CUMULATIVE PATCH FOR VxFEN FOR VERITAS CLUSTER SERVER 5.0MP1 ============================================================ NAME: 5.0MP1CP1 DATE: 2007-August-17 VCS RELEASE: 5.0MP1 LINUX RELEASE: RHEL 4 U4 RELEVANT ARCHITECTURES: x86_64 BUGS FIXED WITH THIS PATCH: The following is a brief description of the bugs fixed in this patch. ----- Incident Number: 1051193 Abstract: vxfen unconfigure uses wrong key to preempt/abort from now faulted paths. Brief Description: When the fencing driver is stopped, it attempts to preempt/abort its own keys registered with the coordinator disks. However, this key may be incorrect. In some setups, this may lead to preempting other nodes keys leading to the panic of all nodes in the cluster. ----- Incident Number: 896781 Abstract: system panics intermitently after disabling array side switch port Brief Description: This bug only manifests if the fencing driver is configured to use DMP paths to coordinator disks. Before we do a preempt and abort, we do dummy IOs to the coordinator disks to ensure that keys are correctly failed over if the paths have failed over before this event. In this operation, there was a bug dur to which the driver could sleep in an interrupt. If this happens, the stack trace is as follows: Call Trace: [c00000000ffef680] [c00000000005bd98] .__might_sleep+0xcc/0xec (unreliable) [c00000000ffef720] [d000000000b7e5ec] .vxfen_add_deblog+0x68/0x130 [vxfen] [c00000000ffef7b0] [d000000000b95e4c] .vxfen_dummy_bio_write_complete+0x90/0x118 [vxfen] [c00000000ffef860] [c0000000000c7b48] .bio_endio+0xc0/0xe0 [c00000000ffef8f0] [d0000000007281c4] .gendmpiodone+0x1c8/0x228 [vxdmp] [c00000000ffef9a0] [d000000000728ea4] .dmpiodone+0xec/0x120 [vxdmp] [c00000000ffefa50] [c0000000000c7b48] .bio_endio+0xc0/0xe0 [c00000000ffefae0] [c0000000001e4638] .__end_that_request_first+0x12c/0x26c [c00000000ffefba0] [d00000000007a488] .scsi_end_request+0x44/0x130 [scsi_mod] [c00000000ffefc40] [d00000000007a97c] .scsi_io_completion+0x224/0x44c [scsi_mod] [c00000000ffefd10] [d00000000003c214] .sd_rw_intr+0x2a8/0x2e0 [sd_mod] [c00000000ffefdc0] [d000000000074318] .scsi_finish_command+0x10c/0x130 [scsi_mod] [c00000000ffefe50] [d000000000074190] .scsi_softirq+0x140/0x168 [scsi_mod] [c00000000ffefef0] [c000000000066084] .__do_softirq+0xa0/0x17c [c00000000ffeff90] [c000000000018580] .call_do_softirq+0x14/0x24 [c00000000250f9e0] [c0000000000144a8] .do_softirq+0x74/0x9c [c00000000250fa70] [c000000000013e64] .do_IRQ+0xe8/0x100 [c00000000250faf0] [c00000000000ae34] HardwareInterrupt_entry+0x8/0x54 ----- Incident Number: 1057418 Abstract: more page 83 changes since PillarData Array's serial number is implemented in an offset that is different from HDS and EMC In 5.0MP1 release, I/O fencing can not retrieve the serial number of luns from an array manufactured by PillarData. ---- CKSUM: ------ 552664454 625700 VRTSvxfen-5.0.10.01-CP_MP1_RHEL4.x86_64.rpm PATCH INSTALLATION INSTRUCTIONS: -------------------------------- The patch needs to be installed after installing Veritas Cluster Server 5.0MP1. ------------------------------------------------- *** Steps to be run on any system in the cluster: ------------------------------------------------- 1. Stop VCS and CVM on all nodes. CVM is usually under VCS control and so will automatically come down when VCS is stopped using the following command on any node in the cluster: "hastop -all" --------------------------------------------------- *** Steps to be run on every system in the cluster: --------------------------------------------------- 2. Log on as superuser on the system where the point patch is to be installed. 3. Stop VxFEN module using the following command: "/etc/init.d/vxfen stop" 4. Upgrade VRTSvxfen package using the following command: "rpm -Uvh VRTSvxfen-5.0.10.01-CP_MP1_RHEL4.x86_64.rpm" Where VRTSvxfen-5.0.10.01-CP_MP1_RHEL4.x86_64.rpm is the package rpm provided in the point patch 5. Bring up VxFEN by using the following command: "/etc/init.d/vxfen start" 6. Bring up VCS on all the nodes "hastart"