running fsck on root filesystem

[Bill Moran]

kanwar at digitarchy.com wrote:
> Hi Bill, 
> 
> Thanks for a good reply. Please see my replies inline.
> 
> akanwar at digitarchy.com wrote:
> 
>>I have a remote machine on whose boot disk I want to run fsck. Is there any
>>way of booting freebsd form the network and then getting into fixit mode
>>without a cdrom or a floppy ?
>>
>>I am using pxeboot so far. What is happening is that when I put
>>"installFixitHoloShell" in the install.cfg then sysinstall throws me to the
>>shell that is VERY barebones. I can't run even a 'ls'. In this case how do
>>I figure out what disk slices I have in the system and on which I want to
>>run fsck ?
> 
> A lot you don't say here.  One thing is _why_ you want to do this.  Another
> is what you want to do.  Your description is pretty vague.
> 
> ---* Sorry I erred on the side of being laconic. Here is the situation: I
> have a file in /usr/something/file. Whenever I try to cp it to any other
> location, the machine freezes! I have tried a tar on the file (machine
> freezes) and a cat filename (same result). Here is a truss on the copy

This doesn't sound good.  Make sure you are getting good backups.

> rack1-104.sjc# truss cp ft-v05.2003-06-30.104501-0700 /tmp
> stat("/tmp",0xbfbffb90)                          = 0 (0x0)
> umask(0x1ff)                                     = 18 (0x12)
> umask(0x12)                                      = 511 (0x1ff)
> readlink("/etc/malloc.conf",0xbfbff984,63)       ERR#2 'No such file or
> directory'
> mmap(0x0,4096,0x3,0x1002,-1,0x0)                 = -2012909568 (0x88057000)
> break(0x806b000)                                 = 0 (0x0)
> break(0x806c000)                                 = 0 (0x0)
> break(0x806d000)                                 = 0 (0x0)
> break(0x806e000)                                 = 0 (0x0)
> stat("ft-v05.2003-06-30.104501-0700",0x806d160)  = 0 (0x0)
> stat("/tmp/ft-v05.2003-06-30.104501-0700",0xbfbffa84) ERR#2 'No such file
> or directory'
> open("ft-v05.2003-06-30.104501-0700",0x0,00)     = 3 (0x3)
> open("/tmp/ft-v05.2003-06-30.104501-0700",0x601,0100644) = 4 (0x4)
> mmap(0x0,5941955,0x1,0x1,3,0x0)                  = -2012905472 (0x88058000)
> <dies>
> 
> If I do a cat here is what happens:
> 
> rack1-104.sjc# truss cat ft-v05.2003-06-30.104501-0700 > /dev/null
> open("ft-v05.2003-06-30.104501-0700",0x0,00)     = 3 (0x3)
> fstat(1,0xbfbffb30)                              = 0 (0x0)
> readlink("/etc/malloc.conf",0xbfbffa90,63)       ERR#2 'No such file or
> directory'
> mmap(0x0,4096,0x3,0x1002,-1,0x0)                 = -2012909568 (0x88057000)
> break(0x805a000)                                 = 0 (0x0)
> break(0x805b000)                                 = 0 (0x0)
> read(0x3,0x805a000,0x1000)                       = 4096 (0x1000)
> write(1,0x805a000,4096)                          = 4096 (0x1000)
> read(0x3,0x805a000,0x1000)                       = 4096 (0x1000)
> write(1,0x805a000,4096)                          = 4096 (0x1000)
> read(0x3,0x805a000,0x1000)                       = 4096 (0x1000)
> write(1,0x805a000,4096)                          = 4096 (0x1000)
> read(0x3,0x805a000,0x1000)                       = 4096 (0x1000)
> write(1,0x805a000,4096)                          = 4096 (0x1000)
> read
> <dies>
> 
> At this point the machine stops responding to any console or network
> connections. After a while it spews ethernet flow control frames,
> indicating that the interrupts from the NIC to the kernel are not being
> serviced.
> 
> I did a memory check on the box and it looks ok. So now I suspected the
> disk...and was looking for a good way to test it.

I think you're already testing it.

> fsck is always run on the root filesystem at boot time, in preen mode. 
> If you're having some sort of filesystem errors, setting fsck_y_enable="YES" in
> /etc/rc.conf will cause the startup scripts to automatically run 'fsck -y' if
> fsck fails in preen mode.
> 
> ---* This is a good suggestion. I need to run it on the /usr slice though.

fsck is run on all filesystems that are listed in /etc/fstab during boot.

> From any shell, you should be able to mount filesystems readonly.  When
> mounted readonly, you can safely run fsck.  Use 'mount -r /mountpoint'. Many
> of your commands are not on the root filesystem, so you may need to mount
> (for example) /usr before you can execute certain commands.
> 
> ---* Problem: here is the fstab:
> # Device                Mountpoint      FStype  Options         Dump   
> Pass#
> /dev/ad0s1a             /               ufs     rw              1       1
> /dev/ad0s1f             /tmp            ufs     rw              2       2
> /dev/ad0s1g             /usr            ufs     rw              2       2
> ...
> 
> When I get to the shell (via sysinstall), I do

What do you mean "via sysinstall"?  Will the machine not boot?  Since this
is the /usr partition, there's nothing magical required.  Boot the system
normally.  Then shut down all processes that might access /usr and be sure
all files on /usr are closed.  Then:
umount /usr
fsck -yf /dev/ad0s1g
Repeat the fsck until it completes without reporting any errors.  Then
'mount /usr' and check to see if you can access the file correctly.
If you have trouble getting /usr unmounted, reboot the system.  At the
countdown press space, then enter 'boot -s' and hit enter.  When asked
for the shell, hit enter.  You're now in "single-user" mode, and /usr
isn't mounted, so you can surely do 'fsck -yf /dev/ad0s1g'.

-- 
Bill Moran
Potential Technologies
http://www.potentialtech.com