Fix bug #11830883, SUPPORT "CORRUPTED" BIT FOR INNODB TABLES AND INDEXES.

Also addressed issues in bug #11745133, where we could mark a table
corrupted instead of crashing the server when found a corrupted buffer/page
if the table created with innodb_file_per_table on.

include/my_base.h

@@ -446,8 +446,9 @@ enum ha_base_keytype {
 #define HA_ERR_FILE_TOO_SHORT	  175	 /* File too short */
 #define HA_ERR_WRONG_CRC	  176	 /* Wrong CRC on page */
 #define HA_ERR_TOO_MANY_CONCURRENT_TRXS 177 /*Too many active concurrent transactions */
-#define HA_ERR_INDEX_COL_TOO_LONG 178	/* Index column length exceeds limit */
-#define HA_ERR_LAST               178    /* Copy of last error nr */
+#define HA_ERR_INDEX_COL_TOO_LONG 178	 /* Index column length exceeds limit */
+#define HA_ERR_INDEX_CORRUPT      179	 /* Index corrupted */
+#define HA_ERR_LAST               179    /* Copy of last error nr */
 
 /* Number of different errors */
 #define HA_ERR_ERRORS            (HA_ERR_LAST - HA_ERR_FIRST + 1)

mysql-test/suite/innodb/r/innodb_corrupt_bit.result

+set names utf8;
+CREATE TABLE corrupt_bit_test_ā(
+a INT AUTO_INCREMENT PRIMARY KEY,
+b CHAR(100),
+c INT,
+z INT,
+INDEX(b))
+ENGINE=InnoDB;
+INSERT INTO corrupt_bit_test_ā VALUES(0,'x',1, 1);
+CREATE UNIQUE INDEX idxā ON corrupt_bit_test_ā(c, b);
+CREATE UNIQUE INDEX idxē ON corrupt_bit_test_ā(z, b);
+SELECT * FROM corrupt_bit_test_ā;
+a	b	c	z
+1	x	1	1
+select @@unique_checks;
+@@unique_checks
+0
+select @@innodb_change_buffering_debug;
+@@innodb_change_buffering_debug
+1
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+1,z+1 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+10,z+10 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+20,z+20 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+50,z+50 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+100,z+100 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+200,z+200 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+400,z+400 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+800,z+800 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+1600,z+1600 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+4000,z+4000 FROM corrupt_bit_test_ā;
+select count(*) from corrupt_bit_test_ā;
+count(*)
+1024
+CREATE INDEX idx3 ON corrupt_bit_test_ā(b, c);
+INSERT INTO corrupt_bit_test_ā VALUES(13000,'x',1,1);
+CREATE INDEX idx4 ON corrupt_bit_test_ā(b, z);
+check table corrupt_bit_test_ā;
+Table	Op	Msg_type	Msg_text
+test.corrupt_bit_test_ā	check	Warning	InnoDB: The B-tree of index "idxā" is corrupted.
+test.corrupt_bit_test_ā	check	Warning	InnoDB: The B-tree of index "idxē" is corrupted.
+test.corrupt_bit_test_ā	check	error	Corrupt
+select c from corrupt_bit_test_ā;
+ERROR HY000: Incorrect key file for table 'corrupt_bit_test_ā'; try to repair it
+select z from corrupt_bit_test_ā;
+ERROR HY000: Incorrect key file for table 'corrupt_bit_test_ā'; try to repair it
+show warnings;
+Level	Code	Message
+Warning	179	InnoDB: Index "idxē" for table "test/corrupt_bit_test_@1s" is marked as corrupted
+Error	1034	Incorrect key file for table 'corrupt_bit_test_ā'; try to repair it
+insert into corrupt_bit_test_ā values (10001, "a", 20001, 20001);
+select * from corrupt_bit_test_ā use index(primary) where a = 10001;
+a	b	c	z
+10001	a	20001	20001
+begin;
+insert into corrupt_bit_test_ā values (10002, "a", 20002, 20002);
+delete from corrupt_bit_test_ā where a = 10001;
+insert into corrupt_bit_test_ā values (10001, "a", 20001, 20001);
+rollback;
+drop index idxā on corrupt_bit_test_ā;
+check table corrupt_bit_test_ā;
+Table	Op	Msg_type	Msg_text
+test.corrupt_bit_test_ā	check	Warning	InnoDB: Index "idxē" is marked as corrupted
+test.corrupt_bit_test_ā	check	error	Corrupt
+set names utf8;
+select z from corrupt_bit_test_ā;
+ERROR HY000: Incorrect key file for table 'corrupt_bit_test_ā'; try to repair it
+drop index idxē on corrupt_bit_test_ā;
+select z from corrupt_bit_test_ā limit 10;
+z
+20001
+1
+1
+2
+11
+12
+21
+22
+31
+32
+drop table corrupt_bit_test_ā;
+SET GLOBAL innodb_change_buffering_debug = 0;

mysql-test/suite/innodb/t/innodb_corrupt_bit.test

+#
+# Test for persistent corrupt bit for corrupted index and table
+#
+-- source include/have_innodb.inc
+
+# This test needs debug server
+--source include/have_debug.inc
+
+-- disable_query_log
+# This test setup is extracted from bug56680.test:
+# The flag innodb_change_buffering_debug is only available in debug builds.
+# It instructs InnoDB to try to evict pages from the buffer pool when
+# change buffering is possible, so that the change buffer will be used
+# whenever possible.
+-- error 0,ER_UNKNOWN_SYSTEM_VARIABLE
+SET @innodb_change_buffering_debug_orig = @@innodb_change_buffering_debug;
+-- error 0,ER_UNKNOWN_SYSTEM_VARIABLE
+SET GLOBAL innodb_change_buffering_debug = 1;
+
+# Turn off Unique Check to create corrupted index with dup key
+SET UNIQUE_CHECKS=0;
+
+-- enable_query_log
+
+set names utf8;
+
+CREATE TABLE corrupt_bit_test_ā(
+       a INT AUTO_INCREMENT PRIMARY KEY,
+       b CHAR(100),
+       c INT,
+       z INT,
+       INDEX(b))
+ENGINE=InnoDB;
+
+INSERT INTO corrupt_bit_test_ā VALUES(0,'x',1, 1);
+
+# This is the first unique index we intend to corrupt
+CREATE UNIQUE INDEX idxā ON corrupt_bit_test_ā(c, b);
+
+# This is the second unique index we intend to corrupt
+CREATE UNIQUE INDEX idxē ON corrupt_bit_test_ā(z, b);
+
+SELECT * FROM corrupt_bit_test_ā;
+
+select @@unique_checks;
+select @@innodb_change_buffering_debug;
+
+# Create enough rows for the table, so that the insert buffer will be
+# used for modifying the secondary index page. There must be multiple
+# index pages, because changes to the root page are never buffered.
+
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+1,z+1 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+10,z+10 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+20,z+20 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+50,z+50 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+100,z+100 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+200,z+200 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+400,z+400 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+800,z+800 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+1600,z+1600 FROM corrupt_bit_test_ā;
+INSERT INTO corrupt_bit_test_ā SELECT 0,b,c+4000,z+4000 FROM corrupt_bit_test_ā;
+
+select count(*) from corrupt_bit_test_ā;
+
+CREATE INDEX idx3 ON corrupt_bit_test_ā(b, c);
+
+# Create a dup key error on index "idxē" and "idxā" by inserting a dup value
+INSERT INTO corrupt_bit_test_ā VALUES(13000,'x',1,1);
+
+# creating an index should succeed even if other secondary indexes are corrupted
+CREATE INDEX idx4 ON corrupt_bit_test_ā(b, z);
+
+# Check table will find the unique indexes corrupted
+# with dup key
+check table corrupt_bit_test_ā;
+
+# This selection intend to use the corrupted index. Expect to fail
+-- error ER_NOT_KEYFILE
+select c from corrupt_bit_test_ā;
+
+-- error ER_NOT_KEYFILE
+select z from corrupt_bit_test_ā;
+
+show warnings;
+
+# Since corrupted index is a secondary index, we only disable such
+# index and allow other DML to proceed
+insert into corrupt_bit_test_ā values (10001, "a", 20001, 20001);
+
+# This does not use the corrupted index, expect to succeed
+select * from corrupt_bit_test_ā use index(primary) where a = 10001;
+
+# Some more DMLs
+begin;
+insert into corrupt_bit_test_ā values (10002, "a", 20002, 20002);
+delete from corrupt_bit_test_ā where a = 10001;
+insert into corrupt_bit_test_ā values (10001, "a", 20001, 20001);
+rollback;
+
+# Drop one corrupted index before reboot
+drop index idxā on corrupt_bit_test_ā;
+
+check table corrupt_bit_test_ā;
+
+# Shut down the server
+-- exec echo "wait" > $MYSQLTEST_VARDIR/tmp/mysqld.1.expect
+-- shutdown_server 20
+-- source include/wait_until_disconnected.inc
+
+# Restart the server
+-- disable_query_log
+--exec echo "restart" > $MYSQLTEST_VARDIR/tmp/mysqld.1.expect
+--enable_reconnect
+--source include/wait_until_connected_again.inc
+--disable_reconnect
+-- enable_query_log
+
+set names utf8;
+
+# The index is marked as suspect in Sys_indexes too, so after server
+# reboot, the attempt to use the index will fail too.
+-- error ER_NOT_KEYFILE
+select z from corrupt_bit_test_ā;
+
+# Drop the corrupted index
+drop index idxē on corrupt_bit_test_ā;
+
+# Now select back to normal
+select z from corrupt_bit_test_ā limit 10;
+
+# Drop table
+drop table corrupt_bit_test_ā;
+
+-- error 0, ER_UNKNOWN_SYSTEM_VARIABLE
+SET GLOBAL innodb_change_buffering_debug = 0;

mysql-test/suite/sys_vars/r/all_vars.result

@@ -11,7 +11,9 @@ There should be *no* long test name listed below:
 select variable_name as `There should be *no* variables listed below:` from t2
 left join t1 on variable_name=test_name where test_name is null;
 There should be *no* variables listed below:
+INNODB_FORCE_LOAD_CORRUPTED
 INNODB_LARGE_PREFIX
+INNODB_FORCE_LOAD_CORRUPTED
 INNODB_LARGE_PREFIX
 drop table t1;
 drop table t2;

mysys/my_handler_errors.h

@@ -81,7 +81,8 @@ static const char *handler_error_messages[]=
   "File to short; Expected more data in file",
   "Read page with wrong checksum",
   "Too many active concurrent transactions",
-  "Index column length exceeds limit"
+  "Index column length exceeds limit",
+  "Index corrupted"
 };
 
 extern void my_handler_error_register(void);

sql/handler.cc

@@ -358,6 +358,7 @@ int ha_init_errors(void)
   SETMSG(HA_ERR_AUTOINC_ERANGE,         ER_DEFAULT(ER_WARN_DATA_OUT_OF_RANGE));
   SETMSG(HA_ERR_TOO_MANY_CONCURRENT_TRXS, ER_DEFAULT(ER_TOO_MANY_CONCURRENT_TRXS));
   SETMSG(HA_ERR_INDEX_COL_TOO_LONG,	ER_DEFAULT(ER_INDEX_COLUMN_TOO_LONG));
+  SETMSG(HA_ERR_INDEX_CORRUPT,		ER_DEFAULT(ER_INDEX_CORRUPT));
 
   /* Register the error messages for use with my_error(). */
   return my_error_register(get_handler_errmsgs, HA_ERR_FIRST, HA_ERR_LAST);
@@ -2865,6 +2866,9 @@ void handler::print_error(int error, myf errflag)
   case HA_ERR_INDEX_COL_TOO_LONG:
     textno= ER_INDEX_COLUMN_TOO_LONG;
     break;
+  case HA_ERR_INDEX_CORRUPT:
+    textno= ER_INDEX_CORRUPT;
+    break;
   default:
     {
       /* The error was "unknown" to this function.

sql/share/errmsg-utf8.txt

@@ -6415,3 +6415,5 @@ ER_ERROR_IN_TRIGGER_BODY
 ER_ERROR_IN_UNKNOWN_TRIGGER_BODY
   eng "Unknown trigger has an error in its body: '%-.256s'"
 
+ER_INDEX_CORRUPT
+  eng "Index %s is corrupted"

storage/innobase/buf/buf0buf.c

@@ -3473,6 +3473,55 @@ buf_page_create(
 	return(block);
 }
 
+/********************************************************************//**
+Mark a table with the specified space pointed by bpage->space corrupted.
+Also remove the bpage from LRU list.
+@return TRUE if successful */
+static
+ibool
+buf_mark_space_corrupt(
+/*===================*/
+	buf_page_t*	bpage)	/*!< in: pointer to the block in question */
+{
+	buf_pool_t*	buf_pool = buf_pool_from_bpage(bpage);
+	const ibool	uncompressed = (buf_page_get_state(bpage)
+					== BUF_BLOCK_FILE_PAGE);
+	ulint		space = bpage->space;
+	ulint		offset = bpage->offset;
+	ibool		ret = TRUE;
+
+	/* First unfix and release lock on the bpage */
+	buf_pool_mutex_enter(buf_pool);
+	mutex_enter(buf_page_get_mutex(bpage));
+	ut_ad(buf_page_get_io_fix(bpage) == BUF_IO_READ);
+	ut_ad(bpage->buf_fix_count == 0);
+
+	/* Set BUF_IO_NONE before we remove the block from LRU list */
+	buf_page_set_io_fix(bpage, BUF_IO_NONE);
+
+	if (uncompressed) {
+		rw_lock_x_unlock_gen(
+			&((buf_block_t*) bpage)->lock,
+			BUF_IO_READ);
+	}
+
+	/* Find the table with specified space id, and mark it corrupted */
+	if (dict_set_corrupted_by_space(space)) {
+		ut_ad(bpage->space == space && bpage->offset == offset);
+		buf_LRU_free_one_page(bpage);
+	} else {
+		ret = FALSE;
+	}
+
+	ut_ad(buf_pool->n_pend_reads > 0);
+	buf_pool->n_pend_reads--;
+
+	mutex_exit(buf_page_get_mutex(bpage));
+	buf_pool_mutex_exit(buf_pool);
+
+	return(ret);
+}
+
 /********************************************************************//**
 Completes an asynchronous read or write request of a file page to or from
 the buffer pool. */
@@ -3598,10 +3647,19 @@ buf_page_io_complete(
 			      "InnoDB: about forcing recovery.\n", stderr);
 
 			if (srv_force_recovery < SRV_FORCE_IGNORE_CORRUPT) {
-				fputs("InnoDB: Ending processing because of"
-				      " a corrupt database page.\n",
-				      stderr);
-				exit(1);
+				/* If page space id is larger than TRX_SYS_SPACE
+				(0), we will attempt to mark the corresponding
+				table as corrupted instead of crashing server */
+				if (bpage->space > TRX_SYS_SPACE
+				    && buf_mark_space_corrupt(bpage)) {
+					return;
+				} else {
+					fputs("InnoDB: Ending processing"
+					      " because of"
+					      " a corrupt database page.\n",
+					      stderr);
+					ut_error;
+				}
 			}
 		}
 

storage/innobase/buf/buf0lru.c

@@ -1885,6 +1885,22 @@ buf_LRU_block_free_hashed_page(
 	buf_LRU_block_free_non_file_page(block);
 }
 
+/******************************************************************//**
+Remove one page from LRU list and put it to free list */
+UNIV_INTERN
+void
+buf_LRU_free_one_page(
+/*==================*/
+	buf_page_t*	bpage)	/*!< in/out: block, must contain a file page and
+				be in a state where it can be freed; there
+				may or may not be a hash index to the page */
+{
+	if (buf_LRU_block_remove_hashed_page(bpage, TRUE)
+	    != BUF_BLOCK_ZIP_FREE) {
+		buf_LRU_block_free_hashed_page((buf_block_t*) bpage);
+	}
+}
+
 /**********************************************************************//**
 Updates buf_pool->LRU_old_ratio for one buffer pool instance.
 @return	updated old_pct */

storage/innobase/dict/dict0dict.c

@@ -54,6 +54,7 @@ UNIV_INTERN dict_index_t*	dict_ind_compact;
 #include "row0merge.h"
 #include "m_ctype.h" /* my_isspace() */
 #include "ha_prototypes.h" /* innobase_strcasecmp(), innobase_casedn_str()*/
+#include "row0upd.h"
 
 #include <ctype.h>
 
@@ -611,8 +612,7 @@ dict_table_get_on_id(
 {
 	dict_table_t*	table;
 
-	if (table_id <= DICT_FIELDS_ID
-	    || trx->dict_operation_lock_mode == RW_X_LATCH) {
+	if (trx->dict_operation_lock_mode == RW_X_LATCH) {
 
 		/* Note: An X latch implies that the transaction
 		already owns the dictionary mutex. */
@@ -5046,4 +5046,179 @@ dict_close(void)
 		rw_lock_free(&dict_table_stats_latches[i]);
 	}
 }
+
+/**********************************************************************//**
+Find a table in dict_sys->table_LRU list with specified space id
+@return table if found, NULL if not */
+static
+dict_table_t*
+dict_find_table_by_space(
+/*=====================*/
+	ulint	space_id)		/*!< in: space ID */
+{
+	dict_table_t*   table;
+	ulint		num_item;
+	ulint		count = 0;
+
+	ut_ad(space_id > 0);
+
+	table = UT_LIST_GET_FIRST(dict_sys->table_LRU);
+	num_item =  UT_LIST_GET_LEN(dict_sys->table_LRU);
+
+	/* This function intentionally does not acquire mutex as it is used
+	by error handling code in deep call stack as last means to avoid
+	killing the server, so it worth to risk some consequencies for
+	the action. */
+	while (table && count < num_item) {
+		if (table->space == space_id) {
+			return(table);
+		}
+
+		table = UT_LIST_GET_NEXT(table_LRU, table);
+		count++;
+	}
+
+	return(NULL);
+}
+
+/**********************************************************************//**
+Flags a table with specified space_id corrupted in the data dictionary
+cache
+@return TRUE if successful */
+UNIV_INTERN
+ibool
+dict_set_corrupted_by_space(
+/*========================*/
+	ulint	space_id)		/*!< in: space ID */
+{
+	dict_table_t*   table;
+
+	table = dict_find_table_by_space(space_id);
+
+	if (!table) {
+		return(FALSE);
+	}
+
+	/* mark the table->corrupted bit only, since the caller
+	could be too deep in the stack for SYS_INDEXES update */
+	table->corrupted = TRUE;
+
+	return(TRUE);
+}
+
+/**********************************************************************//**
+Flags an index corrupted both in the data dictionary cache
+and in the SYS_INDEXES */
+UNIV_INTERN
+void
+dict_set_corrupted(
+/*===============*/
+	dict_index_t*	index)		/*!< in/out: index */
+{
+	mem_heap_t*	heap;
+	mtr_t		mtr;
+	dict_index_t*	sys_index;
+	dtuple_t*	tuple;
+	dfield_t*	dfield;
+	byte*		buf;
+	const char*	status;
+	btr_cur_t	cursor;
+
+	ut_ad(index);
+	ut_ad(mutex_own(&dict_sys->mutex));
+	ut_ad(!dict_table_is_comp(dict_sys->sys_tables));
+	ut_ad(!dict_table_is_comp(dict_sys->sys_indexes));
+
+#ifdef UNIV_SYNC_DEBUG
+        ut_ad(sync_thread_levels_empty_except_dict());
+#endif
+
+	/* Mark the table as corrupted only if the clustered index
+	is corrupted */
+	if (dict_index_is_clust(index)) {
+		index->table->corrupted = TRUE;
+	}
+
+	if (UNIV_UNLIKELY(dict_index_is_corrupted(index))) {
+		/* The index was already flagged corrupted. */
+		ut_ad(index->table->corrupted);
+		return;
+	}
+
+	heap = mem_heap_create(sizeof(dtuple_t) + 2 * (sizeof(dfield_t)
+			       + sizeof(que_fork_t) + sizeof(upd_node_t)
+			       + sizeof(upd_t) + 12));
+	mtr_start(&mtr);
+	index->type |= DICT_CORRUPT;
+
+	sys_index = UT_LIST_GET_FIRST(dict_sys->sys_indexes->indexes);
+
+	/* Find the index row in SYS_INDEXES */
+	tuple = dtuple_create(heap, 2);
+
+	dfield = dtuple_get_nth_field(tuple, 0);
+	buf = mem_heap_alloc(heap, 8);
+	mach_write_to_8(buf, index->table->id);
+	dfield_set_data(dfield, buf, 8);
+
+	dfield = dtuple_get_nth_field(tuple, 1);
+	buf = mem_heap_alloc(heap, 8);
+	mach_write_to_8(buf, index->id);
+	dfield_set_data(dfield, buf, 8);
+
+	dict_index_copy_types(tuple, sys_index, 2);
+
+	btr_cur_search_to_nth_level(sys_index, 0, tuple, PAGE_CUR_GE,
+				    BTR_MODIFY_LEAF,
+				    &cursor, 0, __FILE__, __LINE__, &mtr);
+
+	if (cursor.up_match == dtuple_get_n_fields(tuple)) {
+		/* UPDATE SYS_INDEXES SET TYPE=index->type
+		WHERE TABLE_ID=index->table->id AND INDEX_ID=index->id */
+		ulint	len;
+		byte*	field	= rec_get_nth_field_old(
+			btr_cur_get_rec(&cursor),
+			DICT_SYS_INDEXES_TYPE_FIELD, &len);
+		if (len != 4) {
+			goto fail;
+		}
+		mlog_write_ulint(field, index->type, MLOG_4BYTES, &mtr);
+		status = "  InnoDB: Flagged corruption of ";
+	} else {
+fail:
+		status = "  InnoDB: Unable to flag corruption of ";
+	}
+
+	mtr_commit(&mtr);
+	mem_heap_free(heap);
+
+	ut_print_timestamp(stderr);
+	fputs(status, stderr);
+	dict_index_name_print(stderr, NULL, index);
+	putc('\n', stderr);
+}
+
+/**********************************************************************//**
+Flags an index corrupted in the data dictionary cache only. This
+is used mostly to mark a corrupted index when index's own dictionary
+is corrupted, and we force to load such index for repair purpose */
+UNIV_INTERN
+void
+dict_set_corrupted_index_cache_only(
+/*================================*/
+	dict_index_t*	index)		/*!< in/out: index */
+{
+	ut_ad(index);
+	ut_ad(mutex_own(&dict_sys->mutex));
+	ut_ad(!dict_table_is_comp(dict_sys->sys_tables));
+	ut_ad(!dict_table_is_comp(dict_sys->sys_indexes));
+
+	/* Mark the table as corrupted only if the clustered index
+	is corrupted */
+	if (dict_index_is_clust(index)) {
+		index->table->corrupted = TRUE;
+	}
+
+	index->type |= DICT_CORRUPT;
+}
 #endif /* !UNIV_HOTBACKUP */

storage/innobase/dict/dict0load.c

@@ -52,6 +52,11 @@ static const char* SYSTEM_TABLE_NAME[] = {
 	"SYS_FOREIGN",
 	"SYS_FOREIGN_COLS"
 };
+
+/* If this flag is TRUE, then we will load the cluster index's (and tables')
+metadata even if it is marked as "corrupted". */
+UNIV_INTERN my_bool     srv_load_corrupted = FALSE;
+
 /****************************************************************//**
 Compare the name of an index column.
 @return	TRUE if the i'th column of index is 'name'. */
@@ -1324,6 +1329,9 @@ dict_load_index_low(
 		goto err_len;
 	}
 	type = mach_read_from_4(field);
+	if (UNIV_UNLIKELY(type & (~0 << DICT_IT_BITS))) {
+		return("unknown SYS_INDEXES.TYPE bits");
+	}
 
 	field = rec_get_nth_field_old(rec, 7/*SPACE*/, &len);
 	if (UNIV_UNLIKELY(len != 4)) {
@@ -1423,16 +1431,47 @@ dict_load_indexes(
 			goto next_rec;
 		} else if (err_msg) {
 			fprintf(stderr, "InnoDB: %s\n", err_msg);
+			if (ignore_err & DICT_ERR_IGNORE_CORRUPT) {
+				goto next_rec;
+			}
 			error = DB_CORRUPTION;
 			goto func_exit;
 		}
 
 		ut_ad(index);
 
+		/* Check whether the index is corrupted */
+		if (dict_index_is_corrupted(index)) {
+			ut_print_timestamp(stderr);
+			fputs("  InnoDB: ", stderr);
+			dict_index_name_print(stderr, NULL, index);
+			fputs(" is corrupted\n", stderr);
+
+			if (!srv_load_corrupted
+			    && !(ignore_err & DICT_ERR_IGNORE_CORRUPT)
+			    && dict_index_is_clust(index)) {
+				dict_mem_index_free(index);
+
+				error = DB_INDEX_CORRUPT;
+				goto func_exit;
+			} else {
+				/* We will load the index if
+				1) srv_load_corrupted is TRUE
+				2) ignore_err is set with
+				DICT_ERR_IGNORE_CORRUPT
+				3) if the index corrupted is a secondary
+				index */
+				ut_print_timestamp(stderr);
+				fputs("  InnoDB: load corrupted index ", stderr);
+				dict_index_name_print(stderr, NULL, index);
+				putc('\n', stderr);
+			}
+		}
+
 		/* We check for unsupported types first, so that the
 		subsequent checks are relevant for the supported types. */
-		if (index->type & ~(DICT_CLUSTERED | DICT_UNIQUE)) {
-
+		if (index->type & ~(DICT_CLUSTERED | DICT_UNIQUE
+				    | DICT_CORRUPT)) {
 			fprintf(stderr,
 				"InnoDB: Error: unknown type %lu"
 				" of index %s of table %s\n",
@@ -1453,9 +1492,13 @@ dict_load_indexes(
 				/* If caller can tolerate this error,
 				we will continue to load the index and
 				let caller deal with this error. However
-				mark the index and table corrupted */
-				index->corrupted = TRUE;
-				table->corrupted = TRUE;
+				mark the index and table corrupted. We
+				only need to mark such in the index
+				dictionary cache for such metadata corruption,
+				since we would always be able to set it
+				when loading the dictionary cache */
+				dict_set_corrupted_index_cache_only(index);
+
 				fprintf(stderr,
 					"InnoDB: Index is corrupt but forcing"
 					" load into data dictionary\n");
@@ -1495,9 +1538,10 @@ dict_load_indexes(
 					index->name, table->name);
 
 				/* If the force recovery flag is set, and
-				if the failed index is not the primary index, we
-				will continue and open other indexes */
-				if (srv_force_recovery
+				if the failed index is not the clustered index,
+				we will continue and open other indexes */
+				if ((srv_force_recovery
+				     || srv_load_corrupted)
 				    && !dict_index_is_clust(index)) {
 					error = DB_SUCCESS;
 					goto next_rec;
@@ -1812,6 +1856,30 @@ dict_load_table(
 
 	err = dict_load_indexes(table, heap, ignore_err);
 
+	if (err == DB_INDEX_CORRUPT) {
+		/* Refuse to load the table if the table has a corrupted
+		cluster index */
+		if (!srv_load_corrupted) {
+			fprintf(stderr, "InnoDB: Error: Load table ");
+			ut_print_name(stderr, NULL, TRUE, table->name);
+			fprintf(stderr, " failed, the table has corrupted"
+					" clustered indexes. Turn on"
+					" 'innodb_force_load_corrupted'"
+					" to drop it\n");
+
+			dict_table_remove_from_cache(table);
+			table = NULL;
+			goto func_exit;
+		} else {
+			dict_index_t*	clust_index;
+			clust_index = dict_table_get_first_index(table);
+
+			if (dict_index_is_corrupted(clust_index)) {
+				table->corrupted = TRUE;
+			}
+		}
+	}
+
 	/* Initialize table foreign_child value. Its value could be
 	changed when dict_load_foreigns() is called below */
 	table->fk_max_recusive_level = 0;
@@ -1838,9 +1906,15 @@ dict_load_table(
 		index = dict_table_get_first_index(table);
 
 		if (!srv_force_recovery || !index
-		     || !dict_index_is_clust(index)) {
+		    || !dict_index_is_clust(index)) {
 			dict_table_remove_from_cache(table);
 			table = NULL;
+		} else if (dict_index_is_corrupted(index)) {
+
+			/* It is possible we force to load a corrupted
+			clustered index if srv_load_corrupted is set.
+			Mark the table as corrupted in this case */
+			table->corrupted = TRUE;
 		}
 	}
 #if 0
@@ -1867,6 +1941,7 @@ dict_load_table(
 		mutex_exit(&dict_foreign_err_mutex);
 	}
 #endif /* 0 */
+func_exit:
 	mem_heap_free(heap);
 
 	return(table);

storage/innobase/handler/ha_innodb.cc

@@ -1043,6 +1043,8 @@ convert_error_code_to_mysql(
 #endif /* HA_ERR_TOO_MANY_CONCURRENT_TRXS */
 	case DB_UNSUPPORTED:
 		return(HA_ERR_UNSUPPORTED);
+	case DB_INDEX_CORRUPT:
+		return(HA_ERR_INDEX_CORRUPT);
 	}
 }
 
@@ -2078,6 +2080,27 @@ innobase_convert_name(
 
 }
 
+/*****************************************************************//**
+A wrapper function of innobase_convert_name(), convert a table or
+index name to the MySQL system_charset_info (UTF-8) and quote it if needed.
+@return	pointer to the end of buf */
+static inline
+void
+innobase_format_name(
+/*==================*/
+	char*		buf,	/*!< out: buffer for converted identifier */
+	ulint		buflen,	/*!< in: length of buf, in bytes */
+	const char*	name,	/*!< in: index or table name to format */
+	ibool		is_index_name) /*!< in: index name */
+{
+	const char*     bufend;
+
+	bufend = innobase_convert_name(buf, buflen, name, strlen(name),
+				       NULL, is_index_name);
+
+	buf[bufend - buf] = '\0';
+}
+
 /**********************************************************************//**
 Determines if the currently running transaction has been interrupted.
 @return	TRUE if interrupted */
@@ -5664,12 +5687,14 @@ ha_innobase::index_read(
 
 	index = prebuilt->index;
 
-	if (UNIV_UNLIKELY(index == NULL)) {
+	if (UNIV_UNLIKELY(index == NULL) || dict_index_is_corrupted(index)) {
 		prebuilt->index_usable = FALSE;
 		DBUG_RETURN(HA_ERR_CRASHED);
 	}
 	if (UNIV_UNLIKELY(!prebuilt->index_usable)) {
-		DBUG_RETURN(HA_ERR_TABLE_DEF_CHANGED);
+		DBUG_RETURN(dict_index_is_corrupted(index)
+			    ? HA_ERR_INDEX_CORRUPT
+			    : HA_ERR_TABLE_DEF_CHANGED);
 	}
 
 	/* Note that if the index for which the search template is built is not
@@ -5855,10 +5880,33 @@ ha_innobase::change_active_index(
 							   prebuilt->index);
 
 	if (UNIV_UNLIKELY(!prebuilt->index_usable)) {
-		push_warning_printf(user_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
-				    HA_ERR_TABLE_DEF_CHANGED,
-				    "InnoDB: insufficient history for index %u",
-				    keynr);
+		if (dict_index_is_corrupted(prebuilt->index)) {
+			char	index_name[MAX_FULL_NAME_LEN + 1];
+			char	table_name[MAX_FULL_NAME_LEN + 1];
+
+			innobase_format_name(
+				index_name, sizeof index_name,
+				prebuilt->index->name, TRUE);
+
+			innobase_format_name(
+				table_name, sizeof table_name,
+				prebuilt->index->table->name, FALSE);
+
+			push_warning_printf(
+				user_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+				HA_ERR_INDEX_CORRUPT,
+				"InnoDB: Index %s for table %s is"
+				" marked as corrupted",
+				index_name, table_name);
+			DBUG_RETURN(1);
+		} else {
+			push_warning_printf(
+				user_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+				HA_ERR_TABLE_DEF_CHANGED,
+				"InnoDB: insufficient history for index %u",
+				keynr);
+		}
+
 		/* The caller seems to ignore this.  Thus, we must check
 		this again in row_search_for_mysql(). */
 		DBUG_RETURN(2);
@@ -7518,6 +7566,10 @@ ha_innobase::records_in_range(
 		n_rows = HA_POS_ERROR;
 		goto func_exit;
 	}
+	if (dict_index_is_corrupted(index)) {
+		n_rows = HA_ERR_INDEX_CORRUPT;
+		goto func_exit;
+	}
 	if (UNIV_UNLIKELY(!row_merge_is_index_usable(prebuilt->trx, index))) {
 		n_rows = HA_ERR_TABLE_DEF_CHANGED;
 		goto func_exit;
@@ -8184,6 +8236,7 @@ ha_innobase::check(
 	ulint		n_rows_in_table	= ULINT_UNDEFINED;
 	ibool		is_ok		= TRUE;
 	ulint		old_isolation_level;
+	ibool		table_corrupted;
 
 	DBUG_ENTER("ha_innobase::check");
 	DBUG_ASSERT(thd == ha_thd());
@@ -8225,6 +8278,14 @@ ha_innobase::check(
 
 	prebuilt->trx->isolation_level = TRX_ISO_REPEATABLE_READ;
 
+	/* Check whether the table is already marked as corrupted
+	before running the check table */
+	table_corrupted = prebuilt->table->corrupted;
+
+	/* Reset table->corrupted bit so that check table can proceed to
+	do additional check */
+	prebuilt->table->corrupted = FALSE;
+
 	/* Enlarge the fatal lock wait timeout during CHECK TABLE. */
 	mutex_enter(&kernel_mutex);
 	srv_fatal_semaphore_wait_threshold += 7200; /* 2 hours */
@@ -8233,6 +8294,7 @@ ha_innobase::check(
 	for (index = dict_table_get_first_index(prebuilt->table);
 	     index != NULL;
 	     index = dict_table_get_next_index(index)) {
+		char	index_name[MAX_FULL_NAME_LEN + 1];
 #if 0
 		fputs("Validating index ", stderr);
 		ut_print_name(stderr, trx, FALSE, index->name);
@@ -8241,11 +8303,16 @@ ha_innobase::check(
 
 		if (!btr_validate_index(index, prebuilt->trx)) {
 			is_ok = FALSE;
+
+			innobase_format_name(
+				index_name, sizeof index_name,
+				prebuilt->index->name, TRUE);
+
 			push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
 					    ER_NOT_KEYFILE,
 					    "InnoDB: The B-tree of"
-					    " index '%-.200s' is corrupted.",
-					    index->name);
+					    " index %s is corrupted.",
+					    index_name);
 			continue;
 		}
 
@@ -8258,11 +8325,26 @@ ha_innobase::check(
 			prebuilt->trx, prebuilt->index);
 
 		if (UNIV_UNLIKELY(!prebuilt->index_usable)) {
-			push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
-					    HA_ERR_TABLE_DEF_CHANGED,
-					    "InnoDB: Insufficient history for"
-					    " index '%-.200s'",
-					    index->name);
+			innobase_format_name(
+				index_name, sizeof index_name,
+				prebuilt->index->name, TRUE);
+
+			if (dict_index_is_corrupted(prebuilt->index)) {
+				push_warning_printf(
+					user_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+					HA_ERR_INDEX_CORRUPT,
+					"InnoDB: Index %s is marked as"
+					" corrupted",
+					index_name);
+				is_ok = FALSE;
+			} else {
+				push_warning_printf(
+					thd, MYSQL_ERROR::WARN_LEVEL_WARN,
+					HA_ERR_TABLE_DEF_CHANGED,
+					"InnoDB: Insufficient history for"
+					" index %s",
+					index_name);
+			}
 			continue;
 		}
 
@@ -8276,12 +8358,19 @@ ha_innobase::check(
 		prebuilt->select_lock_type = LOCK_NONE;
 
 		if (!row_check_index_for_mysql(prebuilt, index, &n_rows)) {
+			innobase_format_name(
+				index_name, sizeof index_name,
+				index->name, TRUE);
+
 			push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
 					    ER_NOT_KEYFILE,
 					    "InnoDB: The B-tree of"
-					    " index '%-.200s' is corrupted.",
-					    index->name);
+					    " index %s is corrupted.",
+					    index_name);
 			is_ok = FALSE;
+			row_mysql_lock_data_dictionary(prebuilt->trx);
+			dict_set_corrupted(index);
+			row_mysql_unlock_data_dictionary(prebuilt->trx);
 		}
 
 		if (thd_killed(user_thd)) {
@@ -8308,6 +8397,20 @@ ha_innobase::check(
 		}
 	}
 
+	if (table_corrupted) {
+		/* If some previous operation has marked the table as
+		corrupted in memory, and has not propagated such to
+		clustered index, we will do so here */
+		index = dict_table_get_first_index(prebuilt->table);
+
+		if (!dict_index_is_corrupted(index)) {
+			mutex_enter(&dict_sys->mutex);
+			dict_set_corrupted(index);
+			mutex_exit(&dict_sys->mutex);
+		}
+		prebuilt->table->corrupted = TRUE;
+	}
+
 	/* Restore the original isolation level */
 	prebuilt->trx->isolation_level = old_isolation_level;
 
@@ -11101,6 +11204,11 @@ static MYSQL_SYSVAR_BOOL(large_prefix, innobase_large_prefix,
   "Support large index prefix length of REC_VERSION_56_MAX_INDEX_COL_LEN (3072) bytes.",
   NULL, NULL, FALSE);
 
+static MYSQL_SYSVAR_BOOL(force_load_corrupted, srv_load_corrupted,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Force InnoDB to load metadata of corrupted table.",
+  NULL, NULL, FALSE);
+
 static MYSQL_SYSVAR_BOOL(locks_unsafe_for_binlog, innobase_locks_unsafe_for_binlog,
   PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
   "Force InnoDB to not use next-key locking, to use only row-level locking.",
@@ -11360,6 +11468,7 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(flush_method),
   MYSQL_SYSVAR(force_recovery),
   MYSQL_SYSVAR(large_prefix),
+  MYSQL_SYSVAR(force_load_corrupted),
   MYSQL_SYSVAR(locks_unsafe_for_binlog),
   MYSQL_SYSVAR(lock_wait_timeout),
 #ifdef UNIV_LOG_ARCHIVE

storage/innobase/include/buf0lru.h

@@ -203,6 +203,17 @@ void
 buf_LRU_stat_update(void);
 /*=====================*/
 
+/******************************************************************//**
+Remove one page from LRU list and put it to free list */
+UNIV_INTERN
+void
+buf_LRU_free_one_page(
+/*==================*/
+	buf_page_t*	bpage)	/*!< in/out: block, must contain a file page and
+				be in a state where it can be freed; there
+				may or may not be a hash index to the page */
+	__attribute__((nonnull));
+
 #if defined UNIV_DEBUG || defined UNIV_BUF_DEBUG
 /**********************************************************************//**
 Validates the LRU list.

storage/innobase/include/db0err.h

@@ -110,6 +110,7 @@ enum db_err {
 					foreign keys as its prefix columns */
 	DB_TOO_BIG_INDEX_COL,		/* index column size exceeds maximum
 					limit */
+	DB_INDEX_CORRUPT,		/* we have corrupted index */
 
 	/* The following are partial failure codes */
 	DB_FAIL = 1000,

storage/innobase/include/dict0boot.h

@@ -137,8 +137,10 @@ dict_create(void);
 					header is created */
 /*-------------------------------------------------------------*/
 
-/* The field number of the page number field in the sys_indexes table
-clustered index */
+/* The field numbers in the SYS_TABLES clustered index */
+#define DICT_SYS_TABLES_TYPE_FIELD		5
+
+/* The field numbers in the SYS_INDEXES clustered index */
 #define DICT_SYS_INDEXES_PAGE_NO_FIELD	 8
 #define DICT_SYS_INDEXES_SPACE_NO_FIELD	 7
 #define DICT_SYS_INDEXES_TYPE_FIELD	 6

storage/innobase/include/dict0dict.h

@@ -585,6 +585,20 @@ dict_table_get_next_index(
 # define dict_table_get_next_index(index) UT_LIST_GET_NEXT(indexes, index)
 #endif /* UNIV_DEBUG */
 #endif /* !UNIV_HOTBACKUP */
+
+/* Skip corrupted index */
+#define dict_table_skip_corrupt_index(index)			\
+	while (index && dict_index_is_corrupted(index)) {	\
+		index = dict_table_get_next_index(index);	\
+	}
+
+/* Get the next non-corrupt index */
+#define dict_table_next_uncorrupted_index(index)		\
+do {								\
+	index = dict_table_get_next_index(index);		\
+	dict_table_skip_corrupt_index(index);			\
+} while (0)
+
 /********************************************************************//**
 Check whether the index is the clustered index.
 @return	nonzero for clustered index, zero for other indexes */
@@ -593,7 +607,7 @@ ulint
 dict_index_is_clust(
 /*================*/
 	const dict_index_t*	index)	/*!< in: index */
-	__attribute__((pure));
+	__attribute__((nonnull, pure, warn_unused_result));
 /********************************************************************//**
 Check whether the index is unique.
 @return	nonzero for unique index, zero for other indexes */
@@ -602,7 +616,7 @@ ulint
 dict_index_is_unique(
 /*=================*/
 	const dict_index_t*	index)	/*!< in: index */
-	__attribute__((pure));
+	__attribute__((nonnull, pure, warn_unused_result));
 /********************************************************************//**
 Check whether the index is the insert buffer tree.
 @return	nonzero for insert buffer, zero for other indexes */
@@ -611,7 +625,7 @@ ulint
 dict_index_is_ibuf(
 /*===============*/
 	const dict_index_t*	index)	/*!< in: index */
-	__attribute__((pure));
+	__attribute__((nonnull, pure, warn_unused_result));
 /********************************************************************//**
 Check whether the index is a secondary index or the insert buffer tree.
 @return	nonzero for insert buffer, zero for other indexes */
@@ -620,7 +634,7 @@ ulint
 dict_index_is_sec_or_ibuf(
 /*======================*/
 	const dict_index_t*	index)	/*!< in: index */
-	__attribute__((pure));
+	__attribute__((nonnull, pure, warn_unused_result));
 
 /********************************************************************//**
 Gets the number of user-defined columns in a table in the dictionary
@@ -630,7 +644,8 @@ UNIV_INLINE
 ulint
 dict_table_get_n_user_cols(
 /*=======================*/
-	const dict_table_t*	table);	/*!< in: table */
+	const dict_table_t*	table)	/*!< in: table */
+	__attribute__((nonnull, pure, warn_unused_result));
 /********************************************************************//**
 Gets the number of system columns in a table in the dictionary cache.
 @return	number of system (e.g., ROW_ID) columns of a table */
@@ -638,7 +653,8 @@ UNIV_INLINE
 ulint
 dict_table_get_n_sys_cols(
 /*======================*/
-	const dict_table_t*	table);	/*!< in: table */
+	const dict_table_t*	table)	/*!< in: table */
+	__attribute__((nonnull, pure, warn_unused_result));
 /********************************************************************//**
 Gets the number of all columns (also system) in a table in the dictionary
 cache.
@@ -647,7 +663,8 @@ UNIV_INLINE
 ulint
 dict_table_get_n_cols(
 /*==================*/
-	const dict_table_t*	table);	/*!< in: table */
+	const dict_table_t*	table)	/*!< in: table */
+	__attribute__((nonnull, pure, warn_unused_result));
 #ifdef UNIV_DEBUG
 /********************************************************************//**
 Gets the nth column of a table.
@@ -1243,6 +1260,56 @@ void
 dict_close(void);
 /*============*/
 
+/**********************************************************************//**
+Check whether the table is corrupted.
+@return	nonzero for corrupted table, zero for valid tables */
+UNIV_INLINE
+ulint
+dict_table_is_corrupted(
+/*====================*/
+	const dict_table_t*	table)	/*!< in: table */
+	__attribute__((nonnull, pure, warn_unused_result));
+
+/**********************************************************************//**
+Check whether the index is corrupted.
+@return	nonzero for corrupted index, zero for valid indexes */
+UNIV_INLINE
+ulint
+dict_index_is_corrupted(
+/*====================*/
+	const dict_index_t*	index)	/*!< in: index */
+	__attribute__((nonnull, pure, warn_unused_result));
+
+/**********************************************************************//**
+Flags an index and table corrupted both in the data dictionary cache
+and in the system table SYS_INDEXES. */
+UNIV_INTERN
+void
+dict_set_corrupted(
+/*===============*/
+	dict_index_t*	index)		/*!< in/out: index */
+	UNIV_COLD __attribute__((nonnull));
+
+/**********************************************************************//**
+Flags an index corrupted in the data dictionary cache only. This
+is used mostly to mark a corrupted index when index's own dictionary
+is corrupted, and we force to load such index for repair purpose */
+UNIV_INTERN
+void
+dict_set_corrupted_index_cache_only(
+/*================================*/
+	dict_index_t*	index);		/*!< in/out: index */
+
+/**********************************************************************//**
+Flags a table with specified space_id corrupted in the table dictionary
+cache.
+@return TRUE if successful */
+UNIV_INTERN
+ibool
+dict_set_corrupted_by_space(
+/*========================*/
+	ulint		space_id);	/*!< in: space ID */
+
 #ifndef UNIV_NONINL
 #include "dict0dict.ic"
 #endif

storage/innobase/include/dict0dict.ic

@@ -27,6 +27,7 @@ Created 1/8/1996 Heikki Tuuri
 #ifndef UNIV_HOTBACKUP
 #include "dict0load.h"
 #include "rem0types.h"
+#include "srv0srv.h"
 
 /*********************************************************************//**
 Gets the minimum number of bytes per character.
@@ -828,7 +829,7 @@ dict_table_check_if_in_cache_low(
 }
 
 /**********************************************************************//**
-load a table into dictionary cache, ignore any error specified during load; 
+load a table into dictionary cache, ignore any error specified during load;
 @return	table, NULL if not found */
 UNIV_INLINE
 dict_table_t*
@@ -872,6 +873,18 @@ dict_table_get_low(
 
 	table = dict_table_check_if_in_cache_low(table_name);
 
+	if (table && table->corrupted) {
+		fprintf(stderr, "InnoDB: table");
+		ut_print_name(stderr, NULL, TRUE, table->name);
+		if (srv_load_corrupted) {
+			fputs(" is corrupted, but"
+			      " innodb_force_load_corrupted is set\n", stderr);
+		} else {
+			fputs(" is corrupted\n", stderr);
+			return(NULL);
+		}
+	}
+
 	if (table == NULL) {
 		table = dict_load_table(table_name, TRUE, DICT_ERR_IGNORE_NONE);
 	}
@@ -937,4 +950,35 @@ dict_max_field_len_store_undo(
 	return(prefix_len);
 }
 
+/********************************************************************//**
+Check whether the table is corrupted.
+@return	nonzero for corrupted table, zero for valid tables */
+UNIV_INLINE
+ulint
+dict_table_is_corrupted(
+/*====================*/
+	const dict_table_t*	table)	/*!< in: table */
+{
+	ut_ad(table);
+	ut_ad(table->magic_n == DICT_TABLE_MAGIC_N);
+
+	return(UNIV_UNLIKELY(table->corrupted));
+}
+
+/********************************************************************//**
+Check whether the index is corrupted.
+@return	nonzero for corrupted index, zero for valid indexes */
+UNIV_INLINE
+ulint
+dict_index_is_corrupted(
+/*====================*/
+	const dict_index_t*	index)	/*!< in: index */
+{
+	ut_ad(index);
+	ut_ad(index->magic_n == DICT_INDEX_MAGIC_N);
+
+	return(UNIV_UNLIKELY((index->type & DICT_CORRUPT)
+	       || (index->table && index->table->corrupted)));
+}
+
 #endif /* !UNIV_HOTBACKUP */

storage/innobase/include/dict0mem.h

@@ -51,7 +51,12 @@ combination of types */
 #define DICT_UNIQUE	2	/*!< unique index */
 #define	DICT_UNIVERSAL	4	/*!< index which can contain records from any
 				other index */
-#define	DICT_IBUF 	8	/*!< insert buffer tree */
+#define	DICT_IBUF	8	/*!< insert buffer tree */
+#define	DICT_CORRUPT	16	/*!< bit to store the corrupted flag
+				in SYS_INDEXES.TYPE */
+
+#define	DICT_IT_BITS	5	/*!< number of bits used for
+				SYS_INDEXES.TYPE */
 /* @} */
 
 /** Types for a table object */
@@ -369,8 +374,9 @@ struct dict_index_struct{
 				/*!< space where the index tree is placed */
 	unsigned	page:32;/*!< index tree root page number */
 #endif /* !UNIV_HOTBACKUP */
-	unsigned	type:4;	/*!< index type (DICT_CLUSTERED, DICT_UNIQUE,
-				DICT_UNIVERSAL, DICT_IBUF) */
+	unsigned	type:DICT_IT_BITS;
+				/*!< index type (DICT_CLUSTERED, DICT_UNIQUE,
+				DICT_UNIVERSAL, DICT_IBUF, DICT_CORRUPT) */
 	unsigned	trx_id_offset:10;/*!< position of the trx id column
 				in a clustered index record, if the fields
 				before it are known to be of a fixed size,
@@ -391,8 +397,6 @@ struct dict_index_struct{
 				/*!< TRUE if this index is marked to be
 				dropped in ha_innobase::prepare_drop_index(),
 				otherwise FALSE */
-	unsigned	corrupted:1;
-				/*!< TRUE if the index object is corrupted */
 	dict_field_t*	fields;	/*!< array of field descriptions */
 #ifndef UNIV_HOTBACKUP
 	UT_LIST_NODE_T(dict_index_t)

storage/innobase/include/dict0types.h

@@ -51,7 +51,8 @@ be or-ed together */
 enum dict_err_ignore {
         DICT_ERR_IGNORE_NONE = 0,        /*!< no error to ignore */
         DICT_ERR_IGNORE_INDEX_ROOT = 1, /*!< ignore error if index root
-					page is FIL_NUL or incorrect value */
+					page is FIL_NULL or incorrect value */
+	DICT_ERR_IGNORE_CORRUPT = 2,	/*!< skip corrupted indexes */
         DICT_ERR_IGNORE_ALL = 0xFFFF	/*!< ignore all errors */
 };
 

storage/innobase/include/srv0srv.h

@@ -141,6 +141,10 @@ extern ulint	srv_log_buffer_size;
 extern ulong	srv_flush_log_at_trx_commit;
 extern char	srv_adaptive_flushing;
 
+/* If this flag is TRUE, then we will load the indexes' (and tables') metadata
+even if they are marked as "corrupted". Mostly it is for DBA to process
+corrupted index and table */
+extern my_bool	srv_load_corrupted;
 
 /* The sort order table of the MySQL latin1_swedish_ci character set
 collation */

storage/innobase/pars/pars0opt.c

@@ -568,7 +568,7 @@ opt_search_plan_for_table(
 			best_last_op = last_op;
 		}
 
-		index = dict_table_get_next_index(index);
+		dict_table_next_uncorrupted_index(index);
 	}
 
 	plan->index = best_index;

storage/innobase/row/row0ins.c

@@ -118,6 +118,9 @@ ins_node_create_entry_list(
 					      node->entry_sys_heap);
 		UT_LIST_ADD_LAST(tuple_list, node->entry_list, entry);
 
+		/* We will include all indexes (include those corrupted
+		secondary indexes) in the entry list. Filteration of
+		these corrupted index will be done in row_ins() */
 		index = dict_table_get_next_index(index);
 	}
 }
@@ -2046,7 +2049,6 @@ row_ins_index_entry_low(
 			mtr_start(&mtr);
 
 			if (err != DB_SUCCESS) {
-
 				goto function_exit;
 			}
 
@@ -2431,6 +2433,13 @@ row_ins(
 
 		node->index = dict_table_get_next_index(node->index);
 		node->entry = UT_LIST_GET_NEXT(tuple_list, node->entry);
+
+		/* Skip corrupted secondar index and its entry */
+		while (node->index && dict_index_is_corrupted(node->index)) {
+
+			node->index = dict_table_get_next_index(node->index);
+			node->entry = UT_LIST_GET_NEXT(tuple_list, node->entry);
+		}
 	}
 
 	ut_ad(node->entry == NULL);

storage/innobase/row/row0merge.c

@@ -2554,8 +2554,9 @@ row_merge_is_index_usable(
 	const trx_t*		trx,	/*!< in: transaction */
 	const dict_index_t*	index)	/*!< in: index to check */
 {
-	return(!trx->read_view
-	       || read_view_sees_trx_id(trx->read_view, index->trx_id));
+	return(!dict_index_is_corrupted(index)
+	       && (!trx->read_view
+	           || read_view_sees_trx_id(trx->read_view, index->trx_id)));
 }
 
 /*********************************************************************//**

storage/innobase/row/row0mysql.c

@@ -3098,7 +3098,8 @@ row_drop_table_for_mysql(
 	ut_ad(rw_lock_own(&dict_operation_lock, RW_LOCK_EX));
 #endif /* UNIV_SYNC_DEBUG */
 
-	table = dict_table_get_low_ignore_err(name, DICT_ERR_IGNORE_INDEX_ROOT);
+	table = dict_table_get_low_ignore_err(
+		name, DICT_ERR_IGNORE_INDEX_ROOT | DICT_ERR_IGNORE_CORRUPT);
 
 	if (!table) {
 		err = DB_TABLE_NOT_FOUND;

storage/innobase/row/row0purge.c

@@ -469,6 +469,13 @@ row_purge_del_mark(
 	heap = mem_heap_create(1024);
 
 	while (node->index != NULL) {
+		/* skip corrupted secondary index */
+		dict_table_skip_corrupt_index(node->index);
+
+		if (!node->index) {
+			break;
+		}
+
 		index = node->index;
 
 		/* Build the index entry */
@@ -516,6 +523,12 @@ row_purge_upd_exist_or_extern_func(
 	heap = mem_heap_create(1024);
 
 	while (node->index != NULL) {
+		dict_table_skip_corrupt_index(node->index);
+
+		if (!node->index) {
+			break;
+		}
+
 		index = node->index;
 
 		if (row_upd_changes_ord_field_binary(node->index, node->update,

storage/innobase/row/row0sel.c

@@ -3441,6 +3441,13 @@ row_search_for_mysql(
 		return(DB_MISSING_HISTORY);
 	}
 
+	if (dict_index_is_corrupted(index)) {
+#ifdef UNIV_SYNC_DEBUG
+		ut_ad(!sync_thread_levels_nonempty_trx(trx->has_search_latch));
+#endif /* UNIV_SYNC_DEBUG */
+		return(DB_CORRUPTION);
+	}
+
 	if (UNIV_UNLIKELY(prebuilt->magic_n != ROW_PREBUILT_ALLOCATED)) {
 		fprintf(stderr,
 			"InnoDB: Error: trying to free a corrupt\n"

storage/innobase/row/row0uins.c

@@ -328,6 +328,8 @@ row_undo_ins(
 	node->index = dict_table_get_next_index(
 		dict_table_get_first_index(node->table));
 
+	dict_table_skip_corrupt_index(node->index);
+
 	while (node->index != NULL) {
 		dtuple_t*	entry;
 		ulint		err;
@@ -355,7 +357,7 @@ row_undo_ins(
 			}
 		}
 
-		node->index = dict_table_get_next_index(node->index);
+		dict_table_next_uncorrupted_index(node->index);
 	}
 
 	log_free_check();

storage/innobase/row/row0umod.c

@@ -573,6 +573,14 @@ row_undo_mod_upd_del_sec(
 	heap = mem_heap_create(1024);
 
 	while (node->index != NULL) {
+
+		/* Skip all corrupted secondary index */
+		dict_table_skip_corrupt_index(node->index);
+
+		if (!node->index) {
+			break;
+		}
+
 		index = node->index;
 
 		entry = row_build_index_entry(node->row, node->ext,
@@ -626,6 +634,13 @@ row_undo_mod_del_mark_sec(
 	heap = mem_heap_create(1024);
 
 	while (node->index != NULL) {
+		/* Skip all corrupted secondary index */
+		dict_table_skip_corrupt_index(node->index);
+
+		if (!node->index) {
+			break;
+		}
+
 		index = node->index;
 
 		entry = row_build_index_entry(node->row, node->ext,
@@ -677,6 +692,13 @@ row_undo_mod_upd_exist_sec(
 	heap = mem_heap_create(1024);
 
 	while (node->index != NULL) {
+		/* Skip all corrupted secondary index */
+		dict_table_skip_corrupt_index(node->index);
+
+		if (!node->index) {
+			break;
+		}
+
 		index = node->index;
 
 		if (row_upd_changes_ord_field_binary(node->index, node->update,
@@ -859,6 +881,9 @@ row_undo_mod(
 	node->index = dict_table_get_next_index(
 		dict_table_get_first_index(node->table));
 
+	/* Skip all corrupted secondary index */
+	dict_table_skip_corrupt_index(node->index);
+
 	if (node->rec_type == TRX_UNDO_UPD_EXIST_REC) {
 
 		err = row_undo_mod_upd_exist_sec(node, thr);

storage/innobase/row/row0upd.c

@@ -2320,6 +2320,13 @@ row_upd(
 
 	while (node->index != NULL) {
 
+		/* Skip corrupted index */
+		dict_table_skip_corrupt_index(node->index);
+
+		if (!node->index) {
+			break;
+		}
+
 		log_free_check();
 		err = row_upd_sec_step(node, thr);
 

storage/innobase/ut/ut0ut.c

@@ -712,6 +712,8 @@ ut_strerr(
 		return("No index on referencing keys in referencing table");
 	case DB_PARENT_NO_INDEX:
 		return("No index on referenced keys in referenced table");
+	case DB_INDEX_CORRUPT:
+		return("Index corrupted");
 	case DB_END_OF_INDEX:
 		return("End of index");
 	/* do not add default: in order to produce a warning if new code