[illumos-gate merge]

commit 2ec7644aab2a726a64681fa66c6db8731b160de1
    9580 Add a hash-table on top of nvlist to speed-up operations
commit abe1fd01ce5a83718c5a840daeab4abdaec1c104
    9465 ARC check for 'anon_size > arc_c/2' can stall the system

10 files changed, 392 insertions, 72 deletions

diff --git a/usr/src/common/nvpair/nvpair.c b/usr/src/common/nvpair/nvpair.c
index 5881ba54b7..bffae68ca3 100644
--- a/usr/src/common/nvpair/nvpair.c
+++ b/usr/src/common/nvpair/nvpair.c
@@ -144,6 +144,8 @@ int nvpair_max_recursion = 20;
 int nvpair_max_recursion = 100;
 #endif
 
+uint64_t nvlist_hashtable_init_size = (1 << 4);
+
 int
 nv_alloc_init(nv_alloc_t *nva, const nv_alloc_ops_t *nvo, /* args */ ...)
 {
@@ -251,6 +253,291 @@ nv_priv_alloc_embedded(nvpriv_t *priv)
 	return (emb_priv);
 }
 
+static int
+nvt_tab_alloc(nvpriv_t *priv, uint64_t buckets)
+{
+	ASSERT3P(priv->nvp_hashtable, ==, NULL);
+	ASSERT0(priv->nvp_nbuckets);
+	ASSERT0(priv->nvp_nentries);
+
+	i_nvp_t **tab = nv_mem_zalloc(priv, buckets * sizeof (i_nvp_t *));
+	if (tab == NULL)
+		return (ENOMEM);
+
+	priv->nvp_hashtable = tab;
+	priv->nvp_nbuckets = buckets;
+	return (0);
+}
+
+static void
+nvt_tab_free(nvpriv_t *priv)
+{
+	i_nvp_t **tab = priv->nvp_hashtable;
+	if (tab == NULL) {
+		ASSERT0(priv->nvp_nbuckets);
+		ASSERT0(priv->nvp_nentries);
+		return;
+	}
+
+	nv_mem_free(priv, tab, priv->nvp_nbuckets * sizeof (i_nvp_t *));
+
+	priv->nvp_hashtable = NULL;
+	priv->nvp_nbuckets = 0;
+	priv->nvp_nentries = 0;
+}
+
+static uint32_t
+nvt_hash(const char *p)
+{
+	uint32_t g, hval = 0;
+
+	while (*p) {
+		hval = (hval << 4) + *p++;
+		if ((g = (hval & 0xf0000000)) != 0)
+			hval ^= g >> 24;
+		hval &= ~g;
+	}
+	return (hval);
+}
+
+static boolean_t
+nvt_nvpair_match(nvpair_t *nvp1, nvpair_t *nvp2, uint32_t nvflag)
+{
+	boolean_t match = B_FALSE;
+	if (nvflag & NV_UNIQUE_NAME_TYPE) {
+		if (strcmp(NVP_NAME(nvp1), NVP_NAME(nvp2)) == 0 &&
+		    NVP_TYPE(nvp1) == NVP_TYPE(nvp2))
+			match = B_TRUE;
+	} else {
+		ASSERT(nvflag == 0 || nvflag & NV_UNIQUE_NAME);
+		if (strcmp(NVP_NAME(nvp1), NVP_NAME(nvp2)) == 0)
+			match = B_TRUE;
+	}
+	return (match);
+}
+
+static nvpair_t *
+nvt_lookup_name_type(nvlist_t *nvl, const char *name, data_type_t type)
+{
+	nvpriv_t *priv = (nvpriv_t *)(uintptr_t)nvl->nvl_priv;
+	ASSERT(priv != NULL);
+
+	i_nvp_t **tab = priv->nvp_hashtable;
+
+	if (tab == NULL) {
+		ASSERT3P(priv->nvp_list, ==, NULL);
+		ASSERT0(priv->nvp_nbuckets);
+		ASSERT0(priv->nvp_nentries);
+		return (NULL);
+	} else {
+		ASSERT(priv->nvp_nbuckets != 0);
+	}
+
+	uint64_t hash = nvt_hash(name);
+	uint64_t index = hash & (priv->nvp_nbuckets - 1);
+
+	ASSERT3U(index, <, priv->nvp_nbuckets);
+	i_nvp_t *entry = tab[index];
+
+	for (i_nvp_t *e = entry; e != NULL; e = e->nvi_hashtable_next) {
+		if (strcmp(NVP_NAME(&e->nvi_nvp), name) == 0 &&
+		    (type == DATA_TYPE_DONTCARE ||
+		    NVP_TYPE(&e->nvi_nvp) == type))
+			return (&e->nvi_nvp);
+	}
+	return (NULL);
+}
+
+static nvpair_t *
+nvt_lookup_name(nvlist_t *nvl, const char *name)
+{
+	return (nvt_lookup_name_type(nvl, name, DATA_TYPE_DONTCARE));
+}
+
+static int
+nvt_resize(nvpriv_t *priv, uint32_t new_size)
+{
+	i_nvp_t **tab = priv->nvp_hashtable;
+
+	/*
+	 * Migrate all the entries from the current table
+	 * to a newly-allocated table with the new size by
+	 * re-adjusting the pointers of their entries.
+	 */
+	uint32_t size = priv->nvp_nbuckets;
+	uint32_t new_mask = new_size - 1;
+	ASSERT(ISP2(new_size));
+
+	i_nvp_t **new_tab = nv_mem_zalloc(priv, new_size * sizeof (i_nvp_t *));
+	if (new_tab == NULL)
+		return (ENOMEM);
+
+	uint32_t nentries = 0;
+	for (uint32_t i = 0; i < size; i++) {
+		i_nvp_t *next, *e = tab[i];
+
+		while (e != NULL) {
+			next = e->nvi_hashtable_next;
+
+			uint32_t hash = nvt_hash(NVP_NAME(&e->nvi_nvp));
+			uint32_t index = hash & new_mask;
+
+			e->nvi_hashtable_next = new_tab[index];
+			new_tab[index] = e;
+			nentries++;
+
+			e = next;
+		}
+		tab[i] = NULL;
+	}
+	ASSERT3U(nentries, ==, priv->nvp_nentries);
+
+	nvt_tab_free(priv);
+
+	priv->nvp_hashtable = new_tab;
+	priv->nvp_nbuckets = new_size;
+	priv->nvp_nentries = nentries;
+
+	return (0);
+}
+
+static boolean_t
+nvt_needs_togrow(nvpriv_t *priv)
+{
+	/*
+	 * Grow only when we have more elements than buckets
+	 * and the # of buckets doesn't overflow.
+	 */
+	return (priv->nvp_nentries > priv->nvp_nbuckets &&
+	    (UINT32_MAX >> 1) >= priv->nvp_nbuckets);
+}
+
+/*
+ * Allocate a new table that's twice the size of the old one,
+ * and migrate all the entries from the old one to the new
+ * one by re-adjusting their pointers.
+ */
+static int
+nvt_grow(nvpriv_t *priv)
+{
+	uint32_t current_size = priv->nvp_nbuckets;
+	/* ensure we won't overflow */
+	ASSERT3U(UINT32_MAX >> 1, >=, current_size);
+	return (nvt_resize(priv, current_size << 1));
+}
+
+static boolean_t
+nvt_needs_toshrink(nvpriv_t *priv)
+{
+	/*
+	 * Shrink only when the # of elements is less than or
+	 * equal to 1/4 the # of buckets. Never shrink less than
+	 * nvlist_hashtable_init_size.
+	 */
+	ASSERT3U(priv->nvp_nbuckets, >=, nvlist_hashtable_init_size);
+	if (priv->nvp_nbuckets == nvlist_hashtable_init_size)
+		return (B_FALSE);
+	return (priv->nvp_nentries <= (priv->nvp_nbuckets >> 2));
+}
+
+/*
+ * Allocate a new table that's half the size of the old one,
+ * and migrate all the entries from the old one to the new
+ * one by re-adjusting their pointers.
+ */
+static int
+nvt_shrink(nvpriv_t *priv)
+{
+	uint32_t current_size = priv->nvp_nbuckets;
+	/* ensure we won't overflow */
+	ASSERT3U(current_size, >=, nvlist_hashtable_init_size);
+	return (nvt_resize(priv, current_size >> 1));
+}
+
+static int
+nvt_remove_nvpair(nvlist_t *nvl, nvpair_t *nvp)
+{
+	nvpriv_t *priv = (nvpriv_t *)(uintptr_t)nvl->nvl_priv;
+
+	if (nvt_needs_toshrink(priv)) {
+		int err = nvt_shrink(priv);
+		if (err != 0)
+			return (err);
+	}
+	i_nvp_t **tab = priv->nvp_hashtable;
+
+	char *name = NVP_NAME(nvp);
+	uint64_t hash = nvt_hash(name);
+	uint64_t index = hash & (priv->nvp_nbuckets - 1);
+
+	ASSERT3U(index, <, priv->nvp_nbuckets);
+	i_nvp_t *bucket = tab[index];
+
+	for (i_nvp_t *prev = NULL, *e = bucket;
+	    e != NULL; prev = e, e = e->nvi_hashtable_next) {
+		if (nvt_nvpair_match(&e->nvi_nvp, nvp, nvl->nvl_flag)) {
+			if (prev != NULL) {
+				prev->nvi_hashtable_next =
+				    e->nvi_hashtable_next;
+			} else {
+				ASSERT3P(e, ==, bucket);
+				tab[index] = e->nvi_hashtable_next;
+			}
+			e->nvi_hashtable_next = NULL;
+			priv->nvp_nentries--;
+			break;
+		}
+	}
+
+	return (0);
+}
+
+static int
+nvt_add_nvpair(nvlist_t *nvl, nvpair_t *nvp)
+{
+	nvpriv_t *priv = (nvpriv_t *)(uintptr_t)nvl->nvl_priv;
+
+	/* initialize nvpair table now if it doesn't exist. */
+	if (priv->nvp_hashtable == NULL) {
+		int err = nvt_tab_alloc(priv, nvlist_hashtable_init_size);
+		if (err != 0)
+			return (err);
+	}
+
+	/*
+	 * if we don't allow duplicate entries, make sure to
+	 * unlink any existing entries from the table.
+	 */
+	if (nvl->nvl_nvflag != 0) {
+		int err = nvt_remove_nvpair(nvl, nvp);
+		if (err != 0)
+			return (err);
+	}
+
+	if (nvt_needs_togrow(priv)) {
+		int err = nvt_grow(priv);
+		if (err != 0)
+			return (err);
+	}
+	i_nvp_t **tab = priv->nvp_hashtable;
+
+	char *name = NVP_NAME(nvp);
+	uint64_t hash = nvt_hash(name);
+	uint64_t index = hash & (priv->nvp_nbuckets - 1);
+
+	ASSERT3U(index, <, priv->nvp_nbuckets);
+	i_nvp_t *bucket = tab[index];
+
+	/* insert link at the beginning of the bucket */
+	i_nvp_t *new_entry = NVPAIR2I_NVP(nvp);
+	ASSERT3P(new_entry->nvi_hashtable_next, ==, NULL);
+	new_entry->nvi_hashtable_next = bucket;
+	tab[index] = new_entry;
+
+	priv->nvp_nentries++;
+	return (0);
+}
+
 static void
 nvlist_init(nvlist_t *nvl, uint32_t nvflag, nvpriv_t *priv)
 {
@@ -583,6 +870,7 @@ nvlist_free(nvlist_t *nvl)
 	else
 		nvl->nvl_priv = 0;
 
+	nvt_tab_free(priv);
 	nv_mem_free(priv, priv, sizeof (nvpriv_t));
 }
 
@@ -643,26 +931,14 @@ nvlist_xdup(nvlist_t *nvl, nvlist_t **nvlp, nv_alloc_t *nva)
 int
 nvlist_remove_all(nvlist_t *nvl, const char *name)
 {
-	nvpriv_t *priv;
-	i_nvp_t *curr;
 	int error = ENOENT;
 
-	if (nvl == NULL || name == NULL ||
-	    (priv = (nvpriv_t *)(uintptr_t)nvl->nvl_priv) == NULL)
+	if (nvl == NULL || name == NULL || nvl->nvl_priv == 0)
 		return (EINVAL);
 
-	curr = priv->nvp_list;
-	while (curr != NULL) {
-		nvpair_t *nvp = &curr->nvi_nvp;
-
-		curr = curr->nvi_next;
-		if (strcmp(name, NVP_NAME(nvp)) != 0)
-			continue;
-
-		nvp_buf_unlink(nvl, nvp);
-		nvpair_free(nvp);
-		nvp_buf_free(nvl, nvp);
-
+	nvpair_t *nvp;
+	while ((nvp = nvt_lookup_name(nvl, name)) != NULL) {
+		VERIFY0(nvlist_remove_nvpair(nvl, nvp));
 		error = 0;
 	}
 
@@ -675,28 +951,14 @@ nvlist_remove_all(nvlist_t *nvl, const char *name)
 int
 nvlist_remove(nvlist_t *nvl, const char *name, data_type_t type)
 {
-	nvpriv_t *priv;
-	i_nvp_t *curr;
-
-	if (nvl == NULL || name == NULL ||
-	    (priv = (nvpriv_t *)(uintptr_t)nvl->nvl_priv) == NULL)
+	if (nvl == NULL || name == NULL || nvl->nvl_priv == 0)
 		return (EINVAL);
 
-	curr = priv->nvp_list;
-	while (curr != NULL) {
-		nvpair_t *nvp = &curr->nvi_nvp;
-
-		if (strcmp(name, NVP_NAME(nvp)) == 0 && NVP_TYPE(nvp) == type) {
-			nvp_buf_unlink(nvl, nvp);
-			nvpair_free(nvp);
-			nvp_buf_free(nvl, nvp);
-
-			return (0);
-		}
-		curr = curr->nvi_next;
-	}
+	nvpair_t *nvp = nvt_lookup_name_type(nvl, name, type);
+	if (nvp == NULL)
+		return (ENOENT);
 
-	return (ENOENT);
+	return (nvlist_remove_nvpair(nvl, nvp));
 }
 
 int
@@ -705,6 +967,10 @@ nvlist_remove_nvpair(nvlist_t *nvl, nvpair_t *nvp)
 	if (nvl == NULL || nvp == NULL)
 		return (EINVAL);
 
+	int err = nvt_remove_nvpair(nvl, nvp);
+	if (err != 0)
+		return (err);
+
 	nvp_buf_unlink(nvl, nvp);
 	nvpair_free(nvp);
 	nvp_buf_free(nvl, nvp);
@@ -982,6 +1248,12 @@ nvlist_add_common(nvlist_t *nvl, const char *name,
 	else if (nvl->nvl_nvflag & NV_UNIQUE_NAME_TYPE)
 		(void) nvlist_remove(nvl, name, type);
 
+	err = nvt_add_nvpair(nvl, nvp);
+	if (err != 0) {
+		nvpair_free(nvp);
+		nvp_buf_free(nvl, nvp);
+		return (err);
+	}
 	nvp_buf_link(nvl, nvp);
 
 	return (0);
@@ -1331,25 +1603,17 @@ static int
 nvlist_lookup_common(nvlist_t *nvl, const char *name, data_type_t type,
     uint_t *nelem, void *data)
 {
-	nvpriv_t *priv;
-	nvpair_t *nvp;
-	i_nvp_t *curr;
-
-	if (name == NULL || nvl == NULL ||
-	    (priv = (nvpriv_t *)(uintptr_t)nvl->nvl_priv) == NULL)
+	if (name == NULL || nvl == NULL || nvl->nvl_priv == 0)
 		return (EINVAL);
 
 	if (!(nvl->nvl_nvflag & (NV_UNIQUE_NAME | NV_UNIQUE_NAME_TYPE)))
 		return (ENOTSUP);
 
-	for (curr = priv->nvp_list; curr != NULL; curr = curr->nvi_next) {
-		nvp = &curr->nvi_nvp;
-
-		if (strcmp(name, NVP_NAME(nvp)) == 0 && NVP_TYPE(nvp) == type)
-			return (nvpair_value_common(nvp, type, nelem, data));
-	}
+	nvpair_t *nvp = nvt_lookup_name_type(nvl, name, type);
+	if (nvp == NULL)
+		return (ENOENT);
 
-	return (ENOENT);
+	return (nvpair_value_common(nvp, type, nelem, data));
 }
 
 int
@@ -2107,6 +2371,12 @@ nvs_decode_pairs(nvstream_t *nvs, nvlist_t *nvl)
 			return (EFAULT);
 		}
 
+		err = nvt_add_nvpair(nvl, nvp);
+		if (err != 0) {
+			nvpair_free(nvp);
+			nvp_buf_free(nvl, nvp);
+			return (err);
+		}
 		nvp_buf_link(nvl, nvp);
 	}
 	return (err);
diff --git a/usr/src/lib/libnvpair/nvpair_json.c b/usr/src/lib/libnvpair/nvpair_json.c
index 7ebd1be7a0..3205d229ab 100644
--- a/usr/src/lib/libnvpair/nvpair_json.c
+++ b/usr/src/lib/libnvpair/nvpair_json.c
@@ -10,6 +10,7 @@
  */
 /*
  * Copyright (c) 2014, Joyent, Inc.
+ * Copyright (c) 2017 by Delphix. All rights reserved.
  */
 
 #include <stdio.h>
@@ -461,8 +462,10 @@ nvlist_do_json(nvlist_t *nvl, char **bufp, size_t *blen, off_t *offp)
 		}
 
 		case DATA_TYPE_UNKNOWN:
+		case DATA_TYPE_DONTCARE:
 			return (-1);
 		}
+
 	}
 
 	FPRINTF(bufp, blen, offp, "}");
diff --git a/usr/src/uts/common/fs/zfs/arc.c b/usr/src/uts/common/fs/zfs/arc.c
index 5a899aa1c7..af70eee950 100644
--- a/usr/src/uts/common/fs/zfs/arc.c
+++ b/usr/src/uts/common/fs/zfs/arc.c
@@ -380,6 +380,13 @@ int zfs_arc_shrink_shift = 0;
 int zfs_arc_p_min_shift = 0;
 int zfs_arc_average_blocksize = 8 * 1024; /* 8KB */
 
+/*
+ * ARC dirty data constraints for arc_tempreserve_space() throttle
+ */
+uint_t zfs_arc_dirty_limit_percent = 50;	/* total dirty data limit */
+uint_t zfs_arc_anon_limit_percent = 25;		/* anon block dirty limit */
+uint_t zfs_arc_pool_dirty_percent = 20;		/* each pool's anon allowance */
+
 boolean_t zfs_compressed_arc_enabled = B_TRUE;
 
 /*
@@ -5848,12 +5855,10 @@ arc_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, arc_buf_t *buf,
 }
 
 static int
-arc_memory_throttle(uint64_t reserve, uint64_t txg)
+arc_memory_throttle(spa_t *spa, uint64_t reserve, uint64_t txg)
 {
 #ifdef _KERNEL
 	uint64_t available_memory = ptob(freemem);
-	static uint64_t page_load = 0;
-	static uint64_t last_txg = 0;
 
 #if defined(__i386)
 	available_memory =
@@ -5863,9 +5868,9 @@ arc_memory_throttle(uint64_t reserve, uint64_t txg)
 	if (freemem > physmem * arc_lotsfree_percent / 100)
 		return (0);
 
-	if (txg > last_txg) {
-		last_txg = txg;
-		page_load = 0;
+	if (txg > spa->spa_lowmem_last_txg) {
+		spa->spa_lowmem_last_txg = txg;
+		spa->spa_lowmem_page_load = 0;
 	}
 	/*
 	 * If we are in pageout, we know that memory is already tight,
@@ -5873,18 +5878,19 @@ arc_memory_throttle(uint64_t reserve, uint64_t txg)
 	 * continue to let page writes occur as quickly as possible.
 	 */
 	if (curproc == proc_pageout) {
-		if (page_load > MAX(ptob(minfree), available_memory) / 4)
+		if (spa->spa_lowmem_page_load >
+		    MAX(ptob(minfree), available_memory) / 4)
 			return (SET_ERROR(ERESTART));
 		/* Note: reserve is inflated, so we deflate */
-		page_load += reserve / 8;
+		atomic_add_64(&spa->spa_lowmem_page_load, reserve / 8);
 		return (0);
-	} else if (page_load > 0 && arc_reclaim_needed()) {
+	} else if (spa->spa_lowmem_page_load > 0 && arc_reclaim_needed()) {
 		/* memory is low, delay before restarting */
 		ARCSTAT_INCR(arcstat_memory_throttle_count, 1);
 		return (SET_ERROR(EAGAIN));
 	}
-	page_load = 0;
-#endif
+	spa->spa_lowmem_page_load = 0;
+#endif /* _KERNEL */
 	return (0);
 }
 
@@ -5896,7 +5902,7 @@ arc_tempreserve_clear(uint64_t reserve)
 }
 
 int
-arc_tempreserve_space(uint64_t reserve, uint64_t txg)
+arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg)
 {
 	int error;
 	uint64_t anon_size;
@@ -5923,7 +5929,7 @@ arc_tempreserve_space(uint64_t reserve, uint64_t txg)
 	 * in order to compress/encrypt/etc the data.  We therefore need to
 	 * make sure that there is sufficient available memory for this.
 	 */
-	error = arc_memory_throttle(reserve, txg);
+	error = arc_memory_throttle(spa, reserve, txg);
 	if (error != 0)
 		return (error);
 
@@ -5931,12 +5937,24 @@ arc_tempreserve_space(uint64_t reserve, uint64_t txg)
 	 * Throttle writes when the amount of dirty data in the cache
 	 * gets too large.  We try to keep the cache less than half full
 	 * of dirty blocks so that our sync times don't grow too large.
+	 *
+	 * In the case of one pool being built on another pool, we want
+	 * to make sure we don't end up throttling the lower (backing)
+	 * pool when the upper pool is the majority contributor to dirty
+	 * data. To insure we make forward progress during throttling, we
+	 * also check the current pool's net dirty data and only throttle
+	 * if it exceeds zfs_arc_pool_dirty_percent of the anonymous dirty
+	 * data in the cache.
+	 *
 	 * Note: if two requests come in concurrently, we might let them
 	 * both succeed, when one of them should fail.  Not a huge deal.
 	 */
+	uint64_t total_dirty = reserve + arc_tempreserve + anon_size;
+	uint64_t spa_dirty_anon = spa_dirty_data(spa);
 
-	if (reserve + arc_tempreserve + anon_size > arc_c / 2 &&
-	    anon_size > arc_c / 4) {
+	if (total_dirty > arc_c * zfs_arc_dirty_limit_percent / 100 &&
+	    anon_size > arc_c * zfs_arc_anon_limit_percent / 100 &&
+	    spa_dirty_anon > anon_size * zfs_arc_pool_dirty_percent / 100) {
 		uint64_t meta_esize =
 		    refcount_count(&arc_anon->arcs_esize[ARC_BUFC_METADATA]);
 		uint64_t data_esize =
diff --git a/usr/src/uts/common/fs/zfs/dsl_dir.c b/usr/src/uts/common/fs/zfs/dsl_dir.c
index 99bc468313..35e76e273e 100644
--- a/usr/src/uts/common/fs/zfs/dsl_dir.c
+++ b/usr/src/uts/common/fs/zfs/dsl_dir.c
@@ -1380,7 +1380,7 @@ dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 	    offsetof(struct tempreserve, tr_node));
 	ASSERT3S(asize, >, 0);
 
-	err = arc_tempreserve_space(lsize, tx->tx_txg);
+	err = arc_tempreserve_space(dd->dd_pool->dp_spa, lsize, tx->tx_txg);
 	if (err == 0) {
 		struct tempreserve *tr;
 
diff --git a/usr/src/uts/common/fs/zfs/spa_misc.c b/usr/src/uts/common/fs/zfs/spa_misc.c
index 41342f37ea..87a95f0a36 100644
--- a/usr/src/uts/common/fs/zfs/spa_misc.c
+++ b/usr/src/uts/common/fs/zfs/spa_misc.c
@@ -1910,6 +1910,12 @@ bp_get_dsize(spa_t *spa, const blkptr_t *bp)
 	return (dsize);
 }
 
+uint64_t
+spa_dirty_data(spa_t *spa)
+{
+	return (spa->spa_dsl_pool->dp_dirty_total);
+}
+
 /*
  * ==========================================================================
  * Initialization and Termination
diff --git a/usr/src/uts/common/fs/zfs/sys/arc.h b/usr/src/uts/common/fs/zfs/sys/arc.h
index 10c920ff9d..45db7701e1 100644
--- a/usr/src/uts/common/fs/zfs/sys/arc.h
+++ b/usr/src/uts/common/fs/zfs/sys/arc.h
@@ -190,7 +190,7 @@ void arc_freed(spa_t *spa, const blkptr_t *bp);
 
 void arc_flush(spa_t *spa, boolean_t retry);
 void arc_tempreserve_clear(uint64_t reserve);
-int arc_tempreserve_space(uint64_t reserve, uint64_t txg);
+int arc_tempreserve_space(spa_t *spa, uint64_t reserve, uint64_t txg);
 
 uint64_t arc_max_bytes(void);
 void arc_init(void);
diff --git a/usr/src/uts/common/fs/zfs/sys/spa.h b/usr/src/uts/common/fs/zfs/sys/spa.h
index 34f02ed430..1acbe31377 100644
--- a/usr/src/uts/common/fs/zfs/sys/spa.h
+++ b/usr/src/uts/common/fs/zfs/sys/spa.h
@@ -813,6 +813,7 @@ extern uint64_t spa_bootfs(spa_t *spa);
 extern uint64_t spa_delegation(spa_t *spa);
 extern objset_t *spa_meta_objset(spa_t *spa);
 extern uint64_t spa_deadman_synctime(spa_t *spa);
+extern uint64_t spa_dirty_data(spa_t *spa);
 
 /* Miscellaneous support routines */
 extern void spa_load_failed(spa_t *spa, const char *fmt, ...);
diff --git a/usr/src/uts/common/fs/zfs/sys/spa_impl.h b/usr/src/uts/common/fs/zfs/sys/spa_impl.h
index 4b90e96cfa..ea251cf0c6 100644
--- a/usr/src/uts/common/fs/zfs/sys/spa_impl.h
+++ b/usr/src/uts/common/fs/zfs/sys/spa_impl.h
@@ -373,6 +373,10 @@ struct spa {
 		int spa_queued;
 	} spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE];
 
+	/* arc_memory_throttle() parameters during low memory condition */
+	uint64_t	spa_lowmem_page_load;	/* memory load during txg */
+	uint64_t	spa_lowmem_last_txg;	/* txg window start */
+
 	hrtime_t	spa_ccw_fail_time;	/* Conf cache write fail time */
 
 	/*
diff --git a/usr/src/uts/common/sys/nvpair.h b/usr/src/uts/common/sys/nvpair.h
index e4d637b007..cf3f761c8c 100644
--- a/usr/src/uts/common/sys/nvpair.h
+++ b/usr/src/uts/common/sys/nvpair.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
  */
 
 #ifndef	_SYS_NVPAIR_H
@@ -40,6 +40,7 @@ extern "C" {
 #endif
 
 typedef enum {
+	DATA_TYPE_DONTCARE = -1,
 	DATA_TYPE_UNKNOWN = 0,
 	DATA_TYPE_BOOLEAN,
 	DATA_TYPE_BYTE,
diff --git a/usr/src/uts/common/sys/nvpair_impl.h b/usr/src/uts/common/sys/nvpair_impl.h
index f12dbbfe6e..c9874b3e4d 100644
--- a/usr/src/uts/common/sys/nvpair_impl.h
+++ b/usr/src/uts/common/sys/nvpair_impl.h
@@ -24,11 +24,13 @@
  * Use is subject to license terms.
  */
 
+/*
+ * Copyright (c) 2017 by Delphix. All rights reserved.
+ */
+
 #ifndef	_NVPAIR_IMPL_H
 #define	_NVPAIR_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -47,16 +49,27 @@ typedef struct i_nvp i_nvp_t;
 
 struct i_nvp {
 	union {
-		uint64_t	_nvi_align;	/* ensure alignment */
+		/* ensure alignment */
+		uint64_t	_nvi_align;
+
 		struct {
-			i_nvp_t	*_nvi_next;	/* pointer to next nvpair */
-			i_nvp_t	*_nvi_prev;	/* pointer to prev nvpair */
+			/* pointer to next nvpair */
+			i_nvp_t	*_nvi_next;
+
+			/* pointer to prev nvpair */
+			i_nvp_t	*_nvi_prev;
+
+			/* next pair in table bucket */
+			i_nvp_t	*_nvi_hashtable_next;
 		} _nvi;
 	} _nvi_un;
-	nvpair_t nvi_nvp;			/* nvpair */
+
+	/* nvpair */
+	nvpair_t nvi_nvp;
 };
 #define	nvi_next	_nvi_un._nvi._nvi_next
 #define	nvi_prev	_nvi_un._nvi._nvi_prev
+#define	nvi_hashtable_next	_nvi_un._nvi._nvi_hashtable_next
 
 typedef struct {
 	i_nvp_t		*nvp_list;	/* linked list of nvpairs */
@@ -64,6 +77,10 @@ typedef struct {
 	i_nvp_t		*nvp_curr;	/* current walker nvpair */
 	nv_alloc_t	*nvp_nva;	/* pluggable allocator */
 	uint32_t	nvp_stat;	/* internal state */
+
+	i_nvp_t		**nvp_hashtable; /* table of entries used for lookup */
+	uint32_t	nvp_nbuckets;	/* # of buckets in hash table */
+	uint32_t	nvp_nentries;	/* # of entries in hash table */
 } nvpriv_t;
 
 #ifdef	__cplusplus