Merge jamppa@bk-internal.mysql.com:/home/bk/mysql-maria

into  labbari.dsl.inet.fi:/home/my/bk/mysql-maria.prod


mysys/thr_lock.c:
  Auto merged
storage/csv/ha_tina.cc:
  Auto merged
storage/csv/ha_tina.h:
  Auto merged
storage/maria/ma_blockrec.c:
  Auto merged
storage/maria/ma_check.c:
  Auto merged
storage/maria/ma_loghandler.c:
  Auto merged
storage/maria/trnman.c:
  Auto merged
storage/myisam/mi_locking.c:
  Auto merged
sql/lock.cc:
  SCCS merged

mysys/thr_lock.c

@@ -24,7 +24,7 @@ Locks are prioritized according to:
 
 The current lock types are:
 
-TL_READ	 		# Low priority read
+TL_READ                 # Low priority read
 TL_READ_WITH_SHARED_LOCKS
 TL_READ_HIGH_PRIORITY	# High priority read
 TL_READ_NO_INSERT	# Read without concurrent inserts
@@ -57,8 +57,12 @@ should put a pointer to the following functions in the lock structure:
 	 In MyISAM this is a simple check if the insert can be done
 	 at the end of the datafile.
 update_status:
-	Before a write lock is released, this function is called.
-	In MyISAM this functions updates the count and length of the datafile
+        in thr_reschedule_write_lock(), when an insert delayed thread
+        downgrades TL_WRITE lock to TL_WRITE_DELAYED, to allow SELECT
+        threads to proceed.
+        A storage engine should also call update_status internally
+        in the ::external_lock(F_UNLCK) method.
+        In MyISAM and CSV this functions updates the length of the datafile.
 get_status:
 	When one gets a lock this functions is called.
 	In MyISAM this stores the number of rows and size of the datafile
@@ -762,16 +766,6 @@ void thr_unlock(THR_LOCK_DATA *data)
   }
   else
     lock->write.last=data->prev;
-  if (lock_type >= TL_WRITE_CONCURRENT_INSERT)
-  {
-    if (lock->update_status)
-      (*lock->update_status)(data->status_param);
-  }
-  else
-  {
-    if (lock->restore_status)
-      (*lock->restore_status)(data->status_param);
-  }
   if (lock_type == TL_READ_NO_INSERT)
     lock->read_no_write_count--;
   data->type=TL_UNLOCK;				/* Mark unlocked */

sql/lock.cc

@@ -289,10 +289,10 @@ static int lock_external(THD *thd, TABLE **tables, uint count)
 void mysql_unlock_tables(THD *thd, MYSQL_LOCK *sql_lock)
 {
   DBUG_ENTER("mysql_unlock_tables");
-  if (sql_lock->lock_count)
-    thr_multi_unlock(sql_lock->locks,sql_lock->lock_count);
   if (sql_lock->table_count)
     VOID(unlock_external(thd,sql_lock->table,sql_lock->table_count));
+  if (sql_lock->lock_count)
+    thr_multi_unlock(sql_lock->locks,sql_lock->lock_count);
   my_free((uchar*) sql_lock,MYF(0));
   DBUG_VOID_RETURN;
 }

storage/csv/ha_tina.cc

@@ -441,7 +441,7 @@ ha_tina::ha_tina(handlerton *hton, TABLE_SHARE *table_arg)
   */
   current_position(0), next_position(0), local_saved_data_file_length(0),
   file_buff(0), chain_alloced(0), chain_size(DEFAULT_CHAIN_LENGTH),
-  records_is_known(0)
+  records_is_known(0), curr_lock_type(F_UNLCK)
 {
   /* Set our original buffers from pre-allocated memory */
   buffer.set((char*)byte_buffer, IO_SIZE, &my_charset_bin);
@@ -1395,6 +1395,14 @@ int ha_tina::delete_all_rows()
   DBUG_RETURN(rc);
 }
 
+int ha_tina::external_lock(THD *thd __attribute__((unused)), int lock_type)
+{
+  if (lock_type==F_UNLCK && curr_lock_type == F_WRLCK)
+    update_status();
+  curr_lock_type= lock_type;
+  return 0;
+}
+
 /*
   Called by the database to lock the table. Keep in mind that this
   is an internal lock.
@@ -1409,7 +1417,7 @@ THR_LOCK_DATA **ha_tina::store_lock(THD *thd,
   return to;
 }
 
-/* 
+/*
   Create a table. You do not want to leave the table open after a call to
   this (the database will call ::open() if it needs to).
 */

storage/csv/ha_tina.h

@@ -81,6 +81,8 @@ class ha_tina: public handler
   bool records_is_known;
 
 private:
+  int curr_lock_type;
+
   bool get_write_pos(off_t *end_pos, tina_set *closest_hole);
   int open_update_temp_file_if_needed();
   int init_tina_writer();
@@ -155,6 +157,8 @@ class ha_tina: public handler
   bool check_if_incompatible_data(HA_CREATE_INFO *info,
                                   uint table_changes);
 
+  int external_lock(THD *thd, int lock_type);
+
   THR_LOCK_DATA **store_lock(THD *thd, THR_LOCK_DATA **to,
       enum thr_lock_type lock_type);
 

storage/maria/ma_blockrec.c

@@ -1811,9 +1811,11 @@ static my_bool write_block_record(MARIA_HA *info,
       ulong length;
       ulong data_length= (tmp_data - info->rec_buff);
 
+#ifdef MONTY_WILL_KNOW
 #ifdef SANITY_CHECKS
       if (cur_block->sub_blocks == 1)
         goto crashed;                           /* no reserved full or tails */
+#endif
 #endif
 
       /*

storage/maria/ma_check.c

@@ -1925,7 +1925,11 @@ int maria_chk_data_link(HA_CHECK *param, MARIA_HA *info,int extend)
   Recover old table by reading each record and writing all keys
 
   NOTES
-    Save new datafile-name in temp_filename
+    Save new datafile-name in temp_filename.
+    We overwrite the index file as we go (writekeys() for example), so if we
+    crash during this the table is unusable and user (or Recovery in the
+    future) must repeat the REPAIR/OPTIMIZE operation. We could use a
+    temporary index file in the future (drawback: more disk space).
 
   IMPLEMENTATION (for hard repair with block format)
    - Create new, unrelated MARIA_HA of the table

storage/maria/ma_control_file.h

@@ -18,7 +18,7 @@
   First version written by Guilhem Bichot on 2006-04-27.
 */
 
-#define CONTROL_FILE_BASE_NAME "maria_control"
+#define CONTROL_FILE_BASE_NAME "maria_log_control"
 
 /* Here is the interface of this module */
 

storage/maria/ma_locking.c

@@ -55,9 +55,15 @@ int maria_lock_database(MARIA_HA *info, int lock_type)
     case F_UNLCK:
       maria_ftparser_call_deinitializer(info);
       if (info->lock_type == F_RDLCK)
+      {
 	count= --share->r_locks;
+        _ma_restore_status(info);
+      }
       else
+      {
 	count= --share->w_locks;
+        _ma_update_status(info);
+      }
       --share->tot_locks;
       if (info->lock_type == F_WRLCK && !share->w_locks)
       {

storage/maria/ma_loghandler.c

@@ -5623,7 +5623,7 @@ static my_bool write_hook_for_redo(enum translog_record_type type
     non-transactional log records (REPAIR, CREATE, RENAME, DROP) should not
     call this hook; we trust them but verify ;)
   */
-  DBUG_ASSERT(trn->trid != 0);
+  DBUG_ASSERT(!(maria_multi_threaded && (trn->trid == 0)));
   /*
     If the hook stays so simple, it would be faster to pass
     !trn->rec_lsn ? trn->rec_lsn : some_dummy_lsn
@@ -5650,7 +5650,7 @@ static my_bool write_hook_for_undo(enum translog_record_type type
                                    struct st_translog_parts *parts
                                    __attribute__ ((unused)))
 {
-  DBUG_ASSERT(trn->trid != 0); /* see write_hook_for_redo() */
+  DBUG_ASSERT(!(maria_multi_threaded && (trn->trid == 0)));
   trn->undo_lsn= *lsn;
   if (unlikely(LSN_WITH_FLAGS_TO_LSN(trn->first_undo_lsn) == 0))
     trn->first_undo_lsn=

storage/maria/ma_test_all.sh

@@ -9,6 +9,8 @@
 # Remove # from following line if you need some more information
 #set -x -v -e
 
+set -e # abort at first failure
+
 valgrind="valgrind --alignment=8 --leak-check=yes"
 silent="-s"
 suffix=""
@@ -196,15 +198,19 @@ run_repair_tests "-M -T"
 run_pack_tests "-M -T"
 
 #
-# Tests that gives warnings
+# Tests that gives warnings or errors
 #
 
 $maria_path/ma_test2$suffix $silent -L -K -W -P -S -R1 -m500
 $maria_path/maria_chk$suffix -sm test2
 echo "ma_test2$suffix $silent -L -K -R1 -m2000 ;  Should give error 135"
-$maria_path/ma_test2$suffix $silent -L -K -R1 -m2000
+$maria_path/ma_test2$suffix $silent -L -K -R1 -m2000 >ma_test2_message.txt 2>&1 && false # success is failure
+cat ma_test2_message.txt
+grep "Error: 135" ma_test2_message.txt > /dev/null
 echo "$maria_path/maria_chk$suffix -sm test2 will warn that 'Datafile is almost full'"
-$maria_path/maria_chk$suffix -sm test2
+$maria_path/maria_chk$suffix -sm test2 >ma_test2_message.txt 2>&1
+cat ma_test2_message.txt
+grep "warning: Datafile is almost full" ma_test2_message.txt >/dev/null
 $maria_path/maria_chk$suffix -ssm test2
 
 #

storage/maria/trnman.c

@@ -209,16 +209,21 @@ static TrID new_trid()
 static void set_short_trid(TRN *trn)
 {
   int i= (global_trid_generator + (intptr)trn) * 312089 % SHORT_TRID_MAX + 1;
-  my_atomic_rwlock_wrlock(&LOCK_short_trid_to_trn);
-  for ( ; ; i= i % SHORT_TRID_MAX + 1) /* the range is [1..SHORT_TRID_MAX] */
+  for ( ; !trn->short_id ; i= 1)
   {
-    void *tmp= NULL;
-    if (short_trid_to_active_trn[i] == NULL &&
-        my_atomic_casptr((void **)&short_trid_to_active_trn[i], &tmp, trn))
-      break;
+    my_atomic_rwlock_wrlock(&LOCK_short_trid_to_trn);
+    for ( ; i <= SHORT_TRID_MAX; i++) /* the range is [1..SHORT_TRID_MAX] */
+    {
+      void *tmp= NULL;
+      if (short_trid_to_active_trn[i] == NULL &&
+          my_atomic_casptr((void **)&short_trid_to_active_trn[i], &tmp, trn))
+      {
+        trn->short_id= i;
+        break;
+      }
+    }
+    my_atomic_rwlock_wrunlock(&LOCK_short_trid_to_trn);
   }
-  my_atomic_rwlock_wrunlock(&LOCK_short_trid_to_trn);
-  trn->short_id= i;
 }
 
 /*

storage/maria/unittest/Makefile.am

@@ -84,6 +84,6 @@ ma_pagecache_consist_64kWR_t_big_CPPFLAGS =	$(ma_pagecache_common_cppflags) -DPA
 
 # the generic lock manager may not be used in the end and lockman1-t crashes,
 # so we don't build lockman-t and lockman1-t 
-CLEANFILES =		maria_control page_cache_test_file_1 \
+CLEANFILES =		maria_log_control page_cache_test_file_1 \
 			maria_log.????????
 

storage/myisam/mi_locking.c

@@ -56,9 +56,15 @@ int mi_lock_database(MI_INFO *info, int lock_type)
     case F_UNLCK:
       ftparser_call_deinitializer(info);
       if (info->lock_type == F_RDLCK)
+      {
 	count= --share->r_locks;
+        mi_restore_status(info);
+      }
       else
+      {
 	count= --share->w_locks;
+        mi_update_status(info);
+      }
       --share->tot_locks;
       if (info->lock_type == F_WRLCK && !share->w_locks &&
 	  !share->delay_key_write && flush_key_blocks(share->key_cache,
@@ -84,16 +90,16 @@ int mi_lock_database(MI_INFO *info, int lock_type)
 	if (share->changed && !share->w_locks)
 	{
 #ifdef HAVE_MMAP
-    if ((info->s->mmaped_length != info->s->state.state.data_file_length) &&
-        (info->s->nonmmaped_inserts > MAX_NONMAPPED_INSERTS))
-    {
-      if (info->s->concurrent_insert)
-        rw_wrlock(&info->s->mmap_lock);
-      mi_remap_file(info, info->s->state.state.data_file_length);
-      info->s->nonmmaped_inserts= 0;
-      if (info->s->concurrent_insert)
-        rw_unlock(&info->s->mmap_lock);
-    }
+          if ((info->s->mmaped_length != info->s->state.state.data_file_length) &&
+              (info->s->nonmmaped_inserts > MAX_NONMAPPED_INSERTS))
+          {
+            if (info->s->concurrent_insert)
+              rw_wrlock(&info->s->mmap_lock);
+            mi_remap_file(info, info->s->state.state.data_file_length);
+            info->s->nonmmaped_inserts= 0;
+            if (info->s->concurrent_insert)
+              rw_unlock(&info->s->mmap_lock);
+          }
 #endif
 	  share->state.process= share->last_process=share->this_process;
 	  share->state.unique=   info->last_unique=  info->this_unique;
@@ -300,6 +306,7 @@ void mi_get_status(void* param, int concurrent_insert)
 void mi_update_status(void* param)
 {
   MI_INFO *info=(MI_INFO*) param;
+  DBUG_ENTER("mi_update_status");
   /*
     Because someone may have closed the table we point at, we only
     update the state if its our own state.  This isn't a problem as
@@ -336,20 +343,32 @@ void mi_update_status(void* param)
     }
     info->opt_flag&= ~WRITE_CACHE_USED;
   }
+  DBUG_VOID_RETURN;
 }
 
 
 void mi_restore_status(void *param)
 {
   MI_INFO *info= (MI_INFO*) param;
+  DBUG_ENTER("mi_restore_status");
+  DBUG_PRINT("info",("key_file: %ld  data_file: %ld",
+		     (long) info->s->state.state.key_file_length,
+		     (long) info->s->state.state.data_file_length));
   info->state= &info->s->state.state;
   info->append_insert_at_end= 0;
+  DBUG_VOID_RETURN;
 }
 
 
 void mi_copy_status(void* to,void *from)
 {
-  ((MI_INFO*) to)->state= &((MI_INFO*) from)->save_state;
+  MI_INFO *info= (MI_INFO*) to;
+  DBUG_ENTER("mi_copy_status");
+  info->state= &((MI_INFO*) from)->save_state;
+  DBUG_PRINT("info",("key_file: %ld  data_file: %ld",
+		     (long) info->state->key_file_length,
+		     (long) info->state->data_file_length));
+  DBUG_VOID_RETURN;
 }
 
 
@@ -377,17 +396,18 @@ void mi_copy_status(void* to,void *from)
 my_bool mi_check_status(void *param)
 {
   MI_INFO *info=(MI_INFO*) param;
+  DBUG_ENTER("mi_check_status");
+  DBUG_PRINT("info",("dellink: %ld  r_locks: %u  w_locks: %u",
+                     (long) info->s->state.dellink, (uint) info->s->r_locks,
+                     (uint) info->s->w_locks));
   /*
     The test for w_locks == 1 is here because this thread has already done an
     external lock (in other words: w_locks == 1 means no other threads has
     a write lock)
   */
-  DBUG_PRINT("info",("dellink: %ld  r_locks: %u  w_locks: %u",
-                     (long) info->s->state.dellink, (uint) info->s->r_locks,
-                     (uint) info->s->w_locks));
-  return (my_bool) !(info->s->state.dellink == HA_OFFSET_ERROR ||
+  DBUG_RETURN((my_bool) !(info->s->state.dellink == HA_OFFSET_ERROR ||
                      (myisam_concurrent_insert == 2 && info->s->r_locks &&
-                      info->s->w_locks == 1));
+                      info->s->w_locks == 1)));
 }