Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!rayssd!galaxia!dave
From: dave@galaxia.UUCP (David H. Brierley)
Newsgroups: comp.sys.att
Subject: Re: Missing: 5 MB on a Unix PC.
Message-ID: <397@galaxia.UUCP>
Date: Tue, 10-Nov-87 23:42:17 EST
Article-I.D.: galaxia.397
Posted: Tue Nov 10 23:42:17 1987
Date-Received: Fri, 13-Nov-87 23:54:58 EST
References: <381@sdcjove.CAM.UNISYS.COM> <182@toylnd.UUCP>
Reply-To: dave@galaxia.RAY.COM (David H. Brierley)
Distribution: comp.sys.att
Organization: Dave's Very Own Personal System
Lines: 44

In article <182@toylnd.UUCP> dca@toylnd.UUCP (David C. Albrecht) writes:
>> My question is: "Why is there roughly a 5 MB difference between AT&T's 
>> number (15.5 MB) and my number (10.9 MB)?"
>
>In a word 'Swap tracks'.

In a word: 'bull crap'.  If the problem were 'swap tracks' then the question
would be "how come I say there is 20 meg but df only says 15 meg?".  I did
some research into this while I was writing a replacement for the df command
and discovered the real reason for the difference.  The df and du commands
both report the number of blocks being used and everyone knows that "a block
is a block is a block (is 512 bytes)", right?  Wrong!  The "blocks" that are
reported by the du command are "calculated" by taking the size of the file
in bytes and dividing by 512.  However, the blocks that are reported by df
are "calculated" by extracting the "total blocks used" field from the
super block and multiplying by 2!!  If you read the manual page for "fs"
you will find a field called "s_type" which is defined as either "Fs1b" or
"Fs2b".  On all the machines that I have access to this field is set to Fs2b.
This value indicates that the file system uses 1024 byte blocks!  What all of
this means is that if you allocate a 500 byte file the du command will report
that you have used 1 block but in reality you have used 1 "1024 byte block"
which df will report as 2 blocks.  As a result, any file that du reports as
having an odd number of blocks is wrong.  If you have 1000 files that fall
into this category then the numbers reported by du and df will differ by
1000.

If you dont believe me, try the following simple test:

1. run df and note how many blocks are reported as available

2. create a file containing 1 byte by typing "echo '' > junkfile"

3. run df again and compare the numbers.

4. type "du junkfile" and see how many blocks are reported.

I had to read the manual four times and run a test program a dozen times
before I fully believed that this was really the case.  I had gotten used
to this type of weirdness on BSD systems but I had never seen it on a SysV
machine before.
-- 
Dave Brierley
Home: rayssd!galaxia!dave  -or-  dave%galaxia@rayssd.ray.com
Work: dhb@rayssd.ray.com   -or-  {cbosgd,gatech,ukma,ihnp4}!rayssd!dhb