Fix InnoDB: Assertion failure in thread 2868898624 in file buf0lru.c …

…line 1000 InnoDB: Failing assertion: mutex_own(&buf_pool->LRU_list_mutex) and InnoDB: Assertion failure in thread 2868898624 in file buf0lru.c line 1077 InnoDB: Failing assertion: mutex_own(&buf_pool->LRU_list_mutex) Analysis: Function buf_LRU_free_block might release LRU_list_mutex on same cases to avoid mutex order problems, we need to take it back before accessing list.
MariaDB · Jul 26, 2014 · 8ae2674 · 8ae2674
1 parent be667b8
commit 8ae2674
Show file tree

Hide file tree

Showing 3 changed files with 64 additions and 5 deletions.
diff --git a/storage/xtradb/buf/buf0buf.c b/storage/xtradb/buf/buf0buf.c
@@ -2566,6 +2566,11 @@ buf_page_get_gen(
 		    || mode == BUF_PEEK_IF_IN_POOL
 		    || mode == BUF_GET_IF_IN_POOL_OR_WATCH) {
 
+			if (have_LRU_mutex) {
+				mutex_exit(&buf_pool->LRU_list_mutex);
+				have_LRU_mutex = FALSE;
+			}
+
 			return(NULL);
 		}
 
@@ -2628,13 +2633,24 @@ buf_page_get_gen(
 		//buf_pool_mutex_exit(buf_pool);
 		mutex_exit(block_mutex);
 
+		if (have_LRU_mutex) {
+			mutex_exit(&buf_pool->LRU_list_mutex);
+			have_LRU_mutex = FALSE;
+		}
+
 		return(NULL);
 	}
 
 	if (UNIV_UNLIKELY(block->page.is_corrupt &&
 			  srv_pass_corrupt_table <= 1)) {
 
 		mutex_exit(block_mutex);
+
+		if (have_LRU_mutex) {
+			mutex_exit(&buf_pool->LRU_list_mutex);
+			have_LRU_mutex = FALSE;
+		}
+
 		return(NULL);
 	}
 
@@ -2843,6 +2859,11 @@ buf_page_get_gen(
 		insert buffer (change buffer) as much as possible. */
 		ulint	page_no	= buf_block_get_page_no(block);
 
+		if (!have_LRU_mutex) {
+			mutex_enter(&buf_pool->LRU_list_mutex);
+			have_LRU_mutex = TRUE;
+		}
+
 		if (buf_LRU_free_block(&block->page, (void *)block_mutex, TRUE, &have_LRU_mutex)) {
 			mutex_exit(block_mutex);
 			if (mode == BUF_GET_IF_IN_POOL_OR_WATCH) {
@@ -2867,6 +2888,12 @@ buf_page_get_gen(
 			fprintf(stderr,
 				"innodb_change_buffering_debug evict %u %u\n",
 				(unsigned) space, (unsigned) offset);
+
+			if (have_LRU_mutex){
+				mutex_exit(&buf_pool->LRU_list_mutex);
+				have_LRU_mutex = FALSE;
+			}
+
 			return(NULL);
 		} else if (UNIV_UNLIKELY(buf_block_get_state(block)
 					 != BUF_BLOCK_FILE_PAGE
@@ -2995,6 +3022,11 @@ buf_page_get_gen(
 		_increment_page_get_statistics(block, trx);
 	}
 
+	if (have_LRU_mutex) {
+		mutex_exit(&buf_pool->LRU_list_mutex);
+		have_LRU_mutex = FALSE;
+	}
+
 	return(block);
 }
 

diff --git a/storage/xtradb/buf/buf0lru.c b/storage/xtradb/buf/buf0lru.c
@@ -1013,6 +1013,10 @@ buf_LRU_free_from_unzip_LRU_list(
 				*have_LRU_mutex = FALSE;
 			}
 			return(TRUE);
+		} else if (!*have_LRU_mutex) {
+			*have_LRU_mutex = TRUE;
+			mutex_enter(&buf_pool->LRU_list_mutex);
+			taken_LRU_mutex = TRUE;
 		}
 	}
 
@@ -1097,6 +1101,10 @@ buf_LRU_free_from_common_LRU_list(
 			}
 
 			return(TRUE);
+		} else if (!*have_LRU_mutex) {
+			mutex_enter(&buf_pool->LRU_list_mutex);
+			taken_LRU_mutex = TRUE;
+			*have_LRU_mutex = TRUE;
 		}
 	}
 
@@ -1955,9 +1963,10 @@ buf_LRU_free_block(
 	mutex_exit((mutex_t*)block_mutex);
 
 	if (!*have_LRU_mutex) {
-		mutex_enter(&buf_pool->LRU_list_mutex); /* optimistic */
+		mutex_enter(&buf_pool->LRU_list_mutex);
 		*have_LRU_mutex = TRUE;
 	}
+
 	rw_lock_x_lock(&buf_pool->page_hash_latch);
 	mutex_enter((mutex_t*)block_mutex);
 
@@ -1968,10 +1977,12 @@ buf_LRU_free_block(
 		if (b) {
 			buf_page_free_descriptor(b);
 		}
+
 		if (*have_LRU_mutex) {
 			mutex_exit(&buf_pool->LRU_list_mutex);
 			*have_LRU_mutex = FALSE;
 		}
+
 		rw_lock_x_unlock(&buf_pool->page_hash_latch);
 		return(FALSE);
 	} else if (zip || !bpage->zip.data) {
@@ -2106,10 +2117,12 @@ buf_LRU_free_block(
 		}
 
 		//buf_pool_mutex_exit(buf_pool);
+
 		if (*have_LRU_mutex) {
 			mutex_exit(&buf_pool->LRU_list_mutex);
 			*have_LRU_mutex = FALSE;
 		}
+
 		rw_lock_x_unlock(&buf_pool->page_hash_latch);
 		mutex_exit((mutex_t*)block_mutex);
 
@@ -2143,10 +2156,12 @@ buf_LRU_free_block(
 		}
 
 		//buf_pool_mutex_enter(buf_pool);
+
 		if (!*have_LRU_mutex) {
 			mutex_enter(&buf_pool->LRU_list_mutex);
 			*have_LRU_mutex = TRUE;
 		}
+
 		mutex_enter((mutex_t*)block_mutex);
 
 		if (b) {
@@ -2161,7 +2176,6 @@ buf_LRU_free_block(
 			mutex_exit(&buf_pool->LRU_list_mutex);
 			*have_LRU_mutex = FALSE;
 		}
-
 	} else {
 		/* The block_mutex should have been released by
 		buf_LRU_block_remove_hashed_page() when it returns
@@ -2173,6 +2187,7 @@ buf_LRU_free_block(
 			mutex_exit(&buf_pool->LRU_list_mutex);
 			*have_LRU_mutex = FALSE;
 		}
+
 		rw_lock_x_unlock(&buf_pool->page_hash_latch);
 	}
 

diff --git a/storage/xtradb/sync/sync0arr.c b/storage/xtradb/sync/sync0arr.c
@@ -1013,7 +1013,8 @@ sync_array_print_long_waits(
 	if (noticed) {
 		for (i = 0; i < sync_primary_wait_array->n_cells; i++) {
 			void*	wait_object;
-			os_thread_id_t reserver=0;
+			os_thread_id_t reserver=ULINT_UNDEFINED;
+			ulint loop=0;
 
 			cell = sync_array_get_nth_cell(sync_primary_wait_array, i);
 
@@ -1030,7 +1031,7 @@ sync_array_print_long_waits(
 			noticed = TRUE;
 
 			/* Try to output cell information for writer recursive way */
-			while (reserver != 0) {
+			while (reserver != ULINT_UNDEFINED) {
 				sync_cell_t* reserver_wait;
 
 				reserver_wait = sync_array_find_thread(sync_primary_wait_array, reserver);
@@ -1040,9 +1041,20 @@ sync_array_print_long_waits(
 					reserver_wait->waiting) {
 					fputs("InnoDB: Warning: Writer thread is waiting this semaphore:\n",
 						stderr);
+					reserver = ULINT_UNDEFINED;
 					sync_array_cell_print(stderr, reserver_wait, &reserver);
+					loop++;
+
+					if (reserver_wait->thread == reserver) {
+						reserver = ULINT_UNDEFINED;
+					}
 				} else {
-					reserver = 0;
+					reserver = ULINT_UNDEFINED;
+				}
+				/* This is protection against loop */
+				if (loop > 100) {
+					fputs("InnoDB: Warning: Too many waiting threads.\n", stderr);
+					break;
 				}
 			}
 		}