HP Tru64 UNIX and TruCluster Server Version 5.1.B-4 Patch Summary and Release Notes (13156)
• Increases the default values for udp_ttl and tcp_ttl to 128 hops.
• Fixes a condition that causes the "rdg: unwiring" panic.
• Increases the default limit of DLI packets to 16 KB and makes the limit tunable. This corrects
a problem in which attempts to send packets larger than 5000 bytes (jumbo packets) can fail.
• Improves the fragment gathering mechanism to boost performance.
• Corrects a problem in which the auditd -d command (which flushes the kernel audit buffers)
could cause audit data inconsistencies on a multi-CPU machine on systems generating a
heavy volume of audit events.
• Provides a base system routine for use by the cluster code to determine if a particular mount
has an NFS exported file or directory on it.
• Corrects a problem in which a user process cannot be interrupted, which in some instances
can be utilized in a denial of service attack.
• Corrects a kernel memory fault caused by uninitialized or incorrect parameters being passed
to the setsockopt system calls.
• Fixes a condition that causes a kernel memory fault panic in the IP multicast loopback code.
• Corrects a problem in which multi-CPU sometimes livelock while processing incoming
network traffic. In some cases the live lock can result in a cluster event timeout panic.
• Resolves internet protocol conformance issues and fixes a problem with sending multicast
datagrams.
• Fixes a condition that can cause crashes from within the pshared subsystem
• Fixes several IPMI-related problems, including the following:
— Erroneous fields in 686 OS-detected environmental machine check logout frame
— Unusually large number of 686 sensor timeouts with heavy system load
— IPMI always reporting -48v sensors as broken, seen as "redundant power supply failed"
messages
— An IPMI memory leak
• Corrects problems in UFS extendfs functionality that cause file system metadata
inconsistencies.
• Corrects the default parameter for physio_max_coalescing to 8K.
• Addresses an issue on large systems in which kernel threads might not be executed for
extended periods of time.
• Fixes two small logic errors with the NFS version 3 client that result in unintended, though
correct, behavior.
• Fixes a problem seen with the TAHI IPv6 conformance test, specifically Test 4 for the IPv6
specification.
• Resolves a problem that results in multiple cluster members crashing with kernel memory
fault in rfs_find_fsid().
• Corrects a potential system hang when an error occurs while updating special file or named
pipe access times on an NFS client.
• Fixes an AdvFS asynchronous direct I/O problem that can cause a thread to hang.
• Fixes a problem in which a truncated AdvFS file erroneously zeros data for the remaining
leading segment of the file.
• Fixes a minor problem in the IPv6 subsystem that causes an extra message to be sent upon
startup.
• Corrects a problem that occurs in some low free memory situations, in which a kernel thread
that completes AIO requests may stall on one request, causing other requests (that can
complete) to back up behind it.
• Improves performance for CFS filesets mounted with the server_only option.
• Allows tasks with just one SCS thread to be migrated in the same manner as single threaded
processes in NUMA environments.
3.4 Summary of Base Operating System Patches 95