursor_t zc;
 	zap_attribute_t za;
 
-	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
-	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
-
 	/* We may be unable to read features if pool is suspended. */
 	if (spa_suspended(spa))
-		goto out;
+		return;
 
 	if (spa->spa_feat_for_read_obj != 0) {
 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
@@ -4391,7 +4387,7 @@ spa_add_feature_stats(spa_t *spa, nvlist_t *config)
 		    zap_cursor_advance(&zc)) {
 			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
 			    za.za_num_integers == 1);
-			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
+			VERIFY0(nvlist_add_uint64(features, za.za_name,
 			    za.za_first_integer));
 		}
 		zap_cursor_fini(&zc);
@@ -4404,16 +4400,62 @@ spa_add_feature_stats(spa_t *spa, nvlist_t *config)
 		    zap_cursor_advance(&zc)) {
 			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
 			    za.za_num_integers == 1);
-			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
+			VERIFY0(nvlist_add_uint64(features, za.za_name,
 			    za.za_first_integer));
 		}
 		zap_cursor_fini(&zc);
 	}
+}
 
-out:
-	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
-	    features) == 0);
-	nvlist_free(features);
+static void
+spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
+{
+	int i;
+
+	for (i = 0; i < SPA_FEATURES; i++) {
+		zfeature_info_t feature = spa_feature_table[i];
+		uint64_t refcount;
+
+		if (feature_get_refcount(spa, &feature, &refcount) != 0)
+			continue;
+
+		VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
+	}
+}
+
+/*
+ * Store a list of pool features and their reference counts in the
+ * config.
+ *
+ * The first time this is called on a spa, allocate a new nvlist, fetch
+ * the pool features and reference counts from disk, then save the list
+ * in the spa. In subsequent calls on the same spa use the saved nvlist
+ * and refresh its values from the cached reference counts.  This
+ * ensures we don't block here on I/O on a suspended pool so 'zpool
+ * clear' can resume the pool.
+ */
+static void
+spa_add_feature_stats(spa_t *spa, nvlist_t *config)
+{
+	nvlist_t *features;
+
+	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
+
+	mutex_enter(&spa->spa_feat_stats_lock);
+	features = spa->spa_feat_stats;
+
+	if (features != NULL) {
+		spa_feature_stats_from_cache(spa, features);
+	} else {
+		VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
+		spa->spa_feat_stats = features;
+		spa_feature_stats_from_disk(spa, features);
+	}
+
+	VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
+	    features));
+
+	mutex_exit(&spa->spa_feat_stats_lock);
 }
 
 int

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c

@@ -708,6 +708,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
+	mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
 
 	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
 	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
@@ -881,6 +882,7 @@ spa_remove(spa_t *spa)
 
 	nvlist_free(spa->spa_label_features);
 	nvlist_free(spa->spa_load_info);
+	nvlist_free(spa->spa_feat_stats);
 	spa_config_set(spa, NULL);
 
 #ifdef illumos
@@ -922,6 +924,7 @@ spa_remove(spa_t *spa)
 	mutex_destroy(&spa->spa_scrub_lock);
 	mutex_destroy(&spa->spa_suspend_lock);
 	mutex_destroy(&spa->spa_vdev_top_lock);
+	mutex_destroy(&spa->spa_feat_stats_lock);
 
 	kmem_free(spa, sizeof (spa_t));
 }

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/spa_impl.h

@@ -360,6 +360,8 @@ struct spa {
 	uint64_t	spa_feat_for_read_obj;	/* required to read from pool */
 	uint64_t	spa_feat_desc_obj;	/* Feature descriptions */
 	uint64_t	spa_feat_enabled_txg_obj; /* Feature enabled txg */
+	kmutex_t	spa_feat_stats_lock;	/* protects spa_feat_stats */
+	nvlist_t	*spa_feat_stats;	/* Cache of enabled features */
 	/* cache feature refcounts */
 	uint64_t	spa_feat_refcount_cache[SPA_FEATURES];
 #ifdef illumos

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfeature.c

@@ -220,7 +220,7 @@ spa_features_check(spa_t *spa, boolean_t for_write,
  *
  * Note: well-designed features will not need to use this; they should
  * use spa_feature_is_enabled() and spa_feature_is_active() instead.
- * However, this is non-static for zdb and zhack.
+ * However, this is non-static for zdb, zhack, and spa_add_feature_stats().
  */
 int
 feature_get_refcount(spa_t *spa, zfeature_info_t *feature, uint64_t *res)