Updated: 24-SEP-2003 (Use your browsers' Reload button to ensure you're viewing the most recent version)
TCPIPJAALP_E03A50 TCP/IP(JAPANESE) for OpenVMS V5.0A ECO
*Compaq TCP/IP] TCPIPJAALP_E03A50 TCP/IP(JAPANESE) for OpenVMS V5.0A ECO Summary
Copyright (c) Compaq Computer Corporation 2001. All rights reserved.
PRODUCT: Compaq TCP/IP Services/Japanese for OpenVMS
OP/SYS: OpenVMS Alpha
SOURCE: Compaq Computer Corporation
ECO INFORMATION:
ECO Kit Name: TCPIPJAALP_E03A50
DEC-AXPVMS-TCPIPJA_ECO-V0500-113-4.PCSI
ECO Kits Superseded by This ECO Kit: TCPIPJAALP_E01A50
ECO Kit Approximate Size: 47,664 Blocks
Kit Applies To: Compaq TCP/IP Services/Japanese for OpenVMS V5.0A
OpenVMS Alpha V7.1, V7.1-1H1, V7.1-1H2,
V7.1-2, V7.2, V7.2-1, V7.2-1H1
System/Cluster Reboot Necessary: UNKNOWN
Rolling Re-boot Supported: Information Not Available
Installation Rating: INSTALL_UNKNOWN
Kit Dependencies:
The following remedial kit(s) must be installed BEFORE
installation of this kit:
None
In order to receive all the corrections listed in this
kit, the following remedial kits should also be installed:
None
ECO KIT SUMMARY:
An ECO kit exists for TCP/IP(JAPANESE) V5.0A on OpenVMS Alpha V7.1 through
V7.2-1H1. This kit addresses the following problems.
PREVIOUSLY UNDOCUMENTED BEHAVIOR:
1: PTR 70-5-1299 Retry and error ignore for MIB browser
---------------------------------------------------------------
The DIGITAL TCP/IP Services for OpenVMS Management manual,
Order Number: AA-LU50K-TE, Chapter 11 Configuring SNMP,
Table 11-4 Flags for the snmp_request Command will be updated
with the following new flags:
-i max_ignores: The number of times snmp_request listens for a reply packet
to a request if it receives an invalid packet (invalid packet
id, version, SNMP version/command combination). Value should
be positive; specifying a negative value results in internal
conversion to an unsigned positive; specifying 0 is the same
as omitting this flag, in which case errors are not ignored.
-r max_retries: The number of times snmp_request resends a request packet if it
times out before receiving a reply. Value should
be positive; specifying a negative value results in internal
conversion to an unsigned positive; specifying 0 is the same
as omitting this flag, in which case no retries are tried.
-s sleep_interval: Seconds between iterations of sending a request (for -r)
and listening for a reply (for -i). Default is 1 second.
Ignored unless -r, -i, or both are also specified.
Notes:
- The -s flag is not the same as -w, which is used to adjust timeout
interval for the internal select() call. -s may be needed in
cases where there is a need to enforce a wait between resend, as
when a server agent which is in the process of starting up.
- After a timeout and resend, if a reply packet is received the resend
counter is reset; after another timeout, max_retries request packets
will be sent.
- The same is true for max_ignores. After a bad reply is received, if
a valid reply packet is received, the internal ignore counter is
reset to max_ignores.
- The -s value may be required in cases where a server is in some
transient state, e.g., in the process of coming up.
- Following receipt of an invalid packet, a timeout may occur. In this
case the packet will be resent, the resend counter will be decremented,
and the ignore counter will be reset to max_ignores.
- -i, -r, and -s apply to individual queries. If the -l or -t flags are
also used, the values for -i, -r, and -s are applied afresh to each
iteration for the -l or -t.
- The -i flag does not cause snmp_request to automatically increment input
OID and continue querying a server after an SNMP general error has been
received. For example, on a mibwalk (result of using -t or -l flag)
if a faulty server may with this kind of error even though the end
of MIB view has not been reached. It is still necessary to manually
increment the input OID to skip the error to continue with the mibwalk.
The following examples will be added to 11.6.4 Command Examples
for snmp_request:
Add these 2 bullets at the beginning of the section:
- To define the command snmp_request, you can execute the command:
$ @SYS$STARTUP:TCPIP$DEFINE_COMMANDS.
- In all examples, output may be edited to fit the 80 column format, with
continuation indicated by a "\" at the end of a line
Add the following examples:
9. To get online UNIX style usage information, enter the command without
parameters:
$ snmp_request
Usage: $ snmp_request = \
"$ucxaxp$dkd400:[tcpip_t51_bl9z.][a70.snmp.exe]tcpip$snmp_request.exe"
$ snmp_request agent community request_type [flags] \
[set request parameters]
or
$ mcr \
ucxaxp$dkd400:[tcpip_t51_bl9z.][a70.snmp.exe]tcpip$snmp_request.exe ...
request_type: get|getnext|getbulk|set
flags: [-v version] [-p port] [-l] [-t] [-d]
[-n non_repeaters] [-m max_repetitions]
[-w wait_seconds_max] [-i max_ignores]
[-r max_retries] [-s sleep_interval]
set request parameters: variable [type value]
10. The server is not up when the request is sent.
It comes up, starts responding, and then goes down again.
$ snmp_request moxie.zko.dec.com public getnext -r 3 -l -s 5 1.3
No reply; resending request...
1.3.6.1.2.1.1.1.0 = moxie.zko.dec.com VAX 6000-410 OpenVMS V7.1 \
Compaq TCP/IP Services for OpenVMS
1.3.6.1.2.1.1.2.0 = 1.3.6.1.4.1.36.2.15.22.1
1.3.6.1.2.1.1.3.0 = 198 = 0 d 0:0:1
1.3.6.1.2.1.1.4.0 = Sam Spade
1.3.6.1.2.1.1.5.0 = moxie.zko.dec.com
1.3.6.1.2.1.1.6.0 = ZKO
1.3.6.1.2.1.1.7.0 = 72
No reply; resending request...
No reply; resending request...
No reply; resending request...
No reply; resend limit reached. Exiting.
11. The server is not up when the request is sent.
Then the server comes up, but is not fully initialized when the second
request packet is sent:
$ snmp_request moxie.zko.dec.com public getnext -r 3 1.3
No reply; retrying...
- no such name - returned for variable 1
12. This example is identical to #3, except that the -s 5 causes a
delay in the sending of the second request packet, giving the server time
to fully initialize itself:
$ snmp_request moxie.zko.dec.com public getnext -r 3 -s 5 1.3
No reply; retrying...
1.3.6.1.2.1.1.1.0 = moxie.zko.dec.com VAX 6000-410 OpenVMS V7.1 COMPAQ TCP/IP
Services for OpenVMS
13. The following is happening:
- A bad reply packet is received
- A timeout causes a new request packet to be sent
- A second bad reply packet is received
- Another timeout causes a new request packet to be sent
- A third bad reply packet is received
- The max_tries limit of 2 has been reached, so snmp_request exits.
$ snmp_request moxie.zko.dec.com public getnext -i 3 -r 2 1.3
Unexpected reply; waiting again...
No reply; resending request...
Unexpected reply; waiting again...
No reply; resending request...
Unexpected reply; waiting again...
No reply; resend limit reached. Exiting.
14. This example is identical to #5 except that max_retries
is greater than max_ignores. Only two extra request packets are sent
before the max_ignores limit is reached:
$ snmp_request moxie.zko.dec.com public getnext -i 2 -r 3 1.3
Unexpected reply; waiting again...
No reply; resending request...
Unexpected reply; waiting again...
No reply; resending request...
Unexpected reply; ignore limit reached. Exiting.
15. This example is identical to #5, except that all invalid packets are
received before any timeouts; therefore no resends are tried:
$ snmp_request moxie.zko.dec.com public getnext -i 3 -r 2 1.3
Unexpected reply; waiting again...
Unexpected reply; waiting again...
Unexpected reply; waiting again...
Unexpected reply; ignore limit reached. Exiting.
----------------------------------------------------------------
2: PTR 70-5-1348 LPD print symbiont inserts LF characters
----------------------------------------------------------------
The following two logicals have been added for the LPD print
symbiont to fix the following problem:
LPD print symbiont inserts LF character after setup
module.
Defining the logical TCPIP$LPD_SETUP_NOLF will suppress
the LF characters on a per-system basis.
Defining the logical TCPIP$LPD_<queue>_SETUP_NOLF will
suppress the LF characters on a per-queue basis.
-----------------------------------------------------------------
3: PTR 70-5-1642 Windows NT NFS server can hang w/o /NOADF
-----------------------------------------------------------------
With some Windows NT and Windows 2000 NFS Servers (bundled with
Compaq TaskSmart [WNT 2000 Disk Access]), the OpenVMS NFS
client will hang attempting to reconstruct a file using it's
companion ADF. The Windows NFS server refuses to respond if
the ADF filename is .$ADF$000000. Since this name is a legal
filename, the long term solution is to ask Microsoft to fix their
NFS server. The short term or interim solution is to mount the
device with either the /RETRIES=<small integer> option or the
/NOADF option.
Please note that the default behavior beginning at TCP/IP V5.1
has been changed to /RETRIES=4 instead of /RETRIES=0 (forever).
-----------------------------------------------------------------
4: PTR 30-5-313 RCP copy changes the file attributes
-----------------------------------------------------------------
"$ COPY/RCP" on a binary file changes the file attributes
from fixed length to STREAM LF and corrupts the file.
To resolve this problem, RCP will now treat VMS fixed-length (FIX)
and undefined (UDF) format files as binary and copy them verbatim
without doing the implicit conversion to Stream_LF. Note however
that RCP will NOT preserve file attributes on the target file
(which will still be created as Stream_LF). To fix this is beyond
the scope of an ECO and may not be possible without incompatibly
breaking the RCP protocol. File attributes on the target will have
to manually be corrected by the user.
Lest some customers have come to depend on the original behavior of
RCP, that behavior can be restored by setting the logical
TCPIP$RCP_SEND_FIX_FORMAT_AS_ASCII to "1", "True", or "Yes".
However, if this logical is set to a numeric value that exactly
matches the fixed-length file's record size, that file will not be
converted to Stream_LF, but will be sent as binary.
Note also, the TCPIP$RCP_SEND_FIX_FORMAT_AS_ASCII logical name will
superceed the use of the UCX$RCP_SEND_FIX_FORMAT_AS_ASCII logical
name. Both are accepted logicals.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A ARP Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 19-APR-2000 Alpha and VAX
Images:
TCPIP$ARP.EXE V5.0A-1A
TCPIP$NETSTAT.EXE V5.0A-1A
TCPIP$GATED.EXE V5.0A-1B
TCPIP$UCP.EXE V5.0A-1F
Problem:
System can crash executing TCPIP> netstat -rn
Solution:
The NLIST library was modified necessitating the relink
of ARP, NETSTAT, GATED and UCP.
ECO B 23-JUN-2000 Alpha and VAX
Images:
TCPIP$UCP.EXE V5.0A-1G
TCPIP$NETSTAT.EXE V5.0A-1B
TCPIP$PING.EXE V5.0A-1A
TCPIP$IFCONFIG.EXE V5.0A-1A
TCPIP$SYSCONFIG.EXE V5.0A-1A
TCPIP$ESNMP_SERVER.EXE V5.0A-1I
TCPIP$HR_MIB.EXE V5.0A-1I
TCPIP$OS_MIBS.EXE V5.0A-1I
TCPIP$ARP.EXE V5.0A-1B
TCPIP$TRACEROUTE.EXE V5.0A-1A
Problem:
tcpip show route fails with INSVMEM when route database is large
Solution:
Backported fix from tcpip Version 5.1 that addresses this issue in
the unix_api library.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A BFS Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 24-MAR-2000 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1A
TCPIP$MSG.EXE V5.0A-1C
Problem:
Message: cache_check detected corrupted LRU queue appears
on console after a mount from an nfs client hangs.
Solution:
Fixed miss-compare problem. Used STAT_CHE$W_CLUSTER_COUNT
instead of STAT_CHE$W_CLUSTER_LIMIT.
ECO B 18-MAR-2000 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1B
Problem:
When UCX$CFS_MODUS_OPERANDI is set to 512, and a VMS NFS
client is being used, the NFS server sometimes reports
FCPREADERR ACCVIO ILLIOFUNC. This prevents the client from
accessing some files.
Solution:
Ported from 4.2.
Fixed handling of UCX$CFS_MODUS_OPERANDI 512 to not break the
communication of the VMS file's attributes between the VMS
NFS-Client and the VMS NFS-Server.
ECO C 3-MAY-2000 Alpha and Vax
Images:
TCPIP$CFS_SHR.EXE V5.0A-1C
Problem:
Problem wih BFS not returning an error status for bad
arguments. I.E. The synch status.
Solution:
Fixed bfs$$synch-alpha.m64 to save R0 from $$STARTUP.
ECO D 12-JUN-2000 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1D
Problem:
Backup errors on NFS due to a failure to allocate
space for access control lists (ACL).
Solution:
Allocate space for the ACL and only return an error if unable to
allocate it due to lack of space available in the system.
ECO 3 updates
--------------
ECO E 6-OCT-2000 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1E
Problem:
When copying a file from a NFS client to the NFS Server the server
stops sending data and the thread usage then increases to the
maximum.
Solution:
Fixed NFS server locking problem.
ECO F 17-JAN-2001 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1F
Problem:
NFS problem with file aliases where certain file aliases created
on the VMS side are only partly visable on Unix.
Solution:
Treat all file aliases to directories as directories when
MODUS_OPERANDI bit 16384 is set.
ECO G 30-MAR-2001 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1G
Problem:
For file systems exported with /options=Name_Conversion,
any particularly long filename can map to the VMS file ".;"
which can be disconcerting and confusing.
Solution:
Properly detect when NonVMS-to-VMS name conversion fails and
report this as an appropriate error back to the client (e.g.
NAMETOOLONG). NonVMS filenames that cannot be converted into
a legal VMS ODS-2 filename can only be accommodated within a
Container File System.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A BIND_RESOLVER Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 18-Apr-2000 Alpha and VAX
Images:
TCPIP$BIND_SERVER.EXE V5.0A-1B
TCPIP$BIND_SERVER_CONTROL.EXE V5.0A-1B
Problem:
BIND 8.1.2 did not include dynamic update support for SRV records.
Solution:
Extract and merge the SRV support from BIND 8.2 into our current
sources.
ECO 3 updates
--------------
ECO B 6-NOV-2000 Alpha and VAX
Images:
TCPIP$BIND_SERVER.EXE V5.0A-1B
TCPIP$BIND_SERVER_XFER.EXE V5.0A-1B
TCPIP$NSUPDATE.EXE V5.0A-1B
Problem:
BIND_RESOLVER allocates too much stack, 32 Kb on Alpha
and 8 Kb on VAX.
Solution:
Modified res_init() and res_query() (another culprit) to use
NS_MAXDNAME instead of BUFSIZ when declaring the buf variable.
This will use 1k of stack instead of 32k (or 8k on vax).
ECO C 26-JAN-2001 Alpha and VAX
Images:
TCPIP$IPC_SHR.EXE V5.0A-1C
Problem:
The dynamic updates being sent by the load broker to a
Windows2000 DNS server were failing with an error similiar to:
unknown response: ans=0, auth=1, add=1, rcode=3
Solution:
Modify the check on the response packet from an SOA query to accept
packets that have an empty answer section, auth=1, and add=1.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A BIND_SERVER Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 06-Aug-1999 Alpha and VAX
Images:
TCPIP$BIND_SERVER.EXE V5.0A-1A
Problem:
The BIND server was reporting syntax errors when CIDR notation
was used in the configuration file (TCPIP$BIND.CONF).
Solution:
There were a couple of different reasons for this problem. The
first was that there were two successive calls being made to
ungetc(). This is an unsupported operation and led to
inappropriate behavior. To cure this, an f*() file positioning
function must be made in between calls to ungetc() in order to
clear the pushback buffer. The pushback buffer is where ungetc()
writes it char to. It does not get placed back to the actual file
buffer. The size of the pushback buffer for the DEC C RTL is 1,
and once it is full additional calls to ungetc() will fail. I
chose to use a call to fseek() that would basically just position
the file pointer to its current position (a no-op besides clearing
the buffer).
The other problem was that due to the parser logic behind
determining whether a slash ('/') was part of a comment or part of
a CIDR notated IP address, the file positioning pointer became a
char ahead of where it should have been. This was after a series
of getc() and ungetc() calls. This problem was solved by doing an
fgetpos()/fsetpos() combo to reset the file position indicator.
There was also a bug discovered here where the code still does not
work if it is compiled with /NOOPTIMIZE on a VAX. This happens if
you compile for debug in our build environment (only VAX). Seeing
we do not ship /NOOPTIMIZED code, we will ignore this optimizer bug
and be sure to add a regression test item that will verify that
this continues to work as expected in the future. The bug could
not be reproduced with a simple test program, only in context of
the BIND code.
ECO B 18-MAR-2000 Alpha and VAX
Images:
TCPIP$BIND_SERVER.EXE V5.0A-1B
TCPIP$BIND_SERVER_CONTROL.EXE V5.0A-1B
TCPIP$BIND_SERVER_XFER.EXE V5.0A-1A
TCPIP$NSUPDATE.EXE V5.0A-1B
Problem:
BIND 8.1.2 did not include dynamic update support for SRV records.
Solution:
Extract and merge the SRV support from BIND 8.2 into our current
sources.
ECO 3 updates
--------------
ECO C 10-OCT-2000 Alpha and VAX
Images:
TCPIP$BIND_SERVER.EXE V5.0A-1C
TCPIP$BIND_SERVER_XFER.EXE V5.0A-1C
Problem:
The BIND database was not being updated once its version number
reached the VMS maximum of 32767.
Solution:
If a "rename" call for the database fails, we purge the previous
versions of the database, then rename the current version to have
version number 1.
ECO E 6-NOV-2000 Alpha and VAX
Images:
TCPIP$BIND_SERVER.EXE V5.0A-1E
TCPIP$BIND_SERVER_XFER.EXE V5.0A-1E
TCPIP$NSUPDATE.EXE V5.0A-1E
Problem:
BIND_RESOLVER allocates too much stack, 32 Kb on Alpha
and 8 Kb on VAX.
Solution:
Modified res_init() and res_query() (another culprit) to use
NS_MAXDNAME instead of BUFSIZ when declaring the buf variable.
This will use 1k of stack instead of 32k (or 8k on vax).
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A BIND_TOOLS Images
---------------------------------------------------------------------------
ECO 3 updates
--------------
ECO A 20-APR-2001 Alpha and VAX
Images:
TCPIP$ACCESS_SHR.EXE V5.0A-1C
Problem:
TCPIP> show host ACCVIOs in certain cases when
retrieving the BIND database.
Solution:
NSLOOKUP was modified to deal with large DNS messages.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A DHCP Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 28-MAR-2000 Alpha and VAX
Images:
TCPIP$DHCP_SERVER.EXE V5.0A-1A
Problem:
DHCP does not send the "Bootfile name" with the DHCP BootReply
packet if BootP is redirect to another system.
The DHCPCAP file does not support ":" and "\" characters in the
BootFile field.
Solution:
The DHCP server was modified so that if a TFTP server IP address
is configured, then the code just takes what it has in the
configured Bootfile name verbatim and returns that to the client.
ECO 3 updates
--------------
ECO B 5-DEC-2000 Alpha
Images:
TCPIP$DHCP_BPASCIITODBMOD.EXE V5.0A-1B
TCPIP$DHCP_BPISAMTOASCII.EXE V5.0A-1B
TCPIP$DHCP_DBDUMP.EXE V5.0A-1B
TCPIP$DHCP_DBMODIFY.EXE V5.0A-1B
TCPIP$DHCP_DBREGISTER.EXE V5.0A-1B
TCPIP$DHCP_DBSHOW.EXE V5.0A-1B
TCPIP$DHCP_GUI.EXE V5.0A-1B
TCPIP$DHCP_SERVER.EXE V5.0A-1B
TCPIP$DHCP_SHOWDBS.EXE V5.0A-1B
TCPIP$DHCP_SIGNAL.EXE V5.0A-1B
Problem:
DHCP Server 5.0A, NT client cannot get IP address, when
coming from another subnet.
Solution:
Backport the changes from the V5.1 code which correctly
puts the port number in network byte order.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A DNFS Images
---------------------------------------------------------------------------
ECO 1 updates:
--------------
ECO A 06-Jul-1999 Alpha and VAX
Images:
TCPIP$DNFSACP.EXE V5.0A-1A
Problem:
Beginning with VMS V7.2, running BACKUP on an NFS client
device results in "-SYSTEM-F-BADATTRIB, bad attribute
control list" errors.
Solution:
The client was doing an argument validation on the size of
the buffer provided to receive the ATR$C_ASCNAME
attribute. Prior to VMS V7.2, this was wrong, but had no
practical effect. In V7.2, BACKUP changes to support EFS
make the incorrect validation visibly harmful. The
solution is to eliminate the length validation on
ATR$C_ASCNAME.
ECO 3 updates
--------------
ECO B 1-AUG-2000 Alpha and VAX
Images:
TCPIP$DNFSACP.EXE V5.0A-1B
Problem:
The original problem under 70-5-1203 is a crash in UCX$DNFSACP due
to memory corruption. A diagnostic module was sent which then
hung the NFS client when it caught the udp_recv_ast touching
deallocated memory.
Also found a bug in cnfs-proxy where the zeroing of memory after
the username characters was modifing 1 too many bytes.
Solution:
Avoid freeing the TRANS block if there is still a receive
outstanding.
Changed the condition in the zeroing "for loop" from <= to < so
that only the allocated bytes are modified.
ECO C 28-SEP-2000 Alpha and VAX
Images:
TCPIP$DNFSACP.EXE V5.0A-1C
Problem:
Introduced regression with fix for 70-5-1340.
Regression is present in TCP/IP V5.0A and V5.1.
Solution:
Remove some lines of code that were missed when
70-5-1340 was fixed.
ECO D 19-OCT-2000 Alpha and VAX
Images:
TCPIP$DNFSACP.EXE V5.0A-1D
Problem:
Rename of files that the client provides with default ADFs
get deleted if the owner is not the one renaming.
Solution:
During a rename, avoid creating an ADF file for files that
already have default ADF.
ECO E 29-MAR-2001 Alpha and VAX
Images:
TCPIP$DNFSACP.EXE V5.0A-1E
Problem:
The NFS client does not send a set attribute to the server after
writing a file that is the same size or smaller, letting it default
to 512 bytes. Initial fix attempt caused a data corruption
regression.
Solution:
Send a setattr procedure to set the size after the re-write,
being careful not to confuse this situation with the one that
led to file corruption. Also, backported V5.1 change so the
NFS client now reports the correct allocated size instead of
the EOF mark.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A FTP_CLIENT Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 2-AUG-1999 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1A
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
SYSTEM-F-NOIOCHAN, with FTP and search list logical
Solution:
Implemented changes throughout FTP regarding the use of
of sys$parse() to mitigate implicit channel usage.
ECO A 9-SEP-1999 Alpha only
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1A
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
FTP client "get a.a.a.a" from Unix host fails with
-RMS-E-CRE, ACP file create failed
-SYSTEM-W-BADFILEVER, bad file version number
Problem occurs only with TCPIP V5 on Alpha VMS 7.2. On VAX, or
Alpha VMS V7.1, the transfer succeeds because "a.a.a.a" is
converted to "a_a_a.a;1" to make RMS happy.
Solution:
Moved all filename conversion operations into
FTP$COMMON_HFS_SUPPORT library where knowledge of ODS-2 vs. ODS-5
parsing conventions are known or can be discerned. Enhanced their
functionality to allow ODS-5 conversion of foreign filenames, and
to provide a backstop conversion back to ODS-2 when an attempt is
made to try creating an ODS-5 name on an ODS-2 volume. Both FTP
Client GET and FTP Server PUT operations were affected.
ECO A 4-JAN-2000 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1A
Problem:
The FTP 'mput' command does not search the several
equivalence names defined by a logical name correctly.
Solution:
This problem was partly caused by 70-5-839 fix to
eliminate channel leaks, in which several uses of
SYS$PARSE were changed to be syntax-checking only.
mput no longer passes the expanded result of its
SYS$PARSE down to wldcrd_put_process().
A similar problem with the recv and dir/out= commands,
and with $copy/ftp remote local were also corrected.
ECO A 5-APR-2000 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1A
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
FTP Client occasionally hangs in accept().
Solution:
Use select() to monitor both the data and control ports.
ECO 3 updates
--------------
ECO B 11-JUL-2000 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1B
Problem:
[1] FTP Client assumes that * is invalid for output filename.
This may not be true if Server is non-VMS.
[2] FTP send command ACCVIO's if globbing is off.
Solution:
[1] Allow the FTP Server to determine if special characters
are valid for the output filename if not VMS Plus mode.
[2] Unconditionally set command type.
ECO C 18-JUL-2000 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1C
Problem:
COPY/FTP (get) or FTP> GET client hangs when retrieving
a large file which already exists on the local system.
Solution:
Check if a socket already exists; if yes, then call
function CLOSE_SOCKET_AND_CHAN(data) to close it down.
ECO D 8-NOV-2000 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1D
Problem:
ACCVIO when using FTP/INPUT=<input file>.
Solution:
Modify the client to recognize both an exit statement and
EOF (the end of the /INPUT file).
ECO E 18-DEC-2000 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1E
Problem:
FTP client gives access violation when using stream_LF
input files; problem occurs after installation of ECO 1.
Solution:
ftp$$fgets() was modified to use fgets() to read files
and then parse them depending on the record format and
record attributes.
ECO F 25-APR-2001 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1F
TCPIP$FTP_CHILD.EXE V5.0A-1I
TCPIP$FTP_SERVER.EXE V5.0A-1I
TCPIP$MSG.EXE V5.0A-1D
Problem:
For file larger than 4 GB, the FTP Client incorrectly reports the
number of bytes transferred. The Kbytes/sec rate reported is also
incorrect.
Solution:
The bytes transferred count is now maintained using an extended
precision representation (64-bit long long on Alpha, two 32-bit
longword array on VAX). The Kbytes/sec rate is now calculated
using double floating-point precision.
Also took this opportunity to include an estimate of a
sequential or fixed-length file's byte length in the
"150 Opening data connection for..." message.
Unix servers do this and it gives a GUI FTP client something
to use for progress bars or %complete screen updates.
Note this size is not 100% accurate due to file conversion
that might occur for the transfer.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A FTP_SERVER Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 7-JAN-2000 Alpha and VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1A
Problem:
FTP> del .; command does not works after UCX was upgraded
from 4.0 ECO 5 to UCX 4.2 ECO 3. Error is: "550 The file
specification must contain a name, type and version number
(even if wildcarded)."
Solution:
Modify FTP DELE command to relax this restriction to just
requiring a version number, in order to accommodate files
that consist of just a name, or just a type.
ECO B 14-JUN-1999 Alpha/VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
An issue with FTP server handling of APPEnd was discovered
during analysis/ testing of 70-5-840.
Solution:
Only set "rms_posted = TRUE" if the preceding RMS call returns
SS$_SUCCESS status in *both* appe_ascii() and similar code in
recv_ascii().
ECO B 14-JUN-1999 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1A
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
Many FTP commands unnecessarily consume VMS channels in both
client and server when a search-list logical is involved in
the operation. This eventually results in SYSTEM-F-NOIOCHAN
error. User must start a new FTP session to make further
progress.
Solution:
Fixed FTP (both client and server) to clean up channels
no longer needed after each command.
ECO B 1-SEP-1999 Alpha and VAX
Image:
TCPIP$FTP_SERVER.EXE V5.0A-1B
Problem:
V5.0 Ftp Child Process creating process owner of
asterisk (*).
Solution:
The FTP Server was fixed to behave like it did in UCX V4.
If the prospective user is known to VMS, the child process
will be owned by that user from inception. If the
prospective user is not known to VMS, the short-lived
child process will be owned by the FTP Server (parent)
process ([TCPIP$AUX,TCPIP$FTP] as configured in the
Service database) until it dies due to login failure.
ECO B 03-SEP-1999 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1A
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
[1] Put/Convert inserts extra LineFeeds (LF) when sending
certain VFC files. (70-5-954)
[2] FTP Client does not send STRU R files correctly in
ASCII mode. Regression from UCX V4.0 ECO stream.
(70-5-110)
Solution:
[1] Fixed FTP's conversion algorithm to perform a more
faithful conversion of VFC/PRN prefix/suffix code
combinations with a better mapping to FTP ASCII
records on the wire. This affected both the client
(Put/Convert) and the server (TCPIP$FTP_CONVERT_FILE
logical).
[2] Fixed FTP client to send proper <IAC><01> end-of-
record tag.
ECO B 10-Sep-1999 Alpha and VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
FTP 'sunique' (store unique) put operations could
overwrite an existing version of the specified file
if the target filename includes a version specification.
Solution:
Fixed the FTP server STOU operation to create a new
version for the file regardless of any specified version
or file transfer mode (ASCII, BINARY, VMS-Plus).
Also fixed instances where a filename parsing error
in the server's RETR, STOR, or STOU commands could
result in a corrupted 501 error message.
ECO B 9-SEP-1999 Alpha only
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1B
TCPIP$FTP_CLIENT.EXE V5.0A-1A
Problem:
FTP client "get a.a.a.a" from Unix host fails with
-RMS-E-CRE, ACP file create failed
-SYSTEM-W-BADFILEVER, bad file version number
Problem occurs only with TCPIP V5 on Alpha VMS 7.2. On VAX, or
Alpha VMS V7.1, the transfer succeeds because "a.a.a.a" is
converted to "a_a_a.a;1" to make RMS happy.
Solution:
Fixed FTP to deal with Alpha VMS V7.2 "enhancements" to sys$parse()
which is now happy to parse Extended File System (ODS-5) pathnames
even though the underlying volume specified in the pathname is
still ODS-2.
FTP now catches an attempt to use an ODS-5 filename on an ODS-2
volume, and performs the filename conversion it traditionally did
to make the name ODS-2 conformant.
FTP also now performs an ODS-5 filename conversion rather than a
more restrictive ODS-2 conversion when the underlying volume is
ODS-5. This generally results in a name much closer to the
original than ODS-2 allowed.
ECO B 15-Dec-1999 Alpha and VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1B
Problem:
FTP> del .; command does not works after UCX was upgraded
from 4.0 ECO 5 to UCX 4.2 ECO 3. Error is: "550 The file
specification must contain a name, type and version number
(even if wildcarded)."
Solution:
Modify FTP DELE command to relax this restriction to just
requiring a version number, in order to accommodate files
that consist of just a name, or just a type.
ECO 3 updates
--------------
ECO C 31-JUL-2000 Alpha and VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1C
Problem:
FTP (server) is producing "ls" output that is malformed --
i.e. two filenames which should be on separate lines are not
separated by \r\n but by \0\0 NUL characters.
Solution:
Replace faulty conditional expression and verify that each entry
is now properly appended with \r\n (carriage return - line feed).
ECO D 20-OCT-2000 Alpha and VAX
Images:
TCPIP$FTP_SERVER.EXE V5.0A-1D
Problem:
TCPIP$FTP server process opens the TCPIP$FTP_SERVER_ANNOUNCE
file for each incoming session, but fails to close it properly
resulting in the FILLM (file limit) quota becoming exhausted.
Solution:
Modify finish_anonymous_message() to call sys$disconnect() and
sys$close() with the correct pointers.
ECO E 27-NOV-2000 Alpha and VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1E
Problem:
Using a DECnet node name in the path of the ftp 'get' command
fails with error message:
550-Unable to determine the file type for <uninitialized data>
Solution:
Initialize resultant_name prior to calling check_filetype()
if the sys$search() path in retr() was somehow avoided.
ECO F 16-FEB-2001 VAX ONLY
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1F
Problem:
Fix for 70-5-1308 caused client DIR command to loop in
server producing blank lines.
Solution:
Wrong source got submitted in which a macro was used
instead of a global, causing an address miscompare.
Replaced use of macro with proper global symbol.
ECO G 19-FEB-2001 Alpha and VAX
Images:
TCPIP$FTP_SERVER.EXE V5.0A-1G
Problem:
FTP Server ACCVIO'S because the reallocation of the session
table on the 129th simultaneous connection fails.
Solution:
Dynamically reallocate the session table.
Session table grows by MAX_SESSIONS.
ECO H 24-APR-2001 Alpha and VAX
Images:
TCPIP$FTP_CHILD.EXE V5.0A-1H
Problem:
FTP> DIR search-list-logical can encounter the server errors
"550 file not found" or "550 unsupported file structure level"
when the search-list-logical is composed of concealed members.
Problem only occurs when file details are requested from the
server, and not when just filenames are requested.
Solution:
Modified the FTP server to better detect the transition from
one concealed device entry to another within a search-list-
logical's SYS$SEARCH() results, so it has the proper device
open for obtaining the requested file details.
ECO I 25-APR-2001 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1F
TCPIP$FTP_CHILD.EXE V5.0A-1I
TCPIP$FTP_SERVER.EXE V5.0A-1I
TCPIP$MSG.EXE V5.0A-1D
Problem:
For file larger than 4 GB, the FTP Client incorrectly reports the
number of bytes transferred. The Kbytes/sec rate reported is also
incorrect.
Solution:
The bytes transferred count is now maintained using an extended
precision representation (64-bit long long on Alpha, two 32-bit
longword array on VAX). The Kbytes/sec rate is now calculated
using double floating-point precision.
Also took this opportunity to include an estimate of a
sequential or fixed-length file's byte length in the
"150 Opening data connection for..." message.
Unix servers do this and it gives a GUI FTP client something
to use for progress bars or %complete screen updates.
Note this size is not 100% accurate due to file conversion
that might occur for the transfer.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A GATED Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 24-Aug-1999 Alpha and VAX
Images:
TCPIP$GATED.EXE V5.0A-1A
Problem:
When OSPF is configured, GATED process goes away after
running a little while. TCPIP$GATED_OUTPUT.LOG doesn't
show ACCVIO but TCPIP$GATED.LOG shows odd big number or
garbage characters for loopback interface index number.
Solution:
Fix the loop in krt_ifread() that goes through interface
table. When loopback interface was being processed it allowed
to exit the loop with ifrl pointing to uninitialized
space in memory.
ECO B 19-APR-2000 Alpha and VAX
Images:
TCPIP$ARP.EXE V5.0A-1A
TCPIP$NETSTAT.EXE V5.0A-1A
TCPIP$GATED.EXE V5.0A-1B
TCPIP$UCP.EXE V5.0A-1F
Problem:
System can crash executing TCPIP> netstat -rn
Solution:
The NLIST library was modified necessitating the relink
of ARP, NETSTAT, GATED and UCP.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A IFCONFIG Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 16-JUN-2000 Alpha and VAX
Images:
TCPIP$UCP.EXE V5.0A-1G
TCPIP$NETSTAT.EXE V5.0A-1B
TCPIP$PING.EXE V5.0A-1A
TCPIP$IFCONFIG.EXE V5.0A-1A
TCPIP$SYSCONFIG.EXE V5.0A-1A
TCPIP$ESNMP_SERVER.EXE V5.0A-1I
TCPIP$HR_MIB.EXE V5.0A-1I
TCPIP$OS_MIBS.EXE V5.0A-1I
TCPIP$ARP.EXE V5.0A-1B
TCPIP$TRACEROUTE.EXE V5.0A-1A
Problem:
tcpip show route fails with INSVMEM when route database is large
Solution:
Backported fix from tcpip Version 5.1 that addresses this issue in
the unix_api library.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A INET Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 22-Dec-1999 Alpha only
Images:
TCPIP$INETDRIVER.EXE V5.0A-1A
Problem:
Random unexplained crashes with stack corruption.
The stack appears misaligned, and systems crash
with weird PC's.
Solution:
The SUSPEND code for Alpha left the stack too low
by 12 bytes, hence also misaligned. Eventual
returns out of the thread would return to the AST
dispatcher which would attempt an REI instruction
with an 'off' SP. The results are unpredictable,
but most often the loaded PC causes an access
violation.
ECO B 29-Dec-1999 Alpha only
Images:
TCPIP$INETDRIVER.EXE V5.0A-1B
Problem:
Non-quadword alignment set at thread initialization.
Alpha kernel stack is just under 8KB, and should be
larger.
Solution:
Add an unused longword at the top of the thread structure.
Increase the size of the allocated memory to 16KB.
ECO 3 updates
--------------
ECO C 2-OCT-2000 Alpha and VAX
Images:
TCPIP$INETDRIVER.EXE V5.0A-1C
Problem:
I/O cancel does not return the OpenVMS error SS$_CANCEL as
the TGV NETDRIVER does.
Solution:
Added a macro to all return code translations to check for a
valid Unix exit code and, in the absence of one, return the
OpenVMS error code.
ECO D 31-JAN-2001 Alpha and VAX
Images:
TCPIP$INETDRIVER.EXE V5.0A-1D
Problem:
The INETdriver can cause a PGFIPLHI crash in VMS EXE$CANCEL.
Solution:
Fixed the INETdriver to make sure the CCB is paged into
memory before returning to EXE$CANCEL.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A INSTALL Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 17-APR-2000 Alpha and VAX
Source:
TCPIP$CONFIG.COM
Problem:
When running TCPIP$CONFIG for the first time after upgrading from
UCX V4.*, the user is asked if they wish to convert their old
databases to the new format. This conversion may have failed with
the error:%EDT-F-INTERERR, Internal software error; contact a
Compaq support representative%DCL-W-SKPDAT, image data (records not
beginning with "$") ignored
Solution:
Add switch EDTI/EDT/NOCOMMAND to ensure no INI files are
interpreted when EDT starts.
ECO A 21-APR-2000 VAX and Alpha
Sources:
TCPIP$CLEANUP.COM V5.0-01
TCPIP$SHUTDOWN.COM V5.0-008
TCPIP$UCP_STARTUP.COM V6.0-003
Images:
UCX$RPCXDR_SHR.EXE V5.0A-1
UCX$ACCESS_SHR.EXE V5.0A-1
Problem:
TCPIP V5.0A does not deliver copies of UCX$ACCESS_SHR.EXE and
UCX$RPCXDR_SHR.EXE, required so that images linked under
UCX V4.2 can run.
Solution:
Change build and install files to deliver those files.
UCX$RPCXDR_SHR.EXE is a copy of TCPIP$RPCXDR_SHR.EXE (V5.0A-1)
and UCX$ACCESS_SHR.EXE is a copy of TCPIP$ACCESS_SHR.EXE
(V5.0A-1).
ECO A 1-MAY-2000 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1B
TCPIP$OS_MIBS.EXE V5.0A-1B
Problem:
Customer cannot send SNMPv1 trap to client.
Solution:
Backport configuration option for V5.1. Allows customer to set
up text configuration file to supplement information configured
through TCPIP$CONFIG.COM in standard configuration file
SYS$SYSTEM:TCPIP$CONFIGURATION.DAT.
Customer must do the following:
$ COPY TCPIP$SNMP_EXAMPLES:TCPIP$VMS_SNMP_CONF.DAT -
SYS$SYSDEVICE:[TCPIP$SNMP]*.*;
Follow the instructions in that file to set up SNMPv1 trap(s).
Note that settings from standard TCPIP$*.DAT config files supersede
those in TCPIP$VMS_SNMP_CONF.DAT.
ECO A 12-MAY-2000 Alpha and VAX
Sources:
TCPIP$VMS_SNMP_CONF.DAT V5.0-01
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1B
TCPIP$OS_MIBS.EXE.EXE V5.0A-1B
Problem:
SNMPv2 traps are default; V1 format expected by some SNMP clients
(e.g., NT).
Solution:
Backport configuration option for V5.1 that allows customer
to set V1 as default trap type. With this option, traps
configured through TCPIP$CONFIG.COM will be V1 format.
Customer must edit SYS$SYSDEVICE:[TCPIP$SNMP]TCPIP$VMS_SNMP_CONF.DAT
to remove the leading # in the line to uncomment it:
#config tcpip$snmp_v1_trap_default
ECO B 16-JUN-2000 Alpha and VAX
Images:
TCPIP$TNDRIVER.EXE V5.0A-1I
TCPIP$TELNET.EXE V5.0A-1B
Problem:
reverse telnet hangs on read when created with /noidle
Solution:
A new /PERMANENT option was added to create_session, to
start a network connection upon device assignment and
disconnect after deassignment.
ECO C 20-JUN-2000 Alpha and VAX
Images:
TCPIP$VERSIONS.EXE V5.0A-1C
Problem:
Update the TCPIP version that gets displayed using
'TCPIP SHOW VERSION', to indicate ECO 1.
Solution:
Recompile and link the version module.
ECO 2 updates:
--------------
ECO D 29-DEC-2000 Alpha and VAX
Sources:
TCPIP$CONFIG.COM
Problem:
If in UCX$CONFIG or TCPIP$CONFIG you go into 1-"Core Environment"
and then 4-"Interfaces" and change the internet address to a
different subnet address, the question "Enter broadcast mask for
xxxx" shows the old "Broadcast Address" and does not default to
the new "Broadcast Address". This often means that customers end
up with the wrong "Broadcast Address". This problem also exists
with the network mask default.
Solution:
Make sure the default values for the network mask and broadcast
mask are recalculated after a new IP address is entered.
ECO E 12-JAN-2001 Alpha and VAX
Images:
TCPIP$VERSIONS.EXE V5.0A-1E
Problem:
Update the TCPIP version that gets displayed using
'TCPIP SHOW VERSION', to indicate ECO 2.
Solution:
Recompile and link the version module.
ECO 3 updates:
--------------
ECO F 4-MAY-2001 Alpha and VAX
Images:
TCPIP$VERSIONS.EXE V5.0A-1F
Problem:
Update the TCPIP version that gets displayed using
'TCPIP SHOW VERSION', to indicate ECO 3.
Solution:
Recompile and link the version module.
ECO G 21-MAY-2001 Alpha and VAX
Sources:
TCPIP$INET_STARTUP.COM V6.0-09B3
Problem:
Many logical name translations of TCPIP$DEVICE and BG0
while running the Apache test.
Solution:
Defined the logical with an underscore:
$ DEFINE/SYSTEM/EXE TCPIP$DEVICE _BG0:
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A IPC Images
---------------------------------------------------------------------------
ECO 1 updates:
--------------
ECO A 17-AUG-2000 Alpha and VAX
Images:
TCPIP$IPC_SHR.EXE V5.0A-1A
Problem:
We failed to provide a mechanism for an application's AST routine
to abort mainline execution of a select() call. Some applications
require quick response to both I/O events and other events
without the resource consumption associated with busy-waiting.
Solution:
We have added a new routine to be called by the C RTL upon
receiving a signal. This routine will abort a pending
select() operation, if any, and cause that call to return EINTR
in errno, as it would on many UNIX systems.
ECO 2 updates:
--------------
ECO B 11-JAN-2001 Alpha and VAX
Images:
TCPIP$IPC_SHR.EXE V5.0A-1B
Problem:
Original: The new aborting-select behavior, in conjunction with a C
RTL change to activate it, could break some existing applications.
Current: Due to a build problem, there was no visible entrypoint
for the SELECT_ABORT fix in TCPIP$IPC_SHR.EXE ECO A (V5.0A-1A).
One of the symptoms of this problem is very poor performance.
Solution:
Original: Older applications may assume that select() will not
return upon receipt of a signal. Therefore, we abort only if the
logical name TCPIP$SELECT_ABORT_ON_SIGNAL has been defined
requesting this behavior.
Current: Re-build the ipc component from the srclib phase.
ECO 3 updates
--------------
ECO C 26-JAN-2001 Alpha and VAX
Images:
TCPIP$IPC_SHR.EXE V5.0A-1C
Problem:
The dynamic updates being sent by the load broker to a
Windows2000 DNS server were failing with an error similiar to:
unknown response: ans=0, auth=1, add=1, rcode=3
Solution:
Modify the check on the response packet from an SOA query to accept
packets that have an empty answer section, auth=1, and add=1.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A LBROKER Images
---------------------------------------------------------------------------
ECO 1 updates:
--------------
ECO A 30-Jul-1999 Alpha and VAX
Images:
TCPIP$LBROKER.EXE V5.0A-1A
Problem:
'max-members', a configuration option in TCPIP$LBROKER.CONF was not
working when set to 1. It was erroneously returning A records for
all active members to the BIND server instead of 1.
Solution:
The logic was such that it was checking for maxmembercnt > 1
when itshould have been > 0 to allow the possibility of
returning only 1record. Simply needed to change the 1 to a
zero.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A LPD5 Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 15-SEP-1999 Alpha and VAX
Images:
TCPIP$TELNETSYM.EXE V5.0A-1A
Problem:
TCPIP$TELNETSYM.EXE (Telnet Symbiont Process) crashes
with an access violation, %SYSTEM-F-ACCVIO
Solution:
The solution was to add a state variable so that the "users" of
the request_packets would know when it is necessary to free them
and when to assume that they have already been freed. Also
prevent possible reference to freed memory by using *request_id
instead of request_packet->request_id. And, improved telnet
symbiont process dump debug capability by sharing the BFS
runtime library and adding some debug code to the symbiont.
This enables the telnet symbiont to dump its stack into the
error log if/when the process terminates due to fatal error.
ECO B 29-NOV-1999 Alpha and VAX
Images:
TCPIP$TELNETSYM.EXE V5.0A-1B
Problem:
Forward port of fix to V5.0a caused LPD5 to depend on BFS.
Solution:
Reworked original coding of dump_stack to use the
LIB$GET_CURR_INVO_CONTEXT and LIB$GET_PREV_INVO_CONTEXT for an
alpha build. Changed to use BLISS builtins to determine which
code to use. Removed include files that are not needed. Added
more inline comments. Validate the information returned from
LIB$GET_CURR_INVO_CONTEXT; assume that LIB$GET_PREV_INVO_CONTEXT
returns the same size and version as LIB$GET_CURR_INVO_CONTEXT.
Added decode of frame type that is returned by the LIB$ routines.
Changed all calls to DUMP_STACK to TCPIP$$CALL_TRACE.
ECO C 20-DEC-1999 Alpha and VAX
Images:
TCPIP$LPD_SHR.EXE V5.0A-1C
TCPIP$LPD_SMB.EXE V5.0A-1C
TCPIP$LPD_RCV.EXE V5.0A-1C
TCPIP$LPRM.EXE V5.0A-1C
TCPIP$LPQ.EXE V5.0A-1C
Problem:
LPD can insert an unwanted linefeed at the beginning of a
VFC/PRN format text file. LPD's conversion of such files
is too literal with regard to the file's first VFC prefix.
Solution:
Suppress the unwanted linefeed if the new system logical
is defined: UCX$LPD_1ST_VFC_PREFIX_SPECIAL (UCX 4.x)
TCPIP$LPD_1ST_VFC_PREFIX_SPECIAL (TCPIP 5.x)
ECO D 22-DEC-1999 Alpha and VAX
Images:
TCPIP$LPD_SHR.EXE V5.0A-1D
TCPIP$LPD_SMB.EXE V5.0A-1D
TCPIP$LPD_RCV.EXE V5.0A-1D
TCPIP$LPRM.EXE V5.0A-1D
TCPIP$LPQ.EXE V5.0A-1D
Problem:
LPD jobs are not being relayed with SYSTEM-F-TIMEOUT
errors observed. With LPD_DEBUG = 7, log file shows
(lpd$send.c/1444) result of 'select' on socket: 3,
rfds: 0, status: 0. LPD Symbiont may also report an
ACCVIO if of recent UCX vintage (4.2 ECO 3).
Solution:
LPD was ignoring all but the minutes field in the
UCX$LPD_RETRY_INTERVAL logical. Customer had
configured a retry interval of 50 seconds -- less
than a minute, resulting in a select() timeout of 0
when waiting for remote server to acknowledge a
command.
Fixed LPD to calculate its select timeout based on
all the fields of an interval time specification,
from days down to hundreths of a second.
Also fixed ACCVIO which could occur in this situation,
a problem introduced by earlier 70-5-520 fix.
ECO E 5-APR-2000 Alpha and VAX
Images:
TCPIP$LPD_SMB.EXE V5.0A-1E
Problem:
LPD symbiont doesn't accept jobs submitted with the
/LOWERCASE qualifier.
Solution:
Modified the LPD symbiont to tell job controller we support
/LOWERCASE so jobs can be queued with this attribute.
ECO F 8-JUN-2000 Alpha and VAX
Images:
TCPIP$LPD_SHR.EXE V5.0A-1F
TCPIP$LPD_SMB.EXE V5.0A-1F
Problem:
LPD Getting extra blank lines at beginning of a report.
Solution:
Port the UCX 4.2 fix for 70-5-281 to TCP/IP V5.0A & V5.1.
Defined two logicals to control this behavior on
either a per-system basis (TCPIP$LPD_SETUP_NOLF)
or a per-queue basis (TCPIP$LPD_<queue>_SETUP_NOLF)
ECO 3 updates
--------------
ECO G 23-Oct-2000 Alpha
Images:
TCPIP$LPD_RCV.EXE V5.0A-1G
TCPIP$LPD_SHR.EXE V5.0A-1G
TCPIP$LPD_SMB.EXE V5.0A-1G
TCPIP$LPQ.EXE V5.0A-1G
TCPIP$LPRM.EXE V5.0A-1G
Problem:
Due to a build problem a TCPIP Library module did not
get built correctly causing some LPD5 images for Alpha
ECO B to be missing the latest fixes.
Solution:
The TCPIP Library was rebuilt and the LPD5 images were
relinked.
ECO H 9-NOV-2000 Alpha and VAX
Images:
TCPIP$LPD_SHR.EXE V5.0A-1H
TCPIP$LPD_SMB.EXE V5.0A-1H
Problem:
LPD queue hangs up; restarting it doesn't seem to help.
Solution:
Problem has already been fixed in UCX 4.2. Ported fixes for
QAR 30-5-311, QAR 30-5-333 and PTR 70-5-980 to V5.0A and V5.1
ECO I 30-MAR-2001 Alpha and VAX
Images:
TCPIP$TELNETSYM.EXE V5.0A-1I
Problem:
The Telnet Symbiont has a delay of 5 seconds before printing.
This delay was necessary to accomodate the DECserver 300.
A customer has requested that this delay be removed.
Solution:
Added a new logical, TCPIP$TELNETSYM_NO_DECSERVER_300 that
if defined reduces the delay from 5 seconds to a half second.
ECO K 24-APR-2001 Alpha and VAX
Images:
TCPIP$LPRSETUP.EXE V5.0A-1K
Problem:
When adding a printer via LPRSETUP, the user is allowed to enter
comments for the printer's PRINTCAP.DAT entry. However, if the
comment is longer than 77 characters, the PRINTCAP.DAT file can
be corrupted with an overlong comment line and several lines that
should be comments but aren't prefixed by the comment character
('#'). This leads to further problems when trying to delete the
corrupted entry.
Solution:
When adding a comment via LPRSETUP, automatically wrap comment
lines that are greater than 77 characters onto the next line(s).
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A MSG Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 27-SEP-1999 Alpha and VAX
Images:
UCX$SMTP_MAILSHR.EXE V5.0A-1A
UCX$MSG.EXE V5.0A-1A
Problem:
The TCPIP> ANALYZE MAIL /REPAIR operation does not fix
any inconsistencies, but exits with %TCPIP-E-SMTPERROR
and %TCPIP-W-ANA_QUEACC.
Solution:
Initialized the field passed when writing out this message
to avoid the non-printable characters. Allow the analyze mail
to continue to queue other files despite the error caused by
particular invalid files. In addition, changed the message:
TCPIP-I-ANA_NOENTR, internal consistency check error
back to the UCX 4.2 behavior to display the more informational
message:
TCPIP-I-ANA_NOENTR, no queue entry found for file <filename>
Also, added another informational message which aids in determining
which file and possible cause of the failing file message:
UCX-I-ANA_USER_RNF, $SNDJBCW(CREATE_JOB) reports RMS-E-RNF;
Suspect user name <name> is not valid for file:
<filename>
ECO B 13-JAN-2000 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE V5.0A-1C
TCPIP$SMTP_PARSESHR.EXE V5.0A-1C (VAX only)
TCPIP$SMTP_PARSESHR_TV.EXE V5.0A-1C (Alpha only)
TCPIP$SMTP_RECEIVER.EXE V5.0A-1C
TCPIP$SMTP_SFF.EXE V5.0A-1C
TCPIP$SMTP_SYMBIONT.EXE V5.0A-1C
TCPIP$MSG.EXE V5.0A-1B
Problem:
Enhancement to SMTP to provide support for 'Outbound Alias'.
Solution:
Add support for a user defined logical TCPIP$SMTP_FROM which will
be used for the text of the From: and the Return-Path:
headers. Also added support for a system manager logical
TCPIP$SMTP_PROHIBIT_USER_HEADERS which will cause the
TCPIP$SMTP_FROM logical to be ignored effectively no-oping it.
ECO C 24-MAR-2000 Alpha and VAX
Images:
TCPIP$CFS_SHR.EXE V5.0A-1A
TCPIP$MSG.EXE V5.0A-1C
Problem:
Message: cache_check detected corrupted LRU queue appears
on console after a mount from an nfs client hangs.
Solution:
Fixed miss-compare problem. Used STAT_CHE$W_CLUSTER_COUNT
instead of STAT_CHE$W_CLUSTER_LIMIT.
ECO 3 updates
--------------
ECO D 25-APR-2001 Alpha and VAX
Images:
TCPIP$FTP_CLIENT.EXE V5.0A-1F
TCPIP$FTP_CHILD.EXE V5.0A-1I
TCPIP$FTP_SERVER.EXE V5.0A-1I
TCPIP$MSG.EXE V5.0A-1D
Problem:
For file larger than 4 GB, the FTP Client incorrectly reports the
number of bytes transferred. The Kbytes/sec rate reported is also
incorrect.
Solution:
The bytes transferred count is now maintained using an extended
precision representation (64-bit long long on Alpha, two 32-bit
longword array on VAX). The Kbytes/sec rate is now calculated
using double floating-point precision.
Also took this opportunity to include an estimate of a
sequential or fixed-length file's byte length in the
"150 Opening data connection for..." message.
Unix servers do this and it gives a GUI FTP client something
to use for progress bars or %complete screen updates.
Note this size is not 100% accurate due to file conversion
that might occur for the transfer.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A NET Images
---------------------------------------------------------------------------
ECO 1 updates:
--------------
ECO A 20-Jul-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1A (ALPHA)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1A (VAX)
Problem:
Accepted socket doesn't inherit the probe/drop
(tcp_keepidle,tcp_keepintvl, tcp_keepcnt) settings of the
listener socket. This issue is raised by the customer as
backwards compatibility problem. In UCX v4.2 accepted socket
inherited probe/drop settings of the listener socket.
Solution:
Inherit tcp_keepidle, tcp_keepintvl and tcp_keepcnt options in
tcp_input() routine the same way as tcp_keepinit is inherited.
ECO B 28-Jul-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1B (ALPHA)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1B (VAX)
Problem:
The problem becomes visible when the operating system runs out
of dynamic memory and the reason for this is that we have
accumulated thousands of mbufs on the TCP reassembly queue of
a TCPCB.
The socket in question happens to have both its send and
receive socket buffers completely full, and the application
process is currently waiting for a write to the socket to
complete, which never will because the send window is full and
it looks like the other side of this connection is in the same
state, blocked trying to write.
Solution:
First, the immediate reason that we accumulate all these mbufs
is that we call routine, tcp_reass(), and that routine gets
confused and winds up just queueing all of the mbufs.
This is fixed in newer versions of the source ported from
DUNIX and therefore this particular fix will be picked up by
TCP/IP V5.1 from TRU64 source pool. To fix the problem in
V5.0A we merge the the fix from DUNIX 4.0DSUPPORTOS patch
area.
ECO C 31-Aug-1999 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1C
Problem:
Sometimes, the UCP command:
TCPIP SHOW SERVICE/FULL RLOGIN
causes an ACCVIO.
Solution:
Processig an incoming RLOGIN connection results in
modification of the SERV$L_USER_INFO_D and SERV$L_USER_INFO
fields of the service database for the RLOGIN service.
Although these fields are not relevant to a TCPIP SHOW SERVICE
command for RLOGIN, the INETACP was returning the data in
these fields to the UCP program, which in turn mis-interpreted
them, resulting in an ACCVIO. The fix is to have the INETACP
not return the contents of these irrelevant fields for RLOGIN.
ECO D 8-Sep-1999 Alpha and VAX
Images:
TCPIP$INTERNET_SERVICES.EXE V5.0A-1D
Problems:
The TCP_NODELAY socket option was not properly inherited
from listener sockets.
It was not possible to set the FULL_DUPLEX_CLOSE option.
Solutions:
When cloning a new socket, copy the value of the TF_NODELAY
option flag, just as several others are already copied. Allow
applications to set FULL_DUPLEX_CLOSE, and return an error if
that fails. Note that currently, it will only succeed if the
option value length is 4 bytes, representing a change in
behavior from UCX V4.2 which would accept lengths of less than
4 bytes.
ECO E 5-OCT-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1E
Problem:
With TCP/IP V5.0 and VMS 7.2 it's not possible to setup ATM CLIP
interface IF0, %TCPIP-E-INVINTER, error defining interface: IF0
Solution:
When creation of port for IPv6 fails, avoid branching to label
INET_VCI_OPEN.CRE_PORT_FAIL and instead continue with the setup of
the IP and ARP protocol. Also, when a user tries to create an IPv6
for interface other than PPP and loopback, reject it by returning
the error EPROTONOSUPPORT. When an error is returned from
VCI_PORTMGMT_INITIATE/VCI_PORTMGMT_COMPLETE routines for IPv6,
branch to the new routine called VCI_IPv6_ENABLE_PORTS_FAILURE to
handle this error. This routine will disable and delete the IPv6
port and delete the IPv6 VCIB. It will then branch to continue the
interface setup. Also if an unexpected event is returned through
INET$VCI_REPORT_EVENT for VCI_PORTMGMT_INITIATE, also branch to the
new routine VCI_IPv6_ENABLE_PORTS_FAILURE.
ECO F 20-Oct-1999 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1F
Problem:
When PPP interface is not deleted before shutting down TCPIP,
shut down of TCPIP causes system crash. The crash is in
INETACP_DISMOUNT_WITH_LOCKS.
Solution:
In INETACP_DISMOUNT_WITH_LOCKS verify INETCB is not zero before
using it.
When this fix was applied, the shutdown still caused system crash.
But now it was "Multiple deallocation of nonpaged pool". The
solution to this problem was to find all the places where
INETACP_IOPOST is called and after return from INETACP_IOPOST clear
INETACP$GL_SAVE_IRP if it is equal to INETACP$GL_CRE_DEL_SHUT_IRP
to make sure same IRP pwd is not posted twice.
ECO G 20-Oct-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1G (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1G (VAX)
Problem:
The SYS$REM_ID logical does not get set up for network jobs.
Solution:
In INET_SETMODE, copy (at most 15) bytes from CTL$GQ_REMOTE_UID
to CTL$T_REMOTEID. This is needed because LOGINOUT is still using
the old P1 global.
Local testing after fixing TELNET and RLOGIN problems with this.
ECO H 21-Oct-1999 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1H
Problem:
$tcpip set int pp0 /ser=tta1 /net=x.x.x.x
Crashes the system.
$tcpip set int pp0/ser=ttb0
%TCPIP-E-INVINTER, error defining interface: PP0
-TCPIP-I-ACPQIO, failure on internet ACP QIO
-SYSTEM-W-NOMSG, Message number 0000FCE0
^^^^^^^^
error # not consistent.
Solution:
Both SLIP and PPP make use of INETACP_CREATE_SLIP_INIT to
create slip or ppp interface. INETACP_CREATE_SLIP_INIT
calls CREATE_P5_BUFFER to construct P5 parameter. P5
parameter is constructed from Attritrbute Control Block.
The assumption that the last attribute is garanteed to be
routing data is wrong. Modify CREATE_P5_BUFFER to consider
three cases.
ECO K 22-Oct-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1K (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1K (VAX)
Problem:
The SYS$REM_ID, SYS$REM_NODE do not get set for RLOGIN and TELNET
sessions.
Solution:
Add code to support delayed setting of these logicals into TNDRIVER.
The new QIO for the items will handle this.
ECO L 28-Oct-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1L (ALPHA)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1L (VAX)
Problem:
Infinite loop at IPL 8 on a single CPU system, or a CPUSPINWAIT
crash on a multiple CPU system.
For an as yet unknown reason, but probably the receipt of a RESET,
incoming connections to a TCP listener socket are aborted very soon
after the SYN is received that initiates the connection and leads
to the creation of a socket for the connection. Because these
nascent sockets have not yet been accepted, and sometimes the
connections have not been fully established, the sockets in
question are found on the so_q or the so_q0 queue headers of the
listener socket. These aborted nascent sockets then are left on
these queues indefinitely until the listener socket is closed. At
that time, in an attempt to cleanup these queues the BGDRIVER goes
into an infinite loop.
Solution:
In trying to clear these queues, a call is made to INET_SOABORT,
passing the address of the socket at the head of the queue in
question. The caller of this routine fully expects that the socket
in question will then be deallocated and removed from the queue.
However, the sockets in this case, are in a state, TCPS_FREE, that
causes the above call to degenerate into a NO-OP, resulting in a
loop because the caller then calls again passing what it hopes is
the new head of the queue, but is in fact the same un-abortable
socket.
The solution is to determine after the call if the socket passed in
has been removed from the queue, and if not, it is forceably
removed at this time.
ECO M 09-Nov-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1M (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1M (VAX)
Problem:
INETACP kernel crash when performing a proxy login.
Solution:
Update V5.0 with a correction applied to T5.1 to handle cases
where the upper 16-bits of the length field of the descriptors
passed on setting the SYS$REM* logicals from INETACP are non-zero.
ECO N 11-Nov-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1N (Alpha)
Problem:
This problem became apparent on a muti-home machine where ethernet
interface was in subnet 10.10.4.0 and FDDI interface was in subnet
10.10.3.0. PPP interface was set up to be in 10.10.3.0 subnet,
however when ppp connection came up and Proxy ARP was created for
the other end\ which had IP address 10.10.3.11 and subnet mask
255.255.255.0, Proxy AR\P was not created correctly. Instead of
mapping remote IP address to hard\ware address of FDDI interface
which was in the same subnet as PPP interface\ it mapped it to the
hardware address of ethernet interface which was in 10.10.4.0
subnet.
Solution:
Fix a bug in Proxy ARP logic. Convert IA$L_SUBNETMASK from host
byte order to network byte order when comparing subnets of the
interfaces.
ECO P 11-Nov-1999 Alpha and VAX
Problem:
RLOGIN fails to create the process; internal SS$_INSFMEM error.
Solution:
Various strings were not properly initialized for ALL cases.
Handle this properly.
Images:
TCPIP$INETACP.EXE V5.0A-1P
ECO Q 14-Nov-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1Q (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1Q (VAX)
Problem:
1. Crash (INCONSTATE) in routine, KERNEL_DEALLOC_DCBE_CHAIN.
2. Read requests specifying LOCKBUF may result in indefinite
hangs.
Solution:
1. The solution is to simply clear the MBUF$L_OVMS_DWAB field in
each MBUF allocated in ALLOC_COUNTED_MBUF_CHAIN, at the time the
MBUF is allocated.
2. The problem occurs when a LOCKBUF read request originally finds
that there is some data in the receive sockbuf, but not enough to
fully satisfy the the request AND during the transfer of data from
the sockbuf to the user's buffer, when we are unsynchronized, more
data (exactly enough to satisfy the request but not one byte more)
arrives.
Because less than the amount requested was present at the
beginning, a partial read. At the end of transferrng data, which
in fact transferred all the requested data including the data that
arrived in the middle of the transfer, we failed to test to see if
the request was completely satisfied before going to "wait".
The fix is to test for this condition and terminate the request
when we notice that the request has been completed.
ECO R 16-Nov-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1R (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1R (VAX)
Problem:
Memory leak of M_RTABLE structures.
Solution:
In routine, rt_setgate(), a sockaddr structure is allocated,
although it is characterized as an M_RTABLE structure in the
MALLOC, and its address is recorded in the rt_gateway field of an
rt structure. When the corresponding rt structure is deallocated,
the sockaddr is not deallocated, leading to a memory leak.
The fix for this is to add code to routine, rtfree(), whereby
immediately before deallocating the rt structure, we first
deallocate the sockaddr if it exists.
ECO S 19-Nov-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1S (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1S (VAX)
Problem:
If the sysconfig settable variables tcp_recvspace and tcp_sendspace
are set higher than the compile time constant SB_MAX it becomes
impossible to create TCP sockets. V5.0 was compiled with SB_MAX
defined as 128*1024.
Solution:
SB_MAX is defined in <sys/socketvar.h>. We redefined it to
1024*1024.
ECO T 29-Nov-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1T (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1T (VAX)
Problem:
DECnet over IP connections are not failing when they should. While
running a DECnet over IP application, if the network cable is
disconnected, the DECnet application does not get an error message.
This worked in UCX V4.2 and earlier versions.
Solution:
The problem here arose because of the way that PWIP was implemented
in V5.0 of the TCPIP product. That implementation contains a major
optimization in that PWIP no longer copies DECnet (and PathWorks)
data but rather passes this data straight through to the TCP/IP
code. As a result, in the case of TCP connections, the DECnet
buffers are not returned to DECnet until TCP is done with them,
which means until the other side of the DECnet connection has
acknowledged receipt of the data. When the network cable is
disconnected, TCP eventually times out the connection and informs
DECnet (via PWIP) that this has occurred. DECnet then waits for
its buffers to be returned until it proceeds to inform its user
that this has occurred.
However, TCPIP will not return the buffers until DECnet closes the
socket, and DECnet will not close the socket until its buffers are
returned.
The soultion to this standoff was to introduce code into routine,
sofree(), to recognize this particular case, and to unconditionally
free buffers waiting in the send sockbuf of such a socket. This
change only affects kernel sockets (i.e. those created by PWIP and
the TELNET server) and therefore does not cause major side-effects.
ECO U 3-Dec-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1U (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1U (VAX)
Problem:
Crash at SOREMQUE+0 due to a socket address passed as a parameter.
Solution:
This crash is a direct result of an typo in the fix done in ECO L.
Unfortunately the socket to be removed from the queue was
incorrectly specified. The fix is to specify it correctly.
ECO V 5-Dec-1999 VAX only
Images:
TCPIP$INETACP.EXE V5.0A-1V
Problem:
Only on VAX, where PPP is not supported
$tcpip set int pp0 /ser=tta1 /net=x.x.x.x
Crashes the system.
Solution:
1. In CREATE_P5_BUFFER routine, handle the case when someone
tries to create PPn interface without /host on VAX. PPP is
not supported on VAX and we must return error instead of
continuing under assumption of having valid routing data.
2. In routine INETACP_CREATE_SLIP_INIT. When determining whether
creating PPP interface or SLIP interface, if it is a PPP interface,
continue only in case of ALPHA. Return ss$_unsupported in case of
VAX.
ECO W 14-Dec-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1W (ALPHA)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1W (VAX)
Problem:
In the case of TCP socket, when a call to recv() returns -1 in
UCX V4.2, the same call to recv() returns 0 in TCPIP V5.0A.
Solution:
In routine, INET_SORECEIVE_STREAM, fix a bug in error reporting
in case of an empty sockbuf.
ECO X 19-JAN-2000 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1X
Problem:
If TFTP service is configured with an accept or reject
host/network list, and the very first TFTP connection
comes from a rejected host, all subsequent TFTP requests
are rejected, even from acceptable hosts. A converse
situation also exists: if the first TFTP connection
comes from an accepted host, TFTP will accept all further
requests, even those from rejected hosts.
Solution:
The first problem, a denial-of-service issue, occurred
because the INETACP was leaving the rejected TFTP request
in the TFTP service's receive socket buffer, where it
was reprocessed (and re-rejected) for every new request
that came in. The INETACP now clears rejected requests
from the socket.
The second problem, a security concern, occurs because
the INETACP spawns a TFTP process to service all future
TFTP requests when the first acceptable request comes in,
and it passes off the TFTP service socket to that process.
Accept/Reject processing only occurs within the INETACP
-- there is no code in TFTP to do this itself. This is a
design limitation of the INETACP and TFTP service which is
not fixable within the scope of an ECO. For sites that
consider this a serious security issue, two workarounds
exist: (1) disable the TFTP service entirely, or (2)
configure UCX to reject *all* communications from a
rejected host/network (using SET COMM/ACCEPT/REJECT)
instead of configuring it on a per-service basis.
ECO Y 8-FEB-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1Y (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1Y (VAX)
Problem:
Several machines (VAX and Alpha) in different clusters
crash at the same time.
Solution:
Fixed TCPIP not to mishandle a redundant ICMP REDIRECT
packet in which the source and gateway addresses are
the same.
ECO Z 22-MAR-1999 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1Z
Problem:
With TCP/IP V5.0A PTRV50A-05_AXP7 and VMS 7.2-1 it's not
possible to setup ATM CLIP interface if0, %TCPIP-E-INVINTER,
error defining interface.
The fix for 70-5-1117 that resolved this problem was not
properly built.
Solution:
Rebuild with the kernel module INET_IF_VCI which did not
get in the INETDRV object library when 70-5-1117 was built.
ECO AA 20-APR-2000 Alpha and VAX
Images:
TCPIP$INTERNET_SERVICES.EXE V5.0A-1AA
Problem:
IP multicast applications which ran properly under UCX V4 using
the REUSEADDR socket option no longer work because the new-for-V5
kernel has added a REUSEPORT option which must be set separately.
Solution:
In order to maximize compatibility, automatically set REUSEPORT
when an application sets REUSEADDR on a DGRAM (UDP) socket.
ECO AB 17-MAY-2000 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1AB
Problem:
After upgrading to latest V5.0A patch, Telnet users from a
terminal server who mistype their password enough times can
trigger the VMS Intrusion Detection mechanism and cause all
Telnet users from that terminal server to be locked out.
This problem was introduced by a recent change to Telnet to
"properly" support the DECNET network login logicals
SYS$REM_ID, SYS$REM_NODE, and SYS$NODE_FULLNAME, as was done
in UCX. The intrusion detection mechanism utilizes
SYS$REM_NODE and SYS$REM_ID to identify the "Source" of an
intrusion. When these aren't available, it uses TT_ACCPORNAM
information. Unlike DECNET (or rlogin), the Telnet protocol
provides no way of identifying the particular username on the
remote system which is logging in -- so the SYS$REM_ID value
that Telnet constructs is simply a representation of the IP
address of the remote system. The TT_ACCPORNAM information
also includes the peer's TCP port number, which is generally
unique and ephemeral.
Solution:
Engineering's position is that the recent changes to V5.0A
Telnet allow the OpenVMS intrusion mechanism to better enforce
its configured policies. There are two ways to mitigate the
problem of inadvertant intrusion lockout with Telnet:
1. require clients use Rlogin instead of Telnet.
2. loosen the intrusion detection policies on the system
through appropriate tuning of the SYSGEN LGI* parameters.
However, for those customers for which neither of the above are
palatable, a new system logical has been implemented:
TCPIP$TELNET_NO_REM_ID
Defining this logical to "TRUE" reverts Telnet to its original
V5.0 behavior of not setting any SYS$REM* logical whatsoever.
Customers are warned that the use of this logical will
effectively bypass the intrusion-detection mechanism for Telnet
logins.
ECO AC 18-JUL-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AC (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1AC (VAX)
Problem:
Socket passed to tcp_ctloutput() contained NULL so_pcb field,
which caused the system to crash.
Solution:
Backport V5.0A fix to check for non-NULL so_pcb field before
reference and return an error if it is NULL. Also initialize
the placeholder for the IRP address, which contains the so_pcb
field. This part was also missing in V5.x.
ECO AD 18-AUG-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AD
TCPIP$INETACP.EXE V5.0A-1AD
TCPIP$INTERNET_SERVICES.EXE V5.0A-1AD (VAX)
Problem:
Some socket options described in the documentation have not
been implemented.
Solution:
The followig options were defined:
1. REUSEPORT at socket level
2. TCP_KEEPIDLE, TCP_KEEPINTVL, TCP_KEEPCNT, TCP_KEEPINIT
and TCP_NODELACK at TCP level
3. IP_OPTIONS, IP_HDRINCL, IP_RECVOPTS and IP_RECVDSTADDR
at IP level
ECO 2 updates:
--------------
ECO AE 2-OCT-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AE
Problem:
PGFIPLHI bugcheck, TCPIP$BGDRIVER+14E2C found accessing a
pageable VA at IPL8.
Solution:
A movl instruction referencing R11 was moved outside the lock
IOLOCK8 and the lock was changed to preserve R0.
ECO AF 6-OCT-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AF
TCP/IP has a memory leak of mbufs when path MTU discovery is
happening.
Solution:
Fixed reference-counting problem in path MTU discovery code and in
nolisten UDP service startup code which was preventing socket and
related protocol mbufs from being freed.
ECO AG 9-OCT-2000 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1AG
Problem:
Executing the "show user" command, users connected via telnet
have their IP address appearing instead of their host name.
Solution:
Ported a simple version of the 5.1 fix to 5.0A. This included
moving the hostname to a global area to be mapped later by
INETACP$TN_START_SESSION which formats for printing the host
and port information.
ECO AH 13-OCT-2000 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1AH
Problem:
Unable to suppress several rlogin-related OPCOM messages.
Solution:
rlogin-related OPCOM messages which were produced are
now suppressed when the related service (RSH/RLOGIN/
REXEC) has been configured via SET SERVICE
/LOG_OPTIONS=(NOLOGIN).
ECO AI 20-OCT-2000 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1AI
Problem:
A user doing RLOGIN into VMS system running TCPIP V5.x will
sometimes see the message:
[TNDRIVER: Proxy login for user "xxxxxxx" in progress ...]
Some customers want the ability to suppress this message.
Solution:
The system logical TCPIP$RLOGIN_SILENT can now be used to
indicate whether these messages should be suppressed by
the RLOGIN server. Defining this logical to be "TRUE",
"YES", or "1" will suppress the message.
This logical affects all incoming rlogin users system-wide.
Unfortunately it is not feasible to make this message
suppressable on a per-user or per-group basis.
ECO AK 22-NOV-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AJ (Alpha only)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1AJ (VAX only)
TCPIP$INETACP.EXE V5.0A-1AJ
Problem:
There is a synchronization hole for an incoming UDP datagram to
be replayed twice -- first successfully, then again through a
failure path causing a system crash.
Solution:
Close the synch hole by removing the function for the dequeuing of
the REQCB from the BGdriver and place this function in the INETACP.
ECO AL 4-Dec-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AL
TCPIP$INETACP.EXE V5.0A-1AL
Problem:
The xxx.sin_zero field was not being initialized to zero as it
was in UCX 4.2, causing older UCX application programs to fail
if they did not zero out this field before calling bind.
Solution:
Bind was modified to clear the sin_zero field so that older
application programs that were not clearing this field before
the call would still work as they did with UCX 4.2.
ECO AM 21-DEC-2000 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-AM (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-AM (VAX)
Problem:
UDP packets with message length zero fail in sendto(). This
caused the routine to exit prior to the deallocation of the
associated MT_SONAME mbuf. Eventually, this resource would
be exhausted.
Solution:
Fixed problem of MT_SONAME mbuf resource exhaustion when sending
zero length UDP datagrams, and made sure applications can both
send and receive such datagrams on the wire.
ECO 3 updates
--------------
ECO AN 14-FEB-2001 Alpha only
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AN
TCPIP$INETACP.EXE V5.0A-1AN
Problem:
ARP cache in TCPIP V5.0 cannot be larger than 592 entries.
Solution:
In TCPIP V5.0A, a larger default ARP table is configured for
Alpha systems with 79 buckets * 37 tabs/bucket = 2923 entries.
VAX systems still use 592 entry ARP table.
The sysconfig net subsystem arptab_nb parameter has been
removed -- this parameter was not meant to be tuned
dynamically and a crash could result if it were.
ECO AO 26-FEB-2001 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AO
Problem:
INVEXCEPTN Crash during a free of an mbuf within a freed mbag
while TCPIP is being, or has been, shut down.
Solution:
Change shutdown process to avoid deallocating mbags and reuse
them during initialization on subsequent startups.
ECO AP 3-MAR-2001 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AP
Problem:
In the RSH/REXEC server, the print-format carriage control is not
properly handled. This is particularly a problem for COBOL
applications, whose output is incorrectly displayed with 137 blank
lines between every two lines of output.
Solution:
The RSH/REXEC server has been corrected to deal with certain
carriage control specifications, including those commonly used by
COBOL applications.
ECO AQ 28-MAR-2001 Alpha and VAX
Images:
TCPIP$INETACP.EXE V5.0A-1AQ
Problem:
"portname" string overflowed with unusually long host-domain
name, causing stack overflow and system crash.
Solution:
Limit the size of the portname string to 64 characters.
ECO AR 11-APR-2001 Alpha and VAX
Images:
TCPIP$TNDRIVER.EXE V5.0A-1Y
TCPIP$INETACP.EXE V5.0A-1AR
Note: even though both NET and TELNET_SERVER images are
affected by this set of changes, the two images are NOT
fatally interdependent. Either can safely be shipped to
a customer running older version of the other image;
they just won't get the new XON_ANY_CHAR functionality.
Problem:
70-5-1601:
Telnet client or user was sending ^S XOFF stopping output,
and not following up with a ^Q XON to restart output. This
can lead to a situation where the Telnet server doesn't
detect when client has closed, leaving the TNA device active
and the user process logged in. Only remedy required system
administrator to manually $STOP the affected process.
70-5-1602 & 70-5-1261:
The TNDRIVER crashed due to unexpected recursion resulting
in the same packet being freed twice.
Solution:
70-5-1601:
Telnet server no longer ignores socket events like
closure when unable to pass any further data up to
the TT driver or application because input is disabled
(either due to Typeahead buffer being full or use of
ReadSync terminal characteristic).
Furthermore, to help deal with defective Telnet client
or inattentive user that sends ^S XOFF but doesn't
follow up with ^Q XON, an implicit ^Q XON will occur
upon receipt of any character (not just ^Q) from the
client if the system logical TCPIP$TELNET_XON_ANY_CHAR
is defined as "1". A similar logical has also been
provided for rlogin service: TCPIP$RLOGIN_XON_ANY_CHAR.
70-5-1602 & 70-5-1261:
Prevent unexpected recursion and let original call complete
processing of the packet.
ECO AS 12-APR-2001 Alpha
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AS
TCPIP$INETACP.EXE V5.0A-1AS
Problem:
The BGDRIVER crashes when terminating an I/O request.
Solution:
Within the RCV_XMT_COMMON_PREPROCESSING routine, move code from one
place, where the IRP fields were only conditionally initialized, to the
beginning of the routine, where the fields are sure to be initialized.
ECO AT 18-APR-2001 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AT(ALPHA)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1AT(VAX)
Problem:
SNMP does not report the correct speed for an interface. On V5.0A
all ethernet interfaces report 10MB even gigabit ethernet which
should report 1000MB, while FDDI correctly reported 100MB.
Solution:
Add code to interface initalization code in the Kernel to properly
initialize the if_baudrate field in the IFNET structure.
ECO AU 23-APR-2001 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AU
TCPIP$INETACP.EXE V5.0A-1AU
Problem:
Crash in INET_SEL_BUILD_PAR_LIST.
Solution:
Backport modifications from V5.1's INET_SEL_INIT_SYS_LIST()
routine to V5.0A's INET_SEL_BUILD_PAR_LIST() routine to properly
check that every entry on the list is in fact a BG UCB.
ECO AV 23-APR-2001 Alpha and VAX
Images:
TCPIP$BGDRIVER.EXE V5.0A-1AV (Alpha)
TCPIP$INTERNET_SERVICES.EXE V5.0A-1AV (VAX)
Problem:
Extra initial blank line is output from RSH and REXEC server.
Solution:
Fixed flawed code that had been designed to remove this extra
line feed from the beginning of RSH/REXEC server output.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A NETSTAT Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 19-APR-2000 Alpha and VAX
Images:
TCPIP$ARP.EXE V5.0A-1A
TCPIP$NETSTAT.EXE V5.0A-1A
TCPIP$GATED.EXE V5.0A-1B
TCPIP$UCP.EXE V5.0A-1F
Problem:
System can crash executing TCPIP> netstat -rn
Solution:
The NLIST library was modified necessitating the relink
of ARP, NETSTAT, GATED and UCP.
ECO B 16-JUN-2000 Alpha and VAX
Images:
TCPIP$UCP.EXE V5.0A-1G
TCPIP$NETSTAT.EXE V5.0A-1B
TCPIP$PING.EXE V5.0A-1A
TCPIP$IFCONFIG.EXE V5.0A-1A
TCPIP$SYSCONFIG.EXE V5.0A-1A
TCPIP$ESNMP_SERVER.EXE V5.0A-1I
TCPIP$HR_MIB.EXE V5.0A-1I
TCPIP$OS_MIBS.EXE V5.0A-1I
TCPIP$ARP.EXE V5.0A-1B
TCPIP$TRACEROUTE.EXE V5.0A-1A
Problem:
tcpip show route fails with INSVMEM when route database is large
Solution:
Backported fix from tcpip Version 5.1 that addresses this issue in
the unix_api library.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A NETUTIL6 Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 16-JUN-2000 Alpha and VAX
Images:
TCPIP$UCP.EXE V5.0A-1G
TCPIP$NETSTAT.EXE V5.0A-1B
TCPIP$PING.EXE V5.0A-1A
TCPIP$IFCONFIG.EXE V5.0A-1A
TCPIP$SYSCONFIG.EXE V5.0A-1A
TCPIP$ESNMP_SERVER.EXE V5.0A-1I
TCPIP$HR_MIB.EXE V5.0A-1I
TCPIP$OS_MIBS.EXE V5.0A-1I
TCPIP$ARP.EXE V5.0A-1B
TCPIP$TRACEROUTE.EXE V5.0A-1A
Problem:
tcpip show route fails with INSVMEM when route database is large
Solution:
Backported fix from tcpip Version 5.1 that addresses this issue in
the unix_api library.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A NFS_SERVER Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 3-NOV-1999 Alpha and VAX
Images:
TCPIP$NFS_SERVER.EXE V5.0A-1A
Problem:
"showmount -e <VMS NFS Server>" gets the error: Export list > UDP
max. Data truncated.
Solution:
Added new logical tcpip$nfs_trim_export to reduce the amount of
data a client will receive. If the logical is set to 1, check to
see if the client can mount a given pathname and only return the
pathname and the client name that matched. Do not return all
client names nor all pathnames. If the logical is not set to 1,
return the export list the old way.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A NLIST Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 19-APR-2000 Alpha and VAX
Images:
TCPIP$ARP.EXE V5.0A-1A
TCPIP$NETSTAT.EXE V5.0A-1A
TCPIP$GATED.EXE V5.0A-1B
TCPIP$UCP.EXE V5.0A-1F
Problem:
System can crash executing TCPIP> netstat -rn
Solution:
The NLIST library was modified necessitating the relink
of ARP, NETSTAT, GATED and UCP.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A POP Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 13-JAN-2000 Alpha and VAX
Images:
TCPIP$POP_SERVER.EXE V5.0A-1A
TCPIP$SMTP_MAILSHR.EXE V5.0A-1B
TCPIP$SMTP_PARSESHR.EXE V5.0A-1B (VAX only)
TCPIP$SMTP_PARSESHR_TV.EXE V5.0A-1B (Alpha only)
TCPIP$SMTP_RECEIVER.EXE V5.0A-1B
TCPIP$SMTP_SFF.EXE V5.0A-1B
TCPIP$SMTP_SYMBIONT.EXE V5.0A-1B
Problem:
Images supplied with CFS.70257 do not work correctly. Incoming
mail is not queued into SMTP-Queue. Message number 0000803A.
Solution:
Fixed internal build issues.
ECO B 19-APR-2000 Alpha and VAX
Images:
TCPIP$POP_SERVER.EXE V5.0A-1B
Problem:
POP server consistently causes client error/timeout on
certain messages.
Solution:
Reenable blocking socket I/O when writing the end-of-message
sequence.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A PWIP Images
---------------------------------------------------------------------------
ECO 1 updates:
--------------
ECO A 11-Nov-1999 Alpha and VAX
Images:
TCPIP$PWIPDRIVER.EXE V5.0A-1A
Problem:
When running with V5.0A, DECnet/TCPIP may experience memory
leakage.
Solution:
A fork was removed that had previously been added before dissolving
a Kernel association. This fork was not necessary as associated
NET changes had obivated the need for it.
ECO 3 updates
--------------
ECO B 20-Sep-2000 Alpha and VAX
Images:
TCPIP$PWIPDRIVER.EXE V5.0A-1B
Problem:
After either TCPIP shutdown or PWIP shutdown, PWIP cannot be
restarted. An attempt to restart PWIP results in an entry in the
PWIP logfile that says:
"interfaceInit : PWIPdriver initialization, address vector alloc
faliure: 708"
Solution:
Changes were made in the PWIP startup code to resolve this
problem.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A RCP Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 19-APR-2000 Alpha and VAX
Image:
TCPIP$RCP.EXE V5.0A-1A
Problem:
Using $RCP/Preserve to 'put' a file will hang on OpenVMS VAX
and some other versions of OpenVMS.
The file gets created with correct contents but the client
server handshake gets out of sequence goes into an infinite wait.
Solution:
The code which sets file protection and modification date was
rewritten. The output file revision date will now be correct
(truncated to the second) with /Preserve for both architectures
and all versions of OpenVMS.
ECO 3 updates
--------------
ECO B 27-NOV-2000 Alpha and VAX
Images:
TCPIP$RCP.EXE V5.0A-1B
Problem:
$copy/rcp gives incorrect return status when a file locked
error occurs.
Solution:
Have the open calls return the correct error status, and have
the last error signalled via LIB$SIGNAL be returned as the
error status. Also, set the vaxc$errno status when errors are
returned from C routines, and return that as the error status.
ECO C 13-FEB-2001 Alpha and VAX
Images:
TCPIP$RCP.EXE V5.0A-1C
Problem:
"$ COPY/RCP" on a binary file changes the file attributes
from fixed length to STREAM LF and corrupts the file.
Solution:
Forward port the fixes for 70-5-1006, 70-5-960 and 70-5-1286 from
UCX 4.2.
RCP will now treat VMS fixed-length (FIX) and undefined (UDF)
format files as binary and copy them verbatim without doing the
implicit conversion to Stream_LF. Note however that RCP will NOT
preserve file attributes on the target file (which will still be
created as Stream_LF). To fix this is beyond the scope of an ECO
and may not be possible without incompatibly breaking the RCP
protocol. File attributes on the target will have to manually be
corrected by the user.
Lest some customers have come to depend on the original behavior of
RCP, that behavior can be restored by setting the logical
TCPIP$RCP_SEND_FIX_FORMAT_AS_ASCII to "1", "True", or "Yes".
However, if this logical is set to a numeric value that exactly
matches the fixed-length file's record size, that file will not be
converted to Stream_LF, but will be sent as binary.
Note also, the TCPIP$RCP_SEND_FIX_FORMAT_AS_ASCII logical name will
superceed the use of the UCX$RCP_SEND_FIX_FORMAT_AS_ASCII logical
name. Both are accepted logicals.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A ROUTED Images
---------------------------------------------------------------------------
ECO 3 updates
--------------
ECO A 5-DEC-2000 Alpha and Vax
Images:
TCPIP$ROUTED.EXE V5.0A-1A
Problem:
Routed causes routes to disappear every few minutes, only
to reappear shortly thereafter. This process repeats.
Solution:
Make sure routed is consistent about how it handles the
new sa_len field in the various sockaddr_in structures
passed from routers using the older sockaddr_in definition.
ECO B 26-APR-2001 Alpha and VAX
Images:
TCPIP$ROUTED.EXE V5.0A-1B
Problem:
This problem occurs when the supply option is enabled for
routed. This option is to supply its own Internet addresses
to routing requests made from remote hosts. The RIP's that
are sent to these remote hosts do not have the right packet
format. Thus, this caused all the listeners of RIP to
reject these packets.
Solution:
The fix is to change the outgoing packets to the correct
RIP v1 format. The change would be to set the sa_len field to 0.
Thus have the correct RIP v1 format which would then be accepted
by remote hosts.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A RPC Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 21-APR-2000 VAX and Alpha
Images:
UCX$RPCXDR_SHR.EXE V5.0A-1
UCX$ACCESS_SHR.EXE V5.0A-1
Problem:
TCPIP V5.0A does not deliver copies of UCX$ACCESS_SHR.EXE and
UCX$RPCXDR_SHR.EXE, required so that images linked under
UCX V4.2 can run.
Solution:
Change build and install files to deliver those files.
UCX$RPCXDR_SHR.EXE is a copy of TCPIP$RPCXDR_SHR.EXE (V5.0A-1)
and UCX$ACCESS_SHR.EXE is a copy of TCPIP$ACCESS_SHR.EXE
(V5.0A-1).
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A RSH_CLIENT Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 19-APR-2000 Alpha and VAX
Images:
TCPIP$RSH.EXE V5.0A-1A
Problem:
Output from RSH is sometimes truncated or garbled.
Solution:
The RSH command was corrected to deal with lengthy output
which sometimes caused this problem.
ECO B 20-APR-2000 Alpha and VAX
Images:
TCPIP$RSH.EXE V5.0A-1B
Problem:
RSH a.b.c.d generates ACCVIO if a.b.c.d is an unknown host.
Solution:
Modify RSH to check for a NULL return from gethostbyname()
for an unknown host.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A SMTP Images
---------------------------------------------------------------------------
ECO 1 updates
--------------
ECO A 27-SEP-1999 Alpha and VAX
Images:
UCX$SMTP_MAILSHR.EXE V5.0A-1A
UCX$MSG.EXE V5.0A-1A
Problem:
The TCPIP> ANALYZE MAIL /REPAIR operation does not fix any
inconsistencies, but exits with %TCPIP-E-SMTPERROR and
%TCPIP-W-ANA_QUEACC. These errors show non-printable characters
and customer needs to reset his terminal.
Solution:
Initialized the queue descriptor. In addition, change the message:
TCPIP-I-ANA_NOENTR, internal consistency check error
back to the UCX 4.2 behavior to display the more informational
message:
TCPIP-I-ANA_NOENTR, no queue entry found for file <filename>
Also, added another informational message which aids in determining
which file and possible cause of the failing file message:
TCPIP-I-ANA_USER_RNF, $SNDJBCW(CREATE_JOB) reports RMS-E-RNF;
Suspect user name <name> is not valid for file:
<filename>
Allow the analyze mail to continue to queue other files despite
the error caused by particular invalid files.
ECO B 13-JAN-2000 Alpha and VAX
Images:
TCPIP$POP_SERVER.EXE V5.0A-1A
TCPIP$SMTP_MAILSHR.EXE V5.0A-1B
TCPIP$SMTP_PARSESHR.EXE V5.0A-1B (VAX only)
TCPIP$SMTP_PARSESHR_TV.EXE V5.0A-1B (Alpha only)
TCPIP$SMTP_RECEIVER.EXE V5.0A-1B
TCPIP$SMTP_SFF.EXE V5.0A-1B
TCPIP$SMTP_SYMBIONT.EXE V5.0A-1B
Problem:
Images supplied with CFS.70257 do not work correctly. Incoming
mail is not queued into SMTP-Queue. Message number 0000803A.
Solution:
Fixed internal build issues.
ECO C 14-JAN-2000 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE V5.0A-1C
TCPIP$SMTP_PARSESHR.EXE V5.0A-1C (VAX only)
TCPIP$SMTP_PARSESHR_TV.EXE V5.0A-1C (Alpha only)
TCPIP$SMTP_RECEIVER.EXE V5.0A-1C
TCPIP$SMTP_SFF.EXE V5.0A-1C
TCPIP$SMTP_SYMBIONT.EXE V5.0A-1C
TCPIP$MSG.EXE V5.0A-1B
Problem:
Enhancement to SMTP to provide support for 'Outbound Alias'.
Solution:
Add support for a user defined logical TCPIP$SMTP_FROM which will
be used for the text of the From: and the Return-Path:
headers. Also added support for a system manager logical
TCPIP$SMTP_PROHIBIT_USER_HEADERS which will cause the
TCPIP$SMTP_FROM logical to be ignored effectively no-oping it.
ECO D 21-APR-2000 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE V5.0A-1D
TCPIP$SMTP_PARSESHR.EXE V5.0A-1D (VAX only)
TCPIP$SMTP_PARSESHR_TV.EXE V5.0A-1D (Alpha only)
TCPIP$SMTP_RECEIVER.EXE V5.0A-1D
TCPIP$SMTP_SFF.EXE V5.0A-1D
TCPIP$SMTP_SYMBIONT.EXE V5.0A-1D
Problem:
When composing a mail message our SMTP MAIL$PROTOCOL code strips
off any comments you may enter into the recipient address. This
makes it impossible to use the new mail-to-FAX tool since it uses
commands put into the To line to tell it who the sender of the FAX
is and what the number is to send to.
Solution:
Allow user to tell SMTP to use the text of the recipient address that
they enter verbatim in the To: header rather than stripping off the
comments which is the default behavior.
ECO E 15-JUN-2000 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE; V5.0A-1E
TCPIP$SMTP_PARSESHR.EXE; V5.0A-1E
TCPIP$SMTP_RECEIVER.EXE; V5.0A-1E
TCPIP$SMTP_SFF.EXE; V5.0A-1E
TCPIP$SMTP_SYMBIONT.EXE; V5.0A-1E
Problem:
70-5-1225: TCPIP$SMTP print queue is stopping several days a week.
A process dump is being created each time this happens.
TCPIP_BUGS 1524: A decision was made to remove the "easily
defeated" privilege check for SSF in V5.0A. See TCPIP_BUGS 1524
for the more complex fix that already went into V5.1.
Solution:
70-5-1225: Ensure that SMTP properly truncates all headers longer
than 256 characters.
TCPIP_BUGS 1524: Comment out the privilege check for SSF in TCP/IP
V5.0A.
ECO F 19-JUN-2000 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE V5.0A-1F
Problem:
SMTP doesn't support VMS native MIME generated messages.
Solution:
Have SMTP check for MIME before inserting blank line
between headers.
ECO 3 updates
--------------
ECO G 22-JAN-2001 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE V5.0A-1G
Problem:
Access Violation occurs in MAILSHR when an attempt is made
to mail a file via SMTP, with incorrect file attributes.
Solution:
Bugs were identified in both VMSMail and SMTP which both
conspired to cause these ACCVIOs on Alpha (and other errant
behavior on VAX). In VMSmail, its RMS error AST routine was
fixed to include correct VMSmail context when LIB$SIGNALing
the error which SMTP was reporting via the MAIL$PROTOCOL
interface. Contact VMS Support regarding availability of
this patch.
In SMTP, don't always trust the longest-record-length file
attribute, establishing a minimum 1000-byte buffer for reading
mail files, and don't resignal RMS$_RTB to the MAILSHR condition
handler directly from SMTP.
ECO H 8-Feb-2001 Alpha and VAX
Images:
TCPIP$SMTP_MAILSHR.EXE V5.0A-1H
Problem:
Users that do not have SYSPRV privilege will cause security
alarms when sending SMTP mail.
Solution:
Set SYSPRV around the system call SYS$GETUAI to prevent
security alarms.
---------------------------------------------------------------------------
Corrections for COMPAQ TCP/IP Services V5.0A SNMP Images
---------------------------------------------------------------------------
ECO 1 updates:
--------------
ECO A 9-NOV-1999 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1A
Problem:
VMS SNMP returned a value for standard MIB-II variable sysObjectID
(1.3.6.1.2.1.1.2) of format 1.3.6.1.4.1.36.2.15.22.1 which reflects
standard Compaq (inherited from Digital) hierarchy.
22 is OpenVMS and the final 1 is eSNMP. For complete hierarchy see
url dssr.zko.dec.com/public/public/mib_tree.txt. Compaq Insight
Manager/XE needs this format to properly identify a system as
a server.
Solution:
Hard code the SNMP OID. The value of this does not use the
OpenVMS version number, so is consistent with the convention in
Tru64 UNIX, and requirements of Compaq Insight Manager.
ECO B 26-APR-2000 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1B
TCPIP$OS_MIBS.EXE V5.0A-1B
Problem:
Customer cannot send SNMPv1 trap to client.
Solution:
Backport configuration option for V5.1. Allows customer to set
up text configuration file to supplement information configured
through TCPIP$CONFIG.COM in standard configuration file
SYS$SYSTEM:TCPIP$CONFIGURATION.DAT.
Customer must do the following:
$ COPY TCPIP$SNMP_EXAMPLES:TCPIP$VMS_SNMP_CONF.DAT -
SYS$SYSDEVICE:[TCPIP$SNMP]*.*;
In this file add the following line:
config snmp_v1_trap_default
Note that settings from standard TCPIP$*.DAT config files supersede
those in TCPIP$VMS_SNMP_CONF.DAT.
ECO D 3-MAY-2000 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1D
TCPIP$HR_MIB.EXE V5.0A-1D
TCPIP$OS_MIBS.EXE V5.0A-1D
TCPIP$SNMPIVP.EXE V5.0A-1D
TCPIP$SNMP_REQUEST.EXE V5.0A-1D
TCPIP$SNMP_TRAPRCV.EXE V5.0A-1D
TCPIP$SNMP_TRAPSND.EXE V5.0A-1D
Problem:
Octet string data type does not display correctly. E.g., hex value
23 is printed as "#" (ASCII x23), not "23".
Solution:
Backport code from V5.1, as fixed in Tru64 UNIX, which does
appropriate test for data type and hex value and prints
appropriately.
ECO E 3-MAY-2000 VAX only
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1E
TCPIP$OS_MIBS.EXE V5.0A-1E
TCPIP$HR_MIB.EXE V5.0A-1E
Problem:
For get on interface counts on VAX, in particular TCPIP$OS_MIBS:
1. Decrease each time in process bytlm, eventual exit of agent
process with %SYSTEM-F-EXBYTLM.
2. For some interfaces, exit with SYSTEM-F-ABORT on first get attempt.
Solution:
1. Remove attempt to start interface for VAX in code to get counts.
Was using buffered I/O byte limit each time, not restored after
get attempt. Start not needed to get counts.
2. There is a bug in the driver for certain interfaces; so far ET
and FQ devices affected. The same behavior shows up with
$ MCR LANCP SHOW DEVICE /COUNTER {device}. In SNMP code, continue
if abort status returned; 0 returned for all counters.
NOTE FOR CUSTOMER: A fix for problem 2 is being worked on by
the OpenVMS group.
ECO F 4-MAY-2000 VAX and Alpha
Images:
TCPIP$OS_MIBS.EXE V5.0A-1F
Problem:
Version displayed for MIB-II variable sysDescr does not include
full OpenVMS version information.
Solution:
In itemlist for sys$getsyiw call, change SYI$_NODE_SWVERS to
SYI$_VERSION and in using resulting string, trim trailing blanks.
ECO G 10-MAY-2000 VAX and Alpha
Images:
TCPIP$HR_MIB.EXE V5.0A-1G
Problem:
Possible ACCVIO situation from inadequate memory allocation on
systems with large number of non-local mounted disks, when query
done for information on MIB subtree 1.3.6.1.2.1.25.3, frequently
between SNMP restarts.
Solution:
Reserve adequate space for string concatenation; simplify
code that moves strings between data structures.
ECO H 10-MAY-2000 VAX and Alpha
Images:
TCPIP$SNMP_REQUEST.EXE V5.0A-1H
Problem:
No retry or error ignore option available for MIB browser,
needed by field support.
Solution:
Three additional flags have been added for the snmp_request
command; -i max_ignores, -r max_retries and -s sleep_interval.
See the details in the PREVIOUSLY UNDOCUMENTED BEHAVIOR
section of the release notes under PTR 70-5-1299.
ECO I 16-JUN-2000 Alpha and VAX
Images:
TCPIP$UCP.EXE V5.0A-1G
TCPIP$NETSTAT.EXE V5.0A-1B
TCPIP$PING.EXE V5.0A-1A
TCPIP$IFCONFIG.EXE V5.0A-1A
TCPIP$SYSCONFIG.EXE V5.0A-1A
TCPIP$ESNMP_SERVER.EXE V5.0A-1I
TCPIP$HR_MIB.EXE V5.0A-1I
TCPIP$OS_MIBS.EXE V5.0A-1I
TCPIP$ARP.EXE V5.0A-1B
TCPIP$TRACEROUTE.EXE V5.0A-1A
Problem:
tcpip show route fails with INSVMEM when route database is large
Solution:
Backported fix from tcpip Version 5.1 that addresses this issue in
the unix_api library.
ECO 3 updates
--------------
ECO K 3-MAY-2001 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1K
Problem:
Master agent ACCVIO after a subagent stops unexpectedly
while processing a request.
Solution:
SNMP code did not handle newer DECC$SHR.EXE (OpenvMS V7.3)
implementation for select() handling invalid file descriptor.
ECO K 3-MAY-2001 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1K
TCPIP$HR_MIB.EXE V5.0A-1K
TCPIP$SNMPIVP.EXE V5.0A-1K
TCPIP$MOSY.EXE V5.0A-1K
TCPIP$MIBCOMP.EXE V5.0A-1K
TCPIP$OS_MIBS.EXE V5.0A-1K
TCPIP$SNMP_REQUEST.EXE V5.0A-1K
TCPIP$SNMP_TRAPRCV.EXE V5.0A-1K
TCPIP$SNMP_TRAPSND.EXE V5.0A-1K
Problem:
SNMP master agent does not handle various invalid packets correctly.
Solution:
Update SNMP code to handle Silver Creek test cases correctly.
ECO K 3-MAY-2001 Alpha and VAX
Images:
TCPIP$HR_MIB.EXE V5.0A-1K
Problem:
Handling of variables, performance, and reliability of HR_MIB
subagent needed to be fixed or improved.
Solution:
Variable retrieval, performance and reliability were fixed.
ECO K 3-MAY-2001 Alpha and VAX
Images:
TCPIP$ESNMP_SERVER.EXE V5.0A-1K
TCPIP$HR_MIB.EXE V5.0A-1K
TCPIP$SNMPIVP.EXE V5.0A-1K
TCPIP$MOSY.EXE V5.0A-1K
TCPIP$MIBCOMP.EXE V5.0A-1K
TCPIP$OS_MIBS.EXE V5.0A-1K
TCPIP$SNMP_REQUEST.EXE V5.0A-1K
TCPIP$SNMP_TRAPRCV.EXE V5.0A-1K
TCPIP$SNMP_TRAPSND.EXE V5.0A-1K
Problem:
Customer requested configuration mechanism for maximum SNMP packet
size accepted by the master agent.
Solution:
Configuration options now available to restrict SNMP packet size
allowed by the master agent.
ECO K 3-MAY-2001 Alpha and VAX
Images:
TCPIP$CHESS_SUBAGENT.EXE V5.0A-1K
TCPIP$ESNMP_SERVER.EXE V5.0A-1K
TCPIP$MOSY.EXE V5.0A-1K
TCPIP$OS_MIBS.EXE
TCPIP$SNMPI.EXE V5.0A-1K
TCPIP$SNMPIVP.EXE V5.0A-1K
TCPIP$SNMP_REQUEST.EXE V5.0A-1K
TCPIP$SNMP_TRAPRCV.EXE V5.0A-1K
TCPIP$SNMP_TRAPSND.EXE V5.0A-1K
Problem:
1. C++ compilations of sub-agent code with #include of esnmp.h
indicate mismatched #ifdef/#endif directives.
2. If flag -l or -t is used with snmp_request, and -n <> 0 or
-m <> 1, accvio may result. Also, if command was not getbulk,
any value for -n and -m were being ignored with no message to user.
3. OS_MIBS and HR_MIB subagents show IVDEVNAM error when DE600
interface is configured for TCPIP on system.
4. For long read requests, e.g., get with many variables specified,
error message with genErr is returned; should be tooBig.
Solution:
1. Move location of some directives, also remove redundant #define
lines. (Backported from 5.1)
2. Add code to check for presence of illegal combinations of flags
and either reset value to acceptable one (wi |
