Merge lp:~stewart/percona-server/5.6-corrupt-table-action into lp:percona-server/5.6

[Stewart Smith]

This is just a post-review bug fix, based on one comment by George.

[George Ormond Lorch III (gl-az)]

review: Approve (g2)

[Laurynas Biveinis (laurynas-biveinis)]

Any new 5.6 srv_pass_corrupt_table checks must follow the 5.5 bug 1125248 idiom.

review: Needs Fixing

Unmerged revisions

312. By Stewart Smith on 2013-02-07

post-review fix for corrupt_table_action: btr_height_get call chain could end up in an error condition that wasn't handled. Let's handle it somewhat properly so we don't die in an assert/segv.

=== modified file 'Percona-Server/storage/innobase/dict/dict0stats.cc'
--- Percona-Server/storage/innobase/dict/dict0stats.cc	2013-01-18 04:53:12 +0000
+++ Percona-Server/storage/innobase/dict/dict0stats.cc	2013-03-14 05:45:26 +0000
@@ -1808,6 +1808,14 @@
 
 	root_level = btr_height_get(index, &mtr);
 
+	if (root_level == -1 && srv_pass_corrupt_table) {
+		fprintf(stderr, " InnoDB: Index %s likely corrupted (found in "
+			"%s)\n",
+			index->name, __func__);
+		mtr_commit(&mtr);
+		return;
+	}
+
 	n_uniq = dict_index_get_n_unique(index);
 
 	/* If the tree has just one level (and one page) or if the user
 
=== modified file 'Percona-Server/storage/innobase/include/btr0btr.ic'
--- Percona-Server/storage/innobase/include/btr0btr.ic	2013-03-05 12:46:43 +0000
+++ Percona-Server/storage/innobase/include/btr0btr.ic	2013-03-14 05:45:26 +0000
@@ -120,6 +120,9 @@
 
 	ut_ad(page);
 
+	if (srv_pass_corrupt_table && page == NULL)
+		return(-1);
+
 	level = mach_read_from_2(page + PAGE_HEADER + PAGE_LEVEL);
 
 	ut_ad(level <= BTR_MAX_NODE_LEVEL);