X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=libraries%2Fliblmdb%2Fmdb.c;h=3f78d3c4830225a0356aca53997e7c9c630b552d;hb=b922a5a086bc5f615e69aa82de5dff46b6ea8895;hp=b0faf6d3f55755f8654ae1fe1ea7ed192ef0f636;hpb=6b1df0e4c7fadd21d1233d7157229b2d89ccaa04;p=openldap

diff --git a/libraries/liblmdb/mdb.c b/libraries/liblmdb/mdb.c
index b0faf6d3f5..3f78d3c483 100644
--- a/libraries/liblmdb/mdb.c
+++ b/libraries/liblmdb/mdb.c
@@ -291,8 +291,10 @@ typedef HANDLE mdb_mutex_t, mdb_mutexref_t;
 #define pthread_mutex_lock(x)	WaitForSingleObject(*x, INFINITE)
 #define pthread_cond_signal(x)	SetEvent(*x)
 #define pthread_cond_wait(cond,mutex)	do{SignalObjectAndWait(*mutex, *cond, INFINITE, FALSE); WaitForSingleObject(*mutex, INFINITE);}while(0)
-#define THREAD_CREATE(thr,start,arg)	thr=CreateThread(NULL,0,start,arg,0,NULL)
-#define THREAD_FINISH(thr)	WaitForSingleObject(thr, INFINITE)
+#define THREAD_CREATE(thr,start,arg) \
+	(((thr) = CreateThread(NULL, 0, start, arg, 0, NULL)) ? 0 : ErrCode())
+#define THREAD_FINISH(thr) \
+	(WaitForSingleObject(thr, INFINITE) ? ErrCode() : 0)
 #define LOCK_MUTEX0(mutex)		WaitForSingleObject(mutex, INFINITE)
 #define UNLOCK_MUTEX(mutex)		ReleaseMutex(mutex)
 #define mdb_mutex_consistent(mutex)	0
@@ -1188,6 +1190,21 @@ typedef struct MDB_xcursor {
 	unsigned char mx_dbflag;
 } MDB_xcursor;
 
+	/** Check if there is an inited xcursor, so #XCURSOR_REFRESH() is proper */
+#define XCURSOR_INITED(mc) \
+	((mc)->mc_xcursor && ((mc)->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED))
+
+	/** Update sub-page pointer, if any, in \b mc->mc_xcursor.  Needed
+	 *	when the node which contains the sub-page may have moved.  Called
+	 *	with \b mp = mc->mc_pg[mc->mc_top], \b ki = mc->mc_ki[mc->mc_top].
+	 */
+#define XCURSOR_REFRESH(mc, mp, ki) do { \
+	MDB_page *xr_pg = (mp); \
+	MDB_node *xr_node = NODEPTR(xr_pg, ki); \
+	if ((xr_node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) \
+		(mc)->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(xr_node); \
+} while (0)
+
 	/** State of FreeDB old pages, stored in the MDB_env */
 typedef struct MDB_pgstate {
 	pgno_t		*mf_pghead;	/**< Reclaimed freeDB pages, or NULL before use */
@@ -1422,7 +1439,7 @@ mdb_strerror(int err)
 	 *	and the actual use of the message uses more than 4K of stack.
 	 */
 #define MSGSIZE	1024
-#define PADSIZE 4096
+#define PADSIZE	4096
 	char buf[MSGSIZE+PADSIZE], *ptr = buf;
 #endif
 	int i;
@@ -1618,7 +1635,7 @@ mdb_cursor_chk(MDB_cursor *mc)
 	}
 	if (mc->mc_ki[i] >= NUMKEYS(mc->mc_pg[i]))
 		printf("ack!\n");
-	if (mc->mc_xcursor && (mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
+	if (XCURSOR_INITED(mc)) {
 		node = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
 		if (((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) &&
 			mc->mc_xcursor->mx_cursor.mc_pg[0] != NODEDATA(node)) {
@@ -2433,14 +2450,8 @@ done:
 			if (m2 == mc) continue;
 			if (m2->mc_pg[mc->mc_top] == mp) {
 				m2->mc_pg[mc->mc_top] = np;
-				if ((mc->mc_db->md_flags & MDB_DUPSORT) &&
-					IS_LEAF(np) &&
-					(m2->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED))
-				{
-					MDB_node *leaf = NODEPTR(np, m2->mc_ki[mc->mc_top]);
-					if ((leaf->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA)
-						m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
-				}
+				if (XCURSOR_INITED(m2) && IS_LEAF(np))
+					XCURSOR_REFRESH(m2, np, m2->mc_ki[mc->mc_top]);
 			}
 		}
 	}
@@ -6206,6 +6217,30 @@ fetchm:
 			}
 		}
 		break;
+	case MDB_PREV_MULTIPLE:
+		if (data == NULL) {
+			rc = EINVAL;
+			break;
+		}
+		if (!(mc->mc_db->md_flags & MDB_DUPFIXED)) {
+			rc = MDB_INCOMPATIBLE;
+			break;
+		}
+		if (!(mc->mc_flags & C_INITIALIZED))
+			rc = mdb_cursor_last(mc, key, data);
+		else
+			rc = MDB_SUCCESS;
+		if (rc == MDB_SUCCESS) {
+			MDB_cursor *mx = &mc->mc_xcursor->mx_cursor;
+			if (mx->mc_flags & C_INITIALIZED) {
+				rc = mdb_cursor_sibling(mx, 0);
+				if (rc == MDB_SUCCESS)
+					goto fetchm;
+			} else {
+				rc = MDB_NOTFOUND;
+			}
+		}
+		break;
 	case MDB_NEXT:
 	case MDB_NEXT_DUP:
 	case MDB_NEXT_NODUP:
@@ -6726,11 +6761,8 @@ new_sub:
 				if (m3->mc_ki[i] >= mc->mc_ki[i] && insert_key) {
 					m3->mc_ki[i]++;
 				}
-				if (m3->mc_xcursor && (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
-					MDB_node *n2 = NODEPTR(mp, m3->mc_ki[i]);
-					if ((n2->mn_flags & (F_SUBDATA|F_DUPDATA)) == F_DUPDATA)
-						m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2);
-				}
+				if (XCURSOR_INITED(m3))
+					XCURSOR_REFRESH(m3, mp, m3->mc_ki[i]);
 			}
 		}
 	}
@@ -6781,9 +6813,7 @@ put_sub:
 						if (m2->mc_ki[i] == mc->mc_ki[i]) {
 							mdb_xcursor_init2(m2, mx, new_dupdata);
 						} else if (!insert_key && m2->mc_ki[i] < nkeys) {
-							MDB_node *n2 = NODEPTR(mp, m2->mc_ki[i]);
-							if ((n2->mn_flags & (F_SUBDATA|F_DUPDATA)) == F_DUPDATA)
-								m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2);
+							XCURSOR_REFRESH(m2, mp, m2->mc_ki[i]);
 						}
 					}
 				}
@@ -6888,13 +6918,12 @@ mdb_cursor_del(MDB_cursor *mc, unsigned int flags)
 						if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
 						if (!(m2->mc_flags & C_INITIALIZED)) continue;
 						if (m2->mc_pg[mc->mc_top] == mp) {
-							if (m2->mc_ki[mc->mc_top] == mc->mc_ki[mc->mc_top]) {
-								m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(leaf);
-							} else {
-								MDB_node *n2 = NODEPTR(mp, m2->mc_ki[mc->mc_top]);
-								if (!(n2->mn_flags & F_SUBDATA))
-									m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2);
+							MDB_node *n2 = leaf;
+							if (m2->mc_ki[mc->mc_top] != mc->mc_ki[mc->mc_top]) {
+								n2 = NODEPTR(mp, m2->mc_ki[mc->mc_top]);
+								if (n2->mn_flags & F_SUBDATA) continue;
 							}
+							m2->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(n2);
 						}
 					}
 				}
@@ -7731,12 +7760,8 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 					m3->mc_ki[csrc->mc_top] = cdst->mc_ki[cdst->mc_top];
 					m3->mc_ki[csrc->mc_top-1]++;
 				}
-				if (m3->mc_xcursor && (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) &&
-					IS_LEAF(mps)) {
-					MDB_node *node = NODEPTR(m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]);
-					if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA)
-						m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
-				}
+				if (XCURSOR_INITED(m3) && IS_LEAF(mps))
+					XCURSOR_REFRESH(m3, m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]);
 			}
 		} else
 		/* Adding on the right, bump others down */
@@ -7757,12 +7782,8 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 					} else {
 						m3->mc_ki[csrc->mc_top]--;
 					}
-					if (m3->mc_xcursor && (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) &&
-						IS_LEAF(mps)) {
-						MDB_node *node = NODEPTR(m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]);
-						if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA)
-							m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
-					}
+					if (XCURSOR_INITED(m3) && IS_LEAF(mps))
+						XCURSOR_REFRESH(m3, m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]);
 				}
 			}
 		}
@@ -7963,12 +7984,8 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst)
 				m3->mc_ki[top-1] > csrc->mc_ki[top-1]) {
 				m3->mc_ki[top-1]--;
 			}
-			if (m3->mc_xcursor && (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) &&
-				IS_LEAF(psrc)) {
-				MDB_node *node = NODEPTR(m3->mc_pg[top], m3->mc_ki[top]);
-				if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA)
-					m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
-			}
+			if (XCURSOR_INITED(m3) && IS_LEAF(psrc))
+				XCURSOR_REFRESH(m3, m3->mc_pg[top], m3->mc_ki[top]);
 		}
 	}
 	{
@@ -8230,11 +8247,8 @@ mdb_cursor_del0(MDB_cursor *mc)
 				} else if (m3->mc_ki[mc->mc_top] > ki) {
 					m3->mc_ki[mc->mc_top]--;
 				}
-				if (m3->mc_xcursor && (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)) {
-					MDB_node *node = NODEPTR(m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]);
-					if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA)
-						m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
-				}
+				if (XCURSOR_INITED(m3))
+					XCURSOR_REFRESH(m3, m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]);
 			}
 		}
 	}
@@ -8764,12 +8778,8 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno
 				m3->mc_ki[ptop] >= mc->mc_ki[ptop]) {
 				m3->mc_ki[ptop]++;
 			}
-			if (m3->mc_xcursor && (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) &&
-				IS_LEAF(mp)) {
-				MDB_node *node = NODEPTR(m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]);
-				if ((node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA)
-					m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
-			}
+			if (XCURSOR_INITED(m3) && IS_LEAF(mp))
+				XCURSOR_REFRESH(m3, m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]);
 		}
 	}
 	DPRINTF(("mp left: %d, rp left: %d", SIZELEFT(mp), SIZELEFT(rp)));
@@ -8810,23 +8820,26 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi,
 #ifndef MDB_WBUF
 #define MDB_WBUF	(1024*1024)
 #endif
+#define MDB_EOF		0x10	/**< #mdb_env_copyfd1() is done reading */
 
-	/** State needed for a compacting copy. */
+	/** State needed for a double-buffering compacting copy. */
 typedef struct mdb_copy {
+	MDB_env *mc_env;
+	MDB_txn *mc_txn;
 	pthread_mutex_t mc_mutex;
-	pthread_cond_t mc_cond;
+	pthread_cond_t mc_cond;	/**< Condition variable for #mc_new */
 	char *mc_wbuf[2];
 	char *mc_over[2];
-	MDB_env *mc_env;
-	MDB_txn *mc_txn;
 	int mc_wlen[2];
 	int mc_olen[2];
 	pgno_t mc_next_pgno;
 	HANDLE mc_fd;
-	int mc_status;
-	volatile int mc_new;
-	int mc_toggle;
-
+	int mc_toggle;			/**< Buffer number in provider */
+	int mc_new;				/**< (0-2 buffers to write) | (#MDB_EOF at end) */
+	/** Error code.  Never cleared if set.  Both threads can set nonzero
+	 *	to fail the copy.  Not mutex-protected, LMDB expects atomic int.
+	 */
+	volatile int mc_error;
 } mdb_copy;
 
 	/** Dedicated writer thread for compacting copy. */
@@ -8845,20 +8858,16 @@ mdb_env_copythr(void *arg)
 #endif
 
 	pthread_mutex_lock(&my->mc_mutex);
-	my->mc_new = 0;
-	pthread_cond_signal(&my->mc_cond);
 	for(;;) {
 		while (!my->mc_new)
 			pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
-		if (my->mc_new < 0) {
-			my->mc_new = 0;
+		if (my->mc_new == 0 + MDB_EOF) /* 0 buffers, just EOF */
 			break;
-		}
-		my->mc_new = 0;
 		wsize = my->mc_wlen[toggle];
 		ptr = my->mc_wbuf[toggle];
 again:
-		while (wsize > 0) {
+		rc = MDB_SUCCESS;
+		while (wsize > 0 && !my->mc_error) {
 			DO_WRITE(rc, my->mc_fd, ptr, wsize, len);
 			if (!rc) {
 				rc = ErrCode();
@@ -8874,8 +8883,7 @@ again:
 			}
 		}
 		if (rc) {
-			my->mc_status = rc;
-			break;
+			my->mc_error = rc;
 		}
 		/* If there's an overflow page tail, write it too */
 		if (my->mc_olen[toggle]) {
@@ -8886,34 +8894,41 @@ again:
 		}
 		my->mc_wlen[toggle] = 0;
 		toggle ^= 1;
+		/* Return the empty buffer to provider */
+		my->mc_new--;
 		pthread_cond_signal(&my->mc_cond);
 	}
-	pthread_cond_signal(&my->mc_cond);
 	pthread_mutex_unlock(&my->mc_mutex);
 	return (THREAD_RET)0;
 #undef DO_WRITE
 }
 
-	/** Tell the writer thread there's a buffer ready to write */
+	/** Give buffer and/or #MDB_EOF to writer thread, await unused buffer.
+	 *
+	 * @param[in] my control structure.
+	 * @param[in] adjust (1 to hand off 1 buffer) | (MDB_EOF when ending).
+	 */
 static int ESECT
-mdb_env_cthr_toggle(mdb_copy *my, int st)
+mdb_env_cthr_toggle(mdb_copy *my, int adjust)
 {
-	int toggle = my->mc_toggle ^ 1;
 	pthread_mutex_lock(&my->mc_mutex);
-	if (my->mc_status) {
-		pthread_mutex_unlock(&my->mc_mutex);
-		return my->mc_status;
-	}
-	while (my->mc_new == 1)
-		pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
-	my->mc_new = st;
-	my->mc_toggle = toggle;
+	my->mc_new += adjust;
 	pthread_cond_signal(&my->mc_cond);
+	while (my->mc_new & 2)		/* both buffers in use */
+		pthread_cond_wait(&my->mc_cond, &my->mc_mutex);
 	pthread_mutex_unlock(&my->mc_mutex);
-	return 0;
+
+	my->mc_toggle ^= (adjust & 1);
+	/* Both threads reset mc_wlen, to be safe from threading errors */
+	my->mc_wlen[my->mc_toggle] = 0;
+	return my->mc_error;
 }
 
-	/** Depth-first tree traversal for compacting copy. */
+	/** Depth-first tree traversal for compacting copy.
+	 * @param[in] my control structure.
+	 * @param[in,out] pg database root.
+	 * @param[in] flags includes #F_DUPDATA if it is a sorted-duplicate sub-DB.
+	 */
 static int ESECT
 mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
 {
@@ -8975,6 +8990,7 @@ mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
 						}
 
 						memcpy(&pg, NODEDATA(ni), sizeof(pg));
+						memcpy(NODEDATA(ni), &my->mc_next_pgno, sizeof(pgno_t));
 						rc = mdb_page_get(&mc, pg, &omp, NULL);
 						if (rc)
 							goto done;
@@ -8997,7 +9013,6 @@ mdb_env_cwalk(mdb_copy *my, pgno_t *pg, int flags)
 								goto done;
 							toggle = my->mc_toggle;
 						}
-						memcpy(NODEDATA(ni), &mo->mp_pgno, sizeof(pgno_t));
 					} else if (ni->mn_flags & F_SUBDATA) {
 						MDB_db db;
 
@@ -9075,47 +9090,56 @@ mdb_env_copyfd1(MDB_env *env, HANDLE fd)
 {
 	MDB_meta *mm;
 	MDB_page *mp;
-	mdb_copy my;
+	mdb_copy my = {0};
 	MDB_txn *txn = NULL;
 	pthread_t thr;
-	int rc;
+	pgno_t root, new_root;
+	int rc = MDB_SUCCESS;
 
 #ifdef _WIN32
-	my.mc_mutex = CreateMutex(NULL, FALSE, NULL);
-	my.mc_cond = CreateEvent(NULL, FALSE, FALSE, NULL);
+	if (!(my.mc_mutex = CreateMutex(NULL, FALSE, NULL)) ||
+		!(my.mc_cond = CreateEvent(NULL, FALSE, FALSE, NULL))) {
+		rc = ErrCode();
+		goto done;
+	}
 	my.mc_wbuf[0] = _aligned_malloc(MDB_WBUF*2, env->me_os_psize);
-	if (my.mc_wbuf[0] == NULL)
-		return errno;
+	if (my.mc_wbuf[0] == NULL) {
+		/* _aligned_malloc() sets errno, but we use Windows error codes */
+		rc = ERROR_NOT_ENOUGH_MEMORY;
+		goto done;
+	}
 #else
-	pthread_mutex_init(&my.mc_mutex, NULL);
-	pthread_cond_init(&my.mc_cond, NULL);
+	if ((rc = pthread_mutex_init(&my.mc_mutex, NULL)) != 0)
+		return rc;
+	if ((rc = pthread_cond_init(&my.mc_cond, NULL)) != 0)
+		goto done2;
 #ifdef HAVE_MEMALIGN
 	my.mc_wbuf[0] = memalign(env->me_os_psize, MDB_WBUF*2);
-	if (my.mc_wbuf[0] == NULL)
-		return errno;
+	if (my.mc_wbuf[0] == NULL) {
+		rc = errno;
+		goto done;
+	}
 #else
-	rc = posix_memalign((void **)&my.mc_wbuf[0], env->me_os_psize, MDB_WBUF*2);
-	if (rc)
-		return rc;
+	{
+		void *p;
+		if ((rc = posix_memalign(&p, env->me_os_psize, MDB_WBUF*2)) != 0)
+			goto done;
+		my.mc_wbuf[0] = p;
+	}
 #endif
 #endif
 	memset(my.mc_wbuf[0], 0, MDB_WBUF*2);
 	my.mc_wbuf[1] = my.mc_wbuf[0] + MDB_WBUF;
-	my.mc_wlen[0] = 0;
-	my.mc_wlen[1] = 0;
-	my.mc_olen[0] = 0;
-	my.mc_olen[1] = 0;
 	my.mc_next_pgno = NUM_METAS;
-	my.mc_status = 0;
-	my.mc_new = 1;
-	my.mc_toggle = 0;
 	my.mc_env = env;
 	my.mc_fd = fd;
-	THREAD_CREATE(thr, mdb_env_copythr, &my);
+	rc = THREAD_CREATE(thr, mdb_env_copythr, &my);
+	if (rc)
+		goto done;
 
 	rc = mdb_txn_begin(env, NULL, MDB_RDONLY, &txn);
 	if (rc)
-		return rc;
+		goto finish;
 
 	mp = (MDB_page *)my.mc_wbuf[0];
 	memset(mp, 0, NUM_METAS * env->me_psize);
@@ -9131,57 +9155,64 @@ mdb_env_copyfd1(MDB_env *env, HANDLE fd)
 	*(MDB_meta *)METADATA(mp) = *mm;
 	mm = (MDB_meta *)METADATA(mp);
 
-	/* Count the number of free pages, subtract from lastpg to find
-	 * number of active pages
-	 */
-	{
+	/* Set metapage 1 with current main DB */
+	root = new_root = txn->mt_dbs[MAIN_DBI].md_root;
+	if (root != P_INVALID) {
+		/* Count free pages + freeDB pages.  Subtract from last_pg
+		 * to find the new last_pg, which also becomes the new root.
+		 */
 		MDB_ID freecount = 0;
 		MDB_cursor mc;
 		MDB_val key, data;
 		mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
 		while ((rc = mdb_cursor_get(&mc, &key, &data, MDB_NEXT)) == 0)
 			freecount += *(MDB_ID *)data.mv_data;
+		if (rc != MDB_NOTFOUND)
+			goto finish;
 		freecount += txn->mt_dbs[FREE_DBI].md_branch_pages +
 			txn->mt_dbs[FREE_DBI].md_leaf_pages +
 			txn->mt_dbs[FREE_DBI].md_overflow_pages;
 
-		/* Set metapage 1 */
-		mm->mm_last_pg = txn->mt_next_pgno - freecount - 1;
+		new_root = txn->mt_next_pgno - 1 - freecount;
+		mm->mm_last_pg = new_root;
 		mm->mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI];
-		if (mm->mm_last_pg > NUM_METAS-1) {
-			mm->mm_dbs[MAIN_DBI].md_root = mm->mm_last_pg;
-			mm->mm_txnid = 1;
-		} else {
-			mm->mm_dbs[MAIN_DBI].md_root = P_INVALID;
-		}
+		mm->mm_dbs[MAIN_DBI].md_root = new_root;
+	} else {
+		/* When the DB is empty, handle it specially to
+		 * fix any breakage like page leaks from ITS#8174.
+		 */
+		mm->mm_dbs[MAIN_DBI].md_flags = txn->mt_dbs[MAIN_DBI].md_flags;
+	}
+	if (root != P_INVALID || mm->mm_dbs[MAIN_DBI].md_flags) {
+		mm->mm_txnid = 1;		/* use metapage 1 */
 	}
+
 	my.mc_wlen[0] = env->me_psize * NUM_METAS;
 	my.mc_txn = txn;
-	pthread_mutex_lock(&my.mc_mutex);
-	while(my.mc_new)
-		pthread_cond_wait(&my.mc_cond, &my.mc_mutex);
-	pthread_mutex_unlock(&my.mc_mutex);
-	rc = mdb_env_cwalk(&my, &txn->mt_dbs[MAIN_DBI].md_root, 0);
-	if (rc == MDB_SUCCESS && my.mc_wlen[my.mc_toggle])
-		rc = mdb_env_cthr_toggle(&my, 1);
-	mdb_env_cthr_toggle(&my, -1);
-	pthread_mutex_lock(&my.mc_mutex);
-	while(my.mc_new)
-		pthread_cond_wait(&my.mc_cond, &my.mc_mutex);
-	pthread_mutex_unlock(&my.mc_mutex);
-	THREAD_FINISH(thr);
+	rc = mdb_env_cwalk(&my, &root, 0);
+	if (rc == MDB_SUCCESS && root != new_root) {
+		rc = MDB_INCOMPATIBLE;	/* page leak or corrupt DB */
+	}
 
+finish:
+	if (rc)
+		my.mc_error = rc;
+	mdb_env_cthr_toggle(&my, 1 | MDB_EOF);
+	rc = THREAD_FINISH(thr);
 	mdb_txn_abort(txn);
+
+done:
 #ifdef _WIN32
-	CloseHandle(my.mc_cond);
-	CloseHandle(my.mc_mutex);
-	_aligned_free(my.mc_wbuf[0]);
+	if (my.mc_wbuf[0]) _aligned_free(my.mc_wbuf[0]);
+	if (my.mc_cond)  CloseHandle(my.mc_cond);
+	if (my.mc_mutex) CloseHandle(my.mc_mutex);
 #else
+	free(my.mc_wbuf[0]);
 	pthread_cond_destroy(&my.mc_cond);
+done2:
 	pthread_mutex_destroy(&my.mc_mutex);
-	free(my.mc_wbuf[0]);
 #endif
-	return rc;
+	return rc ? rc : my.mc_error;
 }
 
 	/** Copy environment as-is. */