Carefully examine each fix for applicability before using it.

Fixes marked with "*" have been pre-applied to the currently available distribution file.

• Download Service Pack 1.5F with fixes pre-applied.

Download Fix File

When datagram fragments are reassembled to obtain the complete datagram, certain routing fields are not carried to the finished datagram. This causes subsequent improper routing of the return ACK.

Since the ACK is not received by the sender, the fragmented datagram will be retransmitted a number of times until the sending stack times-out the connection.

This fix corrects the problem.

Download Fix File

A queued connection that is not claimed by an application before the timeout interval has elapsed is terminated as being stale.

An error in the connection manager can falsely trigger the timeout when an extraneous datagram arrives from the remote host. "Extraneous datagrams" include the probes that some Windows-based stacks send after 3 seconds.

This fix corrects the problem.

Download Fix File

During shutdown, all executing channel programs for the CTCA must be ended.

In z/VSE 4.3, the PUBX control block was moved to 31-bit storage. Since the code that halts the I/O is running AMODE 24, a program check results.

This fix causes the PUBX to be referenced while in AMODE 31.

Download Fix File

The SOCKPASS phase is responsible for returning all socket completions to the requesting external callers. Before accessing storage in the requestor's partition, SOCKPASS performs a variety of tests to ensure that the caller is still executing and that the target area is still valid.

An exceptional condition can occur when the external caller or partition changes status between the instructions that check the validity and the instructions that move the data. Usually, this does not happen unless the TCP/IP partition is executing at a lower priority than the application's partition and SOCKPASS has code to protect it from even this low-probability event.

We now find an uncommon situation that can cause SOCKPASS to program check. This will occur when the requestor's TIB no longer resides in addressable storage.

Applying this fix will cause an interrupt "trap" to be set prior to accessing the external task's TIB. If the TIB address is invalid, an appropriate message will be issued, the socket will be dropped without posting, and normal processing will continue.

• It is important that programs do not abandon socket requests until they are posted complete. This is especially true of requests running under CICS, since this can result in CICS storage corruption.

Download Fix File

A too-restrictive timer test can cause the IPNET link driver to erroneously believe that a data packet was lost. The attempt to retransmit the packet can then cause the link to fail.

This fix adds redundant checking to ensure that the packet has actually been lost.

Download Fix File

Inbound datagrams are not correctly stamped by the IPNET link driver upon entry to the stack. If a connection relies on this value to return its response, the outbound datagram may be misrouted.

This fix adds the appropriate source routing information to the inbound datagrams.

Download Fix File

Under some circumstances, it is possible for UDP datagrams to be constructed with a null source IP address.

This fix ensures that the source IP address will be set to the home IP address of the sending LINK or ADAPTER.

Download Fix File

Although protected by a locking mechanism, an internal message routing routine was coded with a single instruction that falls outside the scope of the lock. Depending on timining, and especially during heavy FTP loads, this error can permit a subtask to have its registers scrambled which can then cause the stack to "hang".

This fix moves the instruction to within the scope of the lock.

Download Fix File

The Connection Block service routine can be called from multiple subtasks. Because of this, it is protected by a locking mechanism. However, after the lock is released, a lock-token pointer is cleared.

Under rare circumstances, this area can already be in use by a previously-waiting task, which can cause additional problems.

This fix moves the token to a register and clears the pointer before the lock is released.

Download Fix File

Under very rare circumstances, a critical storage address is being overlaid. This causes a compare-and-swap instruction to specify an odd address as a target. The result is a specification exception and abend.

The condition that causes the storage overlay requires that an IBBLOK be released and that the high two bytes of its address be equal to the low two bytes of the address of the connection block that owns it. The low two bytes of the IBBLOK's address must also equal a specific, but random, value.

This fix removes the possibility of storage overlay and allows processing to continue normally.

Download Fix File

The UDP protocol does not negotiate OPEN and CLOSE for a connection. When an application issues a CLOSE, the UDP connection simply ceases to exist.

Previous releases of TCP/IP permitted a SOCKET ABORT request (similar to TCP), but this was not carried through to 1.5F.

This fix restores the ability for an application to issue a SOCKET ABORT request. When issued, all outstanding socket requests will be posted as complete-with-error (SRCODE=8) and then the ABORT request will be posted as successful (SRCODE=0)

• Since UDP has no concept of "window", outbound data is always sent as soon as it is queued. Therefore, only RECEIVE sockets will be posted when an ABORT is issued.
• Since UDP has no concept of "reset" or "ACK", there is no way that a remote host can know when a connection has been aborted (or closed). Any inbound data on an "aborted" (or closed) UDP connection is simply discarded. This can result in large amounts of bandwidth and CPU being wasted to receive and discard unwanted data.

Download Fix File

An update to statistical counters is being performed following the release of a control block being used for source data. Depending upon timing, this control block may not be available when referenced.

This fix causes all control block references to be made prior to releasing ownership.

Download Fix File

An update to statistical counters is being performed following the release of a control block being used for source data. Depending upon timing, this control block may not be available when referenced.

This fix causes all control block references to be made prior to releasing ownership.

Download Fix File

Problems have been encountered during the conversion of large timer values. Depending upon the point where the problem is encountered, the failure may be trivial or unrecoverable.

This fix corrects the problem.

An additional problem can occur if an internal connection-locator hash table fills. Although this has only been reported once, this fix adds code to periodically validate the hash table contents and remove invalid and obsolete entries.

• This fix, along with ZP15F022 and ZP15F023 replaces ZP15F114.

Download Fix File

Problems have been encountered during the conversion of large timer values. Depending upon the point where the problem is encountered, the failure may be trivial or unrecoverable.

This fix corrects the problem.

• This fix, along with ZP15F021 and ZP15F023 replaces ZP15F114.

Download Fix File

Problems have been encountered during the conversion of large timer values. Depending upon the point where the problem is encountered, the failure may be trivial or unrecoverable.

This fix corrects the problem.

• This fix, along with ZP15F021 and ZP15F022 replaces ZP15F114.

Download Fix File

By RFC, the remote LPR/LPD port must be 515. The local port must be in the range of 721 through 731. However, the user may elect any port value from 0 through 65,535.

A problem occurs when either port is set to a value between 32,767 and 65,535. In this case, a message is displayed that shows a value other than the one specified.

This fix corrects the problem

• This problem is cosmetic only. The value specified by the user is the value used for the connection. Only the display is incorrect.
• Specifying a port outside of the "legal" range may prevent a successful connection with the remote LPD. Check the remote LPD's documentation to determine how closely it conforms to network standards.
• As with all TCP connections, if there is a conflict with the requested local port value, the port is incremented by one and rechecked until the conflict is resolved. This will occur only when there are multiple concurrent sessions with the same remote host and duplicate port numbers are assigned.

Download Fix File

LE/C-based applications load the $EDCTCPV phase to use BSD processing. Corrections have been added to the BSD code and this fix incorporates them into the LE/C interface.

For additional information, see fix ZP15F438.

Download Fix File

When an external connection request to a passive open cannot be completed due to a retransmission time-out on the outbound SYN, cleanup processing may encounter an error.

This fix corrects the problem.

Download Fix File

Outbound UDP datagrams (used for DNS look-up) do not properly resolve the ARP (MAC address) of the DNS server or gateway. This causes the DNS look-up to fail.

Once any other access is made to the first "hop" on the path to the DNS, the ARP address is resolved and all future calls function correctly. Unless your DNS resides on the local Ethernet, any traffic that uses your default gateway will prevent the problem.

The problem may be circumvented by including a PING to the DNS in your initialization deck. To do this, add an "INCLUDE member,DELAY" statement to your init deck ("member" is the name of the L-book that contains the "ping").

Download Fix File

Stage 1 shutdown processing is divided into two parts, A and B. Part A flags all connection to close. Part B flags all Daemons to close.

The problem occurs when a large number of closing connections permits the Telnet Daemons sufficient time to "recover" from the connection-closed condition and attempt to reset to a listen state.

This fix causes the Telnet Daemon to check the system's "quiesce" state, and proceed to shutdown.

Download Fix File

IBM has announced extensions to the OSA Express adapter for the z10 EC server. This fix adds the "OSAPORT=" parameter to the "DEFINE LINK" command.

This fix provides a method for using Port 1 of an OSA Express adapter adapter that has 2 ports per CHPID.

After applying this fix, you can add the "OSAport=n" parameter to the DEFINE LINK command for a TYPE=OSAX link driver. "n" should be 0 or 1 (default is 0).

This is a re-issue of this zap with the change being the prerequisite of ZP15F532.

Download Fix File

The IPNFPOWR File I/O driver sets a timer to ensure that OPEN processing completes in a reasonable time.

The timer is not canceled and may post the ECB after its storage has been reused for other purposes.

This fix cancels the timer once the OPEN completes.

Download Fix File

When an asynchronous I/O Active OPEN requests completes normally, TCP/IP is passing the socket number as the return code. This is required for Passive OPENs but not Active ones.

This fix causes successful Active OPENs to return a code of 0.

Download Fix File

Before returning data to an external application, TCP/IP applies several tests to ensure that the original socket requestor is still executing.

One of these tests is to check the "job start time". If the value has changed since the socket request was issued, the operation is canceled and message IPN855I is issued.

This message shows the expected and found time values.

The abend occurs because an invalid work area address is being passed to the System GETIME routine.

This fix causes a proper work area to be passed to the GETIME routine.

Download Fix File

Connection-related control blocks are located by means of a unique identifier, set during OPEN processing. Normally, this identifier is assigned when control blocks are allocated. However, this is not possible for "queued connections", since the identifier is not known until the connection block is paired with an OPEN request.

The problem occurs when BSD processing issues a STATUS call for a LISTEN that has just been paired with a queued connection block. If the asynchronous processing involved has not yet updated the hash tables with the new identifier, the STATUS request will fail with a "connection not found" error.

This fix causes the hash tables to be updated at the point of assignment so that subsequent STATUS calls will be able to locate the appropriate anchor block immediately.

Download Fix File

An internal table contains the address of each running pseudo task. Various processes (command and shutdown processing) use this table to iterate through all running tasks. The length of this table is being incorrectly set which means that processes that "run the table" stop too soon and miss some tasks.

This fix corrects the table length constant.

Download Fix File

An internal fix marker was not properly set in the phases provided with ZP15F202. Phase load processing notes that the value contained in the phase is not the expected value, and issues a warning message.

This fix corrects the phases' internal identifiers.

Download Fix File

This problem occurs when a remote host establishes an initial window size that is smaller than its requested Maximum Segment Size (MSS) and the first outbound data transmission is larger than the available window.

When this condition occurs, the logic that detects "Silly Window Syndrome" (SWS) prevents transmission of a partial, short datagram and waits for the window to be increased. If the remote host does not increase the window to the smaller of MSS or the queued data size, transmission does not occur until a timer has expired (nominally, 60 seconds).

This fix allows immediate transmission of data if the window is at its maximum advertised value.

Download Fix File

A piece of debugging code was not removed prior to creating the 1.5F installation job. This code causes the stack's internal dispatcher to serialize console messages at every pass.

Under normal conditions, this causes no delays since there are seldom messages queued for display.

However, if a long display is being produced (eg, QUERY ALL) or if some external process causes a console buffer shortage, TCP/IP processing can be severely impacted. One process known to be involved is the VSE ConnectorServer.

This fix removes the debugging code that causes the serialization.

Download Fix File

Problems have been encountered during the conversion of large timer values. Depending upon the point where the problem is encountered, the failure may be trivial or unrecoverable.

This fix corrects the problem.

An additional problem can occur if an internal connection-locator hash table fills. Although this has only been reported once, this fix adds code to periodically validate the hash table contents and remove invalid and obsolete entries.

*------------------------------------------------* &br.This fix erroneously replaces three phases. Since this makes future maintenance of these phases difficult or impossible, this fix is replaced with three individual fixes:

ZP15F021 - IPNIGARB
ZP15F022 - IPDRIVER &br.ZP15F023 - IPNET

Download Fix File

Hash tables are used to efficiently locate connection blocks based upon datagrams' IP addresses and ports. The problem occurs when an inbound SYN request causes a hash entry based on the stack's default IP address (SET IPADDR=) instead of the IP address that belongs to the Link being used. Future inbound datagrams on the connection are then "reset" because the appropriate hash table entry cannot be found.

This fix corrects the problem.

Download Fix File

Continuation of fix 115

Download Fix File

TCP connections are completed normally if and when each side has sent a FIN and had it acknowledged.

Two problems exist in this area.

Once a FIN has been received from the remote host and our own FIN sent, the connection manager does not always wait for an acknowledgement. This can cause a problem if a retransmission were to be required. However, this would be very seldom, since the packet length is quite small.

The second problem is that the connections manager terminates immediately following exchange of FINs and the connection control block is destroyed at the same time. If the remote host sends an extraneous ACK following this, it will not be associated with a connection and will be replied to with a RESET.

This fix corrects both problems. The connection manager will not terminate until an ACK has been received for its FIN (standard retransmission logic will apply). Additionally, a two-second "grace period" will be added to each connection to allow for receipt of a tardy ACK.

• No perceived delays are introduced by this fix. Responses to both the application and remote host occur as before.

Download Fix File

This message is assigned to class "DIAGNOSE" and is also flagged as "No Console". However, the message does get displayed on the operator's console if it is set to receive the "DIAGNOSE" message class.

The display of this message may be considered a nuisance. If so, this fix will correct the message's behaviour and send it only to SYSLST.

Download Fix File

During execution of this command, the stack attempts to display the percentage 31-bit storage available for IBBLOKs. This results in division by zero when no 31-bit storage is allocated to the partition.

This fix will cause the display to show the percentage relative to 24-bit storage when no 31-bit storage is allocated.

Download Fix File

Connections that do not have data moving across them are normally probed at the interval specified by "pulse time".

A problem occurs when the time of the last inbound data is updated without receipt of data. This prevents generation of a probe and possible detection of a lost connection.

This fix corrects the problem.

Download Fix File

FTPBATCH checks the versions of loaded phases against the expected versions, as provided by the stack.

When a phase is replaced as part of a fix, it is erroneously reported as an "incorrect version".

This fix corrects the problem.

• This problem is cosmetic only. Except for the messages, FTPBATCH processing continues normally.

Download Fix File

One of the reasons given by message IPN379I is "Partition start address is inconsistent". This text misleads the reader into believing that something is wrong with the partition.

This fix replaces the reason text with "Application buffer address is invalid."

Download Fix File

When using the TN3270E protocol, the Telnet Daemons are divided into two functional groups: listeners and effectors. A "listener" initiates a new TN3270E session, negotiates the session parameters, and then passes the connection to the appropriate effector Daemon which then handles the actual session. The listener then issues another passive OPEN and waits for the next user.

A problem can occur when session negotiation fails. For example, when the requested LUName is already in session. Under some circumstances, the TCP connection is not closed before the socket is reused, thus stranding the storage.

This fix causes open sockets to be closed prior to issuing the next OPEN.

Download Fix File

If an application attempts to do a RECEIVE and specifies a buffer address of "0", the operation will be suppressed and not posted complete.

This fix causes such a RECEIVE to fail when issued with RC 5, Reason 1.

Download Fix File

LPR documentation states that issuing the SET FCB=*NULL command should also result in an automatic SET TOF=0C0D command.

This is because setting an actual FCB causes TOF=NULL to be in effect.

This fix corrects the problem.

Download Fix File

An error in ZP15F124 causes a check to be made of a random location rather than the socket parameter list

This fix corrects the problem.

Download Fix File

This abend occurs following the issuance of the FLUSH command. An improper use of POST can corrupt an ECB being used by the connection manager.

This fix corrects the problem.

Download Fix File

Upgrades to UPD processing failed to take into account that socket requests might not include the foreign IP and/or port number.

This fix causes the foreign IP and port to always be assigned their established values.

Download Fix File

This fix ensures that IP addresses and ports are copied from inbound datagrams to the proper work areas.

• Fix ZP15F129 must be backed-out before applying this fix.

Download Fix File

This fix ensures that IP addresses and ports are appropriately returned to the application. It also causes the values to be shown in Socket Trace entries.

• Fix ZP15F129 must be backed-out before applying this fix.

Download Fix File

This fix installs a replacement message skeleton phase. It incorporates additional messages and corrections to existing messages.

Download Fix File

This fix corrects an improper message specification. The message is seldom, if ever, issued. However, this fix should be applied against the possibility of its being required.

Download Fix File

This fix corrects the update of "queued bytes" that is reported by IPT361I.

Download Fix File

At regular intervals, TCP/IP tests each entry in its connection hashing table. This includes issuing PAGESTAT to determine the validity of each control block.

The PAGESTAT SVC appears to consume more CPU than previously thought. This can be a drain on resources when there are a considerable number of connections active.

Since there are no known issues with the hashing tables, this fix puts the testing under control of DIAGNOSE CLEANUP.

Download Fix File

This fix corrects the IPT361I to indicate the number of IBBLOKs and data bytes that are queued to the connection. The message can occur twice: Once displaying inbound data not yet passed to the application and once displaying outbound data that has not yet been acknowledged.

This fix also adds an "ident" field to both QUERY CONNECTION and DIAGNOSE PERFORM displays. This is a unique identifier for a connection and can be used to associate display information with traced IBBLOKs and sockets.

Download Fix File

This phase replacement provides new, updated, and corrected message skeletons

Download Fix File

This fix causes an RC 8 to be returned for STATUS calls made after a RESET has been sent on the connection.

In addition, several more statistical values updated.

Download Fix File

This phase replacement provides new and updated diagnostic messages. It also adds the "Ident" field to most connection-related messages.

Flags are now set to ensure that STATUS requests are notified (RC 8) when a RESET has been sent to a remote host.

Download Fix File

Information from the Connection block is being inserted into the Socket block. This is valid and required for TCP connections, but not UDP.

This fix prevents the overlay of non-zero information in the Socket blocks during scheduling.

Download Fix File

As shipped, TCP/IP is configured for a maximum of 5,000 simultaneous connections. If this number is exceeded, it will not be possible to establish and maintain additional connections.

This fix increases the maximum simultaneous connections to 20,000.

Download Fix File

This phase replacement consolidates various fixes for the UDP protocol. It should be applied by all installations.

Download Fix File

This phase replacement consolidates various fixes for the UDP protocol. It should be applied by all installations.

Download Fix File

This phase replacement consolidates various fixes for the UDP protocol. It should be applied by all installations.

Download Fix File

This phase replacement consolidates various fixes for the UDP protocol. It should be applied by all installations.

Download Fix File

An extra debugging message is produced whenever a DEFINE TRACE or DEFINE SOTRACE command is entered.

This fix suppresses the message.

Download Fix File

This phase replacement consolidates various fixes for the UDP protocol. It should be applied by all installations.

Download Fix File

This fix will remove a number of diagnostic tests performed for each I/O request. These tests include a PAGESTAT SVC, which consumes an appreciable amount of CPU.

DIAGNOSE FILEIO re-enables the tests.

Download Fix File

This fix adds the requesting program's name, task ID, and requesting address to storage requests made for FRBLOKS. A Q STOR,SPID=FRBLOK will then be able to identify the process that owns the FRBLOK.

Download Fix File

When a GPS Daemon is started with the LOG=YES option, an FRBLOK is allocated in 24-bit partition GETVIS. If the Daemon is then shutdown, the FRBLOK is not released.

This fix corrects the problem and ensures that the logging file FRBLOK is released when the Daemon terminates.

• This problem can be circumvented by NOT specifying LOG=YES on the DEFINE GPSD command.

Download Fix File

Under some circumstances, the HTTP Daemon is obtaining an "extra" FRBLOK and then not freeing it.

This fix corrects the problem.

Download Fix File

This phase replacement corrects an error in processing Repeat to Address (RA) orders.

Download Fix File

Older releases of VSE do not permit use of Access Register operations without being in supervisor state.

This fix removes an unneeded test for access register mode in the CSOCKET subtask.

Download Fix File

This fix eliminates an unneeded DETACH

Download Fix File

This fix supplies the updated SOCKOPT macro.

Although this macro was distributed as part of fix ZP15F233, that fix was superseded by ZP15F237.

This fix will allow the IPNRBSDC phase and SOCKOPT macro to be maintained independently in any future fixes.

Download Fix File

When a socket ABORT is issued on a connection, subsequent socket requests (ie, CLOSE) may fail to complete.

This is a timing issue. Once an ABORT is issued, a RST is transmitted to the foreign host, all queued requests are flagged "complete with error", and the connection is closed. If the ABORT is quickly followed by CLOSE, the request may be lost.

This fix ensures that the CLOSE request will post completion back to the application.

• If ABORT or CLOSE is issued with FAST=YES, no pending or subsequent request will be "posted".
• CLOSE must still be issued following ABORT, since it is CLOSE processing that releases the socket's working storage in the application's partition.

Download Fix File

When operating in full-buffer mode, RECEIVE sockets do not complete until either the buffer is filled or a FIN is received.

A problem arises when the RECEIVE also carries a time-out value and the buffer is not filled before the time-out occurs. In this case, the RECEIVE completes as a "time-out" with the returned data length set to zero.

This fix modifies processing so that when a RECEIVE times-out, if there is queued data, it is returned and the SRCODE is set to zero.

Download Fix File

When the File I/O substask is attached, a parameter is being passed in an incorrect area.

This fix corrects the problem.

Download Fix File

When the File I/O substask is attached, a parameter is being passed in an incorrect area.

This fix corrects the problem.

Download Fix File

During a passive OPEN (listen), there is a very brief interval where it is possible for a Socket STATUS command to return data that shows the connection is in a "listen" state while certain other fields are not yet finalized. Specifically, this can affect the returned value assigned as "local port".

This fix delays setting the connection status to "listen" until the local port is definitely established.