mdbx: backport - avoid FreeDB corruption due deep recursive rebalance from freelist_save().

Change-Id: I65b0b62c3e787802c0150c74826f181a8f6d7902

.circleci/config.yml

@@ -0,0 +1,9 @@
+version: 2
+jobs:
+  build:
+    docker:
+      - image: circleci/buildpack-deps:artful
+    steps:
+      - checkout
+      - run: make all lmdb
+      - run: make check

CHANGES

@@ -3,7 +3,28 @@ MDBX
 	Add error MDB_PROBLEM, replace some MDB_CORRUPTED
 	Workarounds for glibc bugs: #21031 and 21032.
 
-LMDB 0.9.20 Release Engineering
+LMDB 0.9.24 Engineering
+	ITS#8969 Tweak mdb_page_split
+
+LMDB 0.9.23 Release (2018/12/19)
+	ITS#8756 Fix loose pages in dirty list
+	ITS#8831 Fix mdb_load flag init
+	ITS#8844 Fix mdb_env_close in forked process
+	Documentation
+		ITS#8857 mdb_cursor_del doesn't invalidate cursor
+		ITS#8908 GET_MULTIPLE etc don't change passed in key
+
+LMDB 0.9.22 Release (2018/03/22)
+	Fix MDB_DUPSORT alignment bug (ITS#8819)
+	Fix regression with new db from 0.9.19 (ITS#8760)
+	Fix liblmdb to build on Solaris (ITS#8612)
+	Fix delete behavior with DUPSORT DB (ITS#8622)
+	Fix mdb_cursor_get/mdb_cursor_del behavior (ITS#8722)
+
+LMDB 0.9.21 Release (2017/06/01)
+	Fix xcursor after cursor_del (ITS#8622)
+
+LMDB 0.9.20 (Withdrawn)
 	Fix mdb_load with escaped plaintext (ITS#8558)
 	Fix mdb_cursor_last / mdb_put interaction (ITS#8557)
 

Doxyfile

@@ -57,7 +57,7 @@ CREATE_SUBDIRS         = NO
 # Croatian, Czech, Danish, Dutch, Esperanto, Farsi, Finnish, French, German,
 # Greek, Hungarian, Italian, Japanese, Japanese-en (Japanese with English
 # messages), Korean, Korean-en, Lithuanian, Norwegian, Macedonian, Persian,
-# Polish, Portuguese, Romanian, Russian, Serbian, Serbian-Cyrilic, Slovak,
+# Polish, Portuguese, Romanian, Russian, Serbian, Serbian-Cyrillic, Slovak,
 # Slovene, Spanish, Swedish, Ukrainian, and Vietnamese.
 
 OUTPUT_LANGUAGE        = English
@@ -211,7 +211,7 @@ OPTIMIZE_OUTPUT_VHDL   = NO
 # parses. With this tag you can assign which parser to use for a given extension.
 # Doxygen has a built-in mapping, but you can override or extend it using this
 # tag. The format is ext=language, where ext is a file extension, and language
-# is one of the parsers supported by doxygen: IDL, Java, Javascript, CSharp, C,
+# is one of the parsers supported by doxygen: IDL, Java, JavaScript, CSharp, C,
 # C++, D, PHP, Objective-C, Python, Fortran, VHDL, C, C++. For instance to make
 # doxygen treat .inc files as Fortran files (default is PHP), and .f files as C
 # (default is Fortran), use: inc=Fortran f=C. Note that for custom extensions
@@ -550,7 +550,7 @@ WARN_IF_UNDOCUMENTED   = YES
 
 WARN_IF_DOC_ERROR      = YES
 
-# This WARN_NO_PARAMDOC option can be abled to get warnings for
+# This WARN_NO_PARAMDOC option can be enabled to get warnings for
 # functions that are documented, but have no documentation for their parameters
 # or return value. If set to NO (the default) doxygen will only warn about
 # wrong or incomplete parameter documentation, but not about the absence of
@@ -1051,7 +1051,7 @@ EXT_LINKS_IN_WINDOW    = NO
 
 FORMULA_FONTSIZE       = 10
 
-# Use the FORMULA_TRANPARENT tag to determine whether or not the images
+# Use the FORMULA_TRANSPARENT tag to determine whether or not the images
 # generated for formulas are transparent PNGs. Transparent PNGs are
 # not supported properly for IE 6.0, but are supported on all modern browsers.
 # Note that when changing this option you need to delete any form_*.png files
@@ -1071,7 +1071,7 @@ SEARCHENGINE           = YES
 
 # When the SERVER_BASED_SEARCH tag is enabled the search engine will be
 # implemented using a PHP enabled web server instead of at the web client
-# using Javascript. Doxygen will generate the search PHP script and index
+# using JavaScript. Doxygen will generate the search PHP script and index
 # file to put on the web server. The advantage of the server
 # based approach is that it scales better to large projects and allows
 # full text search. The disadvances is that it is more difficult to setup

Makefile

@@ -1,5 +1,5 @@
 # GNU Makefile for libmdbx (reliable lightning memory-mapped DB library for Linux).
-# https://github.com/ReOpen/libmdbx
+# https://github.com/leo-yuriev/libmdbx
 
 ########################################################################
 # Configuration. The compiler options must enable threaded compilation.
@@ -71,7 +71,7 @@ clean:
 
 tests:	$(TESTS)
 
-check:	tests
+check:	tests mdbx_chk
 	[ -d testdb ] || mkdir testdb && rm -f testdb/* \
 		&& echo "*** LMDB-TEST-0" && ./mtest0 && ./mdbx_chk -v testdb \
 		&& echo "*** LMDB-TEST-1" && ./mtest1 && ./mdbx_chk -v testdb \

README.md

@@ -3,10 +3,18 @@ libmdbx
 Extended LMDB, aka "Расширенная LMDB".
 
 *The Future will Positive. Всё будет хорошо.*
-[![Build Status](https://travis-ci.org/ReOpen/libmdbx.svg?branch=master)](https://travis-ci.org/ReOpen/libmdbx)
+[![Build Status](https://travis-ci.org/leo-yuriev/libmdbx.svg?branch=stable%2F0.0)](https://travis-ci.org/leo-yuriev/libmdbx)
 
-English version by Google [is here](https://translate.googleusercontent.com/translate_c?act=url&ie=UTF8&sl=ru&tl=en&u=https://github.com/ReOpen/libmdbx/tree/master).
+## Project Status for now
 
+ - The stable versions (the _stable/*_ branches) of  are frozen, i.e. no new features or API changes, but only bug fixes.
+ - The next (the _devel_ branch) version **is under active development**, i.e. current API and set of features are extreme volatile.
+ - The immediate goal of development is formation of the stable API and the stable internal database format, which allows realise all planned features.
+ - Planned features: Integrity check by Merkle tree, Support for raw block devices, Separate place for large data items, Using "roaring bitmaps" for garbage collector, Non-linear page reclaiming, Asynchronous lazy data flush to disk(s), etc.
+
+-----
+
+English version by Google [is here](https://translate.googleusercontent.com/translate_c?act=url&ie=UTF8&sl=ru&tl=en&u=https://github.com/leo-yuriev/libmdbx/tree/stable%2F0.0).
 
 ## Кратко
 
@@ -28,7 +36,7 @@ _libmdbx_ является потомком "Lightning Memory-Mapped Database",
 известной под аббревиатурой
 [LMDB](https://en.wikipedia.org/wiki/Lightning_Memory-Mapped_Database).
 Изначально доработка производилась в составе проекта
-[ReOpenLDAP](https://github.com/ReOpen/ReOpenLDAP). Примерно за год
+[ReOpenLDAP](https://github.com/leo-yuriev/ReOpenLDAP). Примерно за год
 работы внесенные изменения приобрели самостоятельную ценность. Осенью
 2015 доработанный движок был выделен в отдельный проект, который был
 [представлен на конференции Highload++

circle.yml

@@ -1,14 +0,0 @@
-machine:
-  timezone:
-    Europe/Moscow
-
-database:
-  override:
-
-compile:
-  override:
-    - make all lmdb
-
-test:
-  override:
-    - make check

lmdb.h

@@ -212,7 +212,7 @@ typedef int mdb_filehandle_t;
 #define MDB_VERSION_DATE	"2017-02-17"
 
 /** A stringifier for the version info */
-#define MDB_VERSTR(a,b,c,d)	"MDBX " #a "." #b "." #c ": (" d ", https://github.com/ReOpen/libmdbx)"
+#define MDB_VERSTR(a,b,c,d)	"MDBX " #a "." #b "." #c ": (" d ", https://github.com/leo-yuriev/libmdbx)"
 
 /** A helper for the stringifier macro */
 #define MDB_VERFOO(a,b,c,d)	MDB_VERSTR(a,b,c,d)
@@ -390,7 +390,7 @@ typedef enum MDB_cursor_op {
 	MDB_GET_BOTH,			/**< Position at key/data pair. Only for #MDB_DUPSORT */
 	MDB_GET_BOTH_RANGE,		/**< position at key, nearest data. Only for #MDB_DUPSORT */
 	MDB_GET_CURRENT,		/**< Return key/data at current cursor position */
-	MDB_GET_MULTIPLE,		/**< Return key and up to a page of duplicate data items
+	MDB_GET_MULTIPLE,		/**< Return up to a page of duplicate data items
 								from current cursor position. Move cursor to prepare
 								for #MDB_NEXT_MULTIPLE. Only for #MDB_DUPFIXED */
 	MDB_LAST,				/**< Position at last key/data item */
@@ -399,7 +399,7 @@ typedef enum MDB_cursor_op {
 	MDB_NEXT,				/**< Position at next data item */
 	MDB_NEXT_DUP,			/**< Position at next data item of current key.
 								Only for #MDB_DUPSORT */
-	MDB_NEXT_MULTIPLE,		/**< Return key and up to a page of duplicate data items
+	MDB_NEXT_MULTIPLE,		/**< Return up to a page of duplicate data items
 								from next cursor position. Move cursor to prepare
 								for #MDB_NEXT_MULTIPLE. Only for #MDB_DUPFIXED */
 	MDB_NEXT_NODUP,			/**< Position at first data item of next key */
@@ -410,7 +410,7 @@ typedef enum MDB_cursor_op {
 	MDB_SET,				/**< Position at specified key */
 	MDB_SET_KEY,			/**< Position at specified key, return key + data */
 	MDB_SET_RANGE,			/**< Position at first key greater than or equal to specified key. */
-	MDB_PREV_MULTIPLE		/**< Position at previous page and return key and up to
+	MDB_PREV_MULTIPLE		/**< Position at previous page and return up to
 								a page of duplicate data items. Only for #MDB_DUPFIXED */
 } MDB_cursor_op;
 
@@ -612,7 +612,7 @@ int  mdb_env_create(MDB_env **env);
 	 *	<li>#MDB_NOTLS
 	 *		Don't use Thread-Local Storage. Tie reader locktable slots to
 	 *		#MDB_txn objects instead of to threads. I.e. #mdb_txn_reset() keeps
-	 *		the slot reseved for the #MDB_txn object. A thread may use parallel
+	 *		the slot reserved for the #MDB_txn object. A thread may use parallel
 	 *		read-only transactions. A read-only transaction may span threads if
 	 *		the user synchronizes its use. Applications that multiplex many
 	 *		user threads over individual OS threads need this option. Such an
@@ -1589,6 +1589,10 @@ int  mdb_cursor_put(MDB_cursor *cursor, MDB_val *key, MDB_val *data,
 	/** @brief Delete current key/data pair
 	 *
 	 * This function deletes the key/data pair to which the cursor refers.
+	 * This does not invalidate the cursor, so operations such as MDB_NEXT
+	 * can still be used on it.
+	 * Both MDB_NEXT and MDB_GET_CURRENT will return the same record after
+	 * this operation.
 	 * @param[in] cursor A cursor handle returned by #mdb_cursor_open()
 	 * @param[in] flags Options for this operation. This parameter
 	 * must be set to 0 or one of the values described here.

mdb.c

@@ -44,10 +44,8 @@
 #	define _GNU_SOURCE
 #endif
 
-/* LY: Please do not ask us for Windows support, just never!
- * But you can make a fork for Windows, or become maintainer for FreeBSD... */
 #ifndef __gnu_linux__
-#	warning "This version of ReOpenMDBX supports only GNU Linux"
+#	warning "This version of libmdbx supports only GNU Linux"
 #endif
 
 #include <stddef.h>
@@ -56,21 +54,21 @@
 #include "./defs.h"
 
 #if !__GNUC_PREREQ(4,2)
-	/* LY: Actualy ReOpenMDBX was not tested with compilers
+	/* LY: Actualy libmdbx was not tested with compilers
 	 *     older than GCC 4.4 (from RHEL6).
 	 * But you could remove this #error and try to continue at your own risk.
 	 * In such case please don't rise up an issues related ONLY to old compilers.
 	 */
-#	warning "ReOpenMDBX required at least GCC 4.2 compatible C/C++ compiler."
+#	warning "libmdbx required at least GCC 4.2 compatible C/C++ compiler."
 #endif
 
 #if !__GLIBC_PREREQ(2,12)
-	/* LY: Actualy ReOpenMDBX was not tested with something
+	/* LY: Actualy libmdbx was not tested with something
 	 *     older than glibc 2.12 (from RHEL6).
 	 * But you could remove this #error and try to continue at your own risk.
 	 * In such case please don't rise up an issues related ONLY to old systems.
 	 */
-#	warning "ReOpenMDBX required at least GLIBC 2.12."
+#	warning "libmdbx required at least GLIBC 2.12."
 #endif
 
 #if MDB_DEBUG
@@ -604,6 +602,8 @@ typedef struct MDB_page {
 	indx_t		mp_ptrs[1];		/**< dynamic size */
 } MDB_page;
 
+#define PAGETYPE(p)	 ((p)->mp_flags & (P_BRANCH | P_LEAF | P_LEAF2 | P_OVERFLOW))
+
 	/** Size of the page header, excluding dynamic data at the end */
 #define PAGEHDRSZ	 ((unsigned) offsetof(MDB_page, mp_ptrs))
 
@@ -1026,17 +1026,19 @@ typedef struct MDB_xcursor {
 	unsigned char mx_dbflag;
 } MDB_xcursor;
 
-	/** Check if there is an inited xcursor, so #XCURSOR_REFRESH() is proper */
+	/** Check if there is an inited xcursor */
 #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
+	/** Update the xcursor's sub-page pointer, if any, in \b mc.  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].
+	 *	with leaf page \b mp = mc->mc_pg[\b top].
 	 */
-#define XCURSOR_REFRESH(mc, mp, ki) do { \
+#define XCURSOR_REFRESH(mc, top, mp) do { \
 	MDB_page *xr_pg = (mp); \
-	MDB_node *xr_node = NODEPTR(xr_pg, ki); \
+	MDB_node *xr_node; \
+	if (!XCURSOR_INITED(mc) || (mc)->mc_ki[top] >= NUMKEYS(xr_pg)) break; \
+	xr_node = NODEPTR(xr_pg, (mc)->mc_ki[top]); \
 	if ((xr_node->mn_flags & (F_DUPDATA|F_SUBDATA)) == F_DUPDATA) \
 		(mc)->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(xr_node); \
 } while (0)
@@ -1048,7 +1050,7 @@ typedef struct MDB_pgstate {
 } MDB_pgstate;
 
 	/** Context for deferred cleanup of reader's threads.
-	 *  to avoid https://github.com/ReOpen/ReOpenLDAP/issues/48 */
+	 *  to avoid https://github.com/leo-yuriev/ReOpenLDAP/issues/48 */
 typedef struct MDBX_rthc {
 	struct MDBX_rthc *rc_next;
 	pthread_t rc_thread;
@@ -1304,9 +1306,6 @@ static void mdb_debug_log(int type, const char *function, int line, const char *
 #	define mdb_assert_enabled() \
 		unlikely(mdb_runtime_flags & MDBX_DBG_ASSERT)
 
-#	define mdb_audit_enabled() \
-		unlikely(mdb_runtime_flags & MDBX_DBG_AUDIT)
-
 #	define mdb_debug_enabled(type) \
 		unlikely(mdb_runtime_flags & \
 			(type & (MDBX_DBG_TRACE | MDBX_DBG_EXTRA)))
@@ -1317,7 +1316,6 @@ static void mdb_debug_log(int type, const char *function, int line, const char *
 #	else
 #		define mdb_debug_enabled(type) (0)
 #	endif
-#	define mdb_audit_enabled() (0)
 #	define mdb_assert_enabled() (0)
 #	define mdb_assert_fail(env, msg, func, line) \
 		__assert_fail(msg, __FILE__, line, func)
@@ -1427,161 +1425,6 @@ mdb_dkey(MDB_val *key, char *buf)
 	return buf;
 }
 
-#if 0 /* LY: debug stuff */
-static const char *
-mdb_leafnode_type(MDB_node *n)
-{
-	static char *const tp[2][2] = {{"", ": DB"}, {": sub-page", ": sub-DB"}};
-	return F_ISSET(n->mn_flags, F_BIGDATA) ? ": overflow page" :
-		tp[F_ISSET(n->mn_flags, F_DUPDATA)][F_ISSET(n->mn_flags, F_SUBDATA)];
-}
-
-/** Display all the keys in the page. */
-static void
-mdb_page_list(MDB_page *mp)
-{
-	pgno_t pgno = mdb_dbg_pgno(mp);
-	const char *type, *state = (mp->mp_flags & P_DIRTY) ? ", dirty" : "";
-	MDB_node *node;
-	unsigned i, nkeys, nsize, total = 0;
-	MDB_val key;
-	DKBUF;
-
-	switch (mp->mp_flags & (P_BRANCH|P_LEAF|P_LEAF2|P_META|P_OVERFLOW|P_SUBP)) {
-	case P_BRANCH:              type = "Branch page";		break;
-	case P_LEAF:                type = "Leaf page";			break;
-	case P_LEAF|P_SUBP:         type = "Sub-page";			break;
-	case P_LEAF|P_LEAF2:        type = "LEAF2 page";		break;
-	case P_LEAF|P_LEAF2|P_SUBP: type = "LEAF2 sub-page";	break;
-	case P_OVERFLOW:
-		mdb_print("Overflow page %zu pages %u%s\n",
-			pgno, mp->mp_pages, state);
-		return;
-	case P_META:
-		mdb_print("Meta-page %zu txnid %zu\n",
-			pgno, ((MDB_meta *)PAGEDATA(mp))->mm_txnid);
-		return;
-	default:
-		mdb_print("Bad page %zu flags 0x%X\n", pgno, mp->mp_flags);
-		return;
-	}
-
-	nkeys = NUMKEYS(mp);
-	mdb_print("%s %zu numkeys %u%s\n", type, pgno, nkeys, state);
-
-	for (i=0; i<nkeys; i++) {
-		if (IS_LEAF2(mp)) {	/* LEAF2 pages have no mp_ptrs[] or node headers */
-			key.mv_size = nsize = mp->mp_leaf2_ksize;
-			key.mv_data = LEAF2KEY(mp, i, nsize);
-			total += nsize;
-			mdb_print("key %u: nsize %u, %s\n", i, nsize, DKEY(&key));
-			continue;
-		}
-		node = NODEPTR(mp, i);
-		key.mv_size = node->mn_ksize;
-		key.mv_data = node->mn_data;
-		nsize = NODESIZE + key.mv_size;
-		if (IS_BRANCH(mp)) {
-			mdb_print("key %u: page %zu, %s\n", i, NODEPGNO(node), DKEY(&key));
-			total += nsize;
-		} else {
-			if (F_ISSET(node->mn_flags, F_BIGDATA))
-				nsize += sizeof(pgno_t);
-			else
-				nsize += NODEDSZ(node);
-			total += nsize;
-			nsize += sizeof(indx_t);
-			mdb_print("key %u: nsize %u, %s%s\n",
-				i, nsize, DKEY(&key), mdb_leafnode_type(node));
-		}
-		total = EVEN(total);
-	}
-	mdb_print("Total: header %u + contents %u + unused %u\n",
-		IS_LEAF2(mp) ? PAGEHDRSZ : PAGEBASE + mp->mp_lower, total, SIZELEFT(mp));
-}
-
-static void
-mdb_cursor_chk(MDB_cursor *mc)
-{
-	unsigned i;
-	MDB_node *node;
-	MDB_page *mp;
-
-	if (!mc->mc_snum || !(mc->mc_flags & C_INITIALIZED)) return;
-	for (i=0; i<mc->mc_top; i++) {
-		mp = mc->mc_pg[i];
-		node = NODEPTR(mp, mc->mc_ki[i]);
-		if (unlikely(NODEPGNO(node) != mc->mc_pg[i+1]->mp_pgno))
-			mdb_print("oops!\n");
-	}
-	if (unlikely(mc->mc_ki[i] >= NUMKEYS(mc->mc_pg[i])))
-		mdb_print("ack!\n");
-	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)) {
-			mdb_print("blah!\n");
-		}
-	}
-}
-#endif /* 0 */
-
-/** Count all the pages in each DB and in the freelist
- *  and make sure it matches the actual number of pages
- *  being used.
- *  All named DBs must be open for a correct count.
- */
-static void mdb_audit(MDB_txn *txn)
-{
-	MDB_cursor mc;
-	MDB_val key, data;
-	MDB_ID freecount, count;
-	MDB_dbi i;
-	int rc;
-
-	freecount = 0;
-	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;
-	mdb_tassert(txn, rc == MDB_NOTFOUND);
-
-	count = 0;
-	for (i = 0; i<txn->mt_numdbs; i++) {
-		MDB_xcursor mx;
-		if (!(txn->mt_dbflags[i] & DB_VALID))
-			continue;
-		mdb_cursor_init(&mc, txn, i, &mx);
-		if (txn->mt_dbs[i].md_root == P_INVALID)
-			continue;
-		count += txn->mt_dbs[i].md_branch_pages +
-			txn->mt_dbs[i].md_leaf_pages +
-			txn->mt_dbs[i].md_overflow_pages;
-		if (txn->mt_dbs[i].md_flags & MDB_DUPSORT) {
-			rc = mdb_page_search(&mc, NULL, MDB_PS_FIRST);
-			for (; rc == MDB_SUCCESS; rc = mdb_cursor_sibling(&mc, 1)) {
-				unsigned j;
-				MDB_page *mp;
-				mp = mc.mc_pg[mc.mc_top];
-				for (j=0; j<NUMKEYS(mp); j++) {
-					MDB_node *leaf = NODEPTR(mp, j);
-					if (leaf->mn_flags & F_SUBDATA) {
-						MDB_db db;
-						memcpy(&db, NODEDATA(leaf), sizeof(db));
-						count += db.md_branch_pages + db.md_leaf_pages +
-							db.md_overflow_pages;
-					}
-				}
-			}
-			mdb_tassert(txn, rc == MDB_NOTFOUND);
-		}
-	}
-	if (freecount + count + NUM_METAS != txn->mt_next_pgno) {
-		mdb_print("audit: %lu freecount: %lu count: %lu total: %lu next_pgno: %lu\n",
-			txn->mt_txnid, freecount, count+NUM_METAS,
-			freecount+count+NUM_METAS, txn->mt_next_pgno);
-	}
-}
-
 int
 mdb_cmp(MDB_txn *txn, MDB_dbi dbi, const MDB_val *a, const MDB_val *b)
 {
@@ -2138,6 +1981,10 @@ mdb_page_alloc(MDB_cursor *mc, int num, MDB_page **mp, int flags)
 			/* If mc is updating the freeDB, then the freelist cannot play
 			 * catch-up with itself by growing while trying to save it. */
 			flags &= ~(MDBX_ALLOC_GC | MDBX_ALLOC_KICK | MDBX_COALESCE | MDBX_LIFORECLAIM);
+		} else if (unlikely(txn->mt_dbs[FREE_DBI].md_entries == 0)) {
+			/* avoid (recursive) search inside empty tree and while tree is updating,
+			 * https://github.com/leo-yuriev/libmdbx/issues/31 */
+			flags &= ~MDBX_ALLOC_GC;
 		}
 	}
 
@@ -2608,8 +2455,8 @@ done:
 			if (m2 == mc) continue;
 			if (m2->mc_pg[mc->mc_top] == mp) {
 				m2->mc_pg[mc->mc_top] = np;
-				if (XCURSOR_INITED(m2) && IS_LEAF(np))
-					XCURSOR_REFRESH(m2, np, m2->mc_ki[mc->mc_top]);
+				if (IS_LEAF(np))
+					XCURSOR_REFRESH(m2, mc->mc_top, np);
 			}
 		}
 	}
@@ -3324,6 +3171,7 @@ mdb_prep_backlog(MDB_txn *txn, MDB_cursor *mc)
 	const int extra = (txn->mt_env->me_flags & MDBX_LIFORECLAIM) ? 2 : 1;
 
 	if (mdb_backlog_size(txn) < mc->mc_db->md_depth + extra) {
+		mc->mc_flags &= ~C_RECLAIMING;
 		int rc = mdb_cursor_touch(mc);
 		if (unlikely(rc))
 			return rc;
@@ -3336,6 +3184,7 @@ mdb_prep_backlog(MDB_txn *txn, MDB_cursor *mc)
 				break;
 			}
 		}
+		mc->  mc_flags |= C_RECLAIMING;
 	}
 
 	return MDB_SUCCESS;
@@ -3360,6 +3209,9 @@ mdb_freelist_save(MDB_txn *txn)
 	const int lifo = (env->me_flags & MDBX_LIFORECLAIM) != 0;
 
 	mdb_cursor_init(&mc, txn, FREE_DBI, NULL);
+	mc.mc_flags |= C_RECLAIMING;
+	mc.mc_next = txn->mt_cursors[FREE_DBI];
+	txn->mt_cursors[FREE_DBI] = &mc;
 
 	/* MDB_RESERVE cancels meminit in ovpage malloc (when no WRITEMAP) */
 	clean_limit = (env->me_flags & (MDB_NOMEMINIT|MDB_WRITEMAP))
@@ -3386,9 +3238,7 @@ again:
 				total_room = head_room = 0;
 				more = 1;
 				mdb_tassert(txn, pglast <= env->me_pglast);
-				mc.mc_flags |= C_RECLAIMING;
 				rc = mdb_cursor_del(&mc, 0);
-				mc.mc_flags &= ~C_RECLAIMING;
 				if (unlikely(rc))
 					goto bailout;
 			}
@@ -3405,9 +3255,7 @@ again:
 					rc = mdb_prep_backlog(txn, &mc);
 					if (unlikely(rc))
 						goto bailout;
-					mc.mc_flags |= C_RECLAIMING;
 					rc = mdb_cursor_del(&mc, 0);
-					mc.mc_flags &= ~C_RECLAIMING;
 					if (unlikely(rc))
 						goto bailout;
 				}
@@ -3430,7 +3278,9 @@ again:
 		if (freecnt < txn->mt_free_pgs[0]) {
 			if (unlikely(!freecnt)) {
 				/* Make sure last page of freeDB is touched and on freelist */
+				mc.mc_flags &= ~C_RECLAIMING;
 				rc = mdb_page_search(&mc, NULL, MDB_PS_LAST|MDB_PS_MODIFY);
+				mc.mc_flags |= C_RECLAIMING;
 				if (unlikely(rc && rc != MDB_NOTFOUND))
 					goto bailout;
 			}
@@ -3484,7 +3334,9 @@ again:
 		if (lifo) {
 			if (refill_idx > (txn->mt_lifo_reclaimed ? txn->mt_lifo_reclaimed[0] : 0)) {
 				/* LY: need just a txn-id for save page list. */
+				mc.mc_flags &= ~C_RECLAIMING;
 				rc = mdb_page_alloc(&mc, 0, NULL, MDBX_ALLOC_GC | MDBX_ALLOC_KICK);
+				mc.mc_flags |= C_RECLAIMING;
 				if (likely(rc == 0))
 					/* LY: ok, reclaimed from freedb. */
 					continue;
@@ -3641,6 +3493,7 @@ bailout:
 			txn->mt_lifo_reclaimed = NULL;
 		}
 	}
+	txn->mt_cursors[FREE_DBI] = mc.mc_next;
 
 	return rc;
 }
@@ -3986,9 +3839,6 @@ mdb_txn_commit(MDB_txn *txn)
 	env->me_pghead = NULL;
 	mdb_midl_shrink(&txn->mt_free_pgs);
 
-	if (mdb_audit_enabled())
-		mdb_audit(txn);
-
 	rc = mdb_page_flush(txn, 0);
 	if (likely(rc == MDB_SUCCESS)) {
 		MDB_meta meta;
@@ -4341,7 +4191,7 @@ mdb_env_create(MDB_env **env)
 }
 
 static int __cold
-mdb_env_map(MDB_env *env, void *addr, size_t usedsize)
+mdb_env_map(MDB_env *env, void *addr)
 {
 	unsigned flags = env->me_flags;
 
@@ -4381,12 +4231,6 @@ mdb_env_map(MDB_env *env, void *addr, size_t usedsize)
 	}
 #endif
 
-#ifdef MADV_REMOVE
-	if (flags & MDB_WRITEMAP) {
-		(void) madvise(env->me_map + usedsize, env->me_mapsize - usedsize, MADV_REMOVE);
-	}
-#endif
-
 	/* Turn on/off readahead. It's harmful when the DB is larger than RAM. */
 	if (madvise(env->me_map, env->me_mapsize, (flags & MDB_NORDAHEAD) ? MADV_RANDOM : MADV_WILLNEED))
 		return errno;
@@ -4439,7 +4283,7 @@ mdb_env_set_mapsize(MDB_env *env, size_t size)
 #endif
 		env->me_mapsize = size;
 		old = (env->me_flags & MDB_FIXEDMAP) ? env->me_map : NULL;
-		rc = mdb_env_map(env, old, usedsize);
+		rc = mdb_env_map(env, old);
 		if (rc)
 			return rc;
 	}
@@ -4557,8 +4401,7 @@ mdb_env_open2(MDB_env *env, MDB_meta *meta)
 		newenv = 0;
 	}
 
-	const size_t usedsize = (meta->mm_last_pg + 1) * env->me_psize;
-	rc = mdb_env_map(env, (flags & MDB_FIXEDMAP) ? meta->mm_address : NULL, usedsize);
+	rc = mdb_env_map(env, (flags & MDB_FIXEDMAP) ? meta->mm_address : NULL);
 	if (rc)
 		return rc;
 
@@ -4631,7 +4474,7 @@ void mdbx_rthc_dtor(void)
 	 *    TSD-деструкторах и поэтому может выгрузить lib.so до того как
 	 *    отработали все деструкторы.
 	 *  - Исходное проявление проблемы было зафиксировано
-	 *    в https://github.com/ReOpen/ReOpenLDAP/issues/48
+	 *    в https://github.com/leo-yuriev/ReOpenLDAP/issues/48
 	 *
 	 * Предыдущее решение посредством выделяемого динамически MDB_rthc
 	 * было не удачным, так как порождало либо утечку памяти,
@@ -6680,6 +6523,11 @@ fetchm:
 			rc = MDB_INCOMPATIBLE;
 			break;
 		}
+		if (mc->mc_ki[mc->mc_top] >= NUMKEYS(mc->mc_pg[mc->mc_top])) {
+			mc->mc_ki[mc->mc_top] = NUMKEYS(mc->mc_pg[mc->mc_top]);
+			rc = MDB_NOTFOUND;
+			break;
+		}
 		{
 			MDB_node *leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
 			if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
@@ -7030,7 +6878,8 @@ more:
 						offset *= 4; /* space for 4 more */
 						break;
 					}
-					/* FALLTHRU: Big enough MDB_DUPFIXED sub-page */
+					/* Big enough MDB_DUPFIXED sub-page */
+					/* fallthrough */
 				case MDB_CURRENT | MDB_NODUPDATA:
 				case MDB_CURRENT:
 					fp->mp_flags |= P_DIRTY;
@@ -7081,8 +6930,9 @@ prep_subDB:
 				} else {
 					memcpy((char *)mp + mp->mp_upper + PAGEBASE, (char *)fp + fp->mp_upper + PAGEBASE,
 						olddata.mv_size - fp->mp_upper - PAGEBASE);
+					memcpy((char *)(&mp->mp_ptrs), (char *)(&fp->mp_ptrs), NUMKEYS(fp) * sizeof(mp->mp_ptrs[0]));
 					for (i=0; i<NUMKEYS(fp); i++)
-						mp->mp_ptrs[i] = fp->mp_ptrs[i] + offset;
+						mp->mp_ptrs[i] += offset;
 				}
 			}
 
@@ -7210,8 +7060,7 @@ new_sub:
 				if (m3->mc_ki[i] >= mc->mc_ki[i] && insert_key) {
 					m3->mc_ki[i]++;
 				}
-				if (XCURSOR_INITED(m3))
-					XCURSOR_REFRESH(m3, mp, m3->mc_ki[i]);
+				XCURSOR_REFRESH(m3, i, mp);
 			}
 		}
 	}
@@ -7252,7 +7101,6 @@ put_sub:
 				MDB_xcursor *mx = mc->mc_xcursor;
 				unsigned i = mc->mc_top;
 				MDB_page *mp = mc->mc_pg[i];
-				int nkeys = NUMKEYS(mp);
 
 				for (m2 = mc->mc_txn->mt_cursors[mc->mc_dbi]; m2; m2=m2->mc_next) {
 					if (m2 == mc || m2->mc_snum < mc->mc_snum) continue;
@@ -7260,8 +7108,8 @@ put_sub:
 					if (m2->mc_pg[i] == mp) {
 						if (m2->mc_ki[i] == mc->mc_ki[i]) {
 							mdb_xcursor_init2(m2, mx, dupdata_flag);
-						} else if (!insert_key && m2->mc_ki[i] < nkeys) {
-							XCURSOR_REFRESH(m2, mp, m2->mc_ki[i]);
+						} else if (!insert_key) {
+							XCURSOR_REFRESH(m2, i, mp);
 						}
 					}
 				}
@@ -7371,12 +7219,7 @@ mdb_cursor_del(MDB_cursor *mc, unsigned 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) {
-							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);
+							XCURSOR_REFRESH(m2, mc->mc_top, mp);
 						}
 					}
 				}
@@ -8169,6 +8012,12 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 
 	DKBUF;
 
+	mdb_tassert(csrc->mc_txn, PAGETYPE(csrc->mc_pg[csrc->mc_top]) == PAGETYPE(cdst->mc_pg[cdst->mc_top]));
+	if (unlikely(PAGETYPE(csrc->mc_pg[csrc->mc_top]) != PAGETYPE(cdst->mc_pg[cdst->mc_top]))) {
+		cdst->mc_txn->mt_flags |= MDB_TXN_ERROR;
+		return MDB_PROBLEM;
+	}
+
 	/* Mark src and dst as dirty. */
 	if (unlikely((rc = mdb_page_touch(csrc)) ||
 		(rc = mdb_page_touch(cdst))))
@@ -8280,8 +8129,8 @@ 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 (XCURSOR_INITED(m3) && IS_LEAF(mps))
-					XCURSOR_REFRESH(m3, m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]);
+				if (IS_LEAF(mps))
+					XCURSOR_REFRESH(m3, csrc->mc_top, m3->mc_pg[csrc->mc_top]);
 			}
 		} else
 		/* Adding on the right, bump others down */
@@ -8302,8 +8151,8 @@ mdb_node_move(MDB_cursor *csrc, MDB_cursor *cdst, int fromleft)
 					} else {
 						m3->mc_ki[csrc->mc_top]--;
 					}
-					if (XCURSOR_INITED(m3) && IS_LEAF(mps))
-						XCURSOR_REFRESH(m3, m3->mc_pg[csrc->mc_top], m3->mc_ki[csrc->mc_top]);
+					if (IS_LEAF(mps))
+						XCURSOR_REFRESH(m3, csrc->mc_top, m3->mc_pg[csrc->mc_top]);
 				}
 			}
 		}
@@ -8399,6 +8248,12 @@ mdb_page_merge(MDB_cursor *csrc, MDB_cursor *cdst)
 	mdb_cassert(csrc, csrc->mc_snum > 1);	/* can't merge root page */
 	mdb_cassert(csrc, cdst->mc_snum > 1);
 
+	mdb_tassert(csrc->mc_txn, PAGETYPE(psrc) == PAGETYPE(pdst));
+	if (unlikely(PAGETYPE(psrc) != PAGETYPE(pdst))) {
+		cdst->mc_txn->mt_flags |= MDB_TXN_ERROR;
+		return MDB_PROBLEM;
+	}
+
 	/* Mark dst as dirty. */
 	if (unlikely(rc = mdb_page_touch(cdst)))
 		return rc;
@@ -8501,8 +8356,8 @@ 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 (XCURSOR_INITED(m3) && IS_LEAF(psrc))
-				XCURSOR_REFRESH(m3, m3->mc_pg[top], m3->mc_ki[top]);
+			if (IS_LEAF(psrc))
+				XCURSOR_REFRESH(m3, top, m3->mc_pg[top]);
 		}
 	}
 	{
@@ -8602,7 +8457,7 @@ mdb_rebalance(MDB_cursor *mc)
 						m3 = &m2->mc_xcursor->mx_cursor;
 					else
 						m3 = m2;
-					if (!(m3->mc_flags & C_INITIALIZED) || (m3->mc_snum < mc->mc_snum))
+					if (m3 == mc || !(m3->mc_flags & C_INITIALIZED))
 						continue;
 					if (m3->mc_pg[0] == mp) {
 						m3->mc_snum = 0;
@@ -8764,8 +8619,7 @@ mdb_cursor_del0(MDB_cursor *mc)
 				} else if (m3->mc_ki[mc->mc_top] > ki) {
 					m3->mc_ki[mc->mc_top]--;
 				}
-				if (XCURSOR_INITED(m3))
-					XCURSOR_REFRESH(m3, m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]);
+				XCURSOR_REFRESH(m3, mc->mc_top, mp);
 			}
 		}
 	}
@@ -8812,8 +8666,10 @@ mdb_cursor_del0(MDB_cursor *mc)
 							if (m3->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED) {
 								if (!(node->mn_flags & F_SUBDATA))
 									m3->mc_xcursor->mx_cursor.mc_pg[0] = NODEDATA(node);
-							} else
+							} else {
 								mdb_xcursor_init1(m3, node);
+								m3->mc_xcursor->mx_cursor.mc_flags |= C_DEL;
+							}
 						}
 					}
 				}
@@ -9068,7 +8924,7 @@ mdb_page_split(MDB_cursor *mc, MDB_val *newkey, MDB_val *newdata, pgno_t newpgno
 			 * the split so the new page is emptier than the old page.
 			 * This yields better packing during sequential inserts.
 			 */
-			if (nkeys < 20 || nsize > pmax/16 || newindx >= nkeys) {
+			if (nkeys < 32 || nsize > pmax/16 || newindx >= nkeys) {
 				/* Find split point */
 				psize = 0;
 				if (newindx <= split_indx || newindx >= nkeys) {
@@ -9304,8 +9160,8 @@ 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 (XCURSOR_INITED(m3) && IS_LEAF(mp))
-				XCURSOR_REFRESH(m3, m3->mc_pg[mc->mc_top], m3->mc_ki[mc->mc_top]);
+			if (IS_LEAF(mp))
+				XCURSOR_REFRESH(m3, mc->mc_top, m3->mc_pg[mc->mc_top]);
 		}
 	}
 	mdb_debug("mp left: %d, rp left: %d", SIZELEFT(mp), SIZELEFT(rp));
@@ -10210,8 +10066,11 @@ int mdb_dbi_open(MDB_txn *txn, const char *name, unsigned flags, MDB_dbi *dbi)
 		MDB_node *node = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]);
 		if (unlikely((node->mn_flags & (F_DUPDATA|F_SUBDATA)) != F_SUBDATA))
 			return MDB_INCOMPATIBLE;
-	} else if (! (rc == MDB_NOTFOUND && (flags & MDB_CREATE))) {
-		return rc;
+	} else {
+		if (rc != MDB_NOTFOUND || !(flags & MDB_CREATE))
+			return rc;
+		if (F_ISSET(txn->mt_flags, MDB_TXN_RDONLY))
+			return EACCES;
 	}
 
 	/* Done here so we cannot fail after creating a new DB */

mdb_dump.c

@@ -186,7 +186,7 @@ int main(int argc, char *argv[])
 			break;
 		case 'l':
 			list = 1;
-			/*FALLTHROUGH*/;
+			/* fallthrough */
 		case 'a':
 			if (subname)
 				usage(prog);

mdbx.c

@@ -530,7 +530,7 @@ int mdbx_replace(MDB_txn *txn, MDB_dbi dbi,
 		rc = mdbx_cursor_get(&mc, &present_key, &present_data, MDB_SET_KEY);
 		if (unlikely(rc != MDB_SUCCESS)) {
 			old_data->iov_base = NULL;
-			old_data->iov_len = rc;
+			old_data->iov_len = 0;
 			if (rc != MDB_NOTFOUND || (flags & MDB_CURRENT))
 				goto bailout;
 		} else if (flags & MDB_NOOVERWRITE) {
@@ -864,13 +864,14 @@ int mdbx_set_attr(MDB_txn *txn, MDB_dbi dbi,
 		return rc;
 	}
 
+	old_attr = 0;
 	rc = mdbx_attr_peek(&old_data, &old_attr);
 	if (unlikely(rc != MDB_SUCCESS))
 		return rc;
 
 	if (old_attr == attr && (!data ||
 			(data->mv_size == old_data.mv_size
-				&& memcpy(data->mv_data, old_data.mv_data, old_data.mv_size) == 0)))
+				&& memcmp(data->mv_data, old_data.mv_data, old_data.mv_size) == 0)))
 		return MDB_SUCCESS;
 
 	mc.mc_next = txn->mt_cursors[dbi];

mtest0.c

@@ -21,6 +21,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #include <pthread.h>

mtest1.c

@@ -20,6 +20,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest2.c

@@ -23,6 +23,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest3.c

@@ -23,6 +23,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest4.c

@@ -23,6 +23,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest5.c

@@ -23,6 +23,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest6.c

@@ -23,6 +23,7 @@
 #include <unistd.h>
 #include <errno.h>
 #include <sys/stat.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest7.c

@@ -18,6 +18,7 @@
 #include <string.h>
 #include <time.h>
 #include <sys/time.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)

mtest8.c

@@ -18,6 +18,7 @@
 #include <string.h>
 #include <time.h>
 #include <sys/time.h>
+#include <sys/sysmacros.h>
 #include "mdbx.h"
 
 #define E(expr) CHECK((rc = (expr)) == MDB_SUCCESS, #expr)