Merge pull request #2354 from wiredtiger/server_21553_30backportmongodb-3.0.8

SERVER-21553 3.0 backport

8 files changed, 121 insertions, 81 deletions

diff --git a/src/btree/bt_delete.c b/src/btree/bt_delete.c
index 7313e31267f..e19085830bd 100644
--- a/src/btree/bt_delete.c
+++ b/src/btree/bt_delete.c
@@ -250,6 +250,18 @@ __wt_delete_page_skip(WT_SESSION_IMPL *session, WT_REF *ref, bool visible_all)
 	    __wt_txn_visible_all(session, ref->page_del->txnid) :
 	    __wt_txn_visible(session, ref->page_del->txnid));
 
+	/*
+	 * The page_del structure can be freed as soon as the delete is stable:
+	 * it is only read when the ref state is WT_REF_DELETED.  It is worth
+	 * checking every time we come through because once this is freed, we
+	 * no longer need synchronization to check the ref.
+	 */
+	if (skip && ref->page_del != NULL && (visible_all ||
+	    __wt_txn_visible_all(session, ref->page_del->txnid))) {
+		__wt_free(session, ref->page_del->update_list);
+		__wt_free(session, ref->page_del);
+	}
+
 	WT_PUBLISH(ref->state, WT_REF_DELETED);
 	return (skip);
 }
diff --git a/src/btree/bt_discard.c b/src/btree/bt_discard.c
index c876da6309c..30e19147e12 100644
--- a/src/btree/bt_discard.c
+++ b/src/btree/bt_discard.c
@@ -252,10 +252,7 @@ __wt_free_ref(
 	}
 
 	/* Free any address allocation. */
-	if (ref->addr != NULL && __wt_off_page(page, ref->addr)) {
-		__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-		__wt_free(session, ref->addr);
-	}
+	__wt_ref_addr_free(session, ref);
 
 	/* Free any page-deleted information. */
 	if (ref->page_del != NULL) {
diff --git a/src/btree/bt_slvg.c b/src/btree/bt_slvg.c
index 89355baeb5c..e4e611f947a 100644
--- a/src/btree/bt_slvg.c
+++ b/src/btree/bt_slvg.c
@@ -1299,9 +1299,7 @@ __slvg_col_build_leaf(WT_SESSION_IMPL *session, WT_TRACK *trk, WT_REF *ref)
 	 * would have been lost.)  Clear the reference addr so eviction doesn't
 	 * free the underlying blocks.
 	 */
-	__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-	__wt_free(session, ref->addr);
-	ref->addr = NULL;
+	__wt_ref_addr_free(session, ref);
 
 	/* Write the new version of the leaf page to disk. */
 	WT_ERR(__slvg_modify_init(session, page));
@@ -2008,9 +2006,7 @@ __slvg_row_build_leaf(
 	 * would have been lost.)  Clear the reference addr so eviction doesn't
 	 * free the underlying blocks.
 	 */
-	__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-	__wt_free(session, ref->addr);
-	ref->addr = NULL;
+	__wt_ref_addr_free(session, ref);
 
 	/* Write the new version of the leaf page to disk. */
 	WT_ERR(__slvg_modify_init(session, page));
diff --git a/src/btree/bt_split.c b/src/btree/bt_split.c
index 9d8e463feb0..6e0436bb01f 100644
--- a/src/btree/bt_split.c
+++ b/src/btree/bt_split.c
@@ -1030,6 +1030,16 @@ __split_parent(WT_SESSION_IMPL *session, WT_REF *ref,
 #endif
 
 	/*
+	 * Page-delete information is only read when the WT_REF state is
+	 * WT_REF_DELETED.  The page-delete memory wasn't added to the
+	 * parent's footprint, ignore it here.
+	 */
+	if (ref->page_del != NULL) {
+		__wt_free(session, ref->page_del->update_list);
+		__wt_free(session, ref->page_del);
+	}
+
+	/*
 	 * Reset the page's original WT_REF field to split.  Threads cursoring
 	 * through the tree were blocked because that WT_REF state was set to
 	 * locked.  This update changes the locked state to split, unblocking
@@ -1090,19 +1100,15 @@ __split_parent(WT_SESSION_IMPL *session, WT_REF *ref,
 				    session, split_gen, 0, ikey, size));
 				parent_decr += size;
 			}
-			/*
-			 * The page_del structure can be freed immediately: it
-			 * is only read when the ref state is WT_REF_DELETED.
-			 * The size of the structure wasn't added to the parent,
-			 * don't decrement.
-			 */
-			if (next_ref->page_del != NULL) {
-				__wt_free(session,
-				    next_ref->page_del->update_list);
-				__wt_free(session, next_ref->page_del);
-			}
 		}
 
+		/*
+		 * If this page was fast-truncated, any attached structure
+		 * should have been freed before now.
+		 */
+		WT_ASSERT(session, next_ref->page_del == NULL);
+
+		WT_TRET(__wt_ref_block_free(session, next_ref));
 		WT_TRET(__split_safe_free(
 		    session, split_gen, 0, next_ref, sizeof(WT_REF)));
 		parent_decr += sizeof(WT_REF);
@@ -1213,21 +1219,30 @@ __split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
 	 * The first page in the split is the current page, but we still have
 	 * to create a replacement WT_REF, the original WT_REF will be set to
 	 * split status and eventually freed.
+	 *
+	 * The new WT_REF is not quite identical: we have to instantiate a key,
+	 * and the new reference is visible to readers once the split completes.
+	 *
+	 * The key-instantiation code checks for races, leave the key fields
+	 * zeroed we don't trigger them.
+	 *
+	 * Don't copy any deleted page state: we may be splitting a page that
+	 * was instantiated after a truncate and that history should not be
+	 * carried onto these new child pages.
 	 */
 	WT_ERR(__wt_calloc_one(session, &split_ref[0]));
 	child = split_ref[0];
-	*child = *ref;
+	child->page = ref->page;
+	child->home = ref->home;
+	child->pindex_hint = ref->pindex_hint;
+	child->state = WT_REF_MEM;
+	child->addr = ref->addr;
 
 	/*
-	 * The new WT_REF is not quite identical: we have to instantiate a key,
-	 * and the new reference is visible to readers once the split completes.
-	 *
-	 * The key-instantiation code checks for races, clear the key fields so
-	 * we don't trigger them.
+	 * The address has moved to the replacement WT_REF.  Make sure it isn't
+	 * freed when the original ref is discarded.
 	 */
-	child->key.recno = 0;
-	child->key.ikey = NULL;
-	child->state = WT_REF_MEM;
+	ref->addr = NULL;
 
 	/*
 	 * Copy the first key from the original page into first ref in the new
@@ -1429,6 +1444,11 @@ __split_insert(WT_SESSION_IMPL *session, WT_REF *ref)
 	return (0);
 
 err:	if (split_ref[0] != NULL) {
+		/*
+		 * The address was moved to the replacement WT_REF, restore it.
+		 */
+		ref->addr = split_ref[0]->addr;
+
 		__wt_free(session, split_ref[0]->key.ikey);
 		__wt_free(session, split_ref[0]);
 	}
diff --git a/src/btree/bt_walk.c b/src/btree/bt_walk.c
index 8e0f4036b79..c7d83d8dfff 100644
--- a/src/btree/bt_walk.c
+++ b/src/btree/bt_walk.c
@@ -244,7 +244,8 @@ ascend:	/*
 			 * If we see any child states other than deleted, the
 			 * page isn't empty.
 			 */
-			if (ref->state != WT_REF_DELETED)
+			if (ref->state != WT_REF_DELETED &&
+			    !LF_ISSET(WT_READ_TRUNCATE))
 				empty_internal = false;
 
 			if (LF_ISSET(WT_READ_CACHE)) {
@@ -270,6 +271,7 @@ ascend:	/*
 				WT_ERR(__wt_delete_page(session, ref, &skip));
 				if (skip)
 					break;
+				empty_internal = false;
 			} else if (LF_ISSET(WT_READ_COMPACT)) {
 				/*
 				 * Skip deleted pages, rewriting them doesn't
diff --git a/src/evict/evict_page.c b/src/evict/evict_page.c
index 9de66922931..046d8bb3eba 100644
--- a/src/evict/evict_page.c
+++ b/src/evict/evict_page.c
@@ -222,19 +222,14 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 {
 	WT_ADDR *addr;
 	WT_DECL_RET;
-	WT_PAGE *parent;
 	WT_PAGE_MODIFY *mod;
 
-	parent = ref->home;
 	mod = ref->page->modify;
 
+	WT_ASSERT(session, ref->addr == NULL);
+
 	switch (mod->rec_result) {
 	case WT_PM_REC_EMPTY:				/* Page is empty */
-		if (ref->addr != NULL && __wt_off_page(parent, ref->addr)) {
-			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-			__wt_free(session, ref->addr);
-		}
-
 		/*
 		 * Update the parent to reference a deleted page.  The fact that
 		 * reconciliation left the page "empty" means there's no older
@@ -261,11 +256,6 @@ __evict_page_dirty_update(WT_SESSION_IMPL *session, WT_REF *ref, bool closing)
 		WT_RET(__wt_split_multi(session, ref, closing));
 		break;
 	case WT_PM_REC_REPLACE: 			/* 1-for-1 page swap */
-		if (ref->addr != NULL && __wt_off_page(parent, ref->addr)) {
-			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-			__wt_free(session, ref->addr);
-		}
-
 		/*
 		 * Update the parent to reference the replacement page.
 		 *
diff --git a/src/include/btree.i b/src/include/btree.i
index 1c416c99e13..4029b29d207 100644
--- a/src/include/btree.i
+++ b/src/include/btree.i
@@ -1008,6 +1008,61 @@ __wt_page_can_split(WT_SESSION_IMPL *session, WT_PAGE *page)
 }
 
 /*
+ * __wt_ref_addr_free --
+ *	Free the address in a reference, if necessary.
+ */
+static inline void
+__wt_ref_addr_free(WT_SESSION_IMPL *session, WT_REF *ref)
+ {
+	if (ref->addr == NULL)
+		return;
+
+	if (ref->home == NULL || __wt_off_page(ref->home, ref->addr)) {
+		__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
+		__wt_free(session, ref->addr);
+	}
+	ref->addr = NULL;
+}
+
+/*
+ * __wt_btree_block_free --
+ *	Helper function to free a block from the current tree.
+ */
+static inline int
+__wt_btree_block_free(
+    WT_SESSION_IMPL *session, const uint8_t *addr, size_t addr_size)
+{
+	WT_BM *bm;
+	WT_BTREE *btree;
+
+	btree = S2BT(session);
+	bm = btree->bm;
+
+	return (bm->free(bm, session, addr, addr_size));
+}
+
+/*
+ * __wt_ref_block_free --
+ *	Free the on-disk block for a reference and clear the address.
+ */
+static inline int
+__wt_ref_block_free(WT_SESSION_IMPL *session, WT_REF *ref)
+{
+	const uint8_t *addr;
+	size_t addr_size;
+
+	if (ref->addr == NULL)
+		return (0);
+
+	WT_RET(__wt_ref_info(session, ref, &addr, &addr_size, NULL));
+	WT_RET(__wt_btree_block_free(session, addr, addr_size));
+
+	/* Clear the address (so we don't free it twice). */
+	__wt_ref_addr_free(session, ref);
+	return (0);
+}
+
+/*
  * __wt_page_can_evict --
  *	Check whether a page can be evicted.
  */
diff --git a/src/reconcile/rec_write.c b/src/reconcile/rec_write.c
index 73b7f4968e9..67b43057c8a 100644
--- a/src/reconcile/rec_write.c
+++ b/src/reconcile/rec_write.c
@@ -1066,10 +1066,7 @@ static int
 __rec_child_deleted(WT_SESSION_IMPL *session,
     WT_RECONCILE *r, WT_REF *ref, WT_CHILD_STATE *statep)
 {
-	WT_BM *bm;
 	WT_PAGE_DELETED *page_del;
-	size_t addr_size;
-	const uint8_t *addr;
 
 	page_del = ref->page_del;
 
@@ -1117,17 +1114,8 @@ __rec_child_deleted(WT_SESSION_IMPL *session,
 	 */
 	if (ref->addr != NULL &&
 	    (page_del == NULL ||
-	    __wt_txn_visible_all(session, page_del->txnid))) {
-		WT_RET(__wt_ref_info(session, ref, &addr, &addr_size, NULL));
-		bm = S2BT(session)->bm;
-		WT_RET(bm->free(bm, session, addr, addr_size));
-
-		if (__wt_off_page(ref->home, ref->addr)) {
-			__wt_free(session, ((WT_ADDR *)ref->addr)->addr);
-			__wt_free(session, ref->addr);
-		}
-		ref->addr = NULL;
-	}
+	    __wt_txn_visible_all(session, page_del->txnid)))
+		WT_RET(__wt_ref_block_free(session, ref));
 
 	/*
 	 * If the original page is gone, we can skip the slot on the internal
@@ -4790,13 +4778,11 @@ __rec_row_leaf_insert(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_INSERT *ins)
 static int
 __rec_split_discard(WT_SESSION_IMPL *session, WT_PAGE *page)
 {
-	WT_BM *bm;
 	WT_DECL_RET;
 	WT_PAGE_MODIFY *mod;
 	WT_MULTI *multi;
 	uint32_t i;
 
-	bm = S2BT(session)->bm;
 	mod = page->modify;
 
 	/*
@@ -4816,7 +4802,7 @@ __rec_split_discard(WT_SESSION_IMPL *session, WT_PAGE *page)
 			if (multi->addr.reuse)
 				multi->addr.addr = NULL;
 			else {
-				WT_RET(bm->free(bm, session,
+				WT_RET(__wt_btree_block_free(session,
 				    multi->addr.addr, multi->addr.size));
 				__wt_free(session, multi->addr.addr);
 			}
@@ -4862,8 +4848,6 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 	WT_MULTI *multi;
 	WT_PAGE_MODIFY *mod;
 	WT_REF *ref;
-	size_t addr_size;
-	const uint8_t *addr;
 
 	btree = S2BT(session);
 	bm = btree->bm;
@@ -4888,21 +4872,7 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 		 */
 		if (__wt_ref_is_root(ref))
 			break;
-		if (ref->addr != NULL) {
-			/*
-			 * Free the page and clear the address (so we don't free
-			 * it twice).
-			 */
-			WT_RET(__wt_ref_info(
-			    session, ref, &addr, &addr_size, NULL));
-			WT_RET(bm->free(bm, session, addr, addr_size));
-			if (__wt_off_page(ref->home, ref->addr)) {
-				__wt_free(
-				    session, ((WT_ADDR *)ref->addr)->addr);
-				__wt_free(session, ref->addr);
-			}
-			ref->addr = NULL;
-		}
+		WT_RET(__wt_ref_block_free(session, ref));
 		break;
 	case WT_PM_REC_EMPTY:				/* Page deleted */
 		break;
@@ -4921,7 +4891,7 @@ __rec_write_wrapup(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 		 * are checkpoints, and must be explicitly dropped.
 		 */
 		if (!__wt_ref_is_root(ref))
-			WT_RET(bm->free(bm, session,
+			WT_RET(__wt_btree_block_free(session,
 			    mod->mod_replace.addr, mod->mod_replace.size));
 
 		/* Discard the replacement page's address. */
@@ -5126,14 +5096,12 @@ err:			__wt_scr_free(session, &tkey);
 static int
 __rec_write_wrapup_err(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 {
-	WT_BM *bm;
 	WT_BOUNDARY *bnd;
 	WT_DECL_RET;
 	WT_MULTI *multi;
 	WT_PAGE_MODIFY *mod;
 	uint32_t i;
 
-	bm = S2BT(session)->bm;
 	mod = page->modify;
 
 	/*
@@ -5164,7 +5132,7 @@ __rec_write_wrapup_err(WT_SESSION_IMPL *session, WT_RECONCILE *r, WT_PAGE *page)
 			if (bnd->addr.reuse)
 				bnd->addr.addr = NULL;
 			else {
-				WT_TRET(bm->free(bm, session,
+				WT_TRET(__wt_btree_block_free(session,
 				    bnd->addr.addr, bnd->addr.size));
 				__wt_free(session, bnd->addr.addr);
 			}