NAME
MQSeries::Config::ChannelTable -- class for reading and writing channel table files of various versions.
SYNOPSIS
use MQSeries::Config::ChannelTable;
#
# To read a channel table file...
#
my @clntconn = ();
eval {
@clntconn = MQSeries::Config::ChannelTable->readFile
(
Filename => "/var/mqm/AMQCLCHL.TAB",
);
};
if ( $@ ) {
# Exception handling goes here...
}
#
# To write a channel table file...
#
eval {
MQSeries::Config::ChannelTable->writeFile
(
Filename => "/some/new/path",
Clntconn => [ @clntconn ],
Version => 4,
);
};
if ( $@ ) {
# Exception handling goes here...
}
DESCRIPTION
This class provides a pair of class methods for reading and writing the MQSeries client channel table file, allowing these important configuration files to be managed without interacting with a queue manager.
Normally, the only way to create a channel table file is to define all of the required CLNTCONN objects, via runmqsc (and an input script, perhaps), or PCF commands, then to copy the file
/var/mqm/qmgrs/QMgrName/@ipcc/AMQCLCHL.TAB
and distribute that to the necessary clients. Generation of the channel table can be very slow, especially if a large number of entries are created. What is much worse, however, is the complex logic necessary to create a set of different channel table files. If this is done on one queue manager, some of the gross inefficiencies in the channel table file may become an issue (see below).
This class makes it trivially easy to both read and write channel table files, of either version 4 (MQSeries 5.0) or 6 (5.1 and 5.2), by working directly with the (unfortunately undocumented) file format directly. You do not need a running queue manager at all.
IMPORTANT: Please understand that IBM does NOT document the format of the channel table file, and the author of this module has repeatedly asked them to disclose it. They have repeatedly refused.
Reverse engineering rules.... :-)
METHODS
Note that all of these methods are class methods, not object methods. Furthermore, both of these methods raise fatal exceptions (via Carp::confess) if errors are encountered, so the developer will have to trap those exceptions via eval(), as shown in the SYNOPSIS.
readFile
This method takes a HASH of key/value pairs as an argument, with the following keys:
Key Value
=== =====
Filename String (pathname to channel table file)
Debug Number
The return value is a list of HASHes, each of which represents a single CLNTCONN entry in the file. See below for the CLNTCONN key/value documentation.
- Filename
-
This must be an existing file, and it must be a valid version 4 or 6 channel table file. If the file can not be parsed successfully, then a fatal exception is raised.
- Debug
-
Setting this will cause this method to spew a lot of debugging data, of interest to noone other than the author. See the code if you care.
writeFile
This method takes a HASH of key/value pairs as an argument, with the following keys:
Key Value
=== =====
Filename String (pathname to channel table file)
Clntconn ARRAY of HASHes (see below)
Version Number (4 or 6)
Debug Number
The return value will be true if the file could be written successfully, and if not, a fatal exception is raised.
- Filename
-
This is the pathname to which to write the new file, which will be overwritten if it exists, mercilessly. Either this parameter, or the 'FileHandle' parameter, must be supplied.
- FileHandle
-
Instead of 'Filename', a previously opened file handle may be supplied. This can for example be used to print to STDOUT in a CGI program delivering channel table files from the web. On Windows, the caller is responsible to call
binmode
on the filehandle. - Clntconn
-
This is an ARRAY of HASHes, each of which represents a single Clntconn definition. See the next section for details of which key/value pairs are supported for the Clntconn definitions.
Note that this must be a complete list of all of the definitions you want to be written to the file. Also note that while the SYSTEM.DEF.CLNTCONN is supposed to be first, this code is forgiving, and will sort it first for you.
- Version
-
This specifies the version of the file to generate.
Version 4 is the version generated by an MQSeries 5.0 queue manager, and it can be read on 5.0 and later clients.
Version 6 is the version generated by an MQSeries 5.1 or 5.2 queue manager, and it can obviously be read by clients of the same release or better, however, it can also be read by 5.0 clients, in most cases.
The exception would appear to be when any of the following CLNTCONN fields are specified:
MsgExit MsgUserData SendExit SendUserData ReceiveExit ReceiveUserData
The reason is that the representation of these attributes hsa changed between these versions. In the 5.0 file, only a single exit can be specified for these attributes, but in 5.1 and later, they can be a list.
It is likely that a 5.0 client will not be able to "see" these attribute values, if the file is generated using the version 6 format. The author does not use any of these exits for client channels, and has not tested this, so as usual, YMMV (your mileage may vary, for the acronym impaired).
Clntconn HASH Specification
Each Clntconn entry in the channel table, as passed back to the caller of readFile, or as passed into writeFile by the application, is represented by a HASH reference of key/value pairs which are a subset of those support by the MQCD structure. Only the keys which are relevant to clntconn channels are used.
Note that these key/value pairs are also identical to those used by the MQSeries::Command channel commands, such as CreateChannel, InquireChannel, etc. and thus the same data structures can be used for both interfaces.
The valid key/value pairs for this HASH are:
Key Value (Max Length of strings)
=== =====
ChannelName String (20)
Version Numeric
ChannelType "Clntconn"
TransportType String (see below)
ChannelDesc String (64)
QMgrName String (48)
ModeName String (8)
TpName String (64)
SecurityExit String (128)
MsgExit String (128)
SendExit String (128)
ReceiveExit String (128)
MaxMsgLength Number
SecurityUserData String (32)
MsgUserData String (32)
SendUserData String (32)
ReceiveUserData String (32)
UserIdentifier String (12)
Password String (12)
ConnectionName String (264)
The specific use of each of these fields is documented in the IBM "MQSeries Programmable System Management" documentation, and the other partner docs. All of the string and number values are fairly obvious, with the following important exceptions:
- ChannelType
-
There is only one supported value for this attribute: "Clntconn". As a result, this is entirely optional, and need not be specified when creating files with writeFile. This key is returned by readFile for completeness, but its use is optional.
- Version
-
This field is returned by readFile(), however, it need not be specified for each individual channel passed to writeFile, since the version is itself an argument to writeFile(). It must be either 4 or 6, since those are the only formats supported by this version of this module.
Since all of the entries in a single file must be the same, specifying this on a per-channel basis is meaningless. They must all be the same anyway, so you only have to specify this once (as an argument to writeFile).
- TransportType
-
Rather than specify the specific binary value of the TransportType via a macro, the following keys can be given as strings:
Key Macro === ===== DECnet MQXPT_DECNET LU62 MQXPT_LU62 NetBIOS MQXPT_NETBIOS SPX MQXPT_SPX TCP MQXPT_TCP UDP MQXPT_UDP
Likewise, when a channel table file is read using readFile, it will map the integer value found in the MQCD structure to the appropriate string above.
Secrets of AMQCLCHL.TAB
Inefficiencies
The primary inefficiency of this format is its size. When this file is managed by the queue manager (which is your only choice without using this perl module), it grows and never shrinks. When channels are deleted, the linked list pointers are modified so that the deleted entry is merely skipped, but it is not removed from the file.
When new entries are added, a deleted entry will be searched for and reused if found. If not, a new entry will be added to the end of the file. The size problem occurs if you ever create a very large channel table file, since that will effectively extend the file size permanently.
In addition, each entry is a complete MQCD structure, even though most of the fields of the MQCD are not relevant for CLNTCONN channels. The total size of the MQCD, version 4, is 1540 bytes, of which only 1148 bytes are used for fields that are relevant to a CLNTCONN channel definitions. Most of that space is taken up by the various string fields which are always fully padded with blanks.
In the version 6 format, add an additional 108 bytes of irrelevant structure padding, and then, if message, send and/or receive exits are used, consider the 480 bytes of fixed space in the MQCD to be wasted, since IBM supports lists of those values in the V6 file format, and they are tacked onto the end of the MQCD structure. That consumes a variable amount of space, since those values do not appear to be padded.
File format
Many thanks to Mark Unger, who reverse engineered the file format. Most of this section is plagiarized from the notes he made, however, he worked for me at the time (December 2000), so I own them ;-)
The channel table is basically a doubly linked list of MQCDs. A few extra strings follow the MQCD as version 6 of the MQCD allows for variable length strings to contain the list of send/receive/message exits. (NOTE: Message exits are not really supported for CLNTCONNs, but the channel table file supports them as lists anyway. IBM has yet to explain this to me).
The channel table is an array of structures that have absolute offsets and relative lengths in some of the fields of each structure. All longwords are in netword byte order (big endian).
If we think of the file as a structure it would be a 4 character magic identifier followed by an array of channel definitions (forward/backward links plus MQCD plus extra strings).
The forward/backwards links are used to traverse the channel definitions during an MQCONN or DISPLAY CHL(*). The file may contain deleted channel definitions since the entire file is not rewritten just to delete/add/modify a channel definition.
To traverse all channel definitions including deleted ones, for instance when adding/modifying a channel definition, the forward link would not be used and the file would just be read as an array of channel definitions with some of them marked as deleted. When adding or modifying a channel, runmqsc will delete the channel by setting the length of the serialized MQCD (offset 4 in the ChannelDef) to zero.
There does not appear to be a free list for the deleted channels so runmqsc must just traverse the file as an array of channels to find deleted channels that it will attempt to reclaim for the modified or new channel definition. If a deleted channel definition is not large enough to contain the new or modified channel definition then it will just append a new channel definition to the end of the file.
The channels are sorted in lexicographic order, with a backwards pointer chain. The entries in the file are therefore in reverse order. Getting the order wrong will cause your MQSeries client library to hang, so you must take care not to modify the sort algorithm. There is one default entry in the file, SYSTEM.DEF.CLNTCONN, which is initialized from is MQCD_CLIENT_CONN_DEFAULT (in cmqxc.h). Initial value of fields in any added channels comes from SYSTEM.DEF.CLNTCONN of course.
Syntax chart:
Offset Type Value Description
====== ==== ===== ===========
File:
0 MQCHAR4 "AMQR" Magic identifier
4 ChannelDefList Channel definition list
ChannelDefList:
[ChannelDef ...] 0 0 terminates list of channel definitions
as the first field of ChannelDef is length
of the ChannelDef
ChannelDef:
0 MQLONG length length of complete channel definition
4 MQLONG length length of serialized MQCD
(0 if channel definition has been deleted)
8 MQLONG 0 (appears to be unused)
12 MQLONG offset forward link to next channel definition
(0 if last channel definition in doubly-linked list)
16 MQLONG offset backward link to previous channel definition
(0 if first channel definition in doubly-linked list)
20 SerializedMQCD MQCD with some additional strings fields appended
SerializedMQCD:
0 MQCD MQCD structure is defined in cmqxc.h
sizeof(MQCD) MQLONG 0
+4 MQLONG 0
+8 MQLONG length length of MQCDExitStrings
+12 MQBYTE52 0
+64 MQLONG timestamp alteration date/time
+68 MQCHAR64 ' ' x 64 unknown spaces
+132 MQCDExitStrings exit names and data lists
MQCDExitStrings:
[MsgExit 02]... 01
[MsgUserData 02]... 01
[SendExit 02]... 01
[SendUserData 02]... 01
[ReceiveExit 02]... 01
[ReceiveUserData 02]... 01
Each exit name and user data in an exit or user data list are terminated by byte 02. Each list is terminated by byte 01.
Notes on MQCD
For version 6 of the MQCD, the 128 byte MsgExit, SendExit, ReceiveExit fields are not used as well as the 32 byte MsgUserData, SendUserData, and ReceiveUserData fields. Those fields are only used if the MQCD is less than version 6. This is because version 6 for MQSeries 5.1 supports a list of exits for each of these and other fields are used to contain this list. The fields that were added to support these lists are MsgExitPtr, MsgUserDataPtr, SendExitPtr, SendUserDataPtr, ReceiveExitPtr, ReceiveUserDataPtr. But since these are pointer fields the value of them is zero in the MQCD and the strings are appended to the channel definition after the MQCD. The value of SendExitsDefined, ... are also zero and irrelevant even though send exits may be defined for the channel.
The IBM documentation is not entirely clear on the use of the MsgExit for CLNTCONN channels. For example, the usage description of the MsgExit attribute for channels in the "MQSeries Programmable Systems Management" document suggests that MsgExits are not supported for CLNTCONNs, yet they are treated specially in the channel table file itself. However, see below...
UserIdentifier and Password fields contains data in ciphered form provided but not opened by IBM.
IBM on channel table file format
Paul Clarke of IBM was kind enough to comment on the channel table file format on the MQSeries mailing list (Friday, August 15, 2003 9:47 AM). His comments are reproduced below:
It is still true that the file does not perform garbage collection
although it does defragment. So, you're right the file will never
shrink. This has not, until now been viewed as an important
requirement. Channel definitions have been treated as relatively
static. If you change an attribute the file will only grow if you
increase the size of channel entry ie. add another channel
exit. Since you should always be storing your channel definitions as
an MQSC script file (or generate them with Perl) then it is a simple
matter to delete the file and regenerate if you believe it is bigger
than necessary.
As far as MsgExits is concerned, it is true that Msg Exits are not
used in a CLNTCONN/SVRCONN connection and not invoked by either of
these channel type. The reason is that a MsgExit takes a pointer to
the MQXQH structure which is not involved at all in a
CLNTCONN/SVRCONN connection. We did debate having an API exit in the
client channels which would be passed pointers to the various
structures involved in whatever API call was being issued. However,
this was superceded by the API crossing exit which does a similar
thing but for all connections both local and client. The cause of
the confusion is that the channel table is the same format as the
channel file. In other words, although I've never used it, I suspect
you could use the MSDW Perl module to create a normal channel file
full of sender and receiver channels. In this case you clearly would
want to be able to add MsgExits.
AUTHORS
Phil Moore
Hildo Biersma
Mike Surikov