ndb dd - fix bug with deletes during LCP, that link between mm and dd could get lost

storage/ndb/src/kernel/blocks/dbtup/Dbtup.hpp

@@ -604,6 +604,7 @@ struct Fragrecord {
   SLList<Page>::Head m_empty_pages; // Empty pages not in logical/physical map
   
   Uint32 m_lcp_scan_op;
+  Uint32 m_lcp_keep_list;
 
   State fragStatus;
   Uint32 fragTableId;
@@ -1194,8 +1195,10 @@ typedef Ptr<HostBuffer> HostBufferPtr;
     STATIC_CONST( ALLOC       = 0x00100000 ); // Is record allocated now
     STATIC_CONST( MM_SHRINK   = 0x00200000 ); // Has MM part shrunk
     STATIC_CONST( MM_GROWN    = 0x00400000 ); // Has MM part grown
-    STATIC_CONST( FREE        = 0x00800000 ); // On free list of page
+    STATIC_CONST( FREED       = 0x00800000 ); // Is freed
     STATIC_CONST( LCP_SKIP    = 0x01000000 ); // Should not be returned in LCP
+    STATIC_CONST( LCP_KEEP    = 0x02000000 ); // Should be returned in LCP
+    STATIC_CONST( FREE        = 0x02800000 ); // Is free
     
     Uint32 get_tuple_version() const { 
       return m_header_bits & TUP_VERSION_MASK;

storage/ndb/src/kernel/blocks/dbtup/DbtupAbort.cpp

@@ -172,7 +172,7 @@ void Dbtup::execTUP_ABORTREQ(Signal* signal)
        */
       ndbout_c("clearing ALLOC");
       tuple_ptr->m_header_bits &= ~(Uint32)Tuple_header::ALLOC;
-      tuple_ptr->m_header_bits |= Tuple_header::FREE;
+      tuple_ptr->m_header_bits |= Tuple_header::FREED;
     }
   }
   

storage/ndb/src/kernel/blocks/dbtup/DbtupCommit.cpp

@@ -52,6 +52,15 @@ void Dbtup::execTUP_DEALLOCREQ(Signal* signal)
     PagePtr pagePtr;
     Tuple_header* ptr= (Tuple_header*)get_ptr(&pagePtr, &tmp, regTabPtr.p);
 
+    ndbassert(ptr->m_header_bits & Tuple_header::FREE);
+
+    if (ptr->m_header_bits & Tuple_header::LCP_KEEP)
+    {
+      ndbassert(ptr.p->m_header_bits & Tuple_header::FREED);
+      ptr->m_header_bits |= Tuple_header::FREED;
+      return;
+    }
+    
     if (regTabPtr.p->m_attributes[MM].m_no_of_varsize)
     {
       ljam();
@@ -140,6 +149,15 @@ void Dbtup::initOpConnection(Operationrec* regOperPtr)
   regOperPtr->m_undo_buffer_space= 0;
 }
 
+static
+inline
+bool
+operator>=(const Local_key& key1, const Local_key& key2)
+{
+  return key1.m_page_no > key2.m_page_no ||
+    (key1.m_page_no == key2.m_page_no && key1.m_page_idx >= key2.m_page_idx);
+}
+
 void
 Dbtup::dealloc_tuple(Signal* signal,
 		     Uint32 gci,
@@ -149,8 +167,12 @@ Dbtup::dealloc_tuple(Signal* signal,
 		     Fragrecord* regFragPtr, 
 		     Tablerec* regTabPtr)
 {
-  ptr->m_header_bits |= Tuple_header::FREE;
-  if (ptr->m_header_bits & Tuple_header::DISK_PART)
+  Uint32 lcpScan_ptr_i= regFragPtr->m_lcp_scan_op;
+  Uint32 lcp_keep_list = regFragPtr->m_lcp_keep_list;
+
+  Uint32 bits = ptr->m_header_bits;
+  Uint32 extra_bits = Tuple_header::FREED;
+  if (bits & Tuple_header::DISK_PART)
   {
     Local_key disk;
     memcpy(&disk, ptr->get_disk_ref_ptr(regTabPtr), sizeof(disk));
@@ -158,6 +180,23 @@ Dbtup::dealloc_tuple(Signal* signal,
 		   &disk, *(PagePtr*)&m_pgman.m_ptr, gci);
   }
   
+  if (! (bits & Tuple_header::LCP_SKIP) && lcpScan_ptr_i != RNIL)
+  {
+    ScanOpPtr scanOp;
+    c_scanOpPool.getPtr(scanOp, lcpScan_ptr_i);
+    Local_key rowid = regOperPtr->m_tuple_location;
+    Local_key scanpos = scanOp.p->m_scanPos.m_key;
+    rowid.m_page_no = page->frag_page_id;
+    if (rowid >= scanpos)
+    {
+      extra_bits = Tuple_header::LCP_KEEP; // Note REMOVE FREE
+      ptr->m_operation_ptr_i = lcp_keep_list;
+      regFragPtr->m_lcp_keep_list = rowid.ref();
+    }
+  }
+  
+  ptr->m_header_bits = bits | extra_bits;
+  
   if (regTabPtr->m_bits & Tablerec::TR_RowGCI)
   {
     jam();
@@ -165,15 +204,6 @@ Dbtup::dealloc_tuple(Signal* signal,
   }
 }
 
-static
-inline
-bool
-operator>=(const Local_key& key1, const Local_key& key2)
-{
-  return key1.m_page_no > key2.m_page_no ||
-    (key1.m_page_no == key2.m_page_no && key1.m_page_idx >= key2.m_page_idx);
-}
-
 void
 Dbtup::commit_operation(Signal* signal,
 			Uint32 gci,

storage/ndb/src/kernel/blocks/dbtup/DbtupMeta.cpp

@@ -138,7 +138,8 @@ void Dbtup::execTUPFRAGREQ(Signal* signal)
   regFragPtr.p->fragmentId= fragId;
   regFragPtr.p->m_tablespace_id= tablespace;
   regFragPtr.p->m_undo_complete= false;
-  regFragPtr.p->m_lcp_scan_op= RNIL; 
+  regFragPtr.p->m_lcp_scan_op = RNIL; 
+  regFragPtr.p->m_lcp_keep_list = RNIL;
   
   Uint32 noAllocatedPages= allocFragPages(regFragPtr.p, pages);
   

storage/ndb/src/kernel/blocks/dbtup/DbtupScan.cpp

@@ -587,8 +587,14 @@ Dbtup::scanNext(Signal* signal, ScanOpPtr scanPtr)
   Uint32 foundGCI;
  
   bool mm = (bits & ScanOp::SCAN_DD);
+  bool lcp = (bits & ScanOp::SCAN_LCP);
+  Uint32 lcp_list = fragPtr.p->m_lcp_keep_list;
   Uint32 size = table.m_offsets[mm].m_fix_header_size +
     (bits & ScanOp::SCAN_VS ? Tuple_header::HeaderSize + 1: 0);
+
+  if (lcp && lcp_list != RNIL)
+    goto found_lcp_keep;
+  
   while (true) {
     switch (pos.m_get) {
     case ScanPos::Get_next_page:
@@ -864,6 +870,53 @@ Dbtup::scanNext(Signal* signal, ScanOpPtr scanPtr)
   signal->theData[1] = scanPtr.i;
   sendSignal(reference(), GSN_CONTINUEB, signal, 2, JBB);
   return false;
+
+found_lcp_keep:
+  Local_key tmp;
+  tmp.assref(lcp_list);
+  tmp.m_page_no = getRealpid(fragPtr.p, tmp.m_page_no);
+  
+  Ptr<Page> pagePtr;
+  c_page_pool.getPtr(pagePtr, tmp.m_page_no);
+  Tuple_header* ptr = (Tuple_header*)
+    ((Fix_page*)pagePtr.p)->get_ptr(tmp.m_page_idx, 0);
+  Uint32 headerbits = ptr->m_header_bits;
+  ndbrequire(headerbits & Tuple_header::LCP_KEEP);
+  
+  Uint32 next = ptr->m_operation_ptr_i;
+  ptr->m_operation_ptr_i = RNIL;
+  ptr->m_header_bits = headerbits & ~(Uint32)Tuple_header::FREE;
+  
+  if (tablePtr.p->m_bits & Tablerec::TR_Checksum) {
+    jam();
+    setChecksum(ptr, tablePtr.p);
+  }
+  
+  NextScanConf* const conf = (NextScanConf*)signal->getDataPtrSend();
+  conf->scanPtr = scan.m_userPtr;
+  conf->accOperationPtr = RNIL + 1;
+  conf->fragId = frag.fragmentId;
+  conf->localKey[0] = lcp_list;
+  conf->localKey[1] = 0;
+  conf->localKeyLength = 1;
+  conf->gci = 0;
+  Uint32 blockNo = refToBlock(scan.m_userRef);
+  EXECUTE_DIRECT(blockNo, GSN_NEXT_SCANCONF, signal, 7);
+  
+  fragPtr.p->m_lcp_keep_list = next;
+  ptr->m_header_bits |= Tuple_header::FREED; // RESTORE free flag
+  if (headerbits & Tuple_header::FREED)
+  {
+    if (tablePtr.p->m_attributes[MM].m_no_of_varsize)
+    {
+      jam();
+      free_var_rec(fragPtr.p, tablePtr.p, &tmp, pagePtr);
+    } else {
+      jam();
+      free_fix_rec(fragPtr.p, tablePtr.p, &tmp, (Fix_page*)pagePtr.p);
+    }
+  }
+  return false;
 }
 
 void