Topic
  • No replies
gpfs@us.ibm.com
gpfs@us.ibm.com
312 Posts

Pinned topic Abstract: GPFS directory corruption with possible undetected data corruption

‏2014-04-09T14:27:20Z |

Abstract: GPFS directory corruption with possible undetected data corruption

Problem Summary: When multiple nodes are updating a shared directory concurrently, the problem could cause incorrect results from directory operations issued on one node, leading to orphaned inodes (files inaccessible from any directory entry), or directory entries pointing to deleted or incorrect files.   This problem could also cause silent data corruption, if any disk contains both GPFS metadata and data, and a stale buffer is written to a disk address that has been freed and reallocated for some other purpose.

Users affected (both of the following conditions must apply for customer to be affected):
1. GPFS service levels 3.4.0.24, 3.4.0.25, 3.4.0.26, 3.4.0.27, 3.5.0.13. 3.5.0.14, 3.5.0.15, or 3.5.0.16.
2. Workload consists of concurrent directory updates from multiple nodes.

Problem Description: See Problem Summary.

Recommendation:  Customers who have run the affected service levels should upgrade to GPFS 3.5.0.17 or 3.4.0.28 (when available) service level updates (go to Fix Central http://www.ibm.com/eserver/support/fixes/), or should apply efixes for the affected service levels.Customers who have seen FSSTRUCT 1124 or 1122 messages, or EIO errors during directory operations, should also run off-line fsck to identify and repair possible directory damage.