mdbx: import intrin-includes from t1ha.

Based on http://www.openldap.org/devel/gitweb.cgi?p=openldap.git;a=commitdiff;h=e77918a903d980ff789b7e32f71725481e870510

Makefile

@@ -32,11 +32,12 @@ endif
 XCFLAGS	?= -DNDEBUG=1 -DMDBX_DEBUG=0 -DLIBMDBX_EXPORTS=1
 CFLAGS	+= -D_GNU_SOURCE=1 -std=gnu11 -pthread $(XCFLAGS)
 CXXFLAGS = -std=c++11 $(filter-out -std=gnu11,$(CFLAGS))
-TESTDB	?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-check.db
-TESTLOG ?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-check.log
+TESTDB	?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-test.db
+TESTLOG ?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-test.log
 
 # LY: '--no-as-needed,-lrt' for ability to built with modern glibc, but then run with the old
 LDFLAGS	?= -Wl,--gc-sections,-z,relro,-O,--no-as-needed,-lrt
+EXE_LDFLAGS ?= $(LDFLAGS) -static
 
 # LY: just for benchmarking
 IOARENA ?= $(shell \
@@ -87,6 +88,12 @@ clean:
 check:	all
 	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; test/test --pathname=$(TESTDB) --dont-cleanup-after basic | tee -a $(TESTLOG) | tail -n 42) && ./mdbx_chk -vvn $(TESTDB)
 
+check-singleprocess:	all
+	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; test/test --pathname=$(TESTDB) --dont-cleanup-after --hill | tee -a $(TESTLOG) | tail -n 42) && ./mdbx_chk -vvn $(TESTDB)
+
+check-fault:	all
+	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; test/test --pathname=$(TESTDB) --inject-writefault=42 --dump-config --dont-cleanup-after basic | tee -a $(TESTLOG) | tail -n 42) && ./mdbx_chk -vvn $(TESTDB)
+
 define core-rule
 $(patsubst %.c,%.o,$(1)): $(1) $(CORE_INC) mdbx.h Makefile
 	$(CC) $(CFLAGS) -c $(1) -o $$@
@@ -108,10 +115,12 @@ libmdbx.so: $(CORE_OBJ)
 	$(CC) $(CFLAGS) -save-temps $^ -pthread -shared $(LDFLAGS) -o $@
 
 mdbx_%:	src/tools/mdbx_%.c libmdbx.a
-	$(CC) $(CFLAGS) $^ $(LDFLAGS) -o $@
+	$(CC) $(CFLAGS) $^ $(EXE_LDFLAGS) -o $@
 
 test/test: $(TEST_OBJ) libmdbx.a
-	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
+	$(CXX) $(CXXFLAGS) $^ $(EXE_LDFLAGS) -o $@
+
+###############################################################################
 
 ifneq ($(wildcard $(IOARENA)),)
 
@@ -160,6 +169,8 @@ bench-quartet: bench-mdbx_$(NN).txt bench-lmdb_$(NN).txt bench-rocksdb_$(NN).txt
 
 endif
 
+###############################################################################
+
 ci-rule = ( CC=$$(which $1); if [ -n "$$CC" ]; then \
 		echo -n "probe by $2 ($$(readlink -f $$(which $$CC))): " && \
 		$(MAKE) clean >$1.log 2>$1.err && \
@@ -175,3 +186,36 @@ ci:
 	@$(call ci-rule,gcc,GCC)
 	@$(call ci-rule,clang,clang LLVM)
 	@$(call ci-rule,icc,Intel C)
+
+###############################################################################
+
+CROSS_LIST = alpha-linux-gnu-gcc mips-linux-gnu-gcc \
+	powerpc64-linux-gnu-gcc powerpc-linux-gnu-gcc \
+	arm-linux-gnueabihf-gcc aarch64-linux-gnu-gcc
+
+# hppa-linux-gnu-gcc		- don't supported by current qemu release
+# s390x-linux-gnu-gcc		- qemu troubles (hang/abort)
+# sh4-linux-gnu-gcc		- qemu troubles (pread syscall, etc)
+# mips64-linux-gnuabi64-gcc	- qemu troubles (pread syscall, etc)
+# sparc64-linux-gnu-gcc		- qemu troubles (fcntl for F_SETLK/F_GETLK)
+CROSS_LIST_NOQEMU = hppa-linux-gnu-gcc s390x-linux-gnu-gcc \
+	sh4-linux-gnu-gcc mips64-linux-gnuabi64-gcc sparc64-linux-gnu-gcc
+
+cross-gcc:
+	@echo "CORRESPONDING CROSS-COMPILERs ARE REQUIRED."
+	@echo "FOR INSTANCE: apt install g++-aarch64-linux-gnu g++-alpha-linux-gnu g++-arm-linux-gnueabihf g++-hppa-linux-gnu g++-mips-linux-gnu g++-mips64-linux-gnuabi64 g++-powerpc-linux-gnu g++-powerpc64-linux-gnu g++-s390x-linux-gnu g++-sh4-linux-gnu"
+	@for CC in $(CROSS_LIST_NOQEMU) $(CROSS_LIST); do \
+		echo "===================== $$CC"; \
+		$(MAKE) clean && CC=$$CC CXX=$$(echo $$CC | sed 's/-gcc/-g++/') EXE_LDFLAGS=-static $(MAKE) all || exit $$?; \
+	done
+
+#
+# Unfortunately qemu don't provide robust support for futexes.
+# Therefore it is impossible to run full multi-process tests.
+cross-qemu:
+	@echo "CORRESPONDING CROSS-COMPILERs AND QEMUs ARE REQUIRED."
+	@echo "FOR INSTANCE: apt install binfmt-support qemu-user-static qemu-user qemu-system-arm qemu-system-mips qemu-system-misc qemu-system-ppc qemu-system-sparc g++-aarch64-linux-gnu g++-alpha-linux-gnu g++-arm-linux-gnueabihf g++-hppa-linux-gnu g++-mips-linux-gnu g++-mips64-linux-gnuabi64 g++-powerpc-linux-gnu g++-powerpc64-linux-gnu g++-s390x-linux-gnu g++-sh4-linux-gnu"
+	@for CC in $(CROSS_LIST); do \
+		echo "===================== $$CC + qemu"; \
+		$(MAKE) clean && CC=$$CC CXX=$$(echo $$CC | sed 's/-gcc/-g++/') EXE_LDFLAGS=-static $(MAKE) check-singleprocess || exit $$?; \
+	done

README.md

@@ -233,7 +233,7 @@ scanning data directory.
    Because of this syncing data to disk might be quite resource intensive and be main performance bottleneck
    during intensive write workload.
    > As compromise _libmdbx_ allows several modes of lazy and/or periodic syncing, including `MAPASYNC` mode, which modificate
-   > data in memory and asynchronously syncs data to disc, moment to sync is picked by OS.
+   > data in memory and asynchronously syncs data to disk, moment to sync is picked by OS.
    >
    > Although this should be used with care, synchronous transactions in a DB with transaction journal will require 2 IOPS
    > minimum (probably 3-4 in practice) because of filesystem overhead, overhead depends on filesystem, not on record
@@ -339,7 +339,7 @@ Improvements over LMDB
 	This allows to minimize reclaim loop and make it execution time independent from total page count.
 
 	This results in OS kernel cache mechanisms working with maximum efficiency.
-	In case of using disc controllers or storages with
+	In case of using disk controllers or storages with
 	[BBWC](https://en.wikipedia.org/wiki/Disk_buffer#Write_acceleration) this may greatly improve
 	write performance.
 
@@ -378,7 +378,7 @@ Improvements over LMDB
 
 9. Check if there is a row with data after current cursor position via `mdbx_cursor_eof()`.
 
-10. Ability to explicitly request update of current record without creating new record. Implemented as `MDBX_CURRENT` flag
+10. Ability to explicitly request update of present record without creating new record. Implemented as `MDBX_CURRENT` flag
     for `mdbx_put()`.
 
 11. Ability to update or delete record and get previous value via `mdbx_replace()` Also can update specific multi-value.
@@ -419,7 +419,7 @@ Improvements over LMDB
 22. Additional error code `MDBX_EMULTIVAL`, which is returned by `mdbx_put()` and
     `mdbx_replace()` in case is ambiguous update or delete.
 
-23. Ability to get value by key and duplicates count by `mdbx_get_ex()`
+23. Ability to get value by key and duplicates count by `mdbx_get_ex()`.
 
 24. Functions `mdbx_cursor_on_first() and mdbx_cursor_on_last(), which allows to know if cursor is currently on first or
     last position respectively.

TODO.md

@@ -21,7 +21,7 @@
 Тесты
 =====
 - [ ] Тестирование поддержки lockless-режима.
-- [ ] Додумать имя и размещение тестовой БД по-умолчанию.
+- [x] Додумать имя и размещение тестовой БД по-умолчанию.
 - [ ] Реализовать cleanup в тесте.
 - [ ] usage для теста.
 - [ ] Логирование в файл, плюс более полный progress bar.

mdbx.h

@@ -982,6 +982,15 @@ LIBMDBX_API int mdbx_txn_begin(MDBX_env *env, MDBX_txn *parent, unsigned flags,
  * [in] txn  A transaction handle returned by mdbx_txn_begin() */
 LIBMDBX_API MDBX_env *mdbx_txn_env(MDBX_txn *txn);
 
+/* Return the transaction's flags.
+ *
+ * This returns the flags associated with this transaction.
+ *
+ * [in] txn A transaction handle returned by mdbx_txn_begin()
+ *
+ * Returns A transaction flags, valid if input is an active transaction. */
+LIBMDBX_API int mdbx_txn_flags(MDBX_txn *txn);
+
 /* Return the transaction's ID.
  *
  * This returns the identifier associated with this transaction. For a
@@ -1818,6 +1827,42 @@ LIBMDBX_API int mdbx_cursor_get_attr(MDBX_cursor *mc, MDBX_val *key,
 LIBMDBX_API int mdbx_get_attr(MDBX_txn *txn, MDBX_dbi dbi, MDBX_val *key,
                               MDBX_val *data, mdbx_attr_t *attrptr);
 
+/*----------------------------------------------------------------------------*/
+/* LY: temporary workaround for Elbrus's memcmp() bug. */
+#ifndef __GLIBC_PREREQ
+#if defined(__GLIBC__) && defined(__GLIBC_MINOR__)
+#define __GLIBC_PREREQ(maj, min)                                               \
+  ((__GLIBC__ << 16) + __GLIBC_MINOR__ >= ((maj) << 16) + (min))
+#else
+#define __GLIBC_PREREQ(maj, min) (0)
+#endif
+#endif /* __GLIBC_PREREQ */
+#if defined(__e2k__) && !__GLIBC_PREREQ(2, 24)
+LIBMDBX_API int mdbx_e2k_memcmp_bug_workaround(const void *s1, const void *s2,
+                                               size_t n);
+LIBMDBX_API int mdbx_e2k_strcmp_bug_workaround(const char *s1, const char *s2);
+LIBMDBX_API int mdbx_e2k_strncmp_bug_workaround(const char *s1, const char *s2,
+                                                size_t n);
+LIBMDBX_API size_t mdbx_e2k_strlen_bug_workaround(const char *s);
+LIBMDBX_API size_t mdbx_e2k_strnlen_bug_workaround(const char *s,
+                                                   size_t maxlen);
+#include <string.h>
+#include <strings.h>
+#undef memcmp
+#define memcmp mdbx_e2k_memcmp_bug_workaround
+#undef bcmp
+#define bcmp mdbx_e2k_memcmp_bug_workaround
+#undef strcmp
+#define strcmp mdbx_e2k_strcmp_bug_workaround
+#undef strncmp
+#define strncmp mdbx_e2k_strncmp_bug_workaround
+#undef strlen
+#define strlen mdbx_e2k_strlen_bug_workaround
+#undef strnlen
+#define strnlen mdbx_e2k_strnlen_bug_workaround
+
+#endif /* Elbrus's memcmp() bug. */
+
 #ifdef __cplusplus
 }
 #endif

src/bits.h

@@ -55,6 +55,7 @@
 #pragma warning(disable : 4310) /* cast truncates constant value */
 #pragma warning(disable : 4820) /* bytes padding added after data member for aligment */
 #pragma warning(disable : 4548) /* expression before comma has no effect; expected expression with side - effect */
+#pragma warning(disable : 4366) /* the result of the unary '&' operator may be unaligned */
 #endif                          /* _MSC_VER (warnings) */
 
 #include "../mdbx.h"
@@ -111,7 +112,7 @@
 #   else
 #       pragma warning disable "alignment-reduction-ignored"
 #   endif
-#endif /* -Wno-constant-logical-operand */
+#endif /* -Walignment-reduction-ignored */
 
 #include "./osal.h"
 
@@ -252,7 +253,7 @@ typedef struct MDBX_reader {
   uint8_t pad[MDBX_CACHELINE_SIZE -
               (sizeof(txnid_t) + sizeof(mdbx_pid_t) + sizeof(mdbx_tid_t)) %
                   MDBX_CACHELINE_SIZE];
-} __cache_aligned MDBX_reader;
+} MDBX_reader;
 
 /* Information about a single database in the environment. */
 typedef struct MDBX_db {
@@ -410,47 +411,53 @@ typedef struct MDBX_lockinfo {
   /* Flags which environment was opened. */
   volatile uint32_t mti_envmode;
 
-  union {
 #ifdef MDBX_OSAL_LOCK
+  /* Mutex protecting write access to this table. */
+  union {
     MDBX_OSAL_LOCK mti_wmutex;
+    uint8_t pad_mti_wmutex[MDBX_OSAL_LOCK_SIZE % sizeof(size_t)];
+  };
 #endif
-    uint64_t align_wmutex;
-  };
+#define MDBX_lockinfo_SIZE_A                                                   \
+  (8 /* mti_magic_and_version */ + 4 /* mti_os_and_format */ +                 \
+   4 /* mti_envmode */ + MDBX_OSAL_LOCK_SIZE /* mti_wmutex */ +                \
+   MDBX_OSAL_LOCK_SIZE % sizeof(size_t) /* pad_mti_wmutex */)
 
-  union {
-    /* The number of slots that have been used in the reader table.
-     * This always records the maximum count, it is not decremented
-     * when readers release their slots. */
-    volatile unsigned __cache_aligned mti_numreaders;
-    uint64_t align_numreaders;
-  };
+  /* cache-line alignment */
+  uint8_t
+      pad_a[MDBX_CACHELINE_SIZE - MDBX_lockinfo_SIZE_A % MDBX_CACHELINE_SIZE];
+
+  /* The number of slots that have been used in the reader table.
+   * This always records the maximum count, it is not decremented
+   * when readers release their slots. */
+  volatile unsigned mti_numreaders;
 
-  union {
 #ifdef MDBX_OSAL_LOCK
-    /* Mutex protecting access to this table. */
+  /* Mutex protecting readers registration access to this table. */
+  union {
     MDBX_OSAL_LOCK mti_rmutex;
+    uint8_t pad_mti_rmutex[MDBX_OSAL_LOCK_SIZE % sizeof(size_t)];
+  };
 #endif
-    uint64_t align_rmutex;
-  };
 
-  union {
-    volatile txnid_t mti_oldest;
-    uint64_t align_oldest;
-  };
+  volatile txnid_t mti_oldest;
+  volatile uint32_t mti_readers_refresh_flag;
 
-  union {
-    volatile uint32_t mti_readers_refresh_flag;
-    uint64_t align_reader_finished_flag;
-  };
+#define MDBX_lockinfo_SIZE_B                                                   \
+  (sizeof(unsigned) /* mti_numreaders */ +                                     \
+   MDBX_OSAL_LOCK_SIZE /* mti_rmutex */ + sizeof(txnid_t) /* mti_oldest */ +   \
+   sizeof(uint32_t) /* mti_readers_refresh_flag */ +                           \
+   MDBX_OSAL_LOCK_SIZE % sizeof(size_t) /* pad_mti_rmutex */)
 
-  uint8_t pad_align[MDBX_CACHELINE_SIZE - sizeof(uint64_t) * 7];
+  /* cache-line alignment */
+  uint8_t
+      pad_b[MDBX_CACHELINE_SIZE - MDBX_lockinfo_SIZE_B % MDBX_CACHELINE_SIZE];
+
+  MDBX_reader mti_readers[1];
 
-  MDBX_reader __cache_aligned mti_readers[1];
 } MDBX_lockinfo;
 
-#ifdef _MSC_VER
 #pragma pack(pop)
-#endif /* MSVC: Enable aligment */
 
 #define MDBX_LOCKINFO_WHOLE_SIZE                                               \
   ((sizeof(MDBX_lockinfo) + MDBX_CACHELINE_SIZE - 1) &                         \

src/defs.h

@@ -192,16 +192,6 @@
 #	endif
 #endif /* __prefetch */
 
-#ifndef __aligned
-#   if defined(__GNUC__) || __has_attribute(aligned)
-#       define __aligned(N) __attribute__((aligned(N)))
-#   elif defined(_MSC_VER)
-#       define __aligned(N) __declspec(align(N))
-#   else
-#       define __aligned(N)
-#   endif
-#endif /* __aligned */
-
 #ifndef __noreturn
 #   if defined(__GNUC__) || __has_attribute(noreturn)
 #       define __noreturn __attribute__((noreturn))

src/mdbx.c

@@ -37,6 +37,117 @@
 
 #include "./bits.h"
 
+/* LY: temporary workaround for Elbrus's memcmp() bug. */
+#if defined(__e2k__) && !__GLIBC_PREREQ(2, 24)
+int __hot mdbx_e2k_memcmp_bug_workaround(const void *s1, const void *s2,
+                                         size_t n) {
+  if (unlikely(n > 42
+               /* LY: align followed access if reasonable possible */ &&
+               (((uintptr_t)s1) & 7) != 0 &&
+               (((uintptr_t)s1) & 7) == (((uintptr_t)s2) & 7))) {
+    if (((uintptr_t)s1) & 1) {
+      const int diff = *(uint8_t *)s1 - *(uint8_t *)s2;
+      if (diff)
+        return diff;
+      s1 = (char *)s1 + 1;
+      s2 = (char *)s2 + 1;
+      n -= 1;
+    }
+
+    if (((uintptr_t)s1) & 2) {
+      const uint16_t a = *(uint16_t *)s1;
+      const uint16_t b = *(uint16_t *)s2;
+      if (likely(a != b))
+        return (__builtin_bswap16(a) > __builtin_bswap16(b)) ? 1 : -1;
+      s1 = (char *)s1 + 2;
+      s2 = (char *)s2 + 2;
+      n -= 2;
+    }
+
+    if (((uintptr_t)s1) & 4) {
+      const uint32_t a = *(uint32_t *)s1;
+      const uint32_t b = *(uint32_t *)s2;
+      if (likely(a != b))
+        return (__builtin_bswap32(a) > __builtin_bswap32(b)) ? 1 : -1;
+      s1 = (char *)s1 + 4;
+      s2 = (char *)s2 + 4;
+      n -= 4;
+    }
+  }
+
+  while (n >= 8) {
+    const uint64_t a = *(uint64_t *)s1;
+    const uint64_t b = *(uint64_t *)s2;
+    if (likely(a != b))
+      return (__builtin_bswap64(a) > __builtin_bswap64(b)) ? 1 : -1;
+    s1 = (char *)s1 + 8;
+    s2 = (char *)s2 + 8;
+    n -= 8;
+  }
+
+  if (n & 4) {
+    const uint32_t a = *(uint32_t *)s1;
+    const uint32_t b = *(uint32_t *)s2;
+    if (likely(a != b))
+      return (__builtin_bswap32(a) > __builtin_bswap32(b)) ? 1 : -1;
+    s1 = (char *)s1 + 4;
+    s2 = (char *)s2 + 4;
+  }
+
+  if (n & 2) {
+    const uint16_t a = *(uint16_t *)s1;
+    const uint16_t b = *(uint16_t *)s2;
+    if (likely(a != b))
+      return (__builtin_bswap16(a) > __builtin_bswap16(b)) ? 1 : -1;
+    s1 = (char *)s1 + 2;
+    s2 = (char *)s2 + 2;
+  }
+
+  return (n & 1) ? *(uint8_t *)s1 - *(uint8_t *)s2 : 0;
+}
+
+int __hot mdbx_e2k_strcmp_bug_workaround(const char *s1, const char *s2) {
+  while (true) {
+    int diff = *(uint8_t *)s1 - *(uint8_t *)s2;
+    if (likely(diff != 0) || *s1 == '\0')
+      return diff;
+    s1 += 1;
+    s2 += 1;
+  }
+}
+
+int __hot mdbx_e2k_strncmp_bug_workaround(const char *s1, const char *s2,
+                                          size_t n) {
+  while (n > 0) {
+    int diff = *(uint8_t *)s1 - *(uint8_t *)s2;
+    if (likely(diff != 0) || *s1 == '\0')
+      return diff;
+    s1 += 1;
+    s2 += 1;
+    n -= 1;
+  }
+  return 0;
+}
+
+size_t __hot mdbx_e2k_strlen_bug_workaround(const char *s) {
+  size_t n = 0;
+  while (*s) {
+    s += 1;
+    n += 1;
+  }
+  return n;
+}
+
+size_t __hot mdbx_e2k_strnlen_bug_workaround(const char *s, size_t maxlen) {
+  size_t n = 0;
+  while (maxlen > n && *s) {
+    s += 1;
+    n += 1;
+  }
+  return n;
+}
+#endif /* Elbrus's memcmp() bug. */
+
 /*----------------------------------------------------------------------------*/
 /* rthc (tls keys and destructors) */
 
@@ -2507,6 +2618,12 @@ static int mdbx_txn_renew0(MDBX_txn *txn, unsigned flags) {
     return MDBX_PANIC;
   }
 
+  STATIC_ASSERT(sizeof(MDBX_reader) == MDBX_CACHELINE_SIZE);
+  STATIC_ASSERT(offsetof(MDBX_lockinfo, mti_numreaders) % MDBX_CACHELINE_SIZE ==
+                0);
+  STATIC_ASSERT(offsetof(MDBX_lockinfo, mti_readers) % MDBX_CACHELINE_SIZE ==
+                0);
+
   pgno_t upper_pgno = 0;
   if (flags & MDBX_TXN_RDONLY) {
     txn->mt_flags = MDBX_TXN_RDONLY;
@@ -2562,9 +2679,6 @@ static int mdbx_txn_renew0(MDBX_txn *txn, unsigned flags) {
         }
       }
 
-      STATIC_ASSERT(sizeof(MDBX_reader) == MDBX_CACHELINE_SIZE);
-      STATIC_ASSERT(
-          offsetof(MDBX_lockinfo, mti_readers) % MDBX_CACHELINE_SIZE == 0);
       r = &env->me_lck->mti_readers[slot];
       /* Claim the reader slot, carefully since other code
        * uses the reader table un-mutexed: First reset the
@@ -2876,6 +2990,13 @@ uint64_t mdbx_txn_id(MDBX_txn *txn) {
   return txn->mt_txnid;
 }
 
+int mdbx_txn_flags(MDBX_txn *txn) {
+  if (unlikely(!txn || txn->mt_signature != MDBX_MT_SIGNATURE))
+    return -1;
+
+  return txn->mt_flags;
+}
+
 /* Export or close DBI handles opened in this txn. */
 static void mdbx_dbis_update(MDBX_txn *txn, int keep) {
   MDBX_dbi n = txn->mt_numdbs;
@@ -4042,7 +4163,7 @@ static int __cold mdbx_read_header(MDBX_env *env, MDBX_meta *meta) {
     STATIC_ASSERT(MIN_MAPSIZE < MAX_MAPSIZE);
     if (mapsize_max > MAX_MAPSIZE ||
         MAX_PAGENO < mdbx_roundup2((size_t)mapsize_max, env->me_os_psize) /
-                         (uint64_t)page.mp_meta.mm_psize) {
+                         (size_t)page.mp_meta.mm_psize) {
       const uint64_t used_bytes =
           page.mp_meta.mm_geo.next * (uint64_t)page.mp_meta.mm_psize;
       if (page.mp_meta.mm_geo.next - 1 > MAX_PAGENO ||
@@ -4644,13 +4765,13 @@ LIBMDBX_API int mdbx_env_set_geometry(MDBX_env *env, intptr_t size_lower,
 
     if (pagesize < 0) {
       pagesize = env->me_os_psize;
-      if (pagesize > MAX_PAGESIZE)
+      if ((uintptr_t)pagesize > MAX_PAGESIZE)
         pagesize = MAX_PAGESIZE;
-      mdbx_assert(env, pagesize >= MIN_PAGESIZE);
+      mdbx_assert(env, (uintptr_t)pagesize >= MIN_PAGESIZE);
     }
   }
 
-  if (pagesize < MIN_PAGESIZE || pagesize > MAX_PAGESIZE ||
+  if (pagesize < (intptr_t)MIN_PAGESIZE || pagesize > (intptr_t)MAX_PAGESIZE ||
       !mdbx_is_power2(pagesize)) {
     rc = MDBX_EINVAL;
     goto bailout;
@@ -4684,7 +4805,7 @@ LIBMDBX_API int mdbx_env_set_geometry(MDBX_env *env, intptr_t size_lower,
       size_upper = pagesize * MAX_PAGENO;
   }
 
-  if (unlikely(size_lower < MIN_MAPSIZE || size_lower > size_upper)) {
+  if (unlikely(size_lower < (intptr_t)MIN_MAPSIZE || size_lower > size_upper)) {
     rc = MDBX_EINVAL;
     goto bailout;
   }
@@ -4991,7 +5112,7 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
                 filesize_before_mmap,
                 bytes2pgno(env, (size_t)filesize_before_mmap));
     } else {
-      mdbx_notice("filesize mismatch (expect %" PRIuPTR "/%" PRIaPGNO
+      mdbx_notice("filesize mismatch (expect %" PRIuSIZE "/%" PRIaPGNO
                   ", have %" PRIu64 "/%" PRIaPGNO ")",
                   expected_bytes, bytes2pgno(env, expected_bytes),
                   filesize_before_mmap,
@@ -5007,11 +5128,11 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
       if (env->me_flags & MDBX_RDONLY) {
         mdbx_notice("ignore filesize mismatch in readonly-mode");
       } else {
-        mdbx_info("resize datafile to %" PRIu64 " bytes, %" PRIaPGNO " pages",
+        mdbx_info("resize datafile to %" PRIuSIZE " bytes, %" PRIaPGNO " pages",
                   expected_bytes, bytes2pgno(env, expected_bytes));
         err = mdbx_ftruncate(env->me_fd, expected_bytes);
         if (unlikely(err != MDBX_SUCCESS)) {
-          mdbx_error("error %d, while resize datafile to %" PRIu64
+          mdbx_error("error %d, while resize datafile to %" PRIuSIZE
                      " bytes, %" PRIaPGNO " pages",
                      rc, expected_bytes, bytes2pgno(env, expected_bytes));
           return err;
@@ -5048,15 +5169,30 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
         return MDBX_WANNA_RECOVERY /* LY: could not recovery/rollback */;
       }
 
+      const MDBX_meta *const meta0 = METAPAGE(env, 0);
+      const MDBX_meta *const meta1 = METAPAGE(env, 1);
+      const MDBX_meta *const meta2 = METAPAGE(env, 2);
+      txnid_t undo_txnid = 0;
+      while (
+          (head != meta0 && mdbx_meta_txnid_fluid(env, meta0) == undo_txnid) ||
+          (head != meta1 && mdbx_meta_txnid_fluid(env, meta1) == undo_txnid) ||
+          (head != meta2 && mdbx_meta_txnid_fluid(env, meta2) == undo_txnid))
+        undo_txnid += 1;
+      if (unlikely(undo_txnid >= meta.mm_txnid_a)) {
+        mdbx_fatal("rollback failed: no suitable txnid (0,1,2) < %" PRIaTXN,
+                   meta.mm_txnid_a);
+        return MDBX_PANIC /* LY: could not recovery/rollback */;
+      }
+
       /* LY: rollback weak checkpoint */
-      mdbx_trace("rollback: from %" PRIaTXN ", to %" PRIaTXN, head_txnid,
-                 meta.mm_txnid_a);
+      mdbx_trace("rollback: from %" PRIaTXN ", to %" PRIaTXN " as %" PRIaTXN,
+                 head_txnid, meta.mm_txnid_a, undo_txnid);
       mdbx_ensure(env, head_txnid == mdbx_meta_txnid_stable(env, head));
 
       if (env->me_flags & MDBX_WRITEMAP) {
-        head->mm_txnid_a = 0;
+        head->mm_txnid_a = undo_txnid;
         head->mm_datasync_sign = MDBX_DATASIGN_WEAK;
-        head->mm_txnid_b = 0;
+        head->mm_txnid_b = undo_txnid;
         const size_t offset =
             ((uint8_t *)container_of(head, MDBX_page, mp_meta)) -
             env->me_dxb_mmap.dxb;
@@ -5066,7 +5202,7 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
         err = mdbx_msync(&env->me_dxb_mmap, paged_offset, paged_length, false);
       } else {
         MDBX_meta rollback = *head;
-        mdbx_meta_set_txnid(env, &rollback, 0);
+        mdbx_meta_set_txnid(env, &rollback, undo_txnid);
         rollback.mm_datasync_sign = MDBX_DATASIGN_WEAK;
         err = mdbx_pwrite(env->me_fd, &rollback, sizeof(MDBX_meta),
                           (uint8_t *)head - (uint8_t *)env->me_map);
@@ -5075,7 +5211,7 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
         return err;
 
       mdbx_invalidate_cache(env->me_map, pgno2bytes(env, NUM_METAS));
-      mdbx_ensure(env, 0 == mdbx_meta_txnid_fluid(env, head));
+      mdbx_ensure(env, undo_txnid == mdbx_meta_txnid_fluid(env, head));
       mdbx_ensure(env, 0 == mdbx_meta_eq_mask(env));
       continue;
     }
@@ -7273,9 +7409,11 @@ int mdbx_cursor_put(MDBX_cursor *mc, MDBX_val *key, MDBX_val *data,
           memcpy((char *)mp + mp->mp_upper + PAGEHDRSZ,
                  (char *)fp + fp->mp_upper + PAGEHDRSZ,
                  olddata.iov_len - fp->mp_upper - PAGEHDRSZ);
+          memcpy((char *)(&mp->mp_ptrs), (char *)(&fp->mp_ptrs),
+                 NUMKEYS(fp) * sizeof(mp->mp_ptrs[0]));
           for (i = 0; i < NUMKEYS(fp); i++) {
-            mdbx_cassert(mc, fp->mp_ptrs[i] + offset <= UINT16_MAX);
-            mp->mp_ptrs[i] = (indx_t)(fp->mp_ptrs[i] + offset);
+            mdbx_cassert(mc, mp->mp_ptrs[i] + offset <= UINT16_MAX);
+            mp->mp_ptrs[i] += (indx_t)offset;
           }
         }
       }
@@ -10349,13 +10487,13 @@ int mdbx_dbi_open_ex(MDBX_txn *txn, const char *table_name, unsigned user_flags,
         !strncmp(table_name, txn->mt_dbxs[scan].md_name.iov_base, len)) {
       *dbi = scan;
       rc = mdbx_dbi_bind(txn, scan, user_flags, keycmp, datacmp);
-      goto unlock_return_rc;
+      goto bailout;
     }
   }
 
   if (unlikely(slot >= env->me_maxdbs)) {
     rc = MDBX_DBS_FULL;
-    goto unlock_return_rc;
+    goto bailout;
   }
 
   unsigned dbflag = DB_FRESH | DB_VALID | DB_USRVALID;
@@ -10405,7 +10543,6 @@ int mdbx_dbi_open_ex(MDBX_txn *txn, const char *table_name, unsigned user_flags,
     *dbi = slot;
   }
 
-unlock_return_rc:
   mdbx_ensure(env, mdbx_fastmutex_release(&env->me_dbi_lock) == MDBX_SUCCESS);
   return rc;
 }

src/osal.h

@@ -103,24 +103,47 @@ typedef struct {
 typedef pthread_mutex_t mdbx_fastmutex_t;
 #endif /* Platform */
 
+/* *INDENT-OFF* */
+/* clang-format off */
+#if defined(HAVE_SYS_STAT_H) || __has_include(<sys/stat.h>)
+#include <sys/stat.h>
+#endif
+#if defined(HAVE_SYS_TYPES_H) || __has_include(<sys/types.h>)
+#include <sys/types.h>
+#endif
+#if defined(HAVE_SYS_FILE_H) || __has_include(<sys/file.h>)
+#include <sys/file.h>
+#endif
+/* *INDENT-ON* */
+/* clang-format on */
+
 #ifndef SSIZE_MAX
 #define SSIZE_MAX INTPTR_MAX
 #endif
 
-#ifdef HAVE_SYS_STAT_H
-#include <sys/stat.h>
-#endif
-#ifdef HAVE_SYS_TYPES_H
-#include <sys/types.h>
-#endif
-#ifdef HAVE_SYS_FILE_H
-#include <sys/file.h>
-#endif
+#if defined(i386) || defined(__386) || defined(__i386) || defined(__i386__) || \
+    defined(i486) || defined(__i486) || defined(__i486__) ||                   \
+    defined(i586) | defined(__i586) || defined(__i586__) || defined(i686) ||   \
+    defined(__i686) || defined(__i686__) || defined(_M_IX86) ||                \
+    defined(_X86_) || defined(__THW_INTEL__) || defined(__I86__) ||            \
+    defined(__INTEL__) || defined(__x86_64) || defined(__x86_64__) ||          \
+    defined(__amd64__) || defined(__amd64) || defined(_M_X64) ||               \
+    defined(_M_AMD64) || defined(__IA32__) || defined(__INTEL__)
+#ifndef __ia32__
+/* LY: define neutral __ia32__ for x86 and x86-64 archs */
+#define __ia32__ 1
+#endif /* __ia32__ */
+#if !defined(__amd64__) && (defined(__x86_64) || defined(__x86_64__) ||        \
+                            defined(__amd64) || defined(_M_X64))
+/* LY: define trusty __amd64__ for all AMD64/x86-64 arch */
+#define __amd64__ 1
+#endif /* __amd64__ */
+#endif /* all x86 */
 
 #if !defined(UNALIGNED_OK)
-#if defined(__i386) || defined(__x86_64__) || defined(_M_IX86) ||              \
-    defined(_M_X64) || defined(i386) || defined(_X86_) || defined(__i386__) || \
-    defined(_X86_64_)
+#if (defined(__ia32__) || defined(__e2k__) ||                                  \
+     defined(__ARM_FEATURE_UNALIGNED)) &&                                      \
+    !defined(__ALIGNED__)
 #define UNALIGNED_OK 1
 #else
 #define UNALIGNED_OK 0
@@ -135,26 +158,15 @@ typedef pthread_mutex_t mdbx_fastmutex_t;
 /*----------------------------------------------------------------------------*/
 /* Compiler's includes for builtins/intrinsics */
 
-#ifdef _MSC_VER
-
-#if _MSC_FULL_VER < 190024215
-#if _MSC_FULL_VER < 180040629 && defined(_M_IX86)
-#error Please use Visual Studio 2015 (MSC 19.0) or newer for 32-bit target.
-#else
-#pragma message(                                                               \
-    "It is recommended to use Visual Studio 2015 (MSC 19.0) or newer.")
-#endif
-#endif
-
+#if defined(_MSC_VER) || defined(__INTEL_COMPILER)
 #include <intrin.h>
-
 #elif __GNUC_PREREQ(4, 4) || defined(__clang__)
-#if defined(__i386__) || defined(__x86_64__)
-#include <cpuid.h>
+#if defined(__ia32__) || defined(__e2k__)
 #include <x86intrin.h>
-#endif
-#elif defined(__INTEL_COMPILER)
-#include <intrin.h>
+#endif /* __ia32__ */
+#if defined(__ia32__)
+#include <cpuid.h>
+#endif /* __ia32__ */
 #elif defined(__SUNPRO_C) || defined(__sun) || defined(sun)
 #include <mbarrier.h>
 #elif (defined(_HPUX_SOURCE) || defined(__hpux) || defined(__HP_aCC)) &&       \
@@ -173,8 +185,10 @@ typedef pthread_mutex_t mdbx_fastmutex_t;
 #pragma gcc_extensions
 #elif defined(__SNC__)
 /* Sony PS3 - troubles ? */
+#elif defined(__hppa__) || defined(__hppa)
+#include <machine/inline.h>
 #else
-#error Unknown C compiler, please use GNU C 5.x or newer
+#error Unsupported C compiler, please use GNU C 4.4 or newer
 #endif /* Compiler */
 
 /*----------------------------------------------------------------------------*/
@@ -183,45 +197,70 @@ typedef pthread_mutex_t mdbx_fastmutex_t;
 #if !defined(__BYTE_ORDER__) || !defined(__ORDER_LITTLE_ENDIAN__) ||           \
     !defined(__ORDER_BIG_ENDIAN__)
 
-#if defined(HAVE_ENDIAN_H)
+/* *INDENT-OFF* */
+/* clang-format off */
+#if defined(__GLIBC__) || defined(__GNU_LIBRARY__) || defined(__ANDROID__) ||  \
+    defined(HAVE_ENDIAN_H) || __has_include(<endian.h>)
 #include <endian.h>
-#elif defined(HAVE_SYS_PARAM_H)
-#include <sys/param.h> /* for endianness */
-#elif defined(HAVE_NETINET_IN_H) && defined(HAVE_RESOLV_H)
-#include <netinet/in.h>
-#include <resolv.h> /* defines BYTE_ORDER on HPUX and Solaris */
-#endif
+#elif defined(__APPLE__) || defined(__MACH__) || defined(__OpenBSD__) ||       \
+    defined(HAVE_MACHINE_ENDIAN_H) || __has_include(<machine/endian.h>)
+#include <machine/endian.h>
+#elif defined(HAVE_SYS_ISA_DEFS_H) || __has_include(<sys/isa_defs.h>)
+#include <sys/isa_defs.h>
+#elif (defined(HAVE_SYS_TYPES_H) && defined(HAVE_SYS_ENDIAN_H)) ||             \
+    (__has_include(<sys/types.h>) && __has_include(<sys/endian.h>))
+#include <sys/endian.h>
+#include <sys/types.h>
+#elif defined(__bsdi__) || defined(__DragonFly__) || defined(__FreeBSD__) ||   \
+    defined(__NETBSD__) || defined(__NetBSD__) ||                              \
+    defined(HAVE_SYS_PARAM_H) || __has_include(<sys/param.h>)
+#include <sys/param.h>
+#endif /* OS */
+/* *INDENT-ON* */
+/* clang-format on */
 
 #if defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && defined(__BIG_ENDIAN)
 #define __ORDER_LITTLE_ENDIAN__ __LITTLE_ENDIAN
 #define __ORDER_BIG_ENDIAN__ __BIG_ENDIAN
 #define __BYTE_ORDER__ __BYTE_ORDER
+#elif defined(_BYTE_ORDER) && defined(_LITTLE_ENDIAN) && defined(_BIG_ENDIAN)
+#define __ORDER_LITTLE_ENDIAN__ _LITTLE_ENDIAN
+#define __ORDER_BIG_ENDIAN__ _BIG_ENDIAN
+#define __BYTE_ORDER__ _BYTE_ORDER
 #else
 #define __ORDER_LITTLE_ENDIAN__ 1234
 #define __ORDER_BIG_ENDIAN__ 4321
-#if defined(__LITTLE_ENDIAN__) || defined(_LITTLE_ENDIAN) ||                   \
+
+#if defined(__LITTLE_ENDIAN__) ||                                              \
+    (defined(_LITTLE_ENDIAN) && !defined(_BIG_ENDIAN)) ||                      \
     defined(__ARMEL__) || defined(__THUMBEL__) || defined(__AARCH64EL__) ||    \
     defined(__MIPSEL__) || defined(_MIPSEL) || defined(__MIPSEL) ||            \
-    defined(__i386) || defined(__x86_64__) || defined(_M_IX86) ||              \
-    defined(_M_X64) || defined(i386) || defined(_X86_) || defined(__i386__) || \
-    defined(_X86_64_) || defined(_M_ARM) || defined(_M_ARM64) ||               \
-    defined(__e2k__)
+    defined(_M_ARM) || defined(_M_ARM64) || defined(__e2k__) ||                \
+    defined(__elbrus_4c__) || defined(__elbrus_8c__) || defined(__bfin__) ||   \
+    defined(__BFIN__) || defined(__ia64__) || defined(_IA64) ||                \
+    defined(__IA64__) || defined(__ia64) || defined(_M_IA64) ||                \
+    defined(__itanium__) || defined(__ia32__) || defined(__CYGWIN__) ||        \
+    defined(_WIN64) || defined(_WIN32) || defined(__TOS_WIN__) ||              \
+    defined(__WINDOWS__)
 #define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
-#elif defined(__BIG_ENDIAN__) || defined(_BIG_ENDIAN) || defined(__ARMEB__) || \
-    defined(__THUMBEB__) || defined(__AARCH64EB__) || defined(__MIPSEB__) ||   \
-    defined(_MIPSEB) || defined(__MIPSEB) || defined(_M_IA64)
+
+#elif defined(__BIG_ENDIAN__) ||                                               \
+    (defined(_BIG_ENDIAN) && !defined(_LITTLE_ENDIAN)) ||                      \
+    defined(__ARMEB__) || defined(__THUMBEB__) || defined(__AARCH64EB__) ||    \
+    defined(__MIPSEB__) || defined(_MIPSEB) || defined(__MIPSEB) ||            \
+    defined(__m68k__) || defined(M68000) || defined(__hppa__) ||               \
+    defined(__hppa) || defined(__HPPA__) || defined(__sparc__) ||              \
+    defined(__sparc) || defined(__370__) || defined(__THW_370__) ||            \
+    defined(__s390__) || defined(__s390x__) || defined(__SYSC_ZARCH__)
 #define __BYTE_ORDER__ __ORDER_BIG_ENDIAN__
+
 #else
 #error __BYTE_ORDER__ should be defined.
-#endif
+#endif /* Arch */
+
 #endif
 #endif /* __BYTE_ORDER__ || __ORDER_LITTLE_ENDIAN__ || __ORDER_BIG_ENDIAN__ */
 
-#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__ &&                               \
-    __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__
-#error Unsupported byte order.
-#endif
-
 /*----------------------------------------------------------------------------*/
 /* Memory/Compiler barriers, cache coherence */
 
@@ -286,17 +325,14 @@ static __inline void mdbx_memory_barrier(void) {
 /*----------------------------------------------------------------------------*/
 /* Cache coherence and invalidation */
 
-#if defined(__i386__) || defined(__x86_64__) || defined(_M_AMD64) ||           \
-    defined(_M_IX86) || defined(__i386) || defined(__amd64) ||                 \
-    defined(i386) || defined(__x86_64) || defined(_AMD64_) || defined(_M_X64)
-#define MDBX_CACHE_IS_COHERENT 1
-#elif defined(__hppa) || defined(__hppa__)
-#define MDBX_CACHE_IS_COHERENT 1
-#endif
-
 #ifndef MDBX_CACHE_IS_COHERENT
+#if defined(__ia32__) || defined(__e2k__) || defined(__hppa) ||                \
+    defined(__hppa__)
+#define MDBX_CACHE_IS_COHERENT 1
+#else
 #define MDBX_CACHE_IS_COHERENT 0
 #endif
+#endif /* MDBX_CACHE_IS_COHERENT */
 
 #ifndef MDBX_CACHELINE_SIZE
 #if defined(SYSTEM_CACHE_ALIGNMENT_SIZE)
@@ -308,29 +344,29 @@ static __inline void mdbx_memory_barrier(void) {
 #endif
 #endif /* MDBX_CACHELINE_SIZE */
 
-#ifndef __cache_aligned
-#define __cache_aligned __aligned(MDBX_CACHELINE_SIZE)
-#endif
-
 #if MDBX_CACHE_IS_COHERENT
 #define mdbx_coherent_barrier() mdbx_compiler_barrier()
 #else
 #define mdbx_coherent_barrier() mdbx_memory_barrier()
 #endif
 
-#if defined(__mips) && defined(__linux)
+#if defined(__mips) || defined(__mips__) || defined(__mips64) ||               \
+    defined(__mips64) || defined(_M_MRX000) || defined(_MIPS_)
 /* Only MIPS has explicit cache control */
-#include <asm/cachectl.h>
+#include <sys/cachectl.h>
 #endif
 
 static __inline void mdbx_invalidate_cache(void *addr, size_t nbytes) {
   mdbx_coherent_barrier();
-#if defined(__mips) && defined(__linux)
+#if defined(__mips) || defined(__mips__) || defined(__mips64) ||               \
+    defined(__mips64) || defined(_M_MRX000) || defined(_MIPS_)
+#if defined(DCACHE)
   /* MIPS has cache coherency issues.
    * Note: for any nbytes >= on-chip cache size, entire is flushed. */
   cacheflush(addr, nbytes, DCACHE);
-#elif defined(_M_MRX000) || defined(_MIPS_)
+#else
 #error "Sorry, cacheflush() for MIPS not implemented"
+#endif /* __mips__ */
 #else
   /* LY: assume no relevant mmap/dcache issues. */
   (void)addr;
@@ -504,7 +540,13 @@ void mdbx_osal_jitter(bool tiny);
 #else
 #define MDBX_OSAL_LOCK pthread_mutex_t
 #define MDBX_OSAL_LOCK_SIGN UINT32_C(0x8017)
-#endif
+#endif /* MDBX_OSAL_LOCK */
+
+#ifdef MDBX_OSAL_LOCK
+#define MDBX_OSAL_LOCK_SIZE sizeof(MDBX_OSAL_LOCK)
+#else
+#define MDBX_OSAL_LOCK_SIZE 0
+#endif /* MDBX_OSAL_LOCK_SIZE */
 
 int mdbx_lck_init(MDBX_env *env);
 
@@ -516,8 +558,8 @@ void mdbx_lck_destroy(MDBX_env *env);
 int mdbx_rdt_lock(MDBX_env *env);
 void mdbx_rdt_unlock(MDBX_env *env);
 
-int mdbx_txn_lock(MDBX_env *env, bool dontwait);
-void mdbx_txn_unlock(MDBX_env *env);
+LIBMDBX_API int mdbx_txn_lock(MDBX_env *env, bool dontwait);
+LIBMDBX_API void mdbx_txn_unlock(MDBX_env *env);
 
 int mdbx_rpid_set(MDBX_env *env);
 int mdbx_rpid_clear(MDBX_env *env);

src/tools/mdbx_chk.c

@@ -77,7 +77,7 @@ int exclusive = 2;
 int envflags = MDBX_RDONLY;
 
 MDBX_env *env;
-MDBX_txn *txn, *locktxn;
+MDBX_txn *txn;
 MDBX_envinfo envinfo;
 MDBX_stat envstat;
 size_t maxkeysize, userdb_count, skipped_subdb;
@@ -763,6 +763,7 @@ int main(int argc, char *argv[]) {
   char *envname;
   int problems_maindb = 0, problems_freedb = 0, problems_meta = 0;
   int dont_traversal = 0;
+  bool locked = false;
 
   double elapsed;
 #if defined(_WIN32) || defined(_WIN64)
@@ -836,8 +837,8 @@ int main(int argc, char *argv[]) {
 #endif /* !WINDOWS */
 
   envname = argv[optind];
-  print("Running mdbx_chk for '%s' in %s mode...\n", envname,
-        (envflags & MDBX_RDONLY) ? "read-only" : "write-lock");
+  print("Running mdbx_chk for 'read-%s' in %s mode...\n", envname,
+        (envflags & MDBX_RDONLY) ? "read" : "write");
   fflush(NULL);
 
   rc = mdbx_env_create(&env);
@@ -862,13 +863,19 @@ int main(int argc, char *argv[]) {
   if (verbose)
     print(" - %s mode\n", exclusive ? "monopolistic" : "cooperative");
 
-  if (!(envflags & MDBX_RDONLY)) {
-    rc = mdbx_txn_begin(env, NULL, 0, &locktxn);
-    if (rc) {
-      error("mdbx_txn_begin(lock-write) failed, error %d %s\n", rc,
-            mdbx_strerror(rc));
+  if ((envflags & MDBX_RDONLY) == 0) {
+    rc = mdbx_txn_lock(env, false);
+    if (rc != MDBX_SUCCESS) {
+      error("mdbx_txn_lock failed, error %d %s\n", rc, mdbx_strerror(rc));
       goto bailout;
     }
+    locked = true;
+  }
+
+  rc = mdbx_txn_begin(env, NULL, MDBX_RDONLY, &txn);
+  if (rc) {
+    error("mdbx_txn_begin() failed, error %d %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
   }
 
   rc = mdbx_env_get_maxkeysize(env);
@@ -879,13 +886,6 @@ int main(int argc, char *argv[]) {
   }
   maxkeysize = rc;
 
-  rc = mdbx_txn_begin(env, NULL, MDBX_RDONLY, &txn);
-  if (rc) {
-    error("mdbx_txn_begin(read-only) failed, error %d %s\n", rc,
-          mdbx_strerror(rc));
-    goto bailout;
-  }
-
   rc = mdbx_env_info(env, &envinfo, sizeof(envinfo));
   if (rc) {
     error("mdbx_env_info failed, error %d %s\n", rc, mdbx_strerror(rc));
@@ -950,14 +950,14 @@ int main(int argc, char *argv[]) {
     if (verbose)
       print(" - performs full check recent-txn-id with meta-pages\n");
     problems_meta += check_meta_head(true);
-  } else if (locktxn) {
+  } else if (locked) {
     if (verbose)
       print(" - performs lite check recent-txn-id with meta-pages (not a "
             "monopolistic mode)\n");
     problems_meta += check_meta_head(false);
   } else if (verbose) {
     print(" - skip check recent-txn-id with meta-pages (monopolistic or "
-          "write-lock mode only)\n");
+          "read-write mode only)\n");
   }
 
   if (!dont_traversal) {
@@ -1079,7 +1079,7 @@ int main(int argc, char *argv[]) {
   }
 
   if (problems_maindb == 0 && problems_freedb == 0) {
-    if (!dont_traversal && (exclusive || locktxn)) {
+    if (!dont_traversal && (exclusive || (envflags & MDBX_RDONLY) == 0)) {
       if (walk.pgcount != lastpgno - freedb_pages) {
         error("used pages mismatch (%" PRIu64 " != %" PRIu64 ")\n",
               walk.pgcount, lastpgno - freedb_pages);
@@ -1090,7 +1090,7 @@ int main(int argc, char *argv[]) {
       }
     } else if (verbose) {
       print(" - skip check used and gc pages (btree-traversal with "
-            "monopolistic or write-lock mode only)\n");
+            "monopolistic or read-write mode only)\n");
     }
 
     if (!process_db(MAIN_DBI, NULL, handle_maindb, true)) {
@@ -1102,8 +1102,8 @@ int main(int argc, char *argv[]) {
 bailout:
   if (txn)
     mdbx_txn_abort(txn);
-  if (locktxn)
-    mdbx_txn_abort(locktxn);
+  if (locked)
+    mdbx_txn_unlock(env);
   if (env)
     mdbx_env_close(env);
   fflush(NULL);

test/base.h

@@ -76,14 +76,6 @@
 #include <unordered_set>
 #include <vector>
 
-#ifdef _MSC_VER
-#include <intrin.h>
-#endif
-
-#if defined(__i386__) || defined(__x86_64__)
-#include <x86intrin.h>
-#endif
-
 #include "../mdbx.h"
 #include "../src/defs.h"
 #include "../src/osal.h"

test/config.cc

@@ -278,8 +278,6 @@ void dump(const char *title) {
   logging::local_suffix indent(title);
 
   for (auto i = global::actors.begin(); i != global::actors.end(); ++i) {
-    const std::string tableid =
-        i->space_id ? "MAINDB" : ("SUB#" + std::to_string(i->space_id));
     log_info("#%u, testcase %s, space_id/table %u\n", i->actor_id,
              testcase2str(i->testcase), i->space_id);
     indent.push();
@@ -326,6 +324,11 @@ void dump(const char *title) {
     else
       log_info("no-delay\n");
 
+    if (i->params.inject_writefaultn)
+      log_info("inject-writefault on %u ops\n", i->params.inject_writefaultn);
+    else
+      log_info("no-inject-writefault\n");
+
     log_info("limits: readers %u, tables %u\n", i->params.max_readers,
              i->params.max_tables);
 

test/config.h

@@ -218,6 +218,7 @@ struct actor_params_pod {
 
   unsigned delaystart;
   unsigned waitfor_nops;
+  unsigned inject_writefaultn;
 
   unsigned max_readers;
   unsigned max_tables;
@@ -244,7 +245,7 @@ void dump(const char *title = "config-dump: ");
 struct actor_params : public config::actor_params_pod {
   std::string pathname_log;
   std::string pathname_db;
-  void set_defaults(void);
+  void set_defaults(const std::string &tmpdir);
 };
 
 struct actor_config : public config::actor_config_pod {

test/log.cc

@@ -285,3 +285,5 @@ void log_trouble(const char *where, const char *what, int errnum) {
 bool log_enabled(const logging::loglevel priority) {
   return (priority >= logging::level);
 }
+
+void log_flush(void) { fflushall(); }

test/log.h

@@ -81,6 +81,7 @@ void __printf_args(1, 2) log_warning(const char *msg, ...);
 void __printf_args(1, 2) log_error(const char *msg, ...);
 
 void log_trouble(const char *where, const char *what, int errnum);
+void log_flush(void);
 bool log_enabled(const logging::loglevel priority);
 
 #ifdef _DEBUG

test/main.cc

@@ -22,21 +22,16 @@ void __noreturn usage(void) {
 
 //-----------------------------------------------------------------------------
 
-void actor_params::set_defaults(void) {
+void actor_params::set_defaults(const std::string &tmpdir) {
   pathname_log = "";
   loglevel =
 #ifdef NDEBUG
-      logging::notice;
+      logging::info;
 #else
       logging::trace;
 #endif
 
-  pathname_db =
-#ifdef __linux__
-      "/dev/shm/test_tmpdb.mdbx";
-#else
-      "test_tmpdb.mdbx";
-#endif
+  pathname_db = tmpdir + "mdbx-test.db";
   mode_flags = MDBX_NOSUBDIR | MDBX_WRITEMAP | MDBX_MAPASYNC | MDBX_NORDAHEAD |
                MDBX_NOMEMINIT | MDBX_COALESCE | MDBX_LIFORECLAIM;
   table_flags = MDBX_DUPSORT;
@@ -65,6 +60,7 @@ void actor_params::set_defaults(void) {
 
   delaystart = 0;
   waitfor_nops = 0;
+  inject_writefaultn = 0;
 
   drop_table = false;
 
@@ -135,7 +131,7 @@ int main(int argc, char *const argv[]) {
                : EXIT_FAILURE;
 
   actor_params params;
-  params.set_defaults();
+  params.set_defaults(osal_tempdir());
   global::config::dump_config = true;
   logging::setup((logging::loglevel)params.loglevel, "main");
   unsigned last_space_id = 0;
@@ -219,6 +215,9 @@ int main(int argc, char *const argv[]) {
     if (config::parse_option(argc, argv, narg, "wait4ops", params.waitfor_nops,
                              config::decimal))
       continue;
+    if (config::parse_option(argc, argv, narg, "inject-writefault",
+                             params.inject_writefaultn, config::decimal))
+      continue;
     if (config::parse_option(argc, argv, narg, "drop", params.drop_table))
       continue;
     if (config::parse_option(argc, argv, narg, "dump-config",

test/osal-unix.cc

@@ -272,3 +272,22 @@ void osal_udelay(unsigned us) {
 }
 
 bool osal_istty(int fd) { return isatty(fd) == 1; }
+
+std::string osal_tempdir(void) {
+  const char *tempdir = getenv("TMPDIR");
+  if (!tempdir)
+    tempdir = getenv("TMP");
+  if (!tempdir)
+    tempdir = getenv("TEMPDIR");
+  if (!tempdir)
+    tempdir = getenv("TEMP");
+  if (tempdir) {
+    std::string dir(tempdir);
+    if (!dir.empty() && dir.at(dir.length() - 1) != '/')
+      dir.append("/");
+    return dir;
+  }
+  if (access("/dev/shm/", R_OK | W_OK | X_OK) == 0)
+    return "/dev/shm/";
+  return "";
+}

test/osal-windows.cc

@@ -305,3 +305,9 @@ void osal_udelay(unsigned us) {
 }
 
 bool osal_istty(int fd) { return _isatty(fd) != 0; }
+
+std::string osal_tempdir(void) {
+  char buf[MAX_PATH + 1];
+  DWORD len = GetTempPathA(sizeof(buf), buf);
+  return std::string(buf, len);
+}

test/osal.h

@@ -31,6 +31,7 @@ int osal_delay(unsigned seconds);
 void osal_udelay(unsigned us);
 void osal_yield(void);
 bool osal_istty(int fd);
+std::string osal_tempdir(void);
 
 #ifdef _MSC_VER
 #ifndef STDIN_FILENO

test/test.cc

@@ -123,7 +123,7 @@ void testcase::db_prepare() {
   if (config.params.loglevel <= logging::verbose)
     mdbx_dbg_opts |= MDBX_DBG_PRINT;
   int rc = mdbx_setup_debug(mdbx_dbg_opts, mdbx_logger);
-  log_info("set mdbx debug-opts: 0x%02x", rc);
+  log_trace("set mdbx debug-opts: 0x%02x", rc);
 
   MDBX_env *env = nullptr;
   rc = mdbx_env_create(&env);
@@ -204,6 +204,7 @@ void testcase::txn_end(bool abort) {
     if (unlikely(rc != MDBX_SUCCESS))
       failure_perror("mdbx_txn_abort()", rc);
   } else {
+    txn_inject_writefault(txn);
     int rc = mdbx_txn_commit(txn);
     if (unlikely(rc != MDBX_SUCCESS))
       failure_perror("mdbx_txn_commit()", rc);
@@ -218,6 +219,27 @@ void testcase::txn_restart(bool abort, bool readonly, unsigned flags) {
   txn_begin(readonly, flags);
 }
 
+void testcase::txn_inject_writefault(void) {
+  if (txn_guard)
+    txn_inject_writefault(txn_guard.get());
+}
+
+void testcase::txn_inject_writefault(MDBX_txn *txn) {
+  if (config.params.inject_writefaultn && txn) {
+    if (config.params.inject_writefaultn <= nops_completed &&
+        (mdbx_txn_flags(txn) & MDBX_RDONLY) == 0) {
+      log_info("== txn_inject_writefault(): got %u nops or more, inject FAULT",
+               config.params.inject_writefaultn);
+      log_flush();
+#if defined(_WIN32) || defined(_WIN64) || defined(_WINDOWS)
+      TerminateProcess(GetCurrentProcess(), 42);
+#else
+      raise(SIGKILL);
+#endif
+    }
+  }
+}
+
 bool testcase::wait4start() {
   if (config.wait4id) {
     log_trace(">> wait4start(%u)", config.wait4id);

test/test.h

@@ -107,6 +107,8 @@ protected:
   void txn_begin(bool readonly, unsigned flags = 0);
   void txn_end(bool abort);
   void txn_restart(bool abort, bool readonly, unsigned flags = 0);
+  void txn_inject_writefault(void);
+  void txn_inject_writefault(MDBX_txn *txn);
   void fetch_canary();
   void update_canary(uint64_t increment);
   void kick_progress(bool active) const;

test/utils.cc

@@ -14,44 +14,10 @@
 
 #include "test.h"
 #include <float.h>
-#ifdef HAVE_IEEE754_H
+#if defined(HAVE_IEEE754_H) || __has_include(<ieee754.h>)
 #include <ieee754.h>
 #endif
 
-/* Compiler's includes for builtins/intrinsics */
-#if defined(_MSC_VER) || defined(__INTEL_COMPILER)
-#include <intrin.h>
-#elif __GNUC_PREREQ(4, 4) || defined(__clang__)
-#if defined(__ia32__) || defined(__e2k__)
-#include <x86intrin.h>
-#endif /* __ia32__ */
-#if defined(__ia32__)
-#include <cpuid.h>
-#endif /* __ia32__ */
-#elif defined(__SUNPRO_C) || defined(__sun) || defined(sun)
-#include <mbarrier.h>
-#elif (defined(_HPUX_SOURCE) || defined(__hpux) || defined(__HP_aCC)) &&       \
-    (defined(HP_IA64) || defined(__ia64))
-#include <machine/sys/inline.h>
-#elif defined(__IBMC__) && defined(__powerpc)
-#include <atomic.h>
-#elif defined(_AIX)
-#include <builtins.h>
-#include <sys/atomic_op.h>
-#elif (defined(__osf__) && defined(__DECC)) || defined(__alpha)
-#include <c_asm.h>
-#include <machine/builtins.h>
-#elif defined(__MWERKS__)
-/* CodeWarrior - troubles ? */
-#pragma gcc_extensions
-#elif defined(__SNC__)
-/* Sony PS3 - troubles ? */
-#elif defined(__hppa__) || defined(__hppa)
-#include <machine/inline.h>
-#else
-#error Unsupported C compiler, please use GNU C 4.4 or newer
-#endif /* Compiler */
-
 std::string format(const char *fmt, ...) {
   va_list ap, ones;
   va_start(ap, fmt);
@@ -127,8 +93,22 @@ bool hex2data(const char *hex_begin, const char *hex_end, void *ptr,
 
 //-----------------------------------------------------------------------------
 
+/* TODO: replace my 'libmera' fomr t1ha. */
 uint64_t entropy_ticks(void) {
+#if defined(EMSCRIPTEN)
+  return (uint64_t)emscripten_get_now();
+#endif /* EMSCRIPTEN */
+
+#if defined(__APPLE__) || defined(__MACH__)
+  return mach_absolute_time();
+#endif /* defined(__APPLE__) || defined(__MACH__) */
+
+#if defined(__sun__) || defined(__sun)
+  return gethrtime();
+#endif /* __sun__ */
+
 #if defined(__GNUC__) || defined(__clang__)
+
 #if defined(__ia64__)
   uint64_t ticks;
   __asm __volatile("mov %0=ar.itc" : "=r"(ticks));
@@ -145,29 +125,81 @@ uint64_t entropy_ticks(void) {
   uint64_t ticks;
   __asm __volatile("rpcc %0" : "=r"(ticks));
   return ticks;
-#elif defined(__sparc_v9__)
-  uint64_t ticks;
-  __asm __volatile("rd %%tick, %0" : "=r"(ticks));
-  return ticks;
-#elif defined(__powerpc64__) || defined(__ppc64__)
+#elif defined(__sparc__) || defined(__sparc) || defined(__sparc64__) ||        \
+    defined(__sparc64) || defined(__sparc_v8plus__) ||                         \
+    defined(__sparc_v8plus) || defined(__sparc_v8plusa__) ||                   \
+    defined(__sparc_v8plusa) || defined(__sparc_v9__) || defined(__sparc_v9)
+
+  union {
+    uint64_t u64;
+    struct {
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+      uint32_t h, l;
+#else
+      uint32_t l, h;
+#endif
+    } u32;
+  } cycles;
+
+#if defined(__sparc_v8plus__) || defined(__sparc_v8plusa__) ||                 \
+    defined(__sparc_v9__) || defined(__sparc_v8plus) ||                        \
+    defined(__sparc_v8plusa) || defined(__sparc_v9)
+
+#if UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul ||                  \
+    defined(__sparc64__) || defined(__sparc64)
+  __asm __volatile("rd %%tick, %0" : "=r"(cycles.u64));
+#else
+  __asm __volatile("rd %%tick, %1; srlx %1, 32, %0"
+                   : "=r"(cycles.u32.h), "=r"(cycles.u32.l));
+#endif /* __sparc64__ */
+
+#else
+  __asm __volatile(".byte 0x83, 0x41, 0x00, 0x00; mov %%g1, %0"
+                   : "=r"(cycles.u64)
+                   :
+                   : "%g1");
+#endif /* __sparc8plus__ || __sparc_v9__ */
+  return cycles.u64;
+
+#elif (defined(__powerpc64__) || defined(__ppc64__) || defined(__ppc64) ||     \
+       defined(__powerpc64))
   uint64_t ticks;
   __asm __volatile("mfspr %0, 268" : "=r"(ticks));
   return ticks;
-#elif defined(__ppc__) || defined(__powerpc__)
-  unsigned tbl, tbu;
-
-  /* LY: Here not a problem if a high-part (tbu)
-   * would been updated during reading. */
-  __asm __volatile("mftb %0" : "=r"(tbl));
-  __asm __volatile("mftbu %0" : "=r"(tbu));
-
-  return (((uin64_t)tbu0) << 32) | tbl;
+#elif (defined(__powerpc__) || defined(__ppc__) || defined(__powerpc) ||       \
+       defined(__ppc))
+#if UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul
+  uint64_t ticks;
+  __asm __volatile("mftb  %0" : "=r"(ticks));
+  *now = ticks;
+#else
+  uint64_t ticks;
+  uint32_t low, high_before, high_after;
+  __asm __volatile("mftbu %0; mftb  %1; mftbu %2"
+                   : "=r"(high_before), "=r"(low), "=r"(high_after));
+  ticks = (uint64_t)high_after << 32;
+  ticks |= low & /* zeroes if high part has changed */
+           ~(high_before - high_after);
+#endif
+#elif defined(__aarch64__) || (defined(__ARM_ARCH) && __ARM_ARCH > 7)
+  uint64_t virtual_timer;
+  __asm __volatile("mrs %0, cntvct_el0" : "=r"(virtual_timer));
+  return virtual_timer;
+#elif defined(__ARM_ARCH) && __ARM_ARCH > 5 && __ARM_ARCH < 8
+  unsigned long pmccntr;
+  __asm __volatile("mrc p15, 0, %0, c9, c13, 0" : "=r"(pmccntr));
+  return pmccntr;
+#elif defined(__mips__) || defined(__mips) || defined(_R4000)
+  unsigned count;
+  __asm __volatile("rdhwr %0, $2" : "=r"(count));
+  return count;
 #endif /* arch selector */
 #endif /* __GNUC__ || __clang__ */
 
-#if defined(__e2k__) || defined(__elbrus__) || defined(_M_IX86) ||             \
-    defined(_M_X64) || defined(__x86_64__) || defined(__i386__)
+#if defined(__e2k__) || defined(__ia32__)
   return __rdtsc();
+#elif defined(_M_ARM)
+  return __rdpmccntr64();
 #elif defined(_WIN32) || defined(_WIN64) || defined(_WINDOWS)
   LARGE_INTEGER PerformanceCount;
   if (QueryPerformanceCounter(&PerformanceCount))

test/utils.h

@@ -17,33 +17,8 @@
 
 #if !defined(__BYTE_ORDER__) || !defined(__ORDER_LITTLE_ENDIAN__) ||           \
     !defined(__ORDER_BIG_ENDIAN__)
-#ifndef _MSC_VER
-#include <sys/param.h> /* for endianness */
-#endif
-#if defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && defined(__BIG_ENDIAN)
-#define __ORDER_LITTLE_ENDIAN__ __LITTLE_ENDIAN
-#define __ORDER_BIG_ENDIAN__ __BIG_ENDIAN
-#define __BYTE_ORDER__ __BYTE_ORDER
-#else
-#define __ORDER_LITTLE_ENDIAN__ 1234
-#define __ORDER_BIG_ENDIAN__ 4321
-#if defined(__LITTLE_ENDIAN__) || defined(_LITTLE_ENDIAN) ||                   \
-    defined(__ARMEL__) || defined(__THUMBEL__) || defined(__AARCH64EL__) ||    \
-    defined(__MIPSEL__) || defined(_MIPSEL) || defined(__MIPSEL) ||            \
-    defined(__i386) || defined(__x86_64__) || defined(_M_IX86) ||              \
-    defined(_M_X64) || defined(i386) || defined(_X86_) || defined(__i386__) || \
-    defined(_X86_64_) || defined(_M_ARM) || defined(_M_ARM64) ||               \
-    defined(__e2k__)
-#define __BYTE_ORDER__ __ORDER_LITTLE_ENDIAN__
-#elif defined(__BIG_ENDIAN__) || defined(_BIG_ENDIAN) || defined(__ARMEB__) || \
-    defined(__THUMBEB__) || defined(__AARCH64EB__) || defined(__MIPSEB__) ||   \
-    defined(_MIPSEB) || defined(__MIPSEB) || defined(_M_IA64)
-#define __BYTE_ORDER__ __ORDER_BIG_ENDIAN__
-#else
 #error __BYTE_ORDER__ should be defined.
 #endif
-#endif
-#endif
 
 #if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__ &&                               \
     __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__
@@ -274,7 +249,7 @@ static __inline void memory_barrier(void) {
 #elif defined(__INTEL_COMPILER) /* LY: Intel Compiler may mimic GCC and MSC */
 #if defined(__ia64__) || defined(__ia64) || defined(_M_IA64)
   __mf();
-#elif defined(__i386__) || defined(__x86_64__)
+#elif defined(__ia32__)
   _mm_mfence();
 #else
 #error "Unknown target for Intel Compiler, please report to us."
@@ -293,8 +268,7 @@ static __inline void memory_barrier(void) {
 }
 
 static __inline void cpu_relax() {
-#if defined(__i386__) || defined(__x86_64__) || defined(_M_IX86) ||            \
-    defined(_M_X64)
+#if defined(__ia32__)
   _mm_pause();
 #elif defined(_WIN32) || defined(_WIN64) || defined(_WINDOWS) ||               \
     defined(YieldProcessor)