Imported from /home/lorry/working-area/delta_berkeleydb/db-6.1.23.tar.gz.HEADdb-6.1.23master

10 files changed, 168 insertions, 65 deletions

diff --git a/src/txn/txn.c b/src/txn/txn.c
index 81225e5c..91652cb7 100644
--- a/src/txn/txn.c
+++ b/src/txn/txn.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  */
 /*
  * Copyright (c) 1995, 1996
@@ -227,8 +227,15 @@ __txn_begin(env, ip, parent, txnpp, flags)
 	if (LF_ISSET(DB_TXN_FAMILY))
 		F_SET(txn, TXN_FAMILY | TXN_INFAMILY | TXN_READONLY);
 	if (LF_ISSET(DB_TXN_SNAPSHOT) || F_ISSET(dbenv, DB_ENV_TXN_SNAPSHOT) ||
-	    (parent != NULL && F_ISSET(parent, TXN_SNAPSHOT)))
-		F_SET(txn, TXN_SNAPSHOT);
+	    (parent != NULL && F_ISSET(parent, TXN_SNAPSHOT))) {
+		if (IS_REP_CLIENT(env)) {
+			__db_errx(env, DB_STR("4572",
+		"DB_TXN_SNAPSHOT may not be used on a replication client"));
+			ret = (EINVAL);
+			goto err;
+		} else
+			F_SET(txn, TXN_SNAPSHOT);
+	}
 	if (LF_ISSET(DB_IGNORE_LEASE))
 		F_SET(txn, TXN_IGNORE_LEASE);
 
@@ -581,8 +588,7 @@ __txn_continue(env, txn, td, ip, add_to_list)
 	txn->set_timeout = __txn_set_timeout;
 	txn->set_txn_lsnp = __txn_set_txn_lsnp;
 
-	/* XXX Do we need to explicitly set a SYNC flag here? */
-	txn->flags = TXN_MALLOC |
+	txn->flags = TXN_MALLOC | TXN_SYNC |
 	    (F_ISSET(td, TXN_DTL_NOWAIT) ? TXN_NOWAIT : 0);
 	txn->xa_thr_status = TXN_XA_THREAD_NOTA;
 
@@ -795,8 +801,9 @@ __txn_commit(txn, flags)
 				if (ret == 0) {
 					DB_LSN s_lsn;
 
-					DB_ASSERT(env, __log_current_lsn_int(
-					    env, &s_lsn, NULL, NULL) == 0);
+					if ((ret = __log_current_lsn_int(
+					    env, &s_lsn, NULL, NULL)) != 0)
+						goto err;
 					DB_ASSERT(env, LOG_COMPARE(
 					    &td->visible_lsn, &s_lsn) <= 0);
 					COMPQUIET(s_lsn.file, 0);
@@ -890,17 +897,16 @@ static int
 __txn_close_cursors(txn)
 	DB_TXN *txn;
 {
-	int ret, tret;
+	int ret, t_ret;
 	DBC *dbc;
 
-	ret = tret = 0;
+	ret = t_ret = 0;
 	dbc = NULL;
 
 	if (txn == NULL)
 		return (0);
 
 	while ((dbc = TAILQ_FIRST(&txn->my_cursors)) != NULL) {
-
 		DB_ASSERT(dbc->env, txn == dbc->txn);
 
 		/*
@@ -913,21 +919,21 @@ __txn_close_cursors(txn)
 
 		/* Removed from the active queue here. */
 		if (F_ISSET(dbc, DBC_ACTIVE))
-			ret = __dbc_close(dbc);
+			t_ret = __dbc_close(dbc);
 
 		dbc->txn = NULL;
 
 		/* We have to close all cursors anyway, so continue on error. */
-		if (ret != 0) {
-			__db_err(dbc->env, ret, "__dbc_close");
-			if (tret == 0)
-				tret = ret;
+		if (t_ret != 0) {
+			__db_err(dbc->env, t_ret, "__dbc_close");
+			if (ret == 0)
+				ret = t_ret;
 		}
 	}
 	txn->my_cursors.tqh_first = NULL;
 	txn->my_cursors.tqh_last = NULL;
 
-	return (tret);/* Return the first error if any. */
+	return (ret);	/* Return the first error, if any. */
 }
 
 /*
@@ -1050,7 +1056,7 @@ __txn_abort(txn)
 	 * it, however make sure that it is aborted when the last process
 	 * tries to abort it.
 	 */
-	if (txn->xa_thr_status != TXN_XA_THREAD_NOTA &&  td->xa_ref > 1) {
+	if (txn->xa_thr_status != TXN_XA_THREAD_NOTA && td->xa_ref > 1) {
 		td->status = TXN_NEED_ABORT;
 		return (0);
 	}
@@ -2165,5 +2171,5 @@ __txn_applied(env, ip, commit_info, timeout)
 	if (renv->envid == commit_info->envid &&
 	    LOG_COMPARE(&commit_info->lsn, &lsn) <= 0)
 		return (0);
-	return (DB_NOTFOUND);
+	return (USR_ERR(env, DB_NOTFOUND));
 }
diff --git a/src/txn/txn.src b/src/txn/txn.src
index 7e82dc82..d9af5318 100644
--- a/src/txn/txn.src
+++ b/src/txn/txn.src
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
diff --git a/src/txn/txn_chkpt.c b/src/txn/txn_chkpt.c
index 73715b10..a909767f 100644
--- a/src/txn/txn_chkpt.c
+++ b/src/txn/txn_chkpt.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  */
 /*
  * Copyright (c) 1995, 1996
@@ -377,7 +377,7 @@ __txn_getckp(env, lsnp)
 	TXN_SYSTEM_UNLOCK(env);
 
 	if (IS_ZERO_LSN(lsn))
-		return (DB_NOTFOUND);
+		return (USR_ERR(env, DB_NOTFOUND));
 
 	*lsnp = lsn;
 	return (0);
diff --git a/src/txn/txn_failchk.c b/src/txn/txn_failchk.c
index b2007ad6..94f22ec2 100644
--- a/src/txn/txn_failchk.c
+++ b/src/txn/txn_failchk.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 2005, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 2005, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
@@ -57,7 +57,7 @@ retry:	TXN_SYSTEM_LOCK(env);
 
 		if (F_ISSET(td, TXN_DTL_INMEMORY)) {
 			TXN_SYSTEM_UNLOCK(env);
-			return (__db_failed(env, DB_STR("4501",
+			return (__db_failed(env, DB_STR("4573",
 			    "Transaction has in memory logs"),
 			     td->pid, td->tid));
 		}
diff --git a/src/txn/txn_method.c b/src/txn/txn_method.c
index 629eac04..357e78c6 100644
--- a/src/txn/txn_method.c
+++ b/src/txn/txn_method.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
diff --git a/src/txn/txn_rec.c b/src/txn/txn_rec.c
index b39d56d1..708af98a 100644
--- a/src/txn/txn_rec.c
+++ b/src/txn/txn_rec.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  */
 /*
  * Copyright (c) 1996
@@ -210,11 +210,12 @@ __txn_prepare_recover(env, dbtp, lsnp, op, info)
 		 */
 		else if ((ret = __db_txnlist_remove(env,
 		    info, argp->txnp->txnid)) != 0) {
-txn_err:		__db_errx(env,
+txn_err:
+			ret = USR_ERR(env, DB_NOTFOUND);
+			__db_errx(env,
 			    DB_STR_A("4515",
 			    "transaction not in list %lx", "%lx"),
 			    (u_long)argp->txnp->txnid);
-			ret = DB_NOTFOUND;
 		} else if (IS_ZERO_LSN(headp->trunc_lsn) ||
 		    LOG_COMPARE(&headp->trunc_lsn, lsnp) >= 0) {
 			if ((ret = __db_txnlist_add(env,
diff --git a/src/txn/txn_recover.c b/src/txn/txn_recover.c
index 67f24439..915a289f 100644
--- a/src/txn/txn_recover.c
+++ b/src/txn/txn_recover.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 2001, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 2001, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
@@ -305,8 +305,8 @@ __txn_openfiles(env, ip, min, force)
 
 	if ((ret = __db_txnlist_init(env, ip, 0, 0, NULL, &txninfo)) != 0)
 		goto err;
-	ret = __env_openfiles(
-	    env, logc, txninfo, &data, &open_lsn, NULL, (double)0, 0);
+	ret = __env_openfiles(env,
+	    logc, txninfo, &data, &open_lsn, NULL, (double)0, 0);
 	if (txninfo != NULL)
 		__db_txnlist_end(env, txninfo);
 
diff --git a/src/txn/txn_region.c b/src/txn/txn_region.c
index 6f43d45f..7fef66e6 100644
--- a/src/txn/txn_region.c
+++ b/src/txn/txn_region.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
@@ -13,6 +13,7 @@
 #include "dbinc/txn.h"
 
 static int __txn_init __P((ENV *, DB_TXNMGR *));
+static int lsn_hi_to_low __P((const void *, const void *));
 
 /*
  * __txn_open --
@@ -57,12 +58,30 @@ __txn_open(env)
 	env->tx_handle = mgr;
 	return (0);
 
-err:	env->tx_handle = NULL;
-	if (mgr->reginfo.addr != NULL)
-		(void)__env_region_detach(env, &mgr->reginfo, 0);
+err:	(void)__mutex_free(env, &mgr->mutex);
+	(void)__txn_region_detach(env, mgr);
 
-	(void)__mutex_free(env, &mgr->mutex);
-	__os_free(env, mgr);
+	return (ret);
+}
+
+/*
+ * __txn_region_detach --
+ *
+ * PUBLIC: int __txn_region_detach __P((ENV *, DB_TXNMGR *));
+ */
+int
+__txn_region_detach(env, mgr)
+	ENV *env;
+	DB_TXNMGR *mgr;
+{
+	int ret;
+
+	ret = 0;
+	if (mgr != NULL) {
+		ret = __env_region_detach(env, &mgr->reginfo, 0);
+		__os_free(env, mgr);
+		env->tx_handle = NULL;
+	}
 	return (ret);
 }
 
@@ -409,39 +428,101 @@ __txn_id_set(env, cur_txnid, max_txnid)
 }
 
 /*
- * __txn_oldest_reader --
- *	 Find the oldest "read LSN" of any active transaction'
- *	 MVCC changes older than this can safely be discarded from the cache.
+ * lsn_hi_to_low --
+ *	Compare lsns, sorting them from high to low. This is the opposite of
+ *	__rep_lsn_cmp.
+ */
+static int
+lsn_hi_to_low(lsn1, lsn2)
+	const void *lsn1, *lsn2;
+{
+	return (LOG_COMPARE((DB_LSN *)lsn2, (DB_LSN *)lsn1));
+}
+
+/*
+ * __txn_get_readers --
+ *	Find the read LSN of all active transactions.
+ *	MVCC versions older than the oldest active transaction can safely be
+ *	discarded from the cache. MVCC versions not quite so old can be
+ *      discarded if they are not visible to any active transaction.
  *
- * PUBLIC: int __txn_oldest_reader __P((ENV *, DB_LSN *));
+ * Returns:
+ *	An error code, or 0.
+ *	If 0 was returned, *readers has been filled in with an __os_malloc()'d
+ *	array of active transactions with read_lsns, sorted from newest
+ *      (largest) to oldest (smallest). *ntxnsp indicates how many are there.
+ *	The last lsn is that of the oldest active mvcc-supporting transaction.
+ *	The caller must __os_free() *readers whenever it is non-NULL.
+ *
+ * PUBLIC: int __txn_get_readers __P((ENV *, DB_LSN **, int *));
  */
+#define	TXN_READERS_SIZE	64 /* Initial number of LSNs to allocate. */
 int
-__txn_oldest_reader(env, lsnp)
+__txn_get_readers(env, readers, ntxnsp)
 	ENV *env;
-	DB_LSN *lsnp;
+	DB_LSN **readers;
+	int *ntxnsp;
 {
-	DB_LSN old_lsn;
+	DB_LSN current, *lsns;
 	DB_TXNMGR *mgr;
 	DB_TXNREGION *region;
 	TXN_DETAIL *td;
-	int ret;
+	int cmp, is_sorted, ret;
+	unsigned count, txnmax;
+
+	*ntxnsp = 0;
+	*readers = NULL;
 
 	if ((mgr = env->tx_handle) == NULL)
 		return (0);
 	region = mgr->reginfo.primary;
+	lsns = NULL;
+
+	if ((ret = __log_current_lsn_int(env, &current, NULL, NULL)) != 0)
+		return (ret);
 
-	if ((ret = __log_current_lsn_int(env, &old_lsn, NULL, NULL)) != 0)
+	txnmax = TXN_READERS_SIZE;
+	if ((ret = __os_malloc(env, txnmax * sizeof(lsns[0]), &lsns)) != 0)
 		return (ret);
 
 	TXN_SYSTEM_LOCK(env);
-	SH_TAILQ_FOREACH(td, &region->active_txn, links, __txn_detail)
-		if (LOG_COMPARE(&td->read_lsn, &old_lsn) < 0)
-			old_lsn = td->read_lsn;
+	/* The array always has at least the current lsn. */
+	lsns[0] = current;
+	count = 1;
+	is_sorted = TRUE;
 
-	*lsnp = old_lsn;
+	/*
+	 * Build up our array in most-recent (largest) to first-started (oldest)
+	 * order. Delete adjacent dups. Detect when the txns need to be sorted.
+	 */
+	SH_TAILQ_FOREACH(td, &region->active_txn, links, __txn_detail) {
+		if (IS_MAX_LSN(td->read_lsn) ||
+		    (cmp = LOG_COMPARE(&td->read_lsn, &lsns[count - 1])) == 0)
+			continue;
+		if (cmp > 0)
+			is_sorted = FALSE;
+		if (count >= txnmax) {
+			txnmax += txnmax;
+			if ((ret = __os_realloc(env,
+			    txnmax * sizeof(lsns[0]), &lsns)) != 0)
+				goto err;
+		}
+		lsns[count] = td->read_lsn;
+		count++;
+	}
+
+err:
 	TXN_SYSTEM_UNLOCK(env);
 
-	return (0);
+	if (ret != 0)
+		__os_free(env, lsns);
+	else {
+		if (!is_sorted)
+			qsort(lsns, count, sizeof(lsns[0]), lsn_hi_to_low);
+		*ntxnsp = (int)count;
+		*readers = lsns;
+	}
+	return (ret);
 }
 
 /*
diff --git a/src/txn/txn_stat.c b/src/txn/txn_stat.c
index 62fe622d..231ac3c5 100644
--- a/src/txn/txn_stat.c
+++ b/src/txn/txn_stat.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 1996, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 1996, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
diff --git a/src/txn/txn_util.c b/src/txn/txn_util.c
index 0ecd7f6c..9f3b8cf6 100644
--- a/src/txn/txn_util.c
+++ b/src/txn/txn_util.c
@@ -1,7 +1,7 @@
 /*-
  * See the file LICENSE for redistribution information.
  *
- * Copyright (c) 2001, 2012 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 2001, 2015 Oracle and/or its affiliates.  All rights reserved.
  *
  * $Id$
  */
@@ -9,6 +9,7 @@
 #include "db_config.h"
 
 #include "db_int.h"
+#include "dbinc/blob.h"
 #include "dbinc/db_page.h"
 #include "dbinc/lock.h"
 #include "dbinc/mp.h"
@@ -209,7 +210,7 @@ __txn_remlock(env, txn, lock, locker)
 
 	for (e = TAILQ_FIRST(&txn->events); e != NULL; e = next_e) {
 		next_e = TAILQ_NEXT(e, links);
-		if ((e->op != TXN_TRADE && e->op != TXN_TRADED && 
+		if ((e->op != TXN_TRADE && e->op != TXN_TRADED &&
 		    e->op != TXN_XTRADE) ||
 		    (e->u.t.lock.off != lock->off && e->u.t.locker != locker))
 			continue;
@@ -280,13 +281,21 @@ __txn_doevents(env, txn, opcode, preprocess)
 		    e != NULL; e = enext) {
 			enext = TAILQ_NEXT(e, links);
 			/*
-			 * Move all exclusive handle locks and 
+			 * Move all exclusive handle locks and
 			 * read handle locks to the handle locker.
 			 */
 			if (!(opcode == TXN_COMMIT && e->op == TXN_XTRADE) &&
-			    (e->op != TXN_TRADE || 
-			    IS_WRITELOCK(e->u.t.lock.mode)))
+			    (e->op != TXN_TRADE ||
+			    IS_WRITELOCK(e->u.t.lock.mode))) {
+				if (opcode == TXN_PREPARE &&
+				    e->op == TXN_REMOVE) {
+					__db_errx(env, DB_STR_A("4501",
+"TXN->prepare is not allowed because this transaction removes \"%s\"", "%s"),
+					    e->u.r.name);
+					return (EINVAL);
+				}
 				continue;
+			}
 			DO_TRADE;
 			if (txn->parent != NULL) {
 				TAILQ_REMOVE(&txn->events, e, links);
@@ -321,17 +330,26 @@ __txn_doevents(env, txn, opcode, preprocess)
 				ret = t_ret;
 			break;
 		case TXN_REMOVE:
-			if (txn->parent != NULL)
+			if (txn->parent != NULL) {
 				TAILQ_INSERT_TAIL(
 				    &txn->parent->events, e, links);
-			else if (e->u.r.fileid != NULL) {
+				continue;
+			} else if (e->u.r.fileid != NULL) {
 				if ((t_ret = __memp_nameop(env,
 				    e->u.r.fileid, NULL, e->u.r.name,
 				    NULL, e->u.r.inmem)) != 0 && ret == 0)
 					ret = t_ret;
-			} else if ((t_ret =
-			    __os_unlink(env, e->u.r.name, 0)) != 0 && ret == 0)
-				ret = t_ret;
+			} else if ((t_ret = __os_unlink(
+			    env, e->u.r.name, 0)) != 0 && ret == 0) {
+				/*
+				 * It is possible for blob files to be deleted
+				 * multiple times when truncating a database,
+				 * so ignore ENOENT errors with blob files.
+				 */
+				if (t_ret != ENOENT || strstr(
+				    e->u.r.name, BLOB_FILE_PREFIX) == NULL)
+					ret = t_ret;
+			}
 			break;
 		case TXN_TRADE:
 		case TXN_XTRADE:
@@ -371,8 +389,6 @@ dofree:
 		/* Free resources here. */
 		switch (e->op) {
 		case TXN_REMOVE:
-			if (txn->parent != NULL)
-				continue;
 			if (e->u.r.fileid != NULL)
 				__os_free(env, e->u.r.fileid);
 			__os_free(env, e->u.r.name);
@@ -548,9 +564,8 @@ __txn_reset_fe_watermarks(txn)
 {
 	DB *db;
 
-	if (txn->parent) {
+	if (txn->parent)
 		DB_ASSERT(txn->mgrp->env, TAILQ_FIRST(&txn->femfs) == NULL);
-	}
 
 	while ((db = TAILQ_FIRST(&txn->femfs)))
 		__clear_fe_watermark(txn, db);