[linux-lvm] New LVM user, couple of problems

[Brian Poole]

On Wed, 12 Jul 2000, Daniel Whicker wrote:

> At 01:22 PM 7/12/2000, you wrote:
> >Setup is a P2 300, 2 Seagate 4.3G SCSI HDs, 128M RAM, BusLogic controller,
> >2.4.0-test3 kernel, and 0.8final LVM tools.
> 
> >The first issue I hit upon is with lv[extend,reduce]. Essentially I will
> >change the size of the lv (eg lvreduce /dev/vg00/lvol1 -L 200m) however
> >when I mount lvol1 the size shown is the same as it was before, according
> >to df. lvdisplay shows the correct size of 200m, but the system seems to
> >disagree saying the lv is still 6.7g which was the original size. [...]
> >This obviously is a bad thing, if diskspace
> >is exceeded it should just say no more diskspace, or similar, not go into
> >a loop that I can't get out of except by reboot ;).
> 
> Brian, when you resized the LV, you only changed the structure of the block 
> device, not the filesystem that sits on top of it.  The filesystem 
> structure still thinks that it was the size the block device was when the 
> filesystem was created.  Before changing the size of the logical volume, 
> you need to use an ext2 resizing tool to resize the filesystem.  If you're 
> using reiserfs, this is actually easier, and you can do it dynamically.

Ahh.. I see this now.

How about an update to the HOWTO to indicate this? Looking back I see the
'alternative' HOWTO does indicate I should be using e2fsadm or similar,
but the HOWTO directly on the LVM homepage does not indicate this (and it
skips 8 in it's list too ;). Or is the alternative HOWTO going to be
rolled into the official HOWTO one of these days so the official one isn't
being updated?

>From the LVM HOWTO:

7. reduce /dev/test_vg/test_lv to 900 logical extents with relative
extents by
   "lvreduce -l-700 /dev/test_vg/test_lv"
   or with absolute extents by
   "lvreduce -l900 /dev/test_vg/test_lv"

// Insert mention of expanding the filesystem here
 
9. rename a VG by deactivating it with
   "vgchange -an test_vg"   # only VGs with _no_ open LVs can be 
deactivated!
   "vgrename test_vg whatever"
   and reactivate it again by
   "vgchange -ay whatever"

9. rename a LV after closing it by

> >feeling malicious in general, I started a *very* large dd (eg dd
> >if=/dev/zero of=/lvol1/really-big-friggin-file bs=1024k count=7000). This
> >proceeded along fine for a while, during which I realized that it should
> >stop at 2G due to FS (ie ext2) limits. At 2G the ls output became rather
> >unusual, but the dd continued and df continued to mark the fs as fuller
> >and fuller. The dd finished succesfully (7000 records in + out) and df
> >says 6.8G of the 7.5G is being used on the lv. 
> 
> I note that you're using a 2.4.0 kernel.  The 2GB file limitation has been 
> removed from the 2.4 kernels.  LVM has nothing to do with how the 
> filesystem itself operates.  LVM is responsible for the block device, not 
> the data on top of it

oi. another good point, had just upgraded to the 2.4 for the first time to
test LVM and had forgotten about this. I guess an upgrade to ls/fileutils
would probably fix the freaky output as well.

> 		Daniel Whicker  (heimdall at mail.org)
 

Thanks for the pointers Daniel, much appreciated (and very happy that LVM
didn't break as easily as I was fearing ;)


-b