mdbx: backport - fix pwrite() for WRITE_MAX.

Change-Id: If4924d20c1e267c2d3a190c860b89fc2fda0d517

mdbx.h

@@ -610,7 +610,7 @@ LIBMDBX_API int mdbx_env_create(MDBX_env **penv);
  *  - MDBX_NOTLS
  *      Don't use Thread-Local Storage. Tie reader locktable slots to
  *      MDBX_txn objects instead of to threads. I.e. mdbx_txn_reset() keeps
- *      the slot reseved for the MDBX_txn object. A thread may use parallel
+ *      the slot reserved for the MDBX_txn object. A thread may use parallel
  *      read-only transactions. A read-only transaction may span threads if
  *      the user synchronizes its use. Applications that multiplex many
  *      user threads over individual OS threads need this option. Such an

src/lck-posix.c

@@ -20,8 +20,8 @@
 #ifndef MDBX_USE_ROBUST
 /* Howard Chu: Android currently lacks Robust Mutex support */
 #if defined(EOWNERDEAD) &&                                                     \
-    !defined(ANDROID) /* LY: glibc before 2.10 has a troubles with Robust      \
+    !defined(__ANDROID__) /* LY: glibc before 2.10 has a troubles              \
-                         Mutex too. */                                         \
+                                 with Robust Mutex too. */                     \
     && __GLIBC_PREREQ(2, 10)
 #define MDBX_USE_ROBUST 1
 #else

src/mdbx.c

@@ -3755,6 +3755,7 @@ static int mdbx_prep_backlog(MDBX_txn *txn, MDBX_cursor *mc) {
   const int extra = mdbx_backlog_extragap(txn->mt_env);
 
   if (mdbx_backlog_size(txn) < mc->mc_db->md_depth + extra) {
+    mc->mc_flags &= ~C_RECLAIMING;
     int rc = mdbx_cursor_touch(mc);
     if (unlikely(rc))
       return rc;
@@ -3768,6 +3769,7 @@ static int mdbx_prep_backlog(MDBX_txn *txn, MDBX_cursor *mc) {
         break;
       }
     }
+    mc->mc_flags |= C_RECLAIMING;
   }
 
   return MDBX_SUCCESS;
@@ -3880,6 +3882,7 @@ static int mdbx_update_gc(MDBX_txn *txn) {
   if (unlikely(rc != MDBX_SUCCESS))
     goto bailout_notracking;
 
+  mc.mc_flags |= C_RECLAIMING;
   mc.mc_next = txn->mt_cursors[FREE_DBI];
   txn->mt_cursors[FREE_DBI] = &mc;
 
@@ -3925,9 +3928,7 @@ retry:
           mdbx_tassert(txn, cleaned_gc_id < *env->me_oldest);
           mdbx_trace("%s.cleanup-reclaimed-id [%u]%" PRIaTXN, dbg_prefix_mode,
                      cleaned_gc_slot, cleaned_gc_id);
-          mc.mc_flags |= C_RECLAIMING;
           rc = mdbx_cursor_del(&mc, 0);
-          mc.mc_flags ^= C_RECLAIMING;
           if (unlikely(rc != MDBX_SUCCESS))
             goto bailout;
         } while (cleaned_gc_slot < MDBX_PNL_SIZE(txn->mt_lifo_reclaimed));
@@ -3949,9 +3950,7 @@ retry:
         mdbx_tassert(txn, cleaned_gc_id < *env->me_oldest);
         mdbx_trace("%s.cleanup-reclaimed-id %" PRIaTXN, dbg_prefix_mode,
                    cleaned_gc_id);
-        mc.mc_flags |= C_RECLAIMING;
         rc = mdbx_cursor_del(&mc, 0);
-        mc.mc_flags ^= C_RECLAIMING;
         if (unlikely(rc != MDBX_SUCCESS))
           goto bailout;
         settled = 0;
@@ -4104,7 +4103,9 @@ retry:
     if (befree_stored < MDBX_PNL_SIZE(txn->mt_befree_pages)) {
       if (unlikely(!befree_stored)) {
         /* Make sure last page of freeDB is touched and on befree-list */
+        mc.mc_flags &= ~C_RECLAIMING;
         rc = mdbx_page_search(&mc, NULL, MDBX_PS_LAST | MDBX_PS_MODIFY);
+        mc.mc_flags |= C_RECLAIMING;
         if (unlikely(rc != MDBX_SUCCESS && rc != MDBX_NOTFOUND))
           goto bailout;
       }
@@ -4198,7 +4199,9 @@ retry:
                   reused_gc_slot) *
                      env->me_maxgc_ov1page) {
         /* LY: need just a txn-id for save page list. */
+        mc.mc_flags &= ~C_RECLAIMING;
         rc = mdbx_page_alloc(&mc, 0, NULL, MDBX_ALLOC_GC | MDBX_ALLOC_KICK);
+        mc.mc_flags |= C_RECLAIMING;
         if (likely(rc == MDBX_SUCCESS)) {
           /* LY: ok, reclaimed from freedb. */
           mdbx_trace("%s: took @%" PRIaTXN " from GC, continue",
@@ -4412,7 +4415,7 @@ retry:
       key.iov_len = sizeof(fill_gc_id);
 
       mdbx_tassert(txn, data.iov_len >= sizeof(pgno_t) * 2);
-      mc.mc_flags |= C_RECLAIMING | C_GCFREEZE;
+      mc.mc_flags |= C_GCFREEZE;
       unsigned chunk = (unsigned)(data.iov_len / sizeof(pgno_t)) - 1;
       if (unlikely(chunk > left)) {
         mdbx_trace("%s: chunk %u > left %u, @%" PRIaTXN, dbg_prefix_mode, chunk,
@@ -4420,12 +4423,12 @@ retry:
         if (loop < 5 || chunk - left > env->me_maxgc_ov1page) {
           data.iov_len = (left + 1) * sizeof(pgno_t);
           if (loop < 21)
-            mc.mc_flags -= C_GCFREEZE;
+            mc.mc_flags &= ~C_GCFREEZE;
         }
         chunk = left;
       }
       rc = mdbx_cursor_put(&mc, &key, &data, MDBX_CURRENT | MDBX_RESERVE);
-      mc.mc_flags &= ~(C_RECLAIMING | C_GCFREEZE);
+      mc.mc_flags &= ~C_GCFREEZE;
       if (unlikely(rc != MDBX_SUCCESS))
         goto bailout;
 
@@ -5127,7 +5130,7 @@ static int __cold mdbx_read_header(MDBX_env *env, MDBX_meta *meta,
 
     /* LY: check and silently put mm_geo.now into [geo.lower...geo.upper].
      *
-     * Copy-with-compaction by previous version of libmfbx could produce DB-file
+     * Copy-with-compaction by previous version of libmdbx could produce DB-file
      * less than meta.geo.lower bound, in case actual filling is low or no data
      * at all. This is not a problem as there is no damage or loss of data.
      * Therefore it is better not to consider such situation as an error, but
@@ -8300,7 +8303,7 @@ int mdbx_cursor_put(MDBX_cursor *mc, MDBX_val *key, MDBX_val *data,
         if (rc > 0) {
           rc = MDBX_NOTFOUND;
           mc->mc_ki[mc->mc_top]++;
-        } else {
+        } else if (unlikely(rc < 0 || (flags & MDBX_APPENDDUP) == 0)) {
           /* new key is <= last key */
           rc = MDBX_EKEYMISMATCH;
         }
@@ -10882,7 +10885,7 @@ static int mdbx_page_split(MDBX_cursor *mc, const MDBX_val *newkey,
        * This yields better packing during sequential inserts.
        */
       int dir;
-      if (nkeys < 20 || nsize > pmax / 16 || newindx >= nkeys) {
+      if (nkeys < 32 || nsize > pmax / 16 || newindx >= nkeys) {
         /* Find split point */
         psize = 0;
         if (newindx <= split_indx || newindx >= nkeys) {

src/osal.c

@@ -581,17 +581,25 @@ int mdbx_pwrite(mdbx_filehandle_t fd, const void *buf, size_t bytes,
     return (bytes == written) ? MDBX_SUCCESS : MDBX_EIO /* ERROR_WRITE_FAULT */;
   return GetLastError();
 #else
-  int rc;
+  while (true) {
-  intptr_t written;
-  do {
     STATIC_ASSERT_MSG(sizeof(off_t) >= sizeof(size_t),
                       "libmdbx requires 64-bit file I/O on 64-bit systems");
-    written = pwrite(fd, buf, bytes, offset);
+    const intptr_t written =
+        pwrite(fd, buf, (bytes <= MAX_WRITE) ? bytes : MAX_WRITE, offset);
     if (likely(bytes == (size_t)written))
       return MDBX_SUCCESS;
-    rc = errno;
+    if (written < 0) {
-  } while (rc == EINTR);
+      const int rc = errno;
-  return (written < 0) ? rc : MDBX_EIO /* Use which error code (ENOSPC)? */;
+      if (rc != EINTR)
+        return rc;
+    } else if (written > 0) {
+      bytes -= written;
+      offset += written;
+      buf = (char *)buf + written;
+    } else {
+      return -1;
+    }
+  }
 #endif
 }
 

src/version.c

@@ -18,8 +18,8 @@
 #error "API version mismatch!"
 #endif
 
-#define MDBX_VERSION_RELEASE 0
+#define MDBX_VERSION_RELEASE 2
-#define MDBX_VERSION_REVISION 2
+#define MDBX_VERSION_REVISION 0
 
 /*LIBMDBX_EXPORTS*/ const mdbx_version_info mdbx_version = {
     MDBX_VERSION_MAJOR,
@@ -30,5 +30,5 @@
      "@MDBX_GIT_DESCRIBE@"}};
 
 /*LIBMDBX_EXPORTS*/ const mdbx_build_info mdbx_build = {
-    "@MDBX_BUILD_TIMESTAMP@", "@MDBX_BUILD_TAGRET@", "@MDBX_BUILD_OPTIONS@",
+    "@MDBX_BUILD_TIMESTAMP@", "@MDBX_BUILD_TARGET@", "@MDBX_BUILD_OPTIONS@",
     "@MDBX_BUILD_COMPILER@", "@MDBX_BUILD_FLAGS@"};

test/main.cc

@@ -182,11 +182,11 @@ int main(int argc, char *const argv[]) {
                              mdbx_limits_dbsize_min(params.pagesize),
                              mdbx_limits_dbsize_max(params.pagesize)))
       continue;
-    if (config::parse_option(argc, argv, narg, "size", params.size_now,
+    if (config::parse_option(argc, argv, narg, "size-upper", params.size_upper,
                              mdbx_limits_dbsize_min(params.pagesize),
                              mdbx_limits_dbsize_max(params.pagesize)))
       continue;
-    if (config::parse_option(argc, argv, narg, "size-upper", params.size_upper,
+    if (config::parse_option(argc, argv, narg, "size", params.size_now,
                              mdbx_limits_dbsize_min(params.pagesize),
                              mdbx_limits_dbsize_max(params.pagesize)))
       continue;