Brocade SAN Switch Fabric Operating System firmware v5.3.2a - IBM BladeCenter

Download Description

• New IBM xSeries firmware and device driver file naming convention

** Important Note: Fabric OS v5.x supports 2-level step migrations.
To upgrade to FOS v5.3.2a, use the following upgrade path: v4.2.1a to v4.4.1a to v5.0.5d to v5.2.1b to v5.3.2a

Severity: Suggested

Version 5.3.2a

• Brocade Fabric OS v5.3.2a is a maintenance release.

Problem(s) Fixed:

• see Brocade Fabric OS v5.3.2a Release Notes v2.0 (brcdFOS_v5.3.2a_releasenotes_v1.0.pdf)

Version 5.3.1

• Brocade Fabric OS v5.3.1 is a maintenance release.

Problem(s) Fixed:

• see Brocade Fabric OS v5.3.1 Release Notes v2.0 (brcdFOS_v5.3.1_releasenotes_v2.0.pdf)

Version 5.3.0a

Problem(s) Fixed:

• Switch panic when doing a data migration using DMM. DMM issues an invalid frame to the switch causing the panic. Made single sequence frame processing logic robust so that the Fibre Channel information unit is not freed before it is fully processed.
• Diagnostic failure and FA4-18 fault after obernight cable pull failover test. Removes EMD reinitialization for framer path test. Updated error handling when link goes down from link up state.
• Unable to configure FCIP tunnels using Web Tools. Added FCIP wizard and stand-alone tab support for near end tunnel configuration.
• In frame redirect mode, the GA_NXT query from initiator returns the VI PID and PWWN instead of host PWWN. Fixed GA_NZT processing login in frame redirect mode.
• In frame redirect mode, the GPN_NN query returns the VI PWWN and PID instead of initiator PWWN and PID. Added translation from PWWN to PID for remote devices in the GPN_NN query logic in frame redirection mode.
• CP reboots after running for 3 days. Fixed memory leak issue in NS API/library that was causing scpd to eventually crash.

Version 5.2.1b

Problem(s) Fixed:

• A PLOGI ACC (port login accept) frame is not routed to the destination virtual port when the host and target are on the same NPIV (N Port ID Virtualization)port. The traffic path between the host and the target will not be established. This does not impact hosts and devices that are on different NPIV ports or fabrics not using NPIV. The code has been fixed to properly set up the content addressable memory (CAM) entry when the host and target are on the same NPIV port.
• Using the API (application program interface) to access default zoning information may cause the switch to panic. If an application is using the API to retreive ZoneCapabilityobj, the swicth reboots with error message: "[KSWD-1003], 2371, FFDC, WARNING,ED 48000B, kSWD: Detected unexpected termination of [22]cald:0'RfP=727,RgP=727,DfP=0,died=1,rt=278." Code has been fixed to avoid using an invalid pointer.
• During a small window, if a device sends its PLOGI very fast and before the name server information has been fully propogated, the PLOGI might be dropped by a switch using domain/port zoning. The host will not see the device. The code has been fixed to ensure that it properly responds to the FCGS command to retrieve the index for the fast PLOGI device used by domain/port zoning.
• Host loses target for 15 seconds after a non-disruptive failover on an Access Gateway-enabled switch. Any event that triggers a hareboot, such as a firmware download or a switch paninc might cause the host to temporarily lose visibility of the target on the Access Gateway port. This will interrupt traffic. The port now relies on the FLOGI timeframe to send the port-on-line update and to properly set up the internal data structure.
• Switch panic when a kernel process is spawned by a non-root user. "free_uid()" is on the console log backtrace. The code will allow the correct tracking of the root process.

Version 5.0.5d

Problem(s) Fixed:

• On 2 Gbps platform a switch panic can occur if an unstable or faulty port generates an excessive number of interrupts. The switch executes a panic reboot because the link generates more interrupts than the switch could process. The firmware now updates the port fault counter when the link stays in an active state and there is a loss of signal to allow a port fault to be triggered during the resource allocation time period.
• Switch panic is issured when the Remote Procedure Call Daemon (RPCD) attempts to retrieve a non-initialized variable. Switch panics with error message "[KSWD-1003], 1869, WARNING, kSWD: Detected unexpected termination of...". The code was changed to properlt initialize data structures after allocation when there is no Link Incident Record Registration (LIRR). The code was also changed to properly process the response from SEC_GET_BANNER.
• When an FCP probe failure occurs, the failing port gets stuck in an "In_Sync" state. As a result the device does not show up in the name server because of the traffic interruption through this port. The switch continuously reports "Probing failed on L-port ALAP ef warnings: [FCPD-1001], 68601, WARNING". The code has been fixed to properly handle the lost probing response sequences on a link that has been reset after the first multi-frame response has been sent.
• The API library can crash during the activation of a big zoneset with 1276 zones. A 3rd party application using the Brocade API to perform zonin administration calls crashes during activation of the big zoneset. The zone library now appends a terminating zero to all active zone database data buffers returned by the API during zone buffer copy.
• The zoneObjectRename command allows the creation of duplicate aliases which causes database corruption. The new zone group that is being renamed will be checked for its existence in the zone database. If the entry exists, zoneObjectRename will not allow duplicate zone aliases.
• If an invalid IP address is in the syslog.conf file, using the echo command caused the raslogd init to fail which causes the switch to panic. The code now prevents the user from accessing root commands.
• A high number of CRC (Cyclic Redundancy Check) and LLI (Low Level Interrupt) errors are reported bt the porterrshow command. The proper but rate was set in the hardware register after port speed is negotiated.
• The WebTools GUI hangs upon invoking fabric wide polling when there is no IP connection between the Web Tools proxy switch and other switches in the fabric. On a large fabric, the GUI hangs upon launch and no access via Web Tools is possible. The firmware now provides two configurable parameters to allow the user to selectively disable fabric wide polling if there is not IP connectivity between all switches.
• Zoning change fails with message "unabortable internal zoning transaction". The new firmware removes a condition where zoning is blocked by the transport layer.
• Switch panics with time service daemon (tsd) core dump. The time service daemon has been changed to update the hardware clock asynchronously.
• The pdshow command displays "Out of Memory: Killed process 638 (nsd0). VM size - 18380 KB, Runtime = 9844 minutes, CPU time = 423 sec." and RASLOG entry similar to: "[KSWD-1003],195, WARNING, Brocade 3900, kSWD: Detected unexpected termination of..." The code has been fixed to free IU when an HBA performs a PLOGI to a target not yet register with the switch name server and the PID in the PLOGI frame has a non-zero Arbitrated Loop Physical Access (ALPA) for the unknown target.
• HBA ocassionally may not log into a 4GB switch during auto topology discovery. The port connecting to a 3rd party HBA may not come up during initialization. This will happen as a result of a maintenance activity or a severe error condition that would cause the link to force a full reset to recover. A fabric may see a few ports not come back online. The code now sends a CLD/LIP/OLS followed by a loop init to bring the port back up.
• Fixes the error message seen with portCfgPort to explain that the command cannot be applied to ports connected to internal server blades.
• Frame may be dropped in NPIV device load test environment may impact a 4Gig switch with multiple devices attached to a single Single host talking to multiple targets in session-based zoning.
• In interop mode, process login accept (PLOGI ACC) frame to n_ports is dropped by switch. The problem occurs in a configuration with a SilkWorm switch running Fabric OS v4.4.0 or later and 3rd party switches in interop mode. The same frame drop does not occur on pre-Fabric OS v4.4.0 switches.
• Host cannot see target after a switch running Fabric OS v5.0.x reboots while establishing session for a specific target. Symptom: If the SCSI query failed while the switch was establishing a session with the target after process login to an n_port (PLOGI), the device was not considered a fibre channel protocol (FCP) device even though the process login (PRLI) ACCEPT assigned the FCP type.
• merging a Fabric OS v5.0.x switch into a fabric running Fabric OS v4.2.x or earlier on making zoning changes using Web Tools on Fabric OS v5.0.x results in the host losing access to the target.
• SilkWorm 3016 over-voltage error: false error is reported to the log.
• SNMP walk on MIB-2 increases the memory usage of SNMP daemon
• API returns error intermittently when trying to establish a session with the proxy switch when FM tries to retrieve Port Stats and End-to-End Monitoring.
• Blade/Brocade 4GB: (20PT) third party 4GB HBA daughter card not working with third party switch chassis.

Enhancements:

• added support for the SilkWorm 4020
• added support for Dynamic Ports on Demand Licensing for the 10-Port 4Gb Brocade Switch Module. Refer to the readme and release notes for more details on using this function.

Version 4.4.1a

• Using cfgEnable to enable a new zoning configuration that exceeds the CAM limit on a port group will cause traffic to be dropped. Issuing a cfgEnable command will cause a port group to stop sending traffic if the new configuration causes the port group to exceed its CAM limit. Many devices will retry and cause traffic to resume after a short pause but some devices may require a port enable/disable to correct the behavior. When CAM entries are full (64 devices per quad on a BLOOM-based environment, and 2048 entries per chip on a CONDOR-environment), the port turned into session-based zoning. When add another entry into the zone, improper check of error return code puts the port into Hardware zoning temporarily, which caused class3 frame drop during the zoning transition. The fix is to check proper error return code to have the correct session-based zoning setup in the switch to eliminate the window.
• A switch will failover or reboot as a result of an SNMP crash. The software watchdog causes an unscheduled automatic failover or reboot on platforms running Fabric OS v4.4.x after a period of 49 days. Fabric OS includes a software watchdog (swd) that periodically verifies the operation of critical software components. As part of this process, each software daemon must check in on a regular basis to inform the watchdog that the process is alive. If the daemon does not check in, the swd will begin recovery procedures to restore operations. Some daemons can only be restarted by causing the switch to reboot or failover. In Fabric OS v4.4.x, the SNMP daemon incorrectly computes its refresh time due to a counter wraparound. As a result, the watchdog timer begins recovery procedures that cause a switch failover or reboot. This rollover will occur at approximately 49-day intervals. The only method to correct the situation is to activate a failover or reboot the switch prior to the 49-day interval.
• Upgrading to FOS v4.4.x will result in a temporary Name Server/Zoning Service incompatibility. In large fabrics, this can also cause traffic to stop. After upgrading from FOS v4.1.x/4.2.x to FOS v4.4.x, the Name Server loses all zoning information and leaves all devices zoned together. As devices go online/offline, a flood of PLOGI requests may be generated as the host driver re-synchronizes its information. In large fabrics, this PLOGI storm can cause the switch to disable interrupts and cause traffic to stop. The condition can be remedied by issuing a cfgEnable command.
• A single corrupted idle sent by a storage device despite subsequent good idles will cause the switch to take the port offline by sending a not operational primitive sequence (NOS). The switch takes the port offline upon seeing a single bad idle frame, causing the host driver to lose connectivity to storage. The current detection of bad idles is too sensitive, causing unnecessary link resets. The software change is made to accommodate temporary loss of sync condition per FC standards. However, FOS 4.4.0b will ignore spurious invalid primitives, it relies on port fault mechanism to reset the link. FOS 4.4.1a and later will not rely on port fault mechanism but will reset link based on loss of signal detection.
• An edge condition could cause the software watchdog to incorrectly reboot a switch due to FSPF path change request storms. The Software Watchdog (swd) could cause a failover or reboot due to an FSPF panic. As part of the software watchdog code review, an edge condition was discovered where a faulty routing driver could cause a condition where FSPF would be overloaded with a flood of path change requests. The fix was to properly handle this storm to avoid the possibility of the software watchdog error.

Version 4.2.1a

• initial release