Free disk space that isn't there!

[Richard Todd]

peter at csn.org (Peter Steinauer) writes:

>In article <1991Jan18.165524.4327 at chinet.chi.il.us> henry at chinet.chi.il.us (Henry C. Schmitt) writes:
>>I recently ran into a problem and wonder if anyone knows a solution.
>>While working on compiling a couple large programs, I started getting
>>the error: Out of disk space on file system /usr (or something to
>>that effect).  I did a df which showed that /usr had over 10,000 free
>>blocks (> 5 MB!!).  However, doing a df as a normal user (I was root
>>for compiling) in another window showed 0 free blocks on /usr!!!

>Berkley unix file systems have a 'dead zone' on each partition that only 
>	root can see. This block is set aside as protection so that the 
>	disk partitions don't fill up beyond a 'safe' level... (Older 
>	versions didn't handle full file systems very well :-<... )

  More precisely, the Berkeley-style filesystem routines include a whole 
bunch of optimizations to try and make sure that file blocks are allocated
near each other (so as to minimize head seek time).  Obviously, such 
complicated allocation strategies can only work well if there's a lot of 
free space to work with; in particular, when the disk is over 90% full, the
optimized allocation strategies start to fail, meaning the system is forced
to allocate space wherever it can find it (even if it's not in an "optimal"
place), and your disk thruput starts to go down the drain.  So by default 
when a Berkeley-style filesystem is made, it's set up so that ordinary users
(not root) can't use up the last 10% of the disk. 

  However, if you, like me, have too many files and not enough cash for a new
disk drive :-), you can alter the default percentage of free space reserved
on that filesystem.  The command you want is "tunefs", used as follows.
Suppose that you've got /usr on slice 2 of SCSI drive 2.  Then you'd do
	umount /usr # very important; you can't (easily) do this on a 
	            # mounted filesystem!
	tunefs -m 0 /dev/rdsk/c2d0s2 # set minfree to 0 on filesystem on
	                             # /dev/rdsk/c2d0s2
	mount /usr

Note that the unmount is very important; as the man. page says:
          This program should work on mounted and active file systems.
          Because the superblock is not kept in the buffer cache, the
          changes only take effect if the program is run on unmounted
          file systems.  To change the root file system, the system
          must be rebooted after the file system is tuned.
And for those of you setting this on the root filesystem, when they say 
"reboot", they don't mean "/etc/reboot" or "shutdown" (apparently these cause
the old superblock info to be flushed to disk trampling over your changes); 
apparently you really have to do the tunefs on root and then hit the switch
on the side.  (Needless to say, you shouldn't try this except in single-user
mode...)  Perhaps in the next release they'll make tunefs a sash(8) utility
so we don't have to deal with this bogosity when doing tunefs on the root
fs...
--
Richard Todd	rmtodd at uokmax.ecn.uoknor.edu  rmtodd at chinet.chi.il.us
	rmtodd at servalan.uucp
"Try looking in the Yellow Pages under 'Psychotics'." -- Michael Santana