From 7a76ded03062ca4ff1499cc6ddc48b5e6a83ee2e Mon Sep 17 00:00:00 2001
From: Howard Chu <hyc@openldap.org>
Date: Mon, 23 Nov 2015 01:30:02 +0000
Subject: [PATCH] ITS#8321 track temporary cursors

In rebalance/split operations, temporary cursors need to be visible
to propagate fixups
---
 libraries/liblmdb/mdb.c | 51 ++++++++++++++++++++++++++++-------------
 1 file changed, 35 insertions(+), 16 deletions(-)

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index 638f5a9..752fdfa 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -7524,6 +7524,22 @@ mdb_update_key(MDB_cursor *mc, MDB_val *key)
 static void
 mdb_cursor_copy(const MDB_cursor *csrc, MDB_cursor *cdst);
 
+/** Track a temporary cursor */
+#define CURSOR_TMP_TRACK(mc, mn, dummy, tracked) \
+	if (mc->mc_flags & C_SUB) { \
+		dummy.mc_flags =  C_INITIALIZED; \
+		dummy.mc_xcursor = (MDB_xcursor *)&mn; \
+		tracked = &dummy; \
+	} else { \
+		tracked = &mn; \
+	} \
+	tracked->mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi]; \
+	mc->mc_txn->mt_cursors[mc->mc_dbi] = tracked
+
+/** Stop tracking a temporary cursor */
+#define CURSOR_TMP_UNTRACK(mc, tracked) \
+	mc->mc_txn->mt_cursors[mc->mc_dbi] = tracked->mc_next
+
 /** Move a node from csrc to cdst.
  */
 static int
@@ -7679,6 +7695,7 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 	 */
 	if (csrc->mc_ki[csrc->mc_top] == 0) {
 		if (csrc->mc_ki[csrc->mc_top-1] != 0) {
+			MDB_cursor dummy, *tracked;
 			if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
 				key.mv_data = LEAF2KEY(csrc->mc_pg[csrc->mc_top], 0, key.mv_size);
 			} else {
@@ -7691,7 +7708,11 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 			mdb_cursor_copy(csrc, &mn);
 			mn.mc_snum--;
 			mn.mc_top--;
-			if ((rc = mdb_update_key(&mn, &key)) != MDB_SUCCESS)
+			/* We want mdb_rebalance to find mn when doing fixups */
+			CURSOR_TMP_TRACK(csrc, mn, dummy, tracked);
+			rc = mdb_update_key(&mn, &key);
+			CURSOR_TMP_UNTRACK(csrc, tracked);
+			if (rc)
 				return rc;
 		}
 		if (IS_BRANCH(csrc->mc_pg[csrc->mc_top])) {
@@ -7707,6 +7728,7 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 
 	if (cdst->mc_ki[cdst->mc_top] == 0) {
 		if (cdst->mc_ki[cdst->mc_top-1] != 0) {
+			MDB_cursor dummy, *tracked;
 			if (IS_LEAF2(csrc->mc_pg[csrc->mc_top])) {
 				key.mv_data = LEAF2KEY(cdst->mc_pg[cdst->mc_top], 0, key.mv_size);
 			} else {
@@ -7719,7 +7741,11 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 			mdb_cursor_copy(cdst, &mn);
 			mn.mc_snum--;
 			mn.mc_top--;
-			if ((rc = mdb_update_key(&mn, &key)) != MDB_SUCCESS)
+			/* We want mdb_rebalance to find mn when doing fixups */
+			CURSOR_TMP_TRACK(cdst, mn, dummy, tracked);
+			rc = mdb_update_key(&mn, &key);
+			CURSOR_TMP_UNTRACK(cdst, tracked);
+			if (rc)
 				return rc;
 		}
 		if (IS_BRANCH(cdst->mc_pg[cdst->mc_top])) {
@@ -8077,24 +8103,13 @@ mdb_rebalance(MDB_cursor *mc)
 		if (!fromleft) {
 			rc = mdb_page_merge(&mn, mc);
 		} else {
-			MDB_cursor dummy;
+			MDB_cursor dummy, *tracked;
 			oldki += NUMKEYS(mn.mc_pg[mn.mc_top]);
 			mn.mc_ki[mn.mc_top] += mc->mc_ki[mn.mc_top] + 1;
 			/* We want mdb_rebalance to find mn when doing fixups */
-			if (mc->mc_flags & C_SUB) {
-				dummy.mc_flags = C_INITIALIZED;
-				dummy.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi];
-				mc->mc_txn->mt_cursors[mc->mc_dbi] = &dummy;
-				dummy.mc_xcursor = (MDB_xcursor *)&mn;
-			} else {
-				mn.mc_next = mc->mc_txn->mt_cursors[mc->mc_dbi];
-				mc->mc_txn->mt_cursors[mc->mc_dbi] = &mn;
-			}
+			CURSOR_TMP_TRACK(mc, mn, dummy, tracked);
 			rc = mdb_page_merge(mc, &mn);
-			if (mc->mc_flags & C_SUB)
-				mc->mc_txn->mt_cursors[mc->mc_dbi] = dummy.mc_next;
-			else
-				mc->mc_txn->mt_cursors[mc->mc_dbi] = mn.mc_next;
+			CURSOR_TMP_UNTRACK(mc, tracked);
 			mdb_cursor_copy(&mn, mc);
 		}
 		mc->mc_flags &= ~C_EOF;
@@ -8460,10 +8475,14 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno
 	 */
 	if (SIZELEFT(mn.mc_pg[ptop]) < mdb_branch_size(env, &sepkey)) {
 		int snum = mc->mc_snum;
+		MDB_cursor dummy, *tracked;
 		mn.mc_snum--;
 		mn.mc_top--;
 		did_split = 1;
+		/* We want other splits to find mn when doing fixups */
+		CURSOR_TMP_TRACK(mc, mn, dummy, tracked);
 		rc = mdb_page_split(&mn, &sepkey, NULL, rp->mp_pgno, 0);
+		CURSOR_TMP_UNTRACK(mc, tracked);
 		if (rc)
 			goto done;