Import wiredtiger-wiredtiger-2.8.0-209-g234b68b.tar.gz from wiredtige…

…r branch mongodb-3.2

ref: 88b898e..234b68b

SERVER-24306  40-second journaling stall from "log files prepared" to checkpoint
WT-2559       Windows segfault in logging code
WT-2560       Stuck trying to update oldest transaction ID
WT-2613       Windows build failing with a C4100 error
WT-2629       Introduction of ppc64le crc32c assembly file has made the stack executable
SERVER-23954  WiredTiger changes for MongoDB 3.2.7

src/third_party/wiredtiger/src/btree/bt_cursor.c

@@ -325,7 +325,7 @@ __wt_btcur_search(WT_CURSOR_BTREE *cbt)
 	valid = false;
 	if (F_ISSET(cbt, WT_CBT_ACTIVE) &&
 	    cbt->ref->page->read_gen != WT_READGEN_OLDEST) {
-		__wt_txn_cursor_op(session);
+		WT_ERR(__wt_txn_cursor_op(session));
 
 		WT_ERR(btree->type == BTREE_ROW ?
 		    __cursor_row_search(session, cbt, cbt->ref, false) :
@@ -405,7 +405,7 @@ __wt_btcur_search_near(WT_CURSOR_BTREE *cbt, int *exactp)
 	if (btree->type == BTREE_ROW &&
 	    F_ISSET(cbt, WT_CBT_ACTIVE) &&
 	    cbt->ref->page->read_gen != WT_READGEN_OLDEST) {
-		__wt_txn_cursor_op(session);
+		WT_ERR(__wt_txn_cursor_op(session));
 
 		WT_ERR(__cursor_row_search(session, cbt, cbt->ref, true));
 

src/third_party/wiredtiger/src/btree/bt_read.c

@@ -326,7 +326,7 @@ __evict_force_check(WT_SESSION_IMPL *session, WT_REF *ref)
 	__wt_page_evict_soon(page);
 
 	/* Bump the oldest ID, we're about to do some visibility checks. */
-	__wt_txn_update_oldest(session, false);
+	WT_RET(__wt_txn_update_oldest(session, false));
 
 	/* If eviction cannot succeed, don't try. */
 	return (__wt_page_can_evict(session, ref, NULL));

src/third_party/wiredtiger/src/btree/bt_sync.c

@@ -81,7 +81,7 @@ __sync_file(WT_SESSION_IMPL *session, WT_CACHE_OP syncop)
 			if (__wt_page_is_modified(page) &&
 			    WT_TXNID_LT(page->modify->update_txn, oldest_id)) {
 				if (txn->isolation == WT_ISO_READ_COMMITTED)
-					__wt_txn_get_snapshot(session);
+					WT_ERR(__wt_txn_get_snapshot(session));
 				leaf_bytes += page->memory_footprint;
 				++leaf_pages;
 				WT_ERR(__wt_reconcile(session, walk, NULL, 0));
@@ -100,7 +100,7 @@ __sync_file(WT_SESSION_IMPL *session, WT_CACHE_OP syncop)
 		 * the metadata shouldn't be that big, and (b) if we do ever
 		 */
 		if (txn->isolation == WT_ISO_READ_COMMITTED)
-			__wt_txn_get_snapshot(session);
+			WT_ERR(__wt_txn_get_snapshot(session));
 
 		/*
 		 * We cannot check the tree modified flag in the case of a

src/third_party/wiredtiger/src/conn/conn_open.c

@@ -93,7 +93,7 @@ __wt_connection_close(WT_CONNECTION_IMPL *conn)
 	 * transaction ID will catch up with the current ID.
 	 */
 	for (;;) {
-		__wt_txn_update_oldest(session, true);
+		WT_TRET(__wt_txn_update_oldest(session, true));
 		if (txn_global->oldest_id == txn_global->current)
 			break;
 		__wt_yield();

src/third_party/wiredtiger/src/cursor/cur_ds.c

@@ -16,7 +16,7 @@ static int
 __curds_txn_enter(WT_SESSION_IMPL *session)
 {
 	session->ncursors++;				/* XXX */
-	__wt_txn_cursor_op(session);
+	WT_RET(__wt_txn_cursor_op(session));
 
 	return (0);
 }

src/third_party/wiredtiger/src/evict/evict_file.c

@@ -26,7 +26,7 @@ __wt_evict_file(WT_SESSION_IMPL *session, WT_CACHE_OP syncop)
 	WT_RET(__wt_evict_file_exclusive_on(session));
 
 	/* Make sure the oldest transaction ID is up-to-date. */
-	__wt_txn_update_oldest(session, true);
+	WT_RET(__wt_txn_update_oldest(session, true));
 
 	/* Walk the tree, discarding pages. */
 	next_ref = NULL;

src/third_party/wiredtiger/src/evict/evict_lru.c

@@ -594,7 +594,7 @@ __evict_pass(WT_SESSION_IMPL *session)
 		 * of whether the cache is full, to prevent the oldest ID
 		 * falling too far behind.
 		 */
-		__wt_txn_update_oldest(session, true);
+		WT_RET(__wt_txn_update_oldest(session, loop > 0));
 
 		if (!__evict_update_work(session))
 			break;

src/third_party/wiredtiger/src/evict/evict_page.c

@@ -420,7 +420,7 @@ __evict_review(
 		 * fallen behind current.
 		 */
 		if (modified)
-			__wt_txn_update_oldest(session, true);
+			WT_RET(__wt_txn_update_oldest(session, false));
 
 		if (!__wt_page_can_evict(session, ref, inmem_splitp))
 			return (EBUSY);

src/third_party/wiredtiger/src/include/cursor.i

@@ -270,7 +270,7 @@ __cursor_func_init(WT_CURSOR_BTREE *cbt, bool reenter)
 	 * to read.
 	 */
 	if (!F_ISSET(cbt, WT_CBT_NO_TXN))
-		__wt_txn_cursor_op(session);
+		WT_RET(__wt_txn_cursor_op(session));
 	return (0);
 }
 

src/third_party/wiredtiger/src/include/extern.h

@@ -676,8 +676,8 @@ extern void __wt_stat_join_clear_single(WT_JOIN_STATS *stats);
 extern void __wt_stat_join_clear_all(WT_JOIN_STATS **stats);
 extern void __wt_stat_join_aggregate( WT_JOIN_STATS **from, WT_JOIN_STATS *to);
 extern void __wt_txn_release_snapshot(WT_SESSION_IMPL *session);
-extern void __wt_txn_get_snapshot(WT_SESSION_IMPL *session);
-extern void __wt_txn_update_oldest(WT_SESSION_IMPL *session, bool force);
+extern int __wt_txn_get_snapshot(WT_SESSION_IMPL *session);
+extern int __wt_txn_update_oldest(WT_SESSION_IMPL *session, bool force);
 extern int __wt_txn_config(WT_SESSION_IMPL *session, const char *cfg[]);
 extern void __wt_txn_release(WT_SESSION_IMPL *session);
 extern int __wt_txn_commit(WT_SESSION_IMPL *session, const char *cfg[]);

src/third_party/wiredtiger/src/include/serial.i

@@ -306,7 +306,7 @@ __wt_update_serial(WT_SESSION_IMPL *session, WT_PAGE *page,
 	if ((txn = page->modify->obsolete_check_txn) != WT_TXN_NONE) {
 		if (!__wt_txn_visible_all(session, txn)) {
 			/* Try to move the oldest ID forward and re-check. */
-			__wt_txn_update_oldest(session, false);
+			WT_RET(__wt_txn_update_oldest(session, false));
 
 			if (!__wt_txn_visible_all(session, txn))
 				return (0);

src/third_party/wiredtiger/src/include/txn.h

@@ -74,16 +74,19 @@ struct __wt_txn_global {
 	volatile uint64_t current;	/* Current transaction ID. */
 
 	/* The oldest running transaction ID (may race). */
-	uint64_t last_running;
+	volatile uint64_t last_running;
 
 	/*
 	 * The oldest transaction ID that is not yet visible to some
 	 * transaction in the system.
 	 */
 	volatile uint64_t oldest_id;
 
-	/* Count of scanning threads, or -1 for exclusive access. */
-	volatile int32_t scan_count;
+	/*
+	 * Prevents the oldest ID moving forwards while threads are scanning
+	 * the global transaction state.
+	 */
+	WT_RWLOCK *scan_rwlock;
 
 	/*
 	 * Track information about the running checkpoint. The transaction

src/third_party/wiredtiger/src/include/txn.i

@@ -261,14 +261,14 @@ __wt_txn_begin(WT_SESSION_IMPL *session, const char *cfg[])
 		 * eviction, it's better to do it beforehand.
 		 */
 		WT_RET(__wt_cache_eviction_check(session, false, NULL));
-
-		__wt_txn_get_snapshot(session);
+		WT_RET(__wt_txn_get_snapshot(session));
 	}
 
 	F_SET(txn, WT_TXN_RUNNING);
 	if (F_ISSET(S2C(session), WT_CONN_READONLY))
 		F_SET(txn, WT_TXN_READONLY);
-	return (false);
+
+	return (0);
 }
 
 /*
@@ -450,7 +450,7 @@ __wt_txn_read_last(WT_SESSION_IMPL *session)
  * __wt_txn_cursor_op --
  *	Called for each cursor operation.
  */
-static inline void
+static inline int
 __wt_txn_cursor_op(WT_SESSION_IMPL *session)
 {
 	WT_TXN *txn;
@@ -482,7 +482,9 @@ __wt_txn_cursor_op(WT_SESSION_IMPL *session)
 		if (txn_state->snap_min == WT_TXN_NONE)
 			txn_state->snap_min = txn_global->last_running;
 	} else if (!F_ISSET(txn, WT_TXN_HAS_SNAPSHOT))
-		__wt_txn_get_snapshot(session);
+		WT_RET(__wt_txn_get_snapshot(session));
+
+	return (0);
 }
 
 /*

src/third_party/wiredtiger/src/log/log.c

@@ -8,6 +8,8 @@
 
 #include "wt_internal.h"
 
+static int __log_openfile(
+	WT_SESSION_IMPL *, bool, WT_FH **, const char *, uint32_t);
 static int __log_write_internal(
 	WT_SESSION_IMPL *, WT_ITEM *, WT_LSN *, uint32_t);
 
@@ -93,8 +95,9 @@ __wt_log_background(WT_SESSION_IMPL *session, WT_LSN *lsn)
 int
 __wt_log_force_sync(WT_SESSION_IMPL *session, WT_LSN *min_lsn)
 {
-	WT_LOG *log;
 	WT_DECL_RET;
+	WT_FH *log_fh;
+	WT_LOG *log;
 
 	log = S2C(session)->log;
 
@@ -129,12 +132,21 @@ __wt_log_force_sync(WT_SESSION_IMPL *session, WT_LSN *min_lsn)
 	 * Sync the log file if needed.
 	 */
 	if (__wt_log_cmp(&log->sync_lsn, min_lsn) < 0) {
+		/*
+		 * Get our own file handle to the log file.  It is possible
+		 * for the file handle in the log structure to change out
+		 * from under us and either be NULL or point to a different
+		 * file than we want.
+		 */
+		WT_ERR(__log_openfile(session,
+		    false, &log_fh, WT_LOG_FILENAME, min_lsn->l.file));
 		WT_ERR(__wt_verbose(session, WT_VERB_LOG,
 		    "log_force_sync: sync %s to LSN %" PRIu32 "/%" PRIu32,
-		    log->log_fh->name, min_lsn->l.file, min_lsn->l.offset));
-		WT_ERR(__wt_fsync(session, log->log_fh, true));
+		    log_fh->name, min_lsn->l.file, min_lsn->l.offset));
+		WT_ERR(__wt_fsync(session, log_fh, true));
 		log->sync_lsn = *min_lsn;
 		WT_STAT_FAST_CONN_INCR(session, log_sync);
+		WT_ERR(__wt_close(session, &log_fh));
 		WT_ERR(__wt_cond_signal(session, log->log_sync_cond));
 	}
 err:
@@ -2128,9 +2140,18 @@ __wt_log_flush(WT_SESSION_IMPL *session, uint32_t flags)
 	 * We need to flush out the current slot first to get the real
 	 * end of log LSN in log->alloc_lsn.
 	 */
-	WT_RET(__wt_log_flush_lsn(session, &lsn, 0));
+	WT_RET(__wt_log_flush_lsn(session, &lsn, false));
 	last_lsn = log->alloc_lsn;
 
+	/*
+	 * If the last write caused a switch to a new log file, we should only
+	 * wait for the last write to be flushed.  Otherwise, if the workload
+	 * is single-threaded we could wait here forever because the write LSN
+	 * doesn't switch into the new file until it contains a record.
+	 */
+	if (last_lsn.l.offset == WT_LOG_FIRST_RECORD)
+		last_lsn = log->log_close_lsn;
+
 	/*
 	 * Wait until all current outstanding writes have been written
 	 * to the file system.

src/third_party/wiredtiger/src/lsm/lsm_cursor.c

@@ -210,7 +210,7 @@ __clsm_enter(WT_CURSOR_LSM *clsm, bool reset, bool update)
 				goto open;
 
 			if (txn->isolation == WT_ISO_SNAPSHOT)
-				__wt_txn_cursor_op(session);
+				WT_RET(__wt_txn_cursor_op(session));
 
 			/*
 			 * Figure out how many updates are required for

src/third_party/wiredtiger/src/lsm/lsm_work_unit.c

@@ -289,7 +289,7 @@ __wt_lsm_checkpoint_chunk(WT_SESSION_IMPL *session,
 	}
 
 	/* Stop if a running transaction needs the chunk. */
-	__wt_txn_update_oldest(session, true);
+	WT_RET(__wt_txn_update_oldest(session, true));
 	if (chunk->switch_txn == WT_TXN_NONE ||
 	    !__wt_txn_visible_all(session, chunk->switch_txn)) {
 		WT_RET(__wt_verbose(session, WT_VERB_LSM,

src/third_party/wiredtiger/src/support/power8/crc32.S

@@ -769,3 +769,10 @@ FUNC_START(__crc32_vpmsum)
 
 FUNC_END(__crc32_vpmsum)
 #endif
+
+/*
+ * Make sure the stack isn't executable with GCC (regardless of platform).
+ */
+#ifndef __clang__
+.section	.note.GNU-stack,"",@progbits
+#endif