mdbx: add MDBX_EXCLUSIVE instead of mdbx_env_open_ex().

Change-Id: I3c817d58d971385bcd07643df14beaf7122c7836

README-RU.md

@@ -591,8 +591,8 @@ _libmdbx_ при этом не ведет WAL, а передает весь ко
 13. Исправленный вариант `mdbx_cursor_count()`, возвращающий корректное
 количество дубликатов для всех типов таблиц и любого положения курсора.
 
-14. Возможность открыть БД в эксклюзивном режиме посредством
+14. Возможность открыть БД в эксклюзивном режиме посредством флага
-`mdbx_env_open_ex()`, например в целях её проверки.
+`MDBX_EXCLUSIVE`, например в целях её проверки.
 
 15. Возможность закрыть БД в "грязном" состоянии (без сброса данных и
 формирования сильной точки фиксации) посредством `mdbx_env_close_ex()`.

README.md

@@ -405,7 +405,7 @@ Improvements over LMDB
 
 13. Fixed `mdbx_cursor_count()`, which returns correct count of duplicated for all table types and any cursor position.
 
-14. Ability to open DB in exclusive mode via `mdbx_env_open_ex()`, e.g. for integrity check.
+14. Ability to open DB in exclusive mode with `MDBX_EXCLUSIVE` flag, e.g. for integrity check.
 
 15. Ability to close DB in "dirty" state (without data flush and creation of steady synchronization point)
     via `mdbx_env_close_ex()`.

mdbx.h

@@ -288,9 +288,8 @@ typedef int(MDBX_cmp_func)(const MDBX_val *a, const MDBX_val *b);
 #define MDBX_MAPASYNC 0x100000u
 /* tie reader locktable slots to MDBX_txn objects instead of to threads */
 #define MDBX_NOTLS 0x200000u
-/* don't do any locking, caller must manage their own locks
+/* open DB in exclusive/monopolistic mode. */
- * WARNING: libmdbx don't support this mode. */
+#define MDBX_EXCLUSIVE 0x400000u
-#define MDBX_NOLOCK__UNSUPPORTED 0x400000u
 /* don't do readahead */
 #define MDBX_NORDAHEAD 0x800000u
 /* don't initialize malloc'd memory before writing to datafile */
@@ -670,8 +669,6 @@ LIBMDBX_API int mdbx_env_create(MDBX_env **penv);
  *   - MDBX_EAGAIN   - the environment was locked by another process. */
 LIBMDBX_API int mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
                               mode_t mode);
-LIBMDBX_API int mdbx_env_open_ex(MDBX_env *env, const char *path,
-                                 unsigned flags, mode_t mode, int *exclusive);
 
 /* Copy an MDBX environment to the specified path, with options.
  *

src/lck-windows.c

@@ -132,7 +132,8 @@ int mdbx_txn_lock(MDBX_env *env, bool dontwait) {
     EnterCriticalSection(&env->me_windowsbug_lock);
   }
 
-  if (flock(env->me_fd,
+  if ((env->me_flags & MDBX_EXCLUSIVE) ||
+      flock(env->me_fd,
             dontwait ? (LCK_EXCLUSIVE | LCK_DONTWAIT)
                      : (LCK_EXCLUSIVE | LCK_WAITFOR),
             LCK_BODY))
@@ -143,7 +144,8 @@ int mdbx_txn_lock(MDBX_env *env, bool dontwait) {
 }
 
 void mdbx_txn_unlock(MDBX_env *env) {
-  int rc = funlock(env->me_fd, LCK_BODY);
+  int rc = (env->me_flags & MDBX_EXCLUSIVE) ? TRUE
+                                            : funlock(env->me_fd, LCK_BODY);
   LeaveCriticalSection(&env->me_windowsbug_lock);
   if (!rc)
     mdbx_panic("%s failed: errcode %u", mdbx_func_, GetLastError());
@@ -166,7 +168,8 @@ int mdbx_rdt_lock(MDBX_env *env) {
     return MDBX_SUCCESS; /* readonly database in readonly filesystem */
 
   /* transite from S-? (used) to S-E (locked), e.g. exclusive lock upper-part */
-  if (flock(env->me_lfd, LCK_EXCLUSIVE | LCK_WAITFOR, LCK_UPPER))
+  if ((env->me_flags & MDBX_EXCLUSIVE) ||
+      flock(env->me_lfd, LCK_EXCLUSIVE | LCK_WAITFOR, LCK_UPPER))
     return MDBX_SUCCESS;
 
   int rc = GetLastError();
@@ -177,7 +180,8 @@ int mdbx_rdt_lock(MDBX_env *env) {
 void mdbx_rdt_unlock(MDBX_env *env) {
   if (env->me_lfd != INVALID_HANDLE_VALUE) {
     /* transite from S-E (locked) to S-? (used), e.g. unlock upper-part */
-    if (!funlock(env->me_lfd, LCK_UPPER))
+    if ((env->me_flags & MDBX_EXCLUSIVE) == 0 &&
+        !funlock(env->me_lfd, LCK_UPPER))
       mdbx_panic("%s failed: errcode %u", mdbx_func_, GetLastError());
   }
   mdbx_srwlock_ReleaseShared(&env->me_remap_guard);
@@ -372,6 +376,9 @@ int mdbx_lck_seize(MDBX_env *env) {
   int rc;
 
   assert(env->me_fd != INVALID_HANDLE_VALUE);
+  if (env->me_flags & MDBX_EXCLUSIVE)
+    return MDBX_RESULT_TRUE /* files were must be opened non-shareable */;
+
   if (env->me_lfd == INVALID_HANDLE_VALUE) {
     /* LY: without-lck mode (e.g. on read-only filesystem) */
     mdbx_jitter4testing(false);
@@ -414,6 +421,9 @@ int mdbx_lck_downgrade(MDBX_env *env, bool complete) {
   assert(env->me_fd != INVALID_HANDLE_VALUE);
   assert(env->me_lfd != INVALID_HANDLE_VALUE);
 
+  if (env->me_flags & MDBX_EXCLUSIVE)
+    return MDBX_SUCCESS /* files were must be opened non-shareable */;
+
   /* 1) must be at E-E (exclusive-write) */
   if (!complete) {
     /* transite from E-E to E_? (exclusive-read) */
@@ -448,6 +458,10 @@ int mdbx_lck_upgrade(MDBX_env *env) {
   /* Transite from locked state (S-E) to exclusive-write (E-E) */
   assert(env->me_fd != INVALID_HANDLE_VALUE);
   assert(env->me_lfd != INVALID_HANDLE_VALUE);
+  assert((env->me_flags & MDBX_EXCLUSIVE) == 0);
+
+  if (env->me_flags & MDBX_EXCLUSIVE)
+    return MDBX_RESULT_TRUE /* files were must be opened non-shareable */;
 
   /* 1) must be at S-E (locked), transite to ?_E (middle) */
   if (!funlock(env->me_lfd, LCK_LOWER))

src/mdbx.c

@@ -5583,7 +5583,8 @@ static int __cold mdbx_setup_lck(MDBX_env *env, char *lck_pathname,
   assert(env->me_fd != INVALID_HANDLE_VALUE);
   assert(env->me_lfd == INVALID_HANDLE_VALUE);
 
-  int err = mdbx_openfile(lck_pathname, O_RDWR | O_CREAT, mode, &env->me_lfd);
+  int err = mdbx_openfile(lck_pathname, O_RDWR | O_CREAT, mode, &env->me_lfd,
+                          (env->me_flags & MDBX_EXCLUSIVE) ? true : false);
   if (err != MDBX_SUCCESS) {
     if (err != MDBX_EROFS || (env->me_flags & MDBX_RDONLY) == 0)
       return err;
@@ -5626,10 +5627,14 @@ static int __cold mdbx_setup_lck(MDBX_env *env, char *lck_pathname,
         return err;
       size = wanna;
     }
-  } else if (size > SSIZE_MAX || (size & (env->me_os_psize - 1)) ||
+  } else {
-             size < env->me_os_psize) {
+    if (env->me_flags & MDBX_EXCLUSIVE)
-    mdbx_notice("lck-file has invalid size %" PRIu64 " bytes", size);
+      return MDBX_BUSY;
-    return MDBX_PROBLEM;
+    if (size > SSIZE_MAX || (size & (env->me_os_psize - 1)) ||
+        size < env->me_os_psize) {
+      mdbx_notice("lck-file has invalid size %" PRIu64 " bytes", size);
+      return MDBX_PROBLEM;
+    }
   }
 
   const size_t maxreaders =
@@ -5699,14 +5704,14 @@ static int __cold mdbx_setup_lck(MDBX_env *env, char *lck_pathname,
    MDBX_COALESCE | MDBX_PAGEPERTURB)
 #define CHANGELESS                                                             \
   (MDBX_NOSUBDIR | MDBX_RDONLY | MDBX_WRITEMAP | MDBX_NOTLS | MDBX_NORDAHEAD | \
-   MDBX_LIFORECLAIM)
+   MDBX_LIFORECLAIM | MDBX_EXCLUSIVE)
 
 #if VALID_FLAGS & PERSISTENT_FLAGS & (CHANGEABLE | CHANGELESS)
 #error "Persistent DB flags & env flags overlap, but both go in mm_flags"
 #endif
 
-int __cold mdbx_env_open_ex(MDBX_env *env, const char *path, unsigned flags,
+int __cold mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
-                            mode_t mode, int *exclusive) {
+                         mode_t mode) {
   if (unlikely(!env || !path))
     return MDBX_EINVAL;
 
@@ -5770,7 +5775,8 @@ int __cold mdbx_env_open_ex(MDBX_env *env, const char *path, unsigned flags,
   else
     oflags = O_RDWR | O_CREAT;
 
-  rc = mdbx_openfile(dxb_pathname, oflags, mode, &env->me_fd);
+  rc = mdbx_openfile(dxb_pathname, oflags, mode, &env->me_fd,
+                     (env->me_flags & MDBX_EXCLUSIVE) ? true : false);
   if (rc != MDBX_SUCCESS)
     goto bailout;
 
@@ -5791,7 +5797,7 @@ int __cold mdbx_env_open_ex(MDBX_env *env, const char *path, unsigned flags,
       MDBX_WRITEMAP | MDBX_NOSYNC | MDBX_NOMETASYNC | MDBX_MAPASYNC;
   if (lck_rc == MDBX_RESULT_TRUE) {
     env->me_lck->mti_envmode = env->me_flags & (mode_flags | MDBX_RDONLY);
-    if (exclusive == NULL || *exclusive < 2) {
+    if ((env->me_flags & MDBX_EXCLUSIVE) == 0) {
       /* LY: downgrade lock only if exclusive access not requested.
        *     in case exclusive==1, just leave value as is. */
       rc = mdbx_lck_downgrade(env, true);
@@ -5803,10 +5809,6 @@ int __cold mdbx_env_open_ex(MDBX_env *env, const char *path, unsigned flags,
     if (rc != MDBX_SUCCESS)
       goto bailout;
   } else {
-    if (exclusive) {
-      /* LY: just indicate that is not an exclusive access. */
-      *exclusive = 0;
-    }
     if ((env->me_flags & MDBX_RDONLY) == 0) {
       while (env->me_lck->mti_envmode == MDBX_RDONLY) {
         if (mdbx_atomic_compare_and_swap32(&env->me_lck->mti_envmode,
@@ -5877,11 +5879,6 @@ bailout:
   return rc;
 }
 
-int __cold mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
-                         mode_t mode) {
-  return mdbx_env_open_ex(env, path, flags, mode, NULL);
-}
-
 /* Destroy resources from mdbx_env_open(), clear our readers & DBIs */
 static void __cold mdbx_env_close0(MDBX_env *env) {
   if (!(env->me_flags & MDBX_ENV_ACTIVE))
@@ -10496,8 +10493,8 @@ int __cold mdbx_env_copy(MDBX_env *env, const char *path, unsigned flags) {
   /* The destination path must exist, but the destination file must not.
    * We don't want the OS to cache the writes, since the source data is
    * already in the OS cache. */
-  int rc =
+  int rc = mdbx_openfile(lck_pathname, O_WRONLY | O_CREAT | O_EXCL, 0666,
-      mdbx_openfile(lck_pathname, O_WRONLY | O_CREAT | O_EXCL, 0666, &newfd);
+                         &newfd, true);
   if (rc == MDBX_SUCCESS) {
     if (env->me_psize >= env->me_os_psize) {
 #ifdef F_NOCACHE /* __APPLE__ */

src/osal.c

@@ -409,19 +409,20 @@ int mdbx_fastmutex_release(mdbx_fastmutex_t *fastmutex) {
 /*----------------------------------------------------------------------------*/
 
 int mdbx_openfile(const char *pathname, int flags, mode_t mode,
-                  mdbx_filehandle_t *fd) {
+                  mdbx_filehandle_t *fd, bool exclusive) {
   *fd = INVALID_HANDLE_VALUE;
 #if defined(_WIN32) || defined(_WIN64)
   (void)mode;
 
-  DWORD DesiredAccess;
+  DWORD DesiredAccess, ShareMode;
-  DWORD ShareMode = FILE_SHARE_READ | FILE_SHARE_WRITE;
   DWORD FlagsAndAttributes = FILE_ATTRIBUTE_NORMAL;
   switch (flags & (O_RDONLY | O_WRONLY | O_RDWR)) {
   default:
     return ERROR_INVALID_PARAMETER;
   case O_RDONLY:
     DesiredAccess = GENERIC_READ;
+    ShareMode =
+        exclusive ? FILE_SHARE_READ : (FILE_SHARE_READ | FILE_SHARE_WRITE);
     break;
   case O_WRONLY: /* assume for MDBX_env_copy() and friends output */
     DesiredAccess = GENERIC_WRITE;
@@ -430,6 +431,7 @@ int mdbx_openfile(const char *pathname, int flags, mode_t mode,
     break;
   case O_RDWR:
     DesiredAccess = GENERIC_READ | GENERIC_WRITE;
+    ShareMode = exclusive ? 0 : (FILE_SHARE_READ | FILE_SHARE_WRITE);
     break;
   }
 
@@ -468,7 +470,7 @@ int mdbx_openfile(const char *pathname, int flags, mode_t mode,
     }
   }
 #else
-
+  (void)exclusive;
 #ifdef O_CLOEXEC
   flags |= O_CLOEXEC;
 #endif

src/osal.h

@@ -477,7 +477,7 @@ int mdbx_filesize_sync(mdbx_filehandle_t fd);
 int mdbx_ftruncate(mdbx_filehandle_t fd, uint64_t length);
 int mdbx_filesize(mdbx_filehandle_t fd, uint64_t *length);
 int mdbx_openfile(const char *pathname, int flags, mode_t mode,
-                  mdbx_filehandle_t *fd);
+                  mdbx_filehandle_t *fd, bool exclusive);
 int mdbx_closefile(mdbx_filehandle_t fd);
 
 typedef struct mdbx_mmap_param {

src/tools/mdbx_chk.c

@@ -73,8 +73,7 @@ struct {
 } walk;
 
 uint64_t total_unused_bytes;
-int exclusive = 2;
+int envflags = MDBX_RDONLY | MDBX_EXCLUSIVE;
-int envflags = MDBX_RDONLY;
 
 MDBX_env *env;
 MDBX_txn *txn;
@@ -706,7 +705,7 @@ void verbose_meta(int num, txnid_t txnid, uint64_t sign) {
     print(", stay");
 
   if (txnid > envinfo.mi_recent_txnid &&
-      (exclusive || (envflags & MDBX_RDONLY) == 0))
+      (envflags & (MDBX_EXCLUSIVE | MDBX_RDONLY)) == MDBX_EXCLUSIVE)
     print(", rolled-back %" PRIu64 " (%" PRIu64 " >>> %" PRIu64 ")",
           txnid - envinfo.mi_recent_txnid, txnid, envinfo.mi_recent_txnid);
   print("\n");
@@ -805,7 +804,7 @@ int main(int argc, char *argv[]) {
       envflags &= ~MDBX_RDONLY;
       break;
     case 'c':
-      exclusive = 0;
+      envflags &= ~MDBX_EXCLUSIVE;
       break;
     case 'd':
       dont_traversal = 1;
@@ -853,7 +852,19 @@ int main(int argc, char *argv[]) {
     goto bailout;
   }
 
-  rc = mdbx_env_open_ex(env, envname, envflags, 0664, &exclusive);
+  rc = mdbx_env_open(env, envname, envflags, 0664);
+  if ((envflags & MDBX_EXCLUSIVE) &&
+      (rc == MDBX_BUSY ||
+#if defined(_WIN32) || defined(_WIN64)
+       rc == ERROR_LOCK_VIOLATION || rc == ERROR_SHARING_VIOLATION
+#else
+       rc == EBUSY
+#endif
+       )) {
+    envflags &= ~MDBX_EXCLUSIVE;
+    rc = mdbx_env_open(env, envname, envflags, 0664);
+  }
+
   if (rc) {
     error("mdbx_env_open failed, error %d %s\n", rc, mdbx_strerror(rc));
     if (rc == MDBX_WANNA_RECOVERY && (envflags & MDBX_RDONLY))
@@ -861,7 +872,8 @@ int main(int argc, char *argv[]) {
     goto bailout;
   }
   if (verbose)
-    print(" - %s mode\n", exclusive ? "monopolistic" : "cooperative");
+    print(" - %s mode\n",
+          (envflags & MDBX_EXCLUSIVE) ? "monopolistic" : "cooperative");
 
   if ((envflags & MDBX_RDONLY) == 0) {
     rc = mdbx_txn_lock(env, false);
@@ -946,7 +958,7 @@ int main(int argc, char *argv[]) {
     ++problems_meta;
   }
 
-  if (exclusive > 1) {
+  if (envflags & MDBX_EXCLUSIVE) {
     if (verbose)
       print(" - performs full check recent-txn-id with meta-pages\n");
     problems_meta += check_meta_head(true);
@@ -1079,7 +1091,8 @@ int main(int argc, char *argv[]) {
   }
 
   if (problems_maindb == 0 && problems_freedb == 0) {
-    if (!dont_traversal && (exclusive || (envflags & MDBX_RDONLY) == 0)) {
+    if (!dont_traversal &&
+        (envflags & (MDBX_EXCLUSIVE | MDBX_RDONLY)) == MDBX_EXCLUSIVE) {
       if (walk.pgcount != lastpgno - freedb_pages) {
         error("used pages mismatch (%" PRIu64 " != %" PRIu64 ")\n",
               walk.pgcount, lastpgno - freedb_pages);