Category: News from the labs

New IFCIDs 316 – 400/401 for DB2 z/OS Audit “on the fly”

I use the Monty Python title as it reminded me of an old telephone book joke in England: look up „boring“ in the Yellow pages – there you would find a little bit of text that simply said „See Civil Engineering.“ It still makes me laugh these days.  Anyway, back to the newsletter…

Auditing is often looked down upon as being boring, tedious and of no worth. This is rubbish of course! Without auditing we would not be allowed to do anything these days. I hardly know any DBAs who still have SYSADM.

It is just too darn powerful!

The statistics also tell us that nearly all “hack” attacks are “inside jobs” from the very people we know and trust. The statistics are also pretty brutal on the Mainframe/Server divide – Mainframes are very rarely successfully attacked, but Servers (see Sony et al.) are all the time. I think there should be a lot more auditing on the little iron really!

So how do you audit on the host side of the street?

Do you actively check what is happening? Or do you just wait for the thought police to arrive?

Enabling Auditing from DB2 10

When the enhanced IFCID 316 and the new IFCIDs 400/401 were introduced, it closed a gap in the ability to actually Audit your system “on the fly”. Using these IFCIDs  you could actually trap/monitor/audit all of the SQL running in your Plex. These IFCIDs are also nearly free as the overhead is “background noise” levels of CPU.

Now, let us imagine that you are capturing all of this data. That you are regularly snapping both the DSC and the SSC, (that’s what I call the EDMPOOL cache for Static SQL statements), that this data is all being rolled up and saved into a DB2 Data warehouse, and that you are triggering Batch jobs to analyse for Audit – reporting using various queries just to see if anything “untoward” is starting/or is happening!

What queries would you want to run?

– I have a few straight off the bat here: Who is reading from the Payroll table?
– Who is updating the Payroll or Employee tables?
– Who is accessing *any* table from the internet?
– Is anybody being really clever and using ODBC to select from my production tables?
– How many userids are out there using my data?
– Has any SYSADM enabled userid done any work on my system today?

All good Audit questions that you could put into operation very simply indeed!

Oh Lucky Man!

Now as luck would have it, we have a software product called SQL WorkloadExpert that actually does all this for you! What’s more you can expand it as much as you like! Cool huh?

Who is accessing *any* table from the Internet?…
Has any SYSADM enabled userid done any work on my system today?…
How many userids are out there using my data?…

Looking at this screen shot you can get an idea of the possibilities – Look at the Workstation name column for instance. “192.xxx” is the intranet. If any other tcp/ip address showed up here, it would be, shall we say, “worrying”. You can also see great stuff like “EXCEL.EXE” in the Transaction name column, and that a certain Mr. Boxwell has been running a few things from lots of data sources, I wonder what he’s up to?

Here you can see a nice list of “Intents” against a given table (in this case SYSIBM.SYSTABLES)

Now you can see who did what type of insert against a given object.

This is a list of *all* Primary Authorization IDs or Collections and Packages that have run
– Any intruders?

This is a list of all User Data Updates done by users with SYSADM authority in the last workload.

Triggering the Auditor

All of this with drill down to the actual SQL that was executed. Cool stuff and very handy indeed! But this is “past the point” and what you really need is a Batch Style interface that runs the SQLs and triggers alarms on the host before someone looks at the pretty GUI!

I would do this with a nice little set of batch Spufi’s that get post processed and either e-mailed directly to the Auditors, or WTO’d  as an alarm action that then triggers a batch job to do something else.

What would you like to Audit? Or what would your Auditors like to see? How do you currently accomplish this?

I would be fascinated to hear from you!

TTFN,

Roy Boxwell

How will you deal with loop-hole usage in production code?

The IFCIDs 366 and 376

Db2 provides many and varied IFCIDs. But for today, I’m most interested in the 366 and 376. The 366 is available in Db2 10 and the 376 in Db2 11. Now I like to call these “BIFCIDs” because they are triggered whenever a BIF is used that will behave differently than it is currently used when moving to the next release of Db2. (It’s also triggered when changing Application Compatibility settings in Db2 11 and higher).

So where’s the BIF?

BIF Usage Video (11min:)       Presentation

Well, a BIF is a Built-In Function such as CHAR, DECIMAL, etc. There are hundreds of them these days. In the last few Db2 releases, IBM has changed a few to make Db2 more compatible with SQL standards. They have actually closed a couple of loop-holes, where “bad” data could be accepted and processed.

BIF usage

Loop-hole user?

What happens is: someone somewhere found this loop-hole and used it in production code. Now when you upgrade your Db2, this code will either fail or give erroneous results – which is never good. Hence IBM created the IFCID 366. This is output every time an SQL statement is PREPARED, or executed, that contains a candidate BIF. There were so many of these, that IBM introduced a sort of condensed version so it only triggered one for the first execution, or prepare, but sadly that IFCID—376—is only for Db2 11.

Where can, or will, this really hurt?

Looking into the documentation for these IFCIDs you will see a long list of when they are written:

***********************************************************************
**  IFCID 0366 is a serviceability trace.                            **
**  It can be used to identify applications that are affected        **
**  by incompatible changes.                                         **
**  The QW0366FN field indicates the type of incompatible Change:    **
**                                                                   **                                                      
**  QW0366FN = 1                                                     **
**  Indicates that the pre Version 10 CHAR built-in function has     **
**  been invoked. There is an incompatible change to the output of   **
**  the CHAR function for some decimal data. The zparm               **
**  BIF_COMPATIBILITY and/or the SYSCOMPAT_V9 schema have been used  **
**  by this application to get the old behavior. Please make the     **
**  appropriate changes and rebind with the SYSCURRENT schema to     **
**  use the Version 10 CHAR(decimal) built-in function.              **
**  (PM29124 V10 only, usermod V8/V9)                                **
**                                                                   **
**  QW0366FN = 2                                                     **
**  Indicates that the pre Version 10 VARCHAR built-in function or   **
**  CAST(decimal AS CHAR or VARCHAR) has been invoked.               **
**                                                                   **
**  QW0366FN = 3                                                     ** 
**  Indicates that an unsupported character representation of a      ** 
**  timestamp string was used. PM48741 V10 only.                     **
**                                                                   ** 
**  QW0366FN = 4                                                     ** 
**  A QW0366FN 4 record indicates that the statement uses the        **     
**  word ARRAY_EXISTS as an unqualified user-defined function Name   **   
**  in a context that may be incompatible with Version 11.           **     
**                                                                   **
**  QW0366FN = 5                                                     ** 
**  A QW0366FN 5 record indicates that the statement uses the        **
**  word CUBE as an unqualified user-defined function Name           **
**  in a context that may be incompatible with Version 11.           **
**                                                                   **
**  QW0366FN = 6                                                     **
**  A QW0366FN 6 record indicates that the statement uses the        **
**  word ROLLUP as an unqualified user-defined function Name         **
**  in a context that may be incompatible with Version 11.           **
**                                                                   ** 
**  QW0366FN = 7                                                     **
**  A QW0366FN 7 record indicates that Db2 for z/OS server issued    **
**  a SQLCODE -301 for incompatible data type conversion from        **
**  string data type (e.g. CHAR, VARCHAR, GRAPHIC, VARGRAPHIC        **
**  etc.) to numeric data type in V10 CM mode when implicit          **
**  cast is not supported or V10 NFM mode when DDF_COMPATIBILITY     **
**  zparm is set to DISABLE_IMPCAST_NJV or SP_PARMS_NJV to           **
**  disable implicit cast, and the client is CLI Driver              **
**  or v11 NFM mode & APPLCOMPAT = V10R1 when DDF_COMPATIBILITY      **
**  is set to SP_PARMS_NJV or DISABLE_IMPCAST_NJV to disable         **
**  implicit cast either from string data type to numeric or         ** 
**  from numeric data type to string data type.                      **
**                                                                   **
**  QW0366FN = 8                                                     **
**  A QW0366FN 8 record indicates that Db2 for z/OS server           **
**  returned output data match the data types of the                 **
**  corresponding CALL statement arguments when DDF_COMPATIBILITY    **
**  zparm is set to SP_PARMS_NJV.                                    **
**                                                                   **
**  QW0366FN = 9                                                     **
**  A QW0366FN 9 record indicates a data type conversion from        **
**  a TIMESTAMP WITH TIME ZONE input to a TIMESTAMP data             **
**  during input host variable bind-in process on server when        **
**  DDF_COMPATIBILITY zparm is set to IGNORE_TZ to ignore the        **
**  time zone information sent by Java IBM Data Server Driver.       **
**                                                                   **
**  QW0366FN = 10                                                    ** 
**  RTRIM, LTRIM or STRIP version 9 being used with mixed data       **
**                                                                   **
**  QW0366FN = 1101                                                  ** 
**  Indicates that the INSERT statement that inserts into an XML     **
**  column without XMLDOCUMENT function has been processed (which    **
**  should result in SQLCODE -20345 when run on Db2 release prior    **
**  to V11). Starting with V11, SQL error will no longer be issued.  **
**  Application will no longer recieve SQLCODE for this Statement.   **
**                                                                   ** 
**  QW0366FN = 1102                                                  **
**  Indicates that V10 XPath evaluation behavior was in effect which **
**  resulted in an error. For instance, a data type conversion error **
**  could have occured for a predicate that would otherwise be       **
**  evaluated to false. Starting from V11, such "irrelevant" Errors  **
**  might be suppressed so an application might no longer recieve    **
**  the SQLCODE for this Statement.                                  **
**                                                                   **
**  QW0366FN = 1103                                                  **
**  Indicates that a dynamic SQL uses the ASUTime limit that has     **
**  been set for the entire thread for RLF reactive governing.       **
**  For instance, when a dynamic SQL is processed from package A,    **
** if the ASUTime limit is already set during other dynamic SQL      ** 
** processing from package B in the same thread, the SQL from        **
** package A will use the ASUTime limit set during the SQL           **
** processing from package B. Stating with v11, dynamic SQLs from    **
** multiple packages will use the ASUTime limit that is set          **
** considering its own package information.                          **
**                                                                   **
** QW0366FN = 1104, 1105, 1106, 1107                                 **
** Indicates that CLIENT special register (CLIENT_USERID,            **
** CLIENT_WRKSTNNAME, CLIENT_APPLNAME, CLIENT_ACCTNG) has been set   **
** to a value that is longer than what is supported prior to V11.    **
** A shorter value has been used instead.                            **
**                                                                   **
** QW0366FN = 1108                                                   **
** Indicates that CLIENT special register (CLIENT_USERID,            **
** CLIENT_WRKSTNNAME, CLIENT_APPLNAME, CLIENT_ACCTNG) has been set   **
** to a value that is longer than what is supported prior to V11.    **
** Truncated values upto the supported lengths prior to v11 have     **
** been used for RLF table search instead.                           **
**                                                                   **
** QW0366FN = 1109                                                   **
** Indicates that CAST(string AS TIMESTAMP) was processed for the    **
** input string of length 8 and input was treated as a store clock   **
** value (or input string was of length 13 and was treated as a      **
** GENERATE_UNIQUE value). This behavior is incorrect for a CAST     **
** and is valid for TIMESTAMP built-in function only. This behavior  **
** is being corrected in Db2 11 so that input to CAST is not         **
** treated as a store clock value nor GENERATE_UNIQUE.               **
**                                                                   **
** QW0366FN = 1110                                                   **
** Indicates the integer argument of SPACE function is greater       **
** than 32764.                                                       **
**                                                                   **
** QW0366FN = 1111                                                   **
** Indicates the optional integer argument of VARCHAR function       **
** has a value greater than 32764. *                                 **
***********************************************************************

Useful stuff indeed!

Phew! Not a bad list, huh? Now you see why these IFCIDs are so useful. It could well be, that you have none of these “alive” in your system today. Or, of course, it could be that you get millions of the things! Somehow you will have to work out a way to save the data, analyse it to get to the root cause, and then, finally, fix the problem(s).

Saved by APPLCOMPAT?

You could argue that the new Db2 11 parameter Application Compatibility will save you, but this is really a false economy. All it enables is the guarantee that the code will still “run”. However, in two more Db2 releases the code will fail and, in two more releases – so about six years – who will even know *how* to change which piece of source code and, perhaps even, where is that source code?

Saved by BIFCIDs

BIF Usage Video (11min:)       Presentation

Personally, what I would do, is : to run our SQL WorkloadExpert tool to trap all the required [B]IFCIDs for a few hours (at first!). Then I would analyse the results, fix the code where it needs fixing – and repeat! I would keep doing this until no IFCID records are coming out and I would be set!

What is even better, is that our SQL WorkloadExpert will work correctly even when any new QW0366FN values appear – so when IBM decides to add another code (Like the new values 9 and 10 above for example) this BIF Usage still works correctly.

Of course, you may have another tool that you use at your site.

Can it see “Where’s the BIF?”

How will you deal with loop-hole usage in production code?

As usual, any question or comments gladly welcome!

TTFN

Roy Boxwell

Can you Track your SQL Back in time? Do you have standards to catch QUERYNO abuse?

This month I’d like to confront you all with a really nasty little bit of truth…

We have had the ability to specify a number to “mark” an SQL in DB2 for years and years now, using the QUERYNO keyword.

This number is used for the QUERYNO columns of the plan tables for the rows that contain information about this SQL statement. But how many of you out there are actually using this? Even if you do use it, how many of you have got standards and practices in place to guarantee that there is no “misuse” of this feature? Yep, this month we are going to take a long, hard look in the mirror…

Definition of QUERYNO

The late great Richard A. Yevich summed it up neatly in 2001 as:

QUERYNO was introduced in V6 as a way to specify a particular QUERYNO for a SQL statement and match it to a HINT, which reference[s] the QUERYNO column in the plan table. It is used at BIND time by DB2, and nothing else.

Regards,
Richard

Of course these days, in fact as of DB2 V7, the QUERYNO has its own column in the SYSIBM.SYSPACKSTMT and can be used for “other things”. You can happily append QUERYNO to static (in fact also Dynamic SQL – but more about that later!) SQL like this:

How it Looks

EXEC SQL
    SELECT MAX(IBMREQD)
    INTO :O2TEST5-COL1:O2TEST5-NULL
    FROM SYSIBM.SYSDUMMY1
    WITH UR
    QUERYNO 777777
END-EXEC

All it does is “tie” the SQL to a user-defined arbitrary number, normally for HINT usage. Now the whole point of this was so that you, the DBA, could see that last week QUERYNO 77777 was working a treat, but this week it has suddenly gone all tablespace scanny on you. Your job, as always, is to fix the problem – so that DB2 can keep on chugging along!

Working without QUERYNO

Let us imagine that you do *not* use QUERYNO, so the SQL looks like:

EXEC SQL
    SELECT MAX(IBMREQD)
    INTO :O2TEST5-COL1:O2TEST5-NULL
    FROM SYSIBM.SYSDUMMY1
    WITH UR
END-EXEC

After long and hard analysis(!) you decide to add a predicate to speed it all up:

EXEC SQL
    SELECT MAX(A.IBMREQD)
    INTO :O2TEST5-COL1:O2TEST5-NULL
    FROM SYSIBM.SYSDUMMY1 A
    WHERE A.COL1 =
          (SELECT B.COL2 FROM ROY.TABLE B WHERE B.COL3 = A.COL2)
    WITH UR
END-EXEC

Can you track this SQL back in time?

Now, however, you trace this SQL you will find it quite hard to extremely impossible, to match this version to the “old” version in any sort of DB2 SQL Data Warehouse. Even our SQLWorkloadExpert will have trouble trying to do so because the SQL has completely changed.  Introducing new tables, changed SQL text etc. however the core competence of the SQL has not changed. All that has changed is the way we get the correct answer.

Now why you should all be using QUERYNO is obvious! The original query:

EXEC SQL
    SELECT MAX(IBMREQD)
    INTO :O2TEST5-COL1:O2TEST5-NULL
    FROM SYSIBM.SYSDUMMY1
    WITH UR
    QUERYNO 777777
END-EXEC

Brave new coding standards

Is now changed to be:

EXEC SQL
    SELECT MAX(A.IBMREQD)
    INTO :O2TEST5-COL1:O2TEST5-NULL
    FROM SYSIBM.SYSDUMMY1 A
    WHERE A.COL1 =
          (SELECT B.COL2 FROM ROY.TABLE B WHERE B.COL3 = A.COL2)
    WITH UR
    QUERYNO 777777
END-EXEC

And now you *can* match this SQL through time with no problem? Neat huh?

Problems problems problems…

Problems however are plentiful… first up is “cut-and-pasteitis”. This is a very nasty, and highly contagious, programmer disease, not just limited to green screen usage but also to GUI developers. It is *very* tempting to just grab a “neat” piece of SQL and drop it into the code and you are done! But if this contains an existing QUERYNO, or even no QUERYNO, then you are destroying your data warehouse.

You must enable QA code checking in order to stop any possible misuse of this Feature.

This entails checking all source code for the existence of a QUERYNO where relevant and needed (Remember that QUERYNO can only be used on DECLARE, SELECT, DELETE, INSERT, MERGE, REFRESH and UPDATE statements). The checking of the numbers used in order to make sure you do not have the number 1234546 55 different times. Also check for the allowance of “gaps”, for the really deranged among us who always want the numbers to go up sequentially in the code (I usually start at 1000 and go up in 1000 increments to start with, as I am indeed one of those poor deranged souls!).

Playing with time travel

Once all this is in place, and you are trapping and mapping your data,you can then do wonderful SQL time line Analysis  to really see what an SQL, or even a group of SQLs, did over time!

– Did the change I do really help?

– By how much?

– Can you match the SQLs next to each other etc.

Not only that, but when you are trapping everything, why not use it to see if QUERYNO is actually being used correctly?

(We are currently building a “Use Case” for our SQL WorkloadExpert to do just that.)

No way for Dynamic?

I mentioned earlier that the Dynamic world is a bit different. Of course each Dynamic SQL gets its own Statement Id when it comes into the cache, (similar to the Statement Id static SQL gets in the BIND and also in the EDMPOOL cache), but you have no “fixed” point to compare to. Unless, that is, you have a nice SQL Data Warehouse where you store all your Dynamic SQL to enable time travel queries here too!

Please note one very important bit of information here

Our question to you for Research purposes

How do you use or *not* use QUERYNO? Do you have standards to catch abuse?

For research purposes please send me an answer with Yes or No

I would dearly love to know!

Looking forward

As simple as it sounds the big problem has always been time…

Who has enough time to add QUERYNO to all their old legacy code? No-one of course!

But then SQL WorkloadExpert can help there too using our “Multi-Row Fetch” Use Case to show you the legacy code that gives the “biggest bang for the buck” and enables you to kill two birds with one stone!

As you change the heavy hitters for MRF, simply use an edit macro to add in QUERYNO to all of the embedded SQL

– Simple as that!

As usual any queries or criticism gladly accepted!

TTFN

Roy Boxwell