[LU-9903] kernel update [RHEL6.9 2.6.32-696.10.1.el6]

• Status: Resolved
• Project: Lustre
• Type: Bug
• Priority: Minor
• Reporter: Bob Glossman (Inactive)
• Assignee: Bob Glossman (Inactive)
• Resolution: Won't Fix
• Severity: 3

Description

This update fixes the following bugs:

When running a Red Hat Enterprise Linux 6.8 VM with audit watches on
directories experiencing a high level activity, the system slowed due
to unnecessary requests for the process' credentials; DB2 workloads in
particular were affected. The provided patch is more selective about
acquiring process credentials,thus reducing the audit overhead and
restoring performance. (BZ#1460164)

With a sunrpc regression in Red Hat Enterprise Linux 6.9, timer code was removed that allowed the NFS client to reset a TCP connection stuck in FIN-WAIT-2 state. This fix adds TCP keepalives for NFS client TCP connections and allows the NFS client to recover the TCP connection if stuck in FIN-WAIT-2. (BZ#1462094)

Due to a sunrpc regression introduced in Red Hat Enterprise Linux 6.9 , an NFS client with TCP timestamps initiated a TCP disconnect sequence: the NFS TCP connection could not be reconnected for 60 seconds during the TIME_WAIT state because the source TCP port could not be re-used. As a consequence, during this 60 second period multiple side-effects occurred including unresponsive NFS mount points, an rpciod kernel thread consuming 100% CPU, the "retrans" number of "nfsstat -r" becoming a very large number. In addition, certain mount options returned "not responding" errors and even I/O errors could occur. With this update, a different source port is selected when an NFS TCP connection needs to reconnect during TIME_WAIT. As a result, the NFS TCP connection can reconnect immediately after a disconnect sequence and no longer waits 60 seconds for TIME_WAIT to complete. (BZ#1472128)

While running automated array reboots on the InfiniBand Host Channel Adapter (HCA), the system experienced a kernel panic with the crash dump reporting the "Hard Lockup". The provided set of patches ensures the reboot path A fails over to the reboot path B without entering a kernel panic. (BZ#1462097)

If multiple tasks attempted to read statistics for Fibre Channel over Ethernet (FCoE) Host Bus Adapter (HBA), the start_req_done completion could be re-initialized while still being used by another task. Consequently, the system crash occurred with the crash dump reporting the "Hard Lockup". This patch adds a mutex to serialize the calls to the bnx2fc_get_host_stats() function, thus fixing this bug. (BZ#1467323)

Issue Links

• is related to: LU-9763 kernel update [RHEL6.9 2.6.32-696.6.3...
• is related to: LU-9979 kernel update [RHEL6.9 2.6.32-696.10....

Comments

[Gerrit Updater]

Bob Glossman ([e-mail]) uploaded a new patch: https://review.whamcloud.com/28684
Subject: LU-9903 kernel: kernel update RHEL6.9 [2.6.32-696.10.1.el6]
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 9db3c03b59fe7179029c849a0bd588e7bbaf4458

[Gerrit Updater]

Bob Glossman ([e-mail]) uploaded a new patch: https://review.whamcloud.com/28685
Subject: LU-9903 kernel: kernel update RHEL6.9 [2.6.32-696.10.1.el6]
Project: fs/lustre-release
Branch: b2_10
Current Patch Set: 1
Commit: eac0c4328656c5d4d0874bcec3f8e89ee3dc873d

[Gerrit Updater]

John L. Hammond ([e-mail]) merged in patch https://review.whamcloud.com/28685/
Subject: LU-9903 kernel: kernel update RHEL6.9 [2.6.32-696.10.1.el6]
Project: fs/lustre-release
Branch: b2_10
Current Patch Set:
Commit: e1d3a0e780ab3af77ff05ad159e453745cf2fe7e

[Bob Glossman (Inactive)]

Replaced by LU-9979, a later kernel version update for el6.9