Add an operation to set a flag atomically, use it to avoid racing on page flags.

6 files changed, 31 insertions, 11 deletions

diff --git a/src/btree/bt_curprev.c b/src/btree/bt_curprev.c
index 39d0aa8aed3..0347a3f836a 100644
--- a/src/btree/bt_curprev.c
+++ b/src/btree/bt_curprev.c
@@ -344,7 +344,7 @@ __cursor_row_prev(WT_CURSOR_BTREE *cbt, int newpage)
 		 * If we haven't instantiated keys on this page, do so, else it
 		 * is a very, very slow traversal.
 		 */
-		if (!F_ISSET(cbt->page, WT_PAGE_BUILD_KEYS))
+		if (!F_ISSET_ATOMIC(cbt->page, WT_PAGE_BUILD_KEYS))
 			WT_RET(__wt_row_leaf_keys(session, cbt->page));
 
 		if (cbt->page->entries == 0)
diff --git a/src/btree/bt_debug.c b/src/btree/bt_debug.c
index 06fc37189a4..2799637c4b3 100644
--- a/src/btree/bt_debug.c
+++ b/src/btree/bt_debug.c
@@ -442,7 +442,7 @@ __debug_page_hdr(WT_DBG *ds, WT_PAGE *page)
 	__dmsg(ds, ": %s", __wt_page_type_string(page->type));
 
 	__dmsg(ds, " (%s", __wt_page_is_modified(page) ? "dirty" : "clean");
-	if (F_ISSET(page, WT_PAGE_BUILD_KEYS))
+	if (F_ISSET_ATOMIC(page, WT_PAGE_BUILD_KEYS))
 		__dmsg(ds, ", keys-built");
 	if (page->modify != NULL) {
 		if (F_ISSET(page->modify, WT_PM_REC_EMPTY))
diff --git a/src/btree/bt_discard.c b/src/btree/bt_discard.c
index 0b65ebd43c8..a564e3cbc9c 100644
--- a/src/btree/bt_discard.c
+++ b/src/btree/bt_discard.c
@@ -32,7 +32,7 @@ __wt_page_out(WT_SESSION_IMPL *session, WT_PAGE *page, uint32_t flags)
 	page->parent = NULL;
 	page->ref = NULL;
 
-	WT_ASSERT(session, !F_ISSET(page, WT_PAGE_EVICT_LRU));
+	WT_ASSERT(session, !F_ISSET_ATOMIC(page, WT_PAGE_EVICT_LRU));
 
 	/* If not a split merged into its parent, the page must be clean. */
 	WT_ASSERT(session,
diff --git a/src/btree/bt_evict.c b/src/btree/bt_evict.c
index cc265215647..80694247207 100644
--- a/src/btree/bt_evict.c
+++ b/src/btree/bt_evict.c
@@ -42,8 +42,8 @@ static inline void
 __evict_clr(WT_SESSION_IMPL *session, WT_EVICT_LIST *e)
 {
 	if (e->page != NULL) {
-		WT_ASSERT(session, F_ISSET(e->page, WT_PAGE_EVICT_LRU));
-		F_CLR(e->page, WT_PAGE_EVICT_LRU);
+		WT_ASSERT(session, F_ISSET_ATOMIC(e->page, WT_PAGE_EVICT_LRU));
+		F_CLR_ATOMIC(e->page, WT_PAGE_EVICT_LRU);
 	}
 	e->page = NULL;
 	e->btree = WT_DEBUG_POINT;
@@ -85,7 +85,7 @@ __wt_evict_clr_page(WT_SESSION_IMPL *session, WT_PAGE *page)
 	    page->ref->state == WT_REF_LOCKED);
 
 	/* Fast path: if the page isn't on the queue, don't bother searching. */
-	if (!F_ISSET(page, WT_PAGE_EVICT_LRU))
+	if (!F_ISSET_ATOMIC(page, WT_PAGE_EVICT_LRU))
 		return;
 
 	cache = S2C(session)->cache;
@@ -98,7 +98,7 @@ __wt_evict_clr_page(WT_SESSION_IMPL *session, WT_PAGE *page)
 			break;
 		}
 
-	WT_ASSERT(session, !F_ISSET(page, WT_PAGE_EVICT_LRU));
+	WT_ASSERT(session, !F_ISSET_ATOMIC(page, WT_PAGE_EVICT_LRU));
 
 	__wt_spin_unlock(session, &cache->lru_lock);
 }
@@ -670,7 +670,7 @@ __evict_walk_file(WT_SESSION_IMPL *session, u_int *slotp)
 		 * really be skipped.
 		 */
 		if (WT_PAGE_IS_ROOT(page) ||
-		    F_ISSET(page, WT_PAGE_EVICT_LRU) ||
+		    F_ISSET_ATOMIC(page, WT_PAGE_EVICT_LRU) ||
 		    (page->modify != NULL &&
 		    F_ISSET(page->modify, WT_PM_REC_SPLIT_MERGE)))
 			continue;
@@ -686,7 +686,7 @@ __evict_walk_file(WT_SESSION_IMPL *session, u_int *slotp)
 		++evict;
 
 		/* Mark the page on the list */
-		F_SET(page, WT_PAGE_EVICT_LRU);
+		F_SET_ATOMIC(page, WT_PAGE_EVICT_LRU);
 	}
 
 	*slotp += (u_int)(evict - start);
diff --git a/src/btree/row_key.c b/src/btree/row_key.c
index 9cff462ed22..fb4458d083d 100644
--- a/src/btree/row_key.c
+++ b/src/btree/row_key.c
@@ -26,7 +26,7 @@ __wt_row_leaf_keys(WT_SESSION_IMPL *session, WT_PAGE *page)
 	ret = 0;
 
 	if (page->entries == 0) {			/* Just checking... */
-		F_SET(page, WT_PAGE_BUILD_KEYS);
+		F_SET_ATOMIC(page, WT_PAGE_BUILD_KEYS);
 		return (0);
 	}
 
@@ -57,7 +57,7 @@ __wt_row_leaf_keys(WT_SESSION_IMPL *session, WT_PAGE *page)
 		if (__bit_test(tmp->mem, i))
 			WT_ERR(__wt_row_key(session, page, rip, NULL));
 
-	F_SET(page, WT_PAGE_BUILD_KEYS);
+	F_SET_ATOMIC(page, WT_PAGE_BUILD_KEYS);
 
 err:	__wt_scr_free(&tmp);
 	return (ret);
diff --git a/src/include/mutex.h b/src/include/mutex.h
index 539795ea404..93f1d597b45 100644
--- a/src/include/mutex.h
+++ b/src/include/mutex.h
@@ -127,6 +127,26 @@
 } while (0)
 
 /*
+ * Atomic versions of F_ISSET, F_SET and F_CLR.
+ * Spin until the new value can be swapped into place.
+ */
+#define F_ISSET_ATOMIC(p, mask)	((p)->flags_atomic & (mask))
+
+#define F_SET_ATOMIC(p, mask)	do {					\
+	uint32_t __orig;						\
+	do {								\
+		__orig = (p)->flags_atomic;				\
+	} while (!WT_ATOMIC_CAS((p)->flags_atomic, __orig, __orig | (mask)));\
+} while (0)
+
+#define F_CLR_ATOMIC(p, mask)	do {					\
+	uint32_t __orig;						\
+	do {								\
+		__orig = (p)->flags_atomic;				\
+	} while (!WT_ATOMIC_CAS((p)->flags_atomic, __orig, __orig & ~(mask)));\
+} while (0)
+
+/*
  * Condition variables:
  *
  * WiredTiger uses standard pthread condition variables to signal between