Re: umount /usr busy -- locale-archive ?



On Wed, 2008-05-28 at 00:02 -0500, Rubens Gomes wrote:
I have CentOS 5 (=RHEL 5). I am trying to umount the /usr in init level "1" but I cannot due to "/usr: device is busy". After "lsof" research I discoverd the following opened files used by the shell (/bin/sh -> bash).COMMAND PID USER FD TYPE DEVICE SIZE NODE NAMEsh 3180 root mem REG 8,3 56459040 199464 /usr/lib/locale/locale-archivesh 3180 root mem REG 8,3 25462 199708 /usr/lib/gconv/gconv-modules.cacheQuestions:1) How do I close/release these locale/gconv files from the sh process?2) If I cannot release the above files from the sh, is there a static version of bash or ksh that I could install/configure for the root? That is a shell that is not dynamically linked to any resources/files on the system?Having dynamically linked shell at level init 1 is *not* acceptable. The root shell should be completelly independent of any dynamic linking especially for level 1 administration. Is this a flaw of RHEL linux? I remember in Sun Solaris I used to have statically versions of ksh a
nd bourne shell under the root account.Thanks.Rubens Gomes

I understand the concern, but the response I have is two fold beginning
with a question: Why exactly do you need to unmount /usr?

A) Solaris 9 and earlier do have /sbin/sh statically linked so that it
is always able to be run. Sun has seen how infrequently that comes into
play and with Solaris 10.

B) Linux is not Solaris. This may seem curt and I do not mean it to be
so, but there are lots of differences between them. In this case the
most pertinent one is recovery. Suppose you do what a friend of mine at
uhacc.org did and install a bad libc. You get a brick. Hurray! You
can always boot off of some other random disk and do surgery to fix the
system. In my data center I have a CD envelope taped to each rack with
a current version of Knoppix in it. If I F*up a clean repair
environment is only a CD boot away. Should I ever find a box toasted
enough to make me unmount /usr I would rather boot off of a clean source
anyway.... God only knows what else is broken and could be further
harmed by the boot sequence.

With both A and B in line I would hesitate to call this a flaw.

Pat


--
redhat-list mailing list
unsubscribe mailto:redhat-list-request@xxxxxxxxxx?subject=unsubscribe
https://www.redhat.com/mailman/listinfo/redhat-list



Relevant Pages

  • Re: sunmanagers Digest, Vol 8, Issue 7 ... Slow Boot Up
    ... Just checking if we have the same cause for slow boot up? ... Solaris 8 Problem After Patches Installed ... Using PuTTY to send Control-? ... After installation of Solaris 8 on E250 I had to install Raptor GFX Open ...
    (SunManagers)
  • Re: sparc 10 hanging
    ... could mount the disk and edit a file to reset the time ... Which is apparently updated every time anything is written to the boot ... So, since they are so similar between SysV (Solaris) and BSD, I ... would expect the entry in the equivalent file for SunOs 4.1.3 (a BSD ...
    (comp.sys.sun.hardware)
  • Disk read error
    ... Solaris 9 SPARC boot server to Solaris 10 SPARC. ... -> OLD-BROADCAST TFTP Ack block 0 ...
    (SunManagers)
  • Re: xp/solaris dual boot problem
    ... I use System Commander boot loader, ... setting the Solaris partition to ID 82 ... >> It seems to me that solaris partition is all fine and healthy as I ... > and Linux. ...
    (comp.unix.solaris)
  • Booting Solaris from Debian GNU/Linux (Sparc)
    ... because of a lost root password. ... boot image doesn't know how to mount UFS partitions. ... Subnet Mask: 255.255.255.248 ... address by DHCP and fetches the inetboot file (from the Solaris 8 CDs). ...
    (SunManagers)