merge with libmdbx

2 files changed, 109 insertions, 37 deletions

diff --git a/plugins/Dbx_mdbx/src/libmdbx/src/mdbx.c b/plugins/Dbx_mdbx/src/libmdbx/src/mdbx.c
index 2eafad6155..fd8c9a9bc7 100644
--- a/plugins/Dbx_mdbx/src/libmdbx/src/mdbx.c
+++ b/plugins/Dbx_mdbx/src/libmdbx/src/mdbx.c
@@ -2775,12 +2775,16 @@ int mdbx_txn_begin(MDBX_env *env, MDBX_txn *parent, unsigned flags,
     size = env->me_maxdbs * (sizeof(MDBX_db) + sizeof(MDBX_cursor *) + 1);
     size += tsize = sizeof(MDBX_ntxn);
   } else if (flags & MDBX_RDONLY) {
+    if (env->me_txn0 && unlikely(env->me_txn0->mt_owner == mdbx_thread_self()))
+      return MDBX_BUSY;
     size = env->me_maxdbs * (sizeof(MDBX_db) + 1);
     size += tsize = sizeof(MDBX_txn);
   } else {
     /* Reuse preallocated write txn. However, do not touch it until
      * mdbx_txn_renew0() succeeds, since it currently may be active. */
     txn = env->me_txn0;
+    if (unlikely(txn->mt_owner == mdbx_thread_self()))
+      return MDBX_BUSY;
     goto renew;
   }
   if (unlikely((txn = calloc(1, size)) == NULL)) {
@@ -3954,30 +3958,6 @@ static int __cold mdbx_read_header(MDBX_env *env, MDBX_meta *meta) {
       continue;
     }
 
-    /* LY: check mapsize limits */
-    const uint64_t mapsize_min =
-        page.mp_meta.mm_geo.lower * (uint64_t)page.mp_meta.mm_psize;
-    const uint64_t mapsize_max =
-        page.mp_meta.mm_geo.upper * (uint64_t)page.mp_meta.mm_psize;
-    STATIC_ASSERT(MAX_MAPSIZE < SSIZE_MAX - MAX_PAGESIZE);
-    STATIC_ASSERT(MIN_MAPSIZE < MAX_MAPSIZE);
-    if (mapsize_min < MIN_MAPSIZE || mapsize_max > MAX_MAPSIZE) {
-      mdbx_notice("meta[%u] has invalid min-mapsize (%" PRIu64 "), skip it",
-                  meta_number, mapsize_min);
-      rc = MDBX_VERSION_MISMATCH;
-      continue;
-    }
-
-    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) {
-      mdbx_notice("meta[%u] has too large max-mapsize (%" PRIu64 "), skip it",
-                  meta_number, mapsize_max);
-      rc = MDBX_TOO_LARGE;
-      continue;
-    }
-
     /* LY: check end_pgno */
     if (page.mp_meta.mm_geo.now < page.mp_meta.mm_geo.lower ||
         page.mp_meta.mm_geo.now > page.mp_meta.mm_geo.upper) {
@@ -3996,6 +3976,43 @@ static int __cold mdbx_read_header(MDBX_env *env, MDBX_meta *meta) {
       continue;
     }
 
+    /* LY: check mapsize limits */
+    const uint64_t mapsize_min =
+        page.mp_meta.mm_geo.lower * (uint64_t)page.mp_meta.mm_psize;
+    STATIC_ASSERT(MAX_MAPSIZE < SSIZE_MAX - MAX_PAGESIZE);
+    STATIC_ASSERT(MIN_MAPSIZE < MAX_MAPSIZE);
+    if (mapsize_min < MIN_MAPSIZE || mapsize_min > MAX_MAPSIZE) {
+      mdbx_notice("meta[%u] has invalid min-mapsize (%" PRIu64 "), skip it",
+                  meta_number, mapsize_min);
+      rc = MDBX_VERSION_MISMATCH;
+      continue;
+    }
+
+    const uint64_t mapsize_max =
+        page.mp_meta.mm_geo.upper * (uint64_t)page.mp_meta.mm_psize;
+    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) {
+      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 ||
+          used_bytes > MAX_MAPSIZE) {
+        mdbx_notice("meta[%u] has too large max-mapsize (%" PRIu64 "), skip it",
+                    meta_number, mapsize_max);
+        rc = MDBX_TOO_LARGE;
+        continue;
+      }
+
+      /* allow to open large DB from a 32-bit environment */
+      mdbx_notice("meta[%u] has too large max-mapsize (%" PRIu64 "), "
+                  "but size of used space still acceptable (%" PRIu64 ")",
+                  meta_number, mapsize_max, used_bytes);
+      page.mp_meta.mm_geo.upper = (pgno_t)(MAX_MAPSIZE / page.mp_meta.mm_psize);
+      if (page.mp_meta.mm_geo.now > page.mp_meta.mm_geo.upper)
+        page.mp_meta.mm_geo.now = page.mp_meta.mm_geo.upper;
+    }
+
     if (page.mp_meta.mm_geo.next > page.mp_meta.mm_geo.now) {
       mdbx_notice("meta[%u] next-pageno (%" PRIaPGNO
                   ") is beyond end-pgno (%" PRIaPGNO "), skip it",
@@ -4649,7 +4666,7 @@ LIBMDBX_API int mdbx_env_set_geometry(MDBX_env *env, intptr_t size_lower,
       goto bailout;
     }
     size_upper -= env->me_os_psize;
-    if ((size_t)size_upper > (size_t)size_lower)
+    if ((size_t)size_upper < (size_t)size_lower)
       size_lower = size_upper;
   }
   mdbx_assert(env, (size_upper - size_lower) % env->me_os_psize == 0);
@@ -4834,6 +4851,7 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
             meta.mm_txnid_a, mdbx_durable_str(&meta));
 
   mdbx_setup_pagesize(env, meta.mm_psize);
+  const size_t used_bytes = pgno2bytes(env, meta.mm_geo.next);
   if ((env->me_flags & MDBX_RDONLY) /* readonly */
       || lck_rc != MDBX_RESULT_TRUE /* not exclusive */) {
     /* use present params from db */
@@ -4849,9 +4867,6 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
     }
   } else if (env->me_dbgeo.now) {
     /* silently growth to last used page */
-    const size_t used_bytes = pgno2bytes(env, meta.mm_geo.next);
-    if (env->me_dbgeo.lower < used_bytes)
-      env->me_dbgeo.lower = used_bytes;
     if (env->me_dbgeo.now < used_bytes)
       env->me_dbgeo.now = used_bytes;
     if (env->me_dbgeo.upper < used_bytes)
@@ -4916,7 +4931,6 @@ static int __cold mdbx_setup_dxb(MDBX_env *env, int lck_rc) {
 
   const size_t expected_bytes =
       mdbx_roundup2(pgno2bytes(env, meta.mm_geo.now), env->me_os_psize);
-  const size_t used_bytes = pgno2bytes(env, meta.mm_geo.next);
   mdbx_ensure(env, expected_bytes >= used_bytes);
   if (filesize_before_mmap != expected_bytes) {
     if (lck_rc != /* lck exclusive */ MDBX_RESULT_TRUE) {
@@ -8082,7 +8096,10 @@ int mdbx_cursor_renew(MDBX_txn *txn, MDBX_cursor *mc) {
     return MDBX_EBADSIGN;
 
   if (unlikely(txn->mt_owner != mdbx_thread_self()))
-    return MDBX_THREAD_MISMATCH;
+    return txn->mt_owner ? MDBX_THREAD_MISMATCH : MDBX_BAD_TXN;
+
+  if (unlikely(txn->mt_flags & (MDBX_TXN_FINISHED | MDBX_TXN_ERROR)))
+    return MDBX_BAD_TXN;
 
   if (unlikely(mc->mc_signature != MDBX_MC_SIGNATURE &&
                mc->mc_signature != MDBX_MC_READY4CLOSE))
@@ -8180,7 +8197,12 @@ void mdbx_cursor_close(MDBX_cursor *mc) {
 MDBX_txn *mdbx_cursor_txn(MDBX_cursor *mc) {
   if (unlikely(!mc || mc->mc_signature != MDBX_MC_SIGNATURE))
     return NULL;
-  return mc->mc_txn;
+  MDBX_txn *txn = mc->mc_txn;
+  if (unlikely(!txn || txn->mt_signature != MDBX_MT_SIGNATURE))
+    return NULL;
+  if (unlikely(txn->mt_flags & MDBX_TXN_FINISHED))
+    return NULL;
+  return txn;
 }
 
 MDBX_dbi mdbx_cursor_dbi(MDBX_cursor *mc) {
diff --git a/plugins/Dbx_mdbx/src/libmdbx/src/osal.c b/plugins/Dbx_mdbx/src/libmdbx/src/osal.c
index fa6fe279da..af0b88ce88 100644
--- a/plugins/Dbx_mdbx/src/libmdbx/src/osal.c
+++ b/plugins/Dbx_mdbx/src/libmdbx/src/osal.c
@@ -99,12 +99,12 @@ extern NTSTATUS NTAPI NtUnmapViewOfSection(IN HANDLE ProcessHandle,
 extern NTSTATUS NTAPI NtClose(HANDLE Handle);
 
 extern NTSTATUS NTAPI NtAllocateVirtualMemory(
-    IN HANDLE ProcessHandle, IN OUT PVOID *BaseAddress, IN ULONG ZeroBits,
-    IN OUT PULONG RegionSize, IN ULONG AllocationType, IN ULONG Protect);
+    IN HANDLE ProcessHandle, IN OUT PVOID *BaseAddress, IN ULONG_PTR ZeroBits,
+    IN OUT PSIZE_T RegionSize, IN ULONG AllocationType, IN ULONG Protect);
 
 extern NTSTATUS NTAPI NtFreeVirtualMemory(IN HANDLE ProcessHandle,
                                           IN PVOID *BaseAddress,
-                                          IN OUT PULONG RegionSize,
+                                          IN OUT PSIZE_T RegionSize,
                                           IN ULONG FreeType);
 
 #ifndef FILE_PROVIDER_CURRENT_VERSION
@@ -870,7 +870,7 @@ int mdbx_mmap(int flags, mdbx_mmap_t *map, size_t size, size_t limit) {
     if (rc == MDBX_SUCCESS)
       map->filesize = size;
     /* ignore error, because Windows unable shrink file
-     * that already mapped (by another process) */;
+     * that already mapped (by another process) */
   }
 
   LARGE_INTEGER SectionSize;
@@ -931,6 +931,12 @@ int mdbx_munmap(mdbx_mmap_t *map) {
   NTSTATUS rc = NtUnmapViewOfSection(GetCurrentProcess(), map->address);
   if (!NT_SUCCESS(rc))
     ntstatus2errcode(rc);
+
+  if (map->filesize != map->current &&
+      mdbx_filesize(map->fd, &map->filesize) == MDBX_SUCCESS &&
+      map->filesize != map->current)
+    (void)mdbx_ftruncate(map->fd, map->current);
+
   map->length = 0;
   map->current = 0;
   map->address = nullptr;
@@ -961,8 +967,23 @@ int mdbx_mresize(int flags, mdbx_mmap_t *map, size_t size, size_t limit) {
     return ntstatus2errcode(status);
   }
 
+  if (limit > map->length) {
+    /* check ability of address space for growth before umnap */
+    PVOID BaseAddress = (PBYTE)map->address + map->length;
+    SIZE_T RegionSize = limit - map->length;
+    status = NtAllocateVirtualMemory(GetCurrentProcess(), &BaseAddress, 0,
+                                     &RegionSize, MEM_RESERVE, PAGE_NOACCESS);
+    if (!NT_SUCCESS(status))
+      return ntstatus2errcode(status);
+
+    status = NtFreeVirtualMemory(GetCurrentProcess(), &BaseAddress, &RegionSize,
+                                 MEM_RELEASE);
+    if (!NT_SUCCESS(status))
+      return ntstatus2errcode(status);
+  }
+
   /* Windows unable:
-    *  - shrinking a mapped file;
+    *  - shrink a mapped file;
     *  - change size of mapped view;
     *  - extend read-only mapping;
     * Therefore we should unmap/map entire section. */
@@ -971,6 +992,8 @@ int mdbx_mresize(int flags, mdbx_mmap_t *map, size_t size, size_t limit) {
     return ntstatus2errcode(status);
   status = NtClose(map->section);
   map->section = NULL;
+  PVOID ReservedAddress = NULL;
+  SIZE_T ReservedSize = limit;
 
   if (!NT_SUCCESS(status)) {
   bailout_ntstatus:
@@ -978,9 +1001,27 @@ int mdbx_mresize(int flags, mdbx_mmap_t *map, size_t size, size_t limit) {
   bailout:
     map->address = NULL;
     map->current = map->length = 0;
+    if (ReservedAddress)
+      (void)NtFreeVirtualMemory(GetCurrentProcess(), &ReservedAddress,
+                                &ReservedSize, MEM_RELEASE);
     return err;
   }
 
+  /* resizing of the file may take a while,
+   * therefore we reserve address space to avoid occupy it by other threads */
+  ReservedAddress = map->address;
+  status = NtAllocateVirtualMemory(GetCurrentProcess(), &ReservedAddress, 0,
+                                   &ReservedSize, MEM_RESERVE, PAGE_NOACCESS);
+  if (!NT_SUCCESS(status)) {
+    ReservedAddress = NULL;
+    if (status != /* STATUS_CONFLICTING_ADDRESSES */ 0xC0000018 ||
+        limit == map->length)
+      goto bailout_ntstatus /* no way to recovery */;
+
+    /* assume we can change base address if mapping size changed */
+    map->address = NULL;
+  }
+
 retry_file_and_section:
   err = mdbx_filesize(map->fd, &map->filesize);
   if (err != MDBX_SUCCESS)
@@ -991,7 +1032,7 @@ retry_file_and_section:
     if (err == MDBX_SUCCESS)
       map->filesize = size;
     /* ignore error, because Windows unable shrink file
-     * that already mapped (by another process) */;
+     * that already mapped (by another process) */
   }
 
   SectionSize.QuadPart = size;
@@ -1010,6 +1051,15 @@ retry_file_and_section:
   if (!NT_SUCCESS(status))
     goto bailout_ntstatus;
 
+  if (ReservedAddress) {
+    /* release reserved address space */
+    status = NtFreeVirtualMemory(GetCurrentProcess(), &ReservedAddress,
+                                 &ReservedSize, MEM_RELEASE);
+    ReservedAddress = NULL;
+    if (!NT_SUCCESS(status))
+      goto bailout_ntstatus;
+  }
+
 retry_mapview:;
   SIZE_T ViewSize = (flags & MDBX_RDONLY) ? size : limit;
   status = NtMapViewOfSection(