mdbx: add MDBX_ACCEDE environment opening flag.

Change-Id: If0a08d6fce127f35ff2992988715b7dc1fdb70a9

mdbx.h

@@ -921,6 +921,22 @@ LIBMDBX_API const char *mdbx_dump_val(const MDBX_val *key, char *const buf,
  * This flag affects only at environment opening but can't be changed after. */
 #define MDBX_EXCLUSIVE 0x400000u
 
+/* MDBX_ACCEDE = using database which already opened by another process(es).
+ *
+ * The MDBX_ACCEDE flag avoid MDBX_INCOMPATIBLE error while opening If the
+ * database is already used by another process(es) and environment mode/flags
+ * isn't compatible. In such cases, when using the MDBX_ACCEDE flag, instead of
+ * the specified incompatible options, the mode in which the database is already
+ * opened by other processes will be used, including MDBX_LIFORECLAIM,
+ * MDBX_COALESCE and MDBX_NORDAHEAD. The MDBX_ACCEDE flag is useful to open a
+ * database that already used by another process(es) and used mode/flags isn't
+ * known.
+ *
+ * MDBX_ACCEDE has no effect if the current process is the only one either
+ * opening the DB in read-only mode or other process(es) uses the DB in
+ * read-only mode. */
+#define MDBX_ACCEDE 0x40000000u
+
 /* MDBX_WRITEMAP = map data into memory with write permission.
  *
  * Use a writeable memory map unless MDBX_RDONLY is set. This uses fewer mallocs

src/elements/core.c

@@ -4666,6 +4666,8 @@ static int mdbx_txn_renew0(MDBX_txn *txn, unsigned flags) {
   STATIC_ASSERT(offsetof(MDBX_lockinfo, mti_readers) % MDBX_CACHELINE_SIZE ==
                 0);
 
+  mdbx_assert(env, (flags & ~(MDBX_TXN_BEGIN_FLAGS | MDBX_TXN_SPILLS |
+                              MDBX_WRITEMAP)) == 0);
   if (flags & MDBX_RDONLY) {
     txn->mt_flags = MDBX_RDONLY | (env->me_flags & MDBX_NOTLS);
     MDBX_reader *r = txn->to.reader;
@@ -5018,8 +5020,7 @@ int mdbx_txn_begin(MDBX_env *env, MDBX_txn *parent, unsigned flags,
       return MDBX_EPERM;
 #endif /* Windows */
 
-    flags |= parent->mt_flags &
+    flags |= parent->mt_flags & (MDBX_TXN_BEGIN_FLAGS | MDBX_TXN_SPILLS);
-             (MDBX_TXN_BEGIN_FLAGS | MDBX_SHRINK_ALLOWED | MDBX_TXN_SPILLS);
     /* Child txns save MDBX_pgstate and use own copy of cursors */
     size = env->me_maxdbs * (sizeof(MDBX_db) + sizeof(MDBX_cursor *) + 1);
     size += tsize = sizeof(MDBX_txn);
@@ -6515,7 +6516,7 @@ int mdbx_txn_commit(MDBX_txn *txn) {
 
     parent->mt_geo = txn->mt_geo;
     parent->mt_canary = txn->mt_canary;
-    parent->mt_flags = txn->mt_flags;
+    parent->mt_flags = txn->mt_flags | (parent->mt_flags & MDBX_SHRINK_ALLOWED);
 
     /* Merge our cursors into parent's and close them */
     mdbx_cursors_eot(txn, 1);
@@ -8660,7 +8661,7 @@ __cold int mdbx_is_readahead_reasonable(size_t volume, intptr_t redundancy) {
  * environment and re-opening it with the new flags. */
 #define CHANGEABLE                                                             \
   (MDBX_NOSYNC | MDBX_NOMETASYNC | MDBX_MAPASYNC | MDBX_NOMEMINIT |            \
-   MDBX_COALESCE | MDBX_PAGEPERTURB)
+   MDBX_COALESCE | MDBX_PAGEPERTURB | MDBX_ACCEDE)
 #define CHANGELESS                                                             \
   (MDBX_NOSUBDIR | MDBX_RDONLY | MDBX_WRITEMAP | MDBX_NOTLS | MDBX_NORDAHEAD | \
    MDBX_LIFORECLAIM | MDBX_EXCLUSIVE)
@@ -8711,18 +8712,22 @@ int __cold mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
     /* LY: silently ignore irrelevant flags when
      * we're only getting read access */
     flags &= ~(MDBX_WRITEMAP | MDBX_MAPASYNC | MDBX_NOSYNC | MDBX_NOMETASYNC |
-               MDBX_COALESCE | MDBX_LIFORECLAIM | MDBX_NOMEMINIT);
+               MDBX_COALESCE | MDBX_LIFORECLAIM | MDBX_NOMEMINIT | MDBX_ACCEDE);
   } else {
 #ifdef __OpenBSD__
     /* Temporary `workaround` for OpenBSD kernel's bug.
      * See https://github.com/leo-yuriev/libmdbx/issues/67 */
     if ((flags & MDBX_WRITEMAP) == 0) {
+      if (flags & MDBX_ACCEDE)
+        flags |= MDBX_WRITEMAP;
+      else {
         mdbx_debug_log(MDBX_LOG_ERROR, __func__, __LINE__,
                        "OpenBSD requires MDBX_WRITEMAP because of an internal "
                        "bug(s) in a file/buffer/page cache.\n");
         rc = 42 /* ENOPROTOOPT */;
         goto bailout;
       }
+    }
 #endif /* __OpenBSD__ */
     env->me_dirtylist = mdbx_calloc(MDBX_DPL_TXNFULL + 1, sizeof(MDBX_DP));
     if (!env->me_dirtylist)
@@ -8793,6 +8798,36 @@ int __cold mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
     goto bailout;
   }
 
+  const unsigned rigorous_flags = MDBX_WRITEMAP | MDBX_NOSYNC | MDBX_MAPASYNC;
+  const unsigned mode_flags = rigorous_flags | MDBX_NOMETASYNC |
+                              MDBX_LIFORECLAIM | MDBX_COALESCE | MDBX_NORDAHEAD;
+
+  if (env->me_lck && lck_rc != MDBX_RESULT_TRUE &&
+      (env->me_flags & MDBX_RDONLY) == 0) {
+    while (env->me_lck->mti_envmode == MDBX_RDONLY) {
+      if (atomic_cas32(&env->me_lck->mti_envmode, MDBX_RDONLY,
+                       env->me_flags & mode_flags))
+        break;
+      atomic_yield();
+    }
+
+    if (env->me_flags & MDBX_ACCEDE) {
+      /* pickup current mode-flags, including MDBX_LIFORECLAIM |
+       * MDBX_COALESCE | MDBX_NORDAHEAD */
+      const unsigned diff =
+          (env->me_lck->mti_envmode ^ env->me_flags) & mode_flags;
+      mdbx_notice("accede mode-flags: 0x%X, 0x%X -> 0x%X", diff, env->me_flags,
+                  env->me_flags ^ diff);
+      env->me_flags ^= diff;
+    }
+
+    if ((env->me_lck->mti_envmode ^ env->me_flags) & rigorous_flags) {
+      mdbx_error("%s", "current mode/flags incompatible with requested");
+      rc = MDBX_INCOMPATIBLE;
+      goto bailout;
+    }
+  }
+
   const int dxb_rc = mdbx_setup_dxb(env, lck_rc);
   if (MDBX_IS_ERROR(dxb_rc)) {
     rc = dxb_rc;
@@ -8801,8 +8836,6 @@ int __cold mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
 
   mdbx_debug("opened dbenv %p", (void *)env);
   if (env->me_lck) {
-    const unsigned mode_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);
       rc = mdbx_lck_downgrade(env);
@@ -8814,19 +8847,6 @@ int __cold mdbx_env_open(MDBX_env *env, const char *path, unsigned flags,
       rc = mdbx_reader_check0(env, false, NULL);
       if (MDBX_IS_ERROR(rc))
         goto bailout;
-      if ((env->me_flags & MDBX_RDONLY) == 0) {
-        while (env->me_lck->mti_envmode == MDBX_RDONLY) {
-          if (atomic_cas32(&env->me_lck->mti_envmode, MDBX_RDONLY,
-                           env->me_flags & mode_flags))
-            break;
-          atomic_yield();
-        }
-        if ((env->me_lck->mti_envmode ^ env->me_flags) & mode_flags) {
-          mdbx_error("%s", "current mode/flags incompatible with requested");
-          rc = MDBX_INCOMPATIBLE;
-          goto bailout;
-        }
-      }
     }
 
     if ((env->me_flags & MDBX_NOTLS) == 0) {

src/tools/mdbx_chk.c

@@ -968,7 +968,7 @@ int main(int argc, char *argv[]) {
       envflags &= ~MDBX_RDONLY;
       break;
     case 'c':
-      envflags &= ~MDBX_EXCLUSIVE;
+      envflags = (envflags & ~MDBX_EXCLUSIVE) | MDBX_ACCEDE;
       break;
     case 'd':
       dont_traversal = true;
@@ -1031,7 +1031,7 @@ int main(int argc, char *argv[]) {
 #endif
        )) {
     envflags &= ~MDBX_EXCLUSIVE;
-    rc = mdbx_env_open(env, envname, envflags, 0664);
+    rc = mdbx_env_open(env, envname, envflags | MDBX_ACCEDE, 0664);
   }
 
   if (rc) {

test/config.cc

@@ -294,14 +294,21 @@ bool parse_option(int argc, char *const argv[], int &narg, const char *option,
 
 //-----------------------------------------------------------------------------
 
-const struct option_verb mode_bits[] = {
+const struct option_verb mode_bits[] = {{"rdonly", MDBX_RDONLY},
-    {"rdonly", MDBX_RDONLY},          {"mapasync", MDBX_MAPASYNC},
+                                        {"mapasync", MDBX_MAPASYNC},
-    {"utterly", MDBX_UTTERLY_NOSYNC}, {"nosubdir", MDBX_NOSUBDIR},
+                                        {"utterly", MDBX_UTTERLY_NOSYNC},
-    {"nosync", MDBX_NOSYNC},          {"nometasync", MDBX_NOMETASYNC},
+                                        {"nosubdir", MDBX_NOSUBDIR},
-    {"writemap", MDBX_WRITEMAP},      {"notls", MDBX_NOTLS},
+                                        {"nosync", MDBX_NOSYNC},
-    {"nordahead", MDBX_NORDAHEAD},    {"nomeminit", MDBX_NOMEMINIT},
+                                        {"nometasync", MDBX_NOMETASYNC},
-    {"coalesce", MDBX_COALESCE},      {"lifo", MDBX_LIFORECLAIM},
+                                        {"writemap", MDBX_WRITEMAP},
-    {"perturb", MDBX_PAGEPERTURB},    {nullptr, 0}};
+                                        {"notls", MDBX_NOTLS},
+                                        {"nordahead", MDBX_NORDAHEAD},
+                                        {"nomeminit", MDBX_NOMEMINIT},
+                                        {"coalesce", MDBX_COALESCE},
+                                        {"lifo", MDBX_LIFORECLAIM},
+                                        {"perturb", MDBX_PAGEPERTURB},
+                                        {"accede", MDBX_ACCEDE},
+                                        {nullptr, 0}};
 
 const struct option_verb table_bits[] = {
     {"key.reverse", MDBX_REVERSEKEY},

test/main.cc

@@ -124,7 +124,7 @@ void actor_params::set_defaults(const std::string &tmpdir) {
 
   pathname_db = tmpdir + "mdbx-test.db";
   mode_flags = MDBX_NOSUBDIR | MDBX_WRITEMAP | MDBX_MAPASYNC | MDBX_NOMEMINIT |
-               MDBX_COALESCE | MDBX_LIFORECLAIM;
+               MDBX_COALESCE | MDBX_LIFORECLAIM | MDBX_ACCEDE;
   table_flags = MDBX_DUPSORT;
 
   size_lower = -1;