HP Tru64 UNIX and TruCluster Server Version 5.1.B-4 Patch Summary and Release Notes (13156)
• Makes Tru64 UNIX V5.1B compatible with V5.1A in regard to the handling of the relationship
between a controlling terminal and the “session” structure.
• Fixes the fastpath send of RPC modules to correct conditions in which it ignores the VMAC
setting and it excludes source and destination UDP port numbers when LAG is enabled.
• Fixes a problem in which the NFS server, in certain cases, does not update the access time
on the files it serves.
• Enables IPv6 MTU to be set up to 9000 bytes on gigabit Ethernet.
• Fixes memory troller panic that can occur when booting in low-memory conditions.
• Addresses two potential causes of various ics timeout panics, such as
“ics_unable_to_make_progress: input tread” and “ics_unable_to_make_progress: netisrs
stalled.”
• Enables allows AutoFS mount-on paths to have the correct maximum length when AutoFS
files are mapped directly.
• Fixes a race condition in VM map-entry fault handling that can result in poor performance
and/or panics.
• Fixes a SierraCluster "bigpage assert failure" panic.
• Corrects the panic:"kernel memory fault" in umc_page_release() due to non-initialization
of variable hpp.
• Fixes a problem that can cause the system panic "cmn_err: ce_panic:
ics_unable_to_make_progress: netisrs stalled."
• Fixes a problem that causes multiple TCP sockets to be created and then left in the
CLOSE_WAIT state on an NFS client in a TruCluster when the NFS server is unavailable.
• Provides improvements to the TCP Selective Acknowledgment (SACK) option.
• Corrects an issue in which tape devices with no ddr entry would not have compression
correctly enabled or disabled as appropriate to the device special file that was used.
• Corrects a problem that causes data transfers to be slow on socket connections when using
the MSG_WAITALL flag in a recv() socket call.
• Corrects a condition to avoid the copying of multicast packets to the networking stack when
packetfilter is enabled and pfilt_copymulti=0; default is 1, meaning always copy multicast
packets.
• Fixes NFS client hangs on systems with no associated memory for RADs.
• Modifies the kernel to correct a problem that could under certain conditions cause the
following:
— User space programs to experience unexpected "interrupt" results (EINTR) returned by
system calls such as exit() or wait3().
— Korn shell scripts to report that programs were terminated with a "Signal 64" when in
fact the program ran normally and no error was present.
— Programs that check the exit status of children to see abnormal results.
• Fixes a problem of ambiguous connection rejection caused by socket database search in a
multi-RAD system.
• Corrects several problems in file region locking that could result in dropped file region locks
or in a process hang.
• Fixes an "m_copydata offset" panic seen on TruClusters and NUMA systems serving NFS
file systems.
• Fixes a race condition in the UBC subsystem that could result in a kernel memory fault.
• Prevents a race condition in the code dealing with kernel address space data structures.
• Improves NFS to help prevent an NFS client from creating more than one TCP connection
for a mount point and to prevent a system panic if the problem occurs.
• Fixes a lockmode 4 only panic due to a lock ordering problem.
• Resolves a problem that could result in freeing active memory buffers.
3.4 Summary of Base Operating System Patches 101