SNA_PRSRV T21E136 DEC SNA Peer Server V1.3 ECO Summary
TITLE: SNA_PRSRV T21E136 DEC SNA Peer Server V1.3 ECO Summary
Copyright (c) Compaq Computer Corporation 1998. All rights reserved.
Modification Date: 11-SEP-1998
Modification Type: Cleaning up Internet
PRODUCT: DEC SNA Peer Server V1.3
OP/SYS: DIGITAL UNIX [R]
SOURCE: Compaq Computer Corporation
ECO INFORMATION:
ECO Kit Name: T21E136
ECO Kits Superseded by This ECO Kit: T21E135
T21E134
ECO Kit Approximate Size: 5083136 Bytes (9928 Blocks)
Kit Applies To: DEC SNA Peer Server V1.3
System/Cluster Reboot Necessary: Yes (See Installation Notes)
NOTE: The T21E136 ECO kit is a cumulative kit which supersedes
the T21E135 ECO kit.
ECO KIT SUMMARY:
An ECO kit exists for DEC SNA Peer Server V1.3 on DIGITAL UNIX V3.2
through V3.2G. The DEC SNA Peer Server Release Notes contain a complete
explanation of this release. Below is a brief description of the problems
fixed by this release.
PROBLEMS ADDRESSED IN T21E136:
o Only issue stop session management commands for independent lu sessions,
if the session state is >= SPAM_C_SCB_PEND_GAP_BIND_RSP.
o For SMP systems, fix t21__cp_plumbsess_rsp that was being scheduled
before the stream-id had been returned to SPAM from CP Services.
o Change t21llc dispatcher routines to use noninterruptable thread blocks.
o Remove t21_timeout/t21_untimeout timer warning messages.
o The t21llc module was not handling disconnects correctly. Change
t21llc close routine to be event driven.
o Peer Server SDLC was retransmitting frames that had already
been acknowledged.
o Fix Peer Server minor device routines to correctly use queue macro's.
o Fix Peer Server to release the LMF capacity unit for servers
running locally on the same node as Peer Server.
o Change Peer Server LU old name strings to non case sensitive
compares.
o Peer Server minor device routines was incorrectly referencing
a queue pointers.
o Fix thread stack overflow when the gateway and client connection
were lost simultaneously
o Fix NOTIFY RSP RU to Client.
o Fix handling of GAP UNBIND arriving from Client when a GAP UNBIND RQ from
Peer Server has already been sent.
PROBLEMS ADDRESSED IN T21E135 (ECO-5):
o Initialize rcvdsc data structure in t21cad_thread. This can
potentially cause t21cad to crash when trying to free the
buffer. This problem is fixed.
o Prevent internal timer from stopping prematurely.
If multiple Transmission Groups are enabled at once and
these transmission groups become completely synchronized
in their sending of NULL XIDs the internal timer the Peer
Server maintains would stop running. This would cause
the Transmission Groups to stop sending NULL XIDs. If the
Transmission Group was disabled and then reenabled exactly
one NULL XID would be transmitted, instead of a sequence of
multiple NULL XIDs. The only recourse was to reboot the sys-
tem. This change fixes this problem by making sure that the
timer isn't prematurely disabled.
o Obsolete subsets removed from kit. Obsolete WAN Drivers and
DECnet subsets are no longer shipped on this kit.
These subsets must now be installed from the WAN Drivers
and DECnet kit distributions.
More information regarding the nature of these problems can be found in
the ECO kit's release notes. Also, previous ECO kit fixes are fully
described, since the ECO's are cumulative fixes.
PROBLEMS ADDRESSED IN T21E134 (ECO-4):
o Fix session cleanup for dependent and independent LU's when
stop session management commands are issued. Management commands
"Stop SNA LU Services LU XXX Session XXX Severity= ???" would not
terminate sessions if there was ongoing outbound segmentation in the RU.
o Fix BIND reject when the BIND arrives before the response to INIT-SELF.
The BIND was rejected with sense code 0801.
o When SDLC was configured for a dial connection with a station role of
"secondary" or "negotiable", XID3 negotiation was not successful. The
address transitioned to the correct address and then incorrectly back to
the broadcast address FF.
o The routine q25_dlpi_bind_req() was accessing deallocated memory.
This could potentially cause a system crash.
o The routine t21__spam_process_sess_route_inop() was accessing
deallocated memory. This could potentially cause a system crash.
o Set pid to -1 for initialization process after it has completed.
Fixes t21mcd crash if UNIX reuses the pid again for a daemon or client
process. When the pid is reused MCD finds the pid on the init queue and
then tries to free the initent-> data structures again.
o On SMP systems there was a window in T21WAD that allowed mblks to be
updated on one thread while another thread deallocated the mblks data
structure. This change fixes the problem reported in
o More information regarding the nature of these problems can be found in
the ECO kit's release notes. Also, previous ECO kit fixes are fully
described, since the ECO's are cumulative fixes.
PROBLEMS ADDRESSED IN T21E133 (ECO-3):
o Resolved a system crash when QLLC station was disabled and enabled.
o Increased Peer Server components flow control high and low watermarks.
o Retransmit unacknowledged SDLC I-frames.
o Enable TN3270 server LUs to be restricted.
o Make the TN3270 server listen over all network adapters.
o Limited Product Set ID Control Vector to 60 bytes.
o SDLC frames were not always transmitted when of maximum
frame size.
o Inbound X.25 QLLC call is rejected with error "No Filters in Use".
o The Peer Server was clearing the ASPI bit (Adaptive Session
Pacing Indicator) in Bind responses when the original Bind
was not extended (no Control Vector 60).
o Changed T21kit builder to not strip debug symbolics from ECO's.
o Keep X25 receive credits at 2 instead of 1.
o Reset signal handlers for synchronous signals to the UNIX
default signal handler in the t21cad process. This over-
rides the DECthreads handler for the synchronous signals
which turns them into exceptions. Resetting the synchronous
signal handler to the default signal handler will preserves
the stack and the PC where the signal occurred which will
provide useful core images.
o Pass unbinds from Access Routines thru the Peer Server un-
changed. Previously the Peer Server mapped all unbinds sent
from an Access Routine to a normal unbind before forwarding
it to the remote LU.
PROBLEMS ADDRESSED IN T21E132 (ECO-2):
o Resolved a system crash when token ring giving errors.
When the Peer Server was utilized with a token ring and the token
ring network began to report errors to the Peer Server the Peer
Server would cause a system crash. The characteristics of this
crash were that it would fail freeing an mbuf at line 1342 in module
t21llc/src/t21llc_dlpi_actn.c.
o Resolved problem with passive connections would hang a session.
The Peer Server was not handling GAP V3 passive connections properly
where no USS data was requested by the client. The symptom of this
problem was that the OpenVMS Printer Emulator V1.3 would hang a
session when that session terminated.
o Improved TCP/IP Transport Performance.
When the Peer Server was accessed by an access routine using TCP/IP
as the transport, performance was non-optimal. This was because the
PAI used in the CAD daemon process used write() instead of writev()
when sending TCP/IP messages to the access routines. The Peer
Server now uses the writev() service so that the message header
doesn't get separated from the message body.
o Corrected software version number in transmitted XID.
The Peer Server was sending an obsolete software version number
(V110) in the XID software subvector for the Product Set ID. Now it
sends the correct current version number.
o Prohibited use of permanent TGs that are in the connecting state.
The Peer Server was attempting to a use a permanent TG when it was
in the connecting protocol state in the expectation it would
eventually become active. Even though this is a valid assumption it
is better to give an error and allow the user to go and understand
why the TG hasn't come up fully.
o Prevented SDLC circuits from resetting every inactivity seconds.
If the Peer Server was operated with a product such as MEGAPAC (an
SDLC spoofing product) that generated multiple SNRMs the SDLC
inactivity timer would fire every inactivity timer seconds and reset
the SDLC station with the result that all sessions would be taken
down and have to be re-established when the station immediately
restarted.
o Fix internal trace so it works on multi-processor systems.
The internal trace facility truncated its output on multi-processor
systems.
o Fix CTF QLLC analysis routine.
CTF would SEGV in some cases when analyzing QLLC trace files.
o Fix GAP version negotiation problem.
If multiple clients connected to the Peer Server simultaneously it
was possible for an incorrect GAP version number to be sent back to
the client.
o Fix assorted memory leaks in t21cad process.
The t21cad processes would fail to return heap memory and threads
resources to the system. This could lead to hanging or unexpected
termination of the t21cad processes.
o Fix incorrect reported byte offset in XID Negotiation Control Vector.
If the Peer Server detected an error in a received XID it would
report an incorrect error byte offset that was always a negative
number.
o Fix memory leak in Common Agent mold process.
Each time an NCL command was issued to the Peer Server the mold
process would grow by about 45kb of virtual memory and utilize more
threads resources. Eventually the mold process would either core
dump or use up all the swap space on the system. This prevented
Peer Server NCL commands from working at all neccessitating either a
restart of the Peer Server or a system boot. This kit includes a
new mold process that will be installed onto the system unless a new
mold has been supplied through a patch to the DIGITAL UNIX base
operating system or the installation of DIGITAL UNIX V3.2G.
o Allow Peer Server to be installed with WAN (X.25) Support V2.0.
The Peer Server would not install correctly if the V2.0 of the
product: Wide Area Networking (WAN) Support for DIGITAL UNIX
(previously known as X.25 for DIGITAL UNIX Systems) was installed.
PROBLEMS ADDRESSED IN T21131 (ECO-1):
o Resolved a UNIX kernel memory corruption problem.
Prior to ECO-01, the Peer Server might corrupt kernel pool during
outbound session allocation. This problem could result in various
unaligned memory fault crashes with "Unaligned kernel space access
from kernel mode" being prevalent.
This problem could occur in any version of DIGITAL UNIX, however the
new memory allocation scheme in DIGITAL UNIX V3.2C made the problem
more likely to occur.
The problem was resolved by allocating the correct buffer size
required for the connect response message.
INSTALLATION NOTES:
NOTE: This is not an update kit and the current version of the Peer Server
must be deleted before installing this ECO. Deleting the current kit
will not remove the current setup, but you will have to rebuild the
kernel and reconfigure the kit, and you will need to redefine any
TN3270 server processes.
The ECO kit is distributed as a tar file containing the kit which can be
unIstalled using the setld command. The tar file should be unpacked to a
suitable directory and then the setld command issued specifying that directory
as the source of the kit.
For example, if the tar file containing the kit has been copied to
/usr/tmp/t21E135.tar:
# mkdir /usr/tmp/t21E136
# cd /usr/tmp/t21E136
# tar xf /usr/tmp/t21E136.tar
# setld -l .
If the kit is delivered on tap11 then the commands would be:
# mkdir /usr/tmp/t21E136
# cd /usr/tmp/t21E136
# tar x
# setld -l
Once the kit is installed, you should follow the instructions in the
installation guide for rebuilding the kernel then reboot the system and
configure the kit with the exception that it is not necessary to rerun the
initial configuration utility.
[R] UNIX is a registered trademark in the United States and other countries
licensed exclusively through X/Open Company Limited.
This patch can be found at any of these sites:
Colorado Site
Georgia Site
Files on this server are as follows:
t21e136.README
t21e136.CHKSUM
t21e136.CVRLET_TXT
t21e136.tar
|
