From 17c6555a7fef4eb8711e826bf3401b8b9b6e93cf Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Mon, 21 Nov 2016 20:50:39 +0300
Subject: [PATCH 01/13] mdbx: add 'canary' support for libfpta.

Change-Id: I62c68f149adf38d65aa9371a1fb3adac405d23ed
---
 mdb.c  | 22 +++++++++++++++++++++-
 mdbx.c | 32 ++++++++++++++++++++++++++++++++
 mdbx.h |  8 ++++++++
 3 files changed, 61 insertions(+), 1 deletion(-)

diff --git a/mdb.c b/mdb.c
index edf088ec..45fcc5e6 100644
--- a/mdb.c
+++ b/mdb.c
@@ -778,6 +778,10 @@ typedef struct MDB_meta {
 	volatile uint64_t	mm_datasync_sign;
 #define META_IS_WEAK(meta) ((meta)->mm_datasync_sign == MDB_DATASIGN_WEAK)
 #define META_IS_STEADY(meta) ((meta)->mm_datasync_sign > MDB_DATASIGN_WEAK)
+
+#if MDBX_MODE_ENABLED
+	volatile mdbx_canary mm_canary;
+#endif
 } MDB_meta;
 
 	/** Buffer for a stack-allocated meta page.
@@ -809,7 +813,7 @@ typedef struct MDB_dbx {
 	 *	Every operation requires a transaction handle.
 	 */
 struct MDB_txn {
-#define MDBX_MT_SIGNATURE 0x706C553B
+#define MDBX_MT_SIGNATURE 0x93D53A31
 	unsigned	mt_signature;
 	MDB_txn		*mt_parent;		/**< parent of a nested txn */
 	/** Nested txn under this txn, set together with flag #MDB_TXN_HAS_CHILD */
@@ -895,6 +899,10 @@ struct MDB_txn {
 	 *	dirty_list into mt_parent after freeing hidden mt_parent pages.
 	 */
 	unsigned	mt_dirty_room;
+
+#if MDBX_MODE_ENABLED
+	mdbx_canary	mt_canary;
+#endif
 };
 
 /** Enough space for 2^32 nodes with minimum of 2 keys per node. I.e., plenty.
@@ -2842,6 +2850,9 @@ mdb_txn_renew0(MDB_txn *txn, unsigned flags)
 				txn->mt_next_pgno = meta->mm_last_pg+1;
 				/* Copy the DB info and flags */
 				memcpy(txn->mt_dbs, meta->mm_dbs, CORE_DBS * sizeof(MDB_db));
+#if MDBX_MODE_ENABLED
+				txn->mt_canary = meta->mm_canary;
+#endif
 				break;
 			}
 		}
@@ -2858,6 +2869,9 @@ mdb_txn_renew0(MDB_txn *txn, unsigned flags)
 		pthread_mutex_lock(&tsan_mutex);
 #endif
 		MDB_meta *meta = mdb_meta_head_w(env);
+#if MDBX_MODE_ENABLED
+		txn->mt_canary = meta->mm_canary;
+#endif
 		txn->mt_txnid = meta->mm_txnid + 1;
 		txn->mt_flags = flags;
 #ifdef __SANITIZE_THREAD__
@@ -3919,6 +3933,9 @@ mdb_txn_commit(MDB_txn *txn)
 		meta.mm_dbs[MAIN_DBI] = txn->mt_dbs[MAIN_DBI];
 		meta.mm_last_pg = txn->mt_next_pgno - 1;
 		meta.mm_txnid = txn->mt_txnid;
+#if MDBX_MODE_ENABLED
+		meta.mm_canary = txn->mt_canary;
+#endif
 
 		rc = mdb_env_sync0(env, env->me_flags | txn->mt_flags, &meta);
 	}
@@ -4155,6 +4172,9 @@ mdb_env_sync0(MDB_env *env, unsigned flags, MDB_meta *pending)
 		target->mm_dbs[FREE_DBI] = pending->mm_dbs[FREE_DBI];
 		target->mm_dbs[MAIN_DBI] = pending->mm_dbs[MAIN_DBI];
 		target->mm_last_pg = pending->mm_last_pg;
+#if MDBX_MODE_ENABLED
+		target->mm_canary = pending->mm_canary;
+#endif
 		/* LY: 'commit' the meta */
 		target->mm_txnid = pending->mm_txnid;
 		target->mm_datasync_sign = pending->mm_datasync_sign;
diff --git a/mdbx.c b/mdbx.c
index fb4aac85..cf5187fe 100644
--- a/mdbx.c
+++ b/mdbx.c
@@ -320,3 +320,35 @@ mdbx_env_pgwalk(MDB_txn *txn, MDBX_pgvisitor_func* visitor, void* user)
 		rc = visitor(P_INVALID, 0, user, NULL, NULL, 0, 0, 0, 0);
 	return rc;
 }
+
+int mdbx_canary_put(MDB_txn *txn, const mdbx_canary* canary)
+{
+	if (unlikely(!txn))
+		return EINVAL;
+
+	if (unlikely(txn->mt_signature != MDBX_MT_SIGNATURE))
+		return MDB_VERSION_MISMATCH;
+
+	if (unlikely(F_ISSET(txn->mt_flags, MDB_TXN_RDONLY)))
+		return EACCES;
+
+	if (likely(canary)) {
+		txn->mt_canary.x = canary->x;
+		txn->mt_canary.y = canary->y;
+		txn->mt_canary.z = canary->z;
+	}
+	txn->mt_canary.v = txn->mt_txnid;
+
+	return MDB_SUCCESS;
+}
+
+size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary)
+{
+	if(unlikely(!txn || txn->mt_signature != MDBX_MT_SIGNATURE))
+		return 0;
+
+	if (likely(canary))
+		*canary = txn->mt_canary;
+
+	return txn->mt_txnid;
+}
diff --git a/mdbx.h b/mdbx.h
index bcbd4f47..d494ef36 100644
--- a/mdbx.h
+++ b/mdbx.h
@@ -211,6 +211,14 @@ typedef int MDBX_pgvisitor_func(size_t pgno, unsigned pgnumber, void* ctx,
 					const char* dbi, const char *type, int nentries,
 					int payload_bytes, int header_bytes, int unused_bytes);
 int mdbx_env_pgwalk(MDB_txn *txn, MDBX_pgvisitor_func* visitor, void* ctx);
+
+typedef struct mdbx_canary {
+	size_t x, y, z, v;
+} mdbx_canary;
+
+int mdbx_canary_put(MDB_txn *txn, const mdbx_canary* canary);
+size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary);
+
 /**	@} */
 
 #ifdef __cplusplus

From f9f132671cc245cfe39519e5dbc9d5044b8bd973 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Tue, 6 Dec 2016 20:08:08 +0300
Subject: [PATCH 02/13] mdbx: explicit overwrite support for mdbx_put().

---
 lmdb.h |  3 ++-
 mdb.c  | 21 ++++++++++++++++++---
 2 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/lmdb.h b/lmdb.h
index 2d2aabb7..1237fdcd 100644
--- a/lmdb.h
+++ b/lmdb.h
@@ -349,7 +349,8 @@ typedef void (MDB_rel_func)(MDB_val *item, void *oldptr, void *newptr, void *rel
  * For mdb_cursor_del: remove all duplicate data items.
  */
 #define MDB_NODUPDATA	0x20
-/** For mdb_cursor_put: overwrite the current key/data pair */
+/** For mdb_cursor_put: overwrite the current key/data pair
+ *  MDBX allows this flag for mdb_put() for explicit overwrite/update without insertion. */
 #define MDB_CURRENT	0x40
 /** For put: Just reserve space for data, don't copy it. Return a
  * pointer to the reserved space.
diff --git a/mdb.c b/mdb.c
index 5e1dcffc..21933378 100644
--- a/mdb.c
+++ b/mdb.c
@@ -9032,7 +9032,6 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi,
 {
 	MDB_cursor mc;
 	MDB_xcursor mx;
-	int rc;
 
 	if (unlikely(!key || !data || !txn))
 		return EINVAL;
@@ -9043,17 +9042,33 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi,
 	if (unlikely(!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)))
 		return EINVAL;
 
-	if (unlikely(flags & ~(MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP)))
+	if (unlikely(flags & ~(MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP
+			/* LY: MDB_CURRENT indicates explicit overwrite (update) for MDBX */
+			| (MDBX_MODE_ENABLED ? MDB_CURRENT : 0))))
 		return EINVAL;
 
 	if (unlikely(txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)))
 		return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
 
+#if MDBX_MODE_ENABLED
+	/* LY: allows update (explicit overwrite) only for unique keys */
+	if ((flags & MDB_CURRENT) && (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT))
+		return EINVAL;
+#endif /* MDBX_MODE_ENABLED */
+
 	mdb_cursor_init(&mc, txn, dbi, &mx);
 	mc.mc_next = txn->mt_cursors[dbi];
 	txn->mt_cursors[dbi] = &mc;
-	rc = mdb_cursor_put(&mc, key, data, flags);
+	int rc = MDB_SUCCESS;
+#if MDBX_MODE_ENABLED
+	/* LY: support for update (explicit overwrite) */
+	if (flags & MDB_CURRENT)
+		rc = mdb_cursor_get(&mc, key, NULL, MDB_SET);
+#endif /* MDBX_MODE_ENABLED */
+	if (likely(rc == MDB_SUCCESS))
+		rc = mdb_cursor_put(&mc, key, data, flags);
 	txn->mt_cursors[dbi] = mc.mc_next;
+
 	return rc;
 }
 

From b980878b0727396fa9c9812c1d879059743a5288 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Tue, 6 Dec 2016 20:32:09 +0300
Subject: [PATCH 03/13] mdbx: adds mdbx_cursor_eof() for libfpta.

---
 mdbx.c | 11 +++++++++++
 mdbx.h |  4 ++++
 2 files changed, 15 insertions(+)

diff --git a/mdbx.c b/mdbx.c
index cf5187fe..d7001778 100644
--- a/mdbx.c
+++ b/mdbx.c
@@ -352,3 +352,14 @@ size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary)
 
 	return txn->mt_txnid;
 }
+
+int mdbx_cursor_eof(MDB_cursor *mc)
+{
+	if (unlikely(mc == NULL))
+		return EINVAL;
+
+	if (unlikely(mc->mc_signature != MDBX_MC_SIGNATURE))
+		return MDB_VERSION_MISMATCH;
+
+	return (mc->mc_flags & (C_INITIALIZED | C_EOF)) != C_INITIALIZED ? 1 : 0;
+}
diff --git a/mdbx.h b/mdbx.h
index d494ef36..8a5c238e 100644
--- a/mdbx.h
+++ b/mdbx.h
@@ -219,6 +219,10 @@ typedef struct mdbx_canary {
 int mdbx_canary_put(MDB_txn *txn, const mdbx_canary* canary);
 size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary);
 
+/** Returns 1 when no more data available or cursor not positioned,
+ * 0 otherwise or less that zero in error case. */
+int mdbx_cursor_eof(MDB_cursor *mc);
+
 /**	@} */
 
 #ifdef __cplusplus

From c4f4d9ebf3f9690ff85151f66f2497b69b3da00a Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Thu, 8 Dec 2016 16:33:17 +0300
Subject: [PATCH 04/13] mdbx: rethink mdbx_cursor_eof() for libfpta.

---
 mdbx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mdbx.c b/mdbx.c
index d7001778..64c57683 100644
--- a/mdbx.c
+++ b/mdbx.c
@@ -361,5 +361,5 @@ int mdbx_cursor_eof(MDB_cursor *mc)
 	if (unlikely(mc->mc_signature != MDBX_MC_SIGNATURE))
 		return MDB_VERSION_MISMATCH;
 
-	return (mc->mc_flags & (C_INITIALIZED | C_EOF)) != C_INITIALIZED ? 1 : 0;
+	return (mc->mc_flags & C_INITIALIZED) ? 0 : 1;
 }

From 5865c74876578f55cf6bee9f94b2782b36655c49 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Thu, 15 Dec 2016 21:55:28 +0300
Subject: [PATCH 05/13] mdbx: rework overwrite support for mdbx_put().

---
 mdb.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)

diff --git a/mdb.c b/mdb.c
index ab72b9e6..16a24bbd 100644
--- a/mdb.c
+++ b/mdb.c
@@ -9042,20 +9042,23 @@ mdb_put(MDB_txn *txn, MDB_dbi dbi,
 	if (unlikely(txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)))
 		return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
 
-#if MDBX_MODE_ENABLED
-	/* LY: allows update (explicit overwrite) only for unique keys */
-	if ((flags & MDB_CURRENT) && (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT))
-		return EINVAL;
-#endif /* MDBX_MODE_ENABLED */
-
 	mdb_cursor_init(&mc, txn, dbi, &mx);
 	mc.mc_next = txn->mt_cursors[dbi];
 	txn->mt_cursors[dbi] = &mc;
 	int rc = MDB_SUCCESS;
 #if MDBX_MODE_ENABLED
 	/* LY: support for update (explicit overwrite) */
-	if (flags & MDB_CURRENT)
+	if (flags & MDB_CURRENT) {
 		rc = mdb_cursor_get(&mc, key, NULL, MDB_SET);
+		if (likely(rc == MDB_SUCCESS) && (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)) {
+			/* LY: allows update (explicit overwrite) only for unique keys */
+			MDB_node *leaf = NODEPTR(mc.mc_pg[mc.mc_top], mc.mc_ki[mc.mc_top]);
+			if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+				mdb_tassert(txn, XCURSOR_INITED(&mc) && mc.mc_xcursor->mx_db.md_entries > 1);
+				rc = MDB_KEYEXIST;
+			}
+		}
+	}
 #endif /* MDBX_MODE_ENABLED */
 	if (likely(rc == MDB_SUCCESS))
 		rc = mdb_cursor_put(&mc, key, data, flags);

From 2956095c6ded72d22e10e6d1cad4a5410ea52994 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Thu, 15 Dec 2016 21:56:45 +0300
Subject: [PATCH 06/13] mdbx: rework MDB_CURRENT handling for libfpta.

---
 mdb.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/mdb.c b/mdb.c
index 447a618d..411f5ca2 100644
--- a/mdb.c
+++ b/mdb.c
@@ -6185,7 +6185,6 @@ set1:
 				rc = 0;
 			}
 			*data = olddata;
-
 		} else {
 			if (mc->mc_xcursor)
 				mc->mc_xcursor->mx_cursor.mc_flags &= ~(C_INITIALIZED|C_EOF);
@@ -6587,7 +6586,7 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 
 	dkey.mv_size = 0;
 
-	if (flags == MDB_CURRENT) {
+	if (flags & MDB_CURRENT) {
 		if (unlikely(!(mc->mc_flags & C_INITIALIZED)))
 			return EINVAL;
 		rc = MDB_SUCCESS;
@@ -6778,6 +6777,7 @@ more:
 						break;
 					}
 					/* FALLTHRU: Big enough MDB_DUPFIXED sub-page */
+				case MDB_CURRENT | MDB_NODUPDATA:
 				case MDB_CURRENT:
 					fp->mp_flags |= P_DIRTY;
 					COPY_PGNO(fp->mp_pgno, mp->mp_pgno);
@@ -6975,12 +6975,15 @@ put_sub:
 			xdata.mv_size = 0;
 			xdata.mv_data = "";
 			leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
+			xflags = MDB_NOSPILL;
+			if (flags & MDB_NODUPDATA)
+				xflags |= MDB_NOOVERWRITE;
+			if (flags & MDB_APPENDDUP)
+				xflags |= MDB_APPEND;
 			if (flags & MDB_CURRENT) {
-				xflags = MDB_CURRENT|MDB_NOSPILL;
+				xflags |= MDB_CURRENT;
 			} else {
 				mdb_xcursor_init1(mc, leaf);
-				xflags = (flags & MDB_NODUPDATA) ?
-					MDB_NOOVERWRITE|MDB_NOSPILL : MDB_NOSPILL;
 			}
 			if (sub_root)
 				mc->mc_xcursor->mx_cursor.mc_pg[0] = sub_root;
@@ -7014,8 +7017,6 @@ put_sub:
 				}
 			}
 			ecount = mc->mc_xcursor->mx_db.md_entries;
-			if (flags & MDB_APPENDDUP)
-				xflags |= MDB_APPEND;
 			rc = mdb_cursor_put(&mc->mc_xcursor->mx_cursor, data, &xdata, xflags);
 			if (flags & F_SUBDATA) {
 				void *db = NODEDATA(leaf);

From 578fe9e2586cb7dbaa38a79ab1943388581782d8 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Sun, 18 Dec 2016 00:58:26 +0300
Subject: [PATCH 07/13] mdbx: allows zero-length keys for libfpta.

---
 mdb.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/mdb.c b/mdb.c
index 411f5ca2..83cd994c 100644
--- a/mdb.c
+++ b/mdb.c
@@ -6011,9 +6011,6 @@ mdb_cursor_set(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 	MDB_node	*leaf = NULL;
 	DKBUF;
 
-	if (unlikely(key->mv_size == 0))
-		return MDB_BAD_VALSIZE;
-
 	if ( (mc->mc_db->md_flags & MDB_INTEGERKEY)
 		&& unlikely( key->mv_size != sizeof(unsigned)
 		&& key->mv_size != sizeof(size_t) )) {
@@ -6556,7 +6553,7 @@ mdb_cursor_put(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 	if (unlikely(mc->mc_txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)))
 		return (mc->mc_txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
 
-	if (unlikely(key->mv_size-1 >= ENV_MAXKEY(env)))
+	if (unlikely(key->mv_size > ENV_MAXKEY(env)))
 		return MDB_BAD_VALSIZE;
 
 #if SIZE_MAX > MAXDATASIZE

From efcf60dfaa0ed33ecab5bd5ccfb93906dd96862c Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Sun, 18 Dec 2016 20:15:27 +0300
Subject: [PATCH 08/13] mdbx: fix MDB_GET_CURRENT for dupsort's subcursor.

---
 mdb.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/mdb.c b/mdb.c
index 83cd994c..1f4cf026 100644
--- a/mdb.c
+++ b/mdb.c
@@ -6326,6 +6326,12 @@ mdb_cursor_get(MDB_cursor *mc, MDB_val *key, MDB_val *data,
 				MDB_GET_KEY(leaf, key);
 				if (data) {
 					if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+						if (unlikely(!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED))) {
+							mdb_xcursor_init1(mc, leaf);
+							rc = mdb_cursor_first(&mc->mc_xcursor->mx_cursor, data, NULL);
+							if (unlikely(rc))
+								break;
+						}
 						rc = mdb_cursor_get(&mc->mc_xcursor->mx_cursor, data, NULL, MDB_GET_CURRENT);
 					} else {
 						rc = mdb_node_read(mc, leaf, data);

From 7e682540510d5fb140fe42e4824a5587f0a3ec62 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Wed, 21 Dec 2016 20:29:19 +0300
Subject: [PATCH 09/13] mdbx: mdb_chk - don't close dbi-handles, set_maxdbs()
 instead.

---
 mdb_chk.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/mdb_chk.c b/mdb_chk.c
index b86b8f96..fd485fe1 100644
--- a/mdb_chk.c
+++ b/mdb_chk.c
@@ -432,7 +432,6 @@ static int process_db(MDB_dbi dbi, char *name, visitor *handler, int silent)
 			fflush(NULL);
 		}
 		skipped_subdb++;
-		mdbx_dbi_close(env, dbi);
 		return MDB_SUCCESS;
 	}
 
@@ -444,14 +443,12 @@ static int process_db(MDB_dbi dbi, char *name, visitor *handler, int silent)
 	rc = mdbx_dbi_flags(txn, dbi, &flags);
 	if (rc) {
 		error(" - mdbx_dbi_flags failed, error %d %s\n", rc, mdbx_strerror(rc));
-		mdbx_dbi_close(env, dbi);
 		return rc;
 	}
 
 	rc = mdbx_stat(txn, dbi, &ms, sizeof(ms));
 	if (rc) {
 		error(" - mdbx_stat failed, error %d %s\n", rc, mdbx_strerror(rc));
-		mdbx_dbi_close(env, dbi);
 		return rc;
 	}
 
@@ -475,7 +472,6 @@ static int process_db(MDB_dbi dbi, char *name, visitor *handler, int silent)
 	rc = mdbx_cursor_open(txn, dbi, &mc);
 	if (rc) {
 		error(" - mdbx_cursor_open failed, error %d %s\n", rc, mdbx_strerror(rc));
-		mdbx_dbi_close(env, dbi);
 		return rc;
 	}
 
@@ -565,7 +561,6 @@ bailout:
 	}
 
 	mdbx_cursor_close(mc);
-	mdbx_dbi_close(env, dbi);
 	return rc || problems_count;
 }
 
@@ -686,7 +681,11 @@ int main(int argc, char *argv[])
 	}
 	maxkeysize = rc;
 
-	mdbx_env_set_maxdbs(env, 3);
+	rc = mdbx_env_set_maxdbs(env, MAX_DBI);
+	if (rc < 0) {
+		error("mdbx_env_set_maxdbs failed, error %d %s\n", rc, mdbx_strerror(rc));
+		goto bailout;
+	}
 
 	rc = mdbx_env_open_ex(env, envname, envflags, 0664, &exclusive);
 	if (rc) {

From 91bb3ab9fa1068bfbdb61455c9045b105f36095e Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Wed, 21 Dec 2016 20:31:46 +0300
Subject: [PATCH 10/13] mdbx: mdb_chk - cosmetics (no extra \n).

---
 mdb_chk.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mdb_chk.c b/mdb_chk.c
index fd485fe1..b61c06b7 100644
--- a/mdb_chk.c
+++ b/mdb_chk.c
@@ -746,7 +746,7 @@ int main(int argc, char *argv[])
 			   meta_lt(info.me_meta1_txnid, info.me_meta1_sign,
 					   info.me_meta2_txnid, info.me_meta2_sign) ? "tail" : "head");
 		if (info.me_meta1_txnid > info.base.me_last_txnid)
-			print(", rolled-back %zu (%zu >>> %zu)\n",
+			print(", rolled-back %zu (%zu >>> %zu)",
 				info.me_meta1_txnid - info.base.me_last_txnid,
 				info.me_meta1_txnid, info.base.me_last_txnid);
 		print("\n");
@@ -756,7 +756,7 @@ int main(int argc, char *argv[])
 			   meta_lt(info.me_meta2_txnid, info.me_meta2_sign,
 					   info.me_meta1_txnid, info.me_meta1_sign) ? "tail" : "head");
 		if (info.me_meta2_txnid > info.base.me_last_txnid)
-			print(", rolled-back %zu (%zu >>> %zu)\n",
+			print(", rolled-back %zu (%zu >>> %zu)",
 				info.me_meta2_txnid - info.base.me_last_txnid,
 				info.me_meta2_txnid, info.base.me_last_txnid);
 		print("\n");

From 9b38d8d422c79db54f4ba39685c956b44fa1a2da Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Wed, 21 Dec 2016 20:32:27 +0300
Subject: [PATCH 11/13] mdbx: adds mdbx_replace() for libfpta.

---
 mdbx.c | 129 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 mdbx.h |   3 ++
 2 files changed, 132 insertions(+)

diff --git a/mdbx.c b/mdbx.c
index ce930336..781450a5 100644
--- a/mdbx.c
+++ b/mdbx.c
@@ -368,3 +368,132 @@ int mdbx_cursor_eof(MDB_cursor *mc)
 
 	return (mc->mc_flags & C_INITIALIZED) ? 0 : 1;
 }
+
+static int mdbx_is_samedata(MDB_val* a, MDB_val* b) {
+	return a->iov_len == b->iov_len
+		&& memcmp(a->iov_base, b->iov_base, a->iov_len) == 0;
+}
+
+/* Позволяет обновить или удалить существующую запись с получением
+ * в old_data предыдущего значения данных. При этом если new_data равен
+ * нулю, то выполняется удаление, иначе обновление/вставка.
+ *
+ * Текущее значение может находиться в уже измененной (грязной) странице.
+ * В этом случае страница будет перезаписана при обновлении, а само старое
+ * значение утрачено. Поэтому исходно в old_data должен быть передан
+ * дополнительный буфер для копирования старого значения.
+ * Если переданный буфер слишком мал, то функция вернет -1, установив
+ * old_data->iov_len в соответствующее значение.
+ *
+ * Для не-уникальных ключей также возможен второй сценарий использования,
+ * когда посредством old_data из записей с одинаковым ключом для
+ * удаления/обновления выбирается конкретная. Для выбора этого сценария
+ * во flags следует одновременно указать MDB_CURRENT и MDB_NOOVERWRITE.
+ *
+ * Функция может быть замещена соответствующими операциями с курсорами
+ * после двух доработок (TODO):
+ *  - внешняя аллокация курсоров, в том числе на стеке (без malloc).
+ *  - получения статуса страницы по адресу (знать о P_DIRTY).
+ */
+int mdbx_replace(MDB_txn *txn, MDB_dbi dbi,
+	MDB_val *key, MDB_val *new_data, MDB_val *old_data, unsigned flags)
+{
+	MDB_cursor mc;
+	MDB_xcursor mx;
+
+	if (unlikely(!key || !old_data || !txn))
+		return EINVAL;
+
+	if (unlikely(txn->mt_signature != MDBX_MT_SIGNATURE))
+		return MDB_VERSION_MISMATCH;
+
+	if (unlikely(old_data->iov_base == NULL && old_data->iov_len))
+		return EINVAL;
+
+	if (unlikely(new_data == NULL && !(flags & MDB_CURRENT)))
+		return EINVAL;
+
+	if (unlikely(!TXN_DBI_EXIST(txn, dbi, DB_USRVALID)))
+		return EINVAL;
+
+	if (unlikely(flags & ~(MDB_NOOVERWRITE|MDB_NODUPDATA|MDB_RESERVE|MDB_APPEND|MDB_APPENDDUP|MDB_CURRENT)))
+		return EINVAL;
+
+	if (unlikely(txn->mt_flags & (MDB_TXN_RDONLY|MDB_TXN_BLOCKED)))
+		return (txn->mt_flags & MDB_TXN_RDONLY) ? EACCES : MDB_BAD_TXN;
+
+	mdb_cursor_init(&mc, txn, dbi, &mx);
+	mc.mc_next = txn->mt_cursors[dbi];
+	txn->mt_cursors[dbi] = &mc;
+
+	int rc;
+	MDB_val present_key = *key;
+	if (F_ISSET(flags, MDB_CURRENT | MDB_NOOVERWRITE)
+			&& (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT)) {
+		/* в old_data значение для выбора конкретного дубликата */
+		rc = mdbx_cursor_get(&mc, &present_key, old_data, MDB_GET_BOTH);
+		if (rc != MDB_SUCCESS)
+			goto bailout;
+		/* если данные совпадают, то ничего делать не надо */
+		if (new_data && mdbx_is_samedata(old_data, new_data))
+			goto bailout;
+	} else {
+		/* в old_data буфер получения предыдущего значения */
+		MDB_val present_data;
+		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;
+			if (rc != MDB_NOTFOUND || (flags & MDB_CURRENT))
+				goto bailout;
+		} else if (flags & MDB_NOOVERWRITE) {
+			rc = MDB_KEYEXIST;
+			*old_data = present_data;
+			goto bailout;
+		} else {
+			MDB_page *page = mc.mc_pg[mc.mc_top];
+			if (txn->mt_dbs[dbi].md_flags & MDB_DUPSORT) {
+				if (flags & MDB_CURRENT) {
+					/* для не-уникальных ключей позволяем update/delete только если ключ один */
+					MDB_node *leaf = NODEPTR(page, mc.mc_ki[mc.mc_top]);
+					if (F_ISSET(leaf->mn_flags, F_DUPDATA)) {
+						mdb_tassert(txn, XCURSOR_INITED(&mc) && mc.mc_xcursor->mx_db.md_entries > 1);
+						rc = MDB_KEYEXIST;
+						goto bailout;
+					}
+					/* если данные совпадают, то ничего делать не надо */
+					if (new_data && mdbx_is_samedata(&present_data, new_data))
+						goto bailout;
+				} else if ((flags & MDB_NODUPDATA) && mdbx_is_samedata(&present_data, new_data)) {
+					/* если данные совпадают и установлен MDB_NODUPDATA */
+					rc = MDB_KEYEXIST;
+					goto bailout;
+				}
+			} else {
+				flags |= MDB_CURRENT;
+			}
+
+			if (page->mp_flags & P_DIRTY) {
+				if (unlikely(old_data->iov_len < present_data.iov_len)) {
+					old_data->iov_base = NULL;
+					old_data->iov_len = present_data.iov_len;
+					rc = -1;
+					goto bailout;
+				}
+				memcpy(old_data->iov_base, present_data.iov_base, present_data.iov_len);
+				old_data->iov_len = present_data.iov_len;
+			} else {
+				*old_data = present_data;
+			}
+		}
+	}
+
+	if (likely(new_data))
+		rc = mdbx_cursor_put(&mc, key, new_data, flags);
+	else
+		rc = mdbx_cursor_del(&mc, 0);
+
+bailout:
+	txn->mt_cursors[dbi] = mc.mc_next;
+	return rc;
+}
diff --git a/mdbx.h b/mdbx.h
index 8a5c238e..033db286 100644
--- a/mdbx.h
+++ b/mdbx.h
@@ -223,6 +223,9 @@ size_t mdbx_canary_get(MDB_txn *txn, mdbx_canary* canary);
  * 0 otherwise or less that zero in error case. */
 int mdbx_cursor_eof(MDB_cursor *mc);
 
+int mdbx_replace(MDB_txn *txn, MDB_dbi dbi,
+	MDB_val *key, MDB_val *new_data, MDB_val *old_data, unsigned flags);
+
 /**	@} */
 
 #ifdef __cplusplus

From ef375647c70ef5ee35f9c23a7fb15a6820ec7f80 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Thu, 22 Dec 2016 20:54:06 +0300
Subject: [PATCH 12/13] mdbx: fix mdbx_replace().

Always return `old_data`, even no changes.
---
 mdbx.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/mdbx.c b/mdbx.c
index 781450a5..3946a50c 100644
--- a/mdbx.c
+++ b/mdbx.c
@@ -369,7 +369,7 @@ int mdbx_cursor_eof(MDB_cursor *mc)
 	return (mc->mc_flags & C_INITIALIZED) ? 0 : 1;
 }
 
-static int mdbx_is_samedata(MDB_val* a, MDB_val* b) {
+static int mdbx_is_samedata(const MDB_val* a, const MDB_val* b) {
 	return a->iov_len == b->iov_len
 		&& memcmp(a->iov_base, b->iov_base, a->iov_len) == 0;
 }
@@ -462,14 +462,21 @@ int mdbx_replace(MDB_txn *txn, MDB_dbi dbi,
 						goto bailout;
 					}
 					/* если данные совпадают, то ничего делать не надо */
-					if (new_data && mdbx_is_samedata(&present_data, new_data))
+					if (new_data && mdbx_is_samedata(&present_data, new_data)) {
+						*old_data = *new_data;
 						goto bailout;
+					}
 				} else if ((flags & MDB_NODUPDATA) && mdbx_is_samedata(&present_data, new_data)) {
 					/* если данные совпадают и установлен MDB_NODUPDATA */
 					rc = MDB_KEYEXIST;
 					goto bailout;
 				}
 			} else {
+				/* если данные совпадают, то ничего делать не надо */
+				if (new_data && mdbx_is_samedata(&present_data, new_data)) {
+					*old_data = *new_data;
+					goto bailout;
+				}
 				flags |= MDB_CURRENT;
 			}
 

From dbc57d3eaf9025b8a5b3d14dd2ecbd0589d1c1b4 Mon Sep 17 00:00:00 2001
From: Leo Yuriev <leo@yuriev.ru>
Date: Fri, 23 Dec 2016 15:35:42 +0300
Subject: [PATCH 13/13] mdbx: fix cursor_count() for libfpta.

---
 mdb.c | 30 +++++++++++++++++++++++-------
 1 file changed, 23 insertions(+), 7 deletions(-)

diff --git a/mdb.c b/mdb.c
index 4976179c..323edf6d 100644
--- a/mdb.c
+++ b/mdb.c
@@ -7713,35 +7713,51 @@ mdb_cursor_renew(MDB_txn *txn, MDB_cursor *mc)
 int
 mdb_cursor_count(MDB_cursor *mc, size_t *countp)
 {
-	MDB_node	*leaf;
-
 	if (unlikely(mc == NULL || countp == NULL))
 		return EINVAL;
 
 	if (unlikely(mc->mc_signature != MDBX_MC_SIGNATURE))
 		return MDB_VERSION_MISMATCH;
 
-	if (unlikely(mc->mc_xcursor == NULL))
-		return MDB_INCOMPATIBLE;
-
 	if (unlikely(mc->mc_txn->mt_flags & MDB_TXN_BLOCKED))
 		return MDB_BAD_TXN;
 
 	if (unlikely(!(mc->mc_flags & C_INITIALIZED)))
 		return EINVAL;
 
+#if MDBX_MODE_ENABLED
+	MDB_page *mp = mc->mc_pg[mc->mc_top];
+	int nkeys = NUMKEYS(mp);
+	if (!nkeys || mc->mc_ki[mc->mc_top] >= nkeys) {
+		*countp = 0;
+		return MDB_NOTFOUND;
+	} else if (mc->mc_xcursor == NULL || IS_LEAF2(mp)) {
+		*countp = 1;
+        } else {
+		MDB_node *leaf = NODEPTR(mp, mc->mc_ki[mc->mc_top]);
+		if (!F_ISSET(leaf->mn_flags, F_DUPDATA))
+			*countp = 1;
+		else if (unlikely(!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)))
+			return EINVAL;
+		else
+			*countp = mc->mc_xcursor->mx_db.md_entries;
+	}
+#else
+        if (unlikely(mc->mc_xcursor == NULL))
+		return MDB_INCOMPATIBLE;
+
 	if (unlikely(!mc->mc_snum || (mc->mc_flags & C_EOF)))
 		return MDB_NOTFOUND;
 
-	leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
+	MDB_node *leaf = NODEPTR(mc->mc_pg[mc->mc_top], mc->mc_ki[mc->mc_top]);
 	if (!F_ISSET(leaf->mn_flags, F_DUPDATA)) {
 		*countp = 1;
 	} else {
 		if (unlikely(!(mc->mc_xcursor->mx_cursor.mc_flags & C_INITIALIZED)))
 			return EINVAL;
-
 		*countp = mc->mc_xcursor->mx_db.md_entries;
 	}
+#endif /* MDBX_MODE_ENABLED */
 	return MDB_SUCCESS;
 }