NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/vmsnet/internal/1196

Internet Message Format | 1992-08-25 | 3.9 KB

Path: sparky!uunet!mcsun!sunic!dkuug!tfl.dk!karsten
From: [e-mail] (Karsten Nyblad)
Newsgroups: vmsnet.internals
Subject: Re: Searching the lock list for locks associated with a file
Message-ID: <[e-mail]>
Date: 26 Aug 92 10:37:20 GMT
References: <920825142952.20201d28@vax4>
Organization: TFL, A Danish Telecommunication Research Laboratory
Lines: 69

In article <920825142952.20201d28@vax4>, [e-mail] writes:
> I am trying to create a utility that will search the lock list database on all
> systems in a cluster and return lock information about all the locks for a
> specified file or records within the file.
>
> I wrote a program that runs on every system in the cluster.  It executes the
> system service $GETLKI in kernal mode to search all the locks on a given system.
> This program looks for a matching resource name.  If one is found, the lock
> information will be displayed.  If there is no match, the lock could possibly
> be for a record within the specified file.  The program will check to see if
> this lock has a parent lock id.  If the id exists, the parent's resource name
> will be checked for a match.  If a match is found, then the lock information
> will be displayed.  If there is no match, then the lock will be skipped.
>
> This solution will not work.  It is very slow.  There are about 30,000 locks
> at any one time in use.  It also appears that my process uses a lot of the CPU
> and goes into RWSCS state quite often.  Therefore, I am looking for an
> alternative.
>
> I would like to be able to write some privileged code to calculate a hash value
> for the specified file and find the resource block from the hash table.
> Once I have this I will be able to find the sub resource blocks and all the
> locks associated with the resource and sub resources.
>
> I would appreciate any help with this matter.
>
>
> Robin Spector
> Bear Stearns Inc. & Co.
> New York, N.Y. 10167
> [phone]

First, listing the locks taken by the XCP wont help you much.  The locks
are on node basis, i.e. owned by the node in the cluster.  Within the
single node, access is synchronize by the XCP.  The easiest way to get
the lock on a file is geting the reference for the FCB, which has a
field containing the lockid on the lock of the file.  The reference
for the FCB is optained by a QIO to the XCP asking for the statistics
block.  Refer to chapter 1 of the I/O control manuals.  To find the single
processes locking a file you need to track the WCBs (Window Control Block),
which lists the processes locking the file on the current node.
I will send you a program that does the trick.  It needs SYSLCK
WORLD and CMKNRL priviledges.  The program is not completed,
but it does what you need on VMS version 5.4 and later.  DEC
cleaned up some symbol names in version 5.4, so it will not
compile on earlier versions.  The program will not list processes
locking the file as a global sections.  The program does not synchronizes
with the XCP, but is written with handlers so that it will not crash
your computer.  Refer to VMS file Systems Internals for details and
for information on how to synchronize properly with the XCP (order
number EY-F575E-DP).

The RMS locks are as far as I know organized into a lock tree with
a parent lock consisting of the prefix RMS$, the file id, and the
volume lock name.  The file id takes up 4 bytes if the file is not
located on a volume set and the file number is less than 2**16.
Otherwise it takes up 6 bytes.  You can chase through the
resource database and find all sublocks in that way.  Refer to
VAX/VMS Internals and Data Structures for details (order number
EY-C171E-DP)

What do you really want?  Finding the process locking the
ressource somebody wants?  It can be done by going through
the list of locks associated to "somebodies" process.  That list is
refered from the PCB.

Karsten Nyblad
TFL, A Danish Telecommunication Research Laboratory
E-mail: [e-mail]