manual.texi website address change

row0sel.c	CHECK TABLE now also for InnoDB, a join speed optimization
trx0trx.c	CHECK TABLE now also for InnoDB, a join speed optimization
rem0cmp.c	CHECK TABLE now also for InnoDB, a join speed optimization
row0mysql.c	CHECK TABLE now also for InnoDB, a join speed optimization
page0page.c	CHECK TABLE now also for InnoDB, a join speed optimization
row0mysql.h	CHECK TABLE now also for InnoDB, a join speed optimization
trx0trx.h	CHECK TABLE now also for InnoDB, a join speed optimization
btr0btr.h	CHECK TABLE now also for InnoDB, a join speed optimization
btr0cur.h	CHECK TABLE now also for InnoDB, a join speed optimization
btr0pcur.h	CHECK TABLE now also for InnoDB, a join speed optimization
btr0pcur.ic	CHECK TABLE now also for InnoDB, a join speed optimization
btr0btr.c	CHECK TABLE now also for InnoDB, a join speed optimization
btr0cur.c	CHECK TABLE now also for InnoDB, a join speed optimization
btr0sea.c	CHECK TABLE now also for InnoDB, a join speed optimization
innodb.result	CHECK TABLE now also for InnoDB, a join speed optimization
ha_innobase.cc	CHECK TABLE now also for InnoDB, a join speed optimization
ha_innobase.h	CHECK TABLE now also for InnoDB, a join speed optimization


sql/ha_innobase.cc:
  CHECK TABLE now also for InnoDB, a join speed optimization
sql/ha_innobase.h:
  CHECK TABLE now also for InnoDB, a join speed optimization
mysql-test/r/innodb.result:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/btr/btr0btr.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/btr/btr0cur.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/btr/btr0sea.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/include/btr0btr.h:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/include/btr0cur.h:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/include/btr0pcur.h:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/include/btr0pcur.ic:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/include/row0mysql.h:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/include/trx0trx.h:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/page/page0page.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/rem/rem0cmp.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/row/row0mysql.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/row/row0sel.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
innobase/trx/trx0trx.c:
  CHECK TABLE now also for InnoDB, a join speed optimization
Docs/manual.texi:
  website address change

Docs/manual.texi

@@ -26712,7 +26712,7 @@ the maximum size for a table. The minimum tablespace size is 10 MB.
 Contact information of Innobase Oy, producer of the InnoDB engine:
 
 @example
-Website: www.innobase.fi
+Website: www.innodb.com
 Heikki.Tuuri@@innobase.inet.fi
 phone: 358-9-6969 3250 (office) 358-40-5617367 (mobile)
 InnoDB Oy Inc.

innobase/btr/btr0btr.c

@@ -2238,12 +2238,93 @@ btr_check_node_ptr(
 	return(TRUE);
 }
 
+/****************************************************************
+Checks the size and number of fields in a record based on the definition of
+the index. */
+static
+ibool
+btr_index_rec_validate(
+/*====================*/
+				/* out: TRUE if ok */
+	rec_t*		rec,	/* in: index record */
+	dict_index_t*	index)	/* in: index */
+{
+	dtype_t* type;
+	byte*	data;
+	ulint	len;
+	ulint	n;
+	ulint	i;
+	
+	n = dict_index_get_n_fields(index);
+
+	if (rec_get_n_fields(rec) != n) {
+		fprintf(stderr, "Record has %lu fields, should have %lu\n",
+				rec_get_n_fields(rec), n);
+
+		return(FALSE);
+	}
+
+	for (i = 0; i < n; i++) {
+		data = rec_get_nth_field(rec, i, &len);
+
+		type = dict_index_get_nth_type(index, i);
+		
+		if (len != UNIV_SQL_NULL && dtype_is_fixed_size(type)
+		    && len != dtype_get_fixed_size(type)) {
+			fprintf(stderr,
+			"Record field %lu len is %lu, should be %lu\n",
+				i, len, dtype_get_fixed_size(type));
+
+			return(FALSE);
+		}
+	}
+
+	return(TRUE);			
+}
+
+/****************************************************************
+Checks the size and number of fields in records based on the definition of
+the index. */
+static
+ibool
+btr_index_page_validate(
+/*====================*/
+				/* out: TRUE if ok */
+	page_t*		page,	/* in: index page */
+	dict_index_t*	index)	/* in: index */
+{
+	rec_t*		rec;
+	page_cur_t 	cur;
+	ibool		ret	= TRUE;
+	
+	page_cur_set_before_first(page, &cur);
+	page_cur_move_to_next(&cur);
+
+	for (;;) {
+		rec = (&cur)->rec;
+
+		if (page_cur_is_after_last(&cur)) {
+			break;
+		}
+
+		if (!btr_index_rec_validate(rec, index)) {
+
+			ret = FALSE;
+		}
+
+		page_cur_move_to_next(&cur);
+	}
+
+	return(ret);	
+}
+
 /****************************************************************
 Validates index tree level. */
 static
-void
+ibool
 btr_validate_level(
 /*===============*/
+				/* out: TRUE if ok */
 	dict_tree_t*	tree,	/* in: index tree */
 	ulint		level)	/* in: level number */
 {
@@ -2260,7 +2341,9 @@ btr_validate_level(
 	page_cur_t	cursor;
 	mem_heap_t*	heap;
 	dtuple_t*	node_ptr_tuple;
-
+	ibool		ret	= TRUE;
+	dict_index_t*	index;
+	
 	mtr_start(&mtr);
 
 	page = btr_root_get(tree, &mtr);
@@ -2278,13 +2361,31 @@ btr_validate_level(
 		page = btr_node_ptr_get_child(node_ptr, &mtr);
 	}
 
+	index = UT_LIST_GET_FIRST(tree->tree_indexes);
+	
 	/* Now we are on the desired level */
 loop:
 	mtr_x_lock(dict_tree_get_lock(tree), &mtr);
 
-	/* Check ordering of records */
-	page_validate(page, UT_LIST_GET_FIRST(tree->tree_indexes));
+	/* Check ordering etc. of records */
+
+	if (!page_validate(page, index)) {
+		fprintf(stderr, "Error in page %lu in index %s\n",
+			buf_frame_get_page_no(page), index->name);
 
+		ret = FALSE;
+	}
+
+	if (level == 0) {
+		if (!btr_index_page_validate(page, index)) {
+ 			fprintf(stderr,
+			"Error in page %lu in index %s\n",
+				buf_frame_get_page_no(page), index->name);
+
+			ret = FALSE;
+		}
+	}
+	
 	ut_a(btr_page_get_level(page, &mtr) == level);
 
 	right_page_no = btr_page_get_next(page, &mtr);
@@ -2374,14 +2475,17 @@ btr_validate_level(
 
 		goto loop;
 	}
+
+	return(ret);
 }
 
 /******************************************************************
 Checks the consistency of an index tree. */
 
-void
+ibool
 btr_validate_tree(
 /*==============*/
+				/* out: TRUE if ok */
 	dict_tree_t*	tree)	/* in: tree */
 {
 	mtr_t	mtr;
@@ -2397,8 +2501,15 @@ btr_validate_tree(
 
 	for (i = 0; i <= n; i++) {
 		
-		btr_validate_level(tree, n - i);
+		if (!btr_validate_level(tree, n - i)) {
+
+			mtr_commit(&mtr);
+
+			return(FALSE);
+		}
 	}
 
 	mtr_commit(&mtr);
+
+	return(TRUE);
 }

innobase/btr/btr0cur.c

@@ -163,9 +163,14 @@ btr_cur_search_to_nth_level(
 				BTR_INSERT and BTR_ESTIMATE;
 				cursor->left_page is used to store a pointer
 				to the left neighbor page, in the cases
-				BTR_SEARCH_PREV and BTR_MODIFY_PREV */
+				BTR_SEARCH_PREV and BTR_MODIFY_PREV;
+				NOTE that if has_search_latch
+				is != 0, we maybe do not have a latch set
+				on the cursor page, we assume
+				the caller uses his search latch
+				to protect the record! */
 	btr_cur_t*	cursor, /* in/out: tree cursor; the cursor page is
-				s- or x-latched */
+				   s- or x-latched, but see also above! */
 	ulint		has_search_latch,/* in: info on the latch mode the
 				caller currently has on btr_search_latch:
 				RW_S_LATCH, or 0 */

innobase/btr/btr0sea.c

@@ -601,7 +601,12 @@ btr_search_guess_on_hash(
 	btr_search_t*	info,		/* in: index search info */
 	dtuple_t*	tuple,		/* in: logical record */
 	ulint		mode,		/* in: PAGE_CUR_L, ... */
-	ulint		latch_mode, 	/* in: BTR_SEARCH_LEAF, ... */
+	ulint		latch_mode, 	/* in: BTR_SEARCH_LEAF, ...;
+					NOTE that only if has_search_latch
+					is 0, we will have a latch set on
+					the cursor page, otherwise we assume
+					the caller uses his search latch
+					to protect the record! */
 	btr_cur_t*	cursor, 	/* out: tree cursor */
 	ulint		has_search_latch,/* in: latch mode the caller
 					currently has on btr_search_latch:
@@ -722,7 +727,9 @@ btr_search_guess_on_hash(
 	}
 	
 	if (!success) {
-		btr_leaf_page_release(page, latch_mode, mtr);
+		if (!has_search_latch) {
+		          btr_leaf_page_release(page, latch_mode, mtr);
+		}
 
 		goto failure;
 	}

innobase/include/btr0btr.h

@@ -376,9 +376,10 @@ btr_print_tree(
 /******************************************************************
 Checks the consistency of an index tree. */
 
-void
+ibool
 btr_validate_tree(
 /*==============*/
+				/* out: TRUE if ok */
 	dict_tree_t*	tree);	/* in: tree */
 
 #define BTR_N_LEAF_PAGES 	1

innobase/include/btr0cur.h

@@ -98,12 +98,18 @@ btr_cur_search_to_nth_level(
 				the previous page of the record! Inserts
 				should always be made using PAGE_CUR_LE to
 				search the position! */
-	ulint		latch_mode, /* in: BTR_SEARCH_LEAF, ...;
+	ulint		latch_mode, /* in: BTR_SEARCH_LEAF, ..., ORed with
+				BTR_INSERT and BTR_ESTIMATE;
 				cursor->left_page is used to store a pointer
 				to the left neighbor page, in the cases
-				BTR_SEARCH_PREV and BTR_MODIFY_PREV */
-	btr_cur_t*	cursor, /* out: tree cursor; the cursor page is s- or
-				x-latched */
+				BTR_SEARCH_PREV and BTR_MODIFY_PREV;
+				NOTE that if has_search_latch
+				is != 0, we maybe do not have a latch set
+				on the cursor page, we assume
+				the caller uses his search latch
+				to protect the record! */
+	btr_cur_t*	cursor, /* in/out: tree cursor; the cursor page is
+				s- or x-latched, but see also above! */
 	ulint		has_search_latch,/* in: latch mode the caller
 				currently has on btr_search_latch:
 				RW_S_LATCH, or 0 */

innobase/include/btr0pcur.h

@@ -87,7 +87,11 @@ btr_pcur_open_with_no_init(
 				PAGE_CUR_LE, not PAGE_CUR_GE, as the latter
 				may end up on the previous page of the
 				record! */
-	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ... */
+	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ...;
+				NOTE that if has_search_latch != 0 then
+				we maybe do not acquire a latch on the cursor
+				page, but assume that the caller uses his
+				btr search latch to protect the record! */
 	btr_pcur_t*	cursor, /* in: memory buffer for persistent cursor */
 	ulint		has_search_latch,/* in: latch mode the caller
 				currently has on btr_search_latch:

innobase/include/btr0pcur.ic

@@ -492,7 +492,11 @@ btr_pcur_open_with_no_init(
 				PAGE_CUR_LE, not PAGE_CUR_GE, as the latter
 				may end up on the previous page of the
 				record! */
-	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ... */
+	ulint		latch_mode,/* in: BTR_SEARCH_LEAF, ...;
+				NOTE that if has_search_latch != 0 then
+				we maybe do not acquire a latch on the cursor
+				page, but assume that the caller uses his
+				btr search latch to protect the record! */
 	btr_pcur_t*	cursor, /* in: memory buffer for persistent cursor */
 	ulint		has_search_latch,/* in: latch mode the caller
 				currently has on btr_search_latch:

innobase/include/row0mysql.h

@@ -229,6 +229,15 @@ row_rename_table_for_mysql(
 	char*	old_name,	/* in: old table name */
 	char*	new_name,	/* in: new table name */
 	trx_t*	trx);		/* in: transaction handle */
+/*************************************************************************
+Checks a table for corruption. */
+
+ulint
+row_check_table_for_mysql(
+/*======================*/
+					/* out: DB_ERROR or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt);	/* in: prebuilt struct in MySQL
+					handle */
 
 /* A struct describing a place for an individual column in the MySQL
 row format which is presented to the table handler in ha_innobase.
@@ -281,7 +290,8 @@ struct row_prebuilt_struct {
 					is set to TRUE */
 	dict_index_t*	index;		/* current index for a search, if any */
 	ulint		template_type;	/* ROW_MYSQL_WHOLE_ROW, 
-					ROW_MYSQL_REC_FIELDS or
+					ROW_MYSQL_REC_FIELDS,
+					ROW_MYSQL_DUMMY_TEMPLATE, or
 					ROW_MYSQL_NO_TEMPLATE */
 	ulint		n_template;	/* number of elements in the
 					template */
@@ -359,6 +369,8 @@ struct row_prebuilt_struct {
 #define ROW_MYSQL_WHOLE_ROW	0
 #define ROW_MYSQL_REC_FIELDS	1
 #define ROW_MYSQL_NO_TEMPLATE	2
+#define ROW_MYSQL_DUMMY_TEMPLATE 3	/* dummy template used in
+					row_scan_and_check_index */
 
 #ifndef UNIV_NONINL
 #include "row0mysql.ic"

innobase/include/trx0trx.h

@@ -24,6 +24,13 @@ saving CPU time. The kernel mutex contention is increased, however. */
 
 extern ulint	trx_n_mysql_transactions;
 
+/************************************************************************
+Releases the search latch if trx has reserved it. */
+
+void
+trx_search_latch_release_if_reserved(
+/*=================================*/
+        trx_t*     trx); /* in: transaction */
 /********************************************************************
 Retrieves the error_info field from a trx. */
 
@@ -282,6 +289,13 @@ struct trx_struct{
 	ulint		n_mysql_tables_in_use; /* number of Innobase tables
 					used in the processing of the current
 					SQL statement in MySQL */
+        ulint           mysql_n_tables_locked;
+                                        /* how many tables the current SQL
+					statement uses, except those
+					in consistent read */
+        ibool           has_search_latch;
+			                /* TRUE if this trx has latched the
+			                search system latch in S-mode */
         ibool           ignore_duplicates_in_insert;
                                         /* in an insert roll back only insert
                                         of the latest row in case

innobase/page/page0page.c

@@ -1199,8 +1199,16 @@ page_rec_validate(
 	n_owned = rec_get_n_owned(rec);
 	heap_no = rec_get_heap_no(rec);
 
-	ut_a(n_owned <= PAGE_DIR_SLOT_MAX_N_OWNED);
-	ut_a(heap_no < page_header_get_field(page, PAGE_N_HEAP));
+	if (!(n_owned <= PAGE_DIR_SLOT_MAX_N_OWNED)) {
+		fprintf(stderr, "Dir slot n owned too big %lu\n", n_owned);
+		return(FALSE);
+	}
+
+	if (!(heap_no < page_header_get_field(page, PAGE_N_HEAP))) {
+		fprintf(stderr, "Heap no too big %lu %lu\n", heap_no,
+				page_header_get_field(page, PAGE_N_HEAP));
+		return(FALSE);
+	}
 	
 	return(TRUE);
 }
@@ -1216,20 +1224,21 @@ page_validate(
 	dict_index_t*	index)	/* in: data dictionary index containing
 				the page record type definition */
 {
+	page_dir_slot_t* slot;
 	mem_heap_t*	heap;
+	page_cur_t 	cur;
 	byte*		buf;
 	ulint		i;
 	ulint		count;
 	ulint		own_count;
 	ulint		slot_no;
 	ulint		data_size;
-	page_cur_t 	cur;
 	rec_t*		rec;
 	rec_t*		old_rec	= NULL;
-	page_dir_slot_t* slot;
 	ulint		offs;
 	ulint		n_slots;
-
+	ibool		ret	= FALSE;
+	
 	heap = mem_heap_create(UNIV_PAGE_SIZE);
 	
 	/* The following buffer is used to check that the
@@ -1244,8 +1253,16 @@ page_validate(
 	overlap. */
 
 	n_slots = page_dir_get_n_slots(page);
-	ut_ad(page_header_get_ptr(page, PAGE_HEAP_TOP) <=
-		page_dir_get_nth_slot(page, n_slots - 1));
+
+	if (!(page_header_get_ptr(page, PAGE_HEAP_TOP) <=
+			page_dir_get_nth_slot(page, n_slots - 1))) {
+		fprintf(stderr,
+       	"Record heap and dir overlap on a page in index %s, %lu, %lu\n",
+       		index->name, page_header_get_ptr(page, PAGE_HEAP_TOP),
+       		page_dir_get_nth_slot(page, n_slots - 1));
+
+       		goto func_exit;
+       	}
 
 	/* Validate the record list in a loop checking also that
 	it is consistent with the directory. */
@@ -1259,11 +1276,20 @@ page_validate(
 
 	for (;;) {
 		rec = (&cur)->rec;
-		page_rec_validate(rec);
+
+		if (!page_rec_validate(rec)) {
+			goto func_exit;
+		}
 		
 		/* Check that the records are in the ascending order */
 		if ((count >= 2) && (!page_cur_is_after_last(&cur))) {
-			ut_a(1 == cmp_rec_rec(rec, old_rec, index));
+			if (!(1 == cmp_rec_rec(rec, old_rec, index))) {
+				fprintf(stderr,
+				"Records in wrong order in index %s\n",
+				index->name);
+
+				goto func_exit;
+			}
 		}
 
 		if ((rec != page_get_supremum_rec(page))
@@ -1275,16 +1301,38 @@ page_validate(
 		offs = rec_get_start(rec) - page;
 		
 		for (i = 0; i < rec_get_size(rec); i++) {
-			ut_a(buf[offs + i] == 0); /* No other record may
-						  overlap this */
+			if (!buf[offs + i] == 0) {
+				/* No other record may overlap this */
+
+				fprintf(stderr,
+				"Record overlaps another in index %s \n",
+				index->name);
+
+				goto func_exit;
+			}
+				
 			buf[offs + i] = 1;
 		}
 		
 		if (rec_get_n_owned(rec) != 0) {
 			/* This is a record pointed to by a dir slot */
-			ut_a(rec_get_n_owned(rec) == own_count);
+			if (rec_get_n_owned(rec) != own_count) {
+				fprintf(stderr,
+				"Wrong owned count %lu, %lu, in index %s\n",
+				rec_get_n_owned(rec), own_count,
+				index->name);
 
-			ut_a(page_dir_slot_get_rec(slot) == rec);
+				goto func_exit;
+			}
+
+			if (page_dir_slot_get_rec(slot) != rec) {
+				fprintf(stderr,
+				"Dir slot does not point to right rec in %s\n",
+				index->name);
+
+				goto func_exit;
+			}
+			
 			page_dir_slot_check(slot);
 			
 			own_count = 0;
@@ -1297,45 +1345,89 @@ page_validate(
 		if (page_cur_is_after_last(&cur)) {
 			break;
 		}
-		
-		count++;
+
+		if (rec_get_next_offs(rec) < FIL_PAGE_DATA
+				|| rec_get_next_offs(rec) >= UNIV_PAGE_SIZE) {
+			fprintf(stderr,
+			  "Next record offset wrong %lu in index %s\n",
+			  rec_get_next_offs(rec), index->name);
+
+			goto func_exit;
+		}
+
+		count++;		
 		page_cur_move_to_next(&cur);
 		own_count++;
 		old_rec = rec;
 	}
 	
-	ut_a(rec_get_n_owned(rec) != 0);
-	ut_a(slot_no == n_slots - 1);
-	ut_a(page_header_get_field(page, PAGE_N_RECS) + 2 == count + 1);
+	if (rec_get_n_owned(rec) == 0) {
+		fprintf(stderr, "n owned is zero in index %s\n", index->name);
+
+		goto func_exit;
+	}
+		
+	if (slot_no != n_slots - 1) {
+		fprintf(stderr, "n slots wrong %lu %lu in index %s\n",
+			slot_no, n_slots - 1, index->name);
+		goto func_exit;
+	}		
+
+	if (page_header_get_field(page, PAGE_N_RECS) + 2 != count + 1) {
+		fprintf(stderr, "n recs wrong %lu %lu in index %s\n",
+		page_header_get_field(page, PAGE_N_RECS) + 2,  count + 1,
+		index->name);
+
+		goto func_exit;
+	}
 
 	if (data_size != page_get_data_size(page)) {
-		printf("Summed data size %lu, returned by func %lu\n",
+		fprintf(stderr, "Summed data size %lu, returned by func %lu\n",
 			data_size, page_get_data_size(page));
-		ut_error;
+		goto func_exit;
 	}
 
 	/* Check then the free list */
 	rec = page_header_get_ptr(page, PAGE_FREE);
 
 	while (rec != NULL) {
-		page_rec_validate(rec);
+		if (!page_rec_validate(rec)) {
+
+			goto func_exit;
+		}
 		
 		count++;	
 		offs = rec_get_start(rec) - page;
 		
 		for (i = 0; i < rec_get_size(rec); i++) {
-			ut_a(buf[offs + i] == 0);
+
+			if (buf[offs + i] != 0) {
+				fprintf(stderr,
+	                "Record overlaps another in free list, index %s \n",
+				index->name);
+
+				goto func_exit;
+			}
+				
 			buf[offs + i] = 1;
 		}
 		
 		rec = page_rec_get_next(rec);
 	}
 	
-	ut_a(page_header_get_field(page, PAGE_N_HEAP) == count + 1);
-	
+	if (page_header_get_field(page, PAGE_N_HEAP) != count + 1) {
+
+		fprintf(stderr, "N heap is wrong %lu %lu in index %s\n",
+		page_header_get_field(page, PAGE_N_HEAP), count + 1,
+		index->name);
+	}
+
+	ret = TRUE;	
+
+func_exit:
 	mem_heap_free(heap);
 
-	return(TRUE);			  
+	return(ret);			  
 }
 
 /*******************************************************************

innobase/rem/rem0cmp.c

@@ -177,7 +177,9 @@ cmp_whole_field(
 				(int)(type->prtype & ~DATA_NOT_NULL),
 				a, a_length, b, b_length));
 	default:
-		assert(0);
+	        fprintf(stderr,
+			"InnoDB: unknown type number %lu\n", data_type);
+	        ut_a(0);
 	}
 
 	return(0);

innobase/row/row0mysql.c

@@ -1129,3 +1129,146 @@ row_rename_table_for_mysql(
 	
 	return((int) err);
 }
+
+/*************************************************************************
+Checks that the index contains entries in an ascending order, unique
+constraint is not broken, and calculates the number of index entries
+in the read view of the current transaction. */
+static
+ibool
+row_scan_and_check_index(
+/*=====================*/
+					/* out: TRUE if ok */
+	row_prebuilt_t*	prebuilt,	/* in: prebuilt struct in MySQL */
+	dict_index_t*	index,		/* in: index */
+	ulint*		n_rows)		/* out: number of entries seen in the
+					current consistent read */
+{
+	mem_heap_t*	heap;
+	dtuple_t*	prev_entry = NULL;
+	ulint		matched_fields;
+	ulint		matched_bytes;
+	byte*		buf;
+	ulint		ret;
+	rec_t*		rec;
+	ibool		is_ok	= TRUE;
+	int		cmp;
+	
+	*n_rows = 0;
+	
+	buf = mem_alloc(UNIV_PAGE_SIZE);
+	heap = mem_heap_create(100);
+	
+	/* Make a dummy template in prebuilt, which we will use
+	in scanning the index entries */
+
+	prebuilt->index = index;
+	prebuilt->sql_stat_start = TRUE;
+	prebuilt->template_type = ROW_MYSQL_DUMMY_TEMPLATE;
+	prebuilt->n_template = 0;
+	prebuilt->need_to_access_clustered = FALSE;
+
+ 	dtuple_set_n_fields(prebuilt->search_tuple, 0);
+
+	prebuilt->select_lock_type = LOCK_NONE;
+
+	ret = row_search_for_mysql(buf, PAGE_CUR_G, prebuilt, 0, 0);
+loop:
+	if (ret != DB_SUCCESS) {
+
+		mem_free(buf);
+		mem_heap_free(heap);
+
+		return(is_ok);
+	}
+
+	*n_rows = *n_rows + 1;
+	
+	/* row_search... returns the index record in buf, record origin offset
+	within buf stored in the first 4 bytes, because we have built a dummy
+	template */
+	
+	rec = buf + mach_read_from_4(buf);
+	
+	if (prev_entry != NULL) {
+		matched_fields = 0;
+		matched_bytes = 0;
+	
+		cmp = cmp_dtuple_rec_with_match(prev_entry, rec,
+						&matched_fields,
+						&matched_bytes);
+		if (cmp > 0) {
+			fprintf(stderr,
+			"Error: index records in a wrong order in index %s\n",
+			index->name);
+
+			is_ok = FALSE;
+		} else if ((index->type & DICT_UNIQUE)
+			   && matched_fields >=
+			   dict_index_get_n_ordering_defined_by_user(index)) {
+			fprintf(stderr,
+			"Error: duplicate key in index %s\n",
+			index->name);
+
+			is_ok = FALSE;			   	
+		}
+	}
+
+	mem_heap_empty(heap);
+	
+	prev_entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap);
+
+	ret = row_search_for_mysql(buf, PAGE_CUR_G, prebuilt, 0, ROW_SEL_NEXT);	
+
+	goto loop;	
+}
+
+/*************************************************************************
+Checks a table for corruption. */
+
+ulint
+row_check_table_for_mysql(
+/*======================*/
+					/* out: DB_ERROR or DB_SUCCESS */
+	row_prebuilt_t*	prebuilt)	/* in: prebuilt struct in MySQL
+					handle */
+{
+	dict_table_t*	table	= prebuilt->table;
+	dict_index_t*	index;
+	ulint		n_rows;
+	ulint		n_rows_in_table;
+	ulint		ret 	= DB_SUCCESS;
+	
+	index = dict_table_get_first_index(table);
+
+	while (index != NULL) {
+      /*        fprintf(stderr, "Validating index %s\n", index->name); */
+	
+		if (!btr_validate_tree(index->tree)) {
+			ret = DB_ERROR;
+		} else {
+			if (!row_scan_and_check_index(prebuilt,
+							index, &n_rows)) {
+				ret = DB_ERROR;
+			}
+
+			/* fprintf(stderr, "%lu entries in index %s\n", n_rows,
+			  index->name); */
+
+			if (index == dict_table_get_first_index(table)) {
+				n_rows_in_table = n_rows;
+			} else if (n_rows != n_rows_in_table) {
+
+				ret = DB_ERROR;
+ 
+				fprintf(stderr,
+		"Error: index %s contains %lu entries, should be %lu\n",
+					index->name, n_rows, n_rows_in_table);
+			}
+		}
+
+		index = dict_table_get_next_index(index);
+	}
+
+	return(ret);
+}

innobase/row/row0sel.c

@@ -2341,6 +2341,65 @@ row_sel_push_cache_row_for_mysql(
 	prebuilt->n_fetch_cached++;
 }
 	
+/*************************************************************************
+Tries to do a shortcut to fetch a clustered index record with a unique key,
+using the hash index if possible (not always). We assume that the search
+mode is PAGE_CUR_GE, it is a consistent read, trx has already a read view,
+btr search latch has been locked in S-mode. */
+static
+ulint
+row_sel_try_search_shortcut_for_mysql(
+/*==================================*/
+				/* out: SEL_FOUND, SEL_EXHAUSTED, SEL_RETRY */
+	rec_t**		out_rec,/* out: record if found */
+	row_prebuilt_t*	prebuilt,/* in: prebuilt struct */
+	mtr_t*		mtr)	/* in: started mtr */
+{
+	dict_index_t*	index		= prebuilt->index;
+	dtuple_t*	search_tuple	= prebuilt->search_tuple;
+	btr_pcur_t*	pcur		= prebuilt->pcur;
+	trx_t*		trx		= prebuilt->trx;
+	rec_t*		rec;
+	
+	ut_ad(index->type & DICT_CLUSTERED);
+
+	btr_pcur_open_with_no_init(index, search_tuple, PAGE_CUR_GE,
+					BTR_SEARCH_LEAF, pcur,
+					RW_S_LATCH, mtr);
+	rec = btr_pcur_get_rec(pcur);
+	
+	if (!page_rec_is_user_rec(rec)) {
+
+		return(SEL_RETRY);
+	}
+
+	/* As the cursor is now placed on a user record after a search with
+	the mode PAGE_CUR_GE, the up_match field in the cursor tells how many
+	fields in the user record matched to the search tuple */ 
+
+	if (btr_pcur_get_up_match(pcur) < dtuple_get_n_fields(search_tuple)) {
+
+		return(SEL_EXHAUSTED);
+	}
+
+	/* This is a non-locking consistent read: if necessary, fetch
+	a previous version of the record */
+			
+	if (!lock_clust_rec_cons_read_sees(rec, index, trx->read_view)) {
+
+		return(SEL_RETRY);
+	}
+
+	if (rec_get_deleted_flag(rec)) {
+
+		return(SEL_EXHAUSTED);
+	}
+
+	*out_rec = rec;
+	
+	return(SEL_FOUND);
+}
+
 /************************************************************************
 Searches for rows in the database. This is used in the interface to
 MySQL. This function opens a cursor, and also implements fetch next
@@ -2387,6 +2446,7 @@ row_search_for_mysql(
 	ibool		cons_read_requires_clust_rec;
 	ibool		was_lock_wait;
 	ulint		ret;
+	ulint		shortcut;
 	ibool		unique_search_from_clust_index	= FALSE;
 	ibool		mtr_has_extra_clust_latch 	= FALSE;
 	ibool		moves_up 			= FALSE;
@@ -2452,6 +2512,8 @@ row_search_for_mysql(
 		mode = pcur->search_mode;
 	}
 
+	mtr_start(&mtr);
+
 	if (match_mode == ROW_SEL_EXACT && index->type & DICT_UNIQUE
 		&& index->type & DICT_CLUSTERED
 		&& dtuple_get_n_fields(search_tuple)
@@ -2464,6 +2526,8 @@ row_search_for_mysql(
 			restore cursor position, and must return
  			immediately */
 
+ 			mtr_commit(&mtr);
+
 			return(DB_RECORD_NOT_FOUND);
 		}
 
@@ -2472,8 +2536,51 @@ row_search_for_mysql(
 		mode = PAGE_CUR_GE;
 
 		unique_search_from_clust_index = TRUE;
+
+		if (trx->mysql_n_tables_locked == 0
+					&& !prebuilt->sql_stat_start) {
+
+			/* This is a SELECT query done as a consistent read,
+			and the read view has already been allocated:
+			let us try a search shortcut through the hash
+			index */
+			
+			if (!trx->has_search_latch) {
+				rw_lock_s_lock(&btr_search_latch);
+				trx->has_search_latch = TRUE;
+
+			} else if (btr_search_latch.writer_is_wait_ex) {
+			        /* There is an x-latch request waiting:
+				release the s-latch for a moment to reduce
+				starvation */
+
+			        rw_lock_s_unlock(&btr_search_latch);
+			        rw_lock_s_lock(&btr_search_latch);
+			}
+
+			shortcut = row_sel_try_search_shortcut_for_mysql(&rec,
+							prebuilt, &mtr);
+			if (shortcut == SEL_FOUND) {
+				row_sel_store_mysql_rec(buf, prebuilt, rec);
+	
+ 				mtr_commit(&mtr);
+
+				return(DB_SUCCESS);
+			
+			} else if (shortcut == SEL_EXHAUSTED) {
+
+ 				mtr_commit(&mtr);
+
+				return(DB_RECORD_NOT_FOUND);
+			}
+		}
 	}
 	
+	if (trx->has_search_latch) {
+		rw_lock_s_unlock(&btr_search_latch);
+		trx->has_search_latch = FALSE;
+	}			
+
 	/* Note that if the search mode was GE or G, then the cursor
 	naturally moves upward (in fetch next) in alphabetical order,
 	otherwise downward */
@@ -2485,8 +2592,6 @@ row_search_for_mysql(
 	} else if (direction == ROW_SEL_NEXT) {
 		moves_up = TRUE;
 	}
-		
-	mtr_start(&mtr);
 
 	thr = que_fork_get_first_thr(prebuilt->sel_graph);
 
@@ -2711,7 +2816,9 @@ row_search_for_mysql(
 	if (prebuilt->n_rows_fetched >= MYSQL_FETCH_CACHE_THRESHOLD
 			&& !prebuilt->templ_contains_blob
 			&& prebuilt->select_lock_type == LOCK_NONE
-			&& !prebuilt->clust_index_was_generated) {
+			&& !prebuilt->clust_index_was_generated
+	                && prebuilt->template_type
+	                                 != ROW_MYSQL_DUMMY_TEMPLATE) {
 
 		/* Inside an update, for example, we do not cache rows,
 		since we may use the cursor position to do the actual
@@ -2726,7 +2833,13 @@ row_search_for_mysql(
 
 		goto next_rec;
 	} else {
-		row_sel_store_mysql_rec(buf, prebuilt, rec);
+		if (prebuilt->template_type == ROW_MYSQL_DUMMY_TEMPLATE) {
+			ut_memcpy(buf + 4, rec - rec_get_extra_size(rec),
+						rec_get_size(rec));
+			mach_write_to_4(buf, rec_get_extra_size(rec) + 4);
+		} else {
+			row_sel_store_mysql_rec(buf, prebuilt, rec);
+		}
 
 		if (prebuilt->clust_index_was_generated) {
 			row_sel_store_row_id_to_prebuilt(prebuilt, index_rec,

innobase/trx/trx0trx.c

@@ -22,6 +22,7 @@ Created 3/26/1996 Heikki Tuuri
 #include "read0read.h"
 #include "srv0srv.h"
 #include "thr0loc.h"
+#include "btr0sea.h"
 
 /* Dummy session used currently in MySQL interface */
 sess_t*		trx_dummy_sess = NULL;
@@ -63,6 +64,7 @@ trx_create(
 	trx->dict_operation = FALSE;
 
 	trx->n_mysql_tables_in_use = 0;
+	trx->mysql_n_tables_locked = 0;
 
 	trx->ignore_duplicates_in_insert = FALSE;
 
@@ -96,6 +98,8 @@ trx_create(
 	trx->lock_heap = mem_heap_create_in_buffer(256);
 	UT_LIST_INIT(trx->trx_locks);
 
+	trx->has_search_latch = FALSE;
+
 	trx->read_view_heap = mem_heap_create(256);
 	trx->read_view = NULL;
 
@@ -132,6 +136,21 @@ trx_allocate_for_mysql(void)
 	return(trx);
 }
 
+/************************************************************************
+Releases the search latch if trx has reserved it. */
+
+void
+trx_search_latch_release_if_reserved(
+/*=================================*/
+        trx_t*     trx) /* in: transaction */
+{
+  if (trx->has_search_latch) {
+    rw_lock_s_unlock(&btr_search_latch);
+
+    trx->has_search_latch = FALSE;
+  }
+}
+
 /************************************************************************
 Frees a transaction object. */
 
@@ -149,6 +168,7 @@ trx_free(
 	ut_a(trx->update_undo == NULL); 
 
 	ut_a(trx->n_mysql_tables_in_use == 0);
+	ut_a(trx->mysql_n_tables_locked == 0);
 	
 	if (trx->undo_no_arr) {
 		trx_undo_arr_free(trx->undo_no_arr);
@@ -160,6 +180,8 @@ trx_free(
 	ut_a(trx->wait_lock == NULL);
 	ut_a(UT_LIST_GET_LEN(trx->wait_thrs) == 0);
 
+	ut_a(!trx->has_search_latch);
+
 	if (trx->lock_heap) {
 		mem_heap_free(trx->lock_heap);
 	}

mysql-test/r/innodb.result

@@ -144,7 +144,7 @@ test.t1	optimize	error	The handler for the table doesn't support check/repair
 a
 2
 Table	Op	Msg_type	Msg_text
-test.t1	check	error	The handler for the table doesn't support check/repair
+test.t1	check	status	OK
 a	b
 2	testing
 Table	Op	Msg_type	Msg_text

sql/ha_innobase.cc

@@ -2142,6 +2142,7 @@ ha_innobase::external_lock(
 	prebuilt->in_update_remember_pos = TRUE;
 
 	if (lock_type == F_WRLCK) {
+
 		/* If this is a SELECT, then it is in UPDATE TABLE ...
 		or SELECT ... FOR UPDATE */
 		prebuilt->select_lock_type = LOCK_X;
@@ -2153,13 +2154,27 @@ ha_innobase::external_lock(
 		}
 
 		trx->n_mysql_tables_in_use++;
+
+		if (prebuilt->select_lock_type != LOCK_NONE) {
+
+		  trx->mysql_n_tables_locked++;
+		}
 	} else {
 		trx->n_mysql_tables_in_use--;
 
-		if (trx->n_mysql_tables_in_use == 0 &&
-		    !(thd->options
-		      & (OPTION_NOT_AUTO_COMMIT | OPTION_BEGIN))) {
-		  innobase_commit(thd, trx);
+		if (trx->n_mysql_tables_in_use == 0) {
+
+		  trx->mysql_n_tables_locked = 0;
+
+		  if (trx->has_search_latch) {
+
+		    trx_search_latch_release_if_reserved(trx);
+		  }
+
+		  if (!(thd->options
+			& (OPTION_NOT_AUTO_COMMIT | OPTION_BEGIN))) {
+		    innobase_commit(thd, trx);
+		  }
 		}
 	}
 
@@ -2690,6 +2705,39 @@ ha_innobase::info(
   	DBUG_VOID_RETURN;
 }
 
+/***********************************************************************
+Tries to check that an InnoDB table is not corrupted. If corruption is
+noticed, prints to stderr information about it. In case of corruption
+may also assert a failure and crash the server. */
+
+int
+ha_innobase::check(
+/*===============*/
+					/* out: HA_ADMIN_CORRUPT or
+					HA_ADMIN_OK */
+	THD* 		thd,		/* in: user thread handle */
+	HA_CHECK_OPT* 	check_opt)	/* in: check options, currently
+					ignored */
+{
+	row_prebuilt_t* prebuilt	= (row_prebuilt_t*) innobase_prebuilt;
+	ulint		ret;
+	
+	if (prebuilt->mysql_template == NULL) {
+		/* Build the template; we will use a dummy template
+		in index scans done in checking */
+
+		build_template(prebuilt, NULL, table, ROW_MYSQL_WHOLE_ROW);
+	}
+
+	ret = row_check_table_for_mysql(prebuilt);
+
+	if (ret == DB_SUCCESS) {
+		return(HA_ADMIN_OK);
+	}
+	
+  	return(HA_ADMIN_CORRUPT); 
+}
+
 /*****************************************************************
 Adds information about free space in the InnoDB tablespace to a
 table comment which is printed out when a user calls SHOW TABLE STATUS. */

sql/ha_innobase.h

@@ -142,7 +142,7 @@ class ha_innobase: public handler
 					HA_CREATE_INFO *create_info);
   	int delete_table(const char *name);
 	int rename_table(const char* from, const char* to);
-
+	int check(THD* thd, HA_CHECK_OPT* check_opt);
         char* update_table_comment(const char* comment);
 
   	THR_LOCK_DATA **store_lock(THD *thd, THR_LOCK_DATA **to,