From 0144da6eab00cf5f321d5c6794175485ba3efefd Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@symas.com>
Date: Sat, 21 Sep 2013 14:37:11 -0700
Subject: [PATCH 01/25] ITS#7701 fix mdb_rebalance

Must copy tmp cursor back to real cursor when merging into tmp cursor.
---
 libraries/liblmdb/mdb.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 9c5e7389de..e5313b0c85 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -7248,8 +7248,11 @@ mdb_rebalance(MDB_cursor *mc)
 	else {
 		if (mc->mc_ki[ptop] == 0)
 			rc = mdb_page_merge(&mn, mc);
-		else
+		else {
+			mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
 			rc = mdb_page_merge(mc, &mn);
+			mdb_cursor_copy(&mn, mc);
+		}
 		mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
 	}
 	return rc;

From 912e09fd04e0512e6f764176a4a0a2ab009c87d1 Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@symas.com>
Date: Sun, 15 Sep 2013 13:08:29 -0700
Subject: [PATCH 02/25] ITS#7681 fix 18a07eb7c2dc33372455a6040984cd6b699b41a5

Set rc=0 when taking the SET_RANGE branch
---
 libraries/liblmdb/mdb.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index e5313b0c85..e30d85c476 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -5231,9 +5231,10 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 		if (!mc->mc_top) {
 			/* There are no other pages */
 			mc->mc_ki[mc->mc_top] = 0;
-			if (op == MDB_SET_RANGE)
+			if (op == MDB_SET_RANGE) {
+				rc = 0;
 				goto set1;
-			else 
+			} else
 				return MDB_NOTFOUND;
 		}
 	}
@@ -5298,6 +5299,7 @@ set1:
 			if (rc) {
 				if (op == MDB_GET_BOTH || rc > 0)
 					return MDB_NOTFOUND;
+				rc = 0;
 			}
 
 		} else {

From 3335b2583490f10c3f273391b48e1e2987022f4c Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:07:29 +0200
Subject: [PATCH 03/25] ITS#7682 mdb_env_copy(): Avoid Linux O_DIRECT bug.

Use fcntl() to set the flag.  Linux open(,O_DIRECT...) can create
the file even on failure, if the filesystem lacks O_DIRECT support.
---
 libraries/liblmdb/mdb.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index e30d85c476..46dea53c87 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -4246,14 +4246,6 @@ mdb_env_copy(MDB_env *env, const char *path)
 	newfd = CreateFile(lpath, GENERIC_WRITE, 0, NULL, CREATE_NEW,
 				FILE_FLAG_NO_BUFFERING|FILE_FLAG_WRITE_THROUGH, NULL);
 #else
-#ifdef O_DIRECT
-	/* The OS supports O_DIRECT, try with it */
-	newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL|O_DIRECT, 0666);
-	/* But open can fail if O_DIRECT isn't supported by the file system
-	 * so retry without the flag
-	 */
-	if (newfd == INVALID_HANDLE_VALUE && ErrCode() == EINVAL)
-#endif
 	newfd = open(lpath, O_WRONLY|O_CREAT|O_EXCL, 0666);
 #endif
 	if (newfd == INVALID_HANDLE_VALUE) {
@@ -4261,6 +4253,11 @@ mdb_env_copy(MDB_env *env, const char *path)
 		goto leave;
 	}
 
+#ifdef O_DIRECT
+	/* Set O_DIRECT if the file system supports it */
+	if ((rc = fcntl(newfd, F_GETFL)) != -1)
+		(void) fcntl(newfd, F_SETFL, rc | O_DIRECT);
+#endif
 #ifdef F_NOCACHE	/* __APPLE__ */
 	rc = fcntl(newfd, F_NOCACHE, 1);
 	if (rc) {

From 5b21307f29ccda563e5f0d82294083c841b0e927 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:09:47 +0200
Subject: [PATCH 04/25] Update MDB documentation and comments.

---
 libraries/liblmdb/lmdb.h | 18 +++++++++++-------
 libraries/liblmdb/mdb.c  | 41 +++++++++++++++++++++++-----------------
 libraries/liblmdb/midl.h |  2 +-
 3 files changed, 36 insertions(+), 25 deletions(-)

diff --git a/libraries/liblmdb/lmdb.h b/libraries/liblmdb/lmdb.h
index 8a46c1d0d1..f637229283 100644
--- a/libraries/liblmdb/lmdb.h
+++ b/libraries/liblmdb/lmdb.h
@@ -216,13 +216,13 @@ typedef struct MDB_cursor MDB_cursor;
 /** @brief Generic structure used for passing keys and data in and out
  * of the database.
  *
- * Key sizes must be between 1 and the liblmdb build-time constant
- * #MDB_MAXKEYSIZE inclusive. This currently defaults to 511. The
- * same applies to data sizes in databases with the #MDB_DUPSORT flag.
- * Other data items can in theory be from 0 to 0xffffffff bytes long.
- *
  * Values returned from the database are valid only until a subsequent
- * update operation, or the end of the transaction.
+ * update operation, or the end of the transaction. Do not modify or
+ * free them, they commonly point into the database itself.
+ *
+ * Key sizes must be between 1 and #mdb_env_get_maxkeysize() inclusive.
+ * The same applies to data sizes in databases with the #MDB_DUPSORT flag.
+ * Other data items can in theory be from 0 to 0xffffffff bytes long.
  */
 typedef struct MDB_val {
 	size_t		 mv_size;	/**< size of the data item */
@@ -486,6 +486,8 @@ int  mdb_env_create(MDB_env **env);
 	 *		and uses fewer mallocs, but loses protection from application bugs
 	 *		like wild pointer writes and other bad updates into the database.
 	 *		Incompatible with nested transactions.
+	 *		Processes with and without MDB_WRITEMAP on the same environment do
+	 *		not cooperate well. 
 	 *	<li>#MDB_NOMETASYNC
 	 *		Flush system buffers to disk only once per transaction, omit the
 	 *		metadata flush. Defer that until the system flushes files to disk,
@@ -733,8 +735,10 @@ int  mdb_env_set_maxdbs(MDB_env *env, MDB_dbi dbs);
 
 	/** @brief Get the maximum size of a key for the environment.
 	 *
+	 * This is the compile-time constant #MDB_MAXKEYSIZE, default 511.
+	 * See @ref MDB_val.
 	 * @param[in] env An environment handle returned by #mdb_env_create()
-	 * @return The maximum size of a key. (#MDB_MAXKEYSIZE)
+	 * @return The maximum size of a key
 	 */
 int  mdb_env_get_maxkeysize(MDB_env *env);
 
diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 46dea53c87..0bc9db798f 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -171,7 +171,7 @@
 #define	Z	"I"
 #else
 
-#define	Z	"z"
+#define	Z	"z"			/**< printf format modifier for size_t */
 
 	/** For MDB_LOCK_FORMAT: True if readers take a pid lock in the lockfile */
 #define MDB_PIDLOCK			1
@@ -600,7 +600,7 @@ typedef struct MDB_page {
 #define	P_LEAF		 0x02		/**< leaf page */
 #define	P_OVERFLOW	 0x04		/**< overflow page */
 #define	P_META		 0x08		/**< meta page */
-#define	P_DIRTY		 0x10		/**< dirty page */
+#define	P_DIRTY		 0x10		/**< dirty page, also set for #P_SUBP pages */
 #define	P_LEAF2		 0x20		/**< for #MDB_DUPFIXED records */
 #define	P_SUBP		 0x40		/**< for #MDB_DUPSORT sub-pages */
 #define	P_KEEP		 0x8000		/**< leave this page alone during spill */
@@ -786,7 +786,10 @@ typedef struct MDB_db {
 	/** Handle for the default DB. */
 #define	MAIN_DBI	1
 
-	/** Meta page content. */
+	/** Meta page content.
+	 *	A meta page is the start point for accessing a database snapshot.
+	 *	Pages 0-1 are meta pages. Transaction N writes meta page #(N % 2).
+	 */
 typedef struct MDB_meta {
 		/** Stamp identifying this as an MDB file. It must be set
 		 *	to #MDB_MAGIC. */
@@ -905,7 +908,14 @@ struct MDB_txn {
 
 struct MDB_xcursor;
 
-	/** Cursors are used for all DB operations */
+	/** Cursors are used for all DB operations.
+	 *	A cursor holds a path of (page pointer, key index) from the DB
+	 *	root to a position in the DB, plus other state. #MDB_DUPSORT
+	 *	cursors include an xcursor to the current data item. Write txns
+	 *	track their cursors and keep them up to date when data moves.
+	 *	Exception: An xcursor's pointer to a #P_SUBP page can be stale.
+	 *	(A node with #F_DUPDATA but no #F_SUBDATA contains a subpage).
+	 */
 struct MDB_cursor {
 	/** Next cursor on this DB in this txn */
 	MDB_cursor	*mc_next;
@@ -1019,8 +1029,8 @@ struct MDB_env {
 
 	/** Nested transaction */
 typedef struct MDB_ntxn {
-	MDB_txn		mnt_txn;		/* the transaction */
-	MDB_pgstate	mnt_pgstate;	/* parent transaction's saved freestate */
+	MDB_txn		mnt_txn;		/**< the transaction */
+	MDB_pgstate	mnt_pgstate;	/**< parent transaction's saved freestate */
 } MDB_ntxn;
 
 	/** max number of pages to commit in one writev() call */
@@ -1329,7 +1339,7 @@ mdb_page_free(MDB_env *env, MDB_page *mp)
 	env->me_dpages = mp;
 }
 
-/* Free a dirty page */
+/** Free a dirty page */
 static void
 mdb_dpage_free(MDB_env *env, MDB_page *dp)
 {
@@ -1356,7 +1366,7 @@ mdb_dlist_free(MDB_txn *txn)
 	dl[0].mid = 0;
 }
 
-/* Set or clear P_KEEP in dirty, non-overflow, non-sub pages watched by txn.
+/** Set or clear P_KEEP in dirty, non-overflow, non-sub pages watched by txn.
  * @param[in] mc A cursor handle for the current operation.
  * @param[in] pflags Flags of the pages to update:
  * P_DIRTY to set P_KEEP, P_DIRTY|P_KEEP to clear it.
@@ -1415,15 +1425,12 @@ static int mdb_page_flush(MDB_txn *txn, int keep);
 /**	Spill pages from the dirty list back to disk.
  * This is intended to prevent running into #MDB_TXN_FULL situations,
  * but note that they may still occur in a few cases:
- *	1) pages in #MDB_DUPSORT sub-DBs are never spilled, so if there
- *	 are too many of these dirtied in one txn, the txn may still get
- *	 too full.
+ *	1) our estimate of the txn size could be too small. Currently this
+ *	 seems unlikely, except with a large number of #MDB_MULTIPLE items.
  *	2) child txns may run out of space if their parents dirtied a
  *	 lot of pages and never spilled them. TODO: we probably should do
  *	 a preemptive spill during #mdb_txn_begin() of a child txn, if
  *	 the parent's dirty_room is below a given threshold.
- *	3) our estimate of the txn size could be too small. At the
- *	 moment this seems unlikely.
  *
  * Otherwise, if not using nested txns, it is expected that apps will
  * not run into #MDB_TXN_FULL any more. The pages are flushed to disk
@@ -2585,7 +2592,7 @@ mdb_freelist_save(MDB_txn *txn)
 		total_room += head_room;
 	}
 
-	/* Fill in the reserved, touched me_pghead records */
+	/* Fill in the reserved me_pghead records */
 	rc = MDB_SUCCESS;
 	if (mop_len) {
 		MDB_val key, data;
@@ -4305,7 +4312,7 @@ mdb_cmp_long(const MDB_val *a, const MDB_val *b)
 		*(size_t *)a->mv_data > *(size_t *)b->mv_data;
 }
 
-/** Compare two items pointing at aligned int's */
+/** Compare two items pointing at aligned unsigned int's */
 static int
 mdb_cmp_int(const MDB_val *a, const MDB_val *b)
 {
@@ -4313,7 +4320,7 @@ mdb_cmp_int(const MDB_val *a, const MDB_val *b)
 		*(unsigned int *)a->mv_data > *(unsigned int *)b->mv_data;
 }
 
-/** Compare two items pointing at ints of unknown alignment.
+/** Compare two items pointing at unsigned ints of unknown alignment.
  *	Nodes and keys are guaranteed to be 2-byte aligned.
  */
 static int
@@ -8270,7 +8277,7 @@ int mdb_reader_list(MDB_env *env, MDB_msg_func *func, void *ctx)
 	return 0;
 }
 
-/* insert pid into list if not already present.
+/** Insert pid into list if not already present.
  * return -1 if already present.
  */
 static int mdb_pid_insert(pid_t *ids, pid_t pid)
diff --git a/libraries/liblmdb/midl.h b/libraries/liblmdb/midl.h
index b0bdff3f49..635cd29e0e 100644
--- a/libraries/liblmdb/midl.h
+++ b/libraries/liblmdb/midl.h
@@ -39,7 +39,7 @@ extern "C" {
 /** @defgroup idls	ID List Management
  *	@{
  */
-	/** A generic ID number. These were entryIDs in back-bdb.
+	/** A generic unsigned ID number. These were entryIDs in back-bdb.
 	 *	Preferably it should have the same size as a pointer.
 	 */
 typedef size_t MDB_ID;

From 5b96d68fafd1732790ac5f887e5050033a48da42 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:13:27 +0200
Subject: [PATCH 05/25] mdb_pages_xkeep(): Reformat, use common flag mask

---
 libraries/liblmdb/mdb.c | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 0bc9db798f..b5a3d8242f 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -1376,10 +1376,11 @@ mdb_dlist_free(MDB_txn *txn)
 static int
 mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
 {
+	enum { Mask = P_SUBP|P_DIRTY|P_KEEP };
 	MDB_txn *txn = mc->mc_txn;
 	MDB_cursor *m3;
 	MDB_xcursor *mx;
-	MDB_page *dp;
+	MDB_page *dp, *mp;
 	unsigned i, j;
 	int rc = MDB_SUCCESS, level;
 
@@ -1389,13 +1390,14 @@ mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
 	for (i = txn->mt_numdbs;; mc = txn->mt_cursors[--i]) {
 		for (; mc; mc=mc->mc_next) {
 			for (m3 = mc; m3->mc_flags & C_INITIALIZED; m3 = &mx->mx_cursor) {
-					for (j=0; j<m3->mc_snum; j++)
-						if ((m3->mc_pg[j]->mp_flags & (P_SUBP|P_DIRTY|P_KEEP))
-								== pflags)
-							m3->mc_pg[j]->mp_flags ^= P_KEEP;
-					mx = m3->mc_xcursor;
-					if (mx == NULL)
-						break;
+				for (j=0; j<m3->mc_snum; j++) {
+					mp = m3->mc_pg[j];
+					if ((mp->mp_flags & Mask) == pflags)
+						mp->mp_flags ^= P_KEEP;
+				}
+				mx = m3->mc_xcursor;
+				if (mx == NULL)
+					break;
 			}
 		}
 		if (i == 0)
@@ -1411,7 +1413,7 @@ mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
 					continue;
 				if ((rc = mdb_page_get(txn, pgno, &dp, &level)) != MDB_SUCCESS)
 					break;
-				if ((dp->mp_flags & (P_DIRTY|P_KEEP)) == pflags && level <= 1)
+				if ((dp->mp_flags & Mask) == pflags && level <= 1)
 					dp->mp_flags ^= P_KEEP;
 			}
 		}

From cfe262dce9a0be4791db4720aa308d5a6ecb8888 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:13:27 +0200
Subject: [PATCH 06/25] ITS#7515 mdb_pages_xkeep(): Careful about xcursors.

Don't prod sub-pages or pages referring to uninitialized xcursors.
---
 libraries/liblmdb/mdb.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index b5a3d8242f..093cb53227 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -1381,6 +1381,7 @@ mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
 	MDB_cursor *m3;
 	MDB_xcursor *mx;
 	MDB_page *dp, *mp;
+	MDB_node *leaf;
 	unsigned i, j;
 	int rc = MDB_SUCCESS, level;
 
@@ -1389,14 +1390,23 @@ mdb_pages_xkeep(MDB_cursor *mc, unsigned pflags, int all)
 		mc = NULL;				/* will find mc in mt_cursors */
 	for (i = txn->mt_numdbs;; mc = txn->mt_cursors[--i]) {
 		for (; mc; mc=mc->mc_next) {
-			for (m3 = mc; m3->mc_flags & C_INITIALIZED; m3 = &mx->mx_cursor) {
+			if (!(mc->mc_flags & C_INITIALIZED))
+				continue;
+			for (m3 = mc;; m3 = &mx->mx_cursor) {
+				mp = NULL;
 				for (j=0; j<m3->mc_snum; j++) {
 					mp = m3->mc_pg[j];
 					if ((mp->mp_flags & Mask) == pflags)
 						mp->mp_flags ^= P_KEEP;
 				}
 				mx = m3->mc_xcursor;
-				if (mx == NULL)
+				/* Proceed to mx if it is at a sub-database */
+				if (! (mx && (mx->mx_cursor.mc_flags & C_INITIALIZED)))
+					break;
+				if (! (mp && (mp->mp_flags & P_LEAF)))
+					break;
+				leaf = NODEPTR(mp, m3->mc_ki[j-1]);
+				if (!(leaf->mn_flags & F_SUBDATA))
 					break;
 			}
 		}

From bc48a40621ee26f15503dbd16321c972648b4769 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:13:27 +0200
Subject: [PATCH 07/25] ITS#7515 Fix mt_dirty_room in nested txns.

Fix description & code: Also ignore dirty pages hidden by
spilled pages, as they won't merge into our dirty_list.
Update it in mdb_page_flush() instead of mdb_page_spill().
---
 libraries/liblmdb/mdb.c | 57 ++++++++++-------------------------------
 1 file changed, 13 insertions(+), 44 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 093cb53227..b49977df8c 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -892,7 +892,11 @@ struct MDB_txn {
 #define MDB_TXN_SPILLS		0x08		/**< txn or a parent has spilled pages */
 /** @} */
 	unsigned int	mt_flags;		/**< @ref mdb_txn */
-	/** dirty_list maxsize - # of allocated pages allowed, including in parent txns */
+	/** dirty_list room: Array size - #dirty pages visible to this txn.
+	 *	Includes ancestor txns' dirty pages not hidden by other txns'
+	 *	dirty/spilled pages. Thus commit(nested txn) has room to merge
+	 *	dirty_list into mt_parent after freeing hidden mt_parent pages.
+	 */
 	unsigned int	mt_dirty_room;
 	/** Tracks which of the two meta pages was used at the start
 	 * 	of this transaction.
@@ -1560,31 +1564,7 @@ mdb_page_spill(MDB_cursor *m0, MDB_val *key, MDB_val *data)
 	rc = mdb_pages_xkeep(m0, P_DIRTY|P_KEEP, i);
 
 done:
-	if (rc == 0) {
-		if (txn->mt_parent) {
-			txn->mt_dirty_room = txn->mt_parent->mt_dirty_room - dl[0].mid;
-			/* dirty pages that are dirty in an ancestor don't
-			 * count against this txn's dirty_room.
-			 */
-			for (i=1; i<=dl[0].mid; i++) {
-				pgno_t pgno = dl[i].mid;
-				MDB_txn *tx2;
-				for (tx2 = txn->mt_parent; tx2; tx2 = tx2->mt_parent) {
-					j = mdb_mid2l_search(tx2->mt_u.dirty_list, pgno);
-					if (j <= tx2->mt_u.dirty_list[0].mid &&
-						tx2->mt_u.dirty_list[j].mid == pgno) {
-						txn->mt_dirty_room++;
-						break;
-					}
-				}
-			}
-		} else {
-			txn->mt_dirty_room = MDB_IDL_UM_MAX - dl[0].mid;
-		}
-		txn->mt_flags |= MDB_TXN_SPILLS;
-	} else {
-		txn->mt_flags |= MDB_TXN_ERROR;
-	}
+	txn->mt_flags |= rc ? MDB_TXN_ERROR : MDB_TXN_SPILLS;
 	return rc;
 }
 
@@ -1829,6 +1809,8 @@ mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
 		if (x <= txn->mt_spill_pgs[0] && txn->mt_spill_pgs[x] == pn) {
 			MDB_page *np;
 			int num;
+			if (txn->mt_dirty_room == 0)
+				return MDB_TXN_FULL;
 			if (IS_OVERFLOW(mp))
 				num = mp->mp_pages;
 			else
@@ -1857,21 +1839,6 @@ mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
 				 * page remains spilled until child commits
 				 */
 
-			if (txn->mt_parent) {
-				MDB_txn *tx2;
-				/* If this page is also in a parent's dirty list, then
-				 * it's already accounted in dirty_room, and we need to
-				 * cancel out the decrement that mdb_page_dirty does.
-				 */
-				for (tx2 = txn->mt_parent; tx2; tx2 = tx2->mt_parent) {
-					x = mdb_mid2l_search(tx2->mt_u.dirty_list, pgno);
-					if (x <= tx2->mt_u.dirty_list[0].mid &&
-						tx2->mt_u.dirty_list[x].mid == pgno) {
-						tx0->mt_dirty_room++;
-						break;
-					}
-				}
-			}
 			mdb_page_dirty(tx0, np);
 			np->mp_flags |= P_DIRTY;
 			*ret = np;
@@ -2674,8 +2641,7 @@ mdb_page_flush(MDB_txn *txn, int keep)
 			}
 			dp->mp_flags &= ~P_DIRTY;
 		}
-		dl[0].mid = j;
-		return MDB_SUCCESS;
+		goto done;
 	}
 
 	/* Write the pages */
@@ -2769,8 +2735,11 @@ mdb_page_flush(MDB_txn *txn, int keep)
 		}
 		mdb_dpage_free(env, dp);
 	}
-	dl[0].mid = j;
 
+done:
+	i--;
+	txn->mt_dirty_room += i - j;
+	dl[0].mid = j;
 	return MDB_SUCCESS;
 }
 

From a3b3482854698f520135a49de930df4b788694dd Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:13:27 +0200
Subject: [PATCH 08/25] ITS#7515 Fix mdb_txn_commit(nested txn with spills).

Catch malloc error.
Fix hunt for dirty vs spilled pages: Don't leave x at a deleted pageno.
Cleanup: Factor out variables, squash pages already marked for deletion.
---
 libraries/liblmdb/mdb.c | 54 +++++++++++++++++++----------------------
 1 file changed, 25 insertions(+), 29 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index b49977df8c..84429ee3d6 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -2779,14 +2779,18 @@ mdb_txn_commit(MDB_txn *txn)
 
 	if (txn->mt_parent) {
 		MDB_txn *parent = txn->mt_parent;
-		unsigned x, y, len;
 		MDB_ID2L dst, src;
+		MDB_IDL pspill;
+		unsigned x, y, len, ps_len;
 
 		/* Append our free list to parent's */
 		rc = mdb_midl_append_list(&parent->mt_free_pgs, txn->mt_free_pgs);
 		if (rc)
 			goto fail;
 		mdb_midl_free(txn->mt_free_pgs);
+		/* Failures after this must either undo the changes
+		 * to the parent or set MDB_TXN_ERROR in the parent.
+		 */
 
 		parent->mt_next_pgno = txn->mt_next_pgno;
 		parent->mt_flags = txn->mt_flags;
@@ -2808,37 +2812,26 @@ mdb_txn_commit(MDB_txn *txn)
 		dst = parent->mt_u.dirty_list;
 		src = txn->mt_u.dirty_list;
 		/* Remove anything in our dirty list from parent's spill list */
-		if (parent->mt_spill_pgs) {
-			x = parent->mt_spill_pgs[0];
-			len = x;
-			/* zero out our dirty pages in parent spill list */
-			for (i=1; i<=src[0].mid; i++) {
+		if ((pspill = parent->mt_spill_pgs) && (ps_len = pspill[0])) {
+			x = y = ps_len;
+			pspill[0] = (pgno_t)-1;
+			/* Mark our dirty pages as deleted in parent spill list */
+			for (i=0, len=src[0].mid; ++i <= len; ) {
 				MDB_ID pn = src[i].mid << 1;
-				if (pn < parent->mt_spill_pgs[x])
-					continue;
-				if (pn > parent->mt_spill_pgs[x]) {
-					if (x <= 1)
-						break;
+				while (pn > pspill[x])
 					x--;
-					continue;
+				if (pn == pspill[x]) {
+					pspill[x] = 1;
+					y = --x;
 				}
-				parent->mt_spill_pgs[x] = 0;
-				len--;
-			}
-			/* OK, we had a few hits, squash zeros from the spill list */
-			if (len < parent->mt_spill_pgs[0]) {
-				x=1;
-				for (y=1; y<=parent->mt_spill_pgs[0]; y++) {
-					if (parent->mt_spill_pgs[y]) {
-						if (y != x) {
-							parent->mt_spill_pgs[x] = parent->mt_spill_pgs[y];
-						}
-						x++;
-					}
-				}
-				parent->mt_spill_pgs[0] = len;
 			}
+			/* Squash deleted pagenums if we deleted any */
+			for (x=y; ++x <= ps_len; )
+				if (!(pspill[x] & 1))
+					pspill[++y] = pspill[x];
+			pspill[0] = y;
 		}
+
 		/* Find len = length of merging our dirty list with parent's */
 		x = dst[0].mid;
 		dst[0].mid = 0;		/* simplify loops */
@@ -2872,7 +2865,10 @@ mdb_txn_commit(MDB_txn *txn)
 		parent->mt_dirty_room = txn->mt_dirty_room;
 		if (txn->mt_spill_pgs) {
 			if (parent->mt_spill_pgs) {
-				mdb_midl_append_list(&parent->mt_spill_pgs, txn->mt_spill_pgs);
+				/* TODO: Prevent failure here, so parent does not fail */
+				rc = mdb_midl_append_list(&parent->mt_spill_pgs, txn->mt_spill_pgs);
+				if (rc)
+					parent->mt_flags |= MDB_TXN_ERROR;
 				mdb_midl_free(txn->mt_spill_pgs);
 				mdb_midl_sort(parent->mt_spill_pgs);
 			} else {
@@ -2883,7 +2879,7 @@ mdb_txn_commit(MDB_txn *txn)
 		parent->mt_child = NULL;
 		mdb_midl_free(((MDB_ntxn *)txn)->mnt_pgstate.mf_pghead);
 		free(txn);
-		return MDB_SUCCESS;
+		return rc;
 	}
 
 	if (txn != env->me_txn) {

From 8e1d10e82887a0b8fdea6151c73377f9cabeb517 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:13:27 +0200
Subject: [PATCH 09/25] ITS#7515 Fix mdb_page_unspill() in nested txn.

Malloc a page in this txn, not in a parent.
---
 libraries/liblmdb/mdb.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 84429ee3d6..21d11b98de 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -1818,7 +1818,7 @@ mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
 			if (env->me_flags & MDB_WRITEMAP) {
 				np = mp;
 			} else {
-				np = mdb_page_malloc(txn, num);
+				np = mdb_page_malloc(tx0, num);
 				if (!np)
 					return ENOMEM;
 				if (num > 1)

From 52cb8b3417386e2cacee98401a797d8dd1a81020 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:13:27 +0200
Subject: [PATCH 10/25] mdb_page_unspill(): Rename local vars.

The names have caused bugs, "txn" was treated as
the current transaction.
---
 libraries/liblmdb/mdb.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 21d11b98de..f5db761e9c 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -1789,24 +1789,24 @@ mdb_page_copy(MDB_page *dst, MDB_page *src, unsigned int psize)
 /** Pull a page off the txn's spill list, if present.
  * If a page being referenced was spilled to disk in this txn, bring
  * it back and make it dirty/writable again.
- * @param[in] tx0 the transaction handle.
+ * @param[in] txn the transaction handle.
  * @param[in] mp the page being referenced.
  * @param[out] ret the writable page, if any. ret is unchanged if
  * mp wasn't spilled.
  */
 static int
-mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
+mdb_page_unspill(MDB_txn *txn, MDB_page *mp, MDB_page **ret)
 {
-	MDB_env *env = tx0->mt_env;
-	MDB_txn *txn;
+	MDB_env *env = txn->mt_env;
+	const MDB_txn *tx2;
 	unsigned x;
 	pgno_t pgno = mp->mp_pgno, pn = pgno << 1;
 
-	for (txn = tx0; txn; txn=txn->mt_parent) {
-		if (!txn->mt_spill_pgs)
+	for (tx2 = txn; tx2; tx2=tx2->mt_parent) {
+		if (!tx2->mt_spill_pgs)
 			continue;
-		x = mdb_midl_search(txn->mt_spill_pgs, pn);
-		if (x <= txn->mt_spill_pgs[0] && txn->mt_spill_pgs[x] == pn) {
+		x = mdb_midl_search(tx2->mt_spill_pgs, pn);
+		if (x <= tx2->mt_spill_pgs[0] && tx2->mt_spill_pgs[x] == pn) {
 			MDB_page *np;
 			int num;
 			if (txn->mt_dirty_room == 0)
@@ -1818,7 +1818,7 @@ mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
 			if (env->me_flags & MDB_WRITEMAP) {
 				np = mp;
 			} else {
-				np = mdb_page_malloc(tx0, num);
+				np = mdb_page_malloc(txn, num);
 				if (!np)
 					return ENOMEM;
 				if (num > 1)
@@ -1826,7 +1826,7 @@ mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
 				else
 					mdb_page_copy(np, mp, env->me_psize);
 			}
-			if (txn == tx0) {
+			if (tx2 == txn) {
 				/* If in current txn, this page is no longer spilled.
 				 * If it happens to be the last page, truncate the spill list.
 				 * Otherwise mark it as deleted by setting the LSB.
@@ -1839,7 +1839,7 @@ mdb_page_unspill(MDB_txn *tx0, MDB_page *mp, MDB_page **ret)
 				 * page remains spilled until child commits
 				 */
 
-			mdb_page_dirty(tx0, np);
+			mdb_page_dirty(txn, np);
 			np->mp_flags |= P_DIRTY;
 			*ret = np;
 			break;

From 7bdb5be0b572a4ff952456a2faca9ecb7a6e8786 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:20:05 +0200
Subject: [PATCH 11/25] Rename SWAP() to avoid conflict with sqlightning.

Happened since sqlightning #includes mdb.c, midl.c.
---
 libraries/liblmdb/midl.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/libraries/liblmdb/midl.c b/libraries/liblmdb/midl.c
index 86e4592d2d..0225af9584 100644
--- a/libraries/liblmdb/midl.c
+++ b/libraries/liblmdb/midl.c
@@ -203,7 +203,7 @@ int mdb_midl_append_range( MDB_IDL *idp, MDB_ID id, unsigned n )
 /* Quicksort + Insertion sort for small arrays */
 
 #define SMALL	8
-#define	SWAP(a,b)	{ itmp=(a); (a)=(b); (b)=itmp; }
+#define	MIDL_SWAP(a,b)	{ itmp=(a); (a)=(b); (b)=itmp; }
 
 void
 mdb_midl_sort( MDB_IDL ids )
@@ -231,15 +231,15 @@ mdb_midl_sort( MDB_IDL ids )
 			l = istack[jstack--];
 		} else {
 			k = (l + ir) >> 1;	/* Choose median of left, center, right */
-			SWAP(ids[k], ids[l+1]);
+			MIDL_SWAP(ids[k], ids[l+1]);
 			if (ids[l] < ids[ir]) {
-				SWAP(ids[l], ids[ir]);
+				MIDL_SWAP(ids[l], ids[ir]);
 			}
 			if (ids[l+1] < ids[ir]) {
-				SWAP(ids[l+1], ids[ir]);
+				MIDL_SWAP(ids[l+1], ids[ir]);
 			}
 			if (ids[l] < ids[l+1]) {
-				SWAP(ids[l], ids[l+1]);
+				MIDL_SWAP(ids[l], ids[l+1]);
 			}
 			i = l+1;
 			j = ir;
@@ -248,7 +248,7 @@ mdb_midl_sort( MDB_IDL ids )
 				do i++; while(ids[i] > a);
 				do j--; while(ids[j] < a);
 				if (j < i) break;
-				SWAP(ids[i],ids[j]);
+				MIDL_SWAP(ids[i],ids[j]);
 			}
 			ids[l+1] = ids[j];
 			ids[j] = a;

From 31cfca9316da2b2b901d384811d80a328eefbcdb Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:20:42 +0200
Subject: [PATCH 12/25] Tweak MDB_DEBUG output

---
 libraries/liblmdb/mdb.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index f5db761e9c..e991a17207 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -4649,7 +4649,7 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
 	}
 
 	DPRINTF(("found leaf page %"Z"u for key [%s]", mp->mp_pgno,
-	    key ? DKEY(key) : NULL));
+	    key ? DKEY(key) : "null"));
 	mc->mc_flags |= C_INITIALIZED;
 	mc->mc_flags &= ~C_EOF;
 
@@ -6245,7 +6245,7 @@ mdb_node_add(MDB_cursor *mc, indx_t indx,
 	    IS_LEAF(mp) ? "leaf" : "branch",
 		IS_SUBP(mp) ? "sub-" : "",
 	    mp->mp_pgno, indx, data ? data->mv_size : 0,
-		key ? key->mv_size : 0, key ? DKEY(key) : NULL));
+		key ? key->mv_size : 0, key ? DKEY(key) : "null"));
 
 	if (IS_LEAF2(mp)) {
 		/* Move higher keys up one slot. */
@@ -8323,6 +8323,8 @@ int mdb_reader_check(MDB_env *env, int *dead)
 					if (!mdb_reader_pid(env, Pidcheck, pid)) {
 						for (j=i; j<rdrs; j++)
 							if (mr[j].mr_pid == pid) {
+								DPRINTF(("clear stale reader pid %u txn %"Z"d",
+									(unsigned) pid, mr[j].mr_txnid));
 								mr[j].mr_pid = 0;
 								count++;
 							}

From 2980bd30f5e7dcb5918ac26d82f658ddb05eabe7 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:21:11 +0200
Subject: [PATCH 13/25] Clean up and simplify mdb_page_search().

Only named DBs can have DB_STALE, and they do not use MDB_PS_MODIFY.
Replace magic key values with flags.  Drop duplicated comments at
mdb_page_search_root() vs. mdb_page_search(), and rephrase.
---
 libraries/liblmdb/mdb.c | 79 +++++++++++++++--------------------------
 1 file changed, 29 insertions(+), 50 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index e991a17207..8d2a43c5db 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -869,8 +869,8 @@ struct MDB_txn {
  * @{
  */
 #define DB_DIRTY	0x01		/**< DB was written in this txn */
-#define DB_STALE	0x02		/**< DB record is older than txnID */
-#define DB_NEW		0x04		/**< DB handle opened in this txn */
+#define DB_STALE	0x02		/**< Named-DB record is older than txnID */
+#define DB_NEW		0x04		/**< Named-DB handle opened in this txn */
 #define DB_VALID	0x08		/**< DB handle is valid, see also #MDB_VALID */
 /** @} */
 	/** In write txns, array of cursors for each DB */
@@ -1056,6 +1056,8 @@ static int  mdb_page_search_root(MDB_cursor *mc,
 			    MDB_val *key, int modify);
 #define MDB_PS_MODIFY	1
 #define MDB_PS_ROOTONLY	2
+#define MDB_PS_FIRST	4
+#define MDB_PS_LAST		8
 static int  mdb_page_search(MDB_cursor *mc,
 			    MDB_val *key, int flags);
 static int	mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst);
@@ -1269,7 +1271,7 @@ static void mdb_audit(MDB_txn *txn)
 			txn->mt_dbs[i].md_leaf_pages +
 			txn->mt_dbs[i].md_overflow_pages;
 		if (txn->mt_dbs[i].md_flags & MDB_DUPSORT) {
-			mdb_page_search(&mc, NULL, 0);
+			mdb_page_search(&mc, NULL, MDB_PS_FIRST);
 			do {
 				unsigned j;
 				MDB_page *mp;
@@ -2474,7 +2476,7 @@ mdb_freelist_save(MDB_txn *txn)
 
 	if (env->me_pghead) {
 		/* Make sure first page of freeDB is touched and on freelist */
-		rc = mdb_page_search(&mc, NULL, MDB_PS_MODIFY);
+		rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST|MDB_PS_MODIFY);
 		if (rc && rc != MDB_NOTFOUND)
 			return rc;
 	}
@@ -2502,9 +2504,7 @@ mdb_freelist_save(MDB_txn *txn)
 		if (freecnt < txn->mt_free_pgs[0]) {
 			if (!freecnt) {
 				/* Make sure last page of freeDB is touched and on freelist */
-				key.mv_size = MDB_MAXKEYSIZE+1;
-				key.mv_data = NULL;
-				rc = mdb_page_search(&mc, &key, MDB_PS_MODIFY);
+				rc = mdb_page_search(&mc, NULL, MDB_PS_LAST|MDB_PS_MODIFY);
 				if (rc && rc != MDB_NOTFOUND)
 					return rc;
 			}
@@ -4579,18 +4579,11 @@ done:
 	return MDB_SUCCESS;
 }
 
-/** Search for the page a given key should be in.
- * Pushes parent pages on the cursor stack. This function continues a
- * search on a cursor that has already been initialized. (Usually by
- * #mdb_page_search() but also by #mdb_node_move().)
- * @param[in,out] mc the cursor for this operation.
- * @param[in] key the key to search for. If NULL, search for the lowest
- * page. (This is used by #mdb_cursor_first().)
- * @param[in] modify If true, visited pages are updated with new page numbers.
- * @return 0 on success, non-zero on failure.
+/** Finish #mdb_page_search() / #mdb_page_search_lowest().
+ *	The cursor is at the root page, set up the rest of it.
  */
 static int
-mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
+mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int flags)
 {
 	MDB_page	*mp = mc->mc_pg[mc->mc_top];
 	int rc;
@@ -4604,11 +4597,10 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
 		assert(NUMKEYS(mp) > 1);
 		DPRINTF(("found index 0 to page %"Z"u", NODEPGNO(NODEPTR(mp, 0))));
 
-		if (key == NULL)	/* Initialize cursor to first page. */
+		if (flags & (MDB_PS_FIRST|MDB_PS_LAST)) {
 			i = 0;
-		else if (key->mv_size > MDB_MAXKEYSIZE && key->mv_data == NULL) {
-							/* cursor to last page */
-			i = NUMKEYS(mp)-1;
+			if (flags & MDB_PS_LAST)
+				i = NUMKEYS(mp) - 1;
 		} else {
 			int	 exact;
 			node = mdb_node_search(mc, key, &exact);
@@ -4621,10 +4613,9 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
 					i--;
 				}
 			}
+			DPRINTF(("following index %u for key [%s]", i, DKEY(key)));
 		}
 
-		if (key)
-			DPRINTF(("following index %u for key [%s]", i, DKEY(key)));
 		assert(i < NUMKEYS(mp));
 		node = NODEPTR(mp, i);
 
@@ -4635,7 +4626,7 @@ mdb_page_search_root(MDB_cursor *mc, MDB_val *key, int modify)
 		if ((rc = mdb_cursor_push(mc, mp)))
 			return rc;
 
-		if (modify) {
+		if (flags & MDB_PS_MODIFY) {
 			if ((rc = mdb_page_touch(mc)) != 0)
 				return rc;
 			mp = mc->mc_pg[mc->mc_top];
@@ -4675,18 +4666,17 @@ mdb_page_search_lowest(MDB_cursor *mc)
 	mc->mc_ki[mc->mc_top] = 0;
 	if ((rc = mdb_cursor_push(mc, mp)))
 		return rc;
-	return mdb_page_search_root(mc, NULL, 0);
+	return mdb_page_search_root(mc, NULL, MDB_PS_FIRST);
 }
 
 /** Search for the page a given key should be in.
- * Pushes parent pages on the cursor stack. This function just sets up
- * the search; it finds the root page for \b mc's database and sets this
- * as the root of the cursor's stack. Then #mdb_page_search_root() is
- * called to complete the search.
+ * Push it and its parent pages on the cursor stack.
  * @param[in,out] mc the cursor for this operation.
- * @param[in] key the key to search for. If NULL, search for the lowest
- * page. (This is used by #mdb_cursor_first().)
- * @param[in] flags If MDB_PS_MODIFY set, visited pages are updated with new page numbers.
+ * @param[in] key the key to search for, or NULL for first/last page.
+ * @param[in] flags If MDB_PS_MODIFY is set, visited pages in the DB
+ *   are touched (updated with new page numbers).
+ *   If MDB_PS_FIRST or MDB_PS_LAST is set, find first or last leaf.
+ *   This is used by #mdb_cursor_first() and #mdb_cursor_last().
  *   If MDB_PS_ROOTONLY set, just fetch root node, no further lookups.
  * @return 0 on success, non-zero on failure.
  */
@@ -4697,23 +4687,20 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags)
 	pgno_t		 root;
 
 	/* Make sure the txn is still viable, then find the root from
-	 * the txn's db table.
+	 * the txn's db table and set it as the root of the cursor's stack.
 	 */
 	if (F_ISSET(mc->mc_txn->mt_flags, MDB_TXN_ERROR)) {
 		DPUTS("transaction has failed, must abort");
 		return MDB_BAD_TXN;
 	} else {
 		/* Make sure we're using an up-to-date root */
-		if (mc->mc_dbi > MAIN_DBI) {
-			if ((*mc->mc_dbflag & DB_STALE) ||
-			((flags & MDB_PS_MODIFY) && !(*mc->mc_dbflag & DB_DIRTY))) {
+		if (*mc->mc_dbflag & DB_STALE) {
 				MDB_cursor mc2;
-				unsigned char dbflag = 0;
 				mdb_cursor_init(&mc2, mc->mc_txn, MAIN_DBI, NULL);
-				rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, flags & MDB_PS_MODIFY);
+				rc = mdb_page_search(&mc2, &mc->mc_dbx->md_name, 0);
 				if (rc)
 					return rc;
-				if (*mc->mc_dbflag & DB_STALE) {
+				{
 					MDB_val data;
 					int exact = 0;
 					uint16_t flags;
@@ -4733,11 +4720,7 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags)
 						return MDB_INCOMPATIBLE;
 					memcpy(mc->mc_db, data.mv_data, sizeof(MDB_db));
 				}
-				if (flags & MDB_PS_MODIFY)
-					dbflag = DB_DIRTY;
 				*mc->mc_dbflag &= ~DB_STALE;
-				*mc->mc_dbflag |= dbflag;
-			}
 		}
 		root = mc->mc_db->md_root;
 
@@ -5310,7 +5293,7 @@ mdb_cursor_first(MDB_cursor *mc, MDB_val *key, MDB_val *data)
 		mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
 
 	if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
-		rc = mdb_page_search(mc, NULL, 0);
+		rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
 		if (rc != MDB_SUCCESS)
 			return rc;
 	}
@@ -5356,11 +5339,7 @@ mdb_cursor_last(MDB_cursor *mc, MDB_val *key, MDB_val *data)
 	if (!(mc->mc_flags & C_EOF)) {
 
 		if (!(mc->mc_flags & C_INITIALIZED) || mc->mc_top) {
-			MDB_val	lkey;
-
-			lkey.mv_size = MDB_MAXKEYSIZE+1;
-			lkey.mv_data = NULL;
-			rc = mdb_page_search(mc, &lkey, 0);
+			rc = mdb_page_search(mc, NULL, MDB_PS_LAST);
 			if (rc != MDB_SUCCESS)
 				return rc;
 		}
@@ -8056,7 +8035,7 @@ mdb_drop0(MDB_cursor *mc, int subs)
 {
 	int rc;
 
-	rc = mdb_page_search(mc, NULL, 0);
+	rc = mdb_page_search(mc, NULL, MDB_PS_FIRST);
 	if (rc == MDB_SUCCESS) {
 		MDB_txn *txn = mc->mc_txn;
 		MDB_node *ni;

From e3f6c152c5ab1a9810a1d3ed7de9a3252ff9e35a Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Mon, 23 Sep 2013 20:21:11 +0200
Subject: [PATCH 14/25] Drop unneeded code.

MDB_txn.mt_toggle: Use (mt_txnid & 1) instead.
Drop error checks which will be repeated.
mdb_cursor_set(): Turn assert into if/return to match the above.
mdb_cursor_del(): 'flags' are now used as bitflags.
---
 libraries/liblmdb/mdb.c | 37 +++++++++++++------------------------
 1 file changed, 13 insertions(+), 24 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 8d2a43c5db..c4db089b32 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -898,10 +898,6 @@ struct MDB_txn {
 	 *	dirty_list into mt_parent after freeing hidden mt_parent pages.
 	 */
 	unsigned int	mt_dirty_room;
-	/** Tracks which of the two meta pages was used at the start
-	 * 	of this transaction.
-	 */
-	unsigned int	mt_toggle;
 };
 
 /** Enough space for 2^32 nodes with minimum of 2 keys per node. I.e., plenty.
@@ -2118,6 +2114,7 @@ static int
 mdb_txn_renew0(MDB_txn *txn)
 {
 	MDB_env *env = txn->mt_env;
+	MDB_meta *meta;
 	unsigned int i;
 	uint16_t x;
 	int rc, new_notls = 0;
@@ -2128,8 +2125,8 @@ mdb_txn_renew0(MDB_txn *txn)
 
 	if (txn->mt_flags & MDB_TXN_RDONLY) {
 		if (!env->me_txns) {
-			i = mdb_env_pick_meta(env);
-			txn->mt_txnid = env->me_metas[i]->mm_txnid;
+			meta = env->me_metas[ mdb_env_pick_meta(env) ];
+			txn->mt_txnid = meta->mm_txnid;
 			txn->mt_u.reader = NULL;
 		} else {
 			MDB_reader *r = (env->me_flags & MDB_NOTLS) ? txn->mt_u.reader :
@@ -2174,13 +2171,13 @@ mdb_txn_renew0(MDB_txn *txn)
 			}
 			txn->mt_txnid = r->mr_txnid = env->me_txns->mti_txnid;
 			txn->mt_u.reader = r;
+			meta = env->me_metas[txn->mt_txnid & 1];
 		}
-		txn->mt_toggle = txn->mt_txnid & 1;
 	} else {
 		LOCK_MUTEX_W(env);
 
 		txn->mt_txnid = env->me_txns->mti_txnid;
-		txn->mt_toggle = txn->mt_txnid & 1;
+		meta = env->me_metas[txn->mt_txnid & 1];
 		txn->mt_txnid++;
 #if MDB_DEBUG
 		if (txn->mt_txnid == mdb_debug_start)
@@ -2196,10 +2193,10 @@ mdb_txn_renew0(MDB_txn *txn)
 	}
 
 	/* Copy the DB info and flags */
-	memcpy(txn->mt_dbs, env->me_metas[txn->mt_toggle]->mm_dbs, 2 * sizeof(MDB_db));
+	memcpy(txn->mt_dbs, meta->mm_dbs, 2 * sizeof(MDB_db));
 
 	/* Moved to here to avoid a data race in read TXNs */
-	txn->mt_next_pgno = env->me_metas[txn->mt_toggle]->mm_last_pg+1;
+	txn->mt_next_pgno = meta->mm_last_pg+1;
 
 	for (i=2; i<txn->mt_numdbs; i++) {
 		x = env->me_dbflags[i];
@@ -2295,7 +2292,6 @@ mdb_txn_begin(MDB_env *env, MDB_txn *parent, unsigned int flags, MDB_txn **ret)
 			return ENOMEM;
 		}
 		txn->mt_txnid = parent->mt_txnid;
-		txn->mt_toggle = parent->mt_toggle;
 		txn->mt_dirty_room = parent->mt_dirty_room;
 		txn->mt_u.dirty_list[0].mid = 0;
 		txn->mt_spill_pgs = NULL;
@@ -3093,7 +3089,7 @@ mdb_env_write_meta(MDB_txn *txn)
 	assert(txn != NULL);
 	assert(txn->mt_env != NULL);
 
-	toggle = !txn->mt_toggle;
+	toggle = txn->mt_txnid & 1;
 	DPRINTF(("writing meta page %d for root page %"Z"u",
 		toggle, txn->mt_dbs[MAIN_DBI].md_root));
 
@@ -4878,7 +4874,7 @@ mdb_get(MDB_txn *txn, MDB_dbi dbi,
 	if (txn->mt_flags & MDB_TXN_ERROR)
 		return MDB_BAD_TXN;
 
-	if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+	if (key->mv_size > MDB_MAXKEYSIZE) {
 		return MDB_BAD_VALSIZE;
 	}
 
@@ -5107,7 +5103,8 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 
 	assert(mc);
 	assert(key);
-	assert(key->mv_size > 0);
+	if (key->mv_size == 0)
+		return MDB_BAD_VALSIZE;
 
 	if (mc->mc_xcursor)
 		mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
@@ -5431,7 +5428,7 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 	case MDB_SET_RANGE:
 		if (key == NULL) {
 			rc = EINVAL;
-		} else if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+		} else if (key->mv_size > MDB_MAXKEYSIZE) {
 			rc = MDB_BAD_VALSIZE;
 		} else if (op == MDB_SET_RANGE)
 			rc = mdb_cursor_set(mc, key, data, op, NULL);
@@ -6051,7 +6048,6 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
 
 	if (!(flags & MDB_NOSPILL) && (rc = mdb_page_spill(mc, NULL, NULL)))
 		return rc;
-	flags &= ~MDB_NOSPILL; /* TODO: Or change (flags != MDB_NODUPDATA) to ~(flags & MDB_NODUPDATA), not looking at the logic of that code just now */
 
 	rc = mdb_cursor_touch(mc);
 	if (rc)
@@ -7301,7 +7297,7 @@ mdb_del(MDB_txn *txn, MDB_dbi dbi,
 	if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
 		return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
 
-	if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
+	if (key->mv_size > MDB_MAXKEYSIZE) {
 		return MDB_BAD_VALSIZE;
 	}
 
@@ -7765,13 +7761,6 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi,
 	if (txn == NULL || !dbi || dbi >= txn->mt_numdbs || !(txn->mt_dbflags[dbi] & DB_VALID))
 		return EINVAL;
 
-	if (txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_ERROR))
-		return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
-
-	if (key->mv_size == 0 || key->mv_size > MDB_MAXKEYSIZE) {
-		return MDB_BAD_VALSIZE;
-	}
-
 	if ((flags & (MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP)) != flags)
 		return EINVAL;
 

From f73994054779be3e177bfd27f44fcd3147e7d08d Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@symas.com>
Date: Tue, 1 Oct 2013 13:16:38 -0700
Subject: [PATCH 15/25] Add MDB_NORDLOCK to omit all reader table usage

Calling app wants to manage its own locking.
---
 libraries/liblmdb/lmdb.h | 2 ++
 libraries/liblmdb/mdb.c  | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/libraries/liblmdb/lmdb.h b/libraries/liblmdb/lmdb.h
index f637229283..848ba635dd 100644
--- a/libraries/liblmdb/lmdb.h
+++ b/libraries/liblmdb/lmdb.h
@@ -269,6 +269,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
 #define MDB_MAPASYNC		0x100000
 	/** tie reader locktable slots to #MDB_txn objects instead of to threads */
 #define MDB_NOTLS		0x200000
+	/** don't use reader locktable at all, caller must manage read/write concurrency */
+#define MDB_NORDLOCK		0x400000
 /** @} */
 
 /**	@defgroup	mdb_dbi_open	Database Flags
diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index c4db089b32..e82335d716 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -3983,7 +3983,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 		goto leave;
 	}
 
-	if (F_ISSET(flags, MDB_RDONLY)) {
+	if ((flags & (MDB_RDONLY|MDB_NORDLOCK)) == MDB_RDONLY) {
 		rc = mdb_env_setup_locks(env, lpath, mode, &excl);
 		if (rc)
 			goto leave;

From a1685c3ef70bc70fcb7b8fde4a2a2e9f6d8e11f0 Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@symas.com>
Date: Tue, 1 Oct 2013 23:36:57 -0700
Subject: [PATCH 16/25] More doc for MDB_NORDLOCK

---
 libraries/liblmdb/lmdb.h | 9 +++++++--
 libraries/liblmdb/mdb.c  | 2 +-
 2 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/libraries/liblmdb/lmdb.h b/libraries/liblmdb/lmdb.h
index 848ba635dd..968862f987 100644
--- a/libraries/liblmdb/lmdb.h
+++ b/libraries/liblmdb/lmdb.h
@@ -265,11 +265,11 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
 #define MDB_NOMETASYNC		0x40000
 	/** use writable mmap */
 #define MDB_WRITEMAP		0x80000
-	/** use asynchronous msync when MDB_WRITEMAP is used */
+	/** use asynchronous msync when #MDB_WRITEMAP is used */
 #define MDB_MAPASYNC		0x100000
 	/** tie reader locktable slots to #MDB_txn objects instead of to threads */
 #define MDB_NOTLS		0x200000
-	/** don't use reader locktable at all, caller must manage read/write concurrency */
+	/** for #MDB_RDONLY env, don't use reader locktable, caller must manage read/write concurrency */
 #define MDB_NORDLOCK		0x400000
 /** @} */
 
@@ -527,6 +527,11 @@ int  mdb_env_create(MDB_env **env);
 	 *		user threads over individual OS threads need this option. Such an
 	 *		application must also serialize the write transactions in an OS
 	 *		thread, since MDB's write locking is unaware of the user threads.
+	 *	<li>#MDB_NORDLOCK
+	 *		Don't use the reader locktable at all. This flag is only valid
+	 *		with #MDB_RDONLY. MDB will use no read locks. If other processes
+	 *		may be opening the environment with write access, the callers
+	 *		must manage read/write locks themselves.
 	 * </ul>
 	 * @param[in] mode The UNIX permissions to set on created files. This parameter
 	 * is ignored on Windows.
diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index e82335d716..637cd08809 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -3900,7 +3900,7 @@ fail:
 	 *	environment and re-opening it with the new flags.
 	 */
 #define	CHANGEABLE	(MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC)
-#define	CHANGELESS	(MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS)
+#define	CHANGELESS	(MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS|MDB_NORDLOCK)
 
 int
 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode)

From 79eac42c0f1fdd6c807f0b601c03ab35bf7d7d8e Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@symas.com>
Date: Thu, 3 Oct 2013 10:26:44 -0700
Subject: [PATCH 17/25] s/MDB_NORDLOCK/MDB_NOLOCK/

Leave all lock management to the caller.
---
 libraries/liblmdb/lmdb.h | 16 ++++++++-------
 libraries/liblmdb/mdb.c  | 43 +++++++++++++++++++++++++---------------
 2 files changed, 36 insertions(+), 23 deletions(-)

diff --git a/libraries/liblmdb/lmdb.h b/libraries/liblmdb/lmdb.h
index 968862f987..8cdac26b06 100644
--- a/libraries/liblmdb/lmdb.h
+++ b/libraries/liblmdb/lmdb.h
@@ -269,8 +269,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
 #define MDB_MAPASYNC		0x100000
 	/** tie reader locktable slots to #MDB_txn objects instead of to threads */
 #define MDB_NOTLS		0x200000
-	/** for #MDB_RDONLY env, don't use reader locktable, caller must manage read/write concurrency */
-#define MDB_NORDLOCK		0x400000
+	/** don't do any locking, caller must manage their own locks */
+#define MDB_NOLOCK		0x400000
 /** @} */
 
 /**	@defgroup	mdb_dbi_open	Database Flags
@@ -527,11 +527,13 @@ int  mdb_env_create(MDB_env **env);
 	 *		user threads over individual OS threads need this option. Such an
 	 *		application must also serialize the write transactions in an OS
 	 *		thread, since MDB's write locking is unaware of the user threads.
-	 *	<li>#MDB_NORDLOCK
-	 *		Don't use the reader locktable at all. This flag is only valid
-	 *		with #MDB_RDONLY. MDB will use no read locks. If other processes
-	 *		may be opening the environment with write access, the callers
-	 *		must manage read/write locks themselves.
+	 *	<li>#MDB_NOLOCK
+	 *		Don't do any locking. If concurrent access is anticipated, the
+	 *		caller must manage all concurrency itself. For proper operation
+	 *		the caller must enforce single-writer semantics, and must ensure
+	 *		that no readers are using old transactions while a writer is
+	 *		active. The simplest approach is to use an exclusive lock so that
+	 *		no readers may be active at all when a writer begins.
 	 * </ul>
 	 * @param[in] mode The UNIX permissions to set on created files. This parameter
 	 * is ignored on Windows.
diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 637cd08809..30b562c43e 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -425,7 +425,8 @@ typedef uint16_t	 indx_t;
  *
  *	If #MDB_NOTLS is set, the slot address is not saved in thread-specific data.
  *
- *	No reader table is used if the database is on a read-only filesystem.
+ *	No reader table is used if the database is on a read-only filesystem, or
+ *	if #MDB_NOLOCK is set.
  *
  *	Since the database uses multi-version concurrency control, readers don't
  *	actually need any locking. This table is used to keep track of which
@@ -1572,12 +1573,14 @@ mdb_find_oldest(MDB_txn *txn)
 {
 	int i;
 	txnid_t mr, oldest = txn->mt_txnid - 1;
-	MDB_reader *r = txn->mt_env->me_txns->mti_readers;
-	for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
-		if (r[i].mr_pid) {
-			mr = r[i].mr_txnid;
-			if (oldest > mr)
-				oldest = mr;
+	if (txn->mt_env->me_txns) {
+		MDB_reader *r = txn->mt_env->me_txns->mti_readers;
+		for (i = txn->mt_env->me_txns->mti_numreaders; --i >= 0; ) {
+			if (r[i].mr_pid) {
+				mr = r[i].mr_txnid;
+				if (oldest > mr)
+					oldest = mr;
+			}
 		}
 	}
 	return oldest;
@@ -2174,10 +2177,15 @@ mdb_txn_renew0(MDB_txn *txn)
 			meta = env->me_metas[txn->mt_txnid & 1];
 		}
 	} else {
-		LOCK_MUTEX_W(env);
+		if (env->me_txns) {
+			LOCK_MUTEX_W(env);
 
-		txn->mt_txnid = env->me_txns->mti_txnid;
-		meta = env->me_metas[txn->mt_txnid & 1];
+			txn->mt_txnid = env->me_txns->mti_txnid;
+			meta = env->me_metas[txn->mt_txnid & 1];
+		} else {
+			meta = env->me_metas[ mdb_env_pick_meta(env) ];
+			txn->mt_txnid = meta->mm_txnid;
+		}
 		txn->mt_txnid++;
 #if MDB_DEBUG
 		if (txn->mt_txnid == mdb_debug_start)
@@ -2417,7 +2425,8 @@ mdb_txn_reset0(MDB_txn *txn, const char *act)
 
 		env->me_txn = NULL;
 		/* The writer mutex was locked in mdb_txn_begin. */
-		UNLOCK_MUTEX_W(env);
+		if (env->me_txns)
+			UNLOCK_MUTEX_W(env);
 	}
 }
 
@@ -2934,7 +2943,8 @@ done:
 	env->me_txn = NULL;
 	mdb_dbis_update(txn, 1);
 
-	UNLOCK_MUTEX_W(env);
+	if (env->me_txns)
+		UNLOCK_MUTEX_W(env);
 	free(txn);
 
 	return MDB_SUCCESS;
@@ -3180,7 +3190,8 @@ done:
 	 * readers will get consistent data regardless of how fresh or
 	 * how stale their view of these values is.
 	 */
-	env->me_txns->mti_txnid = txn->mt_txnid;
+	if (env->me_txns)
+		env->me_txns->mti_txnid = txn->mt_txnid;
 
 	return MDB_SUCCESS;
 }
@@ -3900,7 +3911,7 @@ fail:
 	 *	environment and re-opening it with the new flags.
 	 */
 #define	CHANGEABLE	(MDB_NOSYNC|MDB_NOMETASYNC|MDB_MAPASYNC)
-#define	CHANGELESS	(MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS|MDB_NORDLOCK)
+#define	CHANGELESS	(MDB_FIXEDMAP|MDB_NOSUBDIR|MDB_RDONLY|MDB_WRITEMAP|MDB_NOTLS|MDB_NOLOCK)
 
 int
 mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode)
@@ -3953,7 +3964,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 	}
 
 	/* For RDONLY, get lockfile after we know datafile exists */
-	if (!F_ISSET(flags, MDB_RDONLY)) {
+	if (!(flags & (MDB_RDONLY|MDB_NOLOCK))) {
 		rc = mdb_env_setup_locks(env, lpath, mode, &excl);
 		if (rc)
 			goto leave;
@@ -3983,7 +3994,7 @@ mdb_env_open(MDB_env *env, const char *path, unsigned int flags, mdb_mode_t mode
 		goto leave;
 	}
 
-	if ((flags & (MDB_RDONLY|MDB_NORDLOCK)) == MDB_RDONLY) {
+	if ((flags & (MDB_RDONLY|MDB_NOLOCK)) == MDB_RDONLY) {
 		rc = mdb_env_setup_locks(env, lpath, mode, &excl);
 		if (rc)
 			goto leave;

From 07dc79a7c33121bc1be55bbd81ff643094df6851 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Thu, 3 Oct 2013 23:59:24 +0200
Subject: [PATCH 18/25] Set subDB DBI=parent DBI, fix MDB_DUPSORT delete.

xcursor DBIs were parent DBI+1 for debugging. Instead output
-(parent DBI).  Fixes a crash in mdb_cursor_del0()'s xcursor
tracking, it forgot to subtract 1 for C_SUB cursors.
---
 libraries/liblmdb/mdb.c | 52 ++++++++++++++---------------------------
 1 file changed, 18 insertions(+), 34 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 30b562c43e..cca5fc624c 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -317,6 +317,9 @@ static txnid_t mdb_debug_start;
 	 *	The string is printed literally, with no format processing.
 	 */
 #define DPUTS(arg)	DPRINTF(("%s", arg))
+	/** Debuging output value of a cursor DBI: Negative in a sub-cursor. */
+#define DDBI(mc) \
+	(((mc)->mc_flags & C_SUB) ? -(int)(mc)->mc_dbi : (int)(mc)->mc_dbi)
 /** @} */
 
 	/** A default memory page size.
@@ -1859,7 +1862,6 @@ mdb_page_touch(MDB_cursor *mc)
 	MDB_page *mp = mc->mc_pg[mc->mc_top], *np;
 	MDB_txn *txn = mc->mc_txn;
 	MDB_cursor *m2, *m3;
-	MDB_dbi dbi;
 	pgno_t	pgno;
 	int rc;
 
@@ -1876,7 +1878,8 @@ mdb_page_touch(MDB_cursor *mc)
 			(rc = mdb_page_alloc(mc, 1, &np)))
 			return rc;
 		pgno = np->mp_pgno;
-		DPRINTF(("touched db %u page %"Z"u -> %"Z"u", mc->mc_dbi,mp->mp_pgno,pgno));
+		DPRINTF(("touched db %d page %"Z"u -> %"Z"u", DDBI(mc),
+			mp->mp_pgno, pgno));
 		assert(mp->mp_pgno != pgno);
 		mdb_midl_xappend(txn->mt_free_pgs, mp->mp_pgno);
 		/* Update the parent page, if any, to point to the new page */
@@ -1922,17 +1925,16 @@ mdb_page_touch(MDB_cursor *mc)
 done:
 	/* Adjust cursors pointing to mp */
 	mc->mc_pg[mc->mc_top] = np;
-	dbi = mc->mc_dbi;
+	m2 = txn->mt_cursors[mc->mc_dbi];
 	if (mc->mc_flags & C_SUB) {
-		dbi--;
-		for (m2 = txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+		for (; m2; m2=m2->mc_next) {
 			m3 = &m2->mc_xcursor->mx_cursor;
 			if (m3->mc_snum < mc->mc_snum) continue;
 			if (m3->mc_pg[mc->mc_top] == mp)
 				m3->mc_pg[mc->mc_top] = np;
 		}
 	} else {
-		for (m2 = txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
+		for (; m2; m2=m2->mc_next) {
 			if (m2->mc_snum < mc->mc_snum) continue;
 			if (m2->mc_pg[mc->mc_top] == mp) {
 				m2->mc_pg[mc->mc_top] = np;
@@ -4502,8 +4504,8 @@ mdb_cursor_pop(MDB_cursor *mc)
 		if (mc->mc_snum)
 			mc->mc_top--;
 
-		DPRINTF(("popped page %"Z"u off db %u cursor %p", top->mp_pgno,
-			mc->mc_dbi, (void *) mc));
+		DPRINTF(("popped page %"Z"u off db %d cursor %p", top->mp_pgno,
+			DDBI(mc), (void *) mc));
 	}
 }
 
@@ -4511,8 +4513,8 @@ mdb_cursor_pop(MDB_cursor *mc)
 static int
 mdb_cursor_push(MDB_cursor *mc, MDB_page *mp)
 {
-	DPRINTF(("pushing page %"Z"u on db %u cursor %p", mp->mp_pgno,
-		mc->mc_dbi, (void *) mc));
+	DPRINTF(("pushing page %"Z"u on db %d cursor %p", mp->mp_pgno,
+		DDBI(mc), (void *) mc));
 
 	if (mc->mc_snum >= CURSOR_STACK) {
 		assert(mc->mc_snum < CURSOR_STACK);
@@ -4745,8 +4747,8 @@ mdb_page_search(MDB_cursor *mc, MDB_val *key, int flags)
 	mc->mc_snum = 1;
 	mc->mc_top = 0;
 
-	DPRINTF(("db %u root page %"Z"u has flags 0x%X",
-		mc->mc_dbi, root, mc->mc_pg[0]->mp_flags));
+	DPRINTF(("db %d root page %"Z"u has flags 0x%X",
+		DDBI(mc), root, mc->mc_pg[0]->mp_flags));
 
 	if (flags & MDB_PS_MODIFY) {
 		if ((rc = mdb_page_touch(mc)))
@@ -5620,8 +5622,8 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 		return MDB_BAD_VALSIZE;
 #endif
 
-	DPRINTF(("==> put db %u key [%s], size %"Z"u, data size %"Z"u",
-		mc->mc_dbi, DKEY(key), key ? key->mv_size:0, data->mv_size));
+	DPRINTF(("==> put db %d key [%s], size %"Z"u, data size %"Z"u",
+		DDBI(mc), DKEY(key), key ? key->mv_size : 0, data->mv_size));
 
 	dkey.mv_size = 0;
 
@@ -5950,9 +5952,6 @@ new_sub:
 			unsigned i = mc->mc_top;
 			MDB_page *mp = mc->mc_pg[i];
 
-			if (mc->mc_flags & C_SUB)
-				dbi--;
-
 			for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
 				if (mc->mc_flags & C_SUB)
 					m3 = &m2->mc_xcursor->mx_cursor;
@@ -6462,7 +6461,7 @@ mdb_xcursor_init0(MDB_cursor *mc)
 	mx->mx_cursor.mc_txn = mc->mc_txn;
 	mx->mx_cursor.mc_db = &mx->mx_db;
 	mx->mx_cursor.mc_dbx = &mx->mx_dbx;
-	mx->mx_cursor.mc_dbi = mc->mc_dbi+1;
+	mx->mx_cursor.mc_dbi = mc->mc_dbi;
 	mx->mx_cursor.mc_dbflag = &mx->mx_dbflag;
 	mx->mx_cursor.mc_snum = 0;
 	mx->mx_cursor.mc_top = 0;
@@ -6510,7 +6509,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node)
 				mx->mx_db.md_flags |= MDB_INTEGERKEY;
 		}
 	}
-	DPRINTF(("Sub-db %u for db %u root page %"Z"u", mx->mx_cursor.mc_dbi, mc->mc_dbi,
+	DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
 		mx->mx_db.md_root));
 	mx->mx_dbflag = DB_VALID | (F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY) ?
 		DB_DIRTY : 0);
@@ -6831,9 +6830,6 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst)
 		MDB_dbi dbi = csrc->mc_dbi;
 		MDB_page *mp = csrc->mc_pg[csrc->mc_top];
 
-		if (csrc->mc_flags & C_SUB)
-			dbi--;
-
 		for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
 			if (csrc->mc_flags & C_SUB)
 				m3 = &m2->mc_xcursor->mx_cursor;
@@ -7008,9 +7004,6 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst)
 		MDB_dbi dbi = csrc->mc_dbi;
 		MDB_page *mp = cdst->mc_pg[cdst->mc_top];
 
-		if (csrc->mc_flags & C_SUB)
-			dbi--;
-
 		for (m2 = csrc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
 			if (csrc->mc_flags & C_SUB)
 				m3 = &m2->mc_xcursor->mx_cursor;
@@ -7109,9 +7102,6 @@ mdb_rebalance(MDB_cursor *mc)
 				MDB_cursor *m2, *m3;
 				MDB_dbi dbi = mc->mc_dbi;
 
-				if (mc->mc_flags & C_SUB)
-					dbi--;
-
 				for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
 					if (mc->mc_flags & C_SUB)
 						m3 = &m2->mc_xcursor->mx_cursor;
@@ -7141,9 +7131,6 @@ mdb_rebalance(MDB_cursor *mc)
 				MDB_cursor *m2, *m3;
 				MDB_dbi dbi = mc->mc_dbi;
 
-				if (mc->mc_flags & C_SUB)
-					dbi--;
-
 				for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
 					if (mc->mc_flags & C_SUB)
 						m3 = &m2->mc_xcursor->mx_cursor;
@@ -7712,9 +7699,6 @@ done:
 		MDB_dbi dbi = mc->mc_dbi;
 		int fixup = NUMKEYS(mp);
 
-		if (mc->mc_flags & C_SUB)
-			dbi--;
-
 		for (m2 = mc->mc_txn->mt_cursors[dbi]; m2; m2=m2->mc_next) {
 			if (mc->mc_flags & C_SUB)
 				m3 = &m2->mc_xcursor->mx_cursor;

From 3d67838a59b30b5732c96612cf8ce5bbf59d7441 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Thu, 3 Oct 2013 23:59:24 +0200
Subject: [PATCH 19/25] Set MDB_xcursor DB_DIRTY, clear md_name.

Both were unused and md_name was unmaintained -- except
mdb_cursor_touch(xcursor) would abuse md_name as a key to
touch MAIN_DBI if it could somehow get passed ! DB_DIRTY.
---
 libraries/liblmdb/mdb.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index cca5fc624c..cc7471ac4e 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -872,7 +872,7 @@ struct MDB_txn {
  *	@ingroup internal
  * @{
  */
-#define DB_DIRTY	0x01		/**< DB was written in this txn */
+#define DB_DIRTY	0x01		/**< DB was modified or is DUPSORT data */
 #define DB_STALE	0x02		/**< Named-DB record is older than txnID */
 #define DB_NEW		0x04		/**< Named-DB handle opened in this txn */
 #define DB_VALID	0x08		/**< DB handle is valid, see also #MDB_VALID */
@@ -6466,6 +6466,8 @@ mdb_xcursor_init0(MDB_cursor *mc)
 	mx->mx_cursor.mc_snum = 0;
 	mx->mx_cursor.mc_top = 0;
 	mx->mx_cursor.mc_flags = C_SUB;
+	mx->mx_dbx.md_name.mv_size = 0;
+	mx->mx_dbx.md_name.mv_data = NULL;
 	mx->mx_dbx.md_cmp = mc->mc_dbx->md_dcmp;
 	mx->mx_dbx.md_dcmp = NULL;
 	mx->mx_dbx.md_rel = mc->mc_dbx->md_rel;
@@ -6511,10 +6513,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node)
 	}
 	DPRINTF(("Sub-db -%u root page %"Z"u", mx->mx_cursor.mc_dbi,
 		mx->mx_db.md_root));
-	mx->mx_dbflag = DB_VALID | (F_ISSET(mc->mc_pg[mc->mc_top]->mp_flags, P_DIRTY) ?
-		DB_DIRTY : 0);
-	mx->mx_dbx.md_name.mv_data = NODEKEY(node);
-	mx->mx_dbx.md_name.mv_size = node->mn_ksize;
+	mx->mx_dbflag = DB_VALID|DB_DIRTY; /* DB_DIRTY guides mdb_cursor_touch */
 #if UINT_MAX < SIZE_MAX
 	if (mx->mx_dbx.md_cmp == mdb_cmp_int && mx->mx_db.md_pad == sizeof(size_t))
 #ifdef MISALIGNED_OK

From 3a1d73dafff3a4da3a1ec3841831edadcb7e76be Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Thu, 3 Oct 2013 23:59:24 +0200
Subject: [PATCH 20/25] Optimize code holding reader mutex

---
 libraries/liblmdb/mdb.c | 29 ++++++++++++++++-------------
 1 file changed, 16 insertions(+), 13 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index cc7471ac4e..0ee9f50b36 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -2119,8 +2119,9 @@ static int
 mdb_txn_renew0(MDB_txn *txn)
 {
 	MDB_env *env = txn->mt_env;
+	MDB_txninfo *ti = env->me_txns;
 	MDB_meta *meta;
-	unsigned int i;
+	unsigned int i, nr;
 	uint16_t x;
 	int rc, new_notls = 0;
 
@@ -2129,7 +2130,7 @@ mdb_txn_renew0(MDB_txn *txn)
 	txn->mt_dbxs = env->me_dbxs;	/* mostly static anyway */
 
 	if (txn->mt_flags & MDB_TXN_RDONLY) {
-		if (!env->me_txns) {
+		if (!ti) {
 			meta = env->me_metas[ mdb_env_pick_meta(env) ];
 			txn->mt_txnid = meta->mm_txnid;
 			txn->mt_u.reader = NULL;
@@ -2153,36 +2154,38 @@ mdb_txn_renew0(MDB_txn *txn)
 				}
 
 				LOCK_MUTEX_R(env);
-				for (i=0; i<env->me_txns->mti_numreaders; i++)
-					if (env->me_txns->mti_readers[i].mr_pid == 0)
+				nr = ti->mti_numreaders;
+				for (i=0; i<nr; i++)
+					if (ti->mti_readers[i].mr_pid == 0)
 						break;
 				if (i == env->me_maxreaders) {
 					UNLOCK_MUTEX_R(env);
 					return MDB_READERS_FULL;
 				}
-				env->me_txns->mti_readers[i].mr_pid = pid;
-				env->me_txns->mti_readers[i].mr_tid = tid;
-				if (i >= env->me_txns->mti_numreaders)
-					env->me_txns->mti_numreaders = i+1;
+				ti->mti_readers[i].mr_pid = pid;
+				ti->mti_readers[i].mr_tid = tid;
+				if (i == nr)
+					ti->mti_numreaders = ++nr;
 				/* Save numreaders for un-mutexed mdb_env_close() */
-				env->me_numreaders = env->me_txns->mti_numreaders;
+				env->me_numreaders = nr;
 				UNLOCK_MUTEX_R(env);
-				r = &env->me_txns->mti_readers[i];
+
+				r = &ti->mti_readers[i];
 				new_notls = (env->me_flags & MDB_NOTLS);
 				if (!new_notls && (rc=pthread_setspecific(env->me_txkey, r))) {
 					r->mr_pid = 0;
 					return rc;
 				}
 			}
-			txn->mt_txnid = r->mr_txnid = env->me_txns->mti_txnid;
+			txn->mt_txnid = r->mr_txnid = ti->mti_txnid;
 			txn->mt_u.reader = r;
 			meta = env->me_metas[txn->mt_txnid & 1];
 		}
 	} else {
-		if (env->me_txns) {
+		if (ti) {
 			LOCK_MUTEX_W(env);
 
-			txn->mt_txnid = env->me_txns->mti_txnid;
+			txn->mt_txnid = ti->mti_txnid;
 			meta = env->me_metas[txn->mt_txnid & 1];
 		} else {
 			meta = env->me_metas[ mdb_env_pick_meta(env) ];

From 8f075595a14076fa78d1943f7bbb893031e9a6b2 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Thu, 3 Oct 2013 23:59:25 +0200
Subject: [PATCH 21/25] mdb_node_add(): Plug page leak when MDB_PAGE_FULL.

Do not fail after mdb_page_new() succeeds.
---
 libraries/liblmdb/mdb.c | 27 ++++++++++++++++-----------
 1 file changed, 16 insertions(+), 11 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 0ee9f50b36..f3e3180e00 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -6221,6 +6221,7 @@ mdb_node_add(MDB_cursor *mc, indx_t indx,
 {
 	unsigned int	 i;
 	size_t		 node_size = NODESIZE;
+	ssize_t		 room;
 	indx_t		 ofs;
 	MDB_node	*node;
 	MDB_page	*mp = mc->mc_pg[mc->mc_top];
@@ -6251,9 +6252,9 @@ mdb_node_add(MDB_cursor *mc, indx_t indx,
 		return MDB_SUCCESS;
 	}
 
+	room = (ssize_t)SIZELEFT(mp) - (ssize_t)sizeof(indx_t);
 	if (key != NULL)
 		node_size += key->mv_size;
-
 	if (IS_LEAF(mp)) {
 		assert(data);
 		if (F_ISSET(flags, F_BIGDATA)) {
@@ -6265,26 +6266,23 @@ mdb_node_add(MDB_cursor *mc, indx_t indx,
 			/* Put data on overflow page. */
 			DPRINTF(("data size is %"Z"u, node would be %"Z"u, put data on overflow page",
 			    data->mv_size, node_size+data->mv_size));
-			node_size += sizeof(pgno_t);
+			node_size += sizeof(pgno_t) + (node_size & 1);
+			if ((ssize_t)node_size > room)
+				goto full;
 			if ((rc = mdb_page_new(mc, P_OVERFLOW, ovpages, &ofp)))
 				return rc;
 			DPRINTF(("allocated overflow page %"Z"u", ofp->mp_pgno));
 			flags |= F_BIGDATA;
+			goto update;
 		} else {
 			node_size += data->mv_size;
 		}
 	}
 	node_size += node_size & 1;
+	if ((ssize_t)node_size > room)
+		goto full;
 
-	if (node_size + sizeof(indx_t) > SIZELEFT(mp)) {
-		DPRINTF(("not enough room in page %"Z"u, got %u ptrs",
-		    mp->mp_pgno, NUMKEYS(mp)));
-		DPRINTF(("upper - lower = %u - %u = %u", mp->mp_upper, mp->mp_lower,
-		    mp->mp_upper - mp->mp_lower));
-		DPRINTF(("node size = %"Z"u", node_size));
-		return MDB_PAGE_FULL;
-	}
-
+update:
 	/* Move higher pointers up one slot. */
 	for (i = NUMKEYS(mp); i > indx; i--)
 		mp->mp_ptrs[i] = mp->mp_ptrs[i - 1];
@@ -6330,6 +6328,13 @@ mdb_node_add(MDB_cursor *mc, indx_t indx,
 	}
 
 	return MDB_SUCCESS;
+
+full:
+	DPRINTF(("not enough room in page %"Z"u, got %u ptrs",
+		mp->mp_pgno, NUMKEYS(mp)));
+	DPRINTF(("upper-lower = %u - %u = %"Z"d", mp->mp_upper,mp->mp_lower,room));
+	DPRINTF(("node size = %"Z"u", node_size));
+	return MDB_PAGE_FULL;
 }
 
 /** Delete the specified node from a page.

From 99ea7669a35a79878394b29cd3448b58daf031fb Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Thu, 3 Oct 2013 23:59:25 +0200
Subject: [PATCH 22/25] mdb_cursor_sibling(): Fix error result

---
 libraries/liblmdb/mdb.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index f3e3180e00..e7564be8a0 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -4942,8 +4942,11 @@ mdb_cursor_sibling(MDB_cursor *mc, int move_right)
 	assert(IS_BRANCH(mc->mc_pg[mc->mc_top]));
 
 	indx = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
-	if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL) != 0))
+	if ((rc = mdb_page_get(mc->mc_txn, NODEPGNO(indx), &mp, NULL)) != 0) {
+		/* mc will be inconsistent if caller does mc_snum++ as above */
+		mc->mc_flags &= ~(C_INITIALIZED|C_EOF);
 		return rc;
+	}
 
 	mdb_cursor_push(mc, mp);
 	if (!move_right)

From 0f9b79e12cb9b1b716a0268f6dc70adc41067a09 Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Thu, 3 Oct 2013 23:59:25 +0200
Subject: [PATCH 23/25] Maintain MDB_cursor.mc_top

---
 libraries/liblmdb/mdb.c | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index e7564be8a0..2f671ed0a1 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -5553,14 +5553,14 @@ fetchm:
 	return rc;
 }
 
-/** Touch all the pages in the cursor stack.
+/** Touch all the pages in the cursor stack. Set mc_top.
  *	Makes sure all the pages are writable, before attempting a write operation.
  * @param[in] mc The cursor to operate on.
  */
 static int
 mdb_cursor_touch(MDB_cursor *mc)
 {
-	int rc;
+	int rc = MDB_SUCCESS;
 
 	if (mc->mc_dbi > MAIN_DBI && !(*mc->mc_dbflag & DB_DIRTY)) {
 		MDB_cursor mc2;
@@ -5571,13 +5571,14 @@ mdb_cursor_touch(MDB_cursor *mc)
 			 return rc;
 		*mc->mc_dbflag |= DB_DIRTY;
 	}
-	for (mc->mc_top = 0; mc->mc_top < mc->mc_snum; mc->mc_top++) {
-		rc = mdb_page_touch(mc);
-		if (rc)
-			return rc;
+	mc->mc_top = 0;
+	if (mc->mc_snum) {
+		do {
+			rc = mdb_page_touch(mc);
+		} while (!rc && ++(mc->mc_top) < mc->mc_snum);
+		mc->mc_top = mc->mc_snum-1;
 	}
-	mc->mc_top = mc->mc_snum-1;
-	return MDB_SUCCESS;
+	return rc;
 }
 
 /** Do not spill pages to disk if txn is getting full, may fail instead */
@@ -5640,6 +5641,7 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 	} else if (mc->mc_db->md_root == P_INVALID) {
 		/* new database, cursor has nothing to point to */
 		mc->mc_snum = 0;
+		mc->mc_top = 0;
 		mc->mc_flags &= ~C_INITIALIZED;
 		rc = MDB_NO_ROOT;
 	} else {
@@ -6499,6 +6501,7 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node)
 		memcpy(&mx->mx_db, NODEDATA(node), sizeof(MDB_db));
 		mx->mx_cursor.mc_pg[0] = 0;
 		mx->mx_cursor.mc_snum = 0;
+		mx->mx_cursor.mc_top = 0;
 		mx->mx_cursor.mc_flags = C_SUB;
 	} else {
 		MDB_page *fp = NODEDATA(node);
@@ -6511,8 +6514,8 @@ mdb_xcursor_init1(MDB_cursor *mc, MDB_node *node)
 		mx->mx_db.md_entries = NUMKEYS(fp);
 		COPY_PGNO(mx->mx_db.md_root, fp->mp_pgno);
 		mx->mx_cursor.mc_snum = 1;
-		mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
 		mx->mx_cursor.mc_top = 0;
+		mx->mx_cursor.mc_flags = C_INITIALIZED|C_SUB;
 		mx->mx_cursor.mc_pg[0] = fp;
 		mx->mx_cursor.mc_ki[0] = 0;
 		if (mc->mc_db->md_flags & MDB_DUPFIXED) {

From 9f7ae8925fb896d7de9345ea441c5da3fce6670c Mon Sep 17 00:00:00 2001
From: Hallvard Furuseth <hallvard@openldap.org>
Date: Fri, 4 Oct 2013 00:48:19 +0200
Subject: [PATCH 24/25] Add Caveat: Readers need write access. Whitespace.

---
 libraries/liblmdb/lmdb.h     | 6 +++++-
 libraries/liblmdb/mdb.c      | 2 +-
 libraries/liblmdb/mdb_stat.c | 2 +-
 libraries/liblmdb/midl.c     | 4 ++--
 4 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/libraries/liblmdb/lmdb.h b/libraries/liblmdb/lmdb.h
index 8cdac26b06..f6dfa60c38 100644
--- a/libraries/liblmdb/lmdb.h
+++ b/libraries/liblmdb/lmdb.h
@@ -66,6 +66,10 @@
  *	  BSD systems or when otherwise configured with MDB_USE_POSIX_SEM.
  *	  Multiple users can cause startup to fail later, as noted above.
  *
+ *	- There is normally no pure read-only mode, since readers need write
+ *	  access to locks and lock file. Exceptions: On read-only filesystems
+ *	  or with the #MDB_NOLOCK flag described under #mdb_env_open().
+ *
  *	- A thread can only use one transaction at a time, plus any child
  *	  transactions.  Each transaction belongs to one thread.  See below.
  *	  The #MDB_NOTLS flag changes this for read-only transactions.
@@ -489,7 +493,7 @@ int  mdb_env_create(MDB_env **env);
 	 *		like wild pointer writes and other bad updates into the database.
 	 *		Incompatible with nested transactions.
 	 *		Processes with and without MDB_WRITEMAP on the same environment do
-	 *		not cooperate well. 
+	 *		not cooperate well.
 	 *	<li>#MDB_NOMETASYNC
 	 *		Flush system buffers to disk only once per transaction, omit the
 	 *		metadata flush. Defer that until the system flushes files to disk,
diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 2f671ed0a1..d1cd8643fc 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -8258,7 +8258,7 @@ static int mdb_pid_insert(pid_t *ids, pid_t pid)
 			return -1;
 		}
 	}
-	
+
 	if( val > 0 ) {
 		++cursor;
 	}
diff --git a/libraries/liblmdb/mdb_stat.c b/libraries/liblmdb/mdb_stat.c
index aeb573a495..40bd4ccf1d 100644
--- a/libraries/liblmdb/mdb_stat.c
+++ b/libraries/liblmdb/mdb_stat.c
@@ -199,7 +199,7 @@ int main(int argc, char *argv[])
 		printf("mdb_open failed, error %d %s\n", rc, mdb_strerror(rc));
 		goto txn_abort;
 	}
-   
+
 	rc = mdb_stat(txn, dbi, &mst);
 	if (rc) {
 		printf("mdb_stat failed, error %d %s\n", rc, mdb_strerror(rc));
diff --git a/libraries/liblmdb/midl.c b/libraries/liblmdb/midl.c
index 0225af9584..5ee2129046 100644
--- a/libraries/liblmdb/midl.c
+++ b/libraries/liblmdb/midl.c
@@ -59,7 +59,7 @@ unsigned mdb_midl_search( MDB_IDL ids, MDB_ID id )
 			return cursor;
 		}
 	}
-	
+
 	if( val > 0 ) {
 		++cursor;
 	}
@@ -89,7 +89,7 @@ int mdb_midl_insert( MDB_IDL ids, MDB_ID id )
 		/* no room */
 		--ids[0];
 		return -2;
-	
+
 	} else {
 		/* insert id */
 		for (i=ids[0]; i>x; i--)

From 2520247ba156ae2f881c800ccc47792c622725ae Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@symas.com>
Date: Fri, 4 Oct 2013 02:46:38 -0700
Subject: [PATCH 25/25] ITS#7715 always set filesize for WRITEMAP

---
 libraries/liblmdb/mdb.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index d1cd8643fc..0aae740614 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -3272,7 +3272,7 @@ mdb_env_map(MDB_env *env, void *addr, int newsize)
 	int prot = PROT_READ;
 	if (flags & MDB_WRITEMAP) {
 		prot |= PROT_WRITE;
-		if (newsize && ftruncate(env->me_fd, env->me_mapsize) < 0)
+		if (ftruncate(env->me_fd, env->me_mapsize) < 0)
 			return ErrCode();
 	}
 	env->me_map = mmap(addr, env->me_mapsize, prot, MAP_SHARED,